68 files changed, 8287 insertions, 2286 deletions

diff --git a/arch/tile/include/arch/chip_tilegx.h b/arch/tile/include/arch/chip_tilegx.h
new file mode 100644
index 000000000000..ea8e4f2c9483
--- /dev/null
+++ b/arch/tile/include/arch/chip_tilegx.h
@@ -0,0 +1,258 @@
+/*
+ * Copyright 2011 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+/*
+ * @file
+ * Global header file.
+ * This header file specifies defines for TILE-Gx.
+ */
+
+#ifndef __ARCH_CHIP_H__
+#define __ARCH_CHIP_H__
+
+/** Specify chip version.
+ * When possible, prefer the CHIP_xxx symbols below for future-proofing.
+ * This is intended for cross-compiling; native compilation should
+ * use the predefined __tile_chip__ symbol.
+ */
+#define TILE_CHIP 10
+
+/** Specify chip revision.
+ * This provides for the case of a respin of a particular chip type;
+ * the normal value for this symbol is "0".
+ * This is intended for cross-compiling; native compilation should
+ * use the predefined __tile_chip_rev__ symbol.
+ */
+#define TILE_CHIP_REV 0
+
+/** The name of this architecture. */
+#define CHIP_ARCH_NAME "tilegx"
+
+/** The ELF e_machine type for binaries for this chip. */
+#define CHIP_ELF_TYPE() EM_TILEGX
+
+/** The alternate ELF e_machine type for binaries for this chip. */
+#define CHIP_COMPAT_ELF_TYPE() 0x2597
+
+/** What is the native word size of the machine? */
+#define CHIP_WORD_SIZE() 64
+
+/** How many bits of a virtual address are used. Extra bits must be
+ * the sign extension of the low bits.
+ */
+#define CHIP_VA_WIDTH() 42
+
+/** How many bits are in a physical address? */
+#define CHIP_PA_WIDTH() 40
+
+/** Size of the L2 cache, in bytes. */
+#define CHIP_L2_CACHE_SIZE() 262144
+
+/** Log size of an L2 cache line in bytes. */
+#define CHIP_L2_LOG_LINE_SIZE() 6
+
+/** Size of an L2 cache line, in bytes. */
+#define CHIP_L2_LINE_SIZE() (1 << CHIP_L2_LOG_LINE_SIZE())
+
+/** Associativity of the L2 cache. */
+#define CHIP_L2_ASSOC() 8
+
+/** Size of the L1 data cache, in bytes. */
+#define CHIP_L1D_CACHE_SIZE() 32768
+
+/** Log size of an L1 data cache line in bytes. */
+#define CHIP_L1D_LOG_LINE_SIZE() 6
+
+/** Size of an L1 data cache line, in bytes. */
+#define CHIP_L1D_LINE_SIZE() (1 << CHIP_L1D_LOG_LINE_SIZE())
+
+/** Associativity of the L1 data cache. */
+#define CHIP_L1D_ASSOC() 2
+
+/** Size of the L1 instruction cache, in bytes. */
+#define CHIP_L1I_CACHE_SIZE() 32768
+
+/** Log size of an L1 instruction cache line in bytes. */
+#define CHIP_L1I_LOG_LINE_SIZE() 6
+
+/** Size of an L1 instruction cache line, in bytes. */
+#define CHIP_L1I_LINE_SIZE() (1 << CHIP_L1I_LOG_LINE_SIZE())
+
+/** Associativity of the L1 instruction cache. */
+#define CHIP_L1I_ASSOC() 2
+
+/** Stride with which flush instructions must be issued. */
+#define CHIP_FLUSH_STRIDE() CHIP_L2_LINE_SIZE()
+
+/** Stride with which inv instructions must be issued. */
+#define CHIP_INV_STRIDE() CHIP_L2_LINE_SIZE()
+
+/** Stride with which finv instructions must be issued. */
+#define CHIP_FINV_STRIDE() CHIP_L2_LINE_SIZE()
+
+/** Can the local cache coherently cache data that is homed elsewhere? */
+#define CHIP_HAS_COHERENT_LOCAL_CACHE() 1
+
+/** How many simultaneous outstanding victims can the L2 cache have? */
+#define CHIP_MAX_OUTSTANDING_VICTIMS() 128
+
+/** Does the TLB support the NC and NOALLOC bits? */
+#define CHIP_HAS_NC_AND_NOALLOC_BITS() 1
+
+/** Does the chip support hash-for-home caching? */
+#define CHIP_HAS_CBOX_HOME_MAP() 1
+
+/** Number of entries in the chip's home map tables. */
+#define CHIP_CBOX_HOME_MAP_SIZE() 128
+
+/** Do uncacheable requests miss in the cache regardless of whether
+ * there is matching data? */
+#define CHIP_HAS_ENFORCED_UNCACHEABLE_REQUESTS() 1
+
+/** Does the mf instruction wait for victims? */
+#define CHIP_HAS_MF_WAITS_FOR_VICTIMS() 0
+
+/** Does the chip have an "inv" instruction that doesn't also flush? */
+#define CHIP_HAS_INV() 1
+
+/** Does the chip have a "wh64" instruction? */
+#define CHIP_HAS_WH64() 1
+
+/** Does this chip have a 'dword_align' instruction? */
+#define CHIP_HAS_DWORD_ALIGN() 0
+
+/** Number of performance counters. */
+#define CHIP_PERFORMANCE_COUNTERS() 4
+
+/** Does this chip have auxiliary performance counters? */
+#define CHIP_HAS_AUX_PERF_COUNTERS() 1
+
+/** Is the CBOX_MSR1 SPR supported? */
+#define CHIP_HAS_CBOX_MSR1() 0
+
+/** Is the TILE_RTF_HWM SPR supported? */
+#define CHIP_HAS_TILE_RTF_HWM() 1
+
+/** Is the TILE_WRITE_PENDING SPR supported? */
+#define CHIP_HAS_TILE_WRITE_PENDING() 0
+
+/** Is the PROC_STATUS SPR supported? */
+#define CHIP_HAS_PROC_STATUS_SPR() 1
+
+/** Is the DSTREAM_PF SPR supported? */
+#define CHIP_HAS_DSTREAM_PF() 1
+
+/** Log of the number of mshims we have. */
+#define CHIP_LOG_NUM_MSHIMS() 2
+
+/** Are the bases of the interrupt vector areas fixed? */
+#define CHIP_HAS_FIXED_INTVEC_BASE() 0
+
+/** Are the interrupt masks split up into 2 SPRs? */
+#define CHIP_HAS_SPLIT_INTR_MASK() 0
+
+/** Is the cycle count split up into 2 SPRs? */
+#define CHIP_HAS_SPLIT_CYCLE() 0
+
+/** Does the chip have a static network? */
+#define CHIP_HAS_SN() 0
+
+/** Does the chip have a static network processor? */
+#define CHIP_HAS_SN_PROC() 0
+
+/** Size of the L1 static network processor instruction cache, in bytes. */
+/* #define CHIP_L1SNI_CACHE_SIZE() -- does not apply to chip 10 */
+
+/** Does the chip have DMA support in each tile? */
+#define CHIP_HAS_TILE_DMA() 0
+
+/** Does the chip have the second revision of the directly accessible
+ *  dynamic networks?  This encapsulates a number of characteristics,
+ *  including the absence of the catch-all, the absence of inline message
+ *  tags, the absence of support for network context-switching, and so on.
+ */
+#define CHIP_HAS_REV1_XDN() 1
+
+/** Does the chip have cmpexch and similar (fetchadd, exch, etc.)? */
+#define CHIP_HAS_CMPEXCH() 1
+
+/** Does the chip have memory-mapped I/O support? */
+#define CHIP_HAS_MMIO() 1
+
+/** Does the chip have post-completion interrupts? */
+#define CHIP_HAS_POST_COMPLETION_INTERRUPTS() 1
+
+/** Does the chip have native single step support? */
+#define CHIP_HAS_SINGLE_STEP() 1
+
+#ifndef __OPEN_SOURCE__  /* features only relevant to hypervisor-level code */
+
+/** How many entries are present in the instruction TLB? */
+#define CHIP_ITLB_ENTRIES() 16
+
+/** How many entries are present in the data TLB? */
+#define CHIP_DTLB_ENTRIES() 32
+
+/** How many MAF entries does the XAUI shim have? */
+#define CHIP_XAUI_MAF_ENTRIES() 32
+
+/** Does the memory shim have a source-id table? */
+#define CHIP_HAS_MSHIM_SRCID_TABLE() 0
+
+/** Does the L1 instruction cache clear on reset? */
+#define CHIP_HAS_L1I_CLEAR_ON_RESET() 1
+
+/** Does the chip come out of reset with valid coordinates on all tiles?
+ * Note that if defined, this also implies that the upper left is 1,1.
+ */
+#define CHIP_HAS_VALID_TILE_COORD_RESET() 1
+
+/** Does the chip have unified packet formats? */
+#define CHIP_HAS_UNIFIED_PACKET_FORMATS() 1
+
+/** Does the chip support write reordering? */
+#define CHIP_HAS_WRITE_REORDERING() 1
+
+/** Does the chip support Y-X routing as well as X-Y? */
+#define CHIP_HAS_Y_X_ROUTING() 1
+
+/** Is INTCTRL_3 managed with the correct MPL? */
+#define CHIP_HAS_INTCTRL_3_STATUS_FIX() 1
+
+/** Is it possible to configure the chip to be big-endian? */
+#define CHIP_HAS_BIG_ENDIAN_CONFIG() 1
+
+/** Is the CACHE_RED_WAY_OVERRIDDEN SPR supported? */
+#define CHIP_HAS_CACHE_RED_WAY_OVERRIDDEN() 0
+
+/** Is the DIAG_TRACE_WAY SPR supported? */
+#define CHIP_HAS_DIAG_TRACE_WAY() 0
+
+/** Is the MEM_STRIPE_CONFIG SPR supported? */
+#define CHIP_HAS_MEM_STRIPE_CONFIG() 1
+
+/** Are the TLB_PERF SPRs supported? */
+#define CHIP_HAS_TLB_PERF() 1
+
+/** Is the VDN_SNOOP_SHIM_CTL SPR supported? */
+#define CHIP_HAS_VDN_SNOOP_SHIM_CTL() 0
+
+/** Does the chip support rev1 DMA packets? */
+#define CHIP_HAS_REV1_DMA_PACKETS() 1
+
+/** Does the chip have an IPI shim? */
+#define CHIP_HAS_IPI() 1
+
+#endif /* !__OPEN_SOURCE__ */
+#endif /* __ARCH_CHIP_H__ */
diff --git a/arch/tile/include/arch/icache.h b/arch/tile/include/arch/icache.h
index 5c87c9016338..762eafa8a11e 100644
--- a/arch/tile/include/arch/icache.h
+++ b/arch/tile/include/arch/icache.h
@@ -16,7 +16,7 @@
 /**
  * @file
  *
- * Support for invalidating bytes in the instruction
+ * Support for invalidating bytes in the instruction cache.
  */
 
 #ifndef __ARCH_ICACHE_H__
@@ -30,11 +30,10 @@
  *
  * @param addr The start of memory to be invalidated.
  * @param size The number of bytes to be invalidated.
- * @param page_size The system's page size, typically the PAGE_SIZE constant
- * in sys/page.h.  This value must be a power of two no larger
- * than the page containing the code to be invalidated. If the value
- * is smaller than the actual page size, this function will still
- * work, but may run slower than necessary.
+ * @param page_size The system's page size, e.g. getpagesize() in userspace.
+ * This value must be a power of two no larger than the page containing
+ * the code to be invalidated. If the value is smaller than the actual page
+ * size, this function will still work, but may run slower than necessary.
  */
 static __inline void
 invalidate_icache(const void* addr, unsigned long size,
diff --git a/arch/tile/include/arch/interrupts_32.h b/arch/tile/include/arch/interrupts_32.h
index 9d0bfa7e59be..96b5710505b6 100644
--- a/arch/tile/include/arch/interrupts_32.h
+++ b/arch/tile/include/arch/interrupts_32.h
@@ -16,10 +16,11 @@
 #define __ARCH_INTERRUPTS_H__
 
 /** Mask for an interrupt. */
-#ifdef __ASSEMBLER__
 /* Note: must handle breaking interrupts into high and low words manually. */
-#define INT_MASK(intno) (1 << (intno))
-#else
+#define INT_MASK_LO(intno) (1 << (intno))
+#define INT_MASK_HI(intno) (1 << ((intno) - 32))
+
+#ifndef __ASSEMBLER__
 #define INT_MASK(intno) (1ULL << (intno))
 #endif
 
@@ -89,6 +90,7 @@
 
 #define NUM_INTERRUPTS 49
 
+#ifndef __ASSEMBLER__
 #define QUEUED_INTERRUPTS ( \
     INT_MASK(INT_MEM_ERROR) | \
     INT_MASK(INT_DMATLB_MISS) | \
@@ -301,4 +303,5 @@
     INT_MASK(INT_DOUBLE_FAULT) | \
     INT_MASK(INT_AUX_PERF_COUNT) | \
     0)
+#endif /* !__ASSEMBLER__ */
 #endif /* !__ARCH_INTERRUPTS_H__ */
diff --git a/arch/tile/include/arch/interrupts_64.h b/arch/tile/include/arch/interrupts_64.h
new file mode 100644
index 000000000000..5bb58b2e4e6f
--- /dev/null
+++ b/arch/tile/include/arch/interrupts_64.h
@@ -0,0 +1,276 @@
+/*
+ * Copyright 2011 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#ifndef __ARCH_INTERRUPTS_H__
+#define __ARCH_INTERRUPTS_H__
+
+/** Mask for an interrupt. */
+#ifdef __ASSEMBLER__
+/* Note: must handle breaking interrupts into high and low words manually. */
+#define INT_MASK(intno) (1 << (intno))
+#else
+#define INT_MASK(intno) (1ULL << (intno))
+#endif
+
+
+/** Where a given interrupt executes */
+#define INTERRUPT_VECTOR(i, pl) (0xFC000000 + ((pl) << 24) + ((i) << 8))
+
+/** Where to store a vector for a given interrupt. */
+#define USER_INTERRUPT_VECTOR(i) INTERRUPT_VECTOR(i, 0)
+
+/** The base address of user-level interrupts. */
+#define USER_INTERRUPT_VECTOR_BASE INTERRUPT_VECTOR(0, 0)
+
+
+/** Additional synthetic interrupt. */
+#define INT_BREAKPOINT (63)
+
+#define INT_MEM_ERROR    0
+#define INT_SINGLE_STEP_3    1
+#define INT_SINGLE_STEP_2    2
+#define INT_SINGLE_STEP_1    3
+#define INT_SINGLE_STEP_0    4
+#define INT_IDN_COMPLETE    5
+#define INT_UDN_COMPLETE    6
+#define INT_ITLB_MISS    7
+#define INT_ILL    8
+#define INT_GPV    9
+#define INT_IDN_ACCESS   10
+#define INT_UDN_ACCESS   11
+#define INT_SWINT_3   12
+#define INT_SWINT_2   13
+#define INT_SWINT_1   14
+#define INT_SWINT_0   15
+#define INT_ILL_TRANS   16
+#define INT_UNALIGN_DATA   17
+#define INT_DTLB_MISS   18
+#define INT_DTLB_ACCESS   19
+#define INT_IDN_FIREWALL   20
+#define INT_UDN_FIREWALL   21
+#define INT_TILE_TIMER   22
+#define INT_AUX_TILE_TIMER   23
+#define INT_IDN_TIMER   24
+#define INT_UDN_TIMER   25
+#define INT_IDN_AVAIL   26
+#define INT_UDN_AVAIL   27
+#define INT_IPI_3   28
+#define INT_IPI_2   29
+#define INT_IPI_1   30
+#define INT_IPI_0   31
+#define INT_PERF_COUNT   32
+#define INT_AUX_PERF_COUNT   33
+#define INT_INTCTRL_3   34
+#define INT_INTCTRL_2   35
+#define INT_INTCTRL_1   36
+#define INT_INTCTRL_0   37
+#define INT_BOOT_ACCESS   38
+#define INT_WORLD_ACCESS   39
+#define INT_I_ASID   40
+#define INT_D_ASID   41
+#define INT_DOUBLE_FAULT   42
+
+#define NUM_INTERRUPTS 43
+
+#ifndef __ASSEMBLER__
+#define QUEUED_INTERRUPTS ( \
+    INT_MASK(INT_MEM_ERROR) | \
+    INT_MASK(INT_IDN_COMPLETE) | \
+    INT_MASK(INT_UDN_COMPLETE) | \
+    INT_MASK(INT_IDN_FIREWALL) | \
+    INT_MASK(INT_UDN_FIREWALL) | \
+    INT_MASK(INT_TILE_TIMER) | \
+    INT_MASK(INT_AUX_TILE_TIMER) | \
+    INT_MASK(INT_IDN_TIMER) | \
+    INT_MASK(INT_UDN_TIMER) | \
+    INT_MASK(INT_IDN_AVAIL) | \
+    INT_MASK(INT_UDN_AVAIL) | \
+    INT_MASK(INT_IPI_3) | \
+    INT_MASK(INT_IPI_2) | \
+    INT_MASK(INT_IPI_1) | \
+    INT_MASK(INT_IPI_0) | \
+    INT_MASK(INT_PERF_COUNT) | \
+    INT_MASK(INT_AUX_PERF_COUNT) | \
+    INT_MASK(INT_INTCTRL_3) | \
+    INT_MASK(INT_INTCTRL_2) | \
+    INT_MASK(INT_INTCTRL_1) | \
+    INT_MASK(INT_INTCTRL_0) | \
+    INT_MASK(INT_BOOT_ACCESS) | \
+    INT_MASK(INT_WORLD_ACCESS) | \
+    INT_MASK(INT_I_ASID) | \
+    INT_MASK(INT_D_ASID) | \
+    INT_MASK(INT_DOUBLE_FAULT) | \
+    0)
+#define NONQUEUED_INTERRUPTS ( \
+    INT_MASK(INT_SINGLE_STEP_3) | \
+    INT_MASK(INT_SINGLE_STEP_2) | \
+    INT_MASK(INT_SINGLE_STEP_1) | \
+    INT_MASK(INT_SINGLE_STEP_0) | \
+    INT_MASK(INT_ITLB_MISS) | \
+    INT_MASK(INT_ILL) | \
+    INT_MASK(INT_GPV) | \
+    INT_MASK(INT_IDN_ACCESS) | \
+    INT_MASK(INT_UDN_ACCESS) | \
+    INT_MASK(INT_SWINT_3) | \
+    INT_MASK(INT_SWINT_2) | \
+    INT_MASK(INT_SWINT_1) | \
+    INT_MASK(INT_SWINT_0) | \
+    INT_MASK(INT_ILL_TRANS) | \
+    INT_MASK(INT_UNALIGN_DATA) | \
+    INT_MASK(INT_DTLB_MISS) | \
+    INT_MASK(INT_DTLB_ACCESS) | \
+    0)
+#define CRITICAL_MASKED_INTERRUPTS ( \
+    INT_MASK(INT_MEM_ERROR) | \
+    INT_MASK(INT_SINGLE_STEP_3) | \
+    INT_MASK(INT_SINGLE_STEP_2) | \
+    INT_MASK(INT_SINGLE_STEP_1) | \
+    INT_MASK(INT_SINGLE_STEP_0) | \
+    INT_MASK(INT_IDN_COMPLETE) | \
+    INT_MASK(INT_UDN_COMPLETE) | \
+    INT_MASK(INT_IDN_FIREWALL) | \
+    INT_MASK(INT_UDN_FIREWALL) | \
+    INT_MASK(INT_TILE_TIMER) | \
+    INT_MASK(INT_AUX_TILE_TIMER) | \
+    INT_MASK(INT_IDN_TIMER) | \
+    INT_MASK(INT_UDN_TIMER) | \
+    INT_MASK(INT_IDN_AVAIL) | \
+    INT_MASK(INT_UDN_AVAIL) | \
+    INT_MASK(INT_IPI_3) | \
+    INT_MASK(INT_IPI_2) | \
+    INT_MASK(INT_IPI_1) | \
+    INT_MASK(INT_IPI_0) | \
+    INT_MASK(INT_PERF_COUNT) | \
+    INT_MASK(INT_AUX_PERF_COUNT) | \
+    INT_MASK(INT_INTCTRL_3) | \
+    INT_MASK(INT_INTCTRL_2) | \
+    INT_MASK(INT_INTCTRL_1) | \
+    INT_MASK(INT_INTCTRL_0) | \
+    0)
+#define CRITICAL_UNMASKED_INTERRUPTS ( \
+    INT_MASK(INT_ITLB_MISS) | \
+    INT_MASK(INT_ILL) | \
+    INT_MASK(INT_GPV) | \
+    INT_MASK(INT_IDN_ACCESS) | \
+    INT_MASK(INT_UDN_ACCESS) | \
+    INT_MASK(INT_SWINT_3) | \
+    INT_MASK(INT_SWINT_2) | \
+    INT_MASK(INT_SWINT_1) | \
+    INT_MASK(INT_SWINT_0) | \
+    INT_MASK(INT_ILL_TRANS) | \
+    INT_MASK(INT_UNALIGN_DATA) | \
+    INT_MASK(INT_DTLB_MISS) | \
+    INT_MASK(INT_DTLB_ACCESS) | \
+    INT_MASK(INT_BOOT_ACCESS) | \
+    INT_MASK(INT_WORLD_ACCESS) | \
+    INT_MASK(INT_I_ASID) | \
+    INT_MASK(INT_D_ASID) | \
+    INT_MASK(INT_DOUBLE_FAULT) | \
+    0)
+#define MASKABLE_INTERRUPTS ( \
+    INT_MASK(INT_MEM_ERROR) | \
+    INT_MASK(INT_SINGLE_STEP_3) | \
+    INT_MASK(INT_SINGLE_STEP_2) | \
+    INT_MASK(INT_SINGLE_STEP_1) | \
+    INT_MASK(INT_SINGLE_STEP_0) | \
+    INT_MASK(INT_IDN_COMPLETE) | \
+    INT_MASK(INT_UDN_COMPLETE) | \
+    INT_MASK(INT_IDN_FIREWALL) | \
+    INT_MASK(INT_UDN_FIREWALL) | \
+    INT_MASK(INT_TILE_TIMER) | \
+    INT_MASK(INT_AUX_TILE_TIMER) | \
+    INT_MASK(INT_IDN_TIMER) | \
+    INT_MASK(INT_UDN_TIMER) | \
+    INT_MASK(INT_IDN_AVAIL) | \
+    INT_MASK(INT_UDN_AVAIL) | \
+    INT_MASK(INT_IPI_3) | \
+    INT_MASK(INT_IPI_2) | \
+    INT_MASK(INT_IPI_1) | \
+    INT_MASK(INT_IPI_0) | \
+    INT_MASK(INT_PERF_COUNT) | \
+    INT_MASK(INT_AUX_PERF_COUNT) | \
+    INT_MASK(INT_INTCTRL_3) | \
+    INT_MASK(INT_INTCTRL_2) | \
+    INT_MASK(INT_INTCTRL_1) | \
+    INT_MASK(INT_INTCTRL_0) | \
+    0)
+#define UNMASKABLE_INTERRUPTS ( \
+    INT_MASK(INT_ITLB_MISS) | \
+    INT_MASK(INT_ILL) | \
+    INT_MASK(INT_GPV) | \
+    INT_MASK(INT_IDN_ACCESS) | \
+    INT_MASK(INT_UDN_ACCESS) | \
+    INT_MASK(INT_SWINT_3) | \
+    INT_MASK(INT_SWINT_2) | \
+    INT_MASK(INT_SWINT_1) | \
+    INT_MASK(INT_SWINT_0) | \
+    INT_MASK(INT_ILL_TRANS) | \
+    INT_MASK(INT_UNALIGN_DATA) | \
+    INT_MASK(INT_DTLB_MISS) | \
+    INT_MASK(INT_DTLB_ACCESS) | \
+    INT_MASK(INT_BOOT_ACCESS) | \
+    INT_MASK(INT_WORLD_ACCESS) | \
+    INT_MASK(INT_I_ASID) | \
+    INT_MASK(INT_D_ASID) | \
+    INT_MASK(INT_DOUBLE_FAULT) | \
+    0)
+#define SYNC_INTERRUPTS ( \
+    INT_MASK(INT_SINGLE_STEP_3) | \
+    INT_MASK(INT_SINGLE_STEP_2) | \
+    INT_MASK(INT_SINGLE_STEP_1) | \
+    INT_MASK(INT_SINGLE_STEP_0) | \
+    INT_MASK(INT_IDN_COMPLETE) | \
+    INT_MASK(INT_UDN_COMPLETE) | \
+    INT_MASK(INT_ITLB_MISS) | \
+    INT_MASK(INT_ILL) | \
+    INT_MASK(INT_GPV) | \
+    INT_MASK(INT_IDN_ACCESS) | \
+    INT_MASK(INT_UDN_ACCESS) | \
+    INT_MASK(INT_SWINT_3) | \
+    INT_MASK(INT_SWINT_2) | \
+    INT_MASK(INT_SWINT_1) | \
+    INT_MASK(INT_SWINT_0) | \
+    INT_MASK(INT_ILL_TRANS) | \
+    INT_MASK(INT_UNALIGN_DATA) | \
+    INT_MASK(INT_DTLB_MISS) | \
+    INT_MASK(INT_DTLB_ACCESS) | \
+    0)
+#define NON_SYNC_INTERRUPTS ( \
+    INT_MASK(INT_MEM_ERROR) | \
+    INT_MASK(INT_IDN_FIREWALL) | \
+    INT_MASK(INT_UDN_FIREWALL) | \
+    INT_MASK(INT_TILE_TIMER) | \
+    INT_MASK(INT_AUX_TILE_TIMER) | \
+    INT_MASK(INT_IDN_TIMER) | \
+    INT_MASK(INT_UDN_TIMER) | \
+    INT_MASK(INT_IDN_AVAIL) | \
+    INT_MASK(INT_UDN_AVAIL) | \
+    INT_MASK(INT_IPI_3) | \
+    INT_MASK(INT_IPI_2) | \
+    INT_MASK(INT_IPI_1) | \
+    INT_MASK(INT_IPI_0) | \
+    INT_MASK(INT_PERF_COUNT) | \
+    INT_MASK(INT_AUX_PERF_COUNT) | \
+    INT_MASK(INT_INTCTRL_3) | \
+    INT_MASK(INT_INTCTRL_2) | \
+    INT_MASK(INT_INTCTRL_1) | \
+    INT_MASK(INT_INTCTRL_0) | \
+    INT_MASK(INT_BOOT_ACCESS) | \
+    INT_MASK(INT_WORLD_ACCESS) | \
+    INT_MASK(INT_I_ASID) | \
+    INT_MASK(INT_D_ASID) | \
+    INT_MASK(INT_DOUBLE_FAULT) | \
+    0)
+#endif /* !__ASSEMBLER__ */
+#endif /* !__ARCH_INTERRUPTS_H__ */
diff --git a/arch/tile/include/arch/sim.h b/arch/tile/include/arch/sim.h
new file mode 100644
index 000000000000..e54b7b0527f3
--- /dev/null
+++ b/arch/tile/include/arch/sim.h
@@ -0,0 +1,643 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+/**
+ * @file
+ *
+ * Provides an API for controlling the simulator at runtime.
+ */
+
+/**
+ * @addtogroup arch_sim
+ * @{
+ *
+ * An API for controlling the simulator at runtime.
+ *
+ * The simulator's behavior can be modified while it is running.
+ * For example, human-readable trace output can be enabled and disabled
+ * around code of interest.
+ *
+ * There are two ways to modify simulator behavior:
+ * programmatically, by calling various sim_* functions, and
+ * interactively, by entering commands like "sim set functional true"
+ * at the tile-monitor prompt.  Typing "sim help" at that prompt provides
+ * a list of interactive commands.
+ *
+ * All interactive commands can also be executed programmatically by
+ * passing a string to the sim_command function.
+ */
+
+#ifndef __ARCH_SIM_H__
+#define __ARCH_SIM_H__
+
+#include <arch/sim_def.h>
+#include <arch/abi.h>
+
+#ifndef __ASSEMBLER__
+
+#include <arch/spr_def.h>
+
+
+/**
+ * Return true if the current program is running under a simulator,
+ * rather than on real hardware.  If running on hardware, other "sim_xxx()"
+ * calls have no useful effect.
+ */
+static inline int
+sim_is_simulator(void)
+{
+  return __insn_mfspr(SPR_SIM_CONTROL) != 0;
+}
+
+
+/**
+ * Checkpoint the simulator state to a checkpoint file.
+ *
+ * The checkpoint file name is either the default or the name specified
+ * on the command line with "--checkpoint-file".
+ */
+static __inline void
+sim_checkpoint(void)
+{
+  __insn_mtspr(SPR_SIM_CONTROL, SIM_CONTROL_CHECKPOINT);
+}
+
+
+/**
+ * Report whether or not various kinds of simulator tracing are enabled.
+ *
+ * @return The bitwise OR of these values:
+ *
+ * SIM_TRACE_CYCLES (--trace-cycles),
+ * SIM_TRACE_ROUTER (--trace-router),
+ * SIM_TRACE_REGISTER_WRITES (--trace-register-writes),
+ * SIM_TRACE_DISASM (--trace-disasm),
+ * SIM_TRACE_STALL_INFO (--trace-stall-info)
+ * SIM_TRACE_MEMORY_CONTROLLER (--trace-memory-controller)
+ * SIM_TRACE_L2_CACHE (--trace-l2)
+ * SIM_TRACE_LINES (--trace-lines)
+ */
+static __inline unsigned int
+sim_get_tracing(void)
+{
+  return __insn_mfspr(SPR_SIM_CONTROL) & SIM_TRACE_FLAG_MASK;
+}
+
+
+/**
+ * Turn on or off different kinds of simulator tracing.
+ *
+ * @param mask Either one of these special values:
+ *
+ * SIM_TRACE_NONE (turns off tracing),
+ * SIM_TRACE_ALL (turns on all possible tracing).
+ *
+ * or the bitwise OR of these values:
+ *
+ * SIM_TRACE_CYCLES (--trace-cycles),
+ * SIM_TRACE_ROUTER (--trace-router),
+ * SIM_TRACE_REGISTER_WRITES (--trace-register-writes),
+ * SIM_TRACE_DISASM (--trace-disasm),
+ * SIM_TRACE_STALL_INFO (--trace-stall-info)
+ * SIM_TRACE_MEMORY_CONTROLLER (--trace-memory-controller)
+ * SIM_TRACE_L2_CACHE (--trace-l2)
+ * SIM_TRACE_LINES (--trace-lines)
+ */
+static __inline void
+sim_set_tracing(unsigned int mask)
+{
+  __insn_mtspr(SPR_SIM_CONTROL, SIM_TRACE_SPR_ARG(mask));
+}
+
+
+/**
+ * Request dumping of different kinds of simulator state.
+ *
+ * @param mask Either this special value:
+ *
+ * SIM_DUMP_ALL (dump all known state)
+ *
+ * or the bitwise OR of these values:
+ *
+ * SIM_DUMP_REGS (the register file),
+ * SIM_DUMP_SPRS (the SPRs),
+ * SIM_DUMP_ITLB (the iTLB),
+ * SIM_DUMP_DTLB (the dTLB),
+ * SIM_DUMP_L1I (the L1 I-cache),
+ * SIM_DUMP_L1D (the L1 D-cache),
+ * SIM_DUMP_L2 (the L2 cache),
+ * SIM_DUMP_SNREGS (the switch register file),
+ * SIM_DUMP_SNITLB (the switch iTLB),
+ * SIM_DUMP_SNL1I (the switch L1 I-cache),
+ * SIM_DUMP_BACKTRACE (the current backtrace)
+ */
+static __inline void
+sim_dump(unsigned int mask)
+{
+  __insn_mtspr(SPR_SIM_CONTROL, SIM_DUMP_SPR_ARG(mask));
+}
+
+
+/**
+ * Print a string to the simulator stdout.
+ *
+ * @param str The string to be written.
+ */
+static __inline void
+sim_print(const char* str)
+{
+  for ( ; *str != '\0'; str++)
+  {
+    __insn_mtspr(SPR_SIM_CONTROL, SIM_CONTROL_PUTC |
+                 (*str << _SIM_CONTROL_OPERATOR_BITS));
+  }
+  __insn_mtspr(SPR_SIM_CONTROL, SIM_CONTROL_PUTC |
+               (SIM_PUTC_FLUSH_BINARY << _SIM_CONTROL_OPERATOR_BITS));
+}
+
+
+/**
+ * Print a string to the simulator stdout.
+ *
+ * @param str The string to be written (a newline is automatically added).
+ */
+static __inline void
+sim_print_string(const char* str)
+{
+  for ( ; *str != '\0'; str++)
+  {
+    __insn_mtspr(SPR_SIM_CONTROL, SIM_CONTROL_PUTC |
+                 (*str << _SIM_CONTROL_OPERATOR_BITS));
+  }
+  __insn_mtspr(SPR_SIM_CONTROL, SIM_CONTROL_PUTC |
+               (SIM_PUTC_FLUSH_STRING << _SIM_CONTROL_OPERATOR_BITS));
+}
+
+
+/**
+ * Execute a simulator command string.
+ *
+ * Type 'sim help' at the tile-monitor prompt to learn what commands
+ * are available.  Note the use of the tile-monitor "sim" command to
+ * pass commands to the simulator.
+ *
+ * The argument to sim_command() does not include the leading "sim"
+ * prefix used at the tile-monitor prompt; for example, you might call
+ * sim_command("trace disasm").
+ */
+static __inline void
+sim_command(const char* str)
+{
+  int c;
+  do
+  {
+    c = *str++;
+    __insn_mtspr(SPR_SIM_CONTROL, SIM_CONTROL_COMMAND |
+                 (c << _SIM_CONTROL_OPERATOR_BITS));
+  }
+  while (c);
+}
+
+
+
+#ifndef __DOXYGEN__
+
+/**
+ * The underlying implementation of "_sim_syscall()".
+ *
+ * We use extra "and" instructions to ensure that all the values
+ * we are passing to the simulator are actually valid in the registers
+ * (i.e. returned from memory) prior to the SIM_CONTROL spr.
+ */
+static __inline long _sim_syscall0(int val)
+{
+  long result;
+  __asm__ __volatile__ ("mtspr SIM_CONTROL, r0"
+                        : "=R00" (result) : "R00" (val));
+  return result;
+}
+
+static __inline long _sim_syscall1(int val, long arg1)
+{
+  long result;
+  __asm__ __volatile__ ("{ and zero, r1, r1; mtspr SIM_CONTROL, r0 }"
+                        : "=R00" (result) : "R00" (val), "R01" (arg1));
+  return result;
+}
+
+static __inline long _sim_syscall2(int val, long arg1, long arg2)
+{
+  long result;
+  __asm__ __volatile__ ("{ and zero, r1, r2; mtspr SIM_CONTROL, r0 }"
+                        : "=R00" (result)
+                        : "R00" (val), "R01" (arg1), "R02" (arg2));
+  return result;
+}
+
+/* Note that _sim_syscall3() and higher are technically at risk of
+   receiving an interrupt right before the mtspr bundle, in which case
+   the register values for arguments 3 and up may still be in flight
+   to the core from a stack frame reload. */
+
+static __inline long _sim_syscall3(int val, long arg1, long arg2, long arg3)
+{
+  long result;
+  __asm__ __volatile__ ("{ and zero, r3, r3 };"
+                        "{ and zero, r1, r2; mtspr SIM_CONTROL, r0 }"
+                        : "=R00" (result)
+                        : "R00" (val), "R01" (arg1), "R02" (arg2),
+                          "R03" (arg3));
+  return result;
+}
+
+static __inline long _sim_syscall4(int val, long arg1, long arg2, long arg3,
+                                  long arg4)
+{
+  long result;
+  __asm__ __volatile__ ("{ and zero, r3, r4 };"
+                        "{ and zero, r1, r2; mtspr SIM_CONTROL, r0 }"
+                        : "=R00" (result)
+                        : "R00" (val), "R01" (arg1), "R02" (arg2),
+                          "R03" (arg3), "R04" (arg4));
+  return result;
+}
+
+static __inline long _sim_syscall5(int val, long arg1, long arg2, long arg3,
+                                  long arg4, long arg5)
+{
+  long result;
+  __asm__ __volatile__ ("{ and zero, r3, r4; and zero, r5, r5 };"
+                        "{ and zero, r1, r2; mtspr SIM_CONTROL, r0 }"
+                        : "=R00" (result)
+                        : "R00" (val), "R01" (arg1), "R02" (arg2),
+                          "R03" (arg3), "R04" (arg4), "R05" (arg5));
+  return result;
+}
+
+/**
+ * Make a special syscall to the simulator itself, if running under
+ * simulation. This is used as the implementation of other functions
+ * and should not be used outside this file.
+ *
+ * @param syscall_num The simulator syscall number.
+ * @param nr The number of additional arguments provided.
+ *
+ * @return Varies by syscall.
+ */
+#define _sim_syscall(syscall_num, nr, args...) \
+  _sim_syscall##nr( \
+    ((syscall_num) << _SIM_CONTROL_OPERATOR_BITS) | SIM_CONTROL_SYSCALL, \
+    ##args)
+
+
+/* Values for the "access_mask" parameters below. */
+#define SIM_WATCHPOINT_READ    1
+#define SIM_WATCHPOINT_WRITE   2
+#define SIM_WATCHPOINT_EXECUTE 4
+
+
+static __inline int
+sim_add_watchpoint(unsigned int process_id,
+                   unsigned long address,
+                   unsigned long size,
+                   unsigned int access_mask,
+                   unsigned long user_data)
+{
+  return _sim_syscall(SIM_SYSCALL_ADD_WATCHPOINT, 5, process_id,
+                     address, size, access_mask, user_data);
+}
+
+
+static __inline int
+sim_remove_watchpoint(unsigned int process_id,
+                      unsigned long address,
+                      unsigned long size,
+                      unsigned int access_mask,
+                      unsigned long user_data)
+{
+  return _sim_syscall(SIM_SYSCALL_REMOVE_WATCHPOINT, 5, process_id,
+                     address, size, access_mask, user_data);
+}
+
+
+/**
+ * Return value from sim_query_watchpoint.
+ */
+struct SimQueryWatchpointStatus
+{
+  /**
+   * 0 if a watchpoint fired, 1 if no watchpoint fired, or -1 for
+   * error (meaning a bad process_id).
+   */
+  int syscall_status;
+
+  /**
+   * The address of the watchpoint that fired (this is the address
+   * passed to sim_add_watchpoint, not an address within that range
+   * that actually triggered the watchpoint).
+   */
+  unsigned long address;
+
+  /** The arbitrary user_data installed by sim_add_watchpoint. */
+  unsigned long user_data;
+};
+
+
+static __inline struct SimQueryWatchpointStatus
+sim_query_watchpoint(unsigned int process_id)
+{
+  struct SimQueryWatchpointStatus status;
+  long val = SIM_CONTROL_SYSCALL |
+    (SIM_SYSCALL_QUERY_WATCHPOINT << _SIM_CONTROL_OPERATOR_BITS);
+  __asm__ __volatile__ ("{ and zero, r1, r1; mtspr SIM_CONTROL, r0 }"
+                        : "=R00" (status.syscall_status),
+                          "=R01" (status.address),
+                          "=R02" (status.user_data)
+                        : "R00" (val), "R01" (process_id));
+  return status;
+}
+
+
+/* On the simulator, confirm lines have been evicted everywhere. */
+static __inline void
+sim_validate_lines_evicted(unsigned long long pa, unsigned long length)
+{
+#ifdef __LP64__
+  _sim_syscall(SIM_SYSCALL_VALIDATE_LINES_EVICTED, 2, pa, length);
+#else
+  _sim_syscall(SIM_SYSCALL_VALIDATE_LINES_EVICTED, 4,
+               0 /* dummy */, (long)(pa), (long)(pa >> 32), length);
+#endif
+}
+
+
+/* Return the current CPU speed in cycles per second. */
+static __inline long
+sim_query_cpu_speed(void)
+{
+  return _sim_syscall(SIM_SYSCALL_QUERY_CPU_SPEED, 0);
+}
+
+#endif /* !__DOXYGEN__ */
+
+
+
+
+/**
+ * Modify the shaping parameters of a shim.
+ *
+ * @param shim The shim to modify. One of:
+ *   SIM_CONTROL_SHAPING_GBE_0
+ *   SIM_CONTROL_SHAPING_GBE_1
+ *   SIM_CONTROL_SHAPING_GBE_2
+ *   SIM_CONTROL_SHAPING_GBE_3
+ *   SIM_CONTROL_SHAPING_XGBE_0
+ *   SIM_CONTROL_SHAPING_XGBE_1
+ *
+ * @param type The type of shaping. This should be the same type of
+ * shaping that is already in place on the shim. One of:
+ *   SIM_CONTROL_SHAPING_MULTIPLIER
+ *   SIM_CONTROL_SHAPING_PPS
+ *   SIM_CONTROL_SHAPING_BPS
+ *
+ * @param units The magnitude of the rate. One of:
+ *   SIM_CONTROL_SHAPING_UNITS_SINGLE
+ *   SIM_CONTROL_SHAPING_UNITS_KILO
+ *   SIM_CONTROL_SHAPING_UNITS_MEGA
+ *   SIM_CONTROL_SHAPING_UNITS_GIGA
+ *
+ * @param rate The rate to which to change it. This must fit in
+ * SIM_CONTROL_SHAPING_RATE_BITS bits or a warning is issued and
+ * the shaping is not changed.
+ *
+ * @return 0 if no problems were detected in the arguments to sim_set_shaping
+ * or 1 if problems were detected (for example, rate does not fit in 17 bits).
+ */
+static __inline int
+sim_set_shaping(unsigned shim,
+                unsigned type,
+                unsigned units,
+                unsigned rate)
+{
+  if ((rate & ~((1 << SIM_CONTROL_SHAPING_RATE_BITS) - 1)) != 0)
+    return 1;
+
+  __insn_mtspr(SPR_SIM_CONTROL, SIM_SHAPING_SPR_ARG(shim, type, units, rate));
+  return 0;
+}
+
+#ifdef __tilegx__
+
+/** Enable a set of mPIPE links.  Pass a -1 link_mask to enable all links. */
+static __inline void
+sim_enable_mpipe_links(unsigned mpipe, unsigned long link_mask)
+{
+  __insn_mtspr(SPR_SIM_CONTROL,
+               (SIM_CONTROL_ENABLE_MPIPE_LINK_MAGIC_BYTE |
+                (mpipe << 8) | (1 << 16) | ((uint_reg_t)link_mask << 32)));
+}
+
+/** Disable a set of mPIPE links.  Pass a -1 link_mask to disable all links. */
+static __inline void
+sim_disable_mpipe_links(unsigned mpipe, unsigned long link_mask)
+{
+  __insn_mtspr(SPR_SIM_CONTROL,
+               (SIM_CONTROL_ENABLE_MPIPE_LINK_MAGIC_BYTE |
+                (mpipe << 8) | (0 << 16) | ((uint_reg_t)link_mask << 32)));
+}
+
+#endif /* __tilegx__ */
+
+
+/*
+ * An API for changing "functional" mode.
+ */
+
+#ifndef __DOXYGEN__
+
+#define sim_enable_functional() \
+  __insn_mtspr(SPR_SIM_CONTROL, SIM_CONTROL_ENABLE_FUNCTIONAL)
+
+#define sim_disable_functional() \
+  __insn_mtspr(SPR_SIM_CONTROL, SIM_CONTROL_DISABLE_FUNCTIONAL)
+
+#endif /* __DOXYGEN__ */
+
+
+/*
+ * Profiler support.
+ */
+
+/**
+ * Turn profiling on for the current task.
+ *
+ * Note that this has no effect if run in an environment without
+ * profiling support (thus, the proper flags to the simulator must
+ * be supplied).
+ */
+static __inline void
+sim_profiler_enable(void)
+{
+  __insn_mtspr(SPR_SIM_CONTROL, SIM_CONTROL_PROFILER_ENABLE);
+}
+
+
+/** Turn profiling off for the current task. */
+static __inline void
+sim_profiler_disable(void)
+{
+  __insn_mtspr(SPR_SIM_CONTROL, SIM_CONTROL_PROFILER_DISABLE);
+}
+
+
+/**
+ * Turn profiling on or off for the current task.
+ *
+ * @param enabled If true, turns on profiling. If false, turns it off.
+ *
+ * Note that this has no effect if run in an environment without
+ * profiling support (thus, the proper flags to the simulator must
+ * be supplied).
+ */
+static __inline void
+sim_profiler_set_enabled(int enabled)
+{
+  int val =
+    enabled ? SIM_CONTROL_PROFILER_ENABLE : SIM_CONTROL_PROFILER_DISABLE;
+  __insn_mtspr(SPR_SIM_CONTROL, val);
+}
+
+
+/**
+ * Return true if and only if profiling is currently enabled
+ * for the current task.
+ *
+ * This returns false even if sim_profiler_enable() was called
+ * if the current execution environment does not support profiling.
+ */
+static __inline int
+sim_profiler_is_enabled(void)
+{
+  return ((__insn_mfspr(SPR_SIM_CONTROL) & SIM_PROFILER_ENABLED_MASK) != 0);
+}
+
+
+/**
+ * Reset profiling counters to zero for the current task.
+ *
+ * Resetting can be done while profiling is enabled.  It does not affect
+ * the chip-wide profiling counters.
+ */
+static __inline void
+sim_profiler_clear(void)
+{
+  __insn_mtspr(SPR_SIM_CONTROL, SIM_CONTROL_PROFILER_CLEAR);
+}
+
+
+/**
+ * Enable specified chip-level profiling counters.
+ *
+ * Does not affect the per-task profiling counters.
+ *
+ * @param mask Either this special value:
+ *
+ * SIM_CHIP_ALL (enables all chip-level components).
+ *
+ * or the bitwise OR of these values:
+ *
+ * SIM_CHIP_MEMCTL (enable all memory controllers)
+ * SIM_CHIP_XAUI (enable all XAUI controllers)
+ * SIM_CHIP_MPIPE (enable all MPIPE controllers)
+ */
+static __inline void
+sim_profiler_chip_enable(unsigned int mask)
+{
+  __insn_mtspr(SPR_SIM_CONTROL, SIM_PROFILER_CHIP_ENABLE_SPR_ARG(mask));
+}
+
+
+/**
+ * Disable specified chip-level profiling counters.
+ *
+ * Does not affect the per-task profiling counters.
+ *
+ * @param mask Either this special value:
+ *
+ * SIM_CHIP_ALL (disables all chip-level components).
+ *
+ * or the bitwise OR of these values:
+ *
+ * SIM_CHIP_MEMCTL (disable all memory controllers)
+ * SIM_CHIP_XAUI (disable all XAUI controllers)
+ * SIM_CHIP_MPIPE (disable all MPIPE controllers)
+ */
+static __inline void
+sim_profiler_chip_disable(unsigned int mask)
+{
+  __insn_mtspr(SPR_SIM_CONTROL, SIM_PROFILER_CHIP_DISABLE_SPR_ARG(mask));
+}
+
+
+/**
+ * Reset specified chip-level profiling counters to zero.
+ *
+ * Does not affect the per-task profiling counters.
+ *
+ * @param mask Either this special value:
+ *
+ * SIM_CHIP_ALL (clears all chip-level components).
+ *
+ * or the bitwise OR of these values:
+ *
+ * SIM_CHIP_MEMCTL (clear all memory controllers)
+ * SIM_CHIP_XAUI (clear all XAUI controllers)
+ * SIM_CHIP_MPIPE (clear all MPIPE controllers)
+ */
+static __inline void
+sim_profiler_chip_clear(unsigned int mask)
+{
+  __insn_mtspr(SPR_SIM_CONTROL, SIM_PROFILER_CHIP_CLEAR_SPR_ARG(mask));
+}
+
+
+/*
+ * Event support.
+ */
+
+#ifndef __DOXYGEN__
+
+static __inline void
+sim_event_begin(unsigned int x)
+{
+#if defined(__tile__) && !defined(__NO_EVENT_SPR__)
+  __insn_mtspr(SPR_EVENT_BEGIN, x);
+#endif
+}
+
+static __inline void
+sim_event_end(unsigned int x)
+{
+#if defined(__tile__) && !defined(__NO_EVENT_SPR__)
+  __insn_mtspr(SPR_EVENT_END, x);
+#endif
+}
+
+#endif /* !__DOXYGEN__ */
+
+#endif /* !__ASSEMBLER__ */
+
+#endif /* !__ARCH_SIM_H__ */
+
+/** @} */
diff --git a/arch/tile/include/arch/sim_def.h b/arch/tile/include/arch/sim_def.h
index 6418fbde063e..4b44a2b6a09a 100644
--- a/arch/tile/include/arch/sim_def.h
+++ b/arch/tile/include/arch/sim_def.h
@@ -1,477 +1,464 @@
-// Copyright 2010 Tilera Corporation. All Rights Reserved.
-//
-//   This program is free software; you can redistribute it and/or
-//   modify it under the terms of the GNU General Public License
-//   as published by the Free Software Foundation, version 2.
-//
-//   This program is distributed in the hope that it will be useful, but
-//   WITHOUT ANY WARRANTY; without even the implied warranty of
-//   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
-//   NON INFRINGEMENT.  See the GNU General Public License for
-//   more details.
-
-//! @file
-//!
-//! Some low-level simulator definitions.
-//!
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+/**
+ * @file
+ *
+ * Some low-level simulator definitions.
+ */
 
 #ifndef __ARCH_SIM_DEF_H__
 #define __ARCH_SIM_DEF_H__
 
 
-//! Internal: the low bits of the SIM_CONTROL_* SPR values specify
-//! the operation to perform, and the remaining bits are
-//! an operation-specific parameter (often unused).
-//!
+/**
+ * Internal: the low bits of the SIM_CONTROL_* SPR values specify
+ * the operation to perform, and the remaining bits are
+ * an operation-specific parameter (often unused).
+ */
 #define _SIM_CONTROL_OPERATOR_BITS 8
 
 
-//== Values which can be written to SPR_SIM_CONTROL.
+/*
+ * Values which can be written to SPR_SIM_CONTROL.
+ */
 
-//! If written to SPR_SIM_CONTROL, stops profiling.
-//!
+/** If written to SPR_SIM_CONTROL, stops profiling. */
 #define SIM_CONTROL_PROFILER_DISABLE 0
 
-//! If written to SPR_SIM_CONTROL, starts profiling.
-//!
+/** If written to SPR_SIM_CONTROL, starts profiling. */
 #define SIM_CONTROL_PROFILER_ENABLE 1
 
-//! If written to SPR_SIM_CONTROL, clears profiling counters.
-//!
+/** If written to SPR_SIM_CONTROL, clears profiling counters. */
 #define SIM_CONTROL_PROFILER_CLEAR 2
 
-//! If written to SPR_SIM_CONTROL, checkpoints the simulator.
-//!
+/** If written to SPR_SIM_CONTROL, checkpoints the simulator. */
 #define SIM_CONTROL_CHECKPOINT 3
 
-//! If written to SPR_SIM_CONTROL, combined with a mask (shifted by 8),
-//! sets the tracing mask to the given mask. See "sim_set_tracing()".
-//!
+/**
+ * If written to SPR_SIM_CONTROL, combined with a mask (shifted by 8),
+ * sets the tracing mask to the given mask. See "sim_set_tracing()".
+ */
 #define SIM_CONTROL_SET_TRACING 4
 
-//! If written to SPR_SIM_CONTROL, combined with a mask (shifted by 8),
-//! dumps the requested items of machine state to the log.
-//!
+/**
+ * If written to SPR_SIM_CONTROL, combined with a mask (shifted by 8),
+ * dumps the requested items of machine state to the log.
+ */
 #define SIM_CONTROL_DUMP 5
 
-//! If written to SPR_SIM_CONTROL, clears chip-level profiling counters.
-//!
+/** If written to SPR_SIM_CONTROL, clears chip-level profiling counters. */
 #define SIM_CONTROL_PROFILER_CHIP_CLEAR 6
 
-//! If written to SPR_SIM_CONTROL, disables chip-level profiling.
-//!
+/** If written to SPR_SIM_CONTROL, disables chip-level profiling. */
 #define SIM_CONTROL_PROFILER_CHIP_DISABLE 7
 
-//! If written to SPR_SIM_CONTROL, enables chip-level profiling.
-//!
+/** If written to SPR_SIM_CONTROL, enables chip-level profiling. */
 #define SIM_CONTROL_PROFILER_CHIP_ENABLE 8
 
-//! If written to SPR_SIM_CONTROL, enables chip-level functional mode
-//!
+/** If written to SPR_SIM_CONTROL, enables chip-level functional mode */
 #define SIM_CONTROL_ENABLE_FUNCTIONAL 9
 
-//! If written to SPR_SIM_CONTROL, disables chip-level functional mode.
-//!
+/** If written to SPR_SIM_CONTROL, disables chip-level functional mode. */
 #define SIM_CONTROL_DISABLE_FUNCTIONAL 10
 
-//! If written to SPR_SIM_CONTROL, enables chip-level functional mode.
-//! All tiles must perform this write for functional mode to be enabled.
-//! Ignored in naked boot mode unless --functional is specified.
-//! WARNING: Only the hypervisor startup code should use this!
-//!
+/**
+ * If written to SPR_SIM_CONTROL, enables chip-level functional mode.
+ * All tiles must perform this write for functional mode to be enabled.
+ * Ignored in naked boot mode unless --functional is specified.
+ * WARNING: Only the hypervisor startup code should use this!
+ */
 #define SIM_CONTROL_ENABLE_FUNCTIONAL_BARRIER 11
 
-//! If written to SPR_SIM_CONTROL, combined with a character (shifted by 8),
-//! writes a string directly to the simulator output.  Written to once for
-//! each character in the string, plus a final NUL.  Instead of NUL,
-//! you can also use "SIM_PUTC_FLUSH_STRING" or "SIM_PUTC_FLUSH_BINARY".
-//!
-// ISSUE: Document the meaning of "newline", and the handling of NUL.
-//
+/**
+ * If written to SPR_SIM_CONTROL, combined with a character (shifted by 8),
+ * writes a string directly to the simulator output.  Written to once for
+ * each character in the string, plus a final NUL.  Instead of NUL,
+ * you can also use "SIM_PUTC_FLUSH_STRING" or "SIM_PUTC_FLUSH_BINARY".
+ */
+/* ISSUE: Document the meaning of "newline", and the handling of NUL. */
 #define SIM_CONTROL_PUTC 12
 
-//! If written to SPR_SIM_CONTROL, clears the --grind-coherence state for
-//! this core.  This is intended to be used before a loop that will
-//! invalidate the cache by loading new data and evicting all current data.
-//! Generally speaking, this API should only be used by system code.
-//!
+/**
+ * If written to SPR_SIM_CONTROL, clears the --grind-coherence state for
+ * this core.  This is intended to be used before a loop that will
+ * invalidate the cache by loading new data and evicting all current data.
+ * Generally speaking, this API should only be used by system code.
+ */
 #define SIM_CONTROL_GRINDER_CLEAR 13
 
-//! If written to SPR_SIM_CONTROL, shuts down the simulator.
-//!
+/** If written to SPR_SIM_CONTROL, shuts down the simulator. */
 #define SIM_CONTROL_SHUTDOWN 14
 
-//! If written to SPR_SIM_CONTROL, combined with a pid (shifted by 8),
-//! indicates that a fork syscall just created the given process.
-//!
+/**
+ * If written to SPR_SIM_CONTROL, combined with a pid (shifted by 8),
+ * indicates that a fork syscall just created the given process.
+ */
 #define SIM_CONTROL_OS_FORK 15
 
-//! If written to SPR_SIM_CONTROL, combined with a pid (shifted by 8),
-//! indicates that an exit syscall was just executed by the given process.
-//!
+/**
+ * If written to SPR_SIM_CONTROL, combined with a pid (shifted by 8),
+ * indicates that an exit syscall was just executed by the given process.
+ */
 #define SIM_CONTROL_OS_EXIT 16
 
-//! If written to SPR_SIM_CONTROL, combined with a pid (shifted by 8),
-//! indicates that the OS just switched to the given process.
-//!
+/**
+ * If written to SPR_SIM_CONTROL, combined with a pid (shifted by 8),
+ * indicates that the OS just switched to the given process.
+ */
 #define SIM_CONTROL_OS_SWITCH 17
 
-//! If written to SPR_SIM_CONTROL, combined with a character (shifted by 8),
-//! indicates that an exec syscall was just executed. Written to once for
-//! each character in the executable name, plus a final NUL.
-//!
+/**
+ * If written to SPR_SIM_CONTROL, combined with a character (shifted by 8),
+ * indicates that an exec syscall was just executed. Written to once for
+ * each character in the executable name, plus a final NUL.
+ */
 #define SIM_CONTROL_OS_EXEC 18
 
-//! If written to SPR_SIM_CONTROL, combined with a character (shifted by 8),
-//! indicates that an interpreter (PT_INTERP) was loaded.  Written to once
-//! for each character in "ADDR:PATH", plus a final NUL, where "ADDR" is a
-//! hex load address starting with "0x", and "PATH" is the executable name.
-//!
+/**
+ * If written to SPR_SIM_CONTROL, combined with a character (shifted by 8),
+ * indicates that an interpreter (PT_INTERP) was loaded.  Written to once
+ * for each character in "ADDR:PATH", plus a final NUL, where "ADDR" is a
+ * hex load address starting with "0x", and "PATH" is the executable name.
+ */
 #define SIM_CONTROL_OS_INTERP 19
 
-//! If written to SPR_SIM_CONTROL, combined with a character (shifted by 8),
-//! indicates that a dll was loaded.  Written to once for each character
-//! in "ADDR:PATH", plus a final NUL, where "ADDR" is a hexadecimal load
-//! address starting with "0x", and "PATH" is the executable name.
-//!
+/**
+ * If written to SPR_SIM_CONTROL, combined with a character (shifted by 8),
+ * indicates that a dll was loaded.  Written to once for each character
+ * in "ADDR:PATH", plus a final NUL, where "ADDR" is a hexadecimal load
+ * address starting with "0x", and "PATH" is the executable name.
+ */
 #define SIM_CONTROL_DLOPEN 20
 
-//! If written to SPR_SIM_CONTROL, combined with a character (shifted by 8),
-//! indicates that a dll was unloaded.  Written to once for each character
-//! in "ADDR", plus a final NUL, where "ADDR" is a hexadecimal load
-//! address starting with "0x".
-//!
+/**
+ * If written to SPR_SIM_CONTROL, combined with a character (shifted by 8),
+ * indicates that a dll was unloaded.  Written to once for each character
+ * in "ADDR", plus a final NUL, where "ADDR" is a hexadecimal load
+ * address starting with "0x".
+ */
 #define SIM_CONTROL_DLCLOSE 21
 
-//! If written to SPR_SIM_CONTROL, combined with a flag (shifted by 8),
-//! indicates whether to allow data reads to remotely-cached
-//! dirty cache lines to be cached locally without grinder warnings or
-//! assertions (used by Linux kernel fast memcpy).
-//!
+/**
+ * If written to SPR_SIM_CONTROL, combined with a flag (shifted by 8),
+ * indicates whether to allow data reads to remotely-cached
+ * dirty cache lines to be cached locally without grinder warnings or
+ * assertions (used by Linux kernel fast memcpy).
+ */
 #define SIM_CONTROL_ALLOW_MULTIPLE_CACHING 22
 
-//! If written to SPR_SIM_CONTROL, enables memory tracing.
-//!
+/** If written to SPR_SIM_CONTROL, enables memory tracing. */
 #define SIM_CONTROL_ENABLE_MEM_LOGGING 23
 
-//! If written to SPR_SIM_CONTROL, disables memory tracing.
-//!
+/** If written to SPR_SIM_CONTROL, disables memory tracing. */
 #define SIM_CONTROL_DISABLE_MEM_LOGGING 24
 
-//! If written to SPR_SIM_CONTROL, changes the shaping parameters of one of
-//! the gbe or xgbe shims. Must specify the shim id, the type, the units, and
-//! the rate, as defined in SIM_SHAPING_SPR_ARG.
-//!
+/**
+ * If written to SPR_SIM_CONTROL, changes the shaping parameters of one of
+ * the gbe or xgbe shims. Must specify the shim id, the type, the units, and
+ * the rate, as defined in SIM_SHAPING_SPR_ARG.
+ */
 #define SIM_CONTROL_SHAPING 25
 
-//! If written to SPR_SIM_CONTROL, combined with character (shifted by 8),
-//! requests that a simulator command be executed.  Written to once for each
-//! character in the command, plus a final NUL.
-//!
+/**
+ * If written to SPR_SIM_CONTROL, combined with character (shifted by 8),
+ * requests that a simulator command be executed.  Written to once for each
+ * character in the command, plus a final NUL.
+ */
 #define SIM_CONTROL_COMMAND 26
 
-//! If written to SPR_SIM_CONTROL, indicates that the simulated system
-//! is panicking, to allow debugging via --debug-on-panic.
-//!
+/**
+ * If written to SPR_SIM_CONTROL, indicates that the simulated system
+ * is panicking, to allow debugging via --debug-on-panic.
+ */
 #define SIM_CONTROL_PANIC 27
 
-//! If written to SPR_SIM_CONTROL, triggers a simulator syscall.
-//! See "sim_syscall()" for more info.
-//!
+/**
+ * If written to SPR_SIM_CONTROL, triggers a simulator syscall.
+ * See "sim_syscall()" for more info.
+ */
 #define SIM_CONTROL_SYSCALL 32
 
-//! If written to SPR_SIM_CONTROL, combined with a pid (shifted by 8),
-//! provides the pid that subsequent SIM_CONTROL_OS_FORK writes should
-//! use as the pid, rather than the default previous SIM_CONTROL_OS_SWITCH.
-//!
+/**
+ * If written to SPR_SIM_CONTROL, combined with a pid (shifted by 8),
+ * provides the pid that subsequent SIM_CONTROL_OS_FORK writes should
+ * use as the pid, rather than the default previous SIM_CONTROL_OS_SWITCH.
+ */
 #define SIM_CONTROL_OS_FORK_PARENT 33
 
-//! If written to SPR_SIM_CONTROL, combined with a mPIPE shim number
-//! (shifted by 8), clears the pending magic data section.  The cleared
-//! pending magic data section and any subsequently appended magic bytes
-//! will only take effect when the classifier blast programmer is run.
+/**
+ * If written to SPR_SIM_CONTROL, combined with a mPIPE shim number
+ * (shifted by 8), clears the pending magic data section.  The cleared
+ * pending magic data section and any subsequently appended magic bytes
+ * will only take effect when the classifier blast programmer is run.
+ */
 #define SIM_CONTROL_CLEAR_MPIPE_MAGIC_BYTES 34
 
-//! If written to SPR_SIM_CONTROL, combined with a mPIPE shim number
-//! (shifted by 8) and a byte of data (shifted by 16), appends that byte
-//! to the shim's pending magic data section.  The pending magic data
-//! section takes effect when the classifier blast programmer is run.
+/**
+ * If written to SPR_SIM_CONTROL, combined with a mPIPE shim number
+ * (shifted by 8) and a byte of data (shifted by 16), appends that byte
+ * to the shim's pending magic data section.  The pending magic data
+ * section takes effect when the classifier blast programmer is run.
+ */
 #define SIM_CONTROL_APPEND_MPIPE_MAGIC_BYTE 35
 
-//! If written to SPR_SIM_CONTROL, combined with a mPIPE shim number
-//! (shifted by 8), an enable=1/disable=0 bit (shifted by 16), and a
-//! mask of links (shifted by 32), enable or disable the corresponding
-//! mPIPE links.
+/**
+ * If written to SPR_SIM_CONTROL, combined with a mPIPE shim number
+ * (shifted by 8), an enable=1/disable=0 bit (shifted by 16), and a
+ * mask of links (shifted by 32), enable or disable the corresponding
+ * mPIPE links.
+ */
 #define SIM_CONTROL_ENABLE_MPIPE_LINK_MAGIC_BYTE 36
 
-//== Syscall numbers for use with "sim_syscall()".
 
-//! Syscall number for sim_add_watchpoint().
-//!
+/*
+ * Syscall numbers for use with "sim_syscall()".
+ */
+
+/** Syscall number for sim_add_watchpoint(). */
 #define SIM_SYSCALL_ADD_WATCHPOINT 2
 
-//! Syscall number for sim_remove_watchpoint().
-//!
+/** Syscall number for sim_remove_watchpoint(). */
 #define SIM_SYSCALL_REMOVE_WATCHPOINT 3
 
-//! Syscall number for sim_query_watchpoint().
-//!
+/** Syscall number for sim_query_watchpoint(). */
 #define SIM_SYSCALL_QUERY_WATCHPOINT 4
 
-//! Syscall number that asserts that the cache lines whose 64-bit PA
-//! is passed as the second argument to sim_syscall(), and over a
-//! range passed as the third argument, are no longer in cache.
-//! The simulator raises an error if this is not the case.
-//!
+/**
+ * Syscall number that asserts that the cache lines whose 64-bit PA
+ * is passed as the second argument to sim_syscall(), and over a
+ * range passed as the third argument, are no longer in cache.
+ * The simulator raises an error if this is not the case.
+ */
 #define SIM_SYSCALL_VALIDATE_LINES_EVICTED 5
 
+/** Syscall number for sim_query_cpu_speed(). */
+#define SIM_SYSCALL_QUERY_CPU_SPEED 6
+
 
-//== Bit masks which can be shifted by 8, combined with
-//== SIM_CONTROL_SET_TRACING, and written to SPR_SIM_CONTROL.
+/*
+ * Bit masks which can be shifted by 8, combined with
+ * SIM_CONTROL_SET_TRACING, and written to SPR_SIM_CONTROL.
+ */
 
-//! @addtogroup arch_sim
-//! @{
+/**
+ * @addtogroup arch_sim
+ * @{
+ */
 
-//! Enable --trace-cycle when passed to simulator_set_tracing().
-//!
+/** Enable --trace-cycle when passed to simulator_set_tracing(). */
 #define SIM_TRACE_CYCLES          0x01
 
-//! Enable --trace-router when passed to simulator_set_tracing().
-//!
+/** Enable --trace-router when passed to simulator_set_tracing(). */
 #define SIM_TRACE_ROUTER          0x02
 
-//! Enable --trace-register-writes when passed to simulator_set_tracing().
-//!
+/** Enable --trace-register-writes when passed to simulator_set_tracing(). */
 #define SIM_TRACE_REGISTER_WRITES 0x04
 
-//! Enable --trace-disasm when passed to simulator_set_tracing().
-//!
+/** Enable --trace-disasm when passed to simulator_set_tracing(). */
 #define SIM_TRACE_DISASM          0x08
 
-//! Enable --trace-stall-info when passed to simulator_set_tracing().
-//!
+/** Enable --trace-stall-info when passed to simulator_set_tracing(). */
 #define SIM_TRACE_STALL_INFO      0x10
 
-//! Enable --trace-memory-controller when passed to simulator_set_tracing().
-//!
+/** Enable --trace-memory-controller when passed to simulator_set_tracing(). */
 #define SIM_TRACE_MEMORY_CONTROLLER 0x20
 
-//! Enable --trace-l2 when passed to simulator_set_tracing().
-//!
+/** Enable --trace-l2 when passed to simulator_set_tracing(). */
 #define SIM_TRACE_L2_CACHE 0x40
 
-//! Enable --trace-lines when passed to simulator_set_tracing().
-//!
+/** Enable --trace-lines when passed to simulator_set_tracing(). */
 #define SIM_TRACE_LINES 0x80
 
-//! Turn off all tracing when passed to simulator_set_tracing().
-//!
+/** Turn off all tracing when passed to simulator_set_tracing(). */
 #define SIM_TRACE_NONE 0
 
-//! Turn on all tracing when passed to simulator_set_tracing().
-//!
+/** Turn on all tracing when passed to simulator_set_tracing(). */
 #define SIM_TRACE_ALL (-1)
 
-//! @}
+/** @} */
 
-//! Computes the value to write to SPR_SIM_CONTROL to set tracing flags.
-//!
+/** Computes the value to write to SPR_SIM_CONTROL to set tracing flags. */
 #define SIM_TRACE_SPR_ARG(mask) \
   (SIM_CONTROL_SET_TRACING | ((mask) << _SIM_CONTROL_OPERATOR_BITS))
 
 
-//== Bit masks which can be shifted by 8, combined with
-//== SIM_CONTROL_DUMP, and written to SPR_SIM_CONTROL.
+/*
+ * Bit masks which can be shifted by 8, combined with
+ * SIM_CONTROL_DUMP, and written to SPR_SIM_CONTROL.
+ */
 
-//! @addtogroup arch_sim
-//! @{
+/**
+ * @addtogroup arch_sim
+ * @{
+ */
 
-//! Dump the general-purpose registers.
-//!
+/** Dump the general-purpose registers. */
 #define SIM_DUMP_REGS          0x001
 
-//! Dump the SPRs.
-//!
+/** Dump the SPRs. */
 #define SIM_DUMP_SPRS          0x002
 
-//! Dump the ITLB.
-//!
+/** Dump the ITLB. */
 #define SIM_DUMP_ITLB          0x004
 
-//! Dump the DTLB.
-//!
+/** Dump the DTLB. */
 #define SIM_DUMP_DTLB          0x008
 
-//! Dump the L1 I-cache.
-//!
+/** Dump the L1 I-cache. */
 #define SIM_DUMP_L1I           0x010
 
-//! Dump the L1 D-cache.
-//!
+/** Dump the L1 D-cache. */
 #define SIM_DUMP_L1D           0x020
 
-//! Dump the L2 cache.
-//!
+/** Dump the L2 cache. */
 #define SIM_DUMP_L2            0x040
 
-//! Dump the switch registers.
-//!
+/** Dump the switch registers. */
 #define SIM_DUMP_SNREGS        0x080
 
-//! Dump the switch ITLB.
-//!
+/** Dump the switch ITLB. */
 #define SIM_DUMP_SNITLB        0x100
 
-//! Dump the switch L1 I-cache.
-//!
+/** Dump the switch L1 I-cache. */
 #define SIM_DUMP_SNL1I         0x200
 
-//! Dump the current backtrace.
-//!
+/** Dump the current backtrace. */
 #define SIM_DUMP_BACKTRACE     0x400
 
-//! Only dump valid lines in caches.
-//!
+/** Only dump valid lines in caches. */
 #define SIM_DUMP_VALID_LINES   0x800
 
-//! Dump everything that is dumpable.
-//!
+/** Dump everything that is dumpable. */
 #define SIM_DUMP_ALL (-1 & ~SIM_DUMP_VALID_LINES)
 
-// @}
+/** @} */
 
-//! Computes the value to write to SPR_SIM_CONTROL to dump machine state.
-//!
+/** Computes the value to write to SPR_SIM_CONTROL to dump machine state. */
 #define SIM_DUMP_SPR_ARG(mask) \
   (SIM_CONTROL_DUMP | ((mask) << _SIM_CONTROL_OPERATOR_BITS))
 
 
-//== Bit masks which can be shifted by 8, combined with
-//== SIM_CONTROL_PROFILER_CHIP_xxx, and written to SPR_SIM_CONTROL.
+/*
+ * Bit masks which can be shifted by 8, combined with
+ * SIM_CONTROL_PROFILER_CHIP_xxx, and written to SPR_SIM_CONTROL.
+ */
 
-//! @addtogroup arch_sim
-//! @{
+/**
+ * @addtogroup arch_sim
+ * @{
+ */
 
-//! Use with with SIM_PROFILER_CHIP_xxx to control the memory controllers.
-//!
+/** Use with with SIM_PROFILER_CHIP_xxx to control the memory controllers. */
 #define SIM_CHIP_MEMCTL        0x001
 
-//! Use with with SIM_PROFILER_CHIP_xxx to control the XAUI interface.
-//!
+/** Use with with SIM_PROFILER_CHIP_xxx to control the XAUI interface. */
 #define SIM_CHIP_XAUI          0x002
 
-//! Use with with SIM_PROFILER_CHIP_xxx to control the PCIe interface.
-//!
+/** Use with with SIM_PROFILER_CHIP_xxx to control the PCIe interface. */
 #define SIM_CHIP_PCIE          0x004
 
-//! Use with with SIM_PROFILER_CHIP_xxx to control the MPIPE interface.
-//!
+/** Use with with SIM_PROFILER_CHIP_xxx to control the MPIPE interface. */
 #define SIM_CHIP_MPIPE         0x008
 
-//! Reference all chip devices.
-//!
+/** Use with with SIM_PROFILER_CHIP_xxx to control the TRIO interface. */
+#define SIM_CHIP_TRIO          0x010
+
+/** Reference all chip devices. */
 #define SIM_CHIP_ALL (-1)
 
-//! @}
+/** @} */
 
-//! Computes the value to write to SPR_SIM_CONTROL to clear chip statistics.
-//!
+/** Computes the value to write to SPR_SIM_CONTROL to clear chip statistics. */
 #define SIM_PROFILER_CHIP_CLEAR_SPR_ARG(mask) \
   (SIM_CONTROL_PROFILER_CHIP_CLEAR | ((mask) << _SIM_CONTROL_OPERATOR_BITS))
 
-//! Computes the value to write to SPR_SIM_CONTROL to disable chip statistics.
-//!
+/** Computes the value to write to SPR_SIM_CONTROL to disable chip statistics.*/
 #define SIM_PROFILER_CHIP_DISABLE_SPR_ARG(mask) \
   (SIM_CONTROL_PROFILER_CHIP_DISABLE | ((mask) << _SIM_CONTROL_OPERATOR_BITS))
 
-//! Computes the value to write to SPR_SIM_CONTROL to enable chip statistics.
-//!
+/** Computes the value to write to SPR_SIM_CONTROL to enable chip statistics. */
 #define SIM_PROFILER_CHIP_ENABLE_SPR_ARG(mask) \
   (SIM_CONTROL_PROFILER_CHIP_ENABLE | ((mask) << _SIM_CONTROL_OPERATOR_BITS))
 
 
 
-// Shim bitrate controls.
+/* Shim bitrate controls. */
 
-//! The number of bits used to store the shim id.
-//!
+/** The number of bits used to store the shim id. */
 #define SIM_CONTROL_SHAPING_SHIM_ID_BITS 3
 
-//! @addtogroup arch_sim
-//! @{
+/**
+ * @addtogroup arch_sim
+ * @{
+ */
 
-//! Change the gbe 0 bitrate.
-//!
+/** Change the gbe 0 bitrate. */
 #define SIM_CONTROL_SHAPING_GBE_0 0x0
 
-//! Change the gbe 1 bitrate.
-//!
+/** Change the gbe 1 bitrate. */
 #define SIM_CONTROL_SHAPING_GBE_1 0x1
 
-//! Change the gbe 2 bitrate.
-//!
+/** Change the gbe 2 bitrate. */
 #define SIM_CONTROL_SHAPING_GBE_2 0x2
 
-//! Change the gbe 3 bitrate.
-//!
+/** Change the gbe 3 bitrate. */
 #define SIM_CONTROL_SHAPING_GBE_3 0x3
 
-//! Change the xgbe 0 bitrate.
-//!
+/** Change the xgbe 0 bitrate. */
 #define SIM_CONTROL_SHAPING_XGBE_0 0x4
 
-//! Change the xgbe 1 bitrate.
-//!
+/** Change the xgbe 1 bitrate. */
 #define SIM_CONTROL_SHAPING_XGBE_1 0x5
 
-//! The type of shaping to do.
-//!
+/** The type of shaping to do. */
 #define SIM_CONTROL_SHAPING_TYPE_BITS 2
 
-//! Control the multiplier.
-//!
+/** Control the multiplier. */
 #define SIM_CONTROL_SHAPING_MULTIPLIER 0
 
-//! Control the PPS.
-//!
+/** Control the PPS. */
 #define SIM_CONTROL_SHAPING_PPS 1
 
-//! Control the BPS.
-//!
+/** Control the BPS. */
 #define SIM_CONTROL_SHAPING_BPS 2
 
-//! The number of bits for the units for the shaping parameter.
-//!
+/** The number of bits for the units for the shaping parameter. */
 #define SIM_CONTROL_SHAPING_UNITS_BITS 2
 
-//! Provide a number in single units.
-//!
+/** Provide a number in single units. */
 #define SIM_CONTROL_SHAPING_UNITS_SINGLE 0
 
-//! Provide a number in kilo units.
-//!
+/** Provide a number in kilo units. */
 #define SIM_CONTROL_SHAPING_UNITS_KILO 1
 
-//! Provide a number in mega units.
-//!
+/** Provide a number in mega units. */
 #define SIM_CONTROL_SHAPING_UNITS_MEGA 2
 
-//! Provide a number in giga units.
-//!
+/** Provide a number in giga units. */
 #define SIM_CONTROL_SHAPING_UNITS_GIGA 3
 
-// @}
+/** @} */
 
-//! How many bits are available for the rate.
-//!
+/** How many bits are available for the rate. */
 #define SIM_CONTROL_SHAPING_RATE_BITS \
   (32 - (_SIM_CONTROL_OPERATOR_BITS + \
          SIM_CONTROL_SHAPING_SHIM_ID_BITS + \
          SIM_CONTROL_SHAPING_TYPE_BITS + \
          SIM_CONTROL_SHAPING_UNITS_BITS))
 
-//! Computes the value to write to SPR_SIM_CONTROL to change a bitrate.
-//!
+/** Computes the value to write to SPR_SIM_CONTROL to change a bitrate. */
 #define SIM_SHAPING_SPR_ARG(shim, type, units, rate) \
   (SIM_CONTROL_SHAPING | \
    ((shim) | \
@@ -483,30 +470,36 @@
                SIM_CONTROL_SHAPING_UNITS_BITS))) << _SIM_CONTROL_OPERATOR_BITS)
 
 
-//== Values returned when reading SPR_SIM_CONTROL.
-// ISSUE: These names should share a longer common prefix.
+/*
+ * Values returned when reading SPR_SIM_CONTROL.
+ * ISSUE: These names should share a longer common prefix.
+ */
 
-//! When reading SPR_SIM_CONTROL, the mask of simulator tracing bits
-//! (SIM_TRACE_xxx values).
-//!
+/**
+ * When reading SPR_SIM_CONTROL, the mask of simulator tracing bits
+ * (SIM_TRACE_xxx values).
+ */
 #define SIM_TRACE_FLAG_MASK 0xFFFF
 
-//! When reading SPR_SIM_CONTROL, the mask for whether profiling is enabled.
-//!
+/** When reading SPR_SIM_CONTROL, the mask for whether profiling is enabled. */
 #define SIM_PROFILER_ENABLED_MASK 0x10000
 
 
-//== Special arguments for "SIM_CONTROL_PUTC".
+/*
+ * Special arguments for "SIM_CONTROL_PUTC".
+ */
 
-//! Flag value for forcing a PUTC string-flush, including
-//! coordinate/cycle prefix and newline.
-//!
+/**
+ * Flag value for forcing a PUTC string-flush, including
+ * coordinate/cycle prefix and newline.
+ */
 #define SIM_PUTC_FLUSH_STRING 0x100
 
-//! Flag value for forcing a PUTC binary-data-flush, which skips the
-//! prefix and does not append a newline.
-//!
+/**
+ * Flag value for forcing a PUTC binary-data-flush, which skips the
+ * prefix and does not append a newline.
+ */
 #define SIM_PUTC_FLUSH_BINARY 0x101
 
 
-#endif //__ARCH_SIM_DEF_H__
+#endif /* __ARCH_SIM_DEF_H__ */
diff --git a/arch/tile/include/arch/spr_def.h b/arch/tile/include/arch/spr_def.h
index c8fdbd9a45e6..f548efeb2de3 100644
--- a/arch/tile/include/arch/spr_def.h
+++ b/arch/tile/include/arch/spr_def.h
@@ -12,8 +12,102 @@
  *   more details.
  */
 
+/* Include the proper base SPR definition file. */
 #ifdef __tilegx__
 #include <arch/spr_def_64.h>
 #else
 #include <arch/spr_def_32.h>
 #endif
+
+#ifdef __KERNEL__
+
+/*
+ * In addition to including the proper base SPR definition file, depending
+ * on machine architecture, this file defines several macros which allow
+ * kernel code to use protection-level dependent SPRs without worrying
+ * about which PL it's running at.  In these macros, the PL that the SPR
+ * or interrupt number applies to is replaced by K.
+ */
+
+#if CONFIG_KERNEL_PL != 1 && CONFIG_KERNEL_PL != 2
+#error CONFIG_KERNEL_PL must be 1 or 2
+#endif
+
+/* Concatenate 4 strings. */
+#define __concat4(a, b, c, d) a ## b ## c ## d
+#define _concat4(a, b, c, d)  __concat4(a, b, c, d)
+
+#ifdef __tilegx__
+
+/* TILE-Gx dependent, protection-level dependent SPRs. */
+
+#define SPR_INTERRUPT_MASK_K \
+        _concat4(SPR_INTERRUPT_MASK_, CONFIG_KERNEL_PL,,)
+#define SPR_INTERRUPT_MASK_SET_K \
+        _concat4(SPR_INTERRUPT_MASK_SET_, CONFIG_KERNEL_PL,,)
+#define SPR_INTERRUPT_MASK_RESET_K \
+        _concat4(SPR_INTERRUPT_MASK_RESET_, CONFIG_KERNEL_PL,,)
+#define SPR_INTERRUPT_VECTOR_BASE_K \
+        _concat4(SPR_INTERRUPT_VECTOR_BASE_, CONFIG_KERNEL_PL,,)
+
+#define SPR_IPI_MASK_K \
+        _concat4(SPR_IPI_MASK_, CONFIG_KERNEL_PL,,)
+#define SPR_IPI_MASK_RESET_K \
+        _concat4(SPR_IPI_MASK_RESET_, CONFIG_KERNEL_PL,,)
+#define SPR_IPI_MASK_SET_K \
+        _concat4(SPR_IPI_MASK_SET_, CONFIG_KERNEL_PL,,)
+#define SPR_IPI_EVENT_K \
+        _concat4(SPR_IPI_EVENT_, CONFIG_KERNEL_PL,,)
+#define SPR_IPI_EVENT_RESET_K \
+        _concat4(SPR_IPI_EVENT_RESET_, CONFIG_KERNEL_PL,,)
+#define SPR_IPI_MASK_SET_K \
+        _concat4(SPR_IPI_MASK_SET_, CONFIG_KERNEL_PL,,)
+#define INT_IPI_K \
+        _concat4(INT_IPI_, CONFIG_KERNEL_PL,,)
+
+#define SPR_SINGLE_STEP_CONTROL_K \
+        _concat4(SPR_SINGLE_STEP_CONTROL_, CONFIG_KERNEL_PL,,)
+#define SPR_SINGLE_STEP_EN_K_K \
+        _concat4(SPR_SINGLE_STEP_EN_, CONFIG_KERNEL_PL, _, CONFIG_KERNEL_PL)
+#define INT_SINGLE_STEP_K \
+        _concat4(INT_SINGLE_STEP_, CONFIG_KERNEL_PL,,)
+
+#else
+
+/* TILEPro dependent, protection-level dependent SPRs. */
+
+#define SPR_INTERRUPT_MASK_K_0 \
+        _concat4(SPR_INTERRUPT_MASK_, CONFIG_KERNEL_PL, _0,)
+#define SPR_INTERRUPT_MASK_K_1 \
+        _concat4(SPR_INTERRUPT_MASK_, CONFIG_KERNEL_PL, _1,)
+#define SPR_INTERRUPT_MASK_SET_K_0 \
+        _concat4(SPR_INTERRUPT_MASK_SET_, CONFIG_KERNEL_PL, _0,)
+#define SPR_INTERRUPT_MASK_SET_K_1 \
+        _concat4(SPR_INTERRUPT_MASK_SET_, CONFIG_KERNEL_PL, _1,)
+#define SPR_INTERRUPT_MASK_RESET_K_0 \
+        _concat4(SPR_INTERRUPT_MASK_RESET_, CONFIG_KERNEL_PL, _0,)
+#define SPR_INTERRUPT_MASK_RESET_K_1 \
+        _concat4(SPR_INTERRUPT_MASK_RESET_, CONFIG_KERNEL_PL, _1,)
+
+#endif
+
+/* Generic protection-level dependent SPRs. */
+
+#define SPR_SYSTEM_SAVE_K_0 \
+        _concat4(SPR_SYSTEM_SAVE_, CONFIG_KERNEL_PL, _0,)
+#define SPR_SYSTEM_SAVE_K_1 \
+        _concat4(SPR_SYSTEM_SAVE_, CONFIG_KERNEL_PL, _1,)
+#define SPR_SYSTEM_SAVE_K_2 \
+        _concat4(SPR_SYSTEM_SAVE_, CONFIG_KERNEL_PL, _2,)
+#define SPR_SYSTEM_SAVE_K_3 \
+        _concat4(SPR_SYSTEM_SAVE_, CONFIG_KERNEL_PL, _3,)
+#define SPR_EX_CONTEXT_K_0 \
+        _concat4(SPR_EX_CONTEXT_, CONFIG_KERNEL_PL, _0,)
+#define SPR_EX_CONTEXT_K_1 \
+        _concat4(SPR_EX_CONTEXT_, CONFIG_KERNEL_PL, _1,)
+#define SPR_INTCTRL_K_STATUS \
+        _concat4(SPR_INTCTRL_, CONFIG_KERNEL_PL, _STATUS,)
+#define INT_INTCTRL_K \
+        _concat4(INT_INTCTRL_, CONFIG_KERNEL_PL,,)
+
+#endif /* __KERNEL__ */
diff --git a/arch/tile/include/arch/spr_def_32.h b/arch/tile/include/arch/spr_def_32.h
index b4fc06864df6..bbc1f4c924ee 100644
--- a/arch/tile/include/arch/spr_def_32.h
+++ b/arch/tile/include/arch/spr_def_32.h
@@ -56,58 +56,93 @@
 #define SPR_EX_CONTEXT_1_1__ICS_SHIFT 2
 #define SPR_EX_CONTEXT_1_1__ICS_RMASK 0x1
 #define SPR_EX_CONTEXT_1_1__ICS_MASK  0x4
+#define SPR_EX_CONTEXT_2_0 0x4605
+#define SPR_EX_CONTEXT_2_1 0x4606
+#define SPR_EX_CONTEXT_2_1__PL_SHIFT 0
+#define SPR_EX_CONTEXT_2_1__PL_RMASK 0x3
+#define SPR_EX_CONTEXT_2_1__PL_MASK  0x3
+#define SPR_EX_CONTEXT_2_1__ICS_SHIFT 2
+#define SPR_EX_CONTEXT_2_1__ICS_RMASK 0x1
+#define SPR_EX_CONTEXT_2_1__ICS_MASK  0x4
 #define SPR_FAIL 0x4e09
 #define SPR_INTCTRL_0_STATUS 0x4a07
 #define SPR_INTCTRL_1_STATUS 0x4807
+#define SPR_INTCTRL_2_STATUS 0x4607
 #define SPR_INTERRUPT_CRITICAL_SECTION 0x4e0a
 #define SPR_INTERRUPT_MASK_0_0 0x4a08
 #define SPR_INTERRUPT_MASK_0_1 0x4a09
 #define SPR_INTERRUPT_MASK_1_0 0x4809
 #define SPR_INTERRUPT_MASK_1_1 0x480a
+#define SPR_INTERRUPT_MASK_2_0 0x4608
+#define SPR_INTERRUPT_MASK_2_1 0x4609
 #define SPR_INTERRUPT_MASK_RESET_0_0 0x4a0a
 #define SPR_INTERRUPT_MASK_RESET_0_1 0x4a0b
 #define SPR_INTERRUPT_MASK_RESET_1_0 0x480b
 #define SPR_INTERRUPT_MASK_RESET_1_1 0x480c
+#define SPR_INTERRUPT_MASK_RESET_2_0 0x460a
+#define SPR_INTERRUPT_MASK_RESET_2_1 0x460b
 #define SPR_INTERRUPT_MASK_SET_0_0 0x4a0c
 #define SPR_INTERRUPT_MASK_SET_0_1 0x4a0d
 #define SPR_INTERRUPT_MASK_SET_1_0 0x480d
 #define SPR_INTERRUPT_MASK_SET_1_1 0x480e
+#define SPR_INTERRUPT_MASK_SET_2_0 0x460c
+#define SPR_INTERRUPT_MASK_SET_2_1 0x460d
 #define SPR_MPL_DMA_CPL_SET_0 0x5800
 #define SPR_MPL_DMA_CPL_SET_1 0x5801
+#define SPR_MPL_DMA_CPL_SET_2 0x5802
 #define SPR_MPL_DMA_NOTIFY_SET_0 0x3800
 #define SPR_MPL_DMA_NOTIFY_SET_1 0x3801
+#define SPR_MPL_DMA_NOTIFY_SET_2 0x3802
 #define SPR_MPL_INTCTRL_0_SET_0 0x4a00
 #define SPR_MPL_INTCTRL_0_SET_1 0x4a01
+#define SPR_MPL_INTCTRL_0_SET_2 0x4a02
 #define SPR_MPL_INTCTRL_1_SET_0 0x4800
 #define SPR_MPL_INTCTRL_1_SET_1 0x4801
+#define SPR_MPL_INTCTRL_1_SET_2 0x4802
+#define SPR_MPL_INTCTRL_2_SET_0 0x4600
+#define SPR_MPL_INTCTRL_2_SET_1 0x4601
+#define SPR_MPL_INTCTRL_2_SET_2 0x4602
 #define SPR_MPL_SN_ACCESS_SET_0 0x0800
 #define SPR_MPL_SN_ACCESS_SET_1 0x0801
+#define SPR_MPL_SN_ACCESS_SET_2 0x0802
 #define SPR_MPL_SN_CPL_SET_0 0x5a00
 #define SPR_MPL_SN_CPL_SET_1 0x5a01
+#define SPR_MPL_SN_CPL_SET_2 0x5a02
 #define SPR_MPL_SN_FIREWALL_SET_0 0x2c00
 #define SPR_MPL_SN_FIREWALL_SET_1 0x2c01
+#define SPR_MPL_SN_FIREWALL_SET_2 0x2c02
 #define SPR_MPL_SN_NOTIFY_SET_0 0x2a00
 #define SPR_MPL_SN_NOTIFY_SET_1 0x2a01
+#define SPR_MPL_SN_NOTIFY_SET_2 0x2a02
 #define SPR_MPL_UDN_ACCESS_SET_0 0x0c00
 #define SPR_MPL_UDN_ACCESS_SET_1 0x0c01
+#define SPR_MPL_UDN_ACCESS_SET_2 0x0c02
 #define SPR_MPL_UDN_AVAIL_SET_0 0x4000
 #define SPR_MPL_UDN_AVAIL_SET_1 0x4001
+#define SPR_MPL_UDN_AVAIL_SET_2 0x4002
 #define SPR_MPL_UDN_CA_SET_0 0x3c00
 #define SPR_MPL_UDN_CA_SET_1 0x3c01
+#define SPR_MPL_UDN_CA_SET_2 0x3c02
 #define SPR_MPL_UDN_COMPLETE_SET_0 0x1400
 #define SPR_MPL_UDN_COMPLETE_SET_1 0x1401
+#define SPR_MPL_UDN_COMPLETE_SET_2 0x1402
 #define SPR_MPL_UDN_FIREWALL_SET_0 0x3000
 #define SPR_MPL_UDN_FIREWALL_SET_1 0x3001
+#define SPR_MPL_UDN_FIREWALL_SET_2 0x3002
 #define SPR_MPL_UDN_REFILL_SET_0 0x1000
 #define SPR_MPL_UDN_REFILL_SET_1 0x1001
+#define SPR_MPL_UDN_REFILL_SET_2 0x1002
 #define SPR_MPL_UDN_TIMER_SET_0 0x3600
 #define SPR_MPL_UDN_TIMER_SET_1 0x3601
+#define SPR_MPL_UDN_TIMER_SET_2 0x3602
 #define SPR_MPL_WORLD_ACCESS_SET_0 0x4e00
 #define SPR_MPL_WORLD_ACCESS_SET_1 0x4e01
+#define SPR_MPL_WORLD_ACCESS_SET_2 0x4e02
 #define SPR_PASS 0x4e0b
 #define SPR_PERF_COUNT_0 0x4205
 #define SPR_PERF_COUNT_1 0x4206
 #define SPR_PERF_COUNT_CTL 0x4207
+#define SPR_PERF_COUNT_DN_CTL 0x4210
 #define SPR_PERF_COUNT_STS 0x4208
 #define SPR_PROC_STATUS 0x4f00
 #define SPR_SIM_CONTROL 0x4e0c
@@ -124,6 +159,10 @@
 #define SPR_SYSTEM_SAVE_1_1 0x4901
 #define SPR_SYSTEM_SAVE_1_2 0x4902
 #define SPR_SYSTEM_SAVE_1_3 0x4903
+#define SPR_SYSTEM_SAVE_2_0 0x4700
+#define SPR_SYSTEM_SAVE_2_1 0x4701
+#define SPR_SYSTEM_SAVE_2_2 0x4702
+#define SPR_SYSTEM_SAVE_2_3 0x4703
 #define SPR_TILE_COORD 0x4c17
 #define SPR_TILE_RTF_HWM 0x4e10
 #define SPR_TILE_TIMER_CONTROL 0x3205
diff --git a/arch/tile/include/arch/spr_def_64.h b/arch/tile/include/arch/spr_def_64.h
new file mode 100644
index 000000000000..cd3e5f95d5fd
--- /dev/null
+++ b/arch/tile/include/arch/spr_def_64.h
@@ -0,0 +1,173 @@
+/*
+ * Copyright 2011 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#ifndef __DOXYGEN__
+
+#ifndef __ARCH_SPR_DEF_H__
+#define __ARCH_SPR_DEF_H__
+
+#define SPR_AUX_PERF_COUNT_0 0x2105
+#define SPR_AUX_PERF_COUNT_1 0x2106
+#define SPR_AUX_PERF_COUNT_CTL 0x2107
+#define SPR_AUX_PERF_COUNT_STS 0x2108
+#define SPR_CMPEXCH_VALUE 0x2780
+#define SPR_CYCLE 0x2781
+#define SPR_DONE 0x2705
+#define SPR_DSTREAM_PF 0x2706
+#define SPR_EVENT_BEGIN 0x2782
+#define SPR_EVENT_END 0x2783
+#define SPR_EX_CONTEXT_0_0 0x2580
+#define SPR_EX_CONTEXT_0_1 0x2581
+#define SPR_EX_CONTEXT_0_1__PL_SHIFT 0
+#define SPR_EX_CONTEXT_0_1__PL_RMASK 0x3
+#define SPR_EX_CONTEXT_0_1__PL_MASK  0x3
+#define SPR_EX_CONTEXT_0_1__ICS_SHIFT 2
+#define SPR_EX_CONTEXT_0_1__ICS_RMASK 0x1
+#define SPR_EX_CONTEXT_0_1__ICS_MASK  0x4
+#define SPR_EX_CONTEXT_1_0 0x2480
+#define SPR_EX_CONTEXT_1_1 0x2481
+#define SPR_EX_CONTEXT_1_1__PL_SHIFT 0
+#define SPR_EX_CONTEXT_1_1__PL_RMASK 0x3
+#define SPR_EX_CONTEXT_1_1__PL_MASK  0x3
+#define SPR_EX_CONTEXT_1_1__ICS_SHIFT 2
+#define SPR_EX_CONTEXT_1_1__ICS_RMASK 0x1
+#define SPR_EX_CONTEXT_1_1__ICS_MASK  0x4
+#define SPR_EX_CONTEXT_2_0 0x2380
+#define SPR_EX_CONTEXT_2_1 0x2381
+#define SPR_EX_CONTEXT_2_1__PL_SHIFT 0
+#define SPR_EX_CONTEXT_2_1__PL_RMASK 0x3
+#define SPR_EX_CONTEXT_2_1__PL_MASK  0x3
+#define SPR_EX_CONTEXT_2_1__ICS_SHIFT 2
+#define SPR_EX_CONTEXT_2_1__ICS_RMASK 0x1
+#define SPR_EX_CONTEXT_2_1__ICS_MASK  0x4
+#define SPR_FAIL 0x2707
+#define SPR_ILL_TRANS_REASON__I_STREAM_VA_RMASK 0x1
+#define SPR_INTCTRL_0_STATUS 0x2505
+#define SPR_INTCTRL_1_STATUS 0x2405
+#define SPR_INTCTRL_2_STATUS 0x2305
+#define SPR_INTERRUPT_CRITICAL_SECTION 0x2708
+#define SPR_INTERRUPT_MASK_0 0x2506
+#define SPR_INTERRUPT_MASK_1 0x2406
+#define SPR_INTERRUPT_MASK_2 0x2306
+#define SPR_INTERRUPT_MASK_RESET_0 0x2507
+#define SPR_INTERRUPT_MASK_RESET_1 0x2407
+#define SPR_INTERRUPT_MASK_RESET_2 0x2307
+#define SPR_INTERRUPT_MASK_SET_0 0x2508
+#define SPR_INTERRUPT_MASK_SET_1 0x2408
+#define SPR_INTERRUPT_MASK_SET_2 0x2308
+#define SPR_INTERRUPT_VECTOR_BASE_0 0x2509
+#define SPR_INTERRUPT_VECTOR_BASE_1 0x2409
+#define SPR_INTERRUPT_VECTOR_BASE_2 0x2309
+#define SPR_INTERRUPT_VECTOR_BASE_3 0x2209
+#define SPR_IPI_EVENT_0 0x1f05
+#define SPR_IPI_EVENT_1 0x1e05
+#define SPR_IPI_EVENT_2 0x1d05
+#define SPR_IPI_EVENT_RESET_0 0x1f06
+#define SPR_IPI_EVENT_RESET_1 0x1e06
+#define SPR_IPI_EVENT_RESET_2 0x1d06
+#define SPR_IPI_EVENT_SET_0 0x1f07
+#define SPR_IPI_EVENT_SET_1 0x1e07
+#define SPR_IPI_EVENT_SET_2 0x1d07
+#define SPR_IPI_MASK_0 0x1f08
+#define SPR_IPI_MASK_1 0x1e08
+#define SPR_IPI_MASK_2 0x1d08
+#define SPR_IPI_MASK_RESET_0 0x1f09
+#define SPR_IPI_MASK_RESET_1 0x1e09
+#define SPR_IPI_MASK_RESET_2 0x1d09
+#define SPR_IPI_MASK_SET_0 0x1f0a
+#define SPR_IPI_MASK_SET_1 0x1e0a
+#define SPR_IPI_MASK_SET_2 0x1d0a
+#define SPR_MPL_AUX_TILE_TIMER_SET_0 0x1700
+#define SPR_MPL_AUX_TILE_TIMER_SET_1 0x1701
+#define SPR_MPL_AUX_TILE_TIMER_SET_2 0x1702
+#define SPR_MPL_INTCTRL_0_SET_0 0x2500
+#define SPR_MPL_INTCTRL_0_SET_1 0x2501
+#define SPR_MPL_INTCTRL_0_SET_2 0x2502
+#define SPR_MPL_INTCTRL_1_SET_0 0x2400
+#define SPR_MPL_INTCTRL_1_SET_1 0x2401
+#define SPR_MPL_INTCTRL_1_SET_2 0x2402
+#define SPR_MPL_INTCTRL_2_SET_0 0x2300
+#define SPR_MPL_INTCTRL_2_SET_1 0x2301
+#define SPR_MPL_INTCTRL_2_SET_2 0x2302
+#define SPR_MPL_UDN_ACCESS_SET_0 0x0b00
+#define SPR_MPL_UDN_ACCESS_SET_1 0x0b01
+#define SPR_MPL_UDN_ACCESS_SET_2 0x0b02
+#define SPR_MPL_UDN_AVAIL_SET_0 0x1b00
+#define SPR_MPL_UDN_AVAIL_SET_1 0x1b01
+#define SPR_MPL_UDN_AVAIL_SET_2 0x1b02
+#define SPR_MPL_UDN_COMPLETE_SET_0 0x0600
+#define SPR_MPL_UDN_COMPLETE_SET_1 0x0601
+#define SPR_MPL_UDN_COMPLETE_SET_2 0x0602
+#define SPR_MPL_UDN_FIREWALL_SET_0 0x1500
+#define SPR_MPL_UDN_FIREWALL_SET_1 0x1501
+#define SPR_MPL_UDN_FIREWALL_SET_2 0x1502
+#define SPR_MPL_UDN_TIMER_SET_0 0x1900
+#define SPR_MPL_UDN_TIMER_SET_1 0x1901
+#define SPR_MPL_UDN_TIMER_SET_2 0x1902
+#define SPR_MPL_WORLD_ACCESS_SET_0 0x2700
+#define SPR_MPL_WORLD_ACCESS_SET_1 0x2701
+#define SPR_MPL_WORLD_ACCESS_SET_2 0x2702
+#define SPR_PASS 0x2709
+#define SPR_PERF_COUNT_0 0x2005
+#define SPR_PERF_COUNT_1 0x2006
+#define SPR_PERF_COUNT_CTL 0x2007
+#define SPR_PERF_COUNT_DN_CTL 0x2008
+#define SPR_PERF_COUNT_STS 0x2009
+#define SPR_PROC_STATUS 0x2784
+#define SPR_SIM_CONTROL 0x2785
+#define SPR_SINGLE_STEP_CONTROL_0 0x0405
+#define SPR_SINGLE_STEP_CONTROL_0__CANCELED_MASK  0x1
+#define SPR_SINGLE_STEP_CONTROL_0__INHIBIT_MASK  0x2
+#define SPR_SINGLE_STEP_CONTROL_1 0x0305
+#define SPR_SINGLE_STEP_CONTROL_1__CANCELED_MASK  0x1
+#define SPR_SINGLE_STEP_CONTROL_1__INHIBIT_MASK  0x2
+#define SPR_SINGLE_STEP_CONTROL_2 0x0205
+#define SPR_SINGLE_STEP_CONTROL_2__CANCELED_MASK  0x1
+#define SPR_SINGLE_STEP_CONTROL_2__INHIBIT_MASK  0x2
+#define SPR_SINGLE_STEP_EN_0_0 0x250a
+#define SPR_SINGLE_STEP_EN_0_1 0x240a
+#define SPR_SINGLE_STEP_EN_0_2 0x230a
+#define SPR_SINGLE_STEP_EN_1_0 0x250b
+#define SPR_SINGLE_STEP_EN_1_1 0x240b
+#define SPR_SINGLE_STEP_EN_1_2 0x230b
+#define SPR_SINGLE_STEP_EN_2_0 0x250c
+#define SPR_SINGLE_STEP_EN_2_1 0x240c
+#define SPR_SINGLE_STEP_EN_2_2 0x230c
+#define SPR_SYSTEM_SAVE_0_0 0x2582
+#define SPR_SYSTEM_SAVE_0_1 0x2583
+#define SPR_SYSTEM_SAVE_0_2 0x2584
+#define SPR_SYSTEM_SAVE_0_3 0x2585
+#define SPR_SYSTEM_SAVE_1_0 0x2482
+#define SPR_SYSTEM_SAVE_1_1 0x2483
+#define SPR_SYSTEM_SAVE_1_2 0x2484
+#define SPR_SYSTEM_SAVE_1_3 0x2485
+#define SPR_SYSTEM_SAVE_2_0 0x2382
+#define SPR_SYSTEM_SAVE_2_1 0x2383
+#define SPR_SYSTEM_SAVE_2_2 0x2384
+#define SPR_SYSTEM_SAVE_2_3 0x2385
+#define SPR_TILE_COORD 0x270b
+#define SPR_TILE_RTF_HWM 0x270c
+#define SPR_TILE_TIMER_CONTROL 0x1605
+#define SPR_UDN_AVAIL_EN 0x1b05
+#define SPR_UDN_DATA_AVAIL 0x0b80
+#define SPR_UDN_DEADLOCK_TIMEOUT 0x1906
+#define SPR_UDN_DEMUX_COUNT_0 0x0b05
+#define SPR_UDN_DEMUX_COUNT_1 0x0b06
+#define SPR_UDN_DEMUX_COUNT_2 0x0b07
+#define SPR_UDN_DEMUX_COUNT_3 0x0b08
+#define SPR_UDN_DIRECTION_PROTECT 0x1505
+
+#endif /* !defined(__ARCH_SPR_DEF_H__) */
+
+#endif /* !defined(__DOXYGEN__) */
diff --git a/arch/tile/include/asm/Kbuild b/arch/tile/include/asm/Kbuild
index 3b8f55b82dee..849ab2fa1f5c 100644
--- a/arch/tile/include/asm/Kbuild
+++ b/arch/tile/include/asm/Kbuild
@@ -1,3 +1,4 @@
 include include/asm-generic/Kbuild.asm
 
 header-y += ucontext.h
+header-y += hardwall.h
diff --git a/arch/tile/include/asm/atomic.h b/arch/tile/include/asm/atomic.h
index b8c49f98a44c..739cfe0499d1 100644
--- a/arch/tile/include/asm/atomic.h
+++ b/arch/tile/include/asm/atomic.h
@@ -32,7 +32,7 @@
  */
 static inline int atomic_read(const atomic_t *v)
 {
-       return v->counter;
+        return ACCESS_ONCE(v->counter);
 }
 
 /**
@@ -130,17 +130,52 @@ static inline int atomic_read(const atomic_t *v)
  */
 #define atomic_inc_not_zero(v)          atomic_add_unless((v), 1, 0)
 
-
-/*
- * We define xchg() and cmpxchg() in the included headers.
- * Note that we do not define __HAVE_ARCH_CMPXCHG, since that would imply
- * that cmpxchg() is an efficient operation, which is not particularly true.
- */
-
 /* Nonexistent functions intended to cause link errors. */
 extern unsigned long __xchg_called_with_bad_pointer(void);
 extern unsigned long __cmpxchg_called_with_bad_pointer(void);
 
+#define xchg(ptr, x)                                                    \
+        ({                                                              \
+                typeof(*(ptr)) __x;                                     \
+                switch (sizeof(*(ptr))) {                               \
+                case 4:                                                 \
+                        __x = (typeof(__x))(typeof(__x-__x))atomic_xchg( \
+                                (atomic_t *)(ptr),                      \
+                                (u32)(typeof((x)-(x)))(x));             \
+                        break;                                          \
+                case 8:                                                 \
+                        __x = (typeof(__x))(typeof(__x-__x))atomic64_xchg( \
+                                (atomic64_t *)(ptr),                    \
+                                (u64)(typeof((x)-(x)))(x));             \
+                        break;                                          \
+                default:                                                \
+                        __xchg_called_with_bad_pointer();               \
+                }                                                       \
+                __x;                                                    \
+        })
+
+#define cmpxchg(ptr, o, n)                                              \
+        ({                                                              \
+                typeof(*(ptr)) __x;                                     \
+                switch (sizeof(*(ptr))) {                               \
+                case 4:                                                 \
+                        __x = (typeof(__x))(typeof(__x-__x))atomic_cmpxchg( \
+                                (atomic_t *)(ptr),                      \
+                                (u32)(typeof((o)-(o)))(o),              \
+                                (u32)(typeof((n)-(n)))(n));             \
+                        break;                                          \
+                case 8:                                                 \
+                        __x = (typeof(__x))(typeof(__x-__x))atomic64_cmpxchg( \
+                                (atomic64_t *)(ptr),                    \
+                                (u64)(typeof((o)-(o)))(o),              \
+                                (u64)(typeof((n)-(n)))(n));             \
+                        break;                                          \
+                default:                                                \
+                        __cmpxchg_called_with_bad_pointer();            \
+                }                                                       \
+                __x;                                                    \
+        })
+
 #define tas(ptr) (xchg((ptr), 1))
 
 #endif /* __ASSEMBLY__ */
diff --git a/arch/tile/include/asm/atomic_32.h b/arch/tile/include/asm/atomic_32.h
index ed359aee8837..92a8bee32311 100644
--- a/arch/tile/include/asm/atomic_32.h
+++ b/arch/tile/include/asm/atomic_32.h
@@ -110,16 +110,6 @@ static inline void atomic_set(atomic_t *v, int n)
         _atomic_xchg(v, n);
 }
 
-#define xchg(ptr, x) ((typeof(*(ptr))) \
-  ((sizeof(*(ptr)) == sizeof(atomic_t)) ? \
-   atomic_xchg((atomic_t *)(ptr), (long)(x)) : \
-   __xchg_called_with_bad_pointer()))
-
-#define cmpxchg(ptr, o, n) ((typeof(*(ptr))) \
-  ((sizeof(*(ptr)) == sizeof(atomic_t)) ? \
-   atomic_cmpxchg((atomic_t *)(ptr), (long)(o), (long)(n)) : \
-   __cmpxchg_called_with_bad_pointer()))
-
 /* A 64bit atomic type */
 
 typedef struct {
diff --git a/arch/tile/include/asm/atomic_64.h b/arch/tile/include/asm/atomic_64.h
new file mode 100644
index 000000000000..1c1e60d8ccb6
--- /dev/null
+++ b/arch/tile/include/asm/atomic_64.h
@@ -0,0 +1,156 @@
+/*
+ * Copyright 2011 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ *
+ * Do not include directly; use <asm/atomic.h>.
+ */
+
+#ifndef _ASM_TILE_ATOMIC_64_H
+#define _ASM_TILE_ATOMIC_64_H
+
+#ifndef __ASSEMBLY__
+
+#include <arch/spr_def.h>
+
+/* First, the 32-bit atomic ops that are "real" on our 64-bit platform. */
+
+#define atomic_set(v, i) ((v)->counter = (i))
+
+/*
+ * The smp_mb() operations throughout are to support the fact that
+ * Linux requires memory barriers before and after the operation,
+ * on any routine which updates memory and returns a value.
+ */
+
+static inline int atomic_cmpxchg(atomic_t *v, int o, int n)
+{
+        int val;
+        __insn_mtspr(SPR_CMPEXCH_VALUE, o);
+        smp_mb();  /* barrier for proper semantics */
+        val = __insn_cmpexch4((void *)&v->counter, n);
+        smp_mb();  /* barrier for proper semantics */
+        return val;
+}
+
+static inline int atomic_xchg(atomic_t *v, int n)
+{
+        int val;
+        smp_mb();  /* barrier for proper semantics */
+        val = __insn_exch4((void *)&v->counter, n);
+        smp_mb();  /* barrier for proper semantics */
+        return val;
+}
+
+static inline void atomic_add(int i, atomic_t *v)
+{
+        __insn_fetchadd4((void *)&v->counter, i);
+}
+
+static inline int atomic_add_return(int i, atomic_t *v)
+{
+        int val;
+        smp_mb();  /* barrier for proper semantics */
+        val = __insn_fetchadd4((void *)&v->counter, i) + i;
+        barrier();  /* the "+ i" above will wait on memory */
+        return val;
+}
+
+static inline int atomic_add_unless(atomic_t *v, int a, int u)
+{
+        int guess, oldval = v->counter;
+        do {
+                if (oldval == u)
+                        break;
+                guess = oldval;
+                oldval = atomic_cmpxchg(v, guess, guess + a);
+        } while (guess != oldval);
+        return oldval != u;
+}
+
+/* Now the true 64-bit operations. */
+
+#define ATOMIC64_INIT(i)        { (i) }
+
+#define atomic64_read(v)                ((v)->counter)
+#define atomic64_set(v, i) ((v)->counter = (i))
+
+static inline long atomic64_cmpxchg(atomic64_t *v, long o, long n)
+{
+        long val;
+        smp_mb();  /* barrier for proper semantics */
+        __insn_mtspr(SPR_CMPEXCH_VALUE, o);
+        val = __insn_cmpexch((void *)&v->counter, n);
+        smp_mb();  /* barrier for proper semantics */
+        return val;
+}
+
+static inline long atomic64_xchg(atomic64_t *v, long n)
+{
+        long val;
+        smp_mb();  /* barrier for proper semantics */
+        val = __insn_exch((void *)&v->counter, n);
+        smp_mb();  /* barrier for proper semantics */
+        return val;
+}
+
+static inline void atomic64_add(long i, atomic64_t *v)
+{
+        __insn_fetchadd((void *)&v->counter, i);
+}
+
+static inline long atomic64_add_return(long i, atomic64_t *v)
+{
+        int val;
+        smp_mb();  /* barrier for proper semantics */
+        val = __insn_fetchadd((void *)&v->counter, i) + i;
+        barrier();  /* the "+ i" above will wait on memory */
+        return val;
+}
+
+static inline long atomic64_add_unless(atomic64_t *v, long a, long u)
+{
+        long guess, oldval = v->counter;
+        do {
+                if (oldval == u)
+                        break;
+                guess = oldval;
+                oldval = atomic64_cmpxchg(v, guess, guess + a);
+        } while (guess != oldval);
+        return oldval != u;
+}
+
+#define atomic64_sub_return(i, v)       atomic64_add_return(-(i), (v))
+#define atomic64_sub(i, v)              atomic64_add(-(i), (v))
+#define atomic64_inc_return(v)          atomic64_add_return(1, (v))
+#define atomic64_dec_return(v)          atomic64_sub_return(1, (v))
+#define atomic64_inc(v)                 atomic64_add(1, (v))
+#define atomic64_dec(v)                 atomic64_sub(1, (v))
+
+#define atomic64_inc_and_test(v)        (atomic64_inc_return(v) == 0)
+#define atomic64_dec_and_test(v)        (atomic64_dec_return(v) == 0)
+#define atomic64_sub_and_test(i, v)     (atomic64_sub_return((i), (v)) == 0)
+#define atomic64_add_negative(i, v)     (atomic64_add_return((i), (v)) < 0)
+
+#define atomic64_inc_not_zero(v)        atomic64_add_unless((v), 1, 0)
+
+/* Atomic dec and inc don't implement barrier, so provide them if needed. */
+#define smp_mb__before_atomic_dec()     smp_mb()
+#define smp_mb__after_atomic_dec()      smp_mb()
+#define smp_mb__before_atomic_inc()     smp_mb()
+#define smp_mb__after_atomic_inc()      smp_mb()
+
+/* Define this to indicate that cmpxchg is an efficient operation. */
+#define __HAVE_ARCH_CMPXCHG
+
+#endif /* !__ASSEMBLY__ */
+
+#endif /* _ASM_TILE_ATOMIC_64_H */
diff --git a/arch/tile/include/asm/backtrace.h b/arch/tile/include/asm/backtrace.h
index 758ca4619d50..bd5399a69edf 100644
--- a/arch/tile/include/asm/backtrace.h
+++ b/arch/tile/include/asm/backtrace.h
@@ -12,80 +12,41 @@
  *   more details.
  */
 
-#ifndef _TILE_BACKTRACE_H
-#define _TILE_BACKTRACE_H
-
-
+#ifndef _ASM_TILE_BACKTRACE_H
+#define _ASM_TILE_BACKTRACE_H
 
 #include <linux/types.h>
 
-#include <arch/chip.h>
-
-#if defined(__tile__)
-typedef unsigned long VirtualAddress;
-#elif CHIP_VA_WIDTH() > 32
-typedef unsigned long long VirtualAddress;
-#else
-typedef unsigned int VirtualAddress;
-#endif
-
-
-/** Reads 'size' bytes from 'address' and writes the data to 'result'.
+/* Reads 'size' bytes from 'address' and writes the data to 'result'.
  * Returns true if successful, else false (e.g. memory not readable).
  */
 typedef bool (*BacktraceMemoryReader)(void *result,
-                                      VirtualAddress address,
+                                      unsigned long address,
                                       unsigned int size,
                                       void *extra);
 
 typedef struct {
-        /** Current PC. */
-        VirtualAddress pc;
+        /* Current PC. */
+        unsigned long pc;
 
-        /** Current stack pointer value. */
-        VirtualAddress sp;
+        /* Current stack pointer value. */
+        unsigned long sp;
 
-        /** Current frame pointer value (i.e. caller's stack pointer) */
-        VirtualAddress fp;
+        /* Current frame pointer value (i.e. caller's stack pointer) */
+        unsigned long fp;
 
-        /** Internal use only: caller's PC for first frame. */
-        VirtualAddress initial_frame_caller_pc;
+        /* Internal use only: caller's PC for first frame. */
+        unsigned long initial_frame_caller_pc;
 
-        /** Internal use only: callback to read memory. */
+        /* Internal use only: callback to read memory. */
         BacktraceMemoryReader read_memory_func;
 
-        /** Internal use only: arbitrary argument to read_memory_func. */
+        /* Internal use only: arbitrary argument to read_memory_func. */
         void *read_memory_func_extra;
 
 } BacktraceIterator;
 
 
-/** Initializes a backtracer to start from the given location.
- *
- * If the frame pointer cannot be determined it is set to -1.
- *
- * @param state The state to be filled in.
- * @param read_memory_func A callback that reads memory. If NULL, a default
- *        value is provided.
- * @param read_memory_func_extra An arbitrary argument to read_memory_func.
- * @param pc The current PC.
- * @param lr The current value of the 'lr' register.
- * @param sp The current value of the 'sp' register.
- * @param r52 The current value of the 'r52' register.
- */
-extern void backtrace_init(BacktraceIterator *state,
-                           BacktraceMemoryReader read_memory_func,
-                           void *read_memory_func_extra,
-                           VirtualAddress pc, VirtualAddress lr,
-                           VirtualAddress sp, VirtualAddress r52);
-
-
-/** Advances the backtracing state to the calling frame, returning
- * true iff successful.
- */
-extern bool backtrace_next(BacktraceIterator *state);
-
-
 typedef enum {
 
         /* We have no idea what the caller's pc is. */
@@ -138,7 +99,7 @@ enum {
 };
 
 
-/** Internal constants used to define 'info' operands. */
+/* Internal constants used to define 'info' operands. */
 enum {
         /* 0 and 1 are reserved, as are all negative numbers. */
 
@@ -146,11 +107,11 @@ enum {
 
         CALLER_SP_IN_R52_BASE = 4,
 
-        CALLER_SP_OFFSET_BASE = 8
+        CALLER_SP_OFFSET_BASE = 8,
 };
 
 
-/** Current backtracer state describing where it thinks the caller is. */
+/* Current backtracer state describing where it thinks the caller is. */
 typedef struct {
         /*
          * Public fields
@@ -189,7 +150,13 @@ typedef struct {
 
 } CallerLocation;
 
+extern void backtrace_init(BacktraceIterator *state,
+                          BacktraceMemoryReader read_memory_func,
+                          void *read_memory_func_extra,
+                          unsigned long pc, unsigned long lr,
+                          unsigned long sp, unsigned long r52);
 
 
+extern bool backtrace_next(BacktraceIterator *state);
 
-#endif /* _TILE_BACKTRACE_H */
+#endif /* _ASM_TILE_BACKTRACE_H */
diff --git a/arch/tile/include/asm/bitops.h b/arch/tile/include/asm/bitops.h
index 6832b4be8990..16f1fa51fea1 100644
--- a/arch/tile/include/asm/bitops.h
+++ b/arch/tile/include/asm/bitops.h
@@ -120,8 +120,9 @@ static inline unsigned long __arch_hweight64(__u64 w)
 
 #include <asm-generic/bitops/const_hweight.h>
 #include <asm-generic/bitops/lock.h>
+#include <asm-generic/bitops/find.h>
 #include <asm-generic/bitops/sched.h>
-#include <asm-generic/bitops/ext2-non-atomic.h>
-#include <asm-generic/bitops/minix.h>
+#include <asm-generic/bitops/non-atomic.h>
+#include <asm-generic/bitops/le.h>
 
 #endif /* _ASM_TILE_BITOPS_H */
diff --git a/arch/tile/include/asm/bitops_32.h b/arch/tile/include/asm/bitops_32.h
index 7a93c001ac19..d31ab905cfa7 100644
--- a/arch/tile/include/asm/bitops_32.h
+++ b/arch/tile/include/asm/bitops_32.h
@@ -122,11 +122,10 @@ static inline int test_and_change_bit(unsigned nr,
         return (_atomic_xor(addr, mask) & mask) != 0;
 }
 
-/* See discussion at smp_mb__before_atomic_dec() in <asm/atomic.h>. */
+/* See discussion at smp_mb__before_atomic_dec() in <asm/atomic_32.h>. */
 #define smp_mb__before_clear_bit()      smp_mb()
 #define smp_mb__after_clear_bit()       do {} while (0)
 
-#include <asm-generic/bitops/non-atomic.h>
 #include <asm-generic/bitops/ext2-atomic.h>
 
 #endif /* _ASM_TILE_BITOPS_32_H */
diff --git a/arch/tile/include/asm/bitops_64.h b/arch/tile/include/asm/bitops_64.h
new file mode 100644
index 000000000000..99615e8d2d8b
--- /dev/null
+++ b/arch/tile/include/asm/bitops_64.h
@@ -0,0 +1,105 @@
+/*
+ * Copyright 2011 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#ifndef _ASM_TILE_BITOPS_64_H
+#define _ASM_TILE_BITOPS_64_H
+
+#include <linux/compiler.h>
+#include <asm/atomic.h>
+#include <asm/system.h>
+
+/* See <asm/bitops.h> for API comments. */
+
+static inline void set_bit(unsigned nr, volatile unsigned long *addr)
+{
+        unsigned long mask = (1UL << (nr % BITS_PER_LONG));
+        __insn_fetchor((void *)(addr + nr / BITS_PER_LONG), mask);
+}
+
+static inline void clear_bit(unsigned nr, volatile unsigned long *addr)
+{
+        unsigned long mask = (1UL << (nr % BITS_PER_LONG));
+        __insn_fetchand((void *)(addr + nr / BITS_PER_LONG), ~mask);
+}
+
+#define smp_mb__before_clear_bit()      smp_mb()
+#define smp_mb__after_clear_bit()       smp_mb()
+
+
+static inline void change_bit(unsigned nr, volatile unsigned long *addr)
+{
+        unsigned long old, mask = (1UL << (nr % BITS_PER_LONG));
+        long guess, oldval;
+        addr += nr / BITS_PER_LONG;
+        old = *addr;
+        do {
+                guess = oldval;
+                oldval = atomic64_cmpxchg((atomic64_t *)addr,
+                                          guess, guess ^ mask);
+        } while (guess != oldval);
+}
+
+
+/*
+ * The test_and_xxx_bit() routines require a memory fence before we
+ * start the operation, and after the operation completes.  We use
+ * smp_mb() before, and rely on the "!= 0" comparison, plus a compiler
+ * barrier(), to block until the atomic op is complete.
+ */
+
+static inline int test_and_set_bit(unsigned nr, volatile unsigned long *addr)
+{
+        int val;
+        unsigned long mask = (1UL << (nr % BITS_PER_LONG));
+        smp_mb();  /* barrier for proper semantics */
+        val = (__insn_fetchor((void *)(addr + nr / BITS_PER_LONG), mask)
+               & mask) != 0;
+        barrier();
+        return val;
+}
+
+
+static inline int test_and_clear_bit(unsigned nr, volatile unsigned long *addr)
+{
+        int val;
+        unsigned long mask = (1UL << (nr % BITS_PER_LONG));
+        smp_mb();  /* barrier for proper semantics */
+        val = (__insn_fetchand((void *)(addr + nr / BITS_PER_LONG), ~mask)
+               & mask) != 0;
+        barrier();
+        return val;
+}
+
+
+static inline int test_and_change_bit(unsigned nr,
+                                      volatile unsigned long *addr)
+{
+        unsigned long mask = (1UL << (nr % BITS_PER_LONG));
+        long guess, oldval = *addr;
+        addr += nr / BITS_PER_LONG;
+        oldval = *addr;
+        do {
+                guess = oldval;
+                oldval = atomic64_cmpxchg((atomic64_t *)addr,
+                                          guess, guess ^ mask);
+        } while (guess != oldval);
+        return (oldval & mask) != 0;
+}
+
+#define ext2_set_bit_atomic(lock, nr, addr)                     \
+        test_and_set_bit((nr), (unsigned long *)(addr))
+#define ext2_clear_bit_atomic(lock, nr, addr)                   \
+        test_and_clear_bit((nr), (unsigned long *)(addr))
+
+#endif /* _ASM_TILE_BITOPS_64_H */
diff --git a/arch/tile/include/asm/cache.h b/arch/tile/include/asm/cache.h
index 08a2815b5e4e..392e5333dd8b 100644
--- a/arch/tile/include/asm/cache.h
+++ b/arch/tile/include/asm/cache.h
@@ -40,7 +40,7 @@
 #define INTERNODE_CACHE_BYTES   L2_CACHE_BYTES
 
 /* Group together read-mostly things to avoid cache false sharing */
-#define __read_mostly __attribute__((__section__(".data.read_mostly")))
+#define __read_mostly __attribute__((__section__(".data..read_mostly")))
 
 /*
  * Attribute for data that is kept read/write coherent until the end of
diff --git a/arch/tile/include/asm/cacheflush.h b/arch/tile/include/asm/cacheflush.h
index c5741da4eeac..e925f4bb498f 100644
--- a/arch/tile/include/asm/cacheflush.h
+++ b/arch/tile/include/asm/cacheflush.h
@@ -116,25 +116,40 @@ static inline void __finv_buffer(void *buffer, size_t size)
 }
 
 
-/* Invalidate a VA range, then memory fence. */
+/* Invalidate a VA range and wait for it to be complete. */
 static inline void inv_buffer(void *buffer, size_t size)
 {
         __inv_buffer(buffer, size);
-        mb_incoherent();
+        mb();
 }
 
-/* Flush a VA range, then memory fence. */
-static inline void flush_buffer(void *buffer, size_t size)
+/*
+ * Flush a locally-homecached VA range and wait for the evicted
+ * cachelines to hit memory.
+ */
+static inline void flush_buffer_local(void *buffer, size_t size)
 {
         __flush_buffer(buffer, size);
         mb_incoherent();
 }
 
-/* Flush & invalidate a VA range, then memory fence. */
-static inline void finv_buffer(void *buffer, size_t size)
+/*
+ * Flush and invalidate a locally-homecached VA range and wait for the
+ * evicted cachelines to hit memory.
+ */
+static inline void finv_buffer_local(void *buffer, size_t size)
 {
         __finv_buffer(buffer, size);
         mb_incoherent();
 }
 
+/*
+ * Flush and invalidate a VA range that is homed remotely, waiting
+ * until the memory controller holds the flushed values.  If "hfh" is
+ * true, we will do a more expensive flush involving additional loads
+ * to make sure we have touched all the possible home cpus of a buffer
+ * that is homed with "hash for home".
+ */
+void finv_buffer_remote(void *buffer, size_t size, int hfh);
+
 #endif /* _ASM_TILE_CACHEFLUSH_H */
diff --git a/arch/tile/include/asm/compat.h b/arch/tile/include/asm/compat.h
index 8b60ec8b2d19..bf95f55b82b0 100644
--- a/arch/tile/include/asm/compat.h
+++ b/arch/tile/include/asm/compat.h
@@ -215,16 +215,17 @@ struct compat_sigaction;
 struct compat_siginfo;
 struct compat_sigaltstack;
 long compat_sys_execve(const char __user *path,
-                       const compat_uptr_t __user *argv,
-                       const compat_uptr_t __user *envp);
+                       compat_uptr_t __user *argv,
+                       compat_uptr_t __user *envp, struct pt_regs *);
 long compat_sys_rt_sigaction(int sig, struct compat_sigaction __user *act,
                              struct compat_sigaction __user *oact,
                              size_t sigsetsize);
 long compat_sys_rt_sigqueueinfo(int pid, int sig,
                                 struct compat_siginfo __user *uinfo);
-long compat_sys_rt_sigreturn(void);
+long compat_sys_rt_sigreturn(struct pt_regs *);
 long compat_sys_sigaltstack(const struct compat_sigaltstack __user *uss_ptr,
-                            struct compat_sigaltstack __user *uoss_ptr);
+                            struct compat_sigaltstack __user *uoss_ptr,
+                            struct pt_regs *);
 long compat_sys_truncate64(char __user *filename, u32 dummy, u32 low, u32 high);
 long compat_sys_ftruncate64(unsigned int fd, u32 dummy, u32 low, u32 high);
 long compat_sys_pread64(unsigned int fd, char __user *ubuf, size_t count,
@@ -255,4 +256,12 @@ long tile_compat_sys_ptrace(compat_long_t request, compat_long_t pid,
 /* Tilera Linux syscalls that don't have "compat" versions. */
 #define compat_sys_flush_cache sys_flush_cache
 
+/* These are the intvec_64.S trampolines. */
+long _compat_sys_execve(const char __user *path,
+                        const compat_uptr_t __user *argv,
+                        const compat_uptr_t __user *envp);
+long _compat_sys_sigaltstack(const struct compat_sigaltstack __user *uss_ptr,
+                            struct compat_sigaltstack __user *uoss_ptr);
+long _compat_sys_rt_sigreturn(void);
+
 #endif /* _ASM_TILE_COMPAT_H */
diff --git a/arch/tile/include/asm/dma-mapping.h b/arch/tile/include/asm/dma-mapping.h
index 15e1dceecc64..eaa06d175b39 100644
--- a/arch/tile/include/asm/dma-mapping.h
+++ b/arch/tile/include/asm/dma-mapping.h
@@ -65,7 +65,8 @@ extern void dma_sync_single_range_for_cpu(struct device *, dma_addr_t,
 extern void dma_sync_single_range_for_device(struct device *, dma_addr_t,
                                              unsigned long offset, size_t,
                                              enum dma_data_direction);
-extern void dma_cache_sync(void *vaddr, size_t, enum dma_data_direction);
+extern void dma_cache_sync(struct device *dev, void *vaddr, size_t,
+                           enum dma_data_direction);
 
 static inline int
 dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
diff --git a/arch/tile/include/asm/edac.h b/arch/tile/include/asm/edac.h
new file mode 100644
index 000000000000..87fc83eeaffd
--- /dev/null
+++ b/arch/tile/include/asm/edac.h
@@ -0,0 +1,29 @@
+/*
+ * Copyright 2011 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#ifndef _ASM_TILE_EDAC_H
+#define _ASM_TILE_EDAC_H
+
+/* ECC atomic, DMA, SMP and interrupt safe scrub function */
+
+static inline void atomic_scrub(void *va, u32 size)
+{
+        /*
+         * These is nothing to be done here because CE is
+         * corrected by the mshim.
+         */
+        return;
+}
+
+#endif /* _ASM_TILE_EDAC_H */
diff --git a/arch/tile/include/asm/fb.h b/arch/tile/include/asm/fb.h
new file mode 100644
index 000000000000..3a4988e8df45
--- /dev/null
+++ b/arch/tile/include/asm/fb.h
@@ -0,0 +1 @@
+#include <asm-generic/fb.h>
diff --git a/arch/tile/include/asm/futex.h b/arch/tile/include/asm/futex.h
index fe0d10dcae57..d03ec124a598 100644
--- a/arch/tile/include/asm/futex.h
+++ b/arch/tile/include/asm/futex.h
@@ -29,16 +29,16 @@
 #include <linux/uaccess.h>
 #include <linux/errno.h>
 
-extern struct __get_user futex_set(int __user *v, int i);
-extern struct __get_user futex_add(int __user *v, int n);
-extern struct __get_user futex_or(int __user *v, int n);
-extern struct __get_user futex_andn(int __user *v, int n);
-extern struct __get_user futex_cmpxchg(int __user *v, int o, int n);
+extern struct __get_user futex_set(u32 __user *v, int i);
+extern struct __get_user futex_add(u32 __user *v, int n);
+extern struct __get_user futex_or(u32 __user *v, int n);
+extern struct __get_user futex_andn(u32 __user *v, int n);
+extern struct __get_user futex_cmpxchg(u32 __user *v, int o, int n);
 
 #ifndef __tilegx__
-extern struct __get_user futex_xor(int __user *v, int n);
+extern struct __get_user futex_xor(u32 __user *v, int n);
 #else
-static inline struct __get_user futex_xor(int __user *uaddr, int n)
+static inline struct __get_user futex_xor(u32 __user *uaddr, int n)
 {
         struct __get_user asm_ret = __get_user_4(uaddr);
         if (!asm_ret.err) {
@@ -53,7 +53,7 @@ static inline struct __get_user futex_xor(int __user *uaddr, int n)
 }
 #endif
 
-static inline int futex_atomic_op_inuser(int encoded_op, int __user *uaddr)
+static inline int futex_atomic_op_inuser(int encoded_op, u32 __user *uaddr)
 {
         int op = (encoded_op >> 28) & 7;
         int cmp = (encoded_op >> 24) & 15;
@@ -65,7 +65,7 @@ static inline int futex_atomic_op_inuser(int encoded_op, int __user *uaddr)
         if (encoded_op & (FUTEX_OP_OPARG_SHIFT << 28))
                 oparg = 1 << oparg;
 
-        if (!access_ok(VERIFY_WRITE, uaddr, sizeof(int)))
+        if (!access_ok(VERIFY_WRITE, uaddr, sizeof(u32)))
                 return -EFAULT;
 
         pagefault_disable();
@@ -119,16 +119,17 @@ static inline int futex_atomic_op_inuser(int encoded_op, int __user *uaddr)
         return ret;
 }
 
-static inline int futex_atomic_cmpxchg_inatomic(int __user *uaddr, int oldval,
-                                                int newval)
+static inline int futex_atomic_cmpxchg_inatomic(u32 *uval, u32 __user *uaddr,
+                                                u32 oldval, u32 newval)
 {
         struct __get_user asm_ret;
 
-        if (!access_ok(VERIFY_WRITE, uaddr, sizeof(int)))
+        if (!access_ok(VERIFY_WRITE, uaddr, sizeof(u32)))
                 return -EFAULT;
 
         asm_ret = futex_cmpxchg(uaddr, oldval, newval);
-        return asm_ret.err ? asm_ret.err : asm_ret.val;
+        *uval = asm_ret.val;
+        return asm_ret.err;
 }
 
 #ifndef __tilegx__
diff --git a/arch/tile/include/asm/hardwall.h b/arch/tile/include/asm/hardwall.h
index 0bed3ec7b42c..2ac422848c7d 100644
--- a/arch/tile/include/asm/hardwall.h
+++ b/arch/tile/include/asm/hardwall.h
@@ -40,6 +40,10 @@
 #define HARDWALL_DEACTIVATE \
  _IO(HARDWALL_IOCTL_BASE, _HARDWALL_DEACTIVATE)
 
+#define _HARDWALL_GET_ID 4
+#define HARDWALL_GET_ID \
+ _IO(HARDWALL_IOCTL_BASE, _HARDWALL_GET_ID)
+
 #ifndef __KERNEL__
 
 /* This is the canonical name expected by userspace. */
@@ -47,9 +51,14 @@
 
 #else
 
-/* Hook for /proc/tile/hardwall. */
-struct seq_file;
-int proc_tile_hardwall_show(struct seq_file *sf, void *v);
+/* /proc hooks for hardwall. */
+struct proc_dir_entry;
+#ifdef CONFIG_HARDWALL
+void proc_tile_hardwall_init(struct proc_dir_entry *root);
+int proc_pid_hardwall(struct task_struct *task, char *buffer);
+#else
+static inline void proc_tile_hardwall_init(struct proc_dir_entry *root) {}
+#endif
 
 #endif
 
diff --git a/arch/tile/include/asm/highmem.h b/arch/tile/include/asm/highmem.h
index d155db6fa9bd..b2a6c5de79ab 100644
--- a/arch/tile/include/asm/highmem.h
+++ b/arch/tile/include/asm/highmem.h
@@ -23,7 +23,6 @@
 
 #include <linux/interrupt.h>
 #include <linux/threads.h>
-#include <asm/kmap_types.h>
 #include <asm/tlbflush.h>
 #include <asm/homecache.h>
 
@@ -60,12 +59,12 @@ void *kmap_fix_kpte(struct page *page, int finished);
 /* This macro is used only in map_new_virtual() to map "page". */
 #define kmap_prot page_to_kpgprot(page)
 
-void kunmap_atomic_notypecheck(void *kvaddr, enum km_type type);
-void *kmap_atomic_pfn(unsigned long pfn, enum km_type type);
-void *kmap_atomic_prot_pfn(unsigned long pfn, enum km_type type, pgprot_t prot);
+void *__kmap_atomic(struct page *page);
+void __kunmap_atomic(void *kvaddr);
+void *kmap_atomic_pfn(unsigned long pfn);
+void *kmap_atomic_prot_pfn(unsigned long pfn, pgprot_t prot);
 struct page *kmap_atomic_to_page(void *ptr);
-void *kmap_atomic_prot(struct page *page, enum km_type type, pgprot_t prot);
-void *kmap_atomic(struct page *page, enum km_type type);
+void *kmap_atomic_prot(struct page *page, pgprot_t prot);
 void kmap_atomic_fix_kpte(struct page *page, int finished);
 
 #define flush_cache_kmaps()     do { } while (0)
diff --git a/arch/tile/include/asm/hugetlb.h b/arch/tile/include/asm/hugetlb.h
index 0521c277bbde..d396d1805163 100644
--- a/arch/tile/include/asm/hugetlb.h
+++ b/arch/tile/include/asm/hugetlb.h
@@ -54,7 +54,7 @@ static inline void hugetlb_free_pgd_range(struct mmu_gather *tlb,
 static inline void set_huge_pte_at(struct mm_struct *mm, unsigned long addr,
                                    pte_t *ptep, pte_t pte)
 {
-        set_pte_order(ptep, pte, HUGETLB_PAGE_ORDER);
+        set_pte(ptep, pte);
 }
 
 static inline pte_t huge_ptep_get_and_clear(struct mm_struct *mm,
diff --git a/arch/tile/include/asm/io.h b/arch/tile/include/asm/io.h
index ee43328713ab..c9ea1652af03 100644
--- a/arch/tile/include/asm/io.h
+++ b/arch/tile/include/asm/io.h
@@ -52,12 +52,10 @@ extern void iounmap(volatile void __iomem *addr);
 #endif
 
 #define ioremap_nocache(physaddr, size)         ioremap(physaddr, size)
+#define ioremap_wc(physaddr, size)              ioremap(physaddr, size)
 #define ioremap_writethrough(physaddr, size)    ioremap(physaddr, size)
 #define ioremap_fullcache(physaddr, size)       ioremap(physaddr, size)
 
-void __iomem *ioport_map(unsigned long port, unsigned int len);
-extern inline void ioport_unmap(void __iomem *addr) {}
-
 #define mmiowb()
 
 /* Conversion between virtual and physical mappings.  */
@@ -164,6 +162,15 @@ static inline void _tile_writeq(u64 val, unsigned long addr)
 #define iowrite32 writel
 #define iowrite64 writeq
 
+static inline void memset_io(void *dst, int val, size_t len)
+{
+        int x;
+        BUG_ON((unsigned long)dst & 0x3);
+        val = (val & 0xff) * 0x01010101;
+        for (x = 0; x < len; x += 4)
+                writel(val, dst + x);
+}
+
 static inline void memcpy_fromio(void *dst, const volatile void __iomem *src,
                                  size_t len)
 {
@@ -189,12 +196,22 @@ static inline void memcpy_toio(volatile void __iomem *dst, const void *src,
  * we never run, uses them unconditionally.
  */
 
-static inline int ioport_panic(void)
+static inline long ioport_panic(void)
 {
         panic("inb/outb and friends do not exist on tile");
         return 0;
 }
 
+static inline void __iomem *ioport_map(unsigned long port, unsigned int len)
+{
+        return (void __iomem *) ioport_panic();
+}
+
+static inline void ioport_unmap(void __iomem *addr)
+{
+        ioport_panic();
+}
+
 static inline u8 inb(unsigned long addr)
 {
         return ioport_panic();
@@ -262,6 +279,11 @@ static inline void outsl(unsigned long addr, const void *buffer, int count)
         ioport_panic();
 }
 
+#define ioread16be(addr)        be16_to_cpu(ioread16(addr))
+#define ioread32be(addr)        be32_to_cpu(ioread32(addr))
+#define iowrite16be(v, addr)    iowrite16(be16_to_cpu(v), (addr))
+#define iowrite32be(v, addr)    iowrite32(be32_to_cpu(v), (addr))
+
 #define ioread8_rep(p, dst, count) \
         insb((unsigned long) (p), (dst), (count))
 #define ioread16_rep(p, dst, count) \
@@ -276,4 +298,7 @@ static inline void outsl(unsigned long addr, const void *buffer, int count)
 #define iowrite32_rep(p, src, count) \
         outsl((unsigned long) (p), (src), (count))
 
+#define virt_to_bus     virt_to_phys
+#define bus_to_virt     phys_to_virt
+
 #endif /* _ASM_TILE_IO_H */
diff --git a/arch/tile/include/asm/irq.h b/arch/tile/include/asm/irq.h
index 572fd3ef1d73..94e9a511de84 100644
--- a/arch/tile/include/asm/irq.h
+++ b/arch/tile/include/asm/irq.h
@@ -23,6 +23,8 @@
 /* IRQ numbers used for linux IPIs. */
 #define IRQ_RESCHEDULE 1
 
+#define irq_canonicalize(irq)   (irq)
+
 void ack_bad_irq(unsigned int irq);
 
 /*
diff --git a/arch/tile/include/asm/irqflags.h b/arch/tile/include/asm/irqflags.h
index 45cf67c2f286..5db0ce54284d 100644
--- a/arch/tile/include/asm/irqflags.h
+++ b/arch/tile/include/asm/irqflags.h
@@ -18,6 +18,8 @@
 #include <arch/interrupts.h>
 #include <arch/chip.h>
 
+#if !defined(__tilegx__) && defined(__ASSEMBLY__)
+
 /*
  * The set of interrupts we want to allow when interrupts are nominally
  * disabled.  The remainder are effectively "NMI" interrupts from
@@ -25,6 +27,16 @@
  * interrupts (aka "non-queued") are not blocked by the mask in any case.
  */
 #if CHIP_HAS_AUX_PERF_COUNTERS()
+#define LINUX_MASKABLE_INTERRUPTS_HI \
+       (~(INT_MASK_HI(INT_PERF_COUNT) | INT_MASK_HI(INT_AUX_PERF_COUNT)))
+#else
+#define LINUX_MASKABLE_INTERRUPTS_HI \
+       (~(INT_MASK_HI(INT_PERF_COUNT)))
+#endif
+
+#else
+
+#if CHIP_HAS_AUX_PERF_COUNTERS()
 #define LINUX_MASKABLE_INTERRUPTS \
         (~(INT_MASK(INT_PERF_COUNT) | INT_MASK(INT_AUX_PERF_COUNT)))
 #else
@@ -32,6 +44,8 @@
         (~(INT_MASK(INT_PERF_COUNT)))
 #endif
 
+#endif
+
 #ifndef __ASSEMBLY__
 
 /* NOTE: we can't include <linux/percpu.h> due to #include dependencies. */
@@ -47,53 +61,53 @@
         int __n = (n); \
         int __mask = 1 << (__n & 0x1f); \
         if (__n < 32) \
-                __insn_mtspr(SPR_INTERRUPT_MASK_SET_1_0, __mask); \
+                __insn_mtspr(SPR_INTERRUPT_MASK_SET_K_0, __mask); \
         else \
-                __insn_mtspr(SPR_INTERRUPT_MASK_SET_1_1, __mask); \
+                __insn_mtspr(SPR_INTERRUPT_MASK_SET_K_1, __mask); \
 } while (0)
 #define interrupt_mask_reset(n) do { \
         int __n = (n); \
         int __mask = 1 << (__n & 0x1f); \
         if (__n < 32) \
-                __insn_mtspr(SPR_INTERRUPT_MASK_RESET_1_0, __mask); \
+                __insn_mtspr(SPR_INTERRUPT_MASK_RESET_K_0, __mask); \
         else \
-                __insn_mtspr(SPR_INTERRUPT_MASK_RESET_1_1, __mask); \
+                __insn_mtspr(SPR_INTERRUPT_MASK_RESET_K_1, __mask); \
 } while (0)
 #define interrupt_mask_check(n) ({ \
         int __n = (n); \
         (((__n < 32) ? \
-         __insn_mfspr(SPR_INTERRUPT_MASK_1_0) : \
-         __insn_mfspr(SPR_INTERRUPT_MASK_1_1)) \
+         __insn_mfspr(SPR_INTERRUPT_MASK_K_0) : \
+         __insn_mfspr(SPR_INTERRUPT_MASK_K_1)) \
           >> (__n & 0x1f)) & 1; \
 })
 #define interrupt_mask_set_mask(mask) do { \
         unsigned long long __m = (mask); \
-        __insn_mtspr(SPR_INTERRUPT_MASK_SET_1_0, (unsigned long)(__m)); \
-        __insn_mtspr(SPR_INTERRUPT_MASK_SET_1_1, (unsigned long)(__m>>32)); \
+        __insn_mtspr(SPR_INTERRUPT_MASK_SET_K_0, (unsigned long)(__m)); \
+        __insn_mtspr(SPR_INTERRUPT_MASK_SET_K_1, (unsigned long)(__m>>32)); \
 } while (0)
 #define interrupt_mask_reset_mask(mask) do { \
         unsigned long long __m = (mask); \
-        __insn_mtspr(SPR_INTERRUPT_MASK_RESET_1_0, (unsigned long)(__m)); \
-        __insn_mtspr(SPR_INTERRUPT_MASK_RESET_1_1, (unsigned long)(__m>>32)); \
+        __insn_mtspr(SPR_INTERRUPT_MASK_RESET_K_0, (unsigned long)(__m)); \
+        __insn_mtspr(SPR_INTERRUPT_MASK_RESET_K_1, (unsigned long)(__m>>32)); \
 } while (0)
 #else
 #define interrupt_mask_set(n) \
-        __insn_mtspr(SPR_INTERRUPT_MASK_SET_1, (1UL << (n)))
+        __insn_mtspr(SPR_INTERRUPT_MASK_SET_K, (1UL << (n)))
 #define interrupt_mask_reset(n) \
-        __insn_mtspr(SPR_INTERRUPT_MASK_RESET_1, (1UL << (n)))
+        __insn_mtspr(SPR_INTERRUPT_MASK_RESET_K, (1UL << (n)))
 #define interrupt_mask_check(n) \
-        ((__insn_mfspr(SPR_INTERRUPT_MASK_1) >> (n)) & 1)
+        ((__insn_mfspr(SPR_INTERRUPT_MASK_K) >> (n)) & 1)
 #define interrupt_mask_set_mask(mask) \
-        __insn_mtspr(SPR_INTERRUPT_MASK_SET_1, (mask))
+        __insn_mtspr(SPR_INTERRUPT_MASK_SET_K, (mask))
 #define interrupt_mask_reset_mask(mask) \
-        __insn_mtspr(SPR_INTERRUPT_MASK_RESET_1, (mask))
+        __insn_mtspr(SPR_INTERRUPT_MASK_RESET_K, (mask))
 #endif
 
 /*
  * The set of interrupts we want active if irqs are enabled.
  * Note that in particular, the tile timer interrupt comes and goes
  * from this set, since we have no other way to turn off the timer.
- * Likewise, INTCTRL_1 is removed and re-added during device
+ * Likewise, INTCTRL_K is removed and re-added during device
  * interrupts, as is the the hardwall UDN_FIREWALL interrupt.
  * We use a low bit (MEM_ERROR) as our sentinel value and make sure it
  * is always claimed as an "active interrupt" so we can query that bit
@@ -103,55 +117,57 @@ DECLARE_PER_CPU(unsigned long long, interrupts_enabled_mask);
 #define INITIAL_INTERRUPTS_ENABLED INT_MASK(INT_MEM_ERROR)
 
 /* Disable interrupts. */
-#define raw_local_irq_disable() \
+#define arch_local_irq_disable() \
         interrupt_mask_set_mask(LINUX_MASKABLE_INTERRUPTS)
 
 /* Disable all interrupts, including NMIs. */
-#define raw_local_irq_disable_all() \
+#define arch_local_irq_disable_all() \
         interrupt_mask_set_mask(-1UL)
 
 /* Re-enable all maskable interrupts. */
-#define raw_local_irq_enable() \
+#define arch_local_irq_enable() \
         interrupt_mask_reset_mask(__get_cpu_var(interrupts_enabled_mask))
 
 /* Disable or enable interrupts based on flag argument. */
-#define raw_local_irq_restore(disabled) do { \
+#define arch_local_irq_restore(disabled) do { \
         if (disabled) \
-                raw_local_irq_disable(); \
+                arch_local_irq_disable(); \
         else \
-                raw_local_irq_enable(); \
+                arch_local_irq_enable(); \
 } while (0)
 
 /* Return true if "flags" argument means interrupts are disabled. */
-#define raw_irqs_disabled_flags(flags) ((flags) != 0)
+#define arch_irqs_disabled_flags(flags) ((flags) != 0)
 
 /* Return true if interrupts are currently disabled. */
-#define raw_irqs_disabled() interrupt_mask_check(INT_MEM_ERROR)
+#define arch_irqs_disabled() interrupt_mask_check(INT_MEM_ERROR)
 
 /* Save whether interrupts are currently disabled. */
-#define raw_local_save_flags(flags) ((flags) = raw_irqs_disabled())
+#define arch_local_save_flags() arch_irqs_disabled()
 
 /* Save whether interrupts are currently disabled, then disable them. */
-#define raw_local_irq_save(flags) \
-        do { raw_local_save_flags(flags); raw_local_irq_disable(); } while (0)
+#define arch_local_irq_save() ({ \
+        unsigned long __flags = arch_local_save_flags(); \
+        arch_local_irq_disable(); \
+        __flags; })
 
 /* Prevent the given interrupt from being enabled next time we enable irqs. */
-#define raw_local_irq_mask(interrupt) \
+#define arch_local_irq_mask(interrupt) \
         (__get_cpu_var(interrupts_enabled_mask) &= ~INT_MASK(interrupt))
 
 /* Prevent the given interrupt from being enabled immediately. */
-#define raw_local_irq_mask_now(interrupt) do { \
-        raw_local_irq_mask(interrupt); \
+#define arch_local_irq_mask_now(interrupt) do { \
+        arch_local_irq_mask(interrupt); \
         interrupt_mask_set(interrupt); \
 } while (0)
 
 /* Allow the given interrupt to be enabled next time we enable irqs. */
-#define raw_local_irq_unmask(interrupt) \
+#define arch_local_irq_unmask(interrupt) \
         (__get_cpu_var(interrupts_enabled_mask) |= INT_MASK(interrupt))
 
 /* Allow the given interrupt to be enabled immediately, if !irqs_disabled. */
-#define raw_local_irq_unmask_now(interrupt) do { \
-        raw_local_irq_unmask(interrupt); \
+#define arch_local_irq_unmask_now(interrupt) do { \
+        arch_local_irq_unmask(interrupt); \
         if (!irqs_disabled()) \
                 interrupt_mask_reset(interrupt); \
 } while (0)
@@ -168,14 +184,14 @@ DECLARE_PER_CPU(unsigned long long, interrupts_enabled_mask);
 
 /* Return 0 or 1 to indicate whether interrupts are currently disabled. */
 #define IRQS_DISABLED(tmp)                                      \
-        mfspr   tmp, INTERRUPT_MASK_1;                          \
+        mfspr   tmp, SPR_INTERRUPT_MASK_K;                      \
         andi    tmp, tmp, 1
 
 /* Load up a pointer to &interrupts_enabled_mask. */
 #define GET_INTERRUPTS_ENABLED_MASK_PTR(reg)                    \
-        moveli reg, hw2_last(interrupts_enabled_mask); \
-        shl16insli reg, reg, hw1(interrupts_enabled_mask); \
-        shl16insli reg, reg, hw0(interrupts_enabled_mask); \
+        moveli reg, hw2_last(interrupts_enabled_mask);          \
+        shl16insli reg, reg, hw1(interrupts_enabled_mask);      \
+        shl16insli reg, reg, hw0(interrupts_enabled_mask);      \
         add     reg, reg, tp
 
 /* Disable interrupts. */
@@ -183,18 +199,18 @@ DECLARE_PER_CPU(unsigned long long, interrupts_enabled_mask);
         moveli  tmp0, hw2_last(LINUX_MASKABLE_INTERRUPTS);      \
         shl16insli tmp0, tmp0, hw1(LINUX_MASKABLE_INTERRUPTS);  \
         shl16insli tmp0, tmp0, hw0(LINUX_MASKABLE_INTERRUPTS);  \
-        mtspr   INTERRUPT_MASK_SET_1, tmp0
+        mtspr   SPR_INTERRUPT_MASK_SET_K, tmp0
 
 /* Disable ALL synchronous interrupts (used by NMI entry). */
 #define IRQ_DISABLE_ALL(tmp)                                    \
         movei   tmp, -1;                                        \
-        mtspr   INTERRUPT_MASK_SET_1, tmp
+        mtspr   SPR_INTERRUPT_MASK_SET_K, tmp
 
 /* Enable interrupts. */
 #define IRQ_ENABLE(tmp0, tmp1)                                  \
         GET_INTERRUPTS_ENABLED_MASK_PTR(tmp0);                  \
         ld      tmp0, tmp0;                                     \
-        mtspr   INTERRUPT_MASK_RESET_1, tmp0
+        mtspr   SPR_INTERRUPT_MASK_RESET_K, tmp0
 
 #else /* !__tilegx__ */
 
@@ -208,33 +224,33 @@ DECLARE_PER_CPU(unsigned long long, interrupts_enabled_mask);
  * (making the original code's write of the "high" mask word idempotent).
  */
 #define IRQS_DISABLED(tmp)                                      \
-        mfspr   tmp, INTERRUPT_MASK_1_0;                        \
+        mfspr   tmp, SPR_INTERRUPT_MASK_K_0;                    \
         shri    tmp, tmp, INT_MEM_ERROR;                        \
         andi    tmp, tmp, 1
 
 /* Load up a pointer to &interrupts_enabled_mask. */
 #define GET_INTERRUPTS_ENABLED_MASK_PTR(reg)                    \
-        moveli  reg, lo16(interrupts_enabled_mask);     \
-        auli    reg, reg, ha16(interrupts_enabled_mask);\
+        moveli  reg, lo16(interrupts_enabled_mask);             \
+        auli    reg, reg, ha16(interrupts_enabled_mask);        \
         add     reg, reg, tp
 
 /* Disable interrupts. */
 #define IRQ_DISABLE(tmp0, tmp1)                                 \
         {                                                       \
          movei  tmp0, -1;                                       \
-         moveli tmp1, lo16(LINUX_MASKABLE_INTERRUPTS)           \
+         moveli tmp1, lo16(LINUX_MASKABLE_INTERRUPTS_HI)        \
         };                                                      \
         {                                                       \
-         mtspr  INTERRUPT_MASK_SET_1_0, tmp0;                   \
-         auli   tmp1, tmp1, ha16(LINUX_MASKABLE_INTERRUPTS)     \
+         mtspr  SPR_INTERRUPT_MASK_SET_K_0, tmp0;               \
+         auli   tmp1, tmp1, ha16(LINUX_MASKABLE_INTERRUPTS_HI)  \
         };                                                      \
-        mtspr   INTERRUPT_MASK_SET_1_1, tmp1
+        mtspr   SPR_INTERRUPT_MASK_SET_K_1, tmp1
 
 /* Disable ALL synchronous interrupts (used by NMI entry). */
 #define IRQ_DISABLE_ALL(tmp)                                    \
         movei   tmp, -1;                                        \
-        mtspr   INTERRUPT_MASK_SET_1_0, tmp;                    \
-        mtspr   INTERRUPT_MASK_SET_1_1, tmp
+        mtspr   SPR_INTERRUPT_MASK_SET_K_0, tmp;                \
+        mtspr   SPR_INTERRUPT_MASK_SET_K_1, tmp
 
 /* Enable interrupts. */
 #define IRQ_ENABLE(tmp0, tmp1)                                  \
@@ -244,8 +260,8 @@ DECLARE_PER_CPU(unsigned long long, interrupts_enabled_mask);
          addi   tmp1, tmp0, 4                                   \
         };                                                      \
         lw      tmp1, tmp1;                                     \
-        mtspr   INTERRUPT_MASK_RESET_1_0, tmp0;                 \
-        mtspr   INTERRUPT_MASK_RESET_1_1, tmp1
+        mtspr   SPR_INTERRUPT_MASK_RESET_K_0, tmp0;             \
+        mtspr   SPR_INTERRUPT_MASK_RESET_K_1, tmp1
 #endif
 
 /*
diff --git a/arch/tile/include/asm/kmap_types.h b/arch/tile/include/asm/kmap_types.h
index 1480106d1c05..3d0f20246260 100644
--- a/arch/tile/include/asm/kmap_types.h
+++ b/arch/tile/include/asm/kmap_types.h
@@ -16,28 +16,42 @@
 #define _ASM_TILE_KMAP_TYPES_H
 
 /*
- * In TILE Linux each set of four of these uses another 16MB chunk of
- * address space, given 64 tiles and 64KB pages, so we only enable
- * ones that are required by the kernel configuration.
+ * In 32-bit TILE Linux we have to balance the desire to have a lot of
+ * nested atomic mappings with the fact that large page sizes and many
+ * processors chew up address space quickly.  In a typical
+ * 64-processor, 64KB-page layout build, making KM_TYPE_NR one larger
+ * adds 4MB of required address-space.  For now we leave KM_TYPE_NR
+ * set to depth 8.
  */
 enum km_type {
+        KM_TYPE_NR = 8
+};
+
+/*
+ * We provide dummy definitions of all the stray values that used to be
+ * required for kmap_atomic() and no longer are.
+ */
+enum {
         KM_BOUNCE_READ,
         KM_SKB_SUNRPC_DATA,
         KM_SKB_DATA_SOFTIRQ,
         KM_USER0,
         KM_USER1,
         KM_BIO_SRC_IRQ,
+        KM_BIO_DST_IRQ,
+        KM_PTE0,
+        KM_PTE1,
         KM_IRQ0,
         KM_IRQ1,
         KM_SOFTIRQ0,
         KM_SOFTIRQ1,
-        KM_MEMCPY0,
-        KM_MEMCPY1,
-#if defined(CONFIG_HIGHPTE)
-        KM_PTE0,
-        KM_PTE1,
-#endif
-        KM_TYPE_NR
+        KM_SYNC_ICACHE,
+        KM_SYNC_DCACHE,
+        KM_UML_USERCOPY,
+        KM_IRQ_PTE,
+        KM_NMI,
+        KM_NMI_PTE,
+        KM_KDB
 };
 
 #endif /* _ASM_TILE_KMAP_TYPES_H */
diff --git a/arch/tile/include/asm/mman.h b/arch/tile/include/asm/mman.h
index 4c6811e3e8dc..81b8fc348d63 100644
--- a/arch/tile/include/asm/mman.h
+++ b/arch/tile/include/asm/mman.h
@@ -23,6 +23,7 @@
 #define MAP_POPULATE    0x0040          /* populate (prefault) pagetables */
 #define MAP_NONBLOCK    0x0080          /* do not block on IO */
 #define MAP_GROWSDOWN   0x0100          /* stack-like segment */
+#define MAP_STACK       MAP_GROWSDOWN   /* provide convenience alias */
 #define MAP_LOCKED      0x0200          /* pages are locked */
 #define MAP_NORESERVE   0x0400          /* don't check for reservations */
 #define MAP_DENYWRITE   0x0800          /* ETXTBSY */
diff --git a/arch/tile/include/asm/mmu_context.h b/arch/tile/include/asm/mmu_context.h
index 9bc0d0725c28..15fb24641120 100644
--- a/arch/tile/include/asm/mmu_context.h
+++ b/arch/tile/include/asm/mmu_context.h
@@ -100,8 +100,8 @@ static inline void switch_mm(struct mm_struct *prev, struct mm_struct *next,
                 __get_cpu_var(current_asid) = asid;
 
                 /* Clear cpu from the old mm, and set it in the new one. */
-                cpumask_clear_cpu(cpu, &prev->cpu_vm_mask);
-                cpumask_set_cpu(cpu, &next->cpu_vm_mask);
+                cpumask_clear_cpu(cpu, mm_cpumask(prev));
+                cpumask_set_cpu(cpu, mm_cpumask(next));
 
                 /* Re-load page tables */
                 install_page_table(next->pgd, asid);
diff --git a/arch/tile/include/asm/mmzone.h b/arch/tile/include/asm/mmzone.h
index c6344c4f32ac..9d3dbce8f953 100644
--- a/arch/tile/include/asm/mmzone.h
+++ b/arch/tile/include/asm/mmzone.h
@@ -40,17 +40,6 @@ static inline int pfn_to_nid(unsigned long pfn)
         return highbits_to_node[__pfn_to_highbits(pfn)];
 }
 
-/*
- * Following are macros that each numa implmentation must define.
- */
-
-#define node_start_pfn(nid)     (NODE_DATA(nid)->node_start_pfn)
-#define node_end_pfn(nid)                                               \
-({                                                                      \
-        pg_data_t *__pgdat = NODE_DATA(nid);                            \
-        __pgdat->node_start_pfn + __pgdat->node_spanned_pages;          \
-})
-
 #define kern_addr_valid(kaddr)  virt_addr_valid((void *)kaddr)
 
 static inline int pfn_valid(int pfn)
diff --git a/arch/tile/include/asm/opcode-tile_32.h b/arch/tile/include/asm/opcode-tile_32.h
index eda60ecbae3d..03df7b1e77bf 100644
--- a/arch/tile/include/asm/opcode-tile_32.h
+++ b/arch/tile/include/asm/opcode-tile_32.h
@@ -1502,5 +1502,12 @@ extern int parse_insn_tile(tile_bundle_bits bits,
                            decoded[TILE_MAX_INSTRUCTIONS_PER_BUNDLE]);
 
 
+/* Given a set of bundle bits and a specific pipe, returns which
+ * instruction the bundle contains in that pipe.
+ */
+extern const struct tile_opcode *
+find_opcode(tile_bundle_bits bits, tile_pipeline pipe);
+
+
 
 #endif /* opcode_tile_h */
diff --git a/arch/tile/include/asm/opcode-tile_64.h b/arch/tile/include/asm/opcode-tile_64.h
index eda60ecbae3d..c0633466cd5c 100644
--- a/arch/tile/include/asm/opcode-tile_64.h
+++ b/arch/tile/include/asm/opcode-tile_64.h
@@ -5,863 +5,711 @@
 #ifndef opcode_tile_h
 #define opcode_tile_h
 
-typedef unsigned long long tile_bundle_bits;
+typedef unsigned long long tilegx_bundle_bits;
 
 
 enum
 {
-  TILE_MAX_OPERANDS = 5 /* mm */
+  TILEGX_MAX_OPERANDS = 4 /* bfexts */
 };
 
 typedef enum
 {
-  TILE_OPC_BPT,
-  TILE_OPC_INFO,
-  TILE_OPC_INFOL,
-  TILE_OPC_J,
-  TILE_OPC_JAL,
-  TILE_OPC_MOVE,
-  TILE_OPC_MOVE_SN,
-  TILE_OPC_MOVEI,
-  TILE_OPC_MOVEI_SN,
-  TILE_OPC_MOVELI,
-  TILE_OPC_MOVELI_SN,
-  TILE_OPC_MOVELIS,
-  TILE_OPC_PREFETCH,
-  TILE_OPC_RAISE,
-  TILE_OPC_ADD,
-  TILE_OPC_ADD_SN,
-  TILE_OPC_ADDB,
-  TILE_OPC_ADDB_SN,
-  TILE_OPC_ADDBS_U,
-  TILE_OPC_ADDBS_U_SN,
-  TILE_OPC_ADDH,
-  TILE_OPC_ADDH_SN,
-  TILE_OPC_ADDHS,
-  TILE_OPC_ADDHS_SN,
-  TILE_OPC_ADDI,
-  TILE_OPC_ADDI_SN,
-  TILE_OPC_ADDIB,
-  TILE_OPC_ADDIB_SN,
-  TILE_OPC_ADDIH,
-  TILE_OPC_ADDIH_SN,
-  TILE_OPC_ADDLI,
-  TILE_OPC_ADDLI_SN,
-  TILE_OPC_ADDLIS,
-  TILE_OPC_ADDS,
-  TILE_OPC_ADDS_SN,
-  TILE_OPC_ADIFFB_U,
-  TILE_OPC_ADIFFB_U_SN,
-  TILE_OPC_ADIFFH,
-  TILE_OPC_ADIFFH_SN,
-  TILE_OPC_AND,
-  TILE_OPC_AND_SN,
-  TILE_OPC_ANDI,
-  TILE_OPC_ANDI_SN,
-  TILE_OPC_AULI,
-  TILE_OPC_AVGB_U,
-  TILE_OPC_AVGB_U_SN,
-  TILE_OPC_AVGH,
-  TILE_OPC_AVGH_SN,
-  TILE_OPC_BBNS,
-  TILE_OPC_BBNS_SN,
-  TILE_OPC_BBNST,
-  TILE_OPC_BBNST_SN,
-  TILE_OPC_BBS,
-  TILE_OPC_BBS_SN,
-  TILE_OPC_BBST,
-  TILE_OPC_BBST_SN,
-  TILE_OPC_BGEZ,
-  TILE_OPC_BGEZ_SN,
-  TILE_OPC_BGEZT,
-  TILE_OPC_BGEZT_SN,
-  TILE_OPC_BGZ,
-  TILE_OPC_BGZ_SN,
-  TILE_OPC_BGZT,
-  TILE_OPC_BGZT_SN,
-  TILE_OPC_BITX,
-  TILE_OPC_BITX_SN,
-  TILE_OPC_BLEZ,
-  TILE_OPC_BLEZ_SN,
-  TILE_OPC_BLEZT,
-  TILE_OPC_BLEZT_SN,
-  TILE_OPC_BLZ,
-  TILE_OPC_BLZ_SN,
-  TILE_OPC_BLZT,
-  TILE_OPC_BLZT_SN,
-  TILE_OPC_BNZ,
-  TILE_OPC_BNZ_SN,
-  TILE_OPC_BNZT,
-  TILE_OPC_BNZT_SN,
-  TILE_OPC_BYTEX,
-  TILE_OPC_BYTEX_SN,
-  TILE_OPC_BZ,
-  TILE_OPC_BZ_SN,
-  TILE_OPC_BZT,
-  TILE_OPC_BZT_SN,
-  TILE_OPC_CLZ,
-  TILE_OPC_CLZ_SN,
-  TILE_OPC_CRC32_32,
-  TILE_OPC_CRC32_32_SN,
-  TILE_OPC_CRC32_8,
-  TILE_OPC_CRC32_8_SN,
-  TILE_OPC_CTZ,
-  TILE_OPC_CTZ_SN,
-  TILE_OPC_DRAIN,
-  TILE_OPC_DTLBPR,
-  TILE_OPC_DWORD_ALIGN,
-  TILE_OPC_DWORD_ALIGN_SN,
-  TILE_OPC_FINV,
-  TILE_OPC_FLUSH,
-  TILE_OPC_FNOP,
-  TILE_OPC_ICOH,
-  TILE_OPC_ILL,
-  TILE_OPC_INTHB,
-  TILE_OPC_INTHB_SN,
-  TILE_OPC_INTHH,
-  TILE_OPC_INTHH_SN,
-  TILE_OPC_INTLB,
-  TILE_OPC_INTLB_SN,
-  TILE_OPC_INTLH,
-  TILE_OPC_INTLH_SN,
-  TILE_OPC_INV,
-  TILE_OPC_IRET,
-  TILE_OPC_JALB,
-  TILE_OPC_JALF,
-  TILE_OPC_JALR,
-  TILE_OPC_JALRP,
-  TILE_OPC_JB,
-  TILE_OPC_JF,
-  TILE_OPC_JR,
-  TILE_OPC_JRP,
-  TILE_OPC_LB,
-  TILE_OPC_LB_SN,
-  TILE_OPC_LB_U,
-  TILE_OPC_LB_U_SN,
-  TILE_OPC_LBADD,
-  TILE_OPC_LBADD_SN,
-  TILE_OPC_LBADD_U,
-  TILE_OPC_LBADD_U_SN,
-  TILE_OPC_LH,
-  TILE_OPC_LH_SN,
-  TILE_OPC_LH_U,
-  TILE_OPC_LH_U_SN,
-  TILE_OPC_LHADD,
-  TILE_OPC_LHADD_SN,
-  TILE_OPC_LHADD_U,
-  TILE_OPC_LHADD_U_SN,
-  TILE_OPC_LNK,
-  TILE_OPC_LNK_SN,
-  TILE_OPC_LW,
-  TILE_OPC_LW_SN,
-  TILE_OPC_LW_NA,
-  TILE_OPC_LW_NA_SN,
-  TILE_OPC_LWADD,
-  TILE_OPC_LWADD_SN,
-  TILE_OPC_LWADD_NA,
-  TILE_OPC_LWADD_NA_SN,
-  TILE_OPC_MAXB_U,
-  TILE_OPC_MAXB_U_SN,
-  TILE_OPC_MAXH,
-  TILE_OPC_MAXH_SN,
-  TILE_OPC_MAXIB_U,
-  TILE_OPC_MAXIB_U_SN,
-  TILE_OPC_MAXIH,
-  TILE_OPC_MAXIH_SN,
-  TILE_OPC_MF,
-  TILE_OPC_MFSPR,
-  TILE_OPC_MINB_U,
-  TILE_OPC_MINB_U_SN,
-  TILE_OPC_MINH,
-  TILE_OPC_MINH_SN,
-  TILE_OPC_MINIB_U,
-  TILE_OPC_MINIB_U_SN,
-  TILE_OPC_MINIH,
-  TILE_OPC_MINIH_SN,
-  TILE_OPC_MM,
-  TILE_OPC_MNZ,
-  TILE_OPC_MNZ_SN,
-  TILE_OPC_MNZB,
-  TILE_OPC_MNZB_SN,
-  TILE_OPC_MNZH,
-  TILE_OPC_MNZH_SN,
-  TILE_OPC_MTSPR,
-  TILE_OPC_MULHH_SS,
-  TILE_OPC_MULHH_SS_SN,
-  TILE_OPC_MULHH_SU,
-  TILE_OPC_MULHH_SU_SN,
-  TILE_OPC_MULHH_UU,
-  TILE_OPC_MULHH_UU_SN,
-  TILE_OPC_MULHHA_SS,
-  TILE_OPC_MULHHA_SS_SN,
-  TILE_OPC_MULHHA_SU,
-  TILE_OPC_MULHHA_SU_SN,
-  TILE_OPC_MULHHA_UU,
-  TILE_OPC_MULHHA_UU_SN,
-  TILE_OPC_MULHHSA_UU,
-  TILE_OPC_MULHHSA_UU_SN,
-  TILE_OPC_MULHL_SS,
-  TILE_OPC_MULHL_SS_SN,
-  TILE_OPC_MULHL_SU,
-  TILE_OPC_MULHL_SU_SN,
-  TILE_OPC_MULHL_US,
-  TILE_OPC_MULHL_US_SN,
-  TILE_OPC_MULHL_UU,
-  TILE_OPC_MULHL_UU_SN,
-  TILE_OPC_MULHLA_SS,
-  TILE_OPC_MULHLA_SS_SN,
-  TILE_OPC_MULHLA_SU,
-  TILE_OPC_MULHLA_SU_SN,
-  TILE_OPC_MULHLA_US,
-  TILE_OPC_MULHLA_US_SN,
-  TILE_OPC_MULHLA_UU,
-  TILE_OPC_MULHLA_UU_SN,
-  TILE_OPC_MULHLSA_UU,
-  TILE_OPC_MULHLSA_UU_SN,
-  TILE_OPC_MULLL_SS,
-  TILE_OPC_MULLL_SS_SN,
-  TILE_OPC_MULLL_SU,
-  TILE_OPC_MULLL_SU_SN,
-  TILE_OPC_MULLL_UU,
-  TILE_OPC_MULLL_UU_SN,
-  TILE_OPC_MULLLA_SS,
-  TILE_OPC_MULLLA_SS_SN,
-  TILE_OPC_MULLLA_SU,
-  TILE_OPC_MULLLA_SU_SN,
-  TILE_OPC_MULLLA_UU,
-  TILE_OPC_MULLLA_UU_SN,
-  TILE_OPC_MULLLSA_UU,
-  TILE_OPC_MULLLSA_UU_SN,
-  TILE_OPC_MVNZ,
-  TILE_OPC_MVNZ_SN,
-  TILE_OPC_MVZ,
-  TILE_OPC_MVZ_SN,
-  TILE_OPC_MZ,
-  TILE_OPC_MZ_SN,
-  TILE_OPC_MZB,
-  TILE_OPC_MZB_SN,
-  TILE_OPC_MZH,
-  TILE_OPC_MZH_SN,
-  TILE_OPC_NAP,
-  TILE_OPC_NOP,
-  TILE_OPC_NOR,
-  TILE_OPC_NOR_SN,
-  TILE_OPC_OR,
-  TILE_OPC_OR_SN,
-  TILE_OPC_ORI,
-  TILE_OPC_ORI_SN,
-  TILE_OPC_PACKBS_U,
-  TILE_OPC_PACKBS_U_SN,
-  TILE_OPC_PACKHB,
-  TILE_OPC_PACKHB_SN,
-  TILE_OPC_PACKHS,
-  TILE_OPC_PACKHS_SN,
-  TILE_OPC_PACKLB,
-  TILE_OPC_PACKLB_SN,
-  TILE_OPC_PCNT,
-  TILE_OPC_PCNT_SN,
-  TILE_OPC_RL,
-  TILE_OPC_RL_SN,
-  TILE_OPC_RLI,
-  TILE_OPC_RLI_SN,
-  TILE_OPC_S1A,
-  TILE_OPC_S1A_SN,
-  TILE_OPC_S2A,
-  TILE_OPC_S2A_SN,
-  TILE_OPC_S3A,
-  TILE_OPC_S3A_SN,
-  TILE_OPC_SADAB_U,
-  TILE_OPC_SADAB_U_SN,
-  TILE_OPC_SADAH,
-  TILE_OPC_SADAH_SN,
-  TILE_OPC_SADAH_U,
-  TILE_OPC_SADAH_U_SN,
-  TILE_OPC_SADB_U,
-  TILE_OPC_SADB_U_SN,
-  TILE_OPC_SADH,
-  TILE_OPC_SADH_SN,
-  TILE_OPC_SADH_U,
-  TILE_OPC_SADH_U_SN,
-  TILE_OPC_SB,
-  TILE_OPC_SBADD,
-  TILE_OPC_SEQ,
-  TILE_OPC_SEQ_SN,
-  TILE_OPC_SEQB,
-  TILE_OPC_SEQB_SN,
-  TILE_OPC_SEQH,
-  TILE_OPC_SEQH_SN,
-  TILE_OPC_SEQI,
-  TILE_OPC_SEQI_SN,
-  TILE_OPC_SEQIB,
-  TILE_OPC_SEQIB_SN,
-  TILE_OPC_SEQIH,
-  TILE_OPC_SEQIH_SN,
-  TILE_OPC_SH,
-  TILE_OPC_SHADD,
-  TILE_OPC_SHL,
-  TILE_OPC_SHL_SN,
-  TILE_OPC_SHLB,
-  TILE_OPC_SHLB_SN,
-  TILE_OPC_SHLH,
-  TILE_OPC_SHLH_SN,
-  TILE_OPC_SHLI,
-  TILE_OPC_SHLI_SN,
-  TILE_OPC_SHLIB,
-  TILE_OPC_SHLIB_SN,
-  TILE_OPC_SHLIH,
-  TILE_OPC_SHLIH_SN,
-  TILE_OPC_SHR,
-  TILE_OPC_SHR_SN,
-  TILE_OPC_SHRB,
-  TILE_OPC_SHRB_SN,
-  TILE_OPC_SHRH,
-  TILE_OPC_SHRH_SN,
-  TILE_OPC_SHRI,
-  TILE_OPC_SHRI_SN,
-  TILE_OPC_SHRIB,
-  TILE_OPC_SHRIB_SN,
-  TILE_OPC_SHRIH,
-  TILE_OPC_SHRIH_SN,
-  TILE_OPC_SLT,
-  TILE_OPC_SLT_SN,
-  TILE_OPC_SLT_U,
-  TILE_OPC_SLT_U_SN,
-  TILE_OPC_SLTB,
-  TILE_OPC_SLTB_SN,
-  TILE_OPC_SLTB_U,
-  TILE_OPC_SLTB_U_SN,
-  TILE_OPC_SLTE,
-  TILE_OPC_SLTE_SN,
-  TILE_OPC_SLTE_U,
-  TILE_OPC_SLTE_U_SN,
-  TILE_OPC_SLTEB,
-  TILE_OPC_SLTEB_SN,
-  TILE_OPC_SLTEB_U,
-  TILE_OPC_SLTEB_U_SN,
-  TILE_OPC_SLTEH,
-  TILE_OPC_SLTEH_SN,
-  TILE_OPC_SLTEH_U,
-  TILE_OPC_SLTEH_U_SN,
-  TILE_OPC_SLTH,
-  TILE_OPC_SLTH_SN,
-  TILE_OPC_SLTH_U,
-  TILE_OPC_SLTH_U_SN,
-  TILE_OPC_SLTI,
-  TILE_OPC_SLTI_SN,
-  TILE_OPC_SLTI_U,
-  TILE_OPC_SLTI_U_SN,
-  TILE_OPC_SLTIB,
-  TILE_OPC_SLTIB_SN,
-  TILE_OPC_SLTIB_U,
-  TILE_OPC_SLTIB_U_SN,
-  TILE_OPC_SLTIH,
-  TILE_OPC_SLTIH_SN,
-  TILE_OPC_SLTIH_U,
-  TILE_OPC_SLTIH_U_SN,
-  TILE_OPC_SNE,
-  TILE_OPC_SNE_SN,
-  TILE_OPC_SNEB,
-  TILE_OPC_SNEB_SN,
-  TILE_OPC_SNEH,
-  TILE_OPC_SNEH_SN,
-  TILE_OPC_SRA,
-  TILE_OPC_SRA_SN,
-  TILE_OPC_SRAB,
-  TILE_OPC_SRAB_SN,
-  TILE_OPC_SRAH,
-  TILE_OPC_SRAH_SN,
-  TILE_OPC_SRAI,
-  TILE_OPC_SRAI_SN,
-  TILE_OPC_SRAIB,
-  TILE_OPC_SRAIB_SN,
-  TILE_OPC_SRAIH,
-  TILE_OPC_SRAIH_SN,
-  TILE_OPC_SUB,
-  TILE_OPC_SUB_SN,
-  TILE_OPC_SUBB,
-  TILE_OPC_SUBB_SN,
-  TILE_OPC_SUBBS_U,
-  TILE_OPC_SUBBS_U_SN,
-  TILE_OPC_SUBH,
-  TILE_OPC_SUBH_SN,
-  TILE_OPC_SUBHS,
-  TILE_OPC_SUBHS_SN,
-  TILE_OPC_SUBS,
-  TILE_OPC_SUBS_SN,
-  TILE_OPC_SW,
-  TILE_OPC_SWADD,
-  TILE_OPC_SWINT0,
-  TILE_OPC_SWINT1,
-  TILE_OPC_SWINT2,
-  TILE_OPC_SWINT3,
-  TILE_OPC_TBLIDXB0,
-  TILE_OPC_TBLIDXB0_SN,
-  TILE_OPC_TBLIDXB1,
-  TILE_OPC_TBLIDXB1_SN,
-  TILE_OPC_TBLIDXB2,
-  TILE_OPC_TBLIDXB2_SN,
-  TILE_OPC_TBLIDXB3,
-  TILE_OPC_TBLIDXB3_SN,
-  TILE_OPC_TNS,
-  TILE_OPC_TNS_SN,
-  TILE_OPC_WH64,
-  TILE_OPC_XOR,
-  TILE_OPC_XOR_SN,
-  TILE_OPC_XORI,
-  TILE_OPC_XORI_SN,
-  TILE_OPC_NONE
-} tile_mnemonic;
+  TILEGX_OPC_BPT,
+  TILEGX_OPC_INFO,
+  TILEGX_OPC_INFOL,
+  TILEGX_OPC_MOVE,
+  TILEGX_OPC_MOVEI,
+  TILEGX_OPC_MOVELI,
+  TILEGX_OPC_PREFETCH,
+  TILEGX_OPC_PREFETCH_ADD_L1,
+  TILEGX_OPC_PREFETCH_ADD_L1_FAULT,
+  TILEGX_OPC_PREFETCH_ADD_L2,
+  TILEGX_OPC_PREFETCH_ADD_L2_FAULT,
+  TILEGX_OPC_PREFETCH_ADD_L3,
+  TILEGX_OPC_PREFETCH_ADD_L3_FAULT,
+  TILEGX_OPC_PREFETCH_L1,
+  TILEGX_OPC_PREFETCH_L1_FAULT,
+  TILEGX_OPC_PREFETCH_L2,
+  TILEGX_OPC_PREFETCH_L2_FAULT,
+  TILEGX_OPC_PREFETCH_L3,
+  TILEGX_OPC_PREFETCH_L3_FAULT,
+  TILEGX_OPC_RAISE,
+  TILEGX_OPC_ADD,
+  TILEGX_OPC_ADDI,
+  TILEGX_OPC_ADDLI,
+  TILEGX_OPC_ADDX,
+  TILEGX_OPC_ADDXI,
+  TILEGX_OPC_ADDXLI,
+  TILEGX_OPC_ADDXSC,
+  TILEGX_OPC_AND,
+  TILEGX_OPC_ANDI,
+  TILEGX_OPC_BEQZ,
+  TILEGX_OPC_BEQZT,
+  TILEGX_OPC_BFEXTS,
+  TILEGX_OPC_BFEXTU,
+  TILEGX_OPC_BFINS,
+  TILEGX_OPC_BGEZ,
+  TILEGX_OPC_BGEZT,
+  TILEGX_OPC_BGTZ,
+  TILEGX_OPC_BGTZT,
+  TILEGX_OPC_BLBC,
+  TILEGX_OPC_BLBCT,
+  TILEGX_OPC_BLBS,
+  TILEGX_OPC_BLBST,
+  TILEGX_OPC_BLEZ,
+  TILEGX_OPC_BLEZT,
+  TILEGX_OPC_BLTZ,
+  TILEGX_OPC_BLTZT,
+  TILEGX_OPC_BNEZ,
+  TILEGX_OPC_BNEZT,
+  TILEGX_OPC_CLZ,
+  TILEGX_OPC_CMOVEQZ,
+  TILEGX_OPC_CMOVNEZ,
+  TILEGX_OPC_CMPEQ,
+  TILEGX_OPC_CMPEQI,
+  TILEGX_OPC_CMPEXCH,
+  TILEGX_OPC_CMPEXCH4,
+  TILEGX_OPC_CMPLES,
+  TILEGX_OPC_CMPLEU,
+  TILEGX_OPC_CMPLTS,
+  TILEGX_OPC_CMPLTSI,
+  TILEGX_OPC_CMPLTU,
+  TILEGX_OPC_CMPLTUI,
+  TILEGX_OPC_CMPNE,
+  TILEGX_OPC_CMUL,
+  TILEGX_OPC_CMULA,
+  TILEGX_OPC_CMULAF,
+  TILEGX_OPC_CMULF,
+  TILEGX_OPC_CMULFR,
+  TILEGX_OPC_CMULH,
+  TILEGX_OPC_CMULHR,
+  TILEGX_OPC_CRC32_32,
+  TILEGX_OPC_CRC32_8,
+  TILEGX_OPC_CTZ,
+  TILEGX_OPC_DBLALIGN,
+  TILEGX_OPC_DBLALIGN2,
+  TILEGX_OPC_DBLALIGN4,
+  TILEGX_OPC_DBLALIGN6,
+  TILEGX_OPC_DRAIN,
+  TILEGX_OPC_DTLBPR,
+  TILEGX_OPC_EXCH,
+  TILEGX_OPC_EXCH4,
+  TILEGX_OPC_FDOUBLE_ADD_FLAGS,
+  TILEGX_OPC_FDOUBLE_ADDSUB,
+  TILEGX_OPC_FDOUBLE_MUL_FLAGS,
+  TILEGX_OPC_FDOUBLE_PACK1,
+  TILEGX_OPC_FDOUBLE_PACK2,
+  TILEGX_OPC_FDOUBLE_SUB_FLAGS,
+  TILEGX_OPC_FDOUBLE_UNPACK_MAX,
+  TILEGX_OPC_FDOUBLE_UNPACK_MIN,
+  TILEGX_OPC_FETCHADD,
+  TILEGX_OPC_FETCHADD4,
+  TILEGX_OPC_FETCHADDGEZ,
+  TILEGX_OPC_FETCHADDGEZ4,
+  TILEGX_OPC_FETCHAND,
+  TILEGX_OPC_FETCHAND4,
+  TILEGX_OPC_FETCHOR,
+  TILEGX_OPC_FETCHOR4,
+  TILEGX_OPC_FINV,
+  TILEGX_OPC_FLUSH,
+  TILEGX_OPC_FLUSHWB,
+  TILEGX_OPC_FNOP,
+  TILEGX_OPC_FSINGLE_ADD1,
+  TILEGX_OPC_FSINGLE_ADDSUB2,
+  TILEGX_OPC_FSINGLE_MUL1,
+  TILEGX_OPC_FSINGLE_MUL2,
+  TILEGX_OPC_FSINGLE_PACK1,
+  TILEGX_OPC_FSINGLE_PACK2,
+  TILEGX_OPC_FSINGLE_SUB1,
+  TILEGX_OPC_ICOH,
+  TILEGX_OPC_ILL,
+  TILEGX_OPC_INV,
+  TILEGX_OPC_IRET,
+  TILEGX_OPC_J,
+  TILEGX_OPC_JAL,
+  TILEGX_OPC_JALR,
+  TILEGX_OPC_JALRP,
+  TILEGX_OPC_JR,
+  TILEGX_OPC_JRP,
+  TILEGX_OPC_LD,
+  TILEGX_OPC_LD1S,
+  TILEGX_OPC_LD1S_ADD,
+  TILEGX_OPC_LD1U,
+  TILEGX_OPC_LD1U_ADD,
+  TILEGX_OPC_LD2S,
+  TILEGX_OPC_LD2S_ADD,
+  TILEGX_OPC_LD2U,
+  TILEGX_OPC_LD2U_ADD,
+  TILEGX_OPC_LD4S,
+  TILEGX_OPC_LD4S_ADD,
+  TILEGX_OPC_LD4U,
+  TILEGX_OPC_LD4U_ADD,
+  TILEGX_OPC_LD_ADD,
+  TILEGX_OPC_LDNA,
+  TILEGX_OPC_LDNA_ADD,
+  TILEGX_OPC_LDNT,
+  TILEGX_OPC_LDNT1S,
+  TILEGX_OPC_LDNT1S_ADD,
+  TILEGX_OPC_LDNT1U,
+  TILEGX_OPC_LDNT1U_ADD,
+  TILEGX_OPC_LDNT2S,
+  TILEGX_OPC_LDNT2S_ADD,
+  TILEGX_OPC_LDNT2U,
+  TILEGX_OPC_LDNT2U_ADD,
+  TILEGX_OPC_LDNT4S,
+  TILEGX_OPC_LDNT4S_ADD,
+  TILEGX_OPC_LDNT4U,
+  TILEGX_OPC_LDNT4U_ADD,
+  TILEGX_OPC_LDNT_ADD,
+  TILEGX_OPC_LNK,
+  TILEGX_OPC_MF,
+  TILEGX_OPC_MFSPR,
+  TILEGX_OPC_MM,
+  TILEGX_OPC_MNZ,
+  TILEGX_OPC_MTSPR,
+  TILEGX_OPC_MUL_HS_HS,
+  TILEGX_OPC_MUL_HS_HU,
+  TILEGX_OPC_MUL_HS_LS,
+  TILEGX_OPC_MUL_HS_LU,
+  TILEGX_OPC_MUL_HU_HU,
+  TILEGX_OPC_MUL_HU_LS,
+  TILEGX_OPC_MUL_HU_LU,
+  TILEGX_OPC_MUL_LS_LS,
+  TILEGX_OPC_MUL_LS_LU,
+  TILEGX_OPC_MUL_LU_LU,
+  TILEGX_OPC_MULA_HS_HS,
+  TILEGX_OPC_MULA_HS_HU,
+  TILEGX_OPC_MULA_HS_LS,
+  TILEGX_OPC_MULA_HS_LU,
+  TILEGX_OPC_MULA_HU_HU,
+  TILEGX_OPC_MULA_HU_LS,
+  TILEGX_OPC_MULA_HU_LU,
+  TILEGX_OPC_MULA_LS_LS,
+  TILEGX_OPC_MULA_LS_LU,
+  TILEGX_OPC_MULA_LU_LU,
+  TILEGX_OPC_MULAX,
+  TILEGX_OPC_MULX,
+  TILEGX_OPC_MZ,
+  TILEGX_OPC_NAP,
+  TILEGX_OPC_NOP,
+  TILEGX_OPC_NOR,
+  TILEGX_OPC_OR,
+  TILEGX_OPC_ORI,
+  TILEGX_OPC_PCNT,
+  TILEGX_OPC_REVBITS,
+  TILEGX_OPC_REVBYTES,
+  TILEGX_OPC_ROTL,
+  TILEGX_OPC_ROTLI,
+  TILEGX_OPC_SHL,
+  TILEGX_OPC_SHL16INSLI,
+  TILEGX_OPC_SHL1ADD,
+  TILEGX_OPC_SHL1ADDX,
+  TILEGX_OPC_SHL2ADD,
+  TILEGX_OPC_SHL2ADDX,
+  TILEGX_OPC_SHL3ADD,
+  TILEGX_OPC_SHL3ADDX,
+  TILEGX_OPC_SHLI,
+  TILEGX_OPC_SHLX,
+  TILEGX_OPC_SHLXI,
+  TILEGX_OPC_SHRS,
+  TILEGX_OPC_SHRSI,
+  TILEGX_OPC_SHRU,
+  TILEGX_OPC_SHRUI,
+  TILEGX_OPC_SHRUX,
+  TILEGX_OPC_SHRUXI,
+  TILEGX_OPC_SHUFFLEBYTES,
+  TILEGX_OPC_ST,
+  TILEGX_OPC_ST1,
+  TILEGX_OPC_ST1_ADD,
+  TILEGX_OPC_ST2,
+  TILEGX_OPC_ST2_ADD,
+  TILEGX_OPC_ST4,
+  TILEGX_OPC_ST4_ADD,
+  TILEGX_OPC_ST_ADD,
+  TILEGX_OPC_STNT,
+  TILEGX_OPC_STNT1,
+  TILEGX_OPC_STNT1_ADD,
+  TILEGX_OPC_STNT2,
+  TILEGX_OPC_STNT2_ADD,
+  TILEGX_OPC_STNT4,
+  TILEGX_OPC_STNT4_ADD,
+  TILEGX_OPC_STNT_ADD,
+  TILEGX_OPC_SUB,
+  TILEGX_OPC_SUBX,
+  TILEGX_OPC_SUBXSC,
+  TILEGX_OPC_SWINT0,
+  TILEGX_OPC_SWINT1,
+  TILEGX_OPC_SWINT2,
+  TILEGX_OPC_SWINT3,
+  TILEGX_OPC_TBLIDXB0,
+  TILEGX_OPC_TBLIDXB1,
+  TILEGX_OPC_TBLIDXB2,
+  TILEGX_OPC_TBLIDXB3,
+  TILEGX_OPC_V1ADD,
+  TILEGX_OPC_V1ADDI,
+  TILEGX_OPC_V1ADDUC,
+  TILEGX_OPC_V1ADIFFU,
+  TILEGX_OPC_V1AVGU,
+  TILEGX_OPC_V1CMPEQ,
+  TILEGX_OPC_V1CMPEQI,
+  TILEGX_OPC_V1CMPLES,
+  TILEGX_OPC_V1CMPLEU,
+  TILEGX_OPC_V1CMPLTS,
+  TILEGX_OPC_V1CMPLTSI,
+  TILEGX_OPC_V1CMPLTU,
+  TILEGX_OPC_V1CMPLTUI,
+  TILEGX_OPC_V1CMPNE,
+  TILEGX_OPC_V1DDOTPU,
+  TILEGX_OPC_V1DDOTPUA,
+  TILEGX_OPC_V1DDOTPUS,
+  TILEGX_OPC_V1DDOTPUSA,
+  TILEGX_OPC_V1DOTP,
+  TILEGX_OPC_V1DOTPA,
+  TILEGX_OPC_V1DOTPU,
+  TILEGX_OPC_V1DOTPUA,
+  TILEGX_OPC_V1DOTPUS,
+  TILEGX_OPC_V1DOTPUSA,
+  TILEGX_OPC_V1INT_H,
+  TILEGX_OPC_V1INT_L,
+  TILEGX_OPC_V1MAXU,
+  TILEGX_OPC_V1MAXUI,
+  TILEGX_OPC_V1MINU,
+  TILEGX_OPC_V1MINUI,
+  TILEGX_OPC_V1MNZ,
+  TILEGX_OPC_V1MULTU,
+  TILEGX_OPC_V1MULU,
+  TILEGX_OPC_V1MULUS,
+  TILEGX_OPC_V1MZ,
+  TILEGX_OPC_V1SADAU,
+  TILEGX_OPC_V1SADU,
+  TILEGX_OPC_V1SHL,
+  TILEGX_OPC_V1SHLI,
+  TILEGX_OPC_V1SHRS,
+  TILEGX_OPC_V1SHRSI,
+  TILEGX_OPC_V1SHRU,
+  TILEGX_OPC_V1SHRUI,
+  TILEGX_OPC_V1SUB,
+  TILEGX_OPC_V1SUBUC,
+  TILEGX_OPC_V2ADD,
+  TILEGX_OPC_V2ADDI,
+  TILEGX_OPC_V2ADDSC,
+  TILEGX_OPC_V2ADIFFS,
+  TILEGX_OPC_V2AVGS,
+  TILEGX_OPC_V2CMPEQ,
+  TILEGX_OPC_V2CMPEQI,
+  TILEGX_OPC_V2CMPLES,
+  TILEGX_OPC_V2CMPLEU,
+  TILEGX_OPC_V2CMPLTS,
+  TILEGX_OPC_V2CMPLTSI,
+  TILEGX_OPC_V2CMPLTU,
+  TILEGX_OPC_V2CMPLTUI,
+  TILEGX_OPC_V2CMPNE,
+  TILEGX_OPC_V2DOTP,
+  TILEGX_OPC_V2DOTPA,
+  TILEGX_OPC_V2INT_H,
+  TILEGX_OPC_V2INT_L,
+  TILEGX_OPC_V2MAXS,
+  TILEGX_OPC_V2MAXSI,
+  TILEGX_OPC_V2MINS,
+  TILEGX_OPC_V2MINSI,
+  TILEGX_OPC_V2MNZ,
+  TILEGX_OPC_V2MULFSC,
+  TILEGX_OPC_V2MULS,
+  TILEGX_OPC_V2MULTS,
+  TILEGX_OPC_V2MZ,
+  TILEGX_OPC_V2PACKH,
+  TILEGX_OPC_V2PACKL,
+  TILEGX_OPC_V2PACKUC,
+  TILEGX_OPC_V2SADAS,
+  TILEGX_OPC_V2SADAU,
+  TILEGX_OPC_V2SADS,
+  TILEGX_OPC_V2SADU,
+  TILEGX_OPC_V2SHL,
+  TILEGX_OPC_V2SHLI,
+  TILEGX_OPC_V2SHLSC,
+  TILEGX_OPC_V2SHRS,
+  TILEGX_OPC_V2SHRSI,
+  TILEGX_OPC_V2SHRU,
+  TILEGX_OPC_V2SHRUI,
+  TILEGX_OPC_V2SUB,
+  TILEGX_OPC_V2SUBSC,
+  TILEGX_OPC_V4ADD,
+  TILEGX_OPC_V4ADDSC,
+  TILEGX_OPC_V4INT_H,
+  TILEGX_OPC_V4INT_L,
+  TILEGX_OPC_V4PACKSC,
+  TILEGX_OPC_V4SHL,
+  TILEGX_OPC_V4SHLSC,
+  TILEGX_OPC_V4SHRS,
+  TILEGX_OPC_V4SHRU,
+  TILEGX_OPC_V4SUB,
+  TILEGX_OPC_V4SUBSC,
+  TILEGX_OPC_WH64,
+  TILEGX_OPC_XOR,
+  TILEGX_OPC_XORI,
+  TILEGX_OPC_NONE
+} tilegx_mnemonic;
 
 /* 64-bit pattern for a { bpt ; nop } bundle. */
-#define TILE_BPT_BUNDLE 0x400b3cae70166000ULL
+#define TILEGX_BPT_BUNDLE 0x286a44ae51485000ULL
 
 
-#define TILE_ELF_MACHINE_CODE EM_TILEPRO
+#define TILE_ELF_MACHINE_CODE EM_TILE64
 
-#define TILE_ELF_NAME "elf32-tilepro"
+#define TILE_ELF_NAME "elf32-tile64"
 
 
 static __inline unsigned int
-get_BrOff_SN(tile_bundle_bits num)
+get_BFEnd_X0(tilegx_bundle_bits num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((n >> 0)) & 0x3ff);
+  return (((n >> 12)) & 0x3f);
 }
 
 static __inline unsigned int
-get_BrOff_X1(tile_bundle_bits n)
+get_BFOpcodeExtension_X0(tilegx_bundle_bits num)
 {
-  return (((unsigned int)(n >> 43)) & 0x00007fff) |
-         (((unsigned int)(n >> 20)) & 0x00018000);
+  const unsigned int n = (unsigned int)num;
+  return (((n >> 24)) & 0xf);
 }
 
 static __inline unsigned int
-get_BrType_X1(tile_bundle_bits n)
+get_BFStart_X0(tilegx_bundle_bits num)
 {
-  return (((unsigned int)(n >> 31)) & 0xf);
+  const unsigned int n = (unsigned int)num;
+  return (((n >> 18)) & 0x3f);
 }
 
 static __inline unsigned int
-get_Dest_Imm8_X1(tile_bundle_bits n)
+get_BrOff_X1(tilegx_bundle_bits n)
 {
   return (((unsigned int)(n >> 31)) & 0x0000003f) |
-         (((unsigned int)(n >> 43)) & 0x000000c0);
+         (((unsigned int)(n >> 37)) & 0x0001ffc0);
 }
 
 static __inline unsigned int
-get_Dest_SN(tile_bundle_bits num)
+get_BrType_X1(tilegx_bundle_bits n)
 {
-  const unsigned int n = (unsigned int)num;
-  return (((n >> 2)) & 0x3);
+  return (((unsigned int)(n >> 54)) & 0x1f);
 }
 
 static __inline unsigned int
-get_Dest_X0(tile_bundle_bits num)
+get_Dest_Imm8_X1(tilegx_bundle_bits n)
+{
+  return (((unsigned int)(n >> 31)) & 0x0000003f) |
+         (((unsigned int)(n >> 43)) & 0x000000c0);
+}
+
+static __inline unsigned int
+get_Dest_X0(tilegx_bundle_bits num)
 {
   const unsigned int n = (unsigned int)num;
   return (((n >> 0)) & 0x3f);
 }
 
 static __inline unsigned int
-get_Dest_X1(tile_bundle_bits n)
+get_Dest_X1(tilegx_bundle_bits n)
 {
   return (((unsigned int)(n >> 31)) & 0x3f);
 }
 
 static __inline unsigned int
-get_Dest_Y0(tile_bundle_bits num)
+get_Dest_Y0(tilegx_bundle_bits num)
 {
   const unsigned int n = (unsigned int)num;
   return (((n >> 0)) & 0x3f);
 }
 
 static __inline unsigned int
-get_Dest_Y1(tile_bundle_bits n)
+get_Dest_Y1(tilegx_bundle_bits n)
 {
   return (((unsigned int)(n >> 31)) & 0x3f);
 }
 
 static __inline unsigned int
-get_Imm16_X0(tile_bundle_bits num)
+get_Imm16_X0(tilegx_bundle_bits num)
 {
   const unsigned int n = (unsigned int)num;
   return (((n >> 12)) & 0xffff);
 }
 
 static __inline unsigned int
-get_Imm16_X1(tile_bundle_bits n)
+get_Imm16_X1(tilegx_bundle_bits n)
 {
   return (((unsigned int)(n >> 43)) & 0xffff);
 }
 
 static __inline unsigned int
-get_Imm8_SN(tile_bundle_bits num)
-{
-  const unsigned int n = (unsigned int)num;
-  return (((n >> 0)) & 0xff);
-}
-
-static __inline unsigned int
-get_Imm8_X0(tile_bundle_bits num)
+get_Imm8OpcodeExtension_X0(tilegx_bundle_bits num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((n >> 12)) & 0xff);
+  return (((n >> 20)) & 0xff);
 }
 
 static __inline unsigned int
-get_Imm8_X1(tile_bundle_bits n)
+get_Imm8OpcodeExtension_X1(tilegx_bundle_bits n)
 {
-  return (((unsigned int)(n >> 43)) & 0xff);
+  return (((unsigned int)(n >> 51)) & 0xff);
 }
 
 static __inline unsigned int
-get_Imm8_Y0(tile_bundle_bits num)
+get_Imm8_X0(tilegx_bundle_bits num)
 {
   const unsigned int n = (unsigned int)num;
   return (((n >> 12)) & 0xff);
 }
 
 static __inline unsigned int
-get_Imm8_Y1(tile_bundle_bits n)
+get_Imm8_X1(tilegx_bundle_bits n)
 {
   return (((unsigned int)(n >> 43)) & 0xff);
 }
 
 static __inline unsigned int
-get_ImmOpcodeExtension_X0(tile_bundle_bits num)
-{
-  const unsigned int n = (unsigned int)num;
-  return (((n >> 20)) & 0x7f);
-}
-
-static __inline unsigned int
-get_ImmOpcodeExtension_X1(tile_bundle_bits n)
-{
-  return (((unsigned int)(n >> 51)) & 0x7f);
-}
-
-static __inline unsigned int
-get_ImmRROpcodeExtension_SN(tile_bundle_bits num)
+get_Imm8_Y0(tilegx_bundle_bits num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((n >> 8)) & 0x3);
-}
-
-static __inline unsigned int
-get_JOffLong_X1(tile_bundle_bits n)
-{
-  return (((unsigned int)(n >> 43)) & 0x00007fff) |
-         (((unsigned int)(n >> 20)) & 0x00018000) |
-         (((unsigned int)(n >> 14)) & 0x001e0000) |
-         (((unsigned int)(n >> 16)) & 0x07e00000) |
-         (((unsigned int)(n >> 31)) & 0x18000000);
-}
-
-static __inline unsigned int
-get_JOff_X1(tile_bundle_bits n)
-{
-  return (((unsigned int)(n >> 43)) & 0x00007fff) |
-         (((unsigned int)(n >> 20)) & 0x00018000) |
-         (((unsigned int)(n >> 14)) & 0x001e0000) |
-         (((unsigned int)(n >> 16)) & 0x07e00000) |
-         (((unsigned int)(n >> 31)) & 0x08000000);
-}
-
-static __inline unsigned int
-get_MF_Imm15_X1(tile_bundle_bits n)
-{
-  return (((unsigned int)(n >> 37)) & 0x00003fff) |
-         (((unsigned int)(n >> 44)) & 0x00004000);
+  return (((n >> 12)) & 0xff);
 }
 
 static __inline unsigned int
-get_MMEnd_X0(tile_bundle_bits num)
+get_Imm8_Y1(tilegx_bundle_bits n)
 {
-  const unsigned int n = (unsigned int)num;
-  return (((n >> 18)) & 0x1f);
+  return (((unsigned int)(n >> 43)) & 0xff);
 }
 
 static __inline unsigned int
-get_MMEnd_X1(tile_bundle_bits n)
+get_JumpOff_X1(tilegx_bundle_bits n)
 {
-  return (((unsigned int)(n >> 49)) & 0x1f);
+  return (((unsigned int)(n >> 31)) & 0x7ffffff);
 }
 
 static __inline unsigned int
-get_MMStart_X0(tile_bundle_bits num)
+get_JumpOpcodeExtension_X1(tilegx_bundle_bits n)
 {
-  const unsigned int n = (unsigned int)num;
-  return (((n >> 23)) & 0x1f);
+  return (((unsigned int)(n >> 58)) & 0x1);
 }
 
 static __inline unsigned int
-get_MMStart_X1(tile_bundle_bits n)
+get_MF_Imm14_X1(tilegx_bundle_bits n)
 {
-  return (((unsigned int)(n >> 54)) & 0x1f);
+  return (((unsigned int)(n >> 37)) & 0x3fff);
 }
 
 static __inline unsigned int
-get_MT_Imm15_X1(tile_bundle_bits n)
+get_MT_Imm14_X1(tilegx_bundle_bits n)
 {
   return (((unsigned int)(n >> 31)) & 0x0000003f) |
-         (((unsigned int)(n >> 37)) & 0x00003fc0) |
-         (((unsigned int)(n >> 44)) & 0x00004000);
+         (((unsigned int)(n >> 37)) & 0x00003fc0);
 }
 
 static __inline unsigned int
-get_Mode(tile_bundle_bits n)
+get_Mode(tilegx_bundle_bits n)
 {
-  return (((unsigned int)(n >> 63)) & 0x1);
+  return (((unsigned int)(n >> 62)) & 0x3);
 }
 
 static __inline unsigned int
-get_NoRegOpcodeExtension_SN(tile_bundle_bits num)
-{
-  const unsigned int n = (unsigned int)num;
-  return (((n >> 0)) & 0xf);
-}
-
-static __inline unsigned int
-get_Opcode_SN(tile_bundle_bits num)
-{
-  const unsigned int n = (unsigned int)num;
-  return (((n >> 10)) & 0x3f);
-}
-
-static __inline unsigned int
-get_Opcode_X0(tile_bundle_bits num)
+get_Opcode_X0(tilegx_bundle_bits num)
 {
   const unsigned int n = (unsigned int)num;
   return (((n >> 28)) & 0x7);
 }
 
 static __inline unsigned int
-get_Opcode_X1(tile_bundle_bits n)
+get_Opcode_X1(tilegx_bundle_bits n)
 {
-  return (((unsigned int)(n >> 59)) & 0xf);
+  return (((unsigned int)(n >> 59)) & 0x7);
 }
 
 static __inline unsigned int
-get_Opcode_Y0(tile_bundle_bits num)
+get_Opcode_Y0(tilegx_bundle_bits num)
 {
   const unsigned int n = (unsigned int)num;
   return (((n >> 27)) & 0xf);
 }
 
 static __inline unsigned int
-get_Opcode_Y1(tile_bundle_bits n)
+get_Opcode_Y1(tilegx_bundle_bits n)
 {
-  return (((unsigned int)(n >> 59)) & 0xf);
+  return (((unsigned int)(n >> 58)) & 0xf);
 }
 
 static __inline unsigned int
-get_Opcode_Y2(tile_bundle_bits n)
+get_Opcode_Y2(tilegx_bundle_bits n)
 {
-  return (((unsigned int)(n >> 56)) & 0x7);
-}
-
-static __inline unsigned int
-get_RROpcodeExtension_SN(tile_bundle_bits num)
-{
-  const unsigned int n = (unsigned int)num;
-  return (((n >> 4)) & 0xf);
+  return (((n >> 26)) & 0x00000001) |
+         (((unsigned int)(n >> 56)) & 0x00000002);
 }
 
 static __inline unsigned int
-get_RRROpcodeExtension_X0(tile_bundle_bits num)
+get_RRROpcodeExtension_X0(tilegx_bundle_bits num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((n >> 18)) & 0x1ff);
+  return (((n >> 18)) & 0x3ff);
 }
 
 static __inline unsigned int
-get_RRROpcodeExtension_X1(tile_bundle_bits n)
+get_RRROpcodeExtension_X1(tilegx_bundle_bits n)
 {
-  return (((unsigned int)(n >> 49)) & 0x1ff);
+  return (((unsigned int)(n >> 49)) & 0x3ff);
 }
 
 static __inline unsigned int
-get_RRROpcodeExtension_Y0(tile_bundle_bits num)
+get_RRROpcodeExtension_Y0(tilegx_bundle_bits num)
 {
   const unsigned int n = (unsigned int)num;
   return (((n >> 18)) & 0x3);
 }
 
 static __inline unsigned int
-get_RRROpcodeExtension_Y1(tile_bundle_bits n)
+get_RRROpcodeExtension_Y1(tilegx_bundle_bits n)
 {
   return (((unsigned int)(n >> 49)) & 0x3);
 }
 
 static __inline unsigned int
-get_RouteOpcodeExtension_SN(tile_bundle_bits num)
-{
-  const unsigned int n = (unsigned int)num;
-  return (((n >> 0)) & 0x3ff);
-}
-
-static __inline unsigned int
-get_S_X0(tile_bundle_bits num)
+get_ShAmt_X0(tilegx_bundle_bits num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((n >> 27)) & 0x1);
+  return (((n >> 12)) & 0x3f);
 }
 
 static __inline unsigned int
-get_S_X1(tile_bundle_bits n)
+get_ShAmt_X1(tilegx_bundle_bits n)
 {
-  return (((unsigned int)(n >> 58)) & 0x1);
+  return (((unsigned int)(n >> 43)) & 0x3f);
 }
 
 static __inline unsigned int
-get_ShAmt_X0(tile_bundle_bits num)
+get_ShAmt_Y0(tilegx_bundle_bits num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((n >> 12)) & 0x1f);
+  return (((n >> 12)) & 0x3f);
 }
 
 static __inline unsigned int
-get_ShAmt_X1(tile_bundle_bits n)
+get_ShAmt_Y1(tilegx_bundle_bits n)
 {
-  return (((unsigned int)(n >> 43)) & 0x1f);
+  return (((unsigned int)(n >> 43)) & 0x3f);
 }
 
 static __inline unsigned int
-get_ShAmt_Y0(tile_bundle_bits num)
+get_ShiftOpcodeExtension_X0(tilegx_bundle_bits num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((n >> 12)) & 0x1f);
+  return (((n >> 18)) & 0x3ff);
 }
 
 static __inline unsigned int
-get_ShAmt_Y1(tile_bundle_bits n)
+get_ShiftOpcodeExtension_X1(tilegx_bundle_bits n)
 {
-  return (((unsigned int)(n >> 43)) & 0x1f);
+  return (((unsigned int)(n >> 49)) & 0x3ff);
 }
 
 static __inline unsigned int
-get_SrcA_X0(tile_bundle_bits num)
+get_ShiftOpcodeExtension_Y0(tilegx_bundle_bits num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((n >> 6)) & 0x3f);
+  return (((n >> 18)) & 0x3);
 }
 
 static __inline unsigned int
-get_SrcA_X1(tile_bundle_bits n)
+get_ShiftOpcodeExtension_Y1(tilegx_bundle_bits n)
 {
-  return (((unsigned int)(n >> 37)) & 0x3f);
+  return (((unsigned int)(n >> 49)) & 0x3);
 }
 
 static __inline unsigned int
-get_SrcA_Y0(tile_bundle_bits num)
+get_SrcA_X0(tilegx_bundle_bits num)
 {
   const unsigned int n = (unsigned int)num;
   return (((n >> 6)) & 0x3f);
 }
 
 static __inline unsigned int
-get_SrcA_Y1(tile_bundle_bits n)
+get_SrcA_X1(tilegx_bundle_bits n)
 {
   return (((unsigned int)(n >> 37)) & 0x3f);
 }
 
 static __inline unsigned int
-get_SrcA_Y2(tile_bundle_bits n)
+get_SrcA_Y0(tilegx_bundle_bits num)
 {
-  return (((n >> 26)) & 0x00000001) |
-         (((unsigned int)(n >> 50)) & 0x0000003e);
+  const unsigned int n = (unsigned int)num;
+  return (((n >> 6)) & 0x3f);
 }
 
 static __inline unsigned int
-get_SrcBDest_Y2(tile_bundle_bits num)
+get_SrcA_Y1(tilegx_bundle_bits n)
 {
-  const unsigned int n = (unsigned int)num;
-  return (((n >> 20)) & 0x3f);
+  return (((unsigned int)(n >> 37)) & 0x3f);
 }
 
 static __inline unsigned int
-get_SrcB_X0(tile_bundle_bits num)
+get_SrcA_Y2(tilegx_bundle_bits num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((n >> 12)) & 0x3f);
+  return (((n >> 20)) & 0x3f);
 }
 
 static __inline unsigned int
-get_SrcB_X1(tile_bundle_bits n)
+get_SrcBDest_Y2(tilegx_bundle_bits n)
 {
-  return (((unsigned int)(n >> 43)) & 0x3f);
+  return (((unsigned int)(n >> 51)) & 0x3f);
 }
 
 static __inline unsigned int
-get_SrcB_Y0(tile_bundle_bits num)
+get_SrcB_X0(tilegx_bundle_bits num)
 {
   const unsigned int n = (unsigned int)num;
   return (((n >> 12)) & 0x3f);
 }
 
 static __inline unsigned int
-get_SrcB_Y1(tile_bundle_bits n)
+get_SrcB_X1(tilegx_bundle_bits n)
 {
   return (((unsigned int)(n >> 43)) & 0x3f);
 }
 
 static __inline unsigned int
-get_Src_SN(tile_bundle_bits num)
+get_SrcB_Y0(tilegx_bundle_bits num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((n >> 0)) & 0x3);
-}
-
-static __inline unsigned int
-get_UnOpcodeExtension_X0(tile_bundle_bits num)
-{
-  const unsigned int n = (unsigned int)num;
-  return (((n >> 12)) & 0x1f);
-}
-
-static __inline unsigned int
-get_UnOpcodeExtension_X1(tile_bundle_bits n)
-{
-  return (((unsigned int)(n >> 43)) & 0x1f);
-}
-
-static __inline unsigned int
-get_UnOpcodeExtension_Y0(tile_bundle_bits num)
-{
-  const unsigned int n = (unsigned int)num;
-  return (((n >> 12)) & 0x1f);
+  return (((n >> 12)) & 0x3f);
 }
 
 static __inline unsigned int
-get_UnOpcodeExtension_Y1(tile_bundle_bits n)
+get_SrcB_Y1(tilegx_bundle_bits n)
 {
-  return (((unsigned int)(n >> 43)) & 0x1f);
+  return (((unsigned int)(n >> 43)) & 0x3f);
 }
 
 static __inline unsigned int
-get_UnShOpcodeExtension_X0(tile_bundle_bits num)
+get_UnaryOpcodeExtension_X0(tilegx_bundle_bits num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((n >> 17)) & 0x3ff);
+  return (((n >> 12)) & 0x3f);
 }
 
 static __inline unsigned int
-get_UnShOpcodeExtension_X1(tile_bundle_bits n)
+get_UnaryOpcodeExtension_X1(tilegx_bundle_bits n)
 {
-  return (((unsigned int)(n >> 48)) & 0x3ff);
+  return (((unsigned int)(n >> 43)) & 0x3f);
 }
 
 static __inline unsigned int
-get_UnShOpcodeExtension_Y0(tile_bundle_bits num)
+get_UnaryOpcodeExtension_Y0(tilegx_bundle_bits num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((n >> 17)) & 0x7);
+  return (((n >> 12)) & 0x3f);
 }
 
 static __inline unsigned int
-get_UnShOpcodeExtension_Y1(tile_bundle_bits n)
+get_UnaryOpcodeExtension_Y1(tilegx_bundle_bits n)
 {
-  return (((unsigned int)(n >> 48)) & 0x7);
+  return (((unsigned int)(n >> 43)) & 0x3f);
 }
 
 
@@ -874,546 +722,441 @@ sign_extend(int n, int num_bits)
 
 
 
-static __inline tile_bundle_bits
-create_BrOff_SN(int num)
+static __inline tilegx_bundle_bits
+create_BFEnd_X0(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return ((n & 0x3ff) << 0);
+  return ((n & 0x3f) << 12);
 }
 
-static __inline tile_bundle_bits
-create_BrOff_X1(int num)
+static __inline tilegx_bundle_bits
+create_BFOpcodeExtension_X0(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0x00007fff)) << 43) |
-         (((tile_bundle_bits)(n & 0x00018000)) << 20);
+  return ((n & 0xf) << 24);
 }
 
-static __inline tile_bundle_bits
-create_BrType_X1(int num)
+static __inline tilegx_bundle_bits
+create_BFStart_X0(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0xf)) << 31);
+  return ((n & 0x3f) << 18);
 }
 
-static __inline tile_bundle_bits
-create_Dest_Imm8_X1(int num)
+static __inline tilegx_bundle_bits
+create_BrOff_X1(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0x0000003f)) << 31) |
-         (((tile_bundle_bits)(n & 0x000000c0)) << 43);
+  return (((tilegx_bundle_bits)(n & 0x0000003f)) << 31) |
+         (((tilegx_bundle_bits)(n & 0x0001ffc0)) << 37);
 }
 
-static __inline tile_bundle_bits
-create_Dest_SN(int num)
+static __inline tilegx_bundle_bits
+create_BrType_X1(int num)
+{
+  const unsigned int n = (unsigned int)num;
+  return (((tilegx_bundle_bits)(n & 0x1f)) << 54);
+}
+
+static __inline tilegx_bundle_bits
+create_Dest_Imm8_X1(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return ((n & 0x3) << 2);
+  return (((tilegx_bundle_bits)(n & 0x0000003f)) << 31) |
+         (((tilegx_bundle_bits)(n & 0x000000c0)) << 43);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_Dest_X0(int num)
 {
   const unsigned int n = (unsigned int)num;
   return ((n & 0x3f) << 0);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_Dest_X1(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0x3f)) << 31);
+  return (((tilegx_bundle_bits)(n & 0x3f)) << 31);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_Dest_Y0(int num)
 {
   const unsigned int n = (unsigned int)num;
   return ((n & 0x3f) << 0);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_Dest_Y1(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0x3f)) << 31);
+  return (((tilegx_bundle_bits)(n & 0x3f)) << 31);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_Imm16_X0(int num)
 {
   const unsigned int n = (unsigned int)num;
   return ((n & 0xffff) << 12);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_Imm16_X1(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0xffff)) << 43);
+  return (((tilegx_bundle_bits)(n & 0xffff)) << 43);
 }
 
-static __inline tile_bundle_bits
-create_Imm8_SN(int num)
+static __inline tilegx_bundle_bits
+create_Imm8OpcodeExtension_X0(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return ((n & 0xff) << 0);
+  return ((n & 0xff) << 20);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
+create_Imm8OpcodeExtension_X1(int num)
+{
+  const unsigned int n = (unsigned int)num;
+  return (((tilegx_bundle_bits)(n & 0xff)) << 51);
+}
+
+static __inline tilegx_bundle_bits
 create_Imm8_X0(int num)
 {
   const unsigned int n = (unsigned int)num;
   return ((n & 0xff) << 12);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_Imm8_X1(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0xff)) << 43);
+  return (((tilegx_bundle_bits)(n & 0xff)) << 43);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_Imm8_Y0(int num)
 {
   const unsigned int n = (unsigned int)num;
   return ((n & 0xff) << 12);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_Imm8_Y1(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0xff)) << 43);
-}
-
-static __inline tile_bundle_bits
-create_ImmOpcodeExtension_X0(int num)
-{
-  const unsigned int n = (unsigned int)num;
-  return ((n & 0x7f) << 20);
-}
-
-static __inline tile_bundle_bits
-create_ImmOpcodeExtension_X1(int num)
-{
-  const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0x7f)) << 51);
-}
-
-static __inline tile_bundle_bits
-create_ImmRROpcodeExtension_SN(int num)
-{
-  const unsigned int n = (unsigned int)num;
-  return ((n & 0x3) << 8);
-}
-
-static __inline tile_bundle_bits
-create_JOffLong_X1(int num)
-{
-  const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0x00007fff)) << 43) |
-         (((tile_bundle_bits)(n & 0x00018000)) << 20) |
-         (((tile_bundle_bits)(n & 0x001e0000)) << 14) |
-         (((tile_bundle_bits)(n & 0x07e00000)) << 16) |
-         (((tile_bundle_bits)(n & 0x18000000)) << 31);
-}
-
-static __inline tile_bundle_bits
-create_JOff_X1(int num)
-{
-  const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0x00007fff)) << 43) |
-         (((tile_bundle_bits)(n & 0x00018000)) << 20) |
-         (((tile_bundle_bits)(n & 0x001e0000)) << 14) |
-         (((tile_bundle_bits)(n & 0x07e00000)) << 16) |
-         (((tile_bundle_bits)(n & 0x08000000)) << 31);
-}
-
-static __inline tile_bundle_bits
-create_MF_Imm15_X1(int num)
-{
-  const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0x00003fff)) << 37) |
-         (((tile_bundle_bits)(n & 0x00004000)) << 44);
+  return (((tilegx_bundle_bits)(n & 0xff)) << 43);
 }
 
-static __inline tile_bundle_bits
-create_MMEnd_X0(int num)
+static __inline tilegx_bundle_bits
+create_JumpOff_X1(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return ((n & 0x1f) << 18);
+  return (((tilegx_bundle_bits)(n & 0x7ffffff)) << 31);
 }
 
-static __inline tile_bundle_bits
-create_MMEnd_X1(int num)
+static __inline tilegx_bundle_bits
+create_JumpOpcodeExtension_X1(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0x1f)) << 49);
+  return (((tilegx_bundle_bits)(n & 0x1)) << 58);
 }
 
-static __inline tile_bundle_bits
-create_MMStart_X0(int num)
+static __inline tilegx_bundle_bits
+create_MF_Imm14_X1(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return ((n & 0x1f) << 23);
+  return (((tilegx_bundle_bits)(n & 0x3fff)) << 37);
 }
 
-static __inline tile_bundle_bits
-create_MMStart_X1(int num)
+static __inline tilegx_bundle_bits
+create_MT_Imm14_X1(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0x1f)) << 54);
+  return (((tilegx_bundle_bits)(n & 0x0000003f)) << 31) |
+         (((tilegx_bundle_bits)(n & 0x00003fc0)) << 37);
 }
 
-static __inline tile_bundle_bits
-create_MT_Imm15_X1(int num)
-{
-  const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0x0000003f)) << 31) |
-         (((tile_bundle_bits)(n & 0x00003fc0)) << 37) |
-         (((tile_bundle_bits)(n & 0x00004000)) << 44);
-}
-
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_Mode(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0x1)) << 63);
+  return (((tilegx_bundle_bits)(n & 0x3)) << 62);
 }
 
-static __inline tile_bundle_bits
-create_NoRegOpcodeExtension_SN(int num)
-{
-  const unsigned int n = (unsigned int)num;
-  return ((n & 0xf) << 0);
-}
-
-static __inline tile_bundle_bits
-create_Opcode_SN(int num)
-{
-  const unsigned int n = (unsigned int)num;
-  return ((n & 0x3f) << 10);
-}
-
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_Opcode_X0(int num)
 {
   const unsigned int n = (unsigned int)num;
   return ((n & 0x7) << 28);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_Opcode_X1(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0xf)) << 59);
+  return (((tilegx_bundle_bits)(n & 0x7)) << 59);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_Opcode_Y0(int num)
 {
   const unsigned int n = (unsigned int)num;
   return ((n & 0xf) << 27);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_Opcode_Y1(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0xf)) << 59);
+  return (((tilegx_bundle_bits)(n & 0xf)) << 58);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_Opcode_Y2(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0x7)) << 56);
-}
-
-static __inline tile_bundle_bits
-create_RROpcodeExtension_SN(int num)
-{
-  const unsigned int n = (unsigned int)num;
-  return ((n & 0xf) << 4);
+  return ((n & 0x00000001) << 26) |
+         (((tilegx_bundle_bits)(n & 0x00000002)) << 56);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_RRROpcodeExtension_X0(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return ((n & 0x1ff) << 18);
+  return ((n & 0x3ff) << 18);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_RRROpcodeExtension_X1(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0x1ff)) << 49);
+  return (((tilegx_bundle_bits)(n & 0x3ff)) << 49);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_RRROpcodeExtension_Y0(int num)
 {
   const unsigned int n = (unsigned int)num;
   return ((n & 0x3) << 18);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_RRROpcodeExtension_Y1(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0x3)) << 49);
+  return (((tilegx_bundle_bits)(n & 0x3)) << 49);
 }
 
-static __inline tile_bundle_bits
-create_RouteOpcodeExtension_SN(int num)
+static __inline tilegx_bundle_bits
+create_ShAmt_X0(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return ((n & 0x3ff) << 0);
+  return ((n & 0x3f) << 12);
 }
 
-static __inline tile_bundle_bits
-create_S_X0(int num)
+static __inline tilegx_bundle_bits
+create_ShAmt_X1(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return ((n & 0x1) << 27);
+  return (((tilegx_bundle_bits)(n & 0x3f)) << 43);
 }
 
-static __inline tile_bundle_bits
-create_S_X1(int num)
+static __inline tilegx_bundle_bits
+create_ShAmt_Y0(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0x1)) << 58);
+  return ((n & 0x3f) << 12);
 }
 
-static __inline tile_bundle_bits
-create_ShAmt_X0(int num)
+static __inline tilegx_bundle_bits
+create_ShAmt_Y1(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return ((n & 0x1f) << 12);
+  return (((tilegx_bundle_bits)(n & 0x3f)) << 43);
 }
 
-static __inline tile_bundle_bits
-create_ShAmt_X1(int num)
+static __inline tilegx_bundle_bits
+create_ShiftOpcodeExtension_X0(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0x1f)) << 43);
+  return ((n & 0x3ff) << 18);
 }
 
-static __inline tile_bundle_bits
-create_ShAmt_Y0(int num)
+static __inline tilegx_bundle_bits
+create_ShiftOpcodeExtension_X1(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return ((n & 0x1f) << 12);
+  return (((tilegx_bundle_bits)(n & 0x3ff)) << 49);
 }
 
-static __inline tile_bundle_bits
-create_ShAmt_Y1(int num)
+static __inline tilegx_bundle_bits
+create_ShiftOpcodeExtension_Y0(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0x1f)) << 43);
+  return ((n & 0x3) << 18);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
+create_ShiftOpcodeExtension_Y1(int num)
+{
+  const unsigned int n = (unsigned int)num;
+  return (((tilegx_bundle_bits)(n & 0x3)) << 49);
+}
+
+static __inline tilegx_bundle_bits
 create_SrcA_X0(int num)
 {
   const unsigned int n = (unsigned int)num;
   return ((n & 0x3f) << 6);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_SrcA_X1(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0x3f)) << 37);
+  return (((tilegx_bundle_bits)(n & 0x3f)) << 37);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_SrcA_Y0(int num)
 {
   const unsigned int n = (unsigned int)num;
   return ((n & 0x3f) << 6);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_SrcA_Y1(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0x3f)) << 37);
+  return (((tilegx_bundle_bits)(n & 0x3f)) << 37);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_SrcA_Y2(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return ((n & 0x00000001) << 26) |
-         (((tile_bundle_bits)(n & 0x0000003e)) << 50);
+  return ((n & 0x3f) << 20);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_SrcBDest_Y2(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return ((n & 0x3f) << 20);
+  return (((tilegx_bundle_bits)(n & 0x3f)) << 51);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_SrcB_X0(int num)
 {
   const unsigned int n = (unsigned int)num;
   return ((n & 0x3f) << 12);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_SrcB_X1(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0x3f)) << 43);
+  return (((tilegx_bundle_bits)(n & 0x3f)) << 43);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_SrcB_Y0(int num)
 {
   const unsigned int n = (unsigned int)num;
   return ((n & 0x3f) << 12);
 }
 
-static __inline tile_bundle_bits
+static __inline tilegx_bundle_bits
 create_SrcB_Y1(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0x3f)) << 43);
+  return (((tilegx_bundle_bits)(n & 0x3f)) << 43);
 }
 
-static __inline tile_bundle_bits
-create_Src_SN(int num)
+static __inline tilegx_bundle_bits
+create_UnaryOpcodeExtension_X0(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return ((n & 0x3) << 0);
-}
-
-static __inline tile_bundle_bits
-create_UnOpcodeExtension_X0(int num)
-{
-  const unsigned int n = (unsigned int)num;
-  return ((n & 0x1f) << 12);
-}
-
-static __inline tile_bundle_bits
-create_UnOpcodeExtension_X1(int num)
-{
-  const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0x1f)) << 43);
-}
-
-static __inline tile_bundle_bits
-create_UnOpcodeExtension_Y0(int num)
-{
-  const unsigned int n = (unsigned int)num;
-  return ((n & 0x1f) << 12);
-}
-
-static __inline tile_bundle_bits
-create_UnOpcodeExtension_Y1(int num)
-{
-  const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0x1f)) << 43);
-}
-
-static __inline tile_bundle_bits
-create_UnShOpcodeExtension_X0(int num)
-{
-  const unsigned int n = (unsigned int)num;
-  return ((n & 0x3ff) << 17);
+  return ((n & 0x3f) << 12);
 }
 
-static __inline tile_bundle_bits
-create_UnShOpcodeExtension_X1(int num)
+static __inline tilegx_bundle_bits
+create_UnaryOpcodeExtension_X1(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0x3ff)) << 48);
+  return (((tilegx_bundle_bits)(n & 0x3f)) << 43);
 }
 
-static __inline tile_bundle_bits
-create_UnShOpcodeExtension_Y0(int num)
+static __inline tilegx_bundle_bits
+create_UnaryOpcodeExtension_Y0(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return ((n & 0x7) << 17);
+  return ((n & 0x3f) << 12);
 }
 
-static __inline tile_bundle_bits
-create_UnShOpcodeExtension_Y1(int num)
+static __inline tilegx_bundle_bits
+create_UnaryOpcodeExtension_Y1(int num)
 {
   const unsigned int n = (unsigned int)num;
-  return (((tile_bundle_bits)(n & 0x7)) << 48);
+  return (((tilegx_bundle_bits)(n & 0x3f)) << 43);
 }
 
 
-
 typedef enum
 {
-  TILE_PIPELINE_X0,
-  TILE_PIPELINE_X1,
-  TILE_PIPELINE_Y0,
-  TILE_PIPELINE_Y1,
-  TILE_PIPELINE_Y2,
-} tile_pipeline;
+  TILEGX_PIPELINE_X0,
+  TILEGX_PIPELINE_X1,
+  TILEGX_PIPELINE_Y0,
+  TILEGX_PIPELINE_Y1,
+  TILEGX_PIPELINE_Y2,
+} tilegx_pipeline;
 
-#define tile_is_x_pipeline(p) ((int)(p) <= (int)TILE_PIPELINE_X1)
+#define tilegx_is_x_pipeline(p) ((int)(p) <= (int)TILEGX_PIPELINE_X1)
 
 typedef enum
 {
-  TILE_OP_TYPE_REGISTER,
-  TILE_OP_TYPE_IMMEDIATE,
-  TILE_OP_TYPE_ADDRESS,
-  TILE_OP_TYPE_SPR
-} tile_operand_type;
+  TILEGX_OP_TYPE_REGISTER,
+  TILEGX_OP_TYPE_IMMEDIATE,
+  TILEGX_OP_TYPE_ADDRESS,
+  TILEGX_OP_TYPE_SPR
+} tilegx_operand_type;
 
-/* This is the bit that determines if a bundle is in the Y encoding. */
-#define TILE_BUNDLE_Y_ENCODING_MASK ((tile_bundle_bits)1 << 63)
+/* These are the bits that determine if a bundle is in the X encoding. */
+#define TILEGX_BUNDLE_MODE_MASK ((tilegx_bundle_bits)3 << 62)
 
 enum
 {
   /* Maximum number of instructions in a bundle (2 for X, 3 for Y). */
-  TILE_MAX_INSTRUCTIONS_PER_BUNDLE = 3,
+  TILEGX_MAX_INSTRUCTIONS_PER_BUNDLE = 3,
 
   /* How many different pipeline encodings are there? X0, X1, Y0, Y1, Y2. */
-  TILE_NUM_PIPELINE_ENCODINGS = 5,
+  TILEGX_NUM_PIPELINE_ENCODINGS = 5,
 
-  /* Log base 2 of TILE_BUNDLE_SIZE_IN_BYTES. */
-  TILE_LOG2_BUNDLE_SIZE_IN_BYTES = 3,
+  /* Log base 2 of TILEGX_BUNDLE_SIZE_IN_BYTES. */
+  TILEGX_LOG2_BUNDLE_SIZE_IN_BYTES = 3,
 
   /* Instructions take this many bytes. */
-  TILE_BUNDLE_SIZE_IN_BYTES = 1 << TILE_LOG2_BUNDLE_SIZE_IN_BYTES,
+  TILEGX_BUNDLE_SIZE_IN_BYTES = 1 << TILEGX_LOG2_BUNDLE_SIZE_IN_BYTES,
 
-  /* Log base 2 of TILE_BUNDLE_ALIGNMENT_IN_BYTES. */
-  TILE_LOG2_BUNDLE_ALIGNMENT_IN_BYTES = 3,
+  /* Log base 2 of TILEGX_BUNDLE_ALIGNMENT_IN_BYTES. */
+  TILEGX_LOG2_BUNDLE_ALIGNMENT_IN_BYTES = 3,
 
   /* Bundles should be aligned modulo this number of bytes. */
-  TILE_BUNDLE_ALIGNMENT_IN_BYTES =
-    (1 << TILE_LOG2_BUNDLE_ALIGNMENT_IN_BYTES),
-
-  /* Log base 2 of TILE_SN_INSTRUCTION_SIZE_IN_BYTES. */
-  TILE_LOG2_SN_INSTRUCTION_SIZE_IN_BYTES = 1,
-
-  /* Static network instructions take this many bytes. */
-  TILE_SN_INSTRUCTION_SIZE_IN_BYTES =
-    (1 << TILE_LOG2_SN_INSTRUCTION_SIZE_IN_BYTES),
+  TILEGX_BUNDLE_ALIGNMENT_IN_BYTES =
+    (1 << TILEGX_LOG2_BUNDLE_ALIGNMENT_IN_BYTES),
 
   /* Number of registers (some are magic, such as network I/O). */
-  TILE_NUM_REGISTERS = 64,
-
-  /* Number of static network registers. */
-  TILE_NUM_SN_REGISTERS = 4
+  TILEGX_NUM_REGISTERS = 64,
 };
 
 
-struct tile_operand
+struct tilegx_operand
 {
   /* Is this operand a register, immediate or address? */
-  tile_operand_type type;
+  tilegx_operand_type type;
 
   /* The default relocation type for this operand.  */
   signed int default_reloc : 16;
@@ -1437,27 +1180,27 @@ struct tile_operand
   unsigned int rightshift : 2;
 
   /* Return the bits for this operand to be ORed into an existing bundle. */
-  tile_bundle_bits (*insert) (int op);
+  tilegx_bundle_bits (*insert) (int op);
 
   /* Extract this operand and return it. */
-  unsigned int (*extract) (tile_bundle_bits bundle);
+  unsigned int (*extract) (tilegx_bundle_bits bundle);
 };
 
 
-extern const struct tile_operand tile_operands[];
+extern const struct tilegx_operand tilegx_operands[];
 
 /* One finite-state machine per pipe for rapid instruction decoding. */
 extern const unsigned short * const
-tile_bundle_decoder_fsms[TILE_NUM_PIPELINE_ENCODINGS];
+tilegx_bundle_decoder_fsms[TILEGX_NUM_PIPELINE_ENCODINGS];
 
 
-struct tile_opcode
+struct tilegx_opcode
 {
   /* The opcode mnemonic, e.g. "add" */
   const char *name;
 
   /* The enum value for this mnemonic. */
-  tile_mnemonic mnemonic;
+  tilegx_mnemonic mnemonic;
 
   /* A bit mask of which of the five pipes this instruction
      is compatible with:
@@ -1478,29 +1221,28 @@ struct tile_opcode
   unsigned char can_bundle;
 
   /* The description of the operands. Each of these is an
-   * index into the tile_operands[] table. */
-  unsigned char operands[TILE_NUM_PIPELINE_ENCODINGS][TILE_MAX_OPERANDS];
+   * index into the tilegx_operands[] table. */
+  unsigned char operands[TILEGX_NUM_PIPELINE_ENCODINGS][TILEGX_MAX_OPERANDS];
 
 };
 
-extern const struct tile_opcode tile_opcodes[];
-
+extern const struct tilegx_opcode tilegx_opcodes[];
 
 /* Used for non-textual disassembly into structs. */
-struct tile_decoded_instruction
+struct tilegx_decoded_instruction
 {
-  const struct tile_opcode *opcode;
-  const struct tile_operand *operands[TILE_MAX_OPERANDS];
-  int operand_values[TILE_MAX_OPERANDS];
+  const struct tilegx_opcode *opcode;
+  const struct tilegx_operand *operands[TILEGX_MAX_OPERANDS];
+  long long operand_values[TILEGX_MAX_OPERANDS];
 };
 
 
 /* Disassemble a bundle into a struct for machine processing. */
-extern int parse_insn_tile(tile_bundle_bits bits,
-                           unsigned int pc,
-                           struct tile_decoded_instruction
-                           decoded[TILE_MAX_INSTRUCTIONS_PER_BUNDLE]);
+extern int parse_insn_tilegx(tilegx_bundle_bits bits,
+                             unsigned long long pc,
+                             struct tilegx_decoded_instruction
+                             decoded[TILEGX_MAX_INSTRUCTIONS_PER_BUNDLE]);
 
 
 
-#endif /* opcode_tile_h */
+#endif /* opcode_tilegx_h */
diff --git a/arch/tile/include/asm/opcode_constants_64.h b/arch/tile/include/asm/opcode_constants_64.h
index 227d033b180c..710192869476 100644
--- a/arch/tile/include/asm/opcode_constants_64.h
+++ b/arch/tile/include/asm/opcode_constants_64.h
@@ -1,5 +1,5 @@
 /*
- * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ * Copyright 2011 Tilera Corporation. All Rights Reserved.
  *
  *   This program is free software; you can redistribute it and/or
  *   modify it under the terms of the GNU General Public License
@@ -19,462 +19,591 @@
 #define _TILE_OPCODE_CONSTANTS_H
 enum
 {
-  ADDBS_U_SPECIAL_0_OPCODE_X0 = 98,
-  ADDBS_U_SPECIAL_0_OPCODE_X1 = 68,
-  ADDB_SPECIAL_0_OPCODE_X0 = 1,
-  ADDB_SPECIAL_0_OPCODE_X1 = 1,
-  ADDHS_SPECIAL_0_OPCODE_X0 = 99,
-  ADDHS_SPECIAL_0_OPCODE_X1 = 69,
-  ADDH_SPECIAL_0_OPCODE_X0 = 2,
-  ADDH_SPECIAL_0_OPCODE_X1 = 2,
-  ADDIB_IMM_0_OPCODE_X0 = 1,
-  ADDIB_IMM_0_OPCODE_X1 = 1,
-  ADDIH_IMM_0_OPCODE_X0 = 2,
-  ADDIH_IMM_0_OPCODE_X1 = 2,
-  ADDI_IMM_0_OPCODE_X0 = 3,
-  ADDI_IMM_0_OPCODE_X1 = 3,
-  ADDI_IMM_1_OPCODE_SN = 1,
-  ADDI_OPCODE_Y0 = 9,
-  ADDI_OPCODE_Y1 = 7,
-  ADDLIS_OPCODE_X0 = 1,
-  ADDLIS_OPCODE_X1 = 2,
-  ADDLI_OPCODE_X0 = 2,
-  ADDLI_OPCODE_X1 = 3,
-  ADDS_SPECIAL_0_OPCODE_X0 = 96,
-  ADDS_SPECIAL_0_OPCODE_X1 = 66,
-  ADD_SPECIAL_0_OPCODE_X0 = 3,
-  ADD_SPECIAL_0_OPCODE_X1 = 3,
-  ADD_SPECIAL_0_OPCODE_Y0 = 0,
-  ADD_SPECIAL_0_OPCODE_Y1 = 0,
-  ADIFFB_U_SPECIAL_0_OPCODE_X0 = 4,
-  ADIFFH_SPECIAL_0_OPCODE_X0 = 5,
-  ANDI_IMM_0_OPCODE_X0 = 1,
-  ANDI_IMM_0_OPCODE_X1 = 4,
-  ANDI_OPCODE_Y0 = 10,
-  ANDI_OPCODE_Y1 = 8,
-  AND_SPECIAL_0_OPCODE_X0 = 6,
-  AND_SPECIAL_0_OPCODE_X1 = 4,
-  AND_SPECIAL_2_OPCODE_Y0 = 0,
-  AND_SPECIAL_2_OPCODE_Y1 = 0,
-  AULI_OPCODE_X0 = 3,
-  AULI_OPCODE_X1 = 4,
-  AVGB_U_SPECIAL_0_OPCODE_X0 = 7,
-  AVGH_SPECIAL_0_OPCODE_X0 = 8,
-  BBNST_BRANCH_OPCODE_X1 = 15,
-  BBNS_BRANCH_OPCODE_X1 = 14,
-  BBNS_OPCODE_SN = 63,
-  BBST_BRANCH_OPCODE_X1 = 13,
-  BBS_BRANCH_OPCODE_X1 = 12,
-  BBS_OPCODE_SN = 62,
-  BGEZT_BRANCH_OPCODE_X1 = 7,
-  BGEZ_BRANCH_OPCODE_X1 = 6,
-  BGEZ_OPCODE_SN = 61,
-  BGZT_BRANCH_OPCODE_X1 = 5,
-  BGZ_BRANCH_OPCODE_X1 = 4,
-  BGZ_OPCODE_SN = 58,
-  BITX_UN_0_SHUN_0_OPCODE_X0 = 1,
-  BITX_UN_0_SHUN_0_OPCODE_Y0 = 1,
-  BLEZT_BRANCH_OPCODE_X1 = 11,
-  BLEZ_BRANCH_OPCODE_X1 = 10,
-  BLEZ_OPCODE_SN = 59,
-  BLZT_BRANCH_OPCODE_X1 = 9,
-  BLZ_BRANCH_OPCODE_X1 = 8,
-  BLZ_OPCODE_SN = 60,
-  BNZT_BRANCH_OPCODE_X1 = 3,
-  BNZ_BRANCH_OPCODE_X1 = 2,
-  BNZ_OPCODE_SN = 57,
-  BPT_NOREG_RR_IMM_0_OPCODE_SN = 1,
-  BRANCH_OPCODE_X1 = 5,
-  BYTEX_UN_0_SHUN_0_OPCODE_X0 = 2,
-  BYTEX_UN_0_SHUN_0_OPCODE_Y0 = 2,
-  BZT_BRANCH_OPCODE_X1 = 1,
-  BZ_BRANCH_OPCODE_X1 = 0,
-  BZ_OPCODE_SN = 56,
-  CLZ_UN_0_SHUN_0_OPCODE_X0 = 3,
-  CLZ_UN_0_SHUN_0_OPCODE_Y0 = 3,
-  CRC32_32_SPECIAL_0_OPCODE_X0 = 9,
-  CRC32_8_SPECIAL_0_OPCODE_X0 = 10,
-  CTZ_UN_0_SHUN_0_OPCODE_X0 = 4,
-  CTZ_UN_0_SHUN_0_OPCODE_Y0 = 4,
-  DRAIN_UN_0_SHUN_0_OPCODE_X1 = 1,
-  DTLBPR_UN_0_SHUN_0_OPCODE_X1 = 2,
-  DWORD_ALIGN_SPECIAL_0_OPCODE_X0 = 95,
-  FINV_UN_0_SHUN_0_OPCODE_X1 = 3,
-  FLUSH_UN_0_SHUN_0_OPCODE_X1 = 4,
-  FNOP_NOREG_RR_IMM_0_OPCODE_SN = 3,
-  FNOP_UN_0_SHUN_0_OPCODE_X0 = 5,
-  FNOP_UN_0_SHUN_0_OPCODE_X1 = 5,
-  FNOP_UN_0_SHUN_0_OPCODE_Y0 = 5,
-  FNOP_UN_0_SHUN_0_OPCODE_Y1 = 1,
-  HALT_NOREG_RR_IMM_0_OPCODE_SN = 0,
-  ICOH_UN_0_SHUN_0_OPCODE_X1 = 6,
-  ILL_UN_0_SHUN_0_OPCODE_X1 = 7,
-  ILL_UN_0_SHUN_0_OPCODE_Y1 = 2,
-  IMM_0_OPCODE_SN = 0,
-  IMM_0_OPCODE_X0 = 4,
-  IMM_0_OPCODE_X1 = 6,
-  IMM_1_OPCODE_SN = 1,
-  IMM_OPCODE_0_X0 = 5,
-  INTHB_SPECIAL_0_OPCODE_X0 = 11,
-  INTHB_SPECIAL_0_OPCODE_X1 = 5,
-  INTHH_SPECIAL_0_OPCODE_X0 = 12,
-  INTHH_SPECIAL_0_OPCODE_X1 = 6,
-  INTLB_SPECIAL_0_OPCODE_X0 = 13,
-  INTLB_SPECIAL_0_OPCODE_X1 = 7,
-  INTLH_SPECIAL_0_OPCODE_X0 = 14,
-  INTLH_SPECIAL_0_OPCODE_X1 = 8,
-  INV_UN_0_SHUN_0_OPCODE_X1 = 8,
-  IRET_UN_0_SHUN_0_OPCODE_X1 = 9,
-  JALB_OPCODE_X1 = 13,
-  JALF_OPCODE_X1 = 12,
-  JALRP_SPECIAL_0_OPCODE_X1 = 9,
-  JALRR_IMM_1_OPCODE_SN = 3,
-  JALR_RR_IMM_0_OPCODE_SN = 5,
-  JALR_SPECIAL_0_OPCODE_X1 = 10,
-  JB_OPCODE_X1 = 11,
-  JF_OPCODE_X1 = 10,
-  JRP_SPECIAL_0_OPCODE_X1 = 11,
-  JRR_IMM_1_OPCODE_SN = 2,
-  JR_RR_IMM_0_OPCODE_SN = 4,
-  JR_SPECIAL_0_OPCODE_X1 = 12,
-  LBADD_IMM_0_OPCODE_X1 = 22,
-  LBADD_U_IMM_0_OPCODE_X1 = 23,
-  LB_OPCODE_Y2 = 0,
-  LB_UN_0_SHUN_0_OPCODE_X1 = 10,
-  LB_U_OPCODE_Y2 = 1,
-  LB_U_UN_0_SHUN_0_OPCODE_X1 = 11,
-  LHADD_IMM_0_OPCODE_X1 = 24,
-  LHADD_U_IMM_0_OPCODE_X1 = 25,
-  LH_OPCODE_Y2 = 2,
-  LH_UN_0_SHUN_0_OPCODE_X1 = 12,
-  LH_U_OPCODE_Y2 = 3,
-  LH_U_UN_0_SHUN_0_OPCODE_X1 = 13,
-  LNK_SPECIAL_0_OPCODE_X1 = 13,
-  LWADD_IMM_0_OPCODE_X1 = 26,
-  LWADD_NA_IMM_0_OPCODE_X1 = 27,
-  LW_NA_UN_0_SHUN_0_OPCODE_X1 = 24,
-  LW_OPCODE_Y2 = 4,
-  LW_UN_0_SHUN_0_OPCODE_X1 = 14,
-  MAXB_U_SPECIAL_0_OPCODE_X0 = 15,
-  MAXB_U_SPECIAL_0_OPCODE_X1 = 14,
-  MAXH_SPECIAL_0_OPCODE_X0 = 16,
-  MAXH_SPECIAL_0_OPCODE_X1 = 15,
-  MAXIB_U_IMM_0_OPCODE_X0 = 4,
-  MAXIB_U_IMM_0_OPCODE_X1 = 5,
-  MAXIH_IMM_0_OPCODE_X0 = 5,
-  MAXIH_IMM_0_OPCODE_X1 = 6,
-  MFSPR_IMM_0_OPCODE_X1 = 7,
-  MF_UN_0_SHUN_0_OPCODE_X1 = 15,
-  MINB_U_SPECIAL_0_OPCODE_X0 = 17,
-  MINB_U_SPECIAL_0_OPCODE_X1 = 16,
-  MINH_SPECIAL_0_OPCODE_X0 = 18,
-  MINH_SPECIAL_0_OPCODE_X1 = 17,
-  MINIB_U_IMM_0_OPCODE_X0 = 6,
-  MINIB_U_IMM_0_OPCODE_X1 = 8,
-  MINIH_IMM_0_OPCODE_X0 = 7,
-  MINIH_IMM_0_OPCODE_X1 = 9,
-  MM_OPCODE_X0 = 6,
-  MM_OPCODE_X1 = 7,
-  MNZB_SPECIAL_0_OPCODE_X0 = 19,
-  MNZB_SPECIAL_0_OPCODE_X1 = 18,
-  MNZH_SPECIAL_0_OPCODE_X0 = 20,
-  MNZH_SPECIAL_0_OPCODE_X1 = 19,
-  MNZ_SPECIAL_0_OPCODE_X0 = 21,
-  MNZ_SPECIAL_0_OPCODE_X1 = 20,
-  MNZ_SPECIAL_1_OPCODE_Y0 = 0,
-  MNZ_SPECIAL_1_OPCODE_Y1 = 1,
-  MOVEI_IMM_1_OPCODE_SN = 0,
-  MOVE_RR_IMM_0_OPCODE_SN = 8,
-  MTSPR_IMM_0_OPCODE_X1 = 10,
-  MULHHA_SS_SPECIAL_0_OPCODE_X0 = 22,
-  MULHHA_SS_SPECIAL_7_OPCODE_Y0 = 0,
-  MULHHA_SU_SPECIAL_0_OPCODE_X0 = 23,
-  MULHHA_UU_SPECIAL_0_OPCODE_X0 = 24,
-  MULHHA_UU_SPECIAL_7_OPCODE_Y0 = 1,
-  MULHHSA_UU_SPECIAL_0_OPCODE_X0 = 25,
-  MULHH_SS_SPECIAL_0_OPCODE_X0 = 26,
-  MULHH_SS_SPECIAL_6_OPCODE_Y0 = 0,
-  MULHH_SU_SPECIAL_0_OPCODE_X0 = 27,
-  MULHH_UU_SPECIAL_0_OPCODE_X0 = 28,
-  MULHH_UU_SPECIAL_6_OPCODE_Y0 = 1,
-  MULHLA_SS_SPECIAL_0_OPCODE_X0 = 29,
-  MULHLA_SU_SPECIAL_0_OPCODE_X0 = 30,
-  MULHLA_US_SPECIAL_0_OPCODE_X0 = 31,
-  MULHLA_UU_SPECIAL_0_OPCODE_X0 = 32,
-  MULHLSA_UU_SPECIAL_0_OPCODE_X0 = 33,
-  MULHLSA_UU_SPECIAL_5_OPCODE_Y0 = 0,
-  MULHL_SS_SPECIAL_0_OPCODE_X0 = 34,
-  MULHL_SU_SPECIAL_0_OPCODE_X0 = 35,
-  MULHL_US_SPECIAL_0_OPCODE_X0 = 36,
-  MULHL_UU_SPECIAL_0_OPCODE_X0 = 37,
-  MULLLA_SS_SPECIAL_0_OPCODE_X0 = 38,
-  MULLLA_SS_SPECIAL_7_OPCODE_Y0 = 2,
-  MULLLA_SU_SPECIAL_0_OPCODE_X0 = 39,
-  MULLLA_UU_SPECIAL_0_OPCODE_X0 = 40,
-  MULLLA_UU_SPECIAL_7_OPCODE_Y0 = 3,
-  MULLLSA_UU_SPECIAL_0_OPCODE_X0 = 41,
-  MULLL_SS_SPECIAL_0_OPCODE_X0 = 42,
-  MULLL_SS_SPECIAL_6_OPCODE_Y0 = 2,
-  MULLL_SU_SPECIAL_0_OPCODE_X0 = 43,
-  MULLL_UU_SPECIAL_0_OPCODE_X0 = 44,
-  MULLL_UU_SPECIAL_6_OPCODE_Y0 = 3,
-  MVNZ_SPECIAL_0_OPCODE_X0 = 45,
-  MVNZ_SPECIAL_1_OPCODE_Y0 = 1,
-  MVZ_SPECIAL_0_OPCODE_X0 = 46,
-  MVZ_SPECIAL_1_OPCODE_Y0 = 2,
-  MZB_SPECIAL_0_OPCODE_X0 = 47,
-  MZB_SPECIAL_0_OPCODE_X1 = 21,
-  MZH_SPECIAL_0_OPCODE_X0 = 48,
-  MZH_SPECIAL_0_OPCODE_X1 = 22,
-  MZ_SPECIAL_0_OPCODE_X0 = 49,
-  MZ_SPECIAL_0_OPCODE_X1 = 23,
-  MZ_SPECIAL_1_OPCODE_Y0 = 3,
-  MZ_SPECIAL_1_OPCODE_Y1 = 2,
-  NAP_UN_0_SHUN_0_OPCODE_X1 = 16,
-  NOP_NOREG_RR_IMM_0_OPCODE_SN = 2,
-  NOP_UN_0_SHUN_0_OPCODE_X0 = 6,
-  NOP_UN_0_SHUN_0_OPCODE_X1 = 17,
-  NOP_UN_0_SHUN_0_OPCODE_Y0 = 6,
-  NOP_UN_0_SHUN_0_OPCODE_Y1 = 3,
-  NOREG_RR_IMM_0_OPCODE_SN = 0,
-  NOR_SPECIAL_0_OPCODE_X0 = 50,
-  NOR_SPECIAL_0_OPCODE_X1 = 24,
-  NOR_SPECIAL_2_OPCODE_Y0 = 1,
-  NOR_SPECIAL_2_OPCODE_Y1 = 1,
-  ORI_IMM_0_OPCODE_X0 = 8,
-  ORI_IMM_0_OPCODE_X1 = 11,
-  ORI_OPCODE_Y0 = 11,
-  ORI_OPCODE_Y1 = 9,
-  OR_SPECIAL_0_OPCODE_X0 = 51,
-  OR_SPECIAL_0_OPCODE_X1 = 25,
-  OR_SPECIAL_2_OPCODE_Y0 = 2,
-  OR_SPECIAL_2_OPCODE_Y1 = 2,
-  PACKBS_U_SPECIAL_0_OPCODE_X0 = 103,
-  PACKBS_U_SPECIAL_0_OPCODE_X1 = 73,
-  PACKHB_SPECIAL_0_OPCODE_X0 = 52,
-  PACKHB_SPECIAL_0_OPCODE_X1 = 26,
-  PACKHS_SPECIAL_0_OPCODE_X0 = 102,
-  PACKHS_SPECIAL_0_OPCODE_X1 = 72,
-  PACKLB_SPECIAL_0_OPCODE_X0 = 53,
-  PACKLB_SPECIAL_0_OPCODE_X1 = 27,
-  PCNT_UN_0_SHUN_0_OPCODE_X0 = 7,
-  PCNT_UN_0_SHUN_0_OPCODE_Y0 = 7,
-  RLI_SHUN_0_OPCODE_X0 = 1,
-  RLI_SHUN_0_OPCODE_X1 = 1,
-  RLI_SHUN_0_OPCODE_Y0 = 1,
-  RLI_SHUN_0_OPCODE_Y1 = 1,
-  RL_SPECIAL_0_OPCODE_X0 = 54,
-  RL_SPECIAL_0_OPCODE_X1 = 28,
-  RL_SPECIAL_3_OPCODE_Y0 = 0,
-  RL_SPECIAL_3_OPCODE_Y1 = 0,
-  RR_IMM_0_OPCODE_SN = 0,
-  S1A_SPECIAL_0_OPCODE_X0 = 55,
-  S1A_SPECIAL_0_OPCODE_X1 = 29,
-  S1A_SPECIAL_0_OPCODE_Y0 = 1,
-  S1A_SPECIAL_0_OPCODE_Y1 = 1,
-  S2A_SPECIAL_0_OPCODE_X0 = 56,
-  S2A_SPECIAL_0_OPCODE_X1 = 30,
-  S2A_SPECIAL_0_OPCODE_Y0 = 2,
-  S2A_SPECIAL_0_OPCODE_Y1 = 2,
-  S3A_SPECIAL_0_OPCODE_X0 = 57,
-  S3A_SPECIAL_0_OPCODE_X1 = 31,
-  S3A_SPECIAL_5_OPCODE_Y0 = 1,
-  S3A_SPECIAL_5_OPCODE_Y1 = 1,
-  SADAB_U_SPECIAL_0_OPCODE_X0 = 58,
-  SADAH_SPECIAL_0_OPCODE_X0 = 59,
-  SADAH_U_SPECIAL_0_OPCODE_X0 = 60,
-  SADB_U_SPECIAL_0_OPCODE_X0 = 61,
-  SADH_SPECIAL_0_OPCODE_X0 = 62,
-  SADH_U_SPECIAL_0_OPCODE_X0 = 63,
-  SBADD_IMM_0_OPCODE_X1 = 28,
-  SB_OPCODE_Y2 = 5,
-  SB_SPECIAL_0_OPCODE_X1 = 32,
-  SEQB_SPECIAL_0_OPCODE_X0 = 64,
-  SEQB_SPECIAL_0_OPCODE_X1 = 33,
-  SEQH_SPECIAL_0_OPCODE_X0 = 65,
-  SEQH_SPECIAL_0_OPCODE_X1 = 34,
-  SEQIB_IMM_0_OPCODE_X0 = 9,
-  SEQIB_IMM_0_OPCODE_X1 = 12,
-  SEQIH_IMM_0_OPCODE_X0 = 10,
-  SEQIH_IMM_0_OPCODE_X1 = 13,
-  SEQI_IMM_0_OPCODE_X0 = 11,
-  SEQI_IMM_0_OPCODE_X1 = 14,
-  SEQI_OPCODE_Y0 = 12,
-  SEQI_OPCODE_Y1 = 10,
-  SEQ_SPECIAL_0_OPCODE_X0 = 66,
-  SEQ_SPECIAL_0_OPCODE_X1 = 35,
-  SEQ_SPECIAL_5_OPCODE_Y0 = 2,
-  SEQ_SPECIAL_5_OPCODE_Y1 = 2,
-  SHADD_IMM_0_OPCODE_X1 = 29,
-  SHL8II_IMM_0_OPCODE_SN = 3,
-  SHLB_SPECIAL_0_OPCODE_X0 = 67,
-  SHLB_SPECIAL_0_OPCODE_X1 = 36,
-  SHLH_SPECIAL_0_OPCODE_X0 = 68,
-  SHLH_SPECIAL_0_OPCODE_X1 = 37,
-  SHLIB_SHUN_0_OPCODE_X0 = 2,
-  SHLIB_SHUN_0_OPCODE_X1 = 2,
-  SHLIH_SHUN_0_OPCODE_X0 = 3,
-  SHLIH_SHUN_0_OPCODE_X1 = 3,
-  SHLI_SHUN_0_OPCODE_X0 = 4,
-  SHLI_SHUN_0_OPCODE_X1 = 4,
-  SHLI_SHUN_0_OPCODE_Y0 = 2,
-  SHLI_SHUN_0_OPCODE_Y1 = 2,
-  SHL_SPECIAL_0_OPCODE_X0 = 69,
-  SHL_SPECIAL_0_OPCODE_X1 = 38,
-  SHL_SPECIAL_3_OPCODE_Y0 = 1,
-  SHL_SPECIAL_3_OPCODE_Y1 = 1,
-  SHR1_RR_IMM_0_OPCODE_SN = 9,
-  SHRB_SPECIAL_0_OPCODE_X0 = 70,
-  SHRB_SPECIAL_0_OPCODE_X1 = 39,
-  SHRH_SPECIAL_0_OPCODE_X0 = 71,
-  SHRH_SPECIAL_0_OPCODE_X1 = 40,
-  SHRIB_SHUN_0_OPCODE_X0 = 5,
-  SHRIB_SHUN_0_OPCODE_X1 = 5,
-  SHRIH_SHUN_0_OPCODE_X0 = 6,
-  SHRIH_SHUN_0_OPCODE_X1 = 6,
-  SHRI_SHUN_0_OPCODE_X0 = 7,
-  SHRI_SHUN_0_OPCODE_X1 = 7,
-  SHRI_SHUN_0_OPCODE_Y0 = 3,
-  SHRI_SHUN_0_OPCODE_Y1 = 3,
-  SHR_SPECIAL_0_OPCODE_X0 = 72,
-  SHR_SPECIAL_0_OPCODE_X1 = 41,
-  SHR_SPECIAL_3_OPCODE_Y0 = 2,
-  SHR_SPECIAL_3_OPCODE_Y1 = 2,
-  SHUN_0_OPCODE_X0 = 7,
-  SHUN_0_OPCODE_X1 = 8,
-  SHUN_0_OPCODE_Y0 = 13,
-  SHUN_0_OPCODE_Y1 = 11,
-  SH_OPCODE_Y2 = 6,
-  SH_SPECIAL_0_OPCODE_X1 = 42,
-  SLTB_SPECIAL_0_OPCODE_X0 = 73,
-  SLTB_SPECIAL_0_OPCODE_X1 = 43,
-  SLTB_U_SPECIAL_0_OPCODE_X0 = 74,
-  SLTB_U_SPECIAL_0_OPCODE_X1 = 44,
-  SLTEB_SPECIAL_0_OPCODE_X0 = 75,
-  SLTEB_SPECIAL_0_OPCODE_X1 = 45,
-  SLTEB_U_SPECIAL_0_OPCODE_X0 = 76,
-  SLTEB_U_SPECIAL_0_OPCODE_X1 = 46,
-  SLTEH_SPECIAL_0_OPCODE_X0 = 77,
-  SLTEH_SPECIAL_0_OPCODE_X1 = 47,
-  SLTEH_U_SPECIAL_0_OPCODE_X0 = 78,
-  SLTEH_U_SPECIAL_0_OPCODE_X1 = 48,
-  SLTE_SPECIAL_0_OPCODE_X0 = 79,
-  SLTE_SPECIAL_0_OPCODE_X1 = 49,
-  SLTE_SPECIAL_4_OPCODE_Y0 = 0,
-  SLTE_SPECIAL_4_OPCODE_Y1 = 0,
-  SLTE_U_SPECIAL_0_OPCODE_X0 = 80,
-  SLTE_U_SPECIAL_0_OPCODE_X1 = 50,
-  SLTE_U_SPECIAL_4_OPCODE_Y0 = 1,
-  SLTE_U_SPECIAL_4_OPCODE_Y1 = 1,
-  SLTH_SPECIAL_0_OPCODE_X0 = 81,
-  SLTH_SPECIAL_0_OPCODE_X1 = 51,
-  SLTH_U_SPECIAL_0_OPCODE_X0 = 82,
-  SLTH_U_SPECIAL_0_OPCODE_X1 = 52,
-  SLTIB_IMM_0_OPCODE_X0 = 12,
-  SLTIB_IMM_0_OPCODE_X1 = 15,
-  SLTIB_U_IMM_0_OPCODE_X0 = 13,
-  SLTIB_U_IMM_0_OPCODE_X1 = 16,
-  SLTIH_IMM_0_OPCODE_X0 = 14,
-  SLTIH_IMM_0_OPCODE_X1 = 17,
-  SLTIH_U_IMM_0_OPCODE_X0 = 15,
-  SLTIH_U_IMM_0_OPCODE_X1 = 18,
-  SLTI_IMM_0_OPCODE_X0 = 16,
-  SLTI_IMM_0_OPCODE_X1 = 19,
-  SLTI_OPCODE_Y0 = 14,
-  SLTI_OPCODE_Y1 = 12,
-  SLTI_U_IMM_0_OPCODE_X0 = 17,
-  SLTI_U_IMM_0_OPCODE_X1 = 20,
-  SLTI_U_OPCODE_Y0 = 15,
-  SLTI_U_OPCODE_Y1 = 13,
-  SLT_SPECIAL_0_OPCODE_X0 = 83,
-  SLT_SPECIAL_0_OPCODE_X1 = 53,
-  SLT_SPECIAL_4_OPCODE_Y0 = 2,
-  SLT_SPECIAL_4_OPCODE_Y1 = 2,
-  SLT_U_SPECIAL_0_OPCODE_X0 = 84,
-  SLT_U_SPECIAL_0_OPCODE_X1 = 54,
-  SLT_U_SPECIAL_4_OPCODE_Y0 = 3,
-  SLT_U_SPECIAL_4_OPCODE_Y1 = 3,
-  SNEB_SPECIAL_0_OPCODE_X0 = 85,
-  SNEB_SPECIAL_0_OPCODE_X1 = 55,
-  SNEH_SPECIAL_0_OPCODE_X0 = 86,
-  SNEH_SPECIAL_0_OPCODE_X1 = 56,
-  SNE_SPECIAL_0_OPCODE_X0 = 87,
-  SNE_SPECIAL_0_OPCODE_X1 = 57,
-  SNE_SPECIAL_5_OPCODE_Y0 = 3,
-  SNE_SPECIAL_5_OPCODE_Y1 = 3,
-  SPECIAL_0_OPCODE_X0 = 0,
-  SPECIAL_0_OPCODE_X1 = 1,
-  SPECIAL_0_OPCODE_Y0 = 1,
-  SPECIAL_0_OPCODE_Y1 = 1,
-  SPECIAL_1_OPCODE_Y0 = 2,
-  SPECIAL_1_OPCODE_Y1 = 2,
-  SPECIAL_2_OPCODE_Y0 = 3,
-  SPECIAL_2_OPCODE_Y1 = 3,
-  SPECIAL_3_OPCODE_Y0 = 4,
-  SPECIAL_3_OPCODE_Y1 = 4,
-  SPECIAL_4_OPCODE_Y0 = 5,
-  SPECIAL_4_OPCODE_Y1 = 5,
-  SPECIAL_5_OPCODE_Y0 = 6,
-  SPECIAL_5_OPCODE_Y1 = 6,
-  SPECIAL_6_OPCODE_Y0 = 7,
-  SPECIAL_7_OPCODE_Y0 = 8,
-  SRAB_SPECIAL_0_OPCODE_X0 = 88,
-  SRAB_SPECIAL_0_OPCODE_X1 = 58,
-  SRAH_SPECIAL_0_OPCODE_X0 = 89,
-  SRAH_SPECIAL_0_OPCODE_X1 = 59,
-  SRAIB_SHUN_0_OPCODE_X0 = 8,
-  SRAIB_SHUN_0_OPCODE_X1 = 8,
-  SRAIH_SHUN_0_OPCODE_X0 = 9,
-  SRAIH_SHUN_0_OPCODE_X1 = 9,
-  SRAI_SHUN_0_OPCODE_X0 = 10,
-  SRAI_SHUN_0_OPCODE_X1 = 10,
-  SRAI_SHUN_0_OPCODE_Y0 = 4,
-  SRAI_SHUN_0_OPCODE_Y1 = 4,
-  SRA_SPECIAL_0_OPCODE_X0 = 90,
-  SRA_SPECIAL_0_OPCODE_X1 = 60,
-  SRA_SPECIAL_3_OPCODE_Y0 = 3,
-  SRA_SPECIAL_3_OPCODE_Y1 = 3,
-  SUBBS_U_SPECIAL_0_OPCODE_X0 = 100,
-  SUBBS_U_SPECIAL_0_OPCODE_X1 = 70,
-  SUBB_SPECIAL_0_OPCODE_X0 = 91,
-  SUBB_SPECIAL_0_OPCODE_X1 = 61,
-  SUBHS_SPECIAL_0_OPCODE_X0 = 101,
-  SUBHS_SPECIAL_0_OPCODE_X1 = 71,
-  SUBH_SPECIAL_0_OPCODE_X0 = 92,
-  SUBH_SPECIAL_0_OPCODE_X1 = 62,
-  SUBS_SPECIAL_0_OPCODE_X0 = 97,
-  SUBS_SPECIAL_0_OPCODE_X1 = 67,
-  SUB_SPECIAL_0_OPCODE_X0 = 93,
-  SUB_SPECIAL_0_OPCODE_X1 = 63,
-  SUB_SPECIAL_0_OPCODE_Y0 = 3,
-  SUB_SPECIAL_0_OPCODE_Y1 = 3,
-  SWADD_IMM_0_OPCODE_X1 = 30,
-  SWINT0_UN_0_SHUN_0_OPCODE_X1 = 18,
-  SWINT1_UN_0_SHUN_0_OPCODE_X1 = 19,
-  SWINT2_UN_0_SHUN_0_OPCODE_X1 = 20,
-  SWINT3_UN_0_SHUN_0_OPCODE_X1 = 21,
-  SW_OPCODE_Y2 = 7,
-  SW_SPECIAL_0_OPCODE_X1 = 64,
-  TBLIDXB0_UN_0_SHUN_0_OPCODE_X0 = 8,
-  TBLIDXB0_UN_0_SHUN_0_OPCODE_Y0 = 8,
-  TBLIDXB1_UN_0_SHUN_0_OPCODE_X0 = 9,
-  TBLIDXB1_UN_0_SHUN_0_OPCODE_Y0 = 9,
-  TBLIDXB2_UN_0_SHUN_0_OPCODE_X0 = 10,
-  TBLIDXB2_UN_0_SHUN_0_OPCODE_Y0 = 10,
-  TBLIDXB3_UN_0_SHUN_0_OPCODE_X0 = 11,
-  TBLIDXB3_UN_0_SHUN_0_OPCODE_Y0 = 11,
-  TNS_UN_0_SHUN_0_OPCODE_X1 = 22,
-  UN_0_SHUN_0_OPCODE_X0 = 11,
-  UN_0_SHUN_0_OPCODE_X1 = 11,
-  UN_0_SHUN_0_OPCODE_Y0 = 5,
-  UN_0_SHUN_0_OPCODE_Y1 = 5,
-  WH64_UN_0_SHUN_0_OPCODE_X1 = 23,
-  XORI_IMM_0_OPCODE_X0 = 2,
-  XORI_IMM_0_OPCODE_X1 = 21,
-  XOR_SPECIAL_0_OPCODE_X0 = 94,
-  XOR_SPECIAL_0_OPCODE_X1 = 65,
-  XOR_SPECIAL_2_OPCODE_Y0 = 3,
-  XOR_SPECIAL_2_OPCODE_Y1 = 3
+  ADDI_IMM8_OPCODE_X0 = 1,
+  ADDI_IMM8_OPCODE_X1 = 1,
+  ADDI_OPCODE_Y0 = 0,
+  ADDI_OPCODE_Y1 = 1,
+  ADDLI_OPCODE_X0 = 1,
+  ADDLI_OPCODE_X1 = 0,
+  ADDXI_IMM8_OPCODE_X0 = 2,
+  ADDXI_IMM8_OPCODE_X1 = 2,
+  ADDXI_OPCODE_Y0 = 1,
+  ADDXI_OPCODE_Y1 = 2,
+  ADDXLI_OPCODE_X0 = 2,
+  ADDXLI_OPCODE_X1 = 1,
+  ADDXSC_RRR_0_OPCODE_X0 = 1,
+  ADDXSC_RRR_0_OPCODE_X1 = 1,
+  ADDX_RRR_0_OPCODE_X0 = 2,
+  ADDX_RRR_0_OPCODE_X1 = 2,
+  ADDX_RRR_0_OPCODE_Y0 = 0,
+  ADDX_SPECIAL_0_OPCODE_Y1 = 0,
+  ADD_RRR_0_OPCODE_X0 = 3,
+  ADD_RRR_0_OPCODE_X1 = 3,
+  ADD_RRR_0_OPCODE_Y0 = 1,
+  ADD_SPECIAL_0_OPCODE_Y1 = 1,
+  ANDI_IMM8_OPCODE_X0 = 3,
+  ANDI_IMM8_OPCODE_X1 = 3,
+  ANDI_OPCODE_Y0 = 2,
+  ANDI_OPCODE_Y1 = 3,
+  AND_RRR_0_OPCODE_X0 = 4,
+  AND_RRR_0_OPCODE_X1 = 4,
+  AND_RRR_5_OPCODE_Y0 = 0,
+  AND_RRR_5_OPCODE_Y1 = 0,
+  BEQZT_BRANCH_OPCODE_X1 = 16,
+  BEQZ_BRANCH_OPCODE_X1 = 17,
+  BFEXTS_BF_OPCODE_X0 = 4,
+  BFEXTU_BF_OPCODE_X0 = 5,
+  BFINS_BF_OPCODE_X0 = 6,
+  BF_OPCODE_X0 = 3,
+  BGEZT_BRANCH_OPCODE_X1 = 18,
+  BGEZ_BRANCH_OPCODE_X1 = 19,
+  BGTZT_BRANCH_OPCODE_X1 = 20,
+  BGTZ_BRANCH_OPCODE_X1 = 21,
+  BLBCT_BRANCH_OPCODE_X1 = 22,
+  BLBC_BRANCH_OPCODE_X1 = 23,
+  BLBST_BRANCH_OPCODE_X1 = 24,
+  BLBS_BRANCH_OPCODE_X1 = 25,
+  BLEZT_BRANCH_OPCODE_X1 = 26,
+  BLEZ_BRANCH_OPCODE_X1 = 27,
+  BLTZT_BRANCH_OPCODE_X1 = 28,
+  BLTZ_BRANCH_OPCODE_X1 = 29,
+  BNEZT_BRANCH_OPCODE_X1 = 30,
+  BNEZ_BRANCH_OPCODE_X1 = 31,
+  BRANCH_OPCODE_X1 = 2,
+  CMOVEQZ_RRR_0_OPCODE_X0 = 5,
+  CMOVEQZ_RRR_4_OPCODE_Y0 = 0,
+  CMOVNEZ_RRR_0_OPCODE_X0 = 6,
+  CMOVNEZ_RRR_4_OPCODE_Y0 = 1,
+  CMPEQI_IMM8_OPCODE_X0 = 4,
+  CMPEQI_IMM8_OPCODE_X1 = 4,
+  CMPEQI_OPCODE_Y0 = 3,
+  CMPEQI_OPCODE_Y1 = 4,
+  CMPEQ_RRR_0_OPCODE_X0 = 7,
+  CMPEQ_RRR_0_OPCODE_X1 = 5,
+  CMPEQ_RRR_3_OPCODE_Y0 = 0,
+  CMPEQ_RRR_3_OPCODE_Y1 = 2,
+  CMPEXCH4_RRR_0_OPCODE_X1 = 6,
+  CMPEXCH_RRR_0_OPCODE_X1 = 7,
+  CMPLES_RRR_0_OPCODE_X0 = 8,
+  CMPLES_RRR_0_OPCODE_X1 = 8,
+  CMPLES_RRR_2_OPCODE_Y0 = 0,
+  CMPLES_RRR_2_OPCODE_Y1 = 0,
+  CMPLEU_RRR_0_OPCODE_X0 = 9,
+  CMPLEU_RRR_0_OPCODE_X1 = 9,
+  CMPLEU_RRR_2_OPCODE_Y0 = 1,
+  CMPLEU_RRR_2_OPCODE_Y1 = 1,
+  CMPLTSI_IMM8_OPCODE_X0 = 5,
+  CMPLTSI_IMM8_OPCODE_X1 = 5,
+  CMPLTSI_OPCODE_Y0 = 4,
+  CMPLTSI_OPCODE_Y1 = 5,
+  CMPLTS_RRR_0_OPCODE_X0 = 10,
+  CMPLTS_RRR_0_OPCODE_X1 = 10,
+  CMPLTS_RRR_2_OPCODE_Y0 = 2,
+  CMPLTS_RRR_2_OPCODE_Y1 = 2,
+  CMPLTUI_IMM8_OPCODE_X0 = 6,
+  CMPLTUI_IMM8_OPCODE_X1 = 6,
+  CMPLTU_RRR_0_OPCODE_X0 = 11,
+  CMPLTU_RRR_0_OPCODE_X1 = 11,
+  CMPLTU_RRR_2_OPCODE_Y0 = 3,
+  CMPLTU_RRR_2_OPCODE_Y1 = 3,
+  CMPNE_RRR_0_OPCODE_X0 = 12,
+  CMPNE_RRR_0_OPCODE_X1 = 12,
+  CMPNE_RRR_3_OPCODE_Y0 = 1,
+  CMPNE_RRR_3_OPCODE_Y1 = 3,
+  CMULAF_RRR_0_OPCODE_X0 = 13,
+  CMULA_RRR_0_OPCODE_X0 = 14,
+  CMULFR_RRR_0_OPCODE_X0 = 15,
+  CMULF_RRR_0_OPCODE_X0 = 16,
+  CMULHR_RRR_0_OPCODE_X0 = 17,
+  CMULH_RRR_0_OPCODE_X0 = 18,
+  CMUL_RRR_0_OPCODE_X0 = 19,
+  CNTLZ_UNARY_OPCODE_X0 = 1,
+  CNTLZ_UNARY_OPCODE_Y0 = 1,
+  CNTTZ_UNARY_OPCODE_X0 = 2,
+  CNTTZ_UNARY_OPCODE_Y0 = 2,
+  CRC32_32_RRR_0_OPCODE_X0 = 20,
+  CRC32_8_RRR_0_OPCODE_X0 = 21,
+  DBLALIGN2_RRR_0_OPCODE_X0 = 22,
+  DBLALIGN2_RRR_0_OPCODE_X1 = 13,
+  DBLALIGN4_RRR_0_OPCODE_X0 = 23,
+  DBLALIGN4_RRR_0_OPCODE_X1 = 14,
+  DBLALIGN6_RRR_0_OPCODE_X0 = 24,
+  DBLALIGN6_RRR_0_OPCODE_X1 = 15,
+  DBLALIGN_RRR_0_OPCODE_X0 = 25,
+  DRAIN_UNARY_OPCODE_X1 = 1,
+  DTLBPR_UNARY_OPCODE_X1 = 2,
+  EXCH4_RRR_0_OPCODE_X1 = 16,
+  EXCH_RRR_0_OPCODE_X1 = 17,
+  FDOUBLE_ADDSUB_RRR_0_OPCODE_X0 = 26,
+  FDOUBLE_ADD_FLAGS_RRR_0_OPCODE_X0 = 27,
+  FDOUBLE_MUL_FLAGS_RRR_0_OPCODE_X0 = 28,
+  FDOUBLE_PACK1_RRR_0_OPCODE_X0 = 29,
+  FDOUBLE_PACK2_RRR_0_OPCODE_X0 = 30,
+  FDOUBLE_SUB_FLAGS_RRR_0_OPCODE_X0 = 31,
+  FDOUBLE_UNPACK_MAX_RRR_0_OPCODE_X0 = 32,
+  FDOUBLE_UNPACK_MIN_RRR_0_OPCODE_X0 = 33,
+  FETCHADD4_RRR_0_OPCODE_X1 = 18,
+  FETCHADDGEZ4_RRR_0_OPCODE_X1 = 19,
+  FETCHADDGEZ_RRR_0_OPCODE_X1 = 20,
+  FETCHADD_RRR_0_OPCODE_X1 = 21,
+  FETCHAND4_RRR_0_OPCODE_X1 = 22,
+  FETCHAND_RRR_0_OPCODE_X1 = 23,
+  FETCHOR4_RRR_0_OPCODE_X1 = 24,
+  FETCHOR_RRR_0_OPCODE_X1 = 25,
+  FINV_UNARY_OPCODE_X1 = 3,
+  FLUSHWB_UNARY_OPCODE_X1 = 4,
+  FLUSH_UNARY_OPCODE_X1 = 5,
+  FNOP_UNARY_OPCODE_X0 = 3,
+  FNOP_UNARY_OPCODE_X1 = 6,
+  FNOP_UNARY_OPCODE_Y0 = 3,
+  FNOP_UNARY_OPCODE_Y1 = 8,
+  FSINGLE_ADD1_RRR_0_OPCODE_X0 = 34,
+  FSINGLE_ADDSUB2_RRR_0_OPCODE_X0 = 35,
+  FSINGLE_MUL1_RRR_0_OPCODE_X0 = 36,
+  FSINGLE_MUL2_RRR_0_OPCODE_X0 = 37,
+  FSINGLE_PACK1_UNARY_OPCODE_X0 = 4,
+  FSINGLE_PACK1_UNARY_OPCODE_Y0 = 4,
+  FSINGLE_PACK2_RRR_0_OPCODE_X0 = 38,
+  FSINGLE_SUB1_RRR_0_OPCODE_X0 = 39,
+  ICOH_UNARY_OPCODE_X1 = 7,
+  ILL_UNARY_OPCODE_X1 = 8,
+  ILL_UNARY_OPCODE_Y1 = 9,
+  IMM8_OPCODE_X0 = 4,
+  IMM8_OPCODE_X1 = 3,
+  INV_UNARY_OPCODE_X1 = 9,
+  IRET_UNARY_OPCODE_X1 = 10,
+  JALRP_UNARY_OPCODE_X1 = 11,
+  JALRP_UNARY_OPCODE_Y1 = 10,
+  JALR_UNARY_OPCODE_X1 = 12,
+  JALR_UNARY_OPCODE_Y1 = 11,
+  JAL_JUMP_OPCODE_X1 = 0,
+  JRP_UNARY_OPCODE_X1 = 13,
+  JRP_UNARY_OPCODE_Y1 = 12,
+  JR_UNARY_OPCODE_X1 = 14,
+  JR_UNARY_OPCODE_Y1 = 13,
+  JUMP_OPCODE_X1 = 4,
+  J_JUMP_OPCODE_X1 = 1,
+  LD1S_ADD_IMM8_OPCODE_X1 = 7,
+  LD1S_OPCODE_Y2 = 0,
+  LD1S_UNARY_OPCODE_X1 = 15,
+  LD1U_ADD_IMM8_OPCODE_X1 = 8,
+  LD1U_OPCODE_Y2 = 1,
+  LD1U_UNARY_OPCODE_X1 = 16,
+  LD2S_ADD_IMM8_OPCODE_X1 = 9,
+  LD2S_OPCODE_Y2 = 2,
+  LD2S_UNARY_OPCODE_X1 = 17,
+  LD2U_ADD_IMM8_OPCODE_X1 = 10,
+  LD2U_OPCODE_Y2 = 3,
+  LD2U_UNARY_OPCODE_X1 = 18,
+  LD4S_ADD_IMM8_OPCODE_X1 = 11,
+  LD4S_OPCODE_Y2 = 1,
+  LD4S_UNARY_OPCODE_X1 = 19,
+  LD4U_ADD_IMM8_OPCODE_X1 = 12,
+  LD4U_OPCODE_Y2 = 2,
+  LD4U_UNARY_OPCODE_X1 = 20,
+  LDNA_UNARY_OPCODE_X1 = 21,
+  LDNT1S_ADD_IMM8_OPCODE_X1 = 13,
+  LDNT1S_UNARY_OPCODE_X1 = 22,
+  LDNT1U_ADD_IMM8_OPCODE_X1 = 14,
+  LDNT1U_UNARY_OPCODE_X1 = 23,
+  LDNT2S_ADD_IMM8_OPCODE_X1 = 15,
+  LDNT2S_UNARY_OPCODE_X1 = 24,
+  LDNT2U_ADD_IMM8_OPCODE_X1 = 16,
+  LDNT2U_UNARY_OPCODE_X1 = 25,
+  LDNT4S_ADD_IMM8_OPCODE_X1 = 17,
+  LDNT4S_UNARY_OPCODE_X1 = 26,
+  LDNT4U_ADD_IMM8_OPCODE_X1 = 18,
+  LDNT4U_UNARY_OPCODE_X1 = 27,
+  LDNT_ADD_IMM8_OPCODE_X1 = 19,
+  LDNT_UNARY_OPCODE_X1 = 28,
+  LD_ADD_IMM8_OPCODE_X1 = 20,
+  LD_OPCODE_Y2 = 3,
+  LD_UNARY_OPCODE_X1 = 29,
+  LNK_UNARY_OPCODE_X1 = 30,
+  LNK_UNARY_OPCODE_Y1 = 14,
+  LWNA_ADD_IMM8_OPCODE_X1 = 21,
+  MFSPR_IMM8_OPCODE_X1 = 22,
+  MF_UNARY_OPCODE_X1 = 31,
+  MM_BF_OPCODE_X0 = 7,
+  MNZ_RRR_0_OPCODE_X0 = 40,
+  MNZ_RRR_0_OPCODE_X1 = 26,
+  MNZ_RRR_4_OPCODE_Y0 = 2,
+  MNZ_RRR_4_OPCODE_Y1 = 2,
+  MODE_OPCODE_YA2 = 1,
+  MODE_OPCODE_YB2 = 2,
+  MODE_OPCODE_YC2 = 3,
+  MTSPR_IMM8_OPCODE_X1 = 23,
+  MULAX_RRR_0_OPCODE_X0 = 41,
+  MULAX_RRR_3_OPCODE_Y0 = 2,
+  MULA_HS_HS_RRR_0_OPCODE_X0 = 42,
+  MULA_HS_HS_RRR_9_OPCODE_Y0 = 0,
+  MULA_HS_HU_RRR_0_OPCODE_X0 = 43,
+  MULA_HS_LS_RRR_0_OPCODE_X0 = 44,
+  MULA_HS_LU_RRR_0_OPCODE_X0 = 45,
+  MULA_HU_HU_RRR_0_OPCODE_X0 = 46,
+  MULA_HU_HU_RRR_9_OPCODE_Y0 = 1,
+  MULA_HU_LS_RRR_0_OPCODE_X0 = 47,
+  MULA_HU_LU_RRR_0_OPCODE_X0 = 48,
+  MULA_LS_LS_RRR_0_OPCODE_X0 = 49,
+  MULA_LS_LS_RRR_9_OPCODE_Y0 = 2,
+  MULA_LS_LU_RRR_0_OPCODE_X0 = 50,
+  MULA_LU_LU_RRR_0_OPCODE_X0 = 51,
+  MULA_LU_LU_RRR_9_OPCODE_Y0 = 3,
+  MULX_RRR_0_OPCODE_X0 = 52,
+  MULX_RRR_3_OPCODE_Y0 = 3,
+  MUL_HS_HS_RRR_0_OPCODE_X0 = 53,
+  MUL_HS_HS_RRR_8_OPCODE_Y0 = 0,
+  MUL_HS_HU_RRR_0_OPCODE_X0 = 54,
+  MUL_HS_LS_RRR_0_OPCODE_X0 = 55,
+  MUL_HS_LU_RRR_0_OPCODE_X0 = 56,
+  MUL_HU_HU_RRR_0_OPCODE_X0 = 57,
+  MUL_HU_HU_RRR_8_OPCODE_Y0 = 1,
+  MUL_HU_LS_RRR_0_OPCODE_X0 = 58,
+  MUL_HU_LU_RRR_0_OPCODE_X0 = 59,
+  MUL_LS_LS_RRR_0_OPCODE_X0 = 60,
+  MUL_LS_LS_RRR_8_OPCODE_Y0 = 2,
+  MUL_LS_LU_RRR_0_OPCODE_X0 = 61,
+  MUL_LU_LU_RRR_0_OPCODE_X0 = 62,
+  MUL_LU_LU_RRR_8_OPCODE_Y0 = 3,
+  MZ_RRR_0_OPCODE_X0 = 63,
+  MZ_RRR_0_OPCODE_X1 = 27,
+  MZ_RRR_4_OPCODE_Y0 = 3,
+  MZ_RRR_4_OPCODE_Y1 = 3,
+  NAP_UNARY_OPCODE_X1 = 32,
+  NOP_UNARY_OPCODE_X0 = 5,
+  NOP_UNARY_OPCODE_X1 = 33,
+  NOP_UNARY_OPCODE_Y0 = 5,
+  NOP_UNARY_OPCODE_Y1 = 15,
+  NOR_RRR_0_OPCODE_X0 = 64,
+  NOR_RRR_0_OPCODE_X1 = 28,
+  NOR_RRR_5_OPCODE_Y0 = 1,
+  NOR_RRR_5_OPCODE_Y1 = 1,
+  ORI_IMM8_OPCODE_X0 = 7,
+  ORI_IMM8_OPCODE_X1 = 24,
+  OR_RRR_0_OPCODE_X0 = 65,
+  OR_RRR_0_OPCODE_X1 = 29,
+  OR_RRR_5_OPCODE_Y0 = 2,
+  OR_RRR_5_OPCODE_Y1 = 2,
+  PCNT_UNARY_OPCODE_X0 = 6,
+  PCNT_UNARY_OPCODE_Y0 = 6,
+  REVBITS_UNARY_OPCODE_X0 = 7,
+  REVBITS_UNARY_OPCODE_Y0 = 7,
+  REVBYTES_UNARY_OPCODE_X0 = 8,
+  REVBYTES_UNARY_OPCODE_Y0 = 8,
+  ROTLI_SHIFT_OPCODE_X0 = 1,
+  ROTLI_SHIFT_OPCODE_X1 = 1,
+  ROTLI_SHIFT_OPCODE_Y0 = 0,
+  ROTLI_SHIFT_OPCODE_Y1 = 0,
+  ROTL_RRR_0_OPCODE_X0 = 66,
+  ROTL_RRR_0_OPCODE_X1 = 30,
+  ROTL_RRR_6_OPCODE_Y0 = 0,
+  ROTL_RRR_6_OPCODE_Y1 = 0,
+  RRR_0_OPCODE_X0 = 5,
+  RRR_0_OPCODE_X1 = 5,
+  RRR_0_OPCODE_Y0 = 5,
+  RRR_0_OPCODE_Y1 = 6,
+  RRR_1_OPCODE_Y0 = 6,
+  RRR_1_OPCODE_Y1 = 7,
+  RRR_2_OPCODE_Y0 = 7,
+  RRR_2_OPCODE_Y1 = 8,
+  RRR_3_OPCODE_Y0 = 8,
+  RRR_3_OPCODE_Y1 = 9,
+  RRR_4_OPCODE_Y0 = 9,
+  RRR_4_OPCODE_Y1 = 10,
+  RRR_5_OPCODE_Y0 = 10,
+  RRR_5_OPCODE_Y1 = 11,
+  RRR_6_OPCODE_Y0 = 11,
+  RRR_6_OPCODE_Y1 = 12,
+  RRR_7_OPCODE_Y0 = 12,
+  RRR_7_OPCODE_Y1 = 13,
+  RRR_8_OPCODE_Y0 = 13,
+  RRR_9_OPCODE_Y0 = 14,
+  SHIFT_OPCODE_X0 = 6,
+  SHIFT_OPCODE_X1 = 6,
+  SHIFT_OPCODE_Y0 = 15,
+  SHIFT_OPCODE_Y1 = 14,
+  SHL16INSLI_OPCODE_X0 = 7,
+  SHL16INSLI_OPCODE_X1 = 7,
+  SHL1ADDX_RRR_0_OPCODE_X0 = 67,
+  SHL1ADDX_RRR_0_OPCODE_X1 = 31,
+  SHL1ADDX_RRR_7_OPCODE_Y0 = 1,
+  SHL1ADDX_RRR_7_OPCODE_Y1 = 1,
+  SHL1ADD_RRR_0_OPCODE_X0 = 68,
+  SHL1ADD_RRR_0_OPCODE_X1 = 32,
+  SHL1ADD_RRR_1_OPCODE_Y0 = 0,
+  SHL1ADD_RRR_1_OPCODE_Y1 = 0,
+  SHL2ADDX_RRR_0_OPCODE_X0 = 69,
+  SHL2ADDX_RRR_0_OPCODE_X1 = 33,
+  SHL2ADDX_RRR_7_OPCODE_Y0 = 2,
+  SHL2ADDX_RRR_7_OPCODE_Y1 = 2,
+  SHL2ADD_RRR_0_OPCODE_X0 = 70,
+  SHL2ADD_RRR_0_OPCODE_X1 = 34,
+  SHL2ADD_RRR_1_OPCODE_Y0 = 1,
+  SHL2ADD_RRR_1_OPCODE_Y1 = 1,
+  SHL3ADDX_RRR_0_OPCODE_X0 = 71,
+  SHL3ADDX_RRR_0_OPCODE_X1 = 35,
+  SHL3ADDX_RRR_7_OPCODE_Y0 = 3,
+  SHL3ADDX_RRR_7_OPCODE_Y1 = 3,
+  SHL3ADD_RRR_0_OPCODE_X0 = 72,
+  SHL3ADD_RRR_0_OPCODE_X1 = 36,
+  SHL3ADD_RRR_1_OPCODE_Y0 = 2,
+  SHL3ADD_RRR_1_OPCODE_Y1 = 2,
+  SHLI_SHIFT_OPCODE_X0 = 2,
+  SHLI_SHIFT_OPCODE_X1 = 2,
+  SHLI_SHIFT_OPCODE_Y0 = 1,
+  SHLI_SHIFT_OPCODE_Y1 = 1,
+  SHLXI_SHIFT_OPCODE_X0 = 3,
+  SHLXI_SHIFT_OPCODE_X1 = 3,
+  SHLX_RRR_0_OPCODE_X0 = 73,
+  SHLX_RRR_0_OPCODE_X1 = 37,
+  SHL_RRR_0_OPCODE_X0 = 74,
+  SHL_RRR_0_OPCODE_X1 = 38,
+  SHL_RRR_6_OPCODE_Y0 = 1,
+  SHL_RRR_6_OPCODE_Y1 = 1,
+  SHRSI_SHIFT_OPCODE_X0 = 4,
+  SHRSI_SHIFT_OPCODE_X1 = 4,
+  SHRSI_SHIFT_OPCODE_Y0 = 2,
+  SHRSI_SHIFT_OPCODE_Y1 = 2,
+  SHRS_RRR_0_OPCODE_X0 = 75,
+  SHRS_RRR_0_OPCODE_X1 = 39,
+  SHRS_RRR_6_OPCODE_Y0 = 2,
+  SHRS_RRR_6_OPCODE_Y1 = 2,
+  SHRUI_SHIFT_OPCODE_X0 = 5,
+  SHRUI_SHIFT_OPCODE_X1 = 5,
+  SHRUI_SHIFT_OPCODE_Y0 = 3,
+  SHRUI_SHIFT_OPCODE_Y1 = 3,
+  SHRUXI_SHIFT_OPCODE_X0 = 6,
+  SHRUXI_SHIFT_OPCODE_X1 = 6,
+  SHRUX_RRR_0_OPCODE_X0 = 76,
+  SHRUX_RRR_0_OPCODE_X1 = 40,
+  SHRU_RRR_0_OPCODE_X0 = 77,
+  SHRU_RRR_0_OPCODE_X1 = 41,
+  SHRU_RRR_6_OPCODE_Y0 = 3,
+  SHRU_RRR_6_OPCODE_Y1 = 3,
+  SHUFFLEBYTES_RRR_0_OPCODE_X0 = 78,
+  ST1_ADD_IMM8_OPCODE_X1 = 25,
+  ST1_OPCODE_Y2 = 0,
+  ST1_RRR_0_OPCODE_X1 = 42,
+  ST2_ADD_IMM8_OPCODE_X1 = 26,
+  ST2_OPCODE_Y2 = 1,
+  ST2_RRR_0_OPCODE_X1 = 43,
+  ST4_ADD_IMM8_OPCODE_X1 = 27,
+  ST4_OPCODE_Y2 = 2,
+  ST4_RRR_0_OPCODE_X1 = 44,
+  STNT1_ADD_IMM8_OPCODE_X1 = 28,
+  STNT1_RRR_0_OPCODE_X1 = 45,
+  STNT2_ADD_IMM8_OPCODE_X1 = 29,
+  STNT2_RRR_0_OPCODE_X1 = 46,
+  STNT4_ADD_IMM8_OPCODE_X1 = 30,
+  STNT4_RRR_0_OPCODE_X1 = 47,
+  STNT_ADD_IMM8_OPCODE_X1 = 31,
+  STNT_RRR_0_OPCODE_X1 = 48,
+  ST_ADD_IMM8_OPCODE_X1 = 32,
+  ST_OPCODE_Y2 = 3,
+  ST_RRR_0_OPCODE_X1 = 49,
+  SUBXSC_RRR_0_OPCODE_X0 = 79,
+  SUBXSC_RRR_0_OPCODE_X1 = 50,
+  SUBX_RRR_0_OPCODE_X0 = 80,
+  SUBX_RRR_0_OPCODE_X1 = 51,
+  SUBX_RRR_0_OPCODE_Y0 = 2,
+  SUBX_RRR_0_OPCODE_Y1 = 2,
+  SUB_RRR_0_OPCODE_X0 = 81,
+  SUB_RRR_0_OPCODE_X1 = 52,
+  SUB_RRR_0_OPCODE_Y0 = 3,
+  SUB_RRR_0_OPCODE_Y1 = 3,
+  SWINT0_UNARY_OPCODE_X1 = 34,
+  SWINT1_UNARY_OPCODE_X1 = 35,
+  SWINT2_UNARY_OPCODE_X1 = 36,
+  SWINT3_UNARY_OPCODE_X1 = 37,
+  TBLIDXB0_UNARY_OPCODE_X0 = 9,
+  TBLIDXB0_UNARY_OPCODE_Y0 = 9,
+  TBLIDXB1_UNARY_OPCODE_X0 = 10,
+  TBLIDXB1_UNARY_OPCODE_Y0 = 10,
+  TBLIDXB2_UNARY_OPCODE_X0 = 11,
+  TBLIDXB2_UNARY_OPCODE_Y0 = 11,
+  TBLIDXB3_UNARY_OPCODE_X0 = 12,
+  TBLIDXB3_UNARY_OPCODE_Y0 = 12,
+  UNARY_RRR_0_OPCODE_X0 = 82,
+  UNARY_RRR_0_OPCODE_X1 = 53,
+  UNARY_RRR_1_OPCODE_Y0 = 3,
+  UNARY_RRR_1_OPCODE_Y1 = 3,
+  V1ADDI_IMM8_OPCODE_X0 = 8,
+  V1ADDI_IMM8_OPCODE_X1 = 33,
+  V1ADDUC_RRR_0_OPCODE_X0 = 83,
+  V1ADDUC_RRR_0_OPCODE_X1 = 54,
+  V1ADD_RRR_0_OPCODE_X0 = 84,
+  V1ADD_RRR_0_OPCODE_X1 = 55,
+  V1ADIFFU_RRR_0_OPCODE_X0 = 85,
+  V1AVGU_RRR_0_OPCODE_X0 = 86,
+  V1CMPEQI_IMM8_OPCODE_X0 = 9,
+  V1CMPEQI_IMM8_OPCODE_X1 = 34,
+  V1CMPEQ_RRR_0_OPCODE_X0 = 87,
+  V1CMPEQ_RRR_0_OPCODE_X1 = 56,
+  V1CMPLES_RRR_0_OPCODE_X0 = 88,
+  V1CMPLES_RRR_0_OPCODE_X1 = 57,
+  V1CMPLEU_RRR_0_OPCODE_X0 = 89,
+  V1CMPLEU_RRR_0_OPCODE_X1 = 58,
+  V1CMPLTSI_IMM8_OPCODE_X0 = 10,
+  V1CMPLTSI_IMM8_OPCODE_X1 = 35,
+  V1CMPLTS_RRR_0_OPCODE_X0 = 90,
+  V1CMPLTS_RRR_0_OPCODE_X1 = 59,
+  V1CMPLTUI_IMM8_OPCODE_X0 = 11,
+  V1CMPLTUI_IMM8_OPCODE_X1 = 36,
+  V1CMPLTU_RRR_0_OPCODE_X0 = 91,
+  V1CMPLTU_RRR_0_OPCODE_X1 = 60,
+  V1CMPNE_RRR_0_OPCODE_X0 = 92,
+  V1CMPNE_RRR_0_OPCODE_X1 = 61,
+  V1DDOTPUA_RRR_0_OPCODE_X0 = 161,
+  V1DDOTPUSA_RRR_0_OPCODE_X0 = 93,
+  V1DDOTPUS_RRR_0_OPCODE_X0 = 94,
+  V1DDOTPU_RRR_0_OPCODE_X0 = 162,
+  V1DOTPA_RRR_0_OPCODE_X0 = 95,
+  V1DOTPUA_RRR_0_OPCODE_X0 = 163,
+  V1DOTPUSA_RRR_0_OPCODE_X0 = 96,
+  V1DOTPUS_RRR_0_OPCODE_X0 = 97,
+  V1DOTPU_RRR_0_OPCODE_X0 = 164,
+  V1DOTP_RRR_0_OPCODE_X0 = 98,
+  V1INT_H_RRR_0_OPCODE_X0 = 99,
+  V1INT_H_RRR_0_OPCODE_X1 = 62,
+  V1INT_L_RRR_0_OPCODE_X0 = 100,
+  V1INT_L_RRR_0_OPCODE_X1 = 63,
+  V1MAXUI_IMM8_OPCODE_X0 = 12,
+  V1MAXUI_IMM8_OPCODE_X1 = 37,
+  V1MAXU_RRR_0_OPCODE_X0 = 101,
+  V1MAXU_RRR_0_OPCODE_X1 = 64,
+  V1MINUI_IMM8_OPCODE_X0 = 13,
+  V1MINUI_IMM8_OPCODE_X1 = 38,
+  V1MINU_RRR_0_OPCODE_X0 = 102,
+  V1MINU_RRR_0_OPCODE_X1 = 65,
+  V1MNZ_RRR_0_OPCODE_X0 = 103,
+  V1MNZ_RRR_0_OPCODE_X1 = 66,
+  V1MULTU_RRR_0_OPCODE_X0 = 104,
+  V1MULUS_RRR_0_OPCODE_X0 = 105,
+  V1MULU_RRR_0_OPCODE_X0 = 106,
+  V1MZ_RRR_0_OPCODE_X0 = 107,
+  V1MZ_RRR_0_OPCODE_X1 = 67,
+  V1SADAU_RRR_0_OPCODE_X0 = 108,
+  V1SADU_RRR_0_OPCODE_X0 = 109,
+  V1SHLI_SHIFT_OPCODE_X0 = 7,
+  V1SHLI_SHIFT_OPCODE_X1 = 7,
+  V1SHL_RRR_0_OPCODE_X0 = 110,
+  V1SHL_RRR_0_OPCODE_X1 = 68,
+  V1SHRSI_SHIFT_OPCODE_X0 = 8,
+  V1SHRSI_SHIFT_OPCODE_X1 = 8,
+  V1SHRS_RRR_0_OPCODE_X0 = 111,
+  V1SHRS_RRR_0_OPCODE_X1 = 69,
+  V1SHRUI_SHIFT_OPCODE_X0 = 9,
+  V1SHRUI_SHIFT_OPCODE_X1 = 9,
+  V1SHRU_RRR_0_OPCODE_X0 = 112,
+  V1SHRU_RRR_0_OPCODE_X1 = 70,
+  V1SUBUC_RRR_0_OPCODE_X0 = 113,
+  V1SUBUC_RRR_0_OPCODE_X1 = 71,
+  V1SUB_RRR_0_OPCODE_X0 = 114,
+  V1SUB_RRR_0_OPCODE_X1 = 72,
+  V2ADDI_IMM8_OPCODE_X0 = 14,
+  V2ADDI_IMM8_OPCODE_X1 = 39,
+  V2ADDSC_RRR_0_OPCODE_X0 = 115,
+  V2ADDSC_RRR_0_OPCODE_X1 = 73,
+  V2ADD_RRR_0_OPCODE_X0 = 116,
+  V2ADD_RRR_0_OPCODE_X1 = 74,
+  V2ADIFFS_RRR_0_OPCODE_X0 = 117,
+  V2AVGS_RRR_0_OPCODE_X0 = 118,
+  V2CMPEQI_IMM8_OPCODE_X0 = 15,
+  V2CMPEQI_IMM8_OPCODE_X1 = 40,
+  V2CMPEQ_RRR_0_OPCODE_X0 = 119,
+  V2CMPEQ_RRR_0_OPCODE_X1 = 75,
+  V2CMPLES_RRR_0_OPCODE_X0 = 120,
+  V2CMPLES_RRR_0_OPCODE_X1 = 76,
+  V2CMPLEU_RRR_0_OPCODE_X0 = 121,
+  V2CMPLEU_RRR_0_OPCODE_X1 = 77,
+  V2CMPLTSI_IMM8_OPCODE_X0 = 16,
+  V2CMPLTSI_IMM8_OPCODE_X1 = 41,
+  V2CMPLTS_RRR_0_OPCODE_X0 = 122,
+  V2CMPLTS_RRR_0_OPCODE_X1 = 78,
+  V2CMPLTUI_IMM8_OPCODE_X0 = 17,
+  V2CMPLTUI_IMM8_OPCODE_X1 = 42,
+  V2CMPLTU_RRR_0_OPCODE_X0 = 123,
+  V2CMPLTU_RRR_0_OPCODE_X1 = 79,
+  V2CMPNE_RRR_0_OPCODE_X0 = 124,
+  V2CMPNE_RRR_0_OPCODE_X1 = 80,
+  V2DOTPA_RRR_0_OPCODE_X0 = 125,
+  V2DOTP_RRR_0_OPCODE_X0 = 126,
+  V2INT_H_RRR_0_OPCODE_X0 = 127,
+  V2INT_H_RRR_0_OPCODE_X1 = 81,
+  V2INT_L_RRR_0_OPCODE_X0 = 128,
+  V2INT_L_RRR_0_OPCODE_X1 = 82,
+  V2MAXSI_IMM8_OPCODE_X0 = 18,
+  V2MAXSI_IMM8_OPCODE_X1 = 43,
+  V2MAXS_RRR_0_OPCODE_X0 = 129,
+  V2MAXS_RRR_0_OPCODE_X1 = 83,
+  V2MINSI_IMM8_OPCODE_X0 = 19,
+  V2MINSI_IMM8_OPCODE_X1 = 44,
+  V2MINS_RRR_0_OPCODE_X0 = 130,
+  V2MINS_RRR_0_OPCODE_X1 = 84,
+  V2MNZ_RRR_0_OPCODE_X0 = 131,
+  V2MNZ_RRR_0_OPCODE_X1 = 85,
+  V2MULFSC_RRR_0_OPCODE_X0 = 132,
+  V2MULS_RRR_0_OPCODE_X0 = 133,
+  V2MULTS_RRR_0_OPCODE_X0 = 134,
+  V2MZ_RRR_0_OPCODE_X0 = 135,
+  V2MZ_RRR_0_OPCODE_X1 = 86,
+  V2PACKH_RRR_0_OPCODE_X0 = 136,
+  V2PACKH_RRR_0_OPCODE_X1 = 87,
+  V2PACKL_RRR_0_OPCODE_X0 = 137,
+  V2PACKL_RRR_0_OPCODE_X1 = 88,
+  V2PACKUC_RRR_0_OPCODE_X0 = 138,
+  V2PACKUC_RRR_0_OPCODE_X1 = 89,
+  V2SADAS_RRR_0_OPCODE_X0 = 139,
+  V2SADAU_RRR_0_OPCODE_X0 = 140,
+  V2SADS_RRR_0_OPCODE_X0 = 141,
+  V2SADU_RRR_0_OPCODE_X0 = 142,
+  V2SHLI_SHIFT_OPCODE_X0 = 10,
+  V2SHLI_SHIFT_OPCODE_X1 = 10,
+  V2SHLSC_RRR_0_OPCODE_X0 = 143,
+  V2SHLSC_RRR_0_OPCODE_X1 = 90,
+  V2SHL_RRR_0_OPCODE_X0 = 144,
+  V2SHL_RRR_0_OPCODE_X1 = 91,
+  V2SHRSI_SHIFT_OPCODE_X0 = 11,
+  V2SHRSI_SHIFT_OPCODE_X1 = 11,
+  V2SHRS_RRR_0_OPCODE_X0 = 145,
+  V2SHRS_RRR_0_OPCODE_X1 = 92,
+  V2SHRUI_SHIFT_OPCODE_X0 = 12,
+  V2SHRUI_SHIFT_OPCODE_X1 = 12,
+  V2SHRU_RRR_0_OPCODE_X0 = 146,
+  V2SHRU_RRR_0_OPCODE_X1 = 93,
+  V2SUBSC_RRR_0_OPCODE_X0 = 147,
+  V2SUBSC_RRR_0_OPCODE_X1 = 94,
+  V2SUB_RRR_0_OPCODE_X0 = 148,
+  V2SUB_RRR_0_OPCODE_X1 = 95,
+  V4ADDSC_RRR_0_OPCODE_X0 = 149,
+  V4ADDSC_RRR_0_OPCODE_X1 = 96,
+  V4ADD_RRR_0_OPCODE_X0 = 150,
+  V4ADD_RRR_0_OPCODE_X1 = 97,
+  V4INT_H_RRR_0_OPCODE_X0 = 151,
+  V4INT_H_RRR_0_OPCODE_X1 = 98,
+  V4INT_L_RRR_0_OPCODE_X0 = 152,
+  V4INT_L_RRR_0_OPCODE_X1 = 99,
+  V4PACKSC_RRR_0_OPCODE_X0 = 153,
+  V4PACKSC_RRR_0_OPCODE_X1 = 100,
+  V4SHLSC_RRR_0_OPCODE_X0 = 154,
+  V4SHLSC_RRR_0_OPCODE_X1 = 101,
+  V4SHL_RRR_0_OPCODE_X0 = 155,
+  V4SHL_RRR_0_OPCODE_X1 = 102,
+  V4SHRS_RRR_0_OPCODE_X0 = 156,
+  V4SHRS_RRR_0_OPCODE_X1 = 103,
+  V4SHRU_RRR_0_OPCODE_X0 = 157,
+  V4SHRU_RRR_0_OPCODE_X1 = 104,
+  V4SUBSC_RRR_0_OPCODE_X0 = 158,
+  V4SUBSC_RRR_0_OPCODE_X1 = 105,
+  V4SUB_RRR_0_OPCODE_X0 = 159,
+  V4SUB_RRR_0_OPCODE_X1 = 106,
+  WH64_UNARY_OPCODE_X1 = 38,
+  XORI_IMM8_OPCODE_X0 = 20,
+  XORI_IMM8_OPCODE_X1 = 45,
+  XOR_RRR_0_OPCODE_X0 = 160,
+  XOR_RRR_0_OPCODE_X1 = 107,
+  XOR_RRR_5_OPCODE_Y0 = 3,
+  XOR_RRR_5_OPCODE_Y1 = 3
 };
 
 #endif /* !_TILE_OPCODE_CONSTANTS_H */
diff --git a/arch/tile/include/asm/page.h b/arch/tile/include/asm/page.h
index 7d90641cf18d..db93518fac03 100644
--- a/arch/tile/include/asm/page.h
+++ b/arch/tile/include/asm/page.h
@@ -16,10 +16,12 @@
 #define _ASM_TILE_PAGE_H
 
 #include <linux/const.h>
+#include <hv/hypervisor.h>
+#include <arch/chip.h>
 
 /* PAGE_SHIFT and HPAGE_SHIFT determine the page sizes. */
-#define PAGE_SHIFT      16
-#define HPAGE_SHIFT     24
+#define PAGE_SHIFT      HV_LOG2_PAGE_SIZE_SMALL
+#define HPAGE_SHIFT     HV_LOG2_PAGE_SIZE_LARGE
 
 #define PAGE_SIZE       (_AC(1, UL) << PAGE_SHIFT)
 #define HPAGE_SIZE      (_AC(1, UL) << HPAGE_SHIFT)
@@ -27,25 +29,13 @@
 #define PAGE_MASK       (~(PAGE_SIZE - 1))
 #define HPAGE_MASK      (~(HPAGE_SIZE - 1))
 
-#ifdef __KERNEL__
-
-#include <hv/hypervisor.h>
-#include <arch/chip.h>
-
 /*
- * The {,H}PAGE_SHIFT values must match the HV_LOG2_PAGE_SIZE_xxx
- * definitions in <hv/hypervisor.h>.  We validate this at build time
- * here, and again at runtime during early boot.  We provide a
- * separate definition since userspace doesn't have <hv/hypervisor.h>.
- *
- * Be careful to distinguish PAGE_SHIFT from HV_PTE_INDEX_PFN, since
- * they are the same on i386 but not TILE.
+ * If the Kconfig doesn't specify, set a maximum zone order that
+ * is enough so that we can create huge pages from small pages given
+ * the respective sizes of the two page types.  See <linux/mmzone.h>.
  */
-#if HV_LOG2_PAGE_SIZE_SMALL != PAGE_SHIFT
-# error Small page size mismatch in Linux
-#endif
-#if HV_LOG2_PAGE_SIZE_LARGE != HPAGE_SHIFT
-# error Huge page size mismatch in Linux
+#ifndef CONFIG_FORCE_MAX_ZONEORDER
+#define CONFIG_FORCE_MAX_ZONEORDER (HPAGE_SHIFT - PAGE_SHIFT + 1)
 #endif
 
 #ifndef __ASSEMBLY__
@@ -81,12 +71,6 @@ static inline void copy_user_page(void *to, void *from, unsigned long vaddr,
  * Hypervisor page tables are made of the same basic structure.
  */
 
-typedef __u64 pteval_t;
-typedef __u64 pmdval_t;
-typedef __u64 pudval_t;
-typedef __u64 pgdval_t;
-typedef __u64 pgprotval_t;
-
 typedef HV_PTE pte_t;
 typedef HV_PTE pgd_t;
 typedef HV_PTE pgprot_t;
@@ -103,6 +87,10 @@ typedef struct page *pgtable_t;
 /* Must be a macro since it is used to create constants. */
 #define __pgprot(val) hv_pte(val)
 
+/* Rarely-used initializers, typically with a "zero" value. */
+#define __pte(x) hv_pte(x)
+#define __pgd(x) hv_pte(x)
+
 static inline u64 pgprot_val(pgprot_t pgprot)
 {
         return hv_pte_val(pgprot);
@@ -122,6 +110,8 @@ static inline u64 pgd_val(pgd_t pgd)
 
 typedef HV_PTE pmd_t;
 
+#define __pmd(x) hv_pte(x)
+
 static inline u64 pmd_val(pmd_t pmd)
 {
         return hv_pte_val(pmd);
@@ -199,17 +189,17 @@ static inline __attribute_const__ int get_order(unsigned long size)
  * If you want more physical memory than this then see the CONFIG_HIGHMEM
  * option in the kernel configuration.
  *
- * The top two 16MB chunks in the table below (VIRT and HV) are
- * unavailable to Linux.  Since the kernel interrupt vectors must live
- * at 0xfd000000, we map all of the bottom of RAM at this address with
- * a huge page table entry to minimize its ITLB footprint (as well as
- * at PAGE_OFFSET).  The last architected requirement is that user
- * interrupt vectors live at 0xfc000000, so we make that range of
- * memory available to user processes.  The remaining regions are sized
- * as shown; after the first four addresses, we show "typical" values,
- * since the actual addresses depend on kernel #defines.
+ * The top 16MB chunk in the table below is unavailable to Linux.  Since
+ * the kernel interrupt vectors must live at ether 0xfe000000 or 0xfd000000
+ * (depending on whether the kernel is at PL2 or Pl1), we map all of the
+ * bottom of RAM at this address with a huge page table entry to minimize
+ * its ITLB footprint (as well as at PAGE_OFFSET).  The last architected
+ * requirement is that user interrupt vectors live at 0xfc000000, so we
+ * make that range of memory available to user processes.  The remaining
+ * regions are sized as shown; the first four addresses use the PL 1
+ * values, and after that, we show "typical" values, since the actual
+ * addresses depend on kernel #defines.
  *
- * MEM_VIRT_INTRPT                 0xff000000
  * MEM_HV_INTRPT                   0xfe000000
  * MEM_SV_INTRPT (kernel code)     0xfd000000
  * MEM_USER_INTRPT (user vector)   0xfc000000
@@ -221,9 +211,14 @@ static inline __attribute_const__ int get_order(unsigned long size)
  */
 
 #define MEM_USER_INTRPT         _AC(0xfc000000, UL)
+#if CONFIG_KERNEL_PL == 1
 #define MEM_SV_INTRPT           _AC(0xfd000000, UL)
 #define MEM_HV_INTRPT           _AC(0xfe000000, UL)
-#define MEM_VIRT_INTRPT         _AC(0xff000000, UL)
+#else
+#define MEM_GUEST_INTRPT        _AC(0xfd000000, UL)
+#define MEM_SV_INTRPT           _AC(0xfe000000, UL)
+#define MEM_HV_INTRPT           _AC(0xff000000, UL)
+#endif
 
 #define INTRPT_SIZE             0x4000
 
@@ -325,7 +320,7 @@ static inline int pfn_valid(unsigned long pfn)
 
 /* Provide as macros since these require some other headers included. */
 #define page_to_pa(page) ((phys_addr_t)(page_to_pfn(page)) << PAGE_SHIFT)
-#define virt_to_page(kaddr) pfn_to_page(kaddr_to_pfn(kaddr))
+#define virt_to_page(kaddr) pfn_to_page(kaddr_to_pfn((void *)(kaddr)))
 #define page_to_virt(page) pfn_to_kaddr(page_to_pfn(page))
 
 struct mm_struct;
@@ -338,6 +333,4 @@ extern pte_t *virt_to_pte(struct mm_struct *mm, unsigned long addr);
 
 #include <asm-generic/memory_model.h>
 
-#endif /* __KERNEL__ */
-
 #endif /* _ASM_TILE_PAGE_H */
diff --git a/arch/tile/include/asm/parport.h b/arch/tile/include/asm/parport.h
new file mode 100644
index 000000000000..cf252af64590
--- /dev/null
+++ b/arch/tile/include/asm/parport.h
@@ -0,0 +1 @@
+#include <asm-generic/parport.h>
diff --git a/arch/tile/include/asm/pci-bridge.h b/arch/tile/include/asm/pci-bridge.h
deleted file mode 100644
index e853b0e2793b..000000000000
--- a/arch/tile/include/asm/pci-bridge.h
+++ /dev/null
@@ -1,117 +0,0 @@
-/*
- * Copyright 2010 Tilera Corporation. All Rights Reserved.
- *
- *   This program is free software; you can redistribute it and/or
- *   modify it under the terms of the GNU General Public License
- *   as published by the Free Software Foundation, version 2.
- *
- *   This program is distributed in the hope that it will be useful, but
- *   WITHOUT ANY WARRANTY; without even the implied warranty of
- *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
- *   NON INFRINGEMENT.  See the GNU General Public License for
- *   more details.
- */
-
-#ifndef _ASM_TILE_PCI_BRIDGE_H
-#define _ASM_TILE_PCI_BRIDGE_H
-
-#include <linux/ioport.h>
-#include <linux/pci.h>
-
-struct device_node;
-struct pci_controller;
-
-/*
- * pci_io_base returns the memory address at which you can access
- * the I/O space for PCI bus number `bus' (or NULL on error).
- */
-extern void __iomem *pci_bus_io_base(unsigned int bus);
-extern unsigned long pci_bus_io_base_phys(unsigned int bus);
-extern unsigned long pci_bus_mem_base_phys(unsigned int bus);
-
-/* Allocate a new PCI host bridge structure */
-extern struct pci_controller *pcibios_alloc_controller(void);
-
-/* Helper function for setting up resources */
-extern void pci_init_resource(struct resource *res, unsigned long start,
-                              unsigned long end, int flags, char *name);
-
-/* Get the PCI host controller for a bus */
-extern struct pci_controller *pci_bus_to_hose(int bus);
-
-/*
- * Structure of a PCI controller (host bridge)
- */
-struct pci_controller {
-        int index;              /* PCI domain number */
-        struct pci_bus *root_bus;
-
-        int first_busno;
-        int last_busno;
-
-        int hv_cfg_fd[2];       /* config{0,1} fds for this PCIe controller */
-        int hv_mem_fd;          /* fd to Hypervisor for MMIO operations */
-
-        struct pci_ops *ops;
-
-        int irq_base;           /* Base IRQ from the Hypervisor */
-        int plx_gen1;           /* flag for PLX Gen 1 configuration */
-
-        /* Address ranges that are routed to this controller/bridge. */
-        struct resource mem_resources[3];
-};
-
-static inline struct pci_controller *pci_bus_to_host(struct pci_bus *bus)
-{
-        return bus->sysdata;
-}
-
-extern void setup_indirect_pci_nomap(struct pci_controller *hose,
-                               void __iomem *cfg_addr, void __iomem *cfg_data);
-extern void setup_indirect_pci(struct pci_controller *hose,
-                               u32 cfg_addr, u32 cfg_data);
-extern void setup_grackle(struct pci_controller *hose);
-
-extern unsigned char common_swizzle(struct pci_dev *, unsigned char *);
-
-/*
- *   The following code swizzles for exactly one bridge.  The routine
- *   common_swizzle below handles multiple bridges.  But there are a
- *   some boards that don't follow the PCI spec's suggestion so we
- *   break this piece out separately.
- */
-static inline unsigned char bridge_swizzle(unsigned char pin,
-                unsigned char idsel)
-{
-        return (((pin-1) + idsel) % 4) + 1;
-}
-
-/*
- * The following macro is used to lookup irqs in a standard table
- * format for those PPC systems that do not already have PCI
- * interrupts properly routed.
- */
-/* FIXME - double check this */
-#define PCI_IRQ_TABLE_LOOKUP ({ \
-        long _ctl_ = -1; \
-        if (idsel >= min_idsel && idsel <= max_idsel && pin <= irqs_per_slot) \
-                _ctl_ = pci_irq_table[idsel - min_idsel][pin-1]; \
-        _ctl_; \
-})
-
-/*
- * Scan the buses below a given PCI host bridge and assign suitable
- * resources to all devices found.
- */
-extern int pciauto_bus_scan(struct pci_controller *, int);
-
-#ifdef CONFIG_PCI
-extern unsigned long pci_address_to_pio(phys_addr_t address);
-#else
-static inline unsigned long pci_address_to_pio(phys_addr_t address)
-{
-        return (unsigned long)-1;
-}
-#endif
-
-#endif /* _ASM_TILE_PCI_BRIDGE_H */
diff --git a/arch/tile/include/asm/pci.h b/arch/tile/include/asm/pci.h
index b0c15da2d5d5..7f03cefed1b9 100644
--- a/arch/tile/include/asm/pci.h
+++ b/arch/tile/include/asm/pci.h
@@ -15,7 +15,29 @@
 #ifndef _ASM_TILE_PCI_H
 #define _ASM_TILE_PCI_H
 
-#include <asm/pci-bridge.h>
+#include <linux/pci.h>
+
+/*
+ * Structure of a PCI controller (host bridge)
+ */
+struct pci_controller {
+        int index;              /* PCI domain number */
+        struct pci_bus *root_bus;
+
+        int first_busno;
+        int last_busno;
+
+        int hv_cfg_fd[2];       /* config{0,1} fds for this PCIe controller */
+        int hv_mem_fd;          /* fd to Hypervisor for MMIO operations */
+
+        struct pci_ops *ops;
+
+        int irq_base;           /* Base IRQ from the Hypervisor */
+        int plx_gen1;           /* flag for PLX Gen 1 configuration */
+
+        /* Address ranges that are routed to this controller/bridge. */
+        struct resource mem_resources[3];
+};
 
 /*
  * The hypervisor maps the entirety of CPA-space as bus addresses, so
@@ -24,56 +46,13 @@
  */
 #define PCI_DMA_BUS_IS_PHYS     1
 
-struct pci_controller *pci_bus_to_hose(int bus);
-unsigned char __init common_swizzle(struct pci_dev *dev, unsigned char *pinp);
-int __init tile_pci_init(void);
-void pci_iounmap(struct pci_dev *dev, void __iomem *addr);
-void __iomem *pci_iomap(struct pci_dev *dev, int bar, unsigned long max);
-void __devinit pcibios_fixup_bus(struct pci_bus *bus);
+int __devinit tile_pci_init(void);
+int __devinit pcibios_init(void);
 
-int __devinit _tile_cfg_read(struct pci_controller *hose,
-                                    int bus,
-                                    int slot,
-                                    int function,
-                                    int offset,
-                                    int size,
-                                    u32 *val);
-int __devinit _tile_cfg_write(struct pci_controller *hose,
-                                     int bus,
-                                     int slot,
-                                     int function,
-                                     int offset,
-                                     int size,
-                                     u32 val);
+void __iomem *pci_iomap(struct pci_dev *dev, int bar, unsigned long max);
+static inline void pci_iounmap(struct pci_dev *dev, void __iomem *addr) {}
 
-/*
- * These are used to to config reads and writes in the early stages of
- * setup before the driver infrastructure has been set up enough to be
- * able to do config reads and writes.
- */
-#define early_cfg_read(where, size, value) \
-        _tile_cfg_read(controller, \
-                       current_bus, \
-                       pci_slot, \
-                       pci_fn, \
-                       where, \
-                       size, \
-                       value)
-
-#define early_cfg_write(where, size, value) \
-        _tile_cfg_write(controller, \
-                       current_bus, \
-                       pci_slot, \
-                       pci_fn, \
-                       where, \
-                       size, \
-                       value)
-
-
-
-#define PCICFG_BYTE     1
-#define PCICFG_WORD     2
-#define PCICFG_DWORD    4
+void __devinit pcibios_fixup_bus(struct pci_bus *bus);
 
 #define TILE_NUM_PCIE   2
 
@@ -88,33 +67,33 @@ static inline int pci_proc_domain(struct pci_bus *bus)
 }
 
 /*
- * I/O space is currently not supported.
+ * pcibios_assign_all_busses() tells whether or not the bus numbers
+ * should be reassigned, in case the BIOS didn't do it correctly, or
+ * in case we don't have a BIOS and we want to let Linux do it.
  */
+static inline int pcibios_assign_all_busses(void)
+{
+        return 1;
+}
 
-#define TILE_PCIE_LOWER_IO              0x0
-#define TILE_PCIE_UPPER_IO              0x10000
-#define TILE_PCIE_PCIE_IO_SIZE          0x0000FFFF
-
-#define _PAGE_NO_CACHE          0
-#define _PAGE_GUARDED           0
-
-
-#define pcibios_assign_all_busses()    pci_assign_all_buses
-extern int pci_assign_all_buses;
-
+/*
+ * No special bus mastering setup handling.
+ */
 static inline void pcibios_set_master(struct pci_dev *dev)
 {
-        /* No special bus mastering setup handling */
 }
 
 #define PCIBIOS_MIN_MEM         0
-#define PCIBIOS_MIN_IO          TILE_PCIE_LOWER_IO
+#define PCIBIOS_MIN_IO          0
 
 /*
  * This flag tells if the platform is TILEmpower that needs
  * special configuration for the PLX switch chip.
  */
-extern int blade_pci;
+extern int tile_plx_gen1;
+
+/* Use any cpu for PCI. */
+#define cpumask_of_pcibus(bus) cpu_online_mask
 
 /* implement the pci_ DMA API in terms of the generic device dma_ one */
 #include <asm-generic/pci-dma-compat.h>
@@ -122,7 +101,4 @@ extern int blade_pci;
 /* generic pci stuff */
 #include <asm-generic/pci.h>
 
-/* Use any cpu for PCI. */
-#define cpumask_of_pcibus(bus) cpu_online_mask
-
 #endif /* _ASM_TILE_PCI_H */
diff --git a/arch/tile/include/asm/pgalloc.h b/arch/tile/include/asm/pgalloc.h
index cf52791a5501..e919c0bdc22d 100644
--- a/arch/tile/include/asm/pgalloc.h
+++ b/arch/tile/include/asm/pgalloc.h
@@ -41,9 +41,9 @@
 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
 {
 #ifdef CONFIG_64BIT
-        set_pte_order(pmdp, pmd, L2_USER_PGTABLE_ORDER);
+        set_pte(pmdp, pmd);
 #else
-        set_pte_order(&pmdp->pud.pgd, pmd.pud.pgd, L2_USER_PGTABLE_ORDER);
+        set_pte(&pmdp->pud.pgd, pmd.pud.pgd);
 #endif
 }
 
@@ -100,6 +100,9 @@ pte_t *get_prealloc_pte(unsigned long pfn);
 /* During init, we can shatter kernel huge pages if needed. */
 void shatter_pmd(pmd_t *pmd);
 
+/* After init, a more complex technique is required. */
+void shatter_huge_page(unsigned long addr);
+
 #ifdef __tilegx__
 /* We share a single page allocator for both L1 and L2 page tables. */
 #if HV_L1_SIZE != HV_L2_SIZE
diff --git a/arch/tile/include/asm/pgtable.h b/arch/tile/include/asm/pgtable.h
index b3367379d537..1a20b7ef8ea2 100644
--- a/arch/tile/include/asm/pgtable.h
+++ b/arch/tile/include/asm/pgtable.h
@@ -233,15 +233,23 @@ static inline void __pte_clear(pte_t *ptep)
 #define pgd_ERROR(e) \
         pr_err("%s:%d: bad pgd 0x%016llx.\n", __FILE__, __LINE__, pgd_val(e))
 
+/* Return PA and protection info for a given kernel VA. */
+int va_to_cpa_and_pte(void *va, phys_addr_t *cpa, pte_t *pte);
+
+/*
+ * __set_pte() ensures we write the 64-bit PTE with 32-bit words in
+ * the right order on 32-bit platforms and also allows us to write
+ * hooks to check valid PTEs, etc., if we want.
+ */
+void __set_pte(pte_t *ptep, pte_t pte);
+
 /*
- * set_pte_order() sets the given PTE and also sanity-checks the
+ * set_pte() sets the given PTE and also sanity-checks the
  * requested PTE against the page homecaching.  Unspecified parts
  * of the PTE are filled in when it is written to memory, i.e. all
  * caching attributes if "!forcecache", or the home cpu if "anyhome".
  */
-extern void set_pte_order(pte_t *ptep, pte_t pte, int order);
-
-#define set_pte(ptep, pteval) set_pte_order(ptep, pteval, 0)
+extern void set_pte(pte_t *ptep, pte_t pte);
 #define set_pte_at(mm, addr, ptep, pteval) set_pte(ptep, pteval)
 #define set_pte_atomic(pteptr, pteval) set_pte(pteptr, pteval)
 
@@ -293,21 +301,6 @@ extern void check_mm_caching(struct mm_struct *prev, struct mm_struct *next);
 #define __swp_entry_to_pte(swp) ((pte_t) { (((long long) ((swp).val)) << 32) })
 
 /*
- * clone_pgd_range(pgd_t *dst, pgd_t *src, int count);
- *
- *  dst - pointer to pgd range anwhere on a pgd page
- *  src - ""
- *  count - the number of pgds to copy.
- *
- * dst and src can be on the same page, but the range must not overlap,
- * and must not cross a page boundary.
- */
-static inline void clone_pgd_range(pgd_t *dst, pgd_t *src, int count)
-{
-       memcpy(dst, src, count * sizeof(pgd_t));
-}
-
-/*
  * Conversion functions: convert a page and protection to a page entry,
  * and a page entry and page directory to the page they refer to.
  */
@@ -344,18 +337,11 @@ static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
 #define pgd_offset_k(address) pgd_offset(&init_mm, address)
 
 #if defined(CONFIG_HIGHPTE)
-extern pte_t *_pte_offset_map(pmd_t *, unsigned long address, enum km_type);
-#define pte_offset_map(dir, address) \
-        _pte_offset_map(dir, address, KM_PTE0)
-#define pte_offset_map_nested(dir, address) \
-        _pte_offset_map(dir, address, KM_PTE1)
-#define pte_unmap(pte) kunmap_atomic(pte, KM_PTE0)
-#define pte_unmap_nested(pte) kunmap_atomic(pte, KM_PTE1)
+extern pte_t *pte_offset_map(pmd_t *, unsigned long address);
+#define pte_unmap(pte) kunmap_atomic(pte)
 #else
 #define pte_offset_map(dir, address) pte_offset_kernel(dir, address)
-#define pte_offset_map_nested(dir, address) pte_offset_map(dir, address)
 #define pte_unmap(pte) do { } while (0)
-#define pte_unmap_nested(pte) do { } while (0)
 #endif
 
 /* Clear a non-executable kernel PTE and flush it from the TLB. */
diff --git a/arch/tile/include/asm/pgtable_32.h b/arch/tile/include/asm/pgtable_32.h
index 53ec34884744..9f98529761fd 100644
--- a/arch/tile/include/asm/pgtable_32.h
+++ b/arch/tile/include/asm/pgtable_32.h
@@ -24,6 +24,7 @@
 #define PGDIR_SIZE      HV_PAGE_SIZE_LARGE
 #define PGDIR_MASK      (~(PGDIR_SIZE-1))
 #define PTRS_PER_PGD    (1 << (32 - PGDIR_SHIFT))
+#define SIZEOF_PGD      (PTRS_PER_PGD * sizeof(pgd_t))
 
 /*
  * The level-2 index is defined by the difference between the huge
@@ -33,6 +34,7 @@
  * this nomenclature is somewhat confusing.
  */
 #define PTRS_PER_PTE (1 << (HV_LOG2_PAGE_SIZE_LARGE - HV_LOG2_PAGE_SIZE_SMALL))
+#define SIZEOF_PTE      (PTRS_PER_PTE * sizeof(pte_t))
 
 #ifndef __ASSEMBLY__
 
@@ -94,7 +96,6 @@ static inline int pgd_addr_invalid(unsigned long addr)
  */
 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
-#define __HAVE_ARCH_PTEP_GET_AND_CLEAR
 
 extern int ptep_test_and_clear_young(struct vm_area_struct *,
                                      unsigned long addr, pte_t *);
@@ -110,6 +111,11 @@ static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
         return pte;
 }
 
+static inline void __set_pmd(pmd_t *pmdp, pmd_t pmdval)
+{
+        set_pte(&pmdp->pud.pgd, pmdval.pud.pgd);
+}
+
 /* Create a pmd from a PTFN. */
 static inline pmd_t ptfn_pmd(unsigned long ptfn, pgprot_t prot)
 {
diff --git a/arch/tile/include/asm/pgtable_64.h b/arch/tile/include/asm/pgtable_64.h
new file mode 100644
index 000000000000..fd80328523b4
--- /dev/null
+++ b/arch/tile/include/asm/pgtable_64.h
@@ -0,0 +1,175 @@
+/*
+ * Copyright 2011 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ *
+ */
+
+#ifndef _ASM_TILE_PGTABLE_64_H
+#define _ASM_TILE_PGTABLE_64_H
+
+/* The level-0 page table breaks the address space into 32-bit chunks. */
+#define PGDIR_SHIFT     HV_LOG2_L1_SPAN
+#define PGDIR_SIZE      HV_L1_SPAN
+#define PGDIR_MASK      (~(PGDIR_SIZE-1))
+#define PTRS_PER_PGD    HV_L0_ENTRIES
+#define SIZEOF_PGD      (PTRS_PER_PGD * sizeof(pgd_t))
+
+/*
+ * The level-1 index is defined by the huge page size.  A PMD is composed
+ * of PTRS_PER_PMD pgd_t's and is the middle level of the page table.
+ */
+#define PMD_SHIFT       HV_LOG2_PAGE_SIZE_LARGE
+#define PMD_SIZE        HV_PAGE_SIZE_LARGE
+#define PMD_MASK        (~(PMD_SIZE-1))
+#define PTRS_PER_PMD    (1 << (PGDIR_SHIFT - PMD_SHIFT))
+#define SIZEOF_PMD      (PTRS_PER_PMD * sizeof(pmd_t))
+
+/*
+ * The level-2 index is defined by the difference between the huge
+ * page size and the normal page size.  A PTE is composed of
+ * PTRS_PER_PTE pte_t's and is the bottom level of the page table.
+ * Note that the hypervisor docs use PTE for what we call pte_t, so
+ * this nomenclature is somewhat confusing.
+ */
+#define PTRS_PER_PTE (1 << (HV_LOG2_PAGE_SIZE_LARGE - HV_LOG2_PAGE_SIZE_SMALL))
+#define SIZEOF_PTE      (PTRS_PER_PTE * sizeof(pte_t))
+
+/*
+ * Align the vmalloc area to an L2 page table, and leave a guard page
+ * at the beginning and end.  The vmalloc code also puts in an internal
+ * guard page between each allocation.
+ */
+#define _VMALLOC_END    HUGE_VMAP_BASE
+#define VMALLOC_END     (_VMALLOC_END - PAGE_SIZE)
+#define VMALLOC_START   (_VMALLOC_START + PAGE_SIZE)
+
+#define HUGE_VMAP_END   (HUGE_VMAP_BASE + PGDIR_SIZE)
+
+#ifndef __ASSEMBLY__
+
+/* We have no pud since we are a three-level page table. */
+#include <asm-generic/pgtable-nopud.h>
+
+static inline int pud_none(pud_t pud)
+{
+        return pud_val(pud) == 0;
+}
+
+static inline int pud_present(pud_t pud)
+{
+        return pud_val(pud) & _PAGE_PRESENT;
+}
+
+#define pmd_ERROR(e) \
+        pr_err("%s:%d: bad pmd 0x%016llx.\n", __FILE__, __LINE__, pmd_val(e))
+
+static inline void pud_clear(pud_t *pudp)
+{
+        __pte_clear(&pudp->pgd);
+}
+
+static inline int pud_bad(pud_t pud)
+{
+        return ((pud_val(pud) & _PAGE_ALL) != _PAGE_TABLE);
+}
+
+/* Return the page-table frame number (ptfn) that a pud_t points at. */
+#define pud_ptfn(pud) hv_pte_get_ptfn((pud).pgd)
+
+/*
+ * A given kernel pud_t maps to a kernel pmd_t table at a specific
+ * virtual address.  Since kernel pmd_t tables can be aligned at
+ * sub-page granularity, this macro can return non-page-aligned
+ * pointers, despite its name.
+ */
+#define pud_page_vaddr(pud) \
+        (__va((phys_addr_t)pud_ptfn(pud) << HV_LOG2_PAGE_TABLE_ALIGN))
+
+/*
+ * A pud_t points to a pmd_t array.  Since we can have multiple per
+ * page, we don't have a one-to-one mapping of pud_t's to pages.
+ */
+#define pud_page(pud) pfn_to_page(HV_PTFN_TO_PFN(pud_ptfn(pud)))
+
+static inline unsigned long pud_index(unsigned long address)
+{
+        return (address >> PUD_SHIFT) & (PTRS_PER_PUD - 1);
+}
+
+#define pmd_offset(pud, address) \
+        ((pmd_t *)pud_page_vaddr(*(pud)) + pmd_index(address))
+
+static inline void __set_pmd(pmd_t *pmdp, pmd_t pmdval)
+{
+        set_pte(pmdp, pmdval);
+}
+
+/* Create a pmd from a PTFN and pgprot. */
+static inline pmd_t ptfn_pmd(unsigned long ptfn, pgprot_t prot)
+{
+        return hv_pte_set_ptfn(prot, ptfn);
+}
+
+/* Return the page-table frame number (ptfn) that a pmd_t points at. */
+static inline unsigned long pmd_ptfn(pmd_t pmd)
+{
+        return hv_pte_get_ptfn(pmd);
+}
+
+static inline void pmd_clear(pmd_t *pmdp)
+{
+        __pte_clear(pmdp);
+}
+
+/* Normalize an address to having the correct high bits set. */
+#define pgd_addr_normalize pgd_addr_normalize
+static inline unsigned long pgd_addr_normalize(unsigned long addr)
+{
+        return ((long)addr << (CHIP_WORD_SIZE() - CHIP_VA_WIDTH())) >>
+                (CHIP_WORD_SIZE() - CHIP_VA_WIDTH());
+}
+
+/* We don't define any pgds for these addresses. */
+static inline int pgd_addr_invalid(unsigned long addr)
+{
+        return addr >= MEM_HV_START ||
+                (addr > MEM_LOW_END && addr < MEM_HIGH_START);
+}
+
+/*
+ * Use atomic instructions to provide atomicity against the hypervisor.
+ */
+#define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
+static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
+                                            unsigned long addr, pte_t *ptep)
+{
+        return (__insn_fetchand(&ptep->val, ~HV_PTE_ACCESSED) >>
+                HV_PTE_INDEX_ACCESSED) & 0x1;
+}
+
+#define __HAVE_ARCH_PTEP_SET_WRPROTECT
+static inline void ptep_set_wrprotect(struct mm_struct *mm,
+                                      unsigned long addr, pte_t *ptep)
+{
+        __insn_fetchand(&ptep->val, ~HV_PTE_WRITABLE);
+}
+
+#define __HAVE_ARCH_PTEP_GET_AND_CLEAR
+static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
+                                       unsigned long addr, pte_t *ptep)
+{
+        return hv_pte(__insn_exch(&ptep->val, 0UL));
+}
+
+#endif /* __ASSEMBLY__ */
+
+#endif /* _ASM_TILE_PGTABLE_64_H */
diff --git a/arch/tile/include/asm/processor.h b/arch/tile/include/asm/processor.h
index ccd5f8425688..34c1e01ffb5e 100644
--- a/arch/tile/include/asm/processor.h
+++ b/arch/tile/include/asm/processor.h
@@ -215,6 +215,8 @@ static inline void release_thread(struct task_struct *dead_task)
 
 extern int kernel_thread(int (*fn)(void *), void *arg, unsigned long flags);
 
+extern int do_work_pending(struct pt_regs *regs, u32 flags);
+
 
 /*
  * Return saved (kernel) PC of a blocked thread.
@@ -255,10 +257,6 @@ static inline void cpu_relax(void)
         barrier();
 }
 
-struct siginfo;
-extern void arch_coredump_signal(struct siginfo *, struct pt_regs *);
-#define arch_coredump_signal arch_coredump_signal
-
 /* Info on this processor (see fs/proc/cpuinfo.c) */
 struct seq_operations;
 extern const struct seq_operations cpuinfo_op;
@@ -269,10 +267,6 @@ extern char chip_model[64];
 /* Data on which physical memory controller corresponds to which NUMA node. */
 extern int node_controller[];
 
-
-/* Do we dump information to the console when a user application crashes? */
-extern int show_crashinfo;
-
 #if CHIP_HAS_CBOX_HOME_MAP()
 /* Does the heap allocator return hash-for-home pages by default? */
 extern int hash_default;
@@ -292,8 +286,18 @@ extern int kstack_hash;
 /* Are we using huge pages in the TLB for kernel data? */
 extern int kdata_huge;
 
+/* Support standard Linux prefetching. */
+#define ARCH_HAS_PREFETCH
+#define prefetch(x) __builtin_prefetch(x)
 #define PREFETCH_STRIDE CHIP_L2_LINE_SIZE()
 
+/* Bring a value into the L1D, faulting the TLB if necessary. */
+#ifdef __tilegx__
+#define prefetch_L1(x) __insn_prefetch_l1_fault((void *)(x))
+#else
+#define prefetch_L1(x) __insn_prefetch_L1((void *)(x))
+#endif
+
 #else /* __ASSEMBLY__ */
 
 /* Do some slow action (e.g. read a slow SPR). */
@@ -328,18 +332,21 @@ extern int kdata_huge;
  * Note that assembly code assumes that USER_PL is zero.
  */
 #define USER_PL 0
-#define KERNEL_PL 1
+#if CONFIG_KERNEL_PL == 2
+#define GUEST_PL 1
+#endif
+#define KERNEL_PL CONFIG_KERNEL_PL
 
-/* SYSTEM_SAVE_1_0 holds the current cpu number ORed with ksp0. */
+/* SYSTEM_SAVE_K_0 holds the current cpu number ORed with ksp0. */
 #define CPU_LOG_MASK_VALUE 12
 #define CPU_MASK_VALUE ((1 << CPU_LOG_MASK_VALUE) - 1)
 #if CONFIG_NR_CPUS > CPU_MASK_VALUE
 # error Too many cpus!
 #endif
 #define raw_smp_processor_id() \
-        ((int)__insn_mfspr(SPR_SYSTEM_SAVE_1_0) & CPU_MASK_VALUE)
+        ((int)__insn_mfspr(SPR_SYSTEM_SAVE_K_0) & CPU_MASK_VALUE)
 #define get_current_ksp0() \
-        (__insn_mfspr(SPR_SYSTEM_SAVE_1_0) & ~CPU_MASK_VALUE)
+        (__insn_mfspr(SPR_SYSTEM_SAVE_K_0) & ~CPU_MASK_VALUE)
 #define next_current_ksp0(task) ({ \
         unsigned long __ksp0 = task_ksp0(task); \
         int __cpu = raw_smp_processor_id(); \
diff --git a/arch/tile/include/asm/ptrace.h b/arch/tile/include/asm/ptrace.h
index 4a02bb073979..6be2246e015c 100644
--- a/arch/tile/include/asm/ptrace.h
+++ b/arch/tile/include/asm/ptrace.h
@@ -62,8 +62,8 @@ struct pt_regs {
         pt_reg_t lr;            /* aliases regs[TREG_LR] */
 
         /* Saved special registers. */
-        pt_reg_t pc;            /* stored in EX_CONTEXT_1_0 */
-        pt_reg_t ex1;           /* stored in EX_CONTEXT_1_1 (PL and ICS bit) */
+        pt_reg_t pc;            /* stored in EX_CONTEXT_K_0 */
+        pt_reg_t ex1;           /* stored in EX_CONTEXT_K_1 (PL and ICS bit) */
         pt_reg_t faultnum;      /* fault number (INT_SWINT_1 for syscall) */
         pt_reg_t orig_r0;       /* r0 at syscall entry, else zero */
         pt_reg_t flags;         /* flags (see below) */
@@ -141,6 +141,9 @@ struct single_step_state {
 /* Single-step the instruction at regs->pc */
 extern void single_step_once(struct pt_regs *regs);
 
+/* Clean up after execve(). */
+extern void single_step_execve(void);
+
 struct task_struct;
 
 extern void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs,
diff --git a/arch/tile/include/asm/serial.h b/arch/tile/include/asm/serial.h
new file mode 100644
index 000000000000..a0cb0caff152
--- /dev/null
+++ b/arch/tile/include/asm/serial.h
@@ -0,0 +1 @@
+#include <asm-generic/serial.h>
diff --git a/arch/tile/include/asm/signal.h b/arch/tile/include/asm/signal.h
index c1ee1d61d44c..1e1e616783eb 100644
--- a/arch/tile/include/asm/signal.h
+++ b/arch/tile/include/asm/signal.h
@@ -25,9 +25,13 @@
 
 #if defined(__KERNEL__) && !defined(__ASSEMBLY__)
 struct pt_regs;
-int restore_sigcontext(struct pt_regs *, struct sigcontext __user *, long *);
+int restore_sigcontext(struct pt_regs *, struct sigcontext __user *);
 int setup_sigcontext(struct sigcontext __user *, struct pt_regs *);
 void do_signal(struct pt_regs *regs);
+void signal_fault(const char *type, struct pt_regs *,
+                  void __user *frame, int sig);
+void trace_unhandled_signal(const char *type, struct pt_regs *regs,
+                            unsigned long address, int signo);
 #endif
 
 #endif /* _ASM_TILE_SIGNAL_H */
diff --git a/arch/tile/include/asm/spinlock_32.h b/arch/tile/include/asm/spinlock_32.h
index 88efdde8dd2b..a8f2c6e31a87 100644
--- a/arch/tile/include/asm/spinlock_32.h
+++ b/arch/tile/include/asm/spinlock_32.h
@@ -78,13 +78,6 @@ void arch_spin_unlock_wait(arch_spinlock_t *lock);
 #define _RD_COUNT_SHIFT 24
 #define _RD_COUNT_WIDTH 8
 
-/* Internal functions; do not use. */
-void arch_read_lock_slow(arch_rwlock_t *, u32);
-int arch_read_trylock_slow(arch_rwlock_t *);
-void arch_read_unlock_slow(arch_rwlock_t *);
-void arch_write_lock_slow(arch_rwlock_t *, u32);
-void arch_write_unlock_slow(arch_rwlock_t *, u32);
-
 /**
  * arch_read_can_lock() - would read_trylock() succeed?
  */
@@ -104,94 +97,32 @@ static inline int arch_write_can_lock(arch_rwlock_t *rwlock)
 /**
  * arch_read_lock() - acquire a read lock.
  */
-static inline void arch_read_lock(arch_rwlock_t *rwlock)
-{
-        u32 val = __insn_tns((int *)&rwlock->lock);
-        if (unlikely(val << _RD_COUNT_WIDTH)) {
-                arch_read_lock_slow(rwlock, val);
-                return;
-        }
-        rwlock->lock = val + (1 << _RD_COUNT_SHIFT);
-}
+void arch_read_lock(arch_rwlock_t *rwlock);
 
 /**
- * arch_read_lock() - acquire a write lock.
+ * arch_write_lock() - acquire a write lock.
  */
-static inline void arch_write_lock(arch_rwlock_t *rwlock)
-{
-        u32 val = __insn_tns((int *)&rwlock->lock);
-        if (unlikely(val != 0)) {
-                arch_write_lock_slow(rwlock, val);
-                return;
-        }
-        rwlock->lock = 1 << _WR_NEXT_SHIFT;
-}
+void arch_write_lock(arch_rwlock_t *rwlock);
 
 /**
  * arch_read_trylock() - try to acquire a read lock.
  */
-static inline int arch_read_trylock(arch_rwlock_t *rwlock)
-{
-        int locked;
-        u32 val = __insn_tns((int *)&rwlock->lock);
-        if (unlikely(val & 1))
-                return arch_read_trylock_slow(rwlock);
-        locked = (val << _RD_COUNT_WIDTH) == 0;
-        rwlock->lock = val + (locked << _RD_COUNT_SHIFT);
-        return locked;
-}
+int arch_read_trylock(arch_rwlock_t *rwlock);
 
 /**
  * arch_write_trylock() - try to acquire a write lock.
  */
-static inline int arch_write_trylock(arch_rwlock_t *rwlock)
-{
-        u32 val = __insn_tns((int *)&rwlock->lock);
-
-        /*
-         * If a tns is in progress, or there's a waiting or active locker,
-         * or active readers, we can't take the lock, so give up.
-         */
-        if (unlikely(val != 0)) {
-                if (!(val & 1))
-                        rwlock->lock = val;
-                return 0;
-        }
-
-        /* Set the "next" field to mark it locked. */
-        rwlock->lock = 1 << _WR_NEXT_SHIFT;
-        return 1;
-}
+int arch_write_trylock(arch_rwlock_t *rwlock);
 
 /**
  * arch_read_unlock() - release a read lock.
  */
-static inline void arch_read_unlock(arch_rwlock_t *rwlock)
-{
-        u32 val;
-        mb();  /* guarantee anything modified under the lock is visible */
-        val = __insn_tns((int *)&rwlock->lock);
-        if (unlikely(val & 1)) {
-                arch_read_unlock_slow(rwlock);
-                return;
-        }
-        rwlock->lock = val - (1 << _RD_COUNT_SHIFT);
-}
+void arch_read_unlock(arch_rwlock_t *rwlock);
 
 /**
  * arch_write_unlock() - release a write lock.
  */
-static inline void arch_write_unlock(arch_rwlock_t *rwlock)
-{
-        u32 val;
-        mb();  /* guarantee anything modified under the lock is visible */
-        val = __insn_tns((int *)&rwlock->lock);
-        if (unlikely(val != (1 << _WR_NEXT_SHIFT))) {
-                arch_write_unlock_slow(rwlock, val);
-                return;
-        }
-        rwlock->lock = 0;
-}
+void arch_write_unlock(arch_rwlock_t *rwlock);
 
 #define arch_read_lock_flags(lock, flags) arch_read_lock(lock)
 #define arch_write_lock_flags(lock, flags) arch_write_lock(lock)
diff --git a/arch/tile/include/asm/spinlock_64.h b/arch/tile/include/asm/spinlock_64.h
new file mode 100644
index 000000000000..72be5904e020
--- /dev/null
+++ b/arch/tile/include/asm/spinlock_64.h
@@ -0,0 +1,161 @@
+/*
+ * Copyright 2011 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ *
+ * 64-bit SMP ticket spinlocks, allowing only a single CPU anywhere
+ * (the type definitions are in asm/spinlock_types.h)
+ */
+
+#ifndef _ASM_TILE_SPINLOCK_64_H
+#define _ASM_TILE_SPINLOCK_64_H
+
+/* Shifts and masks for the various fields in "lock". */
+#define __ARCH_SPIN_CURRENT_SHIFT       17
+#define __ARCH_SPIN_NEXT_MASK           0x7fff
+#define __ARCH_SPIN_NEXT_OVERFLOW       0x8000
+
+/*
+ * Return the "current" portion of a ticket lock value,
+ * i.e. the number that currently owns the lock.
+ */
+static inline int arch_spin_current(u32 val)
+{
+        return val >> __ARCH_SPIN_CURRENT_SHIFT;
+}
+
+/*
+ * Return the "next" portion of a ticket lock value,
+ * i.e. the number that the next task to try to acquire the lock will get.
+ */
+static inline int arch_spin_next(u32 val)
+{
+        return val & __ARCH_SPIN_NEXT_MASK;
+}
+
+/* The lock is locked if a task would have to wait to get it. */
+static inline int arch_spin_is_locked(arch_spinlock_t *lock)
+{
+        u32 val = lock->lock;
+        return arch_spin_current(val) != arch_spin_next(val);
+}
+
+/* Bump the current ticket so the next task owns the lock. */
+static inline void arch_spin_unlock(arch_spinlock_t *lock)
+{
+        wmb();  /* guarantee anything modified under the lock is visible */
+        __insn_fetchadd4(&lock->lock, 1U << __ARCH_SPIN_CURRENT_SHIFT);
+}
+
+void arch_spin_unlock_wait(arch_spinlock_t *lock);
+
+void arch_spin_lock_slow(arch_spinlock_t *lock, u32 val);
+
+/* Grab the "next" ticket number and bump it atomically.
+ * If the current ticket is not ours, go to the slow path.
+ * We also take the slow path if the "next" value overflows.
+ */
+static inline void arch_spin_lock(arch_spinlock_t *lock)
+{
+        u32 val = __insn_fetchadd4(&lock->lock, 1);
+        u32 ticket = val & (__ARCH_SPIN_NEXT_MASK | __ARCH_SPIN_NEXT_OVERFLOW);
+        if (unlikely(arch_spin_current(val) != ticket))
+                arch_spin_lock_slow(lock, ticket);
+}
+
+/* Try to get the lock, and return whether we succeeded. */
+int arch_spin_trylock(arch_spinlock_t *lock);
+
+/* We cannot take an interrupt after getting a ticket, so don't enable them. */
+#define arch_spin_lock_flags(lock, flags) arch_spin_lock(lock)
+
+/*
+ * Read-write spinlocks, allowing multiple readers
+ * but only one writer.
+ *
+ * We use fetchadd() for readers, and fetchor() with the sign bit
+ * for writers.
+ */
+
+#define __WRITE_LOCK_BIT (1 << 31)
+
+static inline int arch_write_val_locked(int val)
+{
+        return val < 0;  /* Optimize "val & __WRITE_LOCK_BIT". */
+}
+
+/**
+ * read_can_lock - would read_trylock() succeed?
+ * @lock: the rwlock in question.
+ */
+static inline int arch_read_can_lock(arch_rwlock_t *rw)
+{
+        return !arch_write_val_locked(rw->lock);
+}
+
+/**
+ * write_can_lock - would write_trylock() succeed?
+ * @lock: the rwlock in question.
+ */
+static inline int arch_write_can_lock(arch_rwlock_t *rw)
+{
+        return rw->lock == 0;
+}
+
+extern void __read_lock_failed(arch_rwlock_t *rw);
+
+static inline void arch_read_lock(arch_rwlock_t *rw)
+{
+        u32 val = __insn_fetchaddgez4(&rw->lock, 1);
+        if (unlikely(arch_write_val_locked(val)))
+                __read_lock_failed(rw);
+}
+
+extern void __write_lock_failed(arch_rwlock_t *rw, u32 val);
+
+static inline void arch_write_lock(arch_rwlock_t *rw)
+{
+        u32 val = __insn_fetchor4(&rw->lock, __WRITE_LOCK_BIT);
+        if (unlikely(val != 0))
+                __write_lock_failed(rw, val);
+}
+
+static inline void arch_read_unlock(arch_rwlock_t *rw)
+{
+        __insn_mf();
+        __insn_fetchadd4(&rw->lock, -1);
+}
+
+static inline void arch_write_unlock(arch_rwlock_t *rw)
+{
+        __insn_mf();
+        rw->lock = 0;
+}
+
+static inline int arch_read_trylock(arch_rwlock_t *rw)
+{
+        return !arch_write_val_locked(__insn_fetchaddgez4(&rw->lock, 1));
+}
+
+static inline int arch_write_trylock(arch_rwlock_t *rw)
+{
+        u32 val = __insn_fetchor4(&rw->lock, __WRITE_LOCK_BIT);
+        if (likely(val == 0))
+                return 1;
+        if (!arch_write_val_locked(val))
+                __insn_fetchand4(&rw->lock, ~__WRITE_LOCK_BIT);
+        return 0;
+}
+
+#define arch_read_lock_flags(lock, flags) arch_read_lock(lock)
+#define arch_write_lock_flags(lock, flags) arch_write_lock(lock)
+
+#endif /* _ASM_TILE_SPINLOCK_64_H */
diff --git a/arch/tile/include/asm/stack.h b/arch/tile/include/asm/stack.h
index f908473c322d..4d97a2db932e 100644
--- a/arch/tile/include/asm/stack.h
+++ b/arch/tile/include/asm/stack.h
@@ -18,13 +18,14 @@
 #include <linux/types.h>
 #include <linux/sched.h>
 #include <asm/backtrace.h>
+#include <asm/page.h>
 #include <hv/hypervisor.h>
 
 /* Everything we need to keep track of a backtrace iteration */
 struct KBacktraceIterator {
         BacktraceIterator it;
         struct task_struct *task;     /* task we are backtracing */
-        HV_PTE *pgtable;              /* page table for user space access */
+        pte_t *pgtable;               /* page table for user space access */
         int end;                      /* iteration complete. */
         int new_context;              /* new context is starting */
         int profile;                  /* profiling, so stop on async intrpt */
diff --git a/arch/tile/include/asm/stat.h b/arch/tile/include/asm/stat.h
index 3dc90fa92c70..c0db34d56be3 100644
--- a/arch/tile/include/asm/stat.h
+++ b/arch/tile/include/asm/stat.h
@@ -1 +1,4 @@
+#if defined(__KERNEL__) && defined(CONFIG_COMPAT)
+#define __ARCH_WANT_STAT64      /* Used for compat_sys_stat64() etc. */
+#endif
 #include <asm-generic/stat.h>
diff --git a/arch/tile/include/asm/swab.h b/arch/tile/include/asm/swab.h
index 25c686a00f1d..7c37b38f6c8d 100644
--- a/arch/tile/include/asm/swab.h
+++ b/arch/tile/include/asm/swab.h
@@ -18,12 +18,6 @@
 /* Tile gcc is always >= 4.3.0, so we use __builtin_bswap. */
 #define __arch_swab32(x) __builtin_bswap32(x)
 #define __arch_swab64(x) __builtin_bswap64(x)
-
-/* Use the variant that is natural for the wordsize. */
-#ifdef CONFIG_64BIT
-#define __arch_swab16(x) (__builtin_bswap64(x) >> 48)
-#else
 #define __arch_swab16(x) (__builtin_bswap32(x) >> 16)
-#endif
 
 #endif /* _ASM_TILE_SWAB_H */
diff --git a/arch/tile/include/asm/syscalls.h b/arch/tile/include/asm/syscalls.h
index ce99ffefeacf..3b5507c31eae 100644
--- a/arch/tile/include/asm/syscalls.h
+++ b/arch/tile/include/asm/syscalls.h
@@ -32,8 +32,9 @@ extern void *compat_sys_call_table[];
 
 /*
  * Note that by convention, any syscall which requires the current
- * register set takes an additional "struct pt_regs *" pointer; the
- * sys_xxx() function just adds the pointer and tail-calls to _sys_xxx().
+ * register set takes an additional "struct pt_regs *" pointer; a
+ * _sys_xxx() trampoline in intvec*.S just sets up the pointer and
+ * jumps to sys_xxx().
  */
 
 /* kernel/sys.c */
@@ -43,66 +44,17 @@ long sys32_fadvise64(int fd, u32 offset_lo, u32 offset_hi,
 int sys32_fadvise64_64(int fd, u32 offset_lo, u32 offset_hi,
                        u32 len_lo, u32 len_hi, int advice);
 long sys_flush_cache(void);
-long sys_mmap2(unsigned long addr, unsigned long len,
-               unsigned long prot, unsigned long flags,
-               unsigned long fd, unsigned long pgoff);
-#ifdef __tilegx__
-long sys_mmap(unsigned long addr, unsigned long len,
-              unsigned long prot, unsigned long flags,
-              unsigned long fd, off_t pgoff);
+#ifndef __tilegx__  /* No mmap() in the 32-bit kernel. */
+#define sys_mmap sys_mmap
 #endif
 
-/* kernel/process.c */
-long sys_clone(unsigned long clone_flags, unsigned long newsp,
-               void __user *parent_tid, void __user *child_tid);
-long _sys_clone(unsigned long clone_flags, unsigned long newsp,
-                void __user *parent_tid, void __user *child_tid,
-                struct pt_regs *regs);
-long sys_fork(void);
-long _sys_fork(struct pt_regs *regs);
-long sys_vfork(void);
-long _sys_vfork(struct pt_regs *regs);
-long sys_execve(const char __user *filename,
-                const char __user *const __user *argv,
-                const char __user *const __user *envp);
-long _sys_execve(const char __user *filename,
-                 const char __user *const __user *argv,
-                 const char __user *const __user *envp, struct pt_regs *regs);
-
-/* kernel/signal.c */
-long sys_sigaltstack(const stack_t __user *, stack_t __user *);
-long _sys_sigaltstack(const stack_t __user *, stack_t __user *,
-                      struct pt_regs *);
-long sys_rt_sigreturn(void);
-long _sys_rt_sigreturn(struct pt_regs *regs);
-
-/* platform-independent functions */
-long sys_rt_sigsuspend(sigset_t __user *unewset, size_t sigsetsize);
-long sys_rt_sigaction(int sig, const struct sigaction __user *act,
-                      struct sigaction __user *oact, size_t sigsetsize);
-
 #ifndef __tilegx__
 /* mm/fault.c */
-int sys_cmpxchg_badaddr(unsigned long address);
-int _sys_cmpxchg_badaddr(unsigned long address, struct pt_regs *);
+long sys_cmpxchg_badaddr(unsigned long address, struct pt_regs *);
+long _sys_cmpxchg_badaddr(unsigned long address);
 #endif
 
 #ifdef CONFIG_COMPAT
-long compat_sys_execve(const char __user *path,
-                       const compat_uptr_t __user *argv,
-                       const compat_uptr_t __user *envp);
-long _compat_sys_execve(const char __user *path,
-                        const compat_uptr_t __user *argv,
-                        const compat_uptr_t __user *envp,
-                        struct pt_regs *regs);
-long compat_sys_sigaltstack(const struct compat_sigaltstack __user *uss_ptr,
-                            struct compat_sigaltstack __user *uoss_ptr);
-long _compat_sys_sigaltstack(const struct compat_sigaltstack __user *uss_ptr,
-                             struct compat_sigaltstack __user *uoss_ptr,
-                             struct pt_regs *regs);
-long compat_sys_rt_sigreturn(void);
-long _compat_sys_rt_sigreturn(struct pt_regs *regs);
-
 /* These four are not defined for 64-bit, but serve as "compat" syscalls. */
 long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg);
 long sys_fstat64(unsigned long fd, struct stat64 __user *statbuf);
@@ -110,4 +62,15 @@ long sys_truncate64(const char __user *path, loff_t length);
 long sys_ftruncate64(unsigned int fd, loff_t length);
 #endif
 
+/* These are the intvec*.S trampolines. */
+long _sys_sigaltstack(const stack_t __user *, stack_t __user *);
+long _sys_rt_sigreturn(void);
+long _sys_clone(unsigned long clone_flags, unsigned long newsp,
+                void __user *parent_tid, void __user *child_tid);
+long _sys_execve(const char __user *filename,
+                 const char __user *const __user *argv,
+                 const char __user *const __user *envp);
+
+#include <asm-generic/syscalls.h>
+
 #endif /* _ASM_TILE_SYSCALLS_H */
diff --git a/arch/tile/include/asm/system.h b/arch/tile/include/asm/system.h
index f749be327ce0..23d1842f4839 100644
--- a/arch/tile/include/asm/system.h
+++ b/arch/tile/include/asm/system.h
@@ -89,6 +89,27 @@
 #define get_cycles_low() __insn_mfspr(SPR_CYCLE)   /* just get all 64 bits */
 #endif
 
+#if !CHIP_HAS_MF_WAITS_FOR_VICTIMS()
+#include <hv/syscall_public.h>
+/*
+ * Issue an uncacheable load to each memory controller, then
+ * wait until those loads have completed.
+ */
+static inline void __mb_incoherent(void)
+{
+        long clobber_r10;
+        asm volatile("swint2"
+                     : "=R10" (clobber_r10)
+                     : "R10" (HV_SYS_fence_incoherent)
+                     : "r0", "r1", "r2", "r3", "r4",
+                       "r5", "r6", "r7", "r8", "r9",
+                       "r11", "r12", "r13", "r14",
+                       "r15", "r16", "r17", "r18", "r19",
+                       "r20", "r21", "r22", "r23", "r24",
+                       "r25", "r26", "r27", "r28", "r29");
+}
+#endif
+
 /* Fence to guarantee visibility of stores to incoherent memory. */
 static inline void
 mb_incoherent(void)
@@ -97,7 +118,6 @@ mb_incoherent(void)
 
 #if !CHIP_HAS_MF_WAITS_FOR_VICTIMS()
         {
-                int __mb_incoherent(void);
 #if CHIP_HAS_TILE_WRITE_PENDING()
                 const unsigned long WRITE_TIMEOUT_CYCLES = 400;
                 unsigned long start = get_cycles_low();
@@ -161,7 +181,7 @@ extern struct task_struct *_switch_to(struct task_struct *prev,
 /* Helper function for _switch_to(). */
 extern struct task_struct *__switch_to(struct task_struct *prev,
                                        struct task_struct *next,
-                                       unsigned long new_system_save_1_0);
+                                       unsigned long new_system_save_k_0);
 
 /* Address that switched-away from tasks are at. */
 extern unsigned long get_switch_to_pc(void);
@@ -214,13 +234,6 @@ int hardwall_deactivate(struct task_struct *task);
 } while (0)
 #endif
 
-/* Invoke the simulator "syscall" mechanism (see arch/tile/kernel/entry.S). */
-extern int _sim_syscall(int syscall_num, ...);
-#define sim_syscall(syscall_num, ...) \
-        _sim_syscall(SIM_CONTROL_SYSCALL + \
-                ((syscall_num) << _SIM_CONTROL_OPERATOR_BITS), \
-                ## __VA_ARGS__)
-
 /*
  * Kernel threads can check to see if they need to migrate their
  * stack whenever they return from a context switch; for user
diff --git a/arch/tile/include/asm/thread_info.h b/arch/tile/include/asm/thread_info.h
index 3872f2b345d2..bc4f562bd459 100644
--- a/arch/tile/include/asm/thread_info.h
+++ b/arch/tile/include/asm/thread_info.h
@@ -68,6 +68,7 @@ struct thread_info {
 #else
 #define THREAD_SIZE_ORDER (0)
 #endif
+#define THREAD_SIZE_PAGES (1 << THREAD_SIZE_ORDER)
 
 #define THREAD_SIZE (PAGE_SIZE << THREAD_SIZE_ORDER)
 #define LOG2_THREAD_SIZE (PAGE_SHIFT + THREAD_SIZE_ORDER)
@@ -83,7 +84,7 @@ register unsigned long stack_pointer __asm__("sp");
   ((struct thread_info *)(stack_pointer & -THREAD_SIZE))
 
 #define __HAVE_ARCH_THREAD_INFO_ALLOCATOR
-extern struct thread_info *alloc_thread_info(struct task_struct *task);
+extern struct thread_info *alloc_thread_info_node(struct task_struct *task, int node);
 extern void free_thread_info(struct thread_info *info);
 
 /* Sit on a nap instruction until interrupted. */
@@ -124,6 +125,7 @@ extern void cpu_idle_on_new_stack(struct thread_info *old_ti,
 #define TIF_SYSCALL_AUDIT       5       /* syscall auditing active */
 #define TIF_SECCOMP             6       /* secure computing */
 #define TIF_MEMDIE              7       /* OOM killer at work */
+#define TIF_NOTIFY_RESUME       8       /* callback before returning to user */
 
 #define _TIF_SIGPENDING         (1<<TIF_SIGPENDING)
 #define _TIF_NEED_RESCHED       (1<<TIF_NEED_RESCHED)
@@ -133,10 +135,12 @@ extern void cpu_idle_on_new_stack(struct thread_info *old_ti,
 #define _TIF_SYSCALL_AUDIT      (1<<TIF_SYSCALL_AUDIT)
 #define _TIF_SECCOMP            (1<<TIF_SECCOMP)
 #define _TIF_MEMDIE             (1<<TIF_MEMDIE)
+#define _TIF_NOTIFY_RESUME      (1<<TIF_NOTIFY_RESUME)
 
 /* Work to do on any return to user space. */
 #define _TIF_ALLWORK_MASK \
-  (_TIF_SIGPENDING|_TIF_NEED_RESCHED|_TIF_SINGLESTEP|_TIF_ASYNC_TLB)
+  (_TIF_SIGPENDING|_TIF_NEED_RESCHED|_TIF_SINGLESTEP|\
+   _TIF_ASYNC_TLB|_TIF_NOTIFY_RESUME)
 
 /*
  * Thread-synchronous status.
diff --git a/arch/tile/include/asm/timex.h b/arch/tile/include/asm/timex.h
index 3baf5fc4c0a1..29921f0b86da 100644
--- a/arch/tile/include/asm/timex.h
+++ b/arch/tile/include/asm/timex.h
@@ -38,6 +38,9 @@ static inline cycles_t get_cycles(void)
 
 cycles_t get_clock_rate(void);
 
+/* Convert nanoseconds to core clock cycles. */
+cycles_t ns2cycles(unsigned long nsecs);
+
 /* Called at cpu initialization to set some low-level constants. */
 void setup_clock(void);
 
diff --git a/arch/tile/include/asm/topology.h b/arch/tile/include/asm/topology.h
index 343172d422a9..6fdd0c860193 100644
--- a/arch/tile/include/asm/topology.h
+++ b/arch/tile/include/asm/topology.h
@@ -44,25 +44,64 @@ static inline const struct cpumask *cpumask_of_node(int node)
 /* For now, use numa node -1 for global allocation. */
 #define pcibus_to_node(bus)             ((void)(bus), -1)
 
+/*
+ * TILE architecture has many cores integrated in one processor, so we need
+ * setup bigger balance_interval for both CPU/NODE scheduling domains to
+ * reduce process scheduling costs.
+ */
+
+/* sched_domains SD_CPU_INIT for TILE architecture */
+#define SD_CPU_INIT (struct sched_domain) {                             \
+        .min_interval           = 4,                                    \
+        .max_interval           = 128,                                  \
+        .busy_factor            = 64,                                   \
+        .imbalance_pct          = 125,                                  \
+        .cache_nice_tries       = 1,                                    \
+        .busy_idx               = 2,                                    \
+        .idle_idx               = 1,                                    \
+        .newidle_idx            = 0,                                    \
+        .wake_idx               = 0,                                    \
+        .forkexec_idx           = 0,                                    \
+                                                                        \
+        .flags                  = 1*SD_LOAD_BALANCE                     \
+                                | 1*SD_BALANCE_NEWIDLE                  \
+                                | 1*SD_BALANCE_EXEC                     \
+                                | 1*SD_BALANCE_FORK                     \
+                                | 0*SD_BALANCE_WAKE                     \
+                                | 0*SD_WAKE_AFFINE                      \
+                                | 0*SD_PREFER_LOCAL                     \
+                                | 0*SD_SHARE_CPUPOWER                   \
+                                | 0*SD_SHARE_PKG_RESOURCES              \
+                                | 0*SD_SERIALIZE                        \
+                                ,                                       \
+        .last_balance           = jiffies,                              \
+        .balance_interval       = 32,                                   \
+}
+
 /* sched_domains SD_NODE_INIT for TILE architecture */
-#define SD_NODE_INIT (struct sched_domain) {            \
-        .min_interval           = 8,                    \
-        .max_interval           = 32,                   \
-        .busy_factor            = 32,                   \
-        .imbalance_pct          = 125,                  \
-        .cache_nice_tries       = 1,                    \
-        .busy_idx               = 3,                    \
-        .idle_idx               = 1,                    \
-        .newidle_idx            = 2,                    \
-        .wake_idx               = 1,                    \
-        .flags                  = SD_LOAD_BALANCE       \
-                                | SD_BALANCE_NEWIDLE    \
-                                | SD_BALANCE_EXEC       \
-                                | SD_BALANCE_FORK       \
-                                | SD_WAKE_AFFINE        \
-                                | SD_SERIALIZE,         \
-        .last_balance           = jiffies,              \
-        .balance_interval       = 1,                    \
+#define SD_NODE_INIT (struct sched_domain) {                            \
+        .min_interval           = 16,                                   \
+        .max_interval           = 512,                                  \
+        .busy_factor            = 32,                                   \
+        .imbalance_pct          = 125,                                  \
+        .cache_nice_tries       = 1,                                    \
+        .busy_idx               = 3,                                    \
+        .idle_idx               = 1,                                    \
+        .newidle_idx            = 2,                                    \
+        .wake_idx               = 1,                                    \
+        .flags                  = 1*SD_LOAD_BALANCE                     \
+                                | 1*SD_BALANCE_NEWIDLE                  \
+                                | 1*SD_BALANCE_EXEC                     \
+                                | 1*SD_BALANCE_FORK                     \
+                                | 0*SD_BALANCE_WAKE                     \
+                                | 0*SD_WAKE_AFFINE                      \
+                                | 0*SD_PREFER_LOCAL                     \
+                                | 0*SD_SHARE_CPUPOWER                   \
+                                | 0*SD_SHARE_PKG_RESOURCES              \
+                                | 1*SD_SERIALIZE                        \
+                                ,                                       \
+        .last_balance           = jiffies,                              \
+        .balance_interval       = 128,                                  \
 }
 
 /* By definition, we create nodes based on online memory. */
diff --git a/arch/tile/include/asm/traps.h b/arch/tile/include/asm/traps.h
index 432a9c15c8a2..5f20f920f932 100644
--- a/arch/tile/include/asm/traps.h
+++ b/arch/tile/include/asm/traps.h
@@ -15,10 +15,14 @@
 #ifndef _ASM_TILE_TRAPS_H
 #define _ASM_TILE_TRAPS_H
 
+#include <arch/chip.h>
+
 /* mm/fault.c */
 void do_page_fault(struct pt_regs *, int fault_num,
                    unsigned long address, unsigned long write);
+#if CHIP_HAS_TILE_DMA() || CHIP_HAS_SN_PROC()
 void do_async_page_fault(struct pt_regs *);
+#endif
 
 #ifndef __tilegx__
 /*
@@ -59,4 +63,8 @@ void do_hardwall_trap(struct pt_regs *, int fault_num);
 void do_breakpoint(struct pt_regs *, int fault_num);
 
 
+#ifdef __tilegx__
+void gx_singlestep_handle(struct pt_regs *, int fault_num);
+#endif
+
 #endif /* _ASM_TILE_SYSCALLS_H */
diff --git a/arch/tile/include/asm/unistd.h b/arch/tile/include/asm/unistd.h
index f2e3ff485333..f70bf1c541f1 100644
--- a/arch/tile/include/asm/unistd.h
+++ b/arch/tile/include/asm/unistd.h
@@ -15,7 +15,7 @@
 #if !defined(_ASM_TILE_UNISTD_H) || defined(__SYSCALL)
 #define _ASM_TILE_UNISTD_H
 
-#ifndef __LP64__
+#if !defined(__LP64__) || defined(__SYSCALL_COMPAT)
 /* Use the flavor of this syscall that matches the 32-bit API better. */
 #define __ARCH_WANT_SYNC_FILE_RANGE2
 #endif
@@ -41,6 +41,7 @@ __SYSCALL(__NR_cmpxchg_badaddr, sys_cmpxchg_badaddr)
 #ifdef CONFIG_COMPAT
 #define __ARCH_WANT_SYS_LLSEEK
 #endif
+#define __ARCH_WANT_SYS_NEWFSTATAT
 #endif
 
 #endif /* _ASM_TILE_UNISTD_H */
diff --git a/arch/tile/include/hv/pagesize.h b/arch/tile/include/asm/vga.h
index 58bed114fedd..7b46e754d611 100644
--- a/arch/tile/include/hv/pagesize.h
+++ b/arch/tile/include/asm/vga.h
@@ -10,23 +10,30 @@
  *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
  *   NON INFRINGEMENT.  See the GNU General Public License for
  *   more details.
+ *
+ * Access to VGA videoram.
  */
 
-/**
- * @file pagesize.h
- */
+#ifndef _ASM_TILE_VGA_H
+#define _ASM_TILE_VGA_H
 
-#ifndef _HV_PAGESIZE_H
-#define _HV_PAGESIZE_H
+#include <asm/io.h>
 
-/** The log2 of the size of small pages, in bytes. This value should
- * be verified at runtime by calling hv_sysconf(HV_SYSCONF_PAGE_SIZE_SMALL).
- */
-#define HV_LOG2_PAGE_SIZE_SMALL 16
+#define VT_BUF_HAVE_RW
 
-/** The log2 of the size of large pages, in bytes. This value should be
- * verified at runtime by calling hv_sysconf(HV_SYSCONF_PAGE_SIZE_LARGE).
- */
-#define HV_LOG2_PAGE_SIZE_LARGE 24
+static inline void scr_writew(u16 val, volatile u16 *addr)
+{
+        __raw_writew(val, (volatile u16 __iomem *) addr);
+}
+
+static inline u16 scr_readw(volatile const u16 *addr)
+{
+        return __raw_readw((volatile const u16 __iomem *) addr);
+}
+
+#define vga_readb(a)    readb((u8 __iomem *)(a))
+#define vga_writeb(v,a) writeb(v, (u8 __iomem *)(a))
+
+#define VGA_MAP_MEM(x,s)        ((unsigned long) ioremap(x, s))
 
-#endif /* _HV_PAGESIZE_H */
+#endif
diff --git a/arch/tile/include/hv/drv_mshim_intf.h b/arch/tile/include/hv/drv_mshim_intf.h
new file mode 100644
index 000000000000..c6ef3bdc55cf
--- /dev/null
+++ b/arch/tile/include/hv/drv_mshim_intf.h
@@ -0,0 +1,50 @@
+/*
+ * Copyright 2011 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+/**
+ * @file drv_mshim_intf.h
+ * Interface definitions for the Linux EDAC memory controller driver.
+ */
+
+#ifndef _SYS_HV_INCLUDE_DRV_MSHIM_INTF_H
+#define _SYS_HV_INCLUDE_DRV_MSHIM_INTF_H
+
+/** Number of memory controllers in the public API. */
+#define TILE_MAX_MSHIMS 4
+
+/** Memory info under each memory controller. */
+struct mshim_mem_info
+{
+  uint64_t mem_size;     /**< Total memory size in bytes. */
+  uint8_t mem_type;      /**< Memory type, DDR2 or DDR3. */
+  uint8_t mem_ecc;       /**< Memory supports ECC. */
+};
+
+/**
+ * DIMM error structure.
+ * For now, only correctable errors are counted and the mshim doesn't record
+ * the error PA. HV takes panic upon uncorrectable errors.
+ */
+struct mshim_mem_error
+{
+  uint32_t sbe_count;     /**< Number of single-bit errors. */
+};
+
+/** Read this offset to get the memory info per mshim. */
+#define MSHIM_MEM_INFO_OFF 0x100
+
+/** Read this offset to check DIMM error. */
+#define MSHIM_MEM_ERROR_OFF 0x200
+
+#endif /* _SYS_HV_INCLUDE_DRV_MSHIM_INTF_H */
diff --git a/arch/tile/include/hv/drv_xgbe_impl.h b/arch/tile/include/hv/drv_xgbe_impl.h
new file mode 100644
index 000000000000..3a73b2b44913
--- /dev/null
+++ b/arch/tile/include/hv/drv_xgbe_impl.h
@@ -0,0 +1,300 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+/**
+ * @file drivers/xgbe/impl.h
+ * Implementation details for the NetIO library.
+ */
+
+#ifndef __DRV_XGBE_IMPL_H__
+#define __DRV_XGBE_IMPL_H__
+
+#include <hv/netio_errors.h>
+#include <hv/netio_intf.h>
+#include <hv/drv_xgbe_intf.h>
+
+
+/** How many groups we have (log2). */
+#define LOG2_NUM_GROUPS (12)
+/** How many groups we have. */
+#define NUM_GROUPS (1 << LOG2_NUM_GROUPS)
+
+/** Number of output requests we'll buffer per tile. */
+#define EPP_REQS_PER_TILE (32)
+
+/** Words used in an eDMA command without checksum acceleration. */
+#define EDMA_WDS_NO_CSUM      8
+/** Words used in an eDMA command with checksum acceleration. */
+#define EDMA_WDS_CSUM        10
+/** Total available words in the eDMA command FIFO. */
+#define EDMA_WDS_TOTAL      128
+
+
+/*
+ * FIXME: These definitions are internal and should have underscores!
+ * NOTE: The actual numeric values here are intentional and allow us to
+ * optimize the concept "if small ... else if large ... else ...", by
+ * checking for the low bit being set, and then for non-zero.
+ * These are used as array indices, so they must have the values (0, 1, 2)
+ * in some order.
+ */
+#define SIZE_SMALL (1)       /**< Small packet queue. */
+#define SIZE_LARGE (2)       /**< Large packet queue. */
+#define SIZE_JUMBO (0)       /**< Jumbo packet queue. */
+
+/** The number of "SIZE_xxx" values. */
+#define NETIO_NUM_SIZES 3
+
+
+/*
+ * Default numbers of packets for IPP drivers.  These values are chosen
+ * such that CIPP1 will not overflow its L2 cache.
+ */
+
+/** The default number of small packets. */
+#define NETIO_DEFAULT_SMALL_PACKETS 2750
+/** The default number of large packets. */
+#define NETIO_DEFAULT_LARGE_PACKETS 2500
+/** The default number of jumbo packets. */
+#define NETIO_DEFAULT_JUMBO_PACKETS 250
+
+
+/** Log2 of the size of a memory arena. */
+#define NETIO_ARENA_SHIFT      24      /* 16 MB */
+/** Size of a memory arena. */
+#define NETIO_ARENA_SIZE       (1 << NETIO_ARENA_SHIFT)
+
+
+/** A queue of packets.
+ *
+ * This structure partially defines a queue of packets waiting to be
+ * processed.  The queue as a whole is written to by an interrupt handler and
+ * read by non-interrupt code; this data structure is what's touched by the
+ * interrupt handler.  The other part of the queue state, the read offset, is
+ * kept in user space, not in hypervisor space, so it is in a separate data
+ * structure.
+ *
+ * The read offset (__packet_receive_read in the user part of the queue
+ * structure) points to the next packet to be read. When the read offset is
+ * equal to the write offset, the queue is empty; therefore the queue must
+ * contain one more slot than the required maximum queue size.
+ *
+ * Here's an example of all 3 state variables and what they mean.  All
+ * pointers move left to right.
+ *
+ * @code
+ *   I   I   V   V   V   V   I   I   I   I
+ *   0   1   2   3   4   5   6   7   8   9  10
+ *           ^       ^       ^               ^
+ *           |               |               |
+ *           |               |               __last_packet_plus_one
+ *           |               __buffer_write
+ *           __packet_receive_read
+ * @endcode
+ *
+ * This queue has 10 slots, and thus can hold 9 packets (_last_packet_plus_one
+ * = 10).  The read pointer is at 2, and the write pointer is at 6; thus,
+ * there are valid, unread packets in slots 2, 3, 4, and 5.  The remaining
+ * slots are invalid (do not contain a packet).
+ */
+typedef struct {
+  /** Byte offset of the next notify packet to be written: zero for the first
+   *  packet on the queue, sizeof (netio_pkt_t) for the second packet on the
+   *  queue, etc. */
+  volatile uint32_t __packet_write;
+
+  /** Offset of the packet after the last valid packet (i.e., when any
+   *  pointer is incremented to this value, it wraps back to zero). */
+  uint32_t __last_packet_plus_one;
+}
+__netio_packet_queue_t;
+
+
+/** A queue of buffers.
+ *
+ * This structure partially defines a queue of empty buffers which have been
+ * obtained via requests to the IPP.  (The elements of the queue are packet
+ * handles, which are transformed into a full netio_pkt_t when the buffer is
+ * retrieved.)  The queue as a whole is written to by an interrupt handler and
+ * read by non-interrupt code; this data structure is what's touched by the
+ * interrupt handler.  The other parts of the queue state, the read offset and
+ * requested write offset, are kept in user space, not in hypervisor space, so
+ * they are in a separate data structure.
+ *
+ * The read offset (__buffer_read in the user part of the queue structure)
+ * points to the next buffer to be read. When the read offset is equal to the
+ * write offset, the queue is empty; therefore the queue must contain one more
+ * slot than the required maximum queue size.
+ *
+ * The requested write offset (__buffer_requested_write in the user part of
+ * the queue structure) points to the slot which will hold the next buffer we
+ * request from the IPP, once we get around to sending such a request.  When
+ * the requested write offset is equal to the write offset, no requests for
+ * new buffers are outstanding; when the requested write offset is one greater
+ * than the read offset, no more requests may be sent.
+ *
+ * Note that, unlike the packet_queue, the buffer_queue places incoming
+ * buffers at decreasing addresses.  This makes the check for "is it time to
+ * wrap the buffer pointer" cheaper in the assembly code which receives new
+ * buffers, and means that the value which defines the queue size,
+ * __last_buffer, is different than in the packet queue.  Also, the offset
+ * used in the packet_queue is already scaled by the size of a packet; here we
+ * use unscaled slot indices for the offsets.  (These differences are
+ * historical, and in the future it's possible that the packet_queue will look
+ * more like this queue.)
+ *
+ * @code
+ * Here's an example of all 4 state variables and what they mean.  Remember:
+ * all pointers move right to left.
+ *
+ *   V   V   V   I   I   R   R   V   V   V
+ *   0   1   2   3   4   5   6   7   8   9
+ *           ^       ^       ^           ^
+ *           |       |       |           |
+ *           |       |       |           __last_buffer
+ *           |       |       __buffer_write
+ *           |       __buffer_requested_write
+ *           __buffer_read
+ * @endcode
+ *
+ * This queue has 10 slots, and thus can hold 9 buffers (_last_buffer = 9).
+ * The read pointer is at 2, and the write pointer is at 6; thus, there are
+ * valid, unread buffers in slots 2, 1, 0, 9, 8, and 7.  The requested write
+ * pointer is at 4; thus, requests have been made to the IPP for buffers which
+ * will be placed in slots 6 and 5 when they arrive.  Finally, the remaining
+ * slots are invalid (do not contain a buffer).
+ */
+typedef struct
+{
+  /** Ordinal number of the next buffer to be written: 0 for the first slot in
+   *  the queue, 1 for the second slot in the queue, etc. */
+  volatile uint32_t __buffer_write;
+
+  /** Ordinal number of the last buffer (i.e., when any pointer is decremented
+   *  below zero, it is reloaded with this value). */
+  uint32_t __last_buffer;
+}
+__netio_buffer_queue_t;
+
+
+/**
+ * An object for providing Ethernet packets to a process.
+ */
+typedef struct __netio_queue_impl_t
+{
+  /** The queue of packets waiting to be received. */
+  __netio_packet_queue_t __packet_receive_queue;
+  /** The intr bit mask that IDs this device. */
+  unsigned int __intr_id;
+  /** Offset to queues of empty buffers, one per size. */
+  uint32_t __buffer_queue[NETIO_NUM_SIZES];
+  /** The address of the first EPP tile, or -1 if no EPP. */
+  /* ISSUE: Actually this is always "0" or "~0". */
+  uint32_t __epp_location;
+  /** The queue ID that this queue represents. */
+  unsigned int __queue_id;
+  /** Number of acknowledgements received. */
+  volatile uint32_t __acks_received;
+  /** Last completion number received for packet_sendv. */
+  volatile uint32_t __last_completion_rcv;
+  /** Number of packets allowed to be outstanding. */
+  uint32_t __max_outstanding;
+  /** First VA available for packets. */
+  void* __va_0;
+  /** First VA in second range available for packets. */
+  void* __va_1;
+  /** Padding to align the "__packets" field to the size of a netio_pkt_t. */
+  uint32_t __padding[3];
+  /** The packets themselves. */
+  netio_pkt_t __packets[0];
+}
+netio_queue_impl_t;
+
+
+/**
+ * An object for managing the user end of a NetIO queue.
+ */
+typedef struct __netio_queue_user_impl_t
+{
+  /** The next incoming packet to be read. */
+  uint32_t __packet_receive_read;
+  /** The next empty buffers to be read, one index per size. */
+  uint8_t __buffer_read[NETIO_NUM_SIZES];
+  /** Where the empty buffer we next request from the IPP will go, one index
+   * per size. */
+  uint8_t __buffer_requested_write[NETIO_NUM_SIZES];
+  /** PCIe interface flag. */
+  uint8_t __pcie;
+  /** Number of packets left to be received before we send a credit update. */
+  uint32_t __receive_credit_remaining;
+  /** Value placed in __receive_credit_remaining when it reaches zero. */
+  uint32_t __receive_credit_interval;
+  /** First fast I/O routine index. */
+  uint32_t __fastio_index;
+  /** Number of acknowledgements expected. */
+  uint32_t __acks_outstanding;
+  /** Last completion number requested. */
+  uint32_t __last_completion_req;
+  /** File descriptor for driver. */
+  int __fd;
+}
+netio_queue_user_impl_t;
+
+
+#define NETIO_GROUP_CHUNK_SIZE   64   /**< Max # groups in one IPP request */
+#define NETIO_BUCKET_CHUNK_SIZE  64   /**< Max # buckets in one IPP request */
+
+
+/** Internal structure used to convey packet send information to the
+ * hypervisor.  FIXME: Actually, it's not used for that anymore, but
+ * netio_packet_send() still uses it internally.
+ */
+typedef struct
+{
+  uint16_t flags;              /**< Packet flags (__NETIO_SEND_FLG_xxx) */
+  uint16_t transfer_size;      /**< Size of packet */
+  uint32_t va;                 /**< VA of start of packet */
+  __netio_pkt_handle_t handle; /**< Packet handle */
+  uint32_t csum0;              /**< First checksum word */
+  uint32_t csum1;              /**< Second checksum word */
+}
+__netio_send_cmd_t;
+
+
+/** Flags used in two contexts:
+ *  - As the "flags" member in the __netio_send_cmd_t, above; used only
+ *    for netio_pkt_send_{prepare,commit}.
+ *  - As part of the flags passed to the various send packet fast I/O calls.
+ */
+
+/** Need acknowledgement on this packet.  Note that some code in the
+ *  normal send_pkt fast I/O handler assumes that this is equal to 1. */
+#define __NETIO_SEND_FLG_ACK    0x1
+
+/** Do checksum on this packet.  (Only used with the __netio_send_cmd_t;
+ *  normal packet sends use a special fast I/O index to denote checksumming,
+ *  and multi-segment sends test the checksum descriptor.) */
+#define __NETIO_SEND_FLG_CSUM   0x2
+
+/** Get a completion on this packet.  Only used with multi-segment sends.  */
+#define __NETIO_SEND_FLG_COMPLETION 0x4
+
+/** Position of the number-of-extra-segments value in the flags word.
+    Only used with multi-segment sends. */
+#define __NETIO_SEND_FLG_XSEG_SHIFT 3
+
+/** Width of the number-of-extra-segments value in the flags word. */
+#define __NETIO_SEND_FLG_XSEG_WIDTH 2
+
+#endif /* __DRV_XGBE_IMPL_H__ */
diff --git a/arch/tile/include/hv/drv_xgbe_intf.h b/arch/tile/include/hv/drv_xgbe_intf.h
new file mode 100644
index 000000000000..f13188ac281a
--- /dev/null
+++ b/arch/tile/include/hv/drv_xgbe_intf.h
@@ -0,0 +1,615 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+/**
+ * @file drv_xgbe_intf.h
+ * Interface to the hypervisor XGBE driver.
+ */
+
+#ifndef __DRV_XGBE_INTF_H__
+#define __DRV_XGBE_INTF_H__
+
+/**
+ * An object for forwarding VAs and PAs to the hypervisor.
+ * @ingroup types
+ *
+ * This allows the supervisor to specify a number of areas of memory to
+ * store packet buffers.
+ */
+typedef struct
+{
+  /** The physical address of the memory. */
+  HV_PhysAddr pa;
+  /** Page table entry for the memory.  This is only used to derive the
+   *  memory's caching mode; the PA bits are ignored. */
+  HV_PTE pte;
+  /** The virtual address of the memory. */
+  HV_VirtAddr va;
+  /** Size (in bytes) of the memory area. */
+  int size;
+
+}
+netio_ipp_address_t;
+
+/** The various pread/pwrite offsets into the hypervisor-level driver.
+ * @ingroup types
+ */
+typedef enum
+{
+  /** Inform the Linux driver of the address of the NetIO arena memory.
+   *  This offset is actually only used to convey information from netio
+   *  to the Linux driver; it never makes it from there to the hypervisor.
+   *  Write-only; takes a uint32_t specifying the VA address. */
+  NETIO_FIXED_ADDR               = 0x5000000000000000ULL,
+
+  /** Inform the Linux driver of the size of the NetIO arena memory.
+   *  This offset is actually only used to convey information from netio
+   *  to the Linux driver; it never makes it from there to the hypervisor.
+   *  Write-only; takes a uint32_t specifying the VA size. */
+  NETIO_FIXED_SIZE               = 0x5100000000000000ULL,
+
+  /** Register current tile with IPP.  Write then read: write, takes a
+   *  netio_input_config_t, read returns a pointer to a netio_queue_impl_t. */
+  NETIO_IPP_INPUT_REGISTER_OFF   = 0x6000000000000000ULL,
+
+  /** Unregister current tile from IPP.  Write-only, takes a dummy argument. */
+  NETIO_IPP_INPUT_UNREGISTER_OFF = 0x6100000000000000ULL,
+
+  /** Start packets flowing.  Write-only, takes a dummy argument. */
+  NETIO_IPP_INPUT_INIT_OFF       = 0x6200000000000000ULL,
+
+  /** Stop packets flowing.  Write-only, takes a dummy argument. */
+  NETIO_IPP_INPUT_UNINIT_OFF     = 0x6300000000000000ULL,
+
+  /** Configure group (typically we group on VLAN).  Write-only: takes an
+   *  array of netio_group_t's, low 24 bits of the offset is the base group
+   *  number times the size of a netio_group_t. */
+  NETIO_IPP_INPUT_GROUP_CFG_OFF  = 0x6400000000000000ULL,
+
+  /** Configure bucket.  Write-only: takes an array of netio_bucket_t's, low
+   *  24 bits of the offset is the base bucket number times the size of a
+   *  netio_bucket_t. */
+  NETIO_IPP_INPUT_BUCKET_CFG_OFF = 0x6500000000000000ULL,
+
+  /** Get/set a parameter.  Read or write: read or write data is the parameter
+   *  value, low 32 bits of the offset is a __netio_getset_offset_t. */
+  NETIO_IPP_PARAM_OFF            = 0x6600000000000000ULL,
+
+  /** Get fast I/O index.  Read-only; returns a 4-byte base index value. */
+  NETIO_IPP_GET_FASTIO_OFF       = 0x6700000000000000ULL,
+
+  /** Configure hijack IP address.  Packets with this IPv4 dest address
+   *  go to bucket NETIO_NUM_BUCKETS - 1.  Write-only: takes an IP address
+   *  in some standard form.  FIXME: Define the form! */
+  NETIO_IPP_INPUT_HIJACK_CFG_OFF  = 0x6800000000000000ULL,
+
+  /**
+   * Offsets beyond this point are reserved for the supervisor (although that
+   * enforcement must be done by the supervisor driver itself).
+   */
+  NETIO_IPP_USER_MAX_OFF         = 0x6FFFFFFFFFFFFFFFULL,
+
+  /** Register I/O memory.  Write-only, takes a netio_ipp_address_t. */
+  NETIO_IPP_IOMEM_REGISTER_OFF   = 0x7000000000000000ULL,
+
+  /** Unregister I/O memory.  Write-only, takes a netio_ipp_address_t. */
+  NETIO_IPP_IOMEM_UNREGISTER_OFF = 0x7100000000000000ULL,
+
+  /* Offsets greater than 0x7FFFFFFF can't be used directly from Linux
+   * userspace code due to limitations in the pread/pwrite syscalls. */
+
+  /** Drain LIPP buffers. */
+  NETIO_IPP_DRAIN_OFF              = 0xFA00000000000000ULL,
+
+  /** Supply a netio_ipp_address_t to be used as shared memory for the
+   *  LEPP command queue. */
+  NETIO_EPP_SHM_OFF              = 0xFB00000000000000ULL,
+
+  /* 0xFC... is currently unused. */
+
+  /** Stop IPP/EPP tiles.  Write-only, takes a dummy argument.  */
+  NETIO_IPP_STOP_SHIM_OFF        = 0xFD00000000000000ULL,
+
+  /** Start IPP/EPP tiles.  Write-only, takes a dummy argument.  */
+  NETIO_IPP_START_SHIM_OFF       = 0xFE00000000000000ULL,
+
+  /** Supply packet arena.  Write-only, takes an array of
+    * netio_ipp_address_t values. */
+  NETIO_IPP_ADDRESS_OFF          = 0xFF00000000000000ULL,
+} netio_hv_offset_t;
+
+/** Extract the base offset from an offset */
+#define NETIO_BASE_OFFSET(off)    ((off) & 0xFF00000000000000ULL)
+/** Extract the local offset from an offset */
+#define NETIO_LOCAL_OFFSET(off)   ((off) & 0x00FFFFFFFFFFFFFFULL)
+
+
+/**
+ * Get/set offset.
+ */
+typedef union
+{
+  struct
+  {
+    uint64_t addr:48;        /**< Class-specific address */
+    unsigned int class:8;    /**< Class (e.g., NETIO_PARAM) */
+    unsigned int opcode:8;   /**< High 8 bits of NETIO_IPP_PARAM_OFF */
+  }
+  bits;                      /**< Bitfields */
+  uint64_t word;             /**< Aggregated value to use as the offset */
+}
+__netio_getset_offset_t;
+
+/**
+ * Fast I/O index offsets (must be contiguous).
+ */
+typedef enum
+{
+  NETIO_FASTIO_ALLOCATE         = 0, /**< Get empty packet buffer */
+  NETIO_FASTIO_FREE_BUFFER      = 1, /**< Give buffer back to IPP */
+  NETIO_FASTIO_RETURN_CREDITS   = 2, /**< Give credits to IPP */
+  NETIO_FASTIO_SEND_PKT_NOCK    = 3, /**< Send a packet, no checksum */
+  NETIO_FASTIO_SEND_PKT_CK      = 4, /**< Send a packet, with checksum */
+  NETIO_FASTIO_SEND_PKT_VEC     = 5, /**< Send a vector of packets */
+  NETIO_FASTIO_SENDV_PKT        = 6, /**< Sendv one packet */
+  NETIO_FASTIO_NUM_INDEX        = 7, /**< Total number of fast I/O indices */
+} netio_fastio_index_t;
+
+/** 3-word return type for Fast I/O call. */
+typedef struct
+{
+  int err;            /**< Error code. */
+  uint32_t val0;      /**< Value.  Meaning depends upon the specific call. */
+  uint32_t val1;      /**< Value.  Meaning depends upon the specific call. */
+} netio_fastio_rv3_t;
+
+/** 0-argument fast I/O call */
+int __netio_fastio0(uint32_t fastio_index);
+/** 1-argument fast I/O call */
+int __netio_fastio1(uint32_t fastio_index, uint32_t arg0);
+/** 3-argument fast I/O call, 2-word return value */
+netio_fastio_rv3_t __netio_fastio3_rv3(uint32_t fastio_index, uint32_t arg0,
+                                       uint32_t arg1, uint32_t arg2);
+/** 4-argument fast I/O call */
+int __netio_fastio4(uint32_t fastio_index, uint32_t arg0, uint32_t arg1,
+                    uint32_t arg2, uint32_t arg3);
+/** 6-argument fast I/O call */
+int __netio_fastio6(uint32_t fastio_index, uint32_t arg0, uint32_t arg1,
+                    uint32_t arg2, uint32_t arg3, uint32_t arg4, uint32_t arg5);
+/** 9-argument fast I/O call */
+int __netio_fastio9(uint32_t fastio_index, uint32_t arg0, uint32_t arg1,
+                    uint32_t arg2, uint32_t arg3, uint32_t arg4, uint32_t arg5,
+                    uint32_t arg6, uint32_t arg7, uint32_t arg8);
+
+/** Allocate an empty packet.
+ * @param fastio_index Fast I/O index.
+ * @param size Size of the packet to allocate.
+ */
+#define __netio_fastio_allocate(fastio_index, size) \
+  __netio_fastio1((fastio_index) + NETIO_FASTIO_ALLOCATE, size)
+
+/** Free a buffer.
+ * @param fastio_index Fast I/O index.
+ * @param handle Handle for the packet to free.
+ */
+#define __netio_fastio_free_buffer(fastio_index, handle) \
+  __netio_fastio1((fastio_index) + NETIO_FASTIO_FREE_BUFFER, handle)
+
+/** Increment our receive credits.
+ * @param fastio_index Fast I/O index.
+ * @param credits Number of credits to add.
+ */
+#define __netio_fastio_return_credits(fastio_index, credits) \
+  __netio_fastio1((fastio_index) + NETIO_FASTIO_RETURN_CREDITS, credits)
+
+/** Send packet, no checksum.
+ * @param fastio_index Fast I/O index.
+ * @param ackflag Nonzero if we want an ack.
+ * @param size Size of the packet.
+ * @param va Virtual address of start of packet.
+ * @param handle Packet handle.
+ */
+#define __netio_fastio_send_pkt_nock(fastio_index, ackflag, size, va, handle) \
+  __netio_fastio4((fastio_index) + NETIO_FASTIO_SEND_PKT_NOCK, ackflag, \
+                  size, va, handle)
+
+/** Send packet, calculate checksum.
+ * @param fastio_index Fast I/O index.
+ * @param ackflag Nonzero if we want an ack.
+ * @param size Size of the packet.
+ * @param va Virtual address of start of packet.
+ * @param handle Packet handle.
+ * @param csum0 Shim checksum header.
+ * @param csum1 Checksum seed.
+ */
+#define __netio_fastio_send_pkt_ck(fastio_index, ackflag, size, va, handle, \
+                                   csum0, csum1) \
+  __netio_fastio6((fastio_index) + NETIO_FASTIO_SEND_PKT_CK, ackflag, \
+                  size, va, handle, csum0, csum1)
+
+
+/** Format for the "csum0" argument to the __netio_fastio_send routines
+ * and LEPP.  Note that this is currently exactly identical to the
+ * ShimProtocolOffloadHeader.
+ */
+typedef union
+{
+  struct
+  {
+    unsigned int start_byte:7;       /**< The first byte to be checksummed */
+    unsigned int count:14;           /**< Number of bytes to be checksummed. */
+    unsigned int destination_byte:7; /**< The byte to write the checksum to. */
+    unsigned int reserved:4;         /**< Reserved. */
+  } bits;                            /**< Decomposed method of access. */
+  unsigned int word;                 /**< To send out the IDN. */
+} __netio_checksum_header_t;
+
+
+/** Sendv packet with 1 or 2 segments.
+ * @param fastio_index Fast I/O index.
+ * @param flags Ack/csum/notify flags in low 3 bits; number of segments minus
+ *        1 in next 2 bits; expected checksum in high 16 bits.
+ * @param confno Confirmation number to request, if notify flag set.
+ * @param csum0 Checksum descriptor; if zero, no checksum.
+ * @param va_F Virtual address of first segment.
+ * @param va_L Virtual address of last segment, if 2 segments.
+ * @param len_F_L Length of first segment in low 16 bits; length of last
+ *        segment, if 2 segments, in high 16 bits.
+ */
+#define __netio_fastio_sendv_pkt_1_2(fastio_index, flags, confno, csum0, \
+                                     va_F, va_L, len_F_L) \
+  __netio_fastio6((fastio_index) + NETIO_FASTIO_SENDV_PKT, flags, confno, \
+                  csum0, va_F, va_L, len_F_L)
+
+/** Send packet on PCIe interface.
+ * @param fastio_index Fast I/O index.
+ * @param flags Ack/csum/notify flags in low 3 bits.
+ * @param confno Confirmation number to request, if notify flag set.
+ * @param csum0 Checksum descriptor; Hard wired 0, not needed for PCIe.
+ * @param va_F Virtual address of the packet buffer.
+ * @param va_L Virtual address of last segment, if 2 segments. Hard wired 0.
+ * @param len_F_L Length of the packet buffer in low 16 bits.
+ */
+#define __netio_fastio_send_pcie_pkt(fastio_index, flags, confno, csum0, \
+                                     va_F, va_L, len_F_L) \
+  __netio_fastio6((fastio_index) + PCIE_FASTIO_SENDV_PKT, flags, confno, \
+                  csum0, va_F, va_L, len_F_L)
+
+/** Sendv packet with 3 or 4 segments.
+ * @param fastio_index Fast I/O index.
+ * @param flags Ack/csum/notify flags in low 3 bits; number of segments minus
+ *        1 in next 2 bits; expected checksum in high 16 bits.
+ * @param confno Confirmation number to request, if notify flag set.
+ * @param csum0 Checksum descriptor; if zero, no checksum.
+ * @param va_F Virtual address of first segment.
+ * @param va_L Virtual address of last segment (third segment if 3 segments,
+ *        fourth segment if 4 segments).
+ * @param len_F_L Length of first segment in low 16 bits; length of last
+ *        segment in high 16 bits.
+ * @param va_M0 Virtual address of "middle 0" segment; this segment is sent
+ *        second when there are three segments, and third if there are four.
+ * @param va_M1 Virtual address of "middle 1" segment; this segment is sent
+ *        second when there are four segments.
+ * @param len_M0_M1 Length of middle 0 segment in low 16 bits; length of middle
+ *        1 segment, if 4 segments, in high 16 bits.
+ */
+#define __netio_fastio_sendv_pkt_3_4(fastio_index, flags, confno, csum0, va_F, \
+                                     va_L, len_F_L, va_M0, va_M1, len_M0_M1) \
+  __netio_fastio9((fastio_index) + NETIO_FASTIO_SENDV_PKT, flags, confno, \
+                  csum0, va_F, va_L, len_F_L, va_M0, va_M1, len_M0_M1)
+
+/** Send vector of packets.
+ * @param fastio_index Fast I/O index.
+ * @param seqno Number of packets transmitted so far on this interface;
+ *        used to decide which packets should be acknowledged.
+ * @param nentries Number of entries in vector.
+ * @param va Virtual address of start of vector entry array.
+ * @return 3-word netio_fastio_rv3_t structure.  The structure's err member
+ *         is an error code, or zero if no error.  The val0 member is the
+ *         updated value of seqno; it has been incremented by 1 for each
+ *         packet sent.  That increment may be less than nentries if an
+ *         error occurred, or if some of the entries in the vector contain
+ *         handles equal to NETIO_PKT_HANDLE_NONE.  The val1 member is the
+ *         updated value of nentries; it has been decremented by 1 for each
+ *         vector entry processed.  Again, that decrement may be less than
+ *         nentries (leaving the returned value positive) if an error
+ *         occurred.
+ */
+#define __netio_fastio_send_pkt_vec(fastio_index, seqno, nentries, va) \
+  __netio_fastio3_rv3((fastio_index) + NETIO_FASTIO_SEND_PKT_VEC, seqno, \
+                      nentries, va)
+
+
+/** An egress DMA command for LEPP. */
+typedef struct
+{
+  /** Is this a TSO transfer?
+   *
+   * NOTE: This field is always 0, to distinguish it from
+   * lepp_tso_cmd_t.  It must come first!
+   */
+  uint8_t tso               : 1;
+
+  /** Unused padding bits. */
+  uint8_t _unused           : 3;
+
+  /** Should this packet be sent directly from caches instead of DRAM,
+   * using hash-for-home to locate the packet data?
+   */
+  uint8_t hash_for_home     : 1;
+
+  /** Should we compute a checksum? */
+  uint8_t compute_checksum  : 1;
+
+  /** Is this the final buffer for this packet?
+   *
+   * A single packet can be split over several input buffers (a "gather"
+   * operation).  This flag indicates that this is the last buffer
+   * in a packet.
+   */
+  uint8_t end_of_packet     : 1;
+
+  /** Should LEPP advance 'comp_busy' when this DMA is fully finished? */
+  uint8_t send_completion   : 1;
+
+  /** High bits of Client Physical Address of the start of the buffer
+   *  to be egressed.
+   *
+   *  NOTE: Only 6 bits are actually needed here, as CPAs are
+   *  currently 38 bits.  So two bits could be scavenged from this.
+   */
+  uint8_t cpa_hi;
+
+  /** The number of bytes to be egressed. */
+  uint16_t length;
+
+  /** Low 32 bits of Client Physical Address of the start of the buffer
+   *  to be egressed.
+   */
+  uint32_t cpa_lo;
+
+  /** Checksum information (only used if 'compute_checksum'). */
+  __netio_checksum_header_t checksum_data;
+
+} lepp_cmd_t;
+
+
+/** A chunk of physical memory for a TSO egress. */
+typedef struct
+{
+  /** The low bits of the CPA. */
+  uint32_t cpa_lo;
+  /** The high bits of the CPA. */
+  uint16_t cpa_hi               : 15;
+  /** Should this packet be sent directly from caches instead of DRAM,
+   *  using hash-for-home to locate the packet data?
+   */
+  uint16_t hash_for_home        : 1;
+  /** The length in bytes. */
+  uint16_t length;
+} lepp_frag_t;
+
+
+/** An LEPP command that handles TSO. */
+typedef struct
+{
+  /** Is this a TSO transfer?
+   *
+   *  NOTE: This field is always 1, to distinguish it from
+   *  lepp_cmd_t.  It must come first!
+   */
+  uint8_t tso             : 1;
+
+  /** Unused padding bits. */
+  uint8_t _unused         : 7;
+
+  /** Size of the header[] array in bytes.  It must be in the range
+   *  [40, 127], which are the smallest header for a TCP packet over
+   *  Ethernet and the maximum possible prepend size supported by
+   *  hardware, respectively.  Note that the array storage must be
+   *  padded out to a multiple of four bytes so that the following
+   *  LEPP command is aligned properly.
+   */
+  uint8_t header_size;
+
+  /** Byte offset of the IP header in header[]. */
+  uint8_t ip_offset;
+
+  /** Byte offset of the TCP header in header[]. */
+  uint8_t tcp_offset;
+
+  /** The number of bytes to use for the payload of each packet,
+   *  except of course the last one, which may not have enough bytes.
+   *  This means that each Ethernet packet except the last will have a
+   *  size of header_size + payload_size.
+   */
+  uint16_t payload_size;
+
+  /** The length of the 'frags' array that follows this struct. */
+  uint16_t num_frags;
+
+  /** The actual frags. */
+  lepp_frag_t frags[0 /* Variable-sized; num_frags entries. */];
+
+  /*
+   * The packet header template logically follows frags[],
+   * but you can't declare that in C.
+   *
+   * uint32_t header[header_size_in_words_rounded_up];
+   */
+
+} lepp_tso_cmd_t;
+
+
+/** An LEPP completion ring entry. */
+typedef void* lepp_comp_t;
+
+
+/** Maximum number of frags for one TSO command.  This is adapted from
+ *  linux's "MAX_SKB_FRAGS", and presumably over-estimates by one, for
+ *  our page size of exactly 65536.  We add one for a "body" fragment.
+ */
+#define LEPP_MAX_FRAGS (65536 / HV_PAGE_SIZE_SMALL + 2 + 1)
+
+/** Total number of bytes needed for an lepp_tso_cmd_t. */
+#define LEPP_TSO_CMD_SIZE(num_frags, header_size) \
+  (sizeof(lepp_tso_cmd_t) + \
+   (num_frags) * sizeof(lepp_frag_t) + \
+   (((header_size) + 3) & -4))
+
+/** The size of the lepp "cmd" queue. */
+#define LEPP_CMD_QUEUE_BYTES \
+ (((CHIP_L2_CACHE_SIZE() - 2 * CHIP_L2_LINE_SIZE()) / \
+  (sizeof(lepp_cmd_t) + sizeof(lepp_comp_t))) * sizeof(lepp_cmd_t))
+
+/** The largest possible command that can go in lepp_queue_t::cmds[]. */
+#define LEPP_MAX_CMD_SIZE LEPP_TSO_CMD_SIZE(LEPP_MAX_FRAGS, 128)
+
+/** The largest possible value of lepp_queue_t::cmd_{head, tail} (inclusive).
+ */
+#define LEPP_CMD_LIMIT \
+  (LEPP_CMD_QUEUE_BYTES - LEPP_MAX_CMD_SIZE)
+
+/** The maximum number of completions in an LEPP queue. */
+#define LEPP_COMP_QUEUE_SIZE \
+  ((LEPP_CMD_LIMIT + sizeof(lepp_cmd_t) - 1) / sizeof(lepp_cmd_t))
+
+/** Increment an index modulo the queue size. */
+#define LEPP_QINC(var) \
+  (var = __insn_mnz(var - (LEPP_COMP_QUEUE_SIZE - 1), var + 1))
+
+/** A queue used to convey egress commands from the client to LEPP. */
+typedef struct
+{
+  /** Index of first completion not yet processed by user code.
+   *  If this is equal to comp_busy, there are no such completions.
+   *
+   *  NOTE: This is only read/written by the user.
+   */
+  unsigned int comp_head;
+
+  /** Index of first completion record not yet completed.
+   *  If this is equal to comp_tail, there are no such completions.
+   *  This index gets advanced (modulo LEPP_QUEUE_SIZE) whenever
+   *  a command with the 'completion' bit set is finished.
+   *
+   *  NOTE: This is only written by LEPP, only read by the user.
+   */
+  volatile unsigned int comp_busy;
+
+  /** Index of the first empty slot in the completion ring.
+   *  Entries from this up to but not including comp_head (in ring order)
+   *  can be filled in with completion data.
+   *
+   *  NOTE: This is only read/written by the user.
+   */
+  unsigned int comp_tail;
+
+  /** Byte index of first command enqueued for LEPP but not yet processed.
+   *
+   *  This is always divisible by sizeof(void*) and always <= LEPP_CMD_LIMIT.
+   *
+   *  NOTE: LEPP advances this counter as soon as it no longer needs
+   *  the cmds[] storage for this entry, but the transfer is not actually
+   *  complete (i.e. the buffer pointed to by the command is no longer
+   *  needed) until comp_busy advances.
+   *
+   *  If this is equal to cmd_tail, the ring is empty.
+   *
+   *  NOTE: This is only written by LEPP, only read by the user.
+   */
+  volatile unsigned int cmd_head;
+
+  /** Byte index of first empty slot in the command ring.  This field can
+   *  be incremented up to but not equal to cmd_head (because that would
+   *  mean the ring is empty).
+   *
+   *  This is always divisible by sizeof(void*) and always <= LEPP_CMD_LIMIT.
+   *
+   *  NOTE: This is read/written by the user, only read by LEPP.
+   */
+  volatile unsigned int cmd_tail;
+
+  /** A ring of variable-sized egress DMA commands.
+   *
+   *  NOTE: Only written by the user, only read by LEPP.
+   */
+  char cmds[LEPP_CMD_QUEUE_BYTES]
+    __attribute__((aligned(CHIP_L2_LINE_SIZE())));
+
+  /** A ring of user completion data.
+   *  NOTE: Only read/written by the user.
+   */
+  lepp_comp_t comps[LEPP_COMP_QUEUE_SIZE]
+    __attribute__((aligned(CHIP_L2_LINE_SIZE())));
+} lepp_queue_t;
+
+
+/** An internal helper function for determining the number of entries
+ *  available in a ring buffer, given that there is one sentinel.
+ */
+static inline unsigned int
+_lepp_num_free_slots(unsigned int head, unsigned int tail)
+{
+  /*
+   * One entry is reserved for use as a sentinel, to distinguish
+   * "empty" from "full".  So we compute
+   * (head - tail - 1) % LEPP_QUEUE_SIZE, but without using a slow % operation.
+   */
+  return (head - tail - 1) + ((head <= tail) ? LEPP_COMP_QUEUE_SIZE : 0);
+}
+
+
+/** Returns how many new comp entries can be enqueued. */
+static inline unsigned int
+lepp_num_free_comp_slots(const lepp_queue_t* q)
+{
+  return _lepp_num_free_slots(q->comp_head, q->comp_tail);
+}
+
+static inline int
+lepp_qsub(int v1, int v2)
+{
+  int delta = v1 - v2;
+  return delta + ((delta >> 31) & LEPP_COMP_QUEUE_SIZE);
+}
+
+
+/** FIXME: Check this from linux, via a new "pwrite()" call. */
+#define LIPP_VERSION 1
+
+
+/** We use exactly two bytes of alignment padding. */
+#define LIPP_PACKET_PADDING 2
+
+/** The minimum size of a "small" buffer (including the padding). */
+#define LIPP_SMALL_PACKET_SIZE 128
+
+/*
+ * NOTE: The following two values should total to less than around
+ * 13582, to keep the total size used for "lipp_state_t" below 64K.
+ */
+
+/** The maximum number of "small" buffers.
+ *  This is enough for 53 network cpus with 128 credits.  Note that
+ *  if these are exhausted, we will fall back to using large buffers.
+ */
+#define LIPP_SMALL_BUFFERS 6785
+
+/** The maximum number of "large" buffers.
+ *  This is enough for 53 network cpus with 128 credits.
+ */
+#define LIPP_LARGE_BUFFERS 6785
+
+#endif /* __DRV_XGBE_INTF_H__ */
diff --git a/arch/tile/include/hv/hypervisor.h b/arch/tile/include/hv/hypervisor.h
index 9bd303a141b2..72ec1e972f15 100644
--- a/arch/tile/include/hv/hypervisor.h
+++ b/arch/tile/include/hv/hypervisor.h
@@ -22,8 +22,6 @@
 
 #include <arch/chip.h>
 
-#include <hv/pagesize.h>
-
 /* Linux builds want unsigned long constants, but assembler wants numbers */
 #ifdef __ASSEMBLER__
 /** One, for assembler */
@@ -44,11 +42,21 @@
  */
 #define HV_L1_SPAN (__HV_SIZE_ONE << HV_LOG2_L1_SPAN)
 
+/** The log2 of the size of small pages, in bytes. This value should
+ * be verified at runtime by calling hv_sysconf(HV_SYSCONF_PAGE_SIZE_SMALL).
+ */
+#define HV_LOG2_PAGE_SIZE_SMALL 16
+
 /** The size of small pages, in bytes. This value should be verified
  * at runtime by calling hv_sysconf(HV_SYSCONF_PAGE_SIZE_SMALL).
  */
 #define HV_PAGE_SIZE_SMALL (__HV_SIZE_ONE << HV_LOG2_PAGE_SIZE_SMALL)
 
+/** The log2 of the size of large pages, in bytes. This value should be
+ * verified at runtime by calling hv_sysconf(HV_SYSCONF_PAGE_SIZE_LARGE).
+ */
+#define HV_LOG2_PAGE_SIZE_LARGE 24
+
 /** The size of large pages, in bytes. This value should be verified
  * at runtime by calling hv_sysconf(HV_SYSCONF_PAGE_SIZE_LARGE).
  */
@@ -338,9 +346,10 @@ typedef int HV_Errno;
 #define HV_ENOTREADY   -812  /**< Device not ready */
 #define HV_EIO         -813  /**< I/O error */
 #define HV_ENOMEM      -814  /**< Out of memory */
+#define HV_EAGAIN      -815  /**< Try again */
 
 #define HV_ERR_MAX     -801  /**< Largest HV error code */
-#define HV_ERR_MIN     -814  /**< Smallest HV error code */
+#define HV_ERR_MIN     -815  /**< Smallest HV error code */
 
 #ifndef __ASSEMBLER__
 
@@ -867,6 +876,43 @@ typedef struct
  */
 HV_PhysAddrRange hv_inquire_physical(int idx);
 
+/** Possible DIMM types. */
+typedef enum
+{
+  NO_DIMM                    = 0,  /**< No DIMM */
+  DDR2                       = 1,  /**< DDR2 */
+  DDR3                       = 2   /**< DDR3 */
+} HV_DIMM_Type;
+
+#ifdef __tilegx__
+
+/** Log2 of minimum DIMM bytes supported by the memory controller. */
+#define HV_MSH_MIN_DIMM_SIZE_SHIFT 29
+
+/** Max number of DIMMs contained by one memory controller. */
+#define HV_MSH_MAX_DIMMS 8
+
+#else
+
+/** Log2 of minimum DIMM bytes supported by the memory controller. */
+#define HV_MSH_MIN_DIMM_SIZE_SHIFT 26
+
+/** Max number of DIMMs contained by one memory controller. */
+#define HV_MSH_MAX_DIMMS 2
+
+#endif
+
+/** Number of bits to right-shift to get the DIMM type. */
+#define HV_DIMM_TYPE_SHIFT 0
+
+/** Bits to mask to get the DIMM type. */
+#define HV_DIMM_TYPE_MASK 0xf
+
+/** Number of bits to right-shift to get the DIMM size. */
+#define HV_DIMM_SIZE_SHIFT 4
+
+/** Bits to mask to get the DIMM size. */
+#define HV_DIMM_SIZE_MASK 0xf
 
 /** Memory controller information. */
 typedef struct
@@ -964,6 +1010,11 @@ HV_ASIDRange hv_inquire_asid(int idx);
 
 /** Waits for at least the specified number of nanoseconds then returns.
  *
+ * NOTE: this deprecated function currently assumes a 750 MHz clock,
+ * and is thus not generally suitable for use.  New code should call
+ * hv_sysconf(HV_SYSCONF_CPU_SPEED), compute a cycle count to wait for,
+ * and delay by looping while checking the cycle counter SPR.
+ *
  * @param nanosecs The number of nanoseconds to sleep.
  */
 void hv_nanosleep(int nanosecs);
@@ -1003,41 +1054,42 @@ int hv_console_write(HV_VirtAddr bytes, int len);
  *  when these occur in a client's interrupt critical section, they must
  *  be delivered through the downcall mechanism.
  *
- *  A downcall is initially delivered to the client as an INTCTRL_1
- *  interrupt.  Upon entry to the INTCTRL_1 vector, the client must
- *  immediately invoke the hv_downcall_dispatch service.  This service
- *  will not return; instead it will cause one of the client's actual
- *  downcall-handling interrupt vectors to be entered.  The EX_CONTEXT
- *  registers in the client will be set so that when the client irets,
- *  it will return to the code which was interrupted by the INTCTRL_1
- *  interrupt.
+ *  A downcall is initially delivered to the client as an INTCTRL_CL
+ *  interrupt, where CL is the client's PL.  Upon entry to the INTCTRL_CL
+ *  vector, the client must immediately invoke the hv_downcall_dispatch
+ *  service.  This service will not return; instead it will cause one of
+ *  the client's actual downcall-handling interrupt vectors to be entered.
+ *  The EX_CONTEXT registers in the client will be set so that when the
+ *  client irets, it will return to the code which was interrupted by the
+ *  INTCTRL_CL interrupt.
  *
- *  Under some circumstances, the firing of INTCTRL_1 can race with
+ *  Under some circumstances, the firing of INTCTRL_CL can race with
  *  the lowering of a device interrupt.  In such a case, the
  *  hv_downcall_dispatch service may issue an iret instruction instead
  *  of entering one of the client's actual downcall-handling interrupt
  *  vectors.  This will return execution to the location that was
- *  interrupted by INTCTRL_1.
+ *  interrupted by INTCTRL_CL.
  *
  *  Any saving of registers should be done by the actual handling
- *  vectors; no registers should be changed by the INTCTRL_1 handler.
+ *  vectors; no registers should be changed by the INTCTRL_CL handler.
  *  In particular, the client should not use a jal instruction to invoke
  *  the hv_downcall_dispatch service, as that would overwrite the client's
  *  lr register.  Note that the hv_downcall_dispatch service may overwrite
  *  one or more of the client's system save registers.
  *
- *  The client must not modify the INTCTRL_1_STATUS SPR.  The hypervisor
+ *  The client must not modify the INTCTRL_CL_STATUS SPR.  The hypervisor
  *  will set this register to cause a downcall to happen, and will clear
  *  it when no further downcalls are pending.
  *
- *  When a downcall vector is entered, the INTCTRL_1 interrupt will be
+ *  When a downcall vector is entered, the INTCTRL_CL interrupt will be
  *  masked.  When the client is done processing a downcall, and is ready
  *  to accept another, it must unmask this interrupt; if more downcalls
- *  are pending, this will cause the INTCTRL_1 vector to be reentered.
+ *  are pending, this will cause the INTCTRL_CL vector to be reentered.
  *  Currently the following interrupt vectors can be entered through a
  *  downcall:
  *
  *  INT_MESSAGE_RCV_DWNCL   (hypervisor message available)
+ *  INT_DEV_INTR_DWNCL      (device interrupt)
  *  INT_DMATLB_MISS_DWNCL   (DMA TLB miss)
  *  INT_SNITLB_MISS_DWNCL   (SNI TLB miss)
  *  INT_DMATLB_ACCESS_DWNCL (DMA TLB access violation)
@@ -1296,7 +1348,7 @@ typedef struct
  *  this operation.  If any permanent delivery errors were encountered,
  *  the routine returns HV_ERECIP.  In the event of permanent delivery
  *  errors, it may be the case that delivery was not attempted to all
- *  recipients; if any messages were succesfully delivered, however,
+ *  recipients; if any messages were successfully delivered, however,
  *  recipients' state values will be updated appropriately.
  *
  *  It is explicitly legal to specify a recipient structure whose state
@@ -1315,7 +1367,7 @@ typedef struct
  *  never call hv_receive_message, or could register a different state
  *  buffer, losing the message.
  *
- *  Specifiying the same recipient more than once in the recipient list
+ *  Specifying the same recipient more than once in the recipient list
  *  is an error, which will not result in an error return but which may
  *  or may not result in more than one message being delivered to the
  *  recipient tile.
diff --git a/arch/tile/include/hv/netio_errors.h b/arch/tile/include/hv/netio_errors.h
new file mode 100644
index 000000000000..e1591bff61b5
--- /dev/null
+++ b/arch/tile/include/hv/netio_errors.h
@@ -0,0 +1,122 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+/**
+ * Error codes returned from NetIO routines.
+ */
+
+#ifndef __NETIO_ERRORS_H__
+#define __NETIO_ERRORS_H__
+
+/**
+ * @addtogroup error
+ *
+ * @brief The error codes returned by NetIO functions.
+ *
+ * NetIO functions return 0 (defined as ::NETIO_NO_ERROR) on success, and
+ * a negative value if an error occurs.
+ *
+ * In cases where a NetIO function failed due to a error reported by
+ * system libraries, the error code will be the negation of the
+ * system errno at the time of failure.  The @ref netio_strerror()
+ * function will deliver error strings for both NetIO and system error
+ * codes.
+ *
+ * @{
+ */
+
+/** The set of all NetIO errors. */
+typedef enum
+{
+  /** Operation successfully completed. */
+  NETIO_NO_ERROR        = 0,
+
+  /** A packet was successfully retrieved from an input queue. */
+  NETIO_PKT             = 0,
+
+  /** Largest NetIO error number. */
+  NETIO_ERR_MAX         = -701,
+
+  /** The tile is not registered with the IPP. */
+  NETIO_NOT_REGISTERED  = -701,
+
+  /** No packet was available to retrieve from the input queue. */
+  NETIO_NOPKT           = -702,
+
+  /** The requested function is not implemented. */
+  NETIO_NOT_IMPLEMENTED = -703,
+
+  /** On a registration operation, the target queue already has the maximum
+   *  number of tiles registered for it, and no more may be added.  On a
+   *  packet send operation, the output queue is full and nothing more can
+   *  be queued until some of the queued packets are actually transmitted. */
+  NETIO_QUEUE_FULL      = -704,
+
+  /** The calling process or thread is not bound to exactly one CPU. */
+  NETIO_BAD_AFFINITY    = -705,
+
+  /** Cannot allocate memory on requested controllers. */
+  NETIO_CANNOT_HOME     = -706,
+
+  /** On a registration operation, the IPP specified is not configured
+   *  to support the options requested; for instance, the application
+   *  wants a specific type of tagged headers which the configured IPP
+   *  doesn't support.  Or, the supplied configuration information is
+   *  not self-consistent, or is out of range; for instance, specifying
+   *  both NETIO_RECV and NETIO_NO_RECV, or asking for more than
+   *  NETIO_MAX_SEND_BUFFERS to be preallocated.  On a VLAN or bucket
+   *  configure operation, the number of items, or the base item, was
+   *  out of range.
+   */
+  NETIO_BAD_CONFIG      = -707,
+
+  /** Too many tiles have registered to transmit packets. */
+  NETIO_TOOMANY_XMIT    = -708,
+
+  /** Packet transmission was attempted on a queue which was registered
+      with transmit disabled. */
+  NETIO_UNREG_XMIT      = -709,
+
+  /** This tile is already registered with the IPP. */
+  NETIO_ALREADY_REGISTERED = -710,
+
+  /** The Ethernet link is down. The application should try again later. */
+  NETIO_LINK_DOWN       = -711,
+
+  /** An invalid memory buffer has been specified.  This may be an unmapped
+   * virtual address, or one which does not meet alignment requirements.
+   * For netio_input_register(), this error may be returned when multiple
+   * processes specify different memory regions to be used for NetIO
+   * buffers.  That can happen if these processes specify explicit memory
+   * regions with the ::NETIO_FIXED_BUFFER_VA flag, or if tmc_cmem_init()
+   * has not been called by a common ancestor of the processes.
+   */
+  NETIO_FAULT           = -712,
+
+  /** Cannot combine user-managed shared memory and cache coherence. */
+  NETIO_BAD_CACHE_CONFIG = -713,
+
+  /** Smallest NetIO error number. */
+  NETIO_ERR_MIN         = -713,
+
+#ifndef __DOXYGEN__
+  /** Used internally to mean that no response is needed; never returned to
+   *  an application. */
+  NETIO_NO_RESPONSE     = 1
+#endif
+} netio_error_t;
+
+/** @} */
+
+#endif /* __NETIO_ERRORS_H__ */
diff --git a/arch/tile/include/hv/netio_intf.h b/arch/tile/include/hv/netio_intf.h
new file mode 100644
index 000000000000..8d20972aba2c
--- /dev/null
+++ b/arch/tile/include/hv/netio_intf.h
@@ -0,0 +1,2975 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+/**
+ * NetIO interface structures and macros.
+ */
+
+#ifndef __NETIO_INTF_H__
+#define __NETIO_INTF_H__
+
+#include <hv/netio_errors.h>
+
+#ifdef __KERNEL__
+#include <linux/types.h>
+#else
+#include <stdint.h>
+#endif
+
+#if !defined(__HV__) && !defined(__BOGUX__) && !defined(__KERNEL__)
+#include <assert.h>
+#define netio_assert assert  /**< Enable assertions from macros */
+#else
+#define netio_assert(...) ((void)(0))  /**< Disable assertions from macros */
+#endif
+
+/*
+ * If none of these symbols are defined, we're building libnetio in an
+ * environment where we have pthreads, so we'll enable locking.
+ */
+#if !defined(__HV__) && !defined(__BOGUX__) && !defined(__KERNEL__) && \
+    !defined(__NEWLIB__)
+#define _NETIO_PTHREAD       /**< Include a mutex in netio_queue_t below */
+
+/*
+ * If NETIO_UNLOCKED is defined, we don't do use per-cpu locks on
+ * per-packet NetIO operations.  We still do pthread locking on things
+ * like netio_input_register, though.  This is used for building
+ * libnetio_unlocked.
+ */
+#ifndef NETIO_UNLOCKED
+
+/* Avoid PLT overhead by using our own inlined per-cpu lock. */
+#include <sched.h>
+typedef int _netio_percpu_mutex_t;
+
+static __inline int
+_netio_percpu_mutex_init(_netio_percpu_mutex_t* lock)
+{
+  *lock = 0;
+  return 0;
+}
+
+static __inline int
+_netio_percpu_mutex_lock(_netio_percpu_mutex_t* lock)
+{
+  while (__builtin_expect(__insn_tns(lock), 0))
+    sched_yield();
+  return 0;
+}
+
+static __inline int
+_netio_percpu_mutex_unlock(_netio_percpu_mutex_t* lock)
+{
+  *lock = 0;
+  return 0;
+}
+
+#else /* NETIO_UNLOCKED */
+
+/* Don't do any locking for per-packet NetIO operations. */
+typedef int _netio_percpu_mutex_t;
+#define _netio_percpu_mutex_init(L)
+#define _netio_percpu_mutex_lock(L)
+#define _netio_percpu_mutex_unlock(L)
+
+#endif /* NETIO_UNLOCKED */
+#endif /* !__HV__, !__BOGUX, !__KERNEL__, !__NEWLIB__ */
+
+/** How many tiles can register for a given queue.
+ *  @ingroup setup */
+#define NETIO_MAX_TILES_PER_QUEUE  64
+
+
+/** Largest permissible queue identifier.
+ *  @ingroup setup  */
+#define NETIO_MAX_QUEUE_ID        255
+
+
+#ifndef __DOXYGEN__
+
+/* Metadata packet checksum/ethertype flags. */
+
+/** The L4 checksum has not been calculated. */
+#define _NETIO_PKT_NO_L4_CSUM_SHIFT           0
+#define _NETIO_PKT_NO_L4_CSUM_RMASK           1
+#define _NETIO_PKT_NO_L4_CSUM_MASK \
+         (_NETIO_PKT_NO_L4_CSUM_RMASK << _NETIO_PKT_NO_L4_CSUM_SHIFT)
+
+/** The L3 checksum has not been calculated. */
+#define _NETIO_PKT_NO_L3_CSUM_SHIFT           1
+#define _NETIO_PKT_NO_L3_CSUM_RMASK           1
+#define _NETIO_PKT_NO_L3_CSUM_MASK \
+         (_NETIO_PKT_NO_L3_CSUM_RMASK << _NETIO_PKT_NO_L3_CSUM_SHIFT)
+
+/** The L3 checksum is incorrect (or perhaps has not been calculated). */
+#define _NETIO_PKT_BAD_L3_CSUM_SHIFT          2
+#define _NETIO_PKT_BAD_L3_CSUM_RMASK          1
+#define _NETIO_PKT_BAD_L3_CSUM_MASK \
+         (_NETIO_PKT_BAD_L3_CSUM_RMASK << _NETIO_PKT_BAD_L3_CSUM_SHIFT)
+
+/** The Ethernet packet type is unrecognized. */
+#define _NETIO_PKT_TYPE_UNRECOGNIZED_SHIFT    3
+#define _NETIO_PKT_TYPE_UNRECOGNIZED_RMASK    1
+#define _NETIO_PKT_TYPE_UNRECOGNIZED_MASK \
+         (_NETIO_PKT_TYPE_UNRECOGNIZED_RMASK << \
+          _NETIO_PKT_TYPE_UNRECOGNIZED_SHIFT)
+
+/* Metadata packet type flags. */
+
+/** Where the packet type bits are; this field is the index into
+ *  _netio_pkt_info. */
+#define _NETIO_PKT_TYPE_SHIFT        4
+#define _NETIO_PKT_TYPE_RMASK        0x3F
+
+/** How many VLAN tags the packet has, and, if we have two, which one we
+ *  actually grouped on.  A VLAN within a proprietary (Marvell or Broadcom)
+ *  tag is counted here. */
+#define _NETIO_PKT_VLAN_SHIFT        4
+#define _NETIO_PKT_VLAN_RMASK        0x3
+#define _NETIO_PKT_VLAN_MASK \
+         (_NETIO_PKT_VLAN_RMASK << _NETIO_PKT_VLAN_SHIFT)
+#define _NETIO_PKT_VLAN_NONE         0   /* No VLAN tag. */
+#define _NETIO_PKT_VLAN_ONE          1   /* One VLAN tag. */
+#define _NETIO_PKT_VLAN_TWO_OUTER    2   /* Two VLAN tags, outer one used. */
+#define _NETIO_PKT_VLAN_TWO_INNER    3   /* Two VLAN tags, inner one used. */
+
+/** Which proprietary tags the packet has. */
+#define _NETIO_PKT_TAG_SHIFT         6
+#define _NETIO_PKT_TAG_RMASK         0x3
+#define _NETIO_PKT_TAG_MASK \
+          (_NETIO_PKT_TAG_RMASK << _NETIO_PKT_TAG_SHIFT)
+#define _NETIO_PKT_TAG_NONE          0   /* No proprietary tags. */
+#define _NETIO_PKT_TAG_MRVL          1   /* Marvell HyperG.Stack tags. */
+#define _NETIO_PKT_TAG_MRVL_EXT      2   /* HyperG.Stack extended tags. */
+#define _NETIO_PKT_TAG_BRCM          3   /* Broadcom HiGig tags. */
+
+/** Whether a packet has an LLC + SNAP header. */
+#define _NETIO_PKT_SNAP_SHIFT        8
+#define _NETIO_PKT_SNAP_RMASK        0x1
+#define _NETIO_PKT_SNAP_MASK \
+          (_NETIO_PKT_SNAP_RMASK << _NETIO_PKT_SNAP_SHIFT)
+
+/* NOTE: Bits 9 and 10 are unused. */
+
+/** Length of any custom data before the L2 header, in words. */
+#define _NETIO_PKT_CUSTOM_LEN_SHIFT  11
+#define _NETIO_PKT_CUSTOM_LEN_RMASK  0x1F
+#define _NETIO_PKT_CUSTOM_LEN_MASK \
+          (_NETIO_PKT_CUSTOM_LEN_RMASK << _NETIO_PKT_CUSTOM_LEN_SHIFT)
+
+/** The L4 checksum is incorrect (or perhaps has not been calculated). */
+#define _NETIO_PKT_BAD_L4_CSUM_SHIFT 16
+#define _NETIO_PKT_BAD_L4_CSUM_RMASK 0x1
+#define _NETIO_PKT_BAD_L4_CSUM_MASK \
+          (_NETIO_PKT_BAD_L4_CSUM_RMASK << _NETIO_PKT_BAD_L4_CSUM_SHIFT)
+
+/** Length of the L2 header, in words. */
+#define _NETIO_PKT_L2_LEN_SHIFT  17
+#define _NETIO_PKT_L2_LEN_RMASK  0x1F
+#define _NETIO_PKT_L2_LEN_MASK \
+          (_NETIO_PKT_L2_LEN_RMASK << _NETIO_PKT_L2_LEN_SHIFT)
+
+
+/* Flags in minimal packet metadata. */
+
+/** We need an eDMA checksum on this packet. */
+#define _NETIO_PKT_NEED_EDMA_CSUM_SHIFT            0
+#define _NETIO_PKT_NEED_EDMA_CSUM_RMASK            1
+#define _NETIO_PKT_NEED_EDMA_CSUM_MASK \
+         (_NETIO_PKT_NEED_EDMA_CSUM_RMASK << _NETIO_PKT_NEED_EDMA_CSUM_SHIFT)
+
+/* Data within the packet information table. */
+
+/* Note that, for efficiency, code which uses these fields assumes that none
+ * of the shift values below are zero.  See uses below for an explanation. */
+
+/** Offset within the L2 header of the innermost ethertype (in halfwords). */
+#define _NETIO_PKT_INFO_ETYPE_SHIFT       6
+#define _NETIO_PKT_INFO_ETYPE_RMASK    0x1F
+
+/** Offset within the L2 header of the VLAN tag (in halfwords). */
+#define _NETIO_PKT_INFO_VLAN_SHIFT       11
+#define _NETIO_PKT_INFO_VLAN_RMASK     0x1F
+
+#endif
+
+
+/** The size of a memory buffer representing a small packet.
+ *  @ingroup egress */
+#define SMALL_PACKET_SIZE 256
+
+/** The size of a memory buffer representing a large packet.
+ *  @ingroup egress */
+#define LARGE_PACKET_SIZE 2048
+
+/** The size of a memory buffer representing a jumbo packet.
+ *  @ingroup egress */
+#define JUMBO_PACKET_SIZE (12 * 1024)
+
+
+/* Common ethertypes.
+ * @ingroup ingress */
+/** @{ */
+/** The ethertype of IPv4. */
+#define ETHERTYPE_IPv4 (0x0800)
+/** The ethertype of ARP. */
+#define ETHERTYPE_ARP (0x0806)
+/** The ethertype of VLANs. */
+#define ETHERTYPE_VLAN (0x8100)
+/** The ethertype of a Q-in-Q header. */
+#define ETHERTYPE_Q_IN_Q (0x9100)
+/** The ethertype of IPv6. */
+#define ETHERTYPE_IPv6 (0x86DD)
+/** The ethertype of MPLS. */
+#define ETHERTYPE_MPLS (0x8847)
+/** @} */
+
+
+/** The possible return values of NETIO_PKT_STATUS.
+ * @ingroup ingress
+ */
+typedef enum
+{
+  /** No problems were detected with this packet. */
+  NETIO_PKT_STATUS_OK,
+  /** The packet is undersized; this is expected behavior if the packet's
+    * ethertype is unrecognized, but otherwise the packet is likely corrupt. */
+  NETIO_PKT_STATUS_UNDERSIZE,
+  /** The packet is oversized and some trailing bytes have been discarded.
+      This is expected behavior for short packets, since it's impossible to
+      precisely determine the amount of padding which may have been added to
+      them to make them meet the minimum Ethernet packet size. */
+  NETIO_PKT_STATUS_OVERSIZE,
+  /** The packet was judged to be corrupt by hardware (for instance, it had
+      a bad CRC, or part of it was discarded due to lack of buffer space in
+      the I/O shim) and should be discarded. */
+  NETIO_PKT_STATUS_BAD
+} netio_pkt_status_t;
+
+
+/** Log2 of how many buckets we have. */
+#define NETIO_LOG2_NUM_BUCKETS (10)
+
+/** How many buckets we have.
+ * @ingroup ingress */
+#define NETIO_NUM_BUCKETS (1 << NETIO_LOG2_NUM_BUCKETS)
+
+
+/**
+ * @brief A group-to-bucket identifier.
+ *
+ * @ingroup setup
+ *
+ * This tells us what to do with a given group.
+ */
+typedef union {
+  /** The header broken down into bits. */
+  struct {
+    /** Whether we should balance on L4, if available */
+    unsigned int __balance_on_l4:1;
+    /** Whether we should balance on L3, if available */
+    unsigned int __balance_on_l3:1;
+    /** Whether we should balance on L2, if available */
+    unsigned int __balance_on_l2:1;
+    /** Reserved for future use */
+    unsigned int __reserved:1;
+    /** The base bucket to use to send traffic */
+    unsigned int __bucket_base:NETIO_LOG2_NUM_BUCKETS;
+    /** The mask to apply to the balancing value. This must be one less
+     * than a power of two, e.g. 0x3 or 0xFF.
+     */
+    unsigned int __bucket_mask:NETIO_LOG2_NUM_BUCKETS;
+    /** Pad to 32 bits */
+    unsigned int __padding:(32 - 4 - 2 * NETIO_LOG2_NUM_BUCKETS);
+  } bits;
+  /** To send out the IDN. */
+  unsigned int word;
+}
+netio_group_t;
+
+
+/**
+ * @brief A VLAN-to-bucket identifier.
+ *
+ * @ingroup setup
+ *
+ * This tells us what to do with a given VLAN.
+ */
+typedef netio_group_t netio_vlan_t;
+
+
+/**
+ * A bucket-to-queue mapping.
+ * @ingroup setup
+ */
+typedef unsigned char netio_bucket_t;
+
+
+/**
+ * A packet size can always fit in a netio_size_t.
+ * @ingroup setup
+ */
+typedef unsigned int netio_size_t;
+
+
+/**
+ * @brief Ethernet standard (ingress) packet metadata.
+ *
+ * @ingroup ingress
+ *
+ * This is additional data associated with each packet.
+ * This structure is opaque and accessed through the @ref ingress.
+ *
+ * Also, the buffer population operation currently assumes that standard
+ * metadata is at least as large as minimal metadata, and will need to be
+ * modified if that is no longer the case.
+ */
+typedef struct
+{
+#ifdef __DOXYGEN__
+  /** This structure is opaque. */
+  unsigned char opaque[24];
+#else
+  /** The overall ordinal of the packet */
+  unsigned int __packet_ordinal;
+  /** The ordinal of the packet within the group */
+  unsigned int __group_ordinal;
+  /** The best flow hash IPP could compute. */
+  unsigned int __flow_hash;
+  /** Flags pertaining to checksum calculation, packet type, etc. */
+  unsigned int __flags;
+  /** The first word of "user data". */
+  unsigned int __user_data_0;
+  /** The second word of "user data". */
+  unsigned int __user_data_1;
+#endif
+}
+netio_pkt_metadata_t;
+
+
+/** To ensure that the L3 header is aligned mod 4, the L2 header should be
+ * aligned mod 4 plus 2, since every supported L2 header is 4n + 2 bytes
+ * long.  The standard way to do this is to simply add 2 bytes of padding
+ * before the L2 header.
+ */
+#define NETIO_PACKET_PADDING 2
+
+
+
+/**
+ * @brief Ethernet minimal (egress) packet metadata.
+ *
+ * @ingroup egress
+ *
+ * This structure represents information about packets which have
+ * been processed by @ref netio_populate_buffer() or
+ * @ref netio_populate_prepend_buffer().  This structure is opaque
+ * and accessed through the @ref egress.
+ *
+ * @internal This structure is actually copied into the memory used by
+ * standard metadata, which is assumed to be large enough.
+ */
+typedef struct
+{
+#ifdef __DOXYGEN__
+  /** This structure is opaque. */
+  unsigned char opaque[14];
+#else
+  /** The offset of the L2 header from the start of the packet data. */
+  unsigned short l2_offset;
+  /** The offset of the L3 header from the start of the packet data. */
+  unsigned short l3_offset;
+  /** Where to write the checksum. */
+  unsigned char csum_location;
+  /** Where to start checksumming from. */
+  unsigned char csum_start;
+  /** Flags pertaining to checksum calculation etc. */
+  unsigned short flags;
+  /** The L2 length of the packet. */
+  unsigned short l2_length;
+  /** The checksum with which to seed the checksum generator. */
+  unsigned short csum_seed;
+  /** How much to checksum. */
+  unsigned short csum_length;
+#endif
+}
+netio_pkt_minimal_metadata_t;
+
+
+#ifndef __DOXYGEN__
+
+/**
+ * @brief An I/O notification header.
+ *
+ * This is the first word of data received from an I/O shim in a notification
+ * packet. It contains framing and status information.
+ */
+typedef union
+{
+  unsigned int word; /**< The whole word. */
+  /** The various fields. */
+  struct
+  {
+    unsigned int __channel:7;    /**< Resource channel. */
+    unsigned int __type:4;       /**< Type. */
+    unsigned int __ack:1;        /**< Whether an acknowledgement is needed. */
+    unsigned int __reserved:1;   /**< Reserved. */
+    unsigned int __protocol:1;   /**< A protocol-specific word is added. */
+    unsigned int __status:2;     /**< Status of the transfer. */
+    unsigned int __framing:2;    /**< Framing of the transfer. */
+    unsigned int __transfer_size:14; /**< Transfer size in bytes (total). */
+  } bits;
+}
+__netio_pkt_notif_t;
+
+
+/**
+ * Returns the base address of the packet.
+ */
+#define _NETIO_PKT_HANDLE_BASE(p) \
+  ((unsigned char*)((p).word & 0xFFFFFFC0))
+
+/**
+ * Returns the base address of the packet.
+ */
+#define _NETIO_PKT_BASE(p) \
+  _NETIO_PKT_HANDLE_BASE(p->__packet)
+
+/**
+ * @brief An I/O notification packet (second word)
+ *
+ * This is the second word of data received from an I/O shim in a notification
+ * packet.  This is the virtual address of the packet buffer, plus some flag
+ * bits.  (The virtual address of the packet is always 256-byte aligned so we
+ * have room for 8 bits' worth of flags in the low 8 bits.)
+ *
+ * @internal
+ * NOTE: The low two bits must contain "__queue", so the "packet size"
+ * (SIZE_SMALL, SIZE_LARGE, or SIZE_JUMBO) can be determined quickly.
+ *
+ * If __addr or __offset are moved, _NETIO_PKT_BASE
+ * (defined right below this) must be changed.
+ */
+typedef union
+{
+  unsigned int word; /**< The whole word. */
+  /** The various fields. */
+  struct
+  {
+    /** Which queue the packet will be returned to once it is sent back to
+        the IPP.  This is one of the SIZE_xxx values. */
+    unsigned int __queue:2;
+
+    /** The IPP handle of the sending IPP. */
+    unsigned int __ipp_handle:2;
+
+    /** Reserved for future use. */
+    unsigned int __reserved:1;
+
+    /** If 1, this packet has minimal (egress) metadata; otherwise, it
+        has standard (ingress) metadata. */
+    unsigned int __minimal:1;
+
+    /** Offset of the metadata within the packet.  This value is multiplied
+     *  by 64 and added to the base packet address to get the metadata
+     *  address.  Note that this field is aligned within the word such that
+     *  you can easily extract the metadata address with a 26-bit mask. */
+    unsigned int __offset:2;
+
+    /** The top 24 bits of the packet's virtual address. */
+    unsigned int __addr:24;
+  } bits;
+}
+__netio_pkt_handle_t;
+
+#endif /* !__DOXYGEN__ */
+
+
+/**
+ * @brief A handle for an I/O packet's storage.
+ * @ingroup ingress
+ *
+ * netio_pkt_handle_t encodes the concept of a ::netio_pkt_t with its
+ * packet metadata removed.  It is a much smaller type that exists to
+ * facilitate applications where the full ::netio_pkt_t type is too
+ * large, such as those that cache enormous numbers of packets or wish
+ * to transmit packet descriptors over the UDN.
+ *
+ * Because there is no metadata, most ::netio_pkt_t operations cannot be
+ * performed on a netio_pkt_handle_t.  It supports only
+ * netio_free_handle() (to free the buffer) and
+ * NETIO_PKT_CUSTOM_DATA_H() (to access a pointer to its contents).
+ * The application must acquire any additional metadata it wants from the
+ * original ::netio_pkt_t and record it separately.
+ *
+ * A netio_pkt_handle_t can be extracted from a ::netio_pkt_t by calling
+ * NETIO_PKT_HANDLE().  An invalid handle (analogous to NULL) can be
+ * created by assigning the value ::NETIO_PKT_HANDLE_NONE. A handle can
+ * be tested for validity with NETIO_PKT_HANDLE_IS_VALID().
+ */
+typedef struct
+{
+  unsigned int word; /**< Opaque bits. */
+} netio_pkt_handle_t;
+
+/**
+ * @brief A packet descriptor.
+ *
+ * @ingroup ingress
+ * @ingroup egress
+ *
+ * This data structure represents a packet.  The structure is manipulated
+ * through the @ref ingress and the @ref egress.
+ *
+ * While the contents of a netio_pkt_t are opaque, the structure itself is
+ * portable.  This means that it may be shared between all tiles which have
+ * done a netio_input_register() call for the interface on which the pkt_t
+ * was initially received (via netio_get_packet()) or retrieved (via
+ * netio_get_buffer()).  The contents of a netio_pkt_t can be transmitted to
+ * another tile via shared memory, or via a UDN message, or by other means.
+ * The destination tile may then use the pkt_t as if it had originally been
+ * received locally; it may read or write the packet's data, read its
+ * metadata, free the packet, send the packet, transfer the netio_pkt_t to
+ * yet another tile, and so forth.
+ *
+ * Once a netio_pkt_t has been transferred to a second tile, the first tile
+ * should not reference the original copy; in particular, if more than one
+ * tile frees or sends the same netio_pkt_t, the IPP's packet free lists will
+ * become corrupted.  Note also that each tile which reads or modifies
+ * packet data must obey the memory coherency rules outlined in @ref input.
+ */
+typedef struct
+{
+#ifdef __DOXYGEN__
+  /** This structure is opaque. */
+  unsigned char opaque[32];
+#else
+  /** For an ingress packet (one with standard metadata), this is the
+   *  notification header we got from the I/O shim.  For an egress packet
+   *  (one with minimal metadata), this word is zero if the packet has not
+   *  been populated, and nonzero if it has. */
+  __netio_pkt_notif_t __notif_header;
+
+  /** Virtual address of the packet buffer, plus state flags. */
+  __netio_pkt_handle_t __packet;
+
+  /** Metadata associated with the packet. */
+  netio_pkt_metadata_t __metadata;
+#endif
+}
+netio_pkt_t;
+
+
+#ifndef __DOXYGEN__
+
+#define __NETIO_PKT_NOTIF_HEADER(pkt) ((pkt)->__notif_header)
+#define __NETIO_PKT_IPP_HANDLE(pkt) ((pkt)->__packet.bits.__ipp_handle)
+#define __NETIO_PKT_QUEUE(pkt) ((pkt)->__packet.bits.__queue)
+#define __NETIO_PKT_NOTIF_HEADER_M(mda, pkt) ((pkt)->__notif_header)
+#define __NETIO_PKT_IPP_HANDLE_M(mda, pkt) ((pkt)->__packet.bits.__ipp_handle)
+#define __NETIO_PKT_MINIMAL(pkt) ((pkt)->__packet.bits.__minimal)
+#define __NETIO_PKT_QUEUE_M(mda, pkt) ((pkt)->__packet.bits.__queue)
+#define __NETIO_PKT_FLAGS_M(mda, pkt) ((mda)->__flags)
+
+/* Packet information table, used by the attribute access functions below. */
+extern const uint16_t _netio_pkt_info[];
+
+#endif /* __DOXYGEN__ */
+
+
+#ifndef __DOXYGEN__
+/* These macros are deprecated and will disappear in a future MDE release. */
+#define NETIO_PKT_GOOD_CHECKSUM(pkt) \
+  NETIO_PKT_L4_CSUM_CORRECT(pkt)
+#define NETIO_PKT_GOOD_CHECKSUM_M(mda, pkt) \
+  NETIO_PKT_L4_CSUM_CORRECT_M(mda, pkt)
+#endif /* __DOXYGEN__ */
+
+
+/* Packet attribute access functions. */
+
+/** Return a pointer to the metadata for a packet.
+ * @ingroup ingress
+ *
+ * Calling this function once and passing the result to other retrieval
+ * functions with a "_M" suffix usually improves performance.  This
+ * function must be called on an 'ingress' packet (i.e. one retrieved
+ * by @ref netio_get_packet(), on which @ref netio_populate_buffer() or
+ * @ref netio_populate_prepend_buffer have not been called). Use of this
+ * function on an 'egress' packet will cause an assertion failure.
+ *
+ * @param[in] pkt Packet on which to operate.
+ * @return A pointer to the packet's standard metadata.
+ */
+static __inline netio_pkt_metadata_t*
+NETIO_PKT_METADATA(netio_pkt_t* pkt)
+{
+  netio_assert(!pkt->__packet.bits.__minimal);
+  return &pkt->__metadata;
+}
+
+
+/** Return a pointer to the minimal metadata for a packet.
+ * @ingroup egress
+ *
+ * Calling this function once and passing the result to other retrieval
+ * functions with a "_MM" suffix usually improves performance.  This
+ * function must be called on an 'egress' packet (i.e. one on which
+ * @ref netio_populate_buffer() or @ref netio_populate_prepend_buffer()
+ * have been called, or one retrieved by @ref netio_get_buffer()). Use of
+ * this function on an 'ingress' packet will cause an assertion failure.
+ *
+ * @param[in] pkt Packet on which to operate.
+ * @return A pointer to the packet's standard metadata.
+ */
+static __inline netio_pkt_minimal_metadata_t*
+NETIO_PKT_MINIMAL_METADATA(netio_pkt_t* pkt)
+{
+  netio_assert(pkt->__packet.bits.__minimal);
+  return (netio_pkt_minimal_metadata_t*) &pkt->__metadata;
+}
+
+
+/** Determine whether a packet has 'minimal' metadata.
+ * @ingroup pktfuncs
+ *
+ * This function will return nonzero if the packet is an 'egress'
+ * packet (i.e. one on which @ref netio_populate_buffer() or
+ * @ref netio_populate_prepend_buffer() have been called, or one
+ * retrieved by @ref netio_get_buffer()), and zero if the packet
+ * is an 'ingress' packet (i.e. one retrieved by @ref netio_get_packet(),
+ * which has not been converted into an 'egress' packet).
+ *
+ * @param[in] pkt Packet on which to operate.
+ * @return Nonzero if the packet has minimal metadata.
+ */
+static __inline unsigned int
+NETIO_PKT_IS_MINIMAL(netio_pkt_t* pkt)
+{
+  return pkt->__packet.bits.__minimal;
+}
+
+
+/** Return a handle for a packet's storage.
+ * @ingroup pktfuncs
+ *
+ * @param[in] pkt Packet on which to operate.
+ * @return A handle for the packet's storage.
+ */
+static __inline netio_pkt_handle_t
+NETIO_PKT_HANDLE(netio_pkt_t* pkt)
+{
+  netio_pkt_handle_t h;
+  h.word = pkt->__packet.word;
+  return h;
+}
+
+
+/** A special reserved value indicating the absence of a packet handle.
+ *
+ * @ingroup pktfuncs
+ */
+#define NETIO_PKT_HANDLE_NONE ((netio_pkt_handle_t) { 0 })
+
+
+/** Test whether a packet handle is valid.
+ *
+ * Applications may wish to use the reserved value NETIO_PKT_HANDLE_NONE
+ * to indicate no packet at all.  This function tests to see if a packet
+ * handle is a real handle, not this special reserved value.
+ *
+ * @ingroup pktfuncs
+ *
+ * @param[in] handle Handle on which to operate.
+ * @return One if the packet handle is valid, else zero.
+ */
+static __inline unsigned int
+NETIO_PKT_HANDLE_IS_VALID(netio_pkt_handle_t handle)
+{
+  return handle.word != 0;
+}
+
+
+
+/** Return a pointer to the start of the packet's custom header.
+ *  A custom header may or may not be present, depending upon the IPP; its
+ *  contents and alignment are also IPP-dependent.  Currently, none of the
+ *  standard IPPs supplied by Tilera produce a custom header.  If present,
+ *  the custom header precedes the L2 header in the packet buffer.
+ * @ingroup ingress
+ *
+ * @param[in] handle Handle on which to operate.
+ * @return A pointer to start of the packet.
+ */
+static __inline unsigned char*
+NETIO_PKT_CUSTOM_DATA_H(netio_pkt_handle_t handle)
+{
+  return _NETIO_PKT_HANDLE_BASE(handle) + NETIO_PACKET_PADDING;
+}
+
+
+/** Return the length of the packet's custom header.
+ *  A custom header may or may not be present, depending upon the IPP; its
+ *  contents and alignment are also IPP-dependent.  Currently, none of the
+ *  standard IPPs supplied by Tilera produce a custom header.  If present,
+ *  the custom header precedes the L2 header in the packet buffer.
+ *
+ * @ingroup ingress
+ *
+ * @param[in] mda Pointer to packet's standard metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return The length of the packet's custom header, in bytes.
+ */
+static __inline netio_size_t
+NETIO_PKT_CUSTOM_HEADER_LENGTH_M(netio_pkt_metadata_t* mda, netio_pkt_t* pkt)
+{
+  /*
+   * Note that we effectively need to extract a quantity from the flags word
+   * which is measured in words, and then turn it into bytes by shifting
+   * it left by 2.  We do this all at once by just shifting right two less
+   * bits, and shifting the mask up two bits.
+   */
+  return ((mda->__flags >> (_NETIO_PKT_CUSTOM_LEN_SHIFT - 2)) &
+          (_NETIO_PKT_CUSTOM_LEN_RMASK << 2));
+}
+
+
+/** Return the length of the packet, starting with the custom header.
+ *  A custom header may or may not be present, depending upon the IPP; its
+ *  contents and alignment are also IPP-dependent.  Currently, none of the
+ *  standard IPPs supplied by Tilera produce a custom header.  If present,
+ *  the custom header precedes the L2 header in the packet buffer.
+ * @ingroup ingress
+ *
+ * @param[in] mda Pointer to packet's standard metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return The length of the packet, in bytes.
+ */
+static __inline netio_size_t
+NETIO_PKT_CUSTOM_LENGTH_M(netio_pkt_metadata_t* mda, netio_pkt_t* pkt)
+{
+  return (__NETIO_PKT_NOTIF_HEADER(pkt).bits.__transfer_size -
+          NETIO_PACKET_PADDING);
+}
+
+
+/** Return a pointer to the start of the packet's custom header.
+ *  A custom header may or may not be present, depending upon the IPP; its
+ *  contents and alignment are also IPP-dependent.  Currently, none of the
+ *  standard IPPs supplied by Tilera produce a custom header.  If present,
+ *  the custom header precedes the L2 header in the packet buffer.
+ * @ingroup ingress
+ *
+ * @param[in] mda Pointer to packet's standard metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return A pointer to start of the packet.
+ */
+static __inline unsigned char*
+NETIO_PKT_CUSTOM_DATA_M(netio_pkt_metadata_t* mda, netio_pkt_t* pkt)
+{
+  return NETIO_PKT_CUSTOM_DATA_H(NETIO_PKT_HANDLE(pkt));
+}
+
+
+/** Return the length of the packet's L2 (Ethernet plus VLAN or SNAP) header.
+ * @ingroup ingress
+ *
+ * @param[in] mda Pointer to packet's standard metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return The length of the packet's L2 header, in bytes.
+ */
+static __inline netio_size_t
+NETIO_PKT_L2_HEADER_LENGTH_M(netio_pkt_metadata_t* mda, netio_pkt_t* pkt)
+{
+  /*
+   * Note that we effectively need to extract a quantity from the flags word
+   * which is measured in words, and then turn it into bytes by shifting
+   * it left by 2.  We do this all at once by just shifting right two less
+   * bits, and shifting the mask up two bits.  We then add two bytes.
+   */
+  return ((mda->__flags >> (_NETIO_PKT_L2_LEN_SHIFT - 2)) &
+          (_NETIO_PKT_L2_LEN_RMASK << 2)) + 2;
+}
+
+
+/** Return the length of the packet, starting with the L2 (Ethernet) header.
+ * @ingroup ingress
+ *
+ * @param[in] mda Pointer to packet's standard metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return The length of the packet, in bytes.
+ */
+static __inline netio_size_t
+NETIO_PKT_L2_LENGTH_M(netio_pkt_metadata_t* mda, netio_pkt_t* pkt)
+{
+  return (NETIO_PKT_CUSTOM_LENGTH_M(mda, pkt) -
+          NETIO_PKT_CUSTOM_HEADER_LENGTH_M(mda,pkt));
+}
+
+
+/** Return a pointer to the start of the packet's L2 (Ethernet) header.
+ * @ingroup ingress
+ *
+ * @param[in] mda Pointer to packet's standard metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return A pointer to start of the packet.
+ */
+static __inline unsigned char*
+NETIO_PKT_L2_DATA_M(netio_pkt_metadata_t* mda, netio_pkt_t* pkt)
+{
+  return (NETIO_PKT_CUSTOM_DATA_M(mda, pkt) +
+          NETIO_PKT_CUSTOM_HEADER_LENGTH_M(mda, pkt));
+}
+
+
+/** Retrieve the length of the packet, starting with the L3 (generally,
+ *  the IP) header.
+ * @ingroup ingress
+ *
+ * @param[in] mda Pointer to packet's standard metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return Length of the packet's L3 header and data, in bytes.
+ */
+static __inline netio_size_t
+NETIO_PKT_L3_LENGTH_M(netio_pkt_metadata_t* mda, netio_pkt_t* pkt)
+{
+  return (NETIO_PKT_L2_LENGTH_M(mda, pkt) -
+          NETIO_PKT_L2_HEADER_LENGTH_M(mda,pkt));
+}
+
+
+/** Return a pointer to the packet's L3 (generally, the IP) header.
+ * @ingroup ingress
+ *
+ * Note that we guarantee word alignment of the L3 header.
+ *
+ * @param[in] mda Pointer to packet's standard metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return A pointer to the packet's L3 header.
+ */
+static __inline unsigned char*
+NETIO_PKT_L3_DATA_M(netio_pkt_metadata_t* mda, netio_pkt_t* pkt)
+{
+  return (NETIO_PKT_L2_DATA_M(mda, pkt) +
+          NETIO_PKT_L2_HEADER_LENGTH_M(mda, pkt));
+}
+
+
+/** Return the ordinal of the packet.
+ * @ingroup ingress
+ *
+ * Each packet is given an ordinal number when it is delivered by the IPP.
+ * In the medium term, the ordinal is unique and monotonically increasing,
+ * being incremented by 1 for each packet; the ordinal of the first packet
+ * delivered after the IPP starts is zero.  (Since the ordinal is of finite
+ * size, given enough input packets, it will eventually wrap around to zero;
+ * in the long term, therefore, ordinals are not unique.)  The ordinals
+ * handed out by different IPPs are not disjoint, so two packets from
+ * different IPPs may have identical ordinals.  Packets dropped by the
+ * IPP or by the I/O shim are not assigned ordinals.
+ *
+ * @param[in] mda Pointer to packet's standard metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return The packet's per-IPP packet ordinal.
+ */
+static __inline unsigned int
+NETIO_PKT_ORDINAL_M(netio_pkt_metadata_t* mda, netio_pkt_t* pkt)
+{
+  return mda->__packet_ordinal;
+}
+
+
+/** Return the per-group ordinal of the packet.
+ * @ingroup ingress
+ *
+ * Each packet is given a per-group ordinal number when it is
+ * delivered by the IPP. By default, the group is the packet's VLAN,
+ * although IPP can be recompiled to use different values.  In
+ * the medium term, the ordinal is unique and monotonically
+ * increasing, being incremented by 1 for each packet; the ordinal of
+ * the first packet distributed to a particular group is zero.
+ * (Since the ordinal is of finite size, given enough input packets,
+ * it will eventually wrap around to zero; in the long term,
+ * therefore, ordinals are not unique.)  The ordinals handed out by
+ * different IPPs are not disjoint, so two packets from different IPPs
+ * may have identical ordinals; similarly, packets distributed to
+ * different groups may have identical ordinals.  Packets dropped by
+ * the IPP or by the I/O shim are not assigned ordinals.
+ *
+ * @param[in] mda Pointer to packet's standard metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return The packet's per-IPP, per-group ordinal.
+ */
+static __inline unsigned int
+NETIO_PKT_GROUP_ORDINAL_M(netio_pkt_metadata_t* mda, netio_pkt_t* pkt)
+{
+  return mda->__group_ordinal;
+}
+
+
+/** Return the VLAN ID assigned to the packet.
+ * @ingroup ingress
+ *
+ * This value is usually contained within the packet header.
+ *
+ * This value will be zero if the packet does not have a VLAN tag, or if
+ * this value was not extracted from the packet.
+ *
+ * @param[in] mda Pointer to packet's standard metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return The packet's VLAN ID.
+ */
+static __inline unsigned short
+NETIO_PKT_VLAN_ID_M(netio_pkt_metadata_t* mda, netio_pkt_t* pkt)
+{
+  int vl = (mda->__flags >> _NETIO_PKT_VLAN_SHIFT) & _NETIO_PKT_VLAN_RMASK;
+  unsigned short* pkt_p;
+  int index;
+  unsigned short val;
+
+  if (vl == _NETIO_PKT_VLAN_NONE)
+    return 0;
+
+  pkt_p = (unsigned short*) NETIO_PKT_L2_DATA_M(mda, pkt);
+  index = (mda->__flags >> _NETIO_PKT_TYPE_SHIFT) & _NETIO_PKT_TYPE_RMASK;
+
+  val = pkt_p[(_netio_pkt_info[index] >> _NETIO_PKT_INFO_VLAN_SHIFT) &
+              _NETIO_PKT_INFO_VLAN_RMASK];
+
+#ifdef __TILECC__
+  return (__insn_bytex(val) >> 16) & 0xFFF;
+#else
+  return (__builtin_bswap32(val) >> 16) & 0xFFF;
+#endif
+}
+
+
+/** Return the ethertype of the packet.
+ * @ingroup ingress
+ *
+ * This value is usually contained within the packet header.
+ *
+ * This value is reliable if @ref NETIO_PKT_ETHERTYPE_RECOGNIZED_M()
+ * returns true, and otherwise, may not be well defined.
+ *
+ * @param[in] mda Pointer to packet's standard metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return The packet's ethertype.
+ */
+static __inline unsigned short
+NETIO_PKT_ETHERTYPE_M(netio_pkt_metadata_t* mda, netio_pkt_t* pkt)
+{
+  unsigned short* pkt_p = (unsigned short*) NETIO_PKT_L2_DATA_M(mda, pkt);
+  int index = (mda->__flags >> _NETIO_PKT_TYPE_SHIFT) & _NETIO_PKT_TYPE_RMASK;
+
+  unsigned short val =
+    pkt_p[(_netio_pkt_info[index] >> _NETIO_PKT_INFO_ETYPE_SHIFT) &
+          _NETIO_PKT_INFO_ETYPE_RMASK];
+
+  return __builtin_bswap32(val) >> 16;
+}
+
+
+/** Return the flow hash computed on the packet.
+ * @ingroup ingress
+ *
+ * For TCP and UDP packets, this hash is calculated by hashing together
+ * the "5-tuple" values, specifically the source IP address, destination
+ * IP address, protocol type, source port and destination port.
+ * The hash value is intended to be helpful for millions of distinct
+ * flows.
+ *
+ * For IPv4 or IPv6 packets which are neither TCP nor UDP, the flow hash is
+ * derived by hashing together the source and destination IP addresses.
+ *
+ * For MPLS-encapsulated packets, the flow hash is derived by hashing
+ * the first MPLS label.
+ *
+ * For all other packets the flow hash is computed from the source
+ * and destination Ethernet addresses.
+ *
+ * The hash is symmetric, meaning it produces the same value if the
+ * source and destination are swapped. The only exceptions are
+ * tunneling protocols 0x04 (IP in IP Encapsulation), 0x29 (Simple
+ * Internet Protocol), 0x2F (General Routing Encapsulation) and 0x32
+ * (Encap Security Payload), which use only the destination address
+ * since the source address is not meaningful.
+ *
+ * @param[in] mda Pointer to packet's standard metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return The packet's 32-bit flow hash.
+ */
+static __inline unsigned int
+NETIO_PKT_FLOW_HASH_M(netio_pkt_metadata_t* mda, netio_pkt_t* pkt)
+{
+  return mda->__flow_hash;
+}
+
+
+/** Return the first word of "user data" for the packet.
+ *
+ * The contents of the user data words depend on the IPP.
+ *
+ * When using the standard ipp1, ipp2, or ipp4 sub-drivers, the first
+ * word of user data contains the least significant bits of the 64-bit
+ * arrival cycle count (see @c get_cycle_count_low()).
+ *
+ * See the <em>System Programmer's Guide</em> for details.
+ *
+ * @ingroup ingress
+ *
+ * @param[in] mda Pointer to packet's standard metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return The packet's first word of "user data".
+ */
+static __inline unsigned int
+NETIO_PKT_USER_DATA_0_M(netio_pkt_metadata_t* mda, netio_pkt_t* pkt)
+{
+  return mda->__user_data_0;
+}
+
+
+/** Return the second word of "user data" for the packet.
+ *
+ * The contents of the user data words depend on the IPP.
+ *
+ * When using the standard ipp1, ipp2, or ipp4 sub-drivers, the second
+ * word of user data contains the most significant bits of the 64-bit
+ * arrival cycle count (see @c get_cycle_count_high()).
+ *
+ * See the <em>System Programmer's Guide</em> for details.
+ *
+ * @ingroup ingress
+ *
+ * @param[in] mda Pointer to packet's standard metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return The packet's second word of "user data".
+ */
+static __inline unsigned int
+NETIO_PKT_USER_DATA_1_M(netio_pkt_metadata_t* mda, netio_pkt_t* pkt)
+{
+  return mda->__user_data_1;
+}
+
+
+/** Determine whether the L4 (TCP/UDP) checksum was calculated.
+ * @ingroup ingress
+ *
+ * @param[in] mda Pointer to packet's standard metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return Nonzero if the L4 checksum was calculated.
+ */
+static __inline unsigned int
+NETIO_PKT_L4_CSUM_CALCULATED_M(netio_pkt_metadata_t* mda, netio_pkt_t* pkt)
+{
+  return !(mda->__flags & _NETIO_PKT_NO_L4_CSUM_MASK);
+}
+
+
+/** Determine whether the L4 (TCP/UDP) checksum was calculated and found to
+ *  be correct.
+ * @ingroup ingress
+ *
+ * @param[in] mda Pointer to packet's standard metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return Nonzero if the checksum was calculated and is correct.
+ */
+static __inline unsigned int
+NETIO_PKT_L4_CSUM_CORRECT_M(netio_pkt_metadata_t* mda, netio_pkt_t* pkt)
+{
+  return !(mda->__flags &
+           (_NETIO_PKT_BAD_L4_CSUM_MASK | _NETIO_PKT_NO_L4_CSUM_MASK));
+}
+
+
+/** Determine whether the L3 (IP) checksum was calculated.
+ * @ingroup ingress
+ *
+ * @param[in] mda Pointer to packet's standard metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return Nonzero if the L3 (IP) checksum was calculated.
+*/
+static __inline unsigned int
+NETIO_PKT_L3_CSUM_CALCULATED_M(netio_pkt_metadata_t* mda, netio_pkt_t* pkt)
+{
+  return !(mda->__flags & _NETIO_PKT_NO_L3_CSUM_MASK);
+}
+
+
+/** Determine whether the L3 (IP) checksum was calculated and found to be
+ *  correct.
+ * @ingroup ingress
+ *
+ * @param[in] mda Pointer to packet's standard metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return Nonzero if the checksum was calculated and is correct.
+ */
+static __inline unsigned int
+NETIO_PKT_L3_CSUM_CORRECT_M(netio_pkt_metadata_t* mda, netio_pkt_t* pkt)
+{
+  return !(mda->__flags &
+           (_NETIO_PKT_BAD_L3_CSUM_MASK | _NETIO_PKT_NO_L3_CSUM_MASK));
+}
+
+
+/** Determine whether the ethertype was recognized and L3 packet data was
+ *  processed.
+ * @ingroup ingress
+ *
+ * @param[in] mda Pointer to packet's standard metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return Nonzero if the ethertype was recognized and L3 packet data was
+ *   processed.
+ */
+static __inline unsigned int
+NETIO_PKT_ETHERTYPE_RECOGNIZED_M(netio_pkt_metadata_t* mda, netio_pkt_t* pkt)
+{
+  return !(mda->__flags & _NETIO_PKT_TYPE_UNRECOGNIZED_MASK);
+}
+
+
+/** Retrieve the status of a packet and any errors that may have occurred
+ * during ingress processing (length mismatches, CRC errors, etc.).
+ * @ingroup ingress
+ *
+ * Note that packets for which @ref NETIO_PKT_ETHERTYPE_RECOGNIZED()
+ * returns zero are always reported as underlength, as there is no a priori
+ * means to determine their length.  Normally, applications should use
+ * @ref NETIO_PKT_BAD_M() instead of explicitly checking status with this
+ * function.
+ *
+ * @param[in] mda Pointer to packet's standard metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return The packet's status.
+ */
+static __inline netio_pkt_status_t
+NETIO_PKT_STATUS_M(netio_pkt_metadata_t* mda, netio_pkt_t* pkt)
+{
+  return (netio_pkt_status_t) __NETIO_PKT_NOTIF_HEADER(pkt).bits.__status;
+}
+
+
+/** Report whether a packet is bad (i.e., was shorter than expected based on
+ *  its headers, or had a bad CRC).
+ * @ingroup ingress
+ *
+ * Note that this function does not verify L3 or L4 checksums.
+ *
+ * @param[in] mda Pointer to packet's standard metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return Nonzero if the packet is bad and should be discarded.
+ */
+static __inline unsigned int
+NETIO_PKT_BAD_M(netio_pkt_metadata_t* mda, netio_pkt_t* pkt)
+{
+  return ((NETIO_PKT_STATUS_M(mda, pkt) & 1) &&
+          (NETIO_PKT_ETHERTYPE_RECOGNIZED_M(mda, pkt) ||
+           NETIO_PKT_STATUS_M(mda, pkt) == NETIO_PKT_STATUS_BAD));
+}
+
+
+/** Return the length of the packet, starting with the L2 (Ethernet) header.
+ * @ingroup egress
+ *
+ * @param[in] mmd Pointer to packet's minimal metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return The length of the packet, in bytes.
+ */
+static __inline netio_size_t
+NETIO_PKT_L2_LENGTH_MM(netio_pkt_minimal_metadata_t* mmd, netio_pkt_t* pkt)
+{
+  return mmd->l2_length;
+}
+
+
+/** Return the length of the L2 (Ethernet) header.
+ * @ingroup egress
+ *
+ * @param[in] mmd Pointer to packet's minimal metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return The length of the packet's L2 header, in bytes.
+ */
+static __inline netio_size_t
+NETIO_PKT_L2_HEADER_LENGTH_MM(netio_pkt_minimal_metadata_t* mmd,
+                              netio_pkt_t* pkt)
+{
+  return mmd->l3_offset - mmd->l2_offset;
+}
+
+
+/** Return the length of the packet, starting with the L3 (IP) header.
+ * @ingroup egress
+ *
+ * @param[in] mmd Pointer to packet's minimal metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return Length of the packet's L3 header and data, in bytes.
+ */
+static __inline netio_size_t
+NETIO_PKT_L3_LENGTH_MM(netio_pkt_minimal_metadata_t* mmd, netio_pkt_t* pkt)
+{
+  return (NETIO_PKT_L2_LENGTH_MM(mmd, pkt) -
+          NETIO_PKT_L2_HEADER_LENGTH_MM(mmd, pkt));
+}
+
+
+/** Return a pointer to the packet's L3 (generally, the IP) header.
+ * @ingroup egress
+ *
+ * Note that we guarantee word alignment of the L3 header.
+ *
+ * @param[in] mmd Pointer to packet's minimal metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return A pointer to the packet's L3 header.
+ */
+static __inline unsigned char*
+NETIO_PKT_L3_DATA_MM(netio_pkt_minimal_metadata_t* mmd, netio_pkt_t* pkt)
+{
+  return _NETIO_PKT_BASE(pkt) + mmd->l3_offset;
+}
+
+
+/** Return a pointer to the packet's L2 (Ethernet) header.
+ * @ingroup egress
+ *
+ * @param[in] mmd Pointer to packet's minimal metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @return A pointer to start of the packet.
+ */
+static __inline unsigned char*
+NETIO_PKT_L2_DATA_MM(netio_pkt_minimal_metadata_t* mmd, netio_pkt_t* pkt)
+{
+  return _NETIO_PKT_BASE(pkt) + mmd->l2_offset;
+}
+
+
+/** Retrieve the status of a packet and any errors that may have occurred
+ * during ingress processing (length mismatches, CRC errors, etc.).
+ * @ingroup ingress
+ *
+ * Note that packets for which @ref NETIO_PKT_ETHERTYPE_RECOGNIZED()
+ * returns zero are always reported as underlength, as there is no a priori
+ * means to determine their length.  Normally, applications should use
+ * @ref NETIO_PKT_BAD() instead of explicitly checking status with this
+ * function.
+ *
+ * @param[in] pkt Packet on which to operate.
+ * @return The packet's status.
+ */
+static __inline netio_pkt_status_t
+NETIO_PKT_STATUS(netio_pkt_t* pkt)
+{
+  netio_assert(!pkt->__packet.bits.__minimal);
+
+  return (netio_pkt_status_t) __NETIO_PKT_NOTIF_HEADER(pkt).bits.__status;
+}
+
+
+/** Report whether a packet is bad (i.e., was shorter than expected based on
+ *  its headers, or had a bad CRC).
+ * @ingroup ingress
+ *
+ * Note that this function does not verify L3 or L4 checksums.
+ *
+ * @param[in] pkt Packet on which to operate.
+ * @return Nonzero if the packet is bad and should be discarded.
+ */
+static __inline unsigned int
+NETIO_PKT_BAD(netio_pkt_t* pkt)
+{
+  netio_pkt_metadata_t* mda = NETIO_PKT_METADATA(pkt);
+
+  return NETIO_PKT_BAD_M(mda, pkt);
+}
+
+
+/** Return the length of the packet's custom header.
+ *  A custom header may or may not be present, depending upon the IPP; its
+ *  contents and alignment are also IPP-dependent.  Currently, none of the
+ *  standard IPPs supplied by Tilera produce a custom header.  If present,
+ *  the custom header precedes the L2 header in the packet buffer.
+ * @ingroup pktfuncs
+ *
+ * @param[in] pkt Packet on which to operate.
+ * @return The length of the packet's custom header, in bytes.
+ */
+static __inline netio_size_t
+NETIO_PKT_CUSTOM_HEADER_LENGTH(netio_pkt_t* pkt)
+{
+  netio_pkt_metadata_t* mda = NETIO_PKT_METADATA(pkt);
+
+  return NETIO_PKT_CUSTOM_HEADER_LENGTH_M(mda, pkt);
+}
+
+
+/** Return the length of the packet, starting with the custom header.
+ *  A custom header may or may not be present, depending upon the IPP; its
+ *  contents and alignment are also IPP-dependent.  Currently, none of the
+ *  standard IPPs supplied by Tilera produce a custom header.  If present,
+ *  the custom header precedes the L2 header in the packet buffer.
+ * @ingroup pktfuncs
+ *
+ * @param[in] pkt Packet on which to operate.
+ * @return  The length of the packet, in bytes.
+ */
+static __inline netio_size_t
+NETIO_PKT_CUSTOM_LENGTH(netio_pkt_t* pkt)
+{
+  netio_pkt_metadata_t* mda = NETIO_PKT_METADATA(pkt);
+
+  return NETIO_PKT_CUSTOM_LENGTH_M(mda, pkt);
+}
+
+
+/** Return a pointer to the packet's custom header.
+ *  A custom header may or may not be present, depending upon the IPP; its
+ *  contents and alignment are also IPP-dependent.  Currently, none of the
+ *  standard IPPs supplied by Tilera produce a custom header.  If present,
+ *  the custom header precedes the L2 header in the packet buffer.
+ * @ingroup pktfuncs
+ *
+ * @param[in] pkt Packet on which to operate.
+ * @return A pointer to start of the packet.
+ */
+static __inline unsigned char*
+NETIO_PKT_CUSTOM_DATA(netio_pkt_t* pkt)
+{
+  netio_pkt_metadata_t* mda = NETIO_PKT_METADATA(pkt);
+
+  return NETIO_PKT_CUSTOM_DATA_M(mda, pkt);
+}
+
+
+/** Return the length of the packet's L2 (Ethernet plus VLAN or SNAP) header.
+ * @ingroup pktfuncs
+ *
+ * @param[in] pkt Packet on which to operate.
+ * @return The length of the packet's L2 header, in bytes.
+ */
+static __inline netio_size_t
+NETIO_PKT_L2_HEADER_LENGTH(netio_pkt_t* pkt)
+{
+  if (NETIO_PKT_IS_MINIMAL(pkt))
+  {
+    netio_pkt_minimal_metadata_t* mmd = NETIO_PKT_MINIMAL_METADATA(pkt);
+
+    return NETIO_PKT_L2_HEADER_LENGTH_MM(mmd, pkt);
+  }
+  else
+  {
+    netio_pkt_metadata_t* mda = NETIO_PKT_METADATA(pkt);
+
+    return NETIO_PKT_L2_HEADER_LENGTH_M(mda, pkt);
+  }
+}
+
+
+/** Return the length of the packet, starting with the L2 (Ethernet) header.
+ * @ingroup pktfuncs
+ *
+ * @param[in] pkt Packet on which to operate.
+ * @return  The length of the packet, in bytes.
+ */
+static __inline netio_size_t
+NETIO_PKT_L2_LENGTH(netio_pkt_t* pkt)
+{
+  if (NETIO_PKT_IS_MINIMAL(pkt))
+  {
+    netio_pkt_minimal_metadata_t* mmd = NETIO_PKT_MINIMAL_METADATA(pkt);
+
+    return NETIO_PKT_L2_LENGTH_MM(mmd, pkt);
+  }
+  else
+  {
+    netio_pkt_metadata_t* mda = NETIO_PKT_METADATA(pkt);
+
+    return NETIO_PKT_L2_LENGTH_M(mda, pkt);
+  }
+}
+
+
+/** Return a pointer to the packet's L2 (Ethernet) header.
+ * @ingroup pktfuncs
+ *
+ * @param[in] pkt Packet on which to operate.
+ * @return A pointer to start of the packet.
+ */
+static __inline unsigned char*
+NETIO_PKT_L2_DATA(netio_pkt_t* pkt)
+{
+  if (NETIO_PKT_IS_MINIMAL(pkt))
+  {
+    netio_pkt_minimal_metadata_t* mmd = NETIO_PKT_MINIMAL_METADATA(pkt);
+
+    return NETIO_PKT_L2_DATA_MM(mmd, pkt);
+  }
+  else
+  {
+    netio_pkt_metadata_t* mda = NETIO_PKT_METADATA(pkt);
+
+    return NETIO_PKT_L2_DATA_M(mda, pkt);
+  }
+}
+
+
+/** Retrieve the length of the packet, starting with the L3 (generally, the IP)
+ * header.
+ * @ingroup pktfuncs
+ *
+ * @param[in] pkt Packet on which to operate.
+ * @return Length of the packet's L3 header and data, in bytes.
+ */
+static __inline netio_size_t
+NETIO_PKT_L3_LENGTH(netio_pkt_t* pkt)
+{
+  if (NETIO_PKT_IS_MINIMAL(pkt))
+  {
+    netio_pkt_minimal_metadata_t* mmd = NETIO_PKT_MINIMAL_METADATA(pkt);
+
+    return NETIO_PKT_L3_LENGTH_MM(mmd, pkt);
+  }
+  else
+  {
+    netio_pkt_metadata_t* mda = NETIO_PKT_METADATA(pkt);
+
+    return NETIO_PKT_L3_LENGTH_M(mda, pkt);
+  }
+}
+
+
+/** Return a pointer to the packet's L3 (generally, the IP) header.
+ * @ingroup pktfuncs
+ *
+ * Note that we guarantee word alignment of the L3 header.
+ *
+ * @param[in] pkt Packet on which to operate.
+ * @return A pointer to the packet's L3 header.
+ */
+static __inline unsigned char*
+NETIO_PKT_L3_DATA(netio_pkt_t* pkt)
+{
+  if (NETIO_PKT_IS_MINIMAL(pkt))
+  {
+    netio_pkt_minimal_metadata_t* mmd = NETIO_PKT_MINIMAL_METADATA(pkt);
+
+    return NETIO_PKT_L3_DATA_MM(mmd, pkt);
+  }
+  else
+  {
+    netio_pkt_metadata_t* mda = NETIO_PKT_METADATA(pkt);
+
+    return NETIO_PKT_L3_DATA_M(mda, pkt);
+  }
+}
+
+
+/** Return the ordinal of the packet.
+ * @ingroup ingress
+ *
+ * Each packet is given an ordinal number when it is delivered by the IPP.
+ * In the medium term, the ordinal is unique and monotonically increasing,
+ * being incremented by 1 for each packet; the ordinal of the first packet
+ * delivered after the IPP starts is zero.  (Since the ordinal is of finite
+ * size, given enough input packets, it will eventually wrap around to zero;
+ * in the long term, therefore, ordinals are not unique.)  The ordinals
+ * handed out by different IPPs are not disjoint, so two packets from
+ * different IPPs may have identical ordinals.  Packets dropped by the
+ * IPP or by the I/O shim are not assigned ordinals.
+ *
+ *
+ * @param[in] pkt Packet on which to operate.
+ * @return The packet's per-IPP packet ordinal.
+ */
+static __inline unsigned int
+NETIO_PKT_ORDINAL(netio_pkt_t* pkt)
+{
+  netio_pkt_metadata_t* mda = NETIO_PKT_METADATA(pkt);
+
+  return NETIO_PKT_ORDINAL_M(mda, pkt);
+}
+
+
+/** Return the per-group ordinal of the packet.
+ * @ingroup ingress
+ *
+ * Each packet is given a per-group ordinal number when it is
+ * delivered by the IPP. By default, the group is the packet's VLAN,
+ * although IPP can be recompiled to use different values.  In
+ * the medium term, the ordinal is unique and monotonically
+ * increasing, being incremented by 1 for each packet; the ordinal of
+ * the first packet distributed to a particular group is zero.
+ * (Since the ordinal is of finite size, given enough input packets,
+ * it will eventually wrap around to zero; in the long term,
+ * therefore, ordinals are not unique.)  The ordinals handed out by
+ * different IPPs are not disjoint, so two packets from different IPPs
+ * may have identical ordinals; similarly, packets distributed to
+ * different groups may have identical ordinals.  Packets dropped by
+ * the IPP or by the I/O shim are not assigned ordinals.
+ *
+ * @param[in] pkt Packet on which to operate.
+ * @return The packet's per-IPP, per-group ordinal.
+ */
+static __inline unsigned int
+NETIO_PKT_GROUP_ORDINAL(netio_pkt_t* pkt)
+{
+  netio_pkt_metadata_t* mda = NETIO_PKT_METADATA(pkt);
+
+  return NETIO_PKT_GROUP_ORDINAL_M(mda, pkt);
+}
+
+
+/** Return the VLAN ID assigned to the packet.
+ * @ingroup ingress
+ *
+ * This is usually also contained within the packet header.  If the packet
+ * does not have a VLAN tag, the VLAN ID returned by this function is zero.
+ *
+ * @param[in] pkt Packet on which to operate.
+ * @return The packet's VLAN ID.
+ */
+static __inline unsigned short
+NETIO_PKT_VLAN_ID(netio_pkt_t* pkt)
+{
+  netio_pkt_metadata_t* mda = NETIO_PKT_METADATA(pkt);
+
+  return NETIO_PKT_VLAN_ID_M(mda, pkt);
+}
+
+
+/** Return the ethertype of the packet.
+ * @ingroup ingress
+ *
+ * This value is reliable if @ref NETIO_PKT_ETHERTYPE_RECOGNIZED()
+ * returns true, and otherwise, may not be well defined.
+ *
+ * @param[in] pkt Packet on which to operate.
+ * @return The packet's ethertype.
+ */
+static __inline unsigned short
+NETIO_PKT_ETHERTYPE(netio_pkt_t* pkt)
+{
+  netio_pkt_metadata_t* mda = NETIO_PKT_METADATA(pkt);
+
+  return NETIO_PKT_ETHERTYPE_M(mda, pkt);
+}
+
+
+/** Return the flow hash computed on the packet.
+ * @ingroup ingress
+ *
+ * For TCP and UDP packets, this hash is calculated by hashing together
+ * the "5-tuple" values, specifically the source IP address, destination
+ * IP address, protocol type, source port and destination port.
+ * The hash value is intended to be helpful for millions of distinct
+ * flows.
+ *
+ * For IPv4 or IPv6 packets which are neither TCP nor UDP, the flow hash is
+ * derived by hashing together the source and destination IP addresses.
+ *
+ * For MPLS-encapsulated packets, the flow hash is derived by hashing
+ * the first MPLS label.
+ *
+ * For all other packets the flow hash is computed from the source
+ * and destination Ethernet addresses.
+ *
+ * The hash is symmetric, meaning it produces the same value if the
+ * source and destination are swapped. The only exceptions are
+ * tunneling protocols 0x04 (IP in IP Encapsulation), 0x29 (Simple
+ * Internet Protocol), 0x2F (General Routing Encapsulation) and 0x32
+ * (Encap Security Payload), which use only the destination address
+ * since the source address is not meaningful.
+ *
+ * @param[in] pkt Packet on which to operate.
+ * @return The packet's 32-bit flow hash.
+ */
+static __inline unsigned int
+NETIO_PKT_FLOW_HASH(netio_pkt_t* pkt)
+{
+  netio_pkt_metadata_t* mda = NETIO_PKT_METADATA(pkt);
+
+  return NETIO_PKT_FLOW_HASH_M(mda, pkt);
+}
+
+
+/** Return the first word of "user data" for the packet.
+ *
+ * The contents of the user data words depend on the IPP.
+ *
+ * When using the standard ipp1, ipp2, or ipp4 sub-drivers, the first
+ * word of user data contains the least significant bits of the 64-bit
+ * arrival cycle count (see @c get_cycle_count_low()).
+ *
+ * See the <em>System Programmer's Guide</em> for details.
+ *
+ * @ingroup ingress
+ *
+ * @param[in] pkt Packet on which to operate.
+ * @return The packet's first word of "user data".
+ */
+static __inline unsigned int
+NETIO_PKT_USER_DATA_0(netio_pkt_t* pkt)
+{
+  netio_pkt_metadata_t* mda = NETIO_PKT_METADATA(pkt);
+
+  return NETIO_PKT_USER_DATA_0_M(mda, pkt);
+}
+
+
+/** Return the second word of "user data" for the packet.
+ *
+ * The contents of the user data words depend on the IPP.
+ *
+ * When using the standard ipp1, ipp2, or ipp4 sub-drivers, the second
+ * word of user data contains the most significant bits of the 64-bit
+ * arrival cycle count (see @c get_cycle_count_high()).
+ *
+ * See the <em>System Programmer's Guide</em> for details.
+ *
+ * @ingroup ingress
+ *
+ * @param[in] pkt Packet on which to operate.
+ * @return The packet's second word of "user data".
+ */
+static __inline unsigned int
+NETIO_PKT_USER_DATA_1(netio_pkt_t* pkt)
+{
+  netio_pkt_metadata_t* mda = NETIO_PKT_METADATA(pkt);
+
+  return NETIO_PKT_USER_DATA_1_M(mda, pkt);
+}
+
+
+/** Determine whether the L4 (TCP/UDP) checksum was calculated.
+ * @ingroup ingress
+ *
+ * @param[in] pkt Packet on which to operate.
+ * @return Nonzero if the L4 checksum was calculated.
+ */
+static __inline unsigned int
+NETIO_PKT_L4_CSUM_CALCULATED(netio_pkt_t* pkt)
+{
+  netio_pkt_metadata_t* mda = NETIO_PKT_METADATA(pkt);
+
+  return NETIO_PKT_L4_CSUM_CALCULATED_M(mda, pkt);
+}
+
+
+/** Determine whether the L4 (TCP/UDP) checksum was calculated and found to
+ *  be correct.
+ * @ingroup ingress
+ *
+ * @param[in] pkt Packet on which to operate.
+ * @return Nonzero if the checksum was calculated and is correct.
+ */
+static __inline unsigned int
+NETIO_PKT_L4_CSUM_CORRECT(netio_pkt_t* pkt)
+{
+  netio_pkt_metadata_t* mda = NETIO_PKT_METADATA(pkt);
+
+  return NETIO_PKT_L4_CSUM_CORRECT_M(mda, pkt);
+}
+
+
+/** Determine whether the L3 (IP) checksum was calculated.
+ * @ingroup ingress
+ *
+ * @param[in] pkt Packet on which to operate.
+ * @return Nonzero if the L3 (IP) checksum was calculated.
+*/
+static __inline unsigned int
+NETIO_PKT_L3_CSUM_CALCULATED(netio_pkt_t* pkt)
+{
+  netio_pkt_metadata_t* mda = NETIO_PKT_METADATA(pkt);
+
+  return NETIO_PKT_L3_CSUM_CALCULATED_M(mda, pkt);
+}
+
+
+/** Determine whether the L3 (IP) checksum was calculated and found to be
+ *  correct.
+ * @ingroup ingress
+ *
+ * @param[in] pkt Packet on which to operate.
+ * @return Nonzero if the checksum was calculated and is correct.
+ */
+static __inline unsigned int
+NETIO_PKT_L3_CSUM_CORRECT(netio_pkt_t* pkt)
+{
+  netio_pkt_metadata_t* mda = NETIO_PKT_METADATA(pkt);
+
+  return NETIO_PKT_L3_CSUM_CORRECT_M(mda, pkt);
+}
+
+
+/** Determine whether the Ethertype was recognized and L3 packet data was
+ *  processed.
+ * @ingroup ingress
+ *
+ * @param[in] pkt Packet on which to operate.
+ * @return Nonzero if the Ethertype was recognized and L3 packet data was
+ *   processed.
+ */
+static __inline unsigned int
+NETIO_PKT_ETHERTYPE_RECOGNIZED(netio_pkt_t* pkt)
+{
+  netio_pkt_metadata_t* mda = NETIO_PKT_METADATA(pkt);
+
+  return NETIO_PKT_ETHERTYPE_RECOGNIZED_M(mda, pkt);
+}
+
+
+/** Set an egress packet's L2 length, using a metadata pointer to speed the
+ * computation.
+ * @ingroup egress
+ *
+ * @param[in,out] mmd Pointer to packet's minimal metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @param[in] len Packet L2 length, in bytes.
+ */
+static __inline void
+NETIO_PKT_SET_L2_LENGTH_MM(netio_pkt_minimal_metadata_t* mmd, netio_pkt_t* pkt,
+                           int len)
+{
+  mmd->l2_length = len;
+}
+
+
+/** Set an egress packet's L2 length.
+ * @ingroup egress
+ *
+ * @param[in,out] pkt Packet on which to operate.
+ * @param[in] len Packet L2 length, in bytes.
+ */
+static __inline void
+NETIO_PKT_SET_L2_LENGTH(netio_pkt_t* pkt, int len)
+{
+  netio_pkt_minimal_metadata_t* mmd = NETIO_PKT_MINIMAL_METADATA(pkt);
+
+  NETIO_PKT_SET_L2_LENGTH_MM(mmd, pkt, len);
+}
+
+
+/** Set an egress packet's L2 header length, using a metadata pointer to
+ *  speed the computation.
+ * @ingroup egress
+ *
+ * It is not normally necessary to call this routine; only the L2 length,
+ * not the header length, is needed to transmit a packet.  It may be useful if
+ * the egress packet will later be processed by code which expects to use
+ * functions like @ref NETIO_PKT_L3_DATA() to get a pointer to the L3 payload.
+ *
+ * @param[in,out] mmd Pointer to packet's minimal metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @param[in] len Packet L2 header length, in bytes.
+ */
+static __inline void
+NETIO_PKT_SET_L2_HEADER_LENGTH_MM(netio_pkt_minimal_metadata_t* mmd,
+                                  netio_pkt_t* pkt, int len)
+{
+  mmd->l3_offset = mmd->l2_offset + len;
+}
+
+
+/** Set an egress packet's L2 header length.
+ * @ingroup egress
+ *
+ * It is not normally necessary to call this routine; only the L2 length,
+ * not the header length, is needed to transmit a packet.  It may be useful if
+ * the egress packet will later be processed by code which expects to use
+ * functions like @ref NETIO_PKT_L3_DATA() to get a pointer to the L3 payload.
+ *
+ * @param[in,out] pkt Packet on which to operate.
+ * @param[in] len Packet L2 header length, in bytes.
+ */
+static __inline void
+NETIO_PKT_SET_L2_HEADER_LENGTH(netio_pkt_t* pkt, int len)
+{
+  netio_pkt_minimal_metadata_t* mmd = NETIO_PKT_MINIMAL_METADATA(pkt);
+
+  NETIO_PKT_SET_L2_HEADER_LENGTH_MM(mmd, pkt, len);
+}
+
+
+/** Set up an egress packet for hardware checksum computation, using a
+ *  metadata pointer to speed the operation.
+ * @ingroup egress
+ *
+ *  NetIO provides the ability to automatically calculate a standard
+ *  16-bit Internet checksum on transmitted packets.  The application
+ *  may specify the point in the packet where the checksum starts, the
+ *  number of bytes to be checksummed, and the two bytes in the packet
+ *  which will be replaced with the completed checksum.  (If the range
+ *  of bytes to be checksummed includes the bytes to be replaced, the
+ *  initial values of those bytes will be included in the checksum.)
+ *
+ *  For some protocols, the packet checksum covers data which is not present
+ *  in the packet, or is at least not contiguous to the main data payload.
+ *  For instance, the TCP checksum includes a "pseudo-header" which includes
+ *  the source and destination IP addresses of the packet.  To accommodate
+ *  this, the checksum engine may be "seeded" with an initial value, which
+ *  the application would need to compute based on the specific protocol's
+ *  requirements.  Note that the seed is given in host byte order (little-
+ *  endian), not network byte order (big-endian); code written to compute a
+ *  pseudo-header checksum in network byte order will need to byte-swap it
+ *  before use as the seed.
+ *
+ *  Note that the checksum is computed as part of the transmission process,
+ *  so it will not be present in the packet upon completion of this routine.
+ *
+ * @param[in,out] mmd Pointer to packet's minimal metadata.
+ * @param[in] pkt Packet on which to operate.
+ * @param[in] start Offset within L2 packet of the first byte to include in
+ *   the checksum.
+ * @param[in] length Number of bytes to include in the checksum.
+ *   the checksum.
+ * @param[in] location Offset within L2 packet of the first of the two bytes
+ *   to be replaced with the calculated checksum.
+ * @param[in] seed Initial value of the running checksum before any of the
+ *   packet data is added.
+ */
+static __inline void
+NETIO_PKT_DO_EGRESS_CSUM_MM(netio_pkt_minimal_metadata_t* mmd,
+                            netio_pkt_t* pkt, int start, int length,
+                            int location, uint16_t seed)
+{
+  mmd->csum_start = start;
+  mmd->csum_length = length;
+  mmd->csum_location = location;
+  mmd->csum_seed = seed;
+  mmd->flags |= _NETIO_PKT_NEED_EDMA_CSUM_MASK;
+}
+
+
+/** Set up an egress packet for hardware checksum computation.
+ * @ingroup egress
+ *
+ *  NetIO provides the ability to automatically calculate a standard
+ *  16-bit Internet checksum on transmitted packets.  The application
+ *  may specify the point in the packet where the checksum starts, the
+ *  number of bytes to be checksummed, and the two bytes in the packet
+ *  which will be replaced with the completed checksum.  (If the range
+ *  of bytes to be checksummed includes the bytes to be replaced, the
+ *  initial values of those bytes will be included in the checksum.)
+ *
+ *  For some protocols, the packet checksum covers data which is not present
+ *  in the packet, or is at least not contiguous to the main data payload.
+ *  For instance, the TCP checksum includes a "pseudo-header" which includes
+ *  the source and destination IP addresses of the packet.  To accommodate
+ *  this, the checksum engine may be "seeded" with an initial value, which
+ *  the application would need to compute based on the specific protocol's
+ *  requirements.  Note that the seed is given in host byte order (little-
+ *  endian), not network byte order (big-endian); code written to compute a
+ *  pseudo-header checksum in network byte order will need to byte-swap it
+ *  before use as the seed.
+ *
+ *  Note that the checksum is computed as part of the transmission process,
+ *  so it will not be present in the packet upon completion of this routine.
+ *
+ * @param[in,out] pkt Packet on which to operate.
+ * @param[in] start Offset within L2 packet of the first byte to include in
+ *   the checksum.
+ * @param[in] length Number of bytes to include in the checksum.
+ *   the checksum.
+ * @param[in] location Offset within L2 packet of the first of the two bytes
+ *   to be replaced with the calculated checksum.
+ * @param[in] seed Initial value of the running checksum before any of the
+ *   packet data is added.
+ */
+static __inline void
+NETIO_PKT_DO_EGRESS_CSUM(netio_pkt_t* pkt, int start, int length,
+                         int location, uint16_t seed)
+{
+  netio_pkt_minimal_metadata_t* mmd = NETIO_PKT_MINIMAL_METADATA(pkt);
+
+  NETIO_PKT_DO_EGRESS_CSUM_MM(mmd, pkt, start, length, location, seed);
+}
+
+
+/** Return the number of bytes which could be prepended to a packet, using a
+ *  metadata pointer to speed the operation.
+ *  See @ref netio_populate_prepend_buffer() to get a full description of
+ *  prepending.
+ *
+ * @param[in,out] mda Pointer to packet's standard metadata.
+ * @param[in] pkt Packet on which to operate.
+ */
+static __inline int
+NETIO_PKT_PREPEND_AVAIL_M(netio_pkt_metadata_t* mda, netio_pkt_t* pkt)
+{
+  return (pkt->__packet.bits.__offset << 6) +
+         NETIO_PKT_CUSTOM_HEADER_LENGTH_M(mda, pkt);
+}
+
+
+/** Return the number of bytes which could be prepended to a packet, using a
+ *  metadata pointer to speed the operation.
+ *  See @ref netio_populate_prepend_buffer() to get a full description of
+ *  prepending.
+ * @ingroup egress
+ *
+ * @param[in,out] mmd Pointer to packet's minimal metadata.
+ * @param[in] pkt Packet on which to operate.
+ */
+static __inline int
+NETIO_PKT_PREPEND_AVAIL_MM(netio_pkt_minimal_metadata_t* mmd, netio_pkt_t* pkt)
+{
+  return (pkt->__packet.bits.__offset << 6) + mmd->l2_offset;
+}
+
+
+/** Return the number of bytes which could be prepended to a packet.
+ *  See @ref netio_populate_prepend_buffer() to get a full description of
+ *  prepending.
+ * @ingroup egress
+ *
+ * @param[in] pkt Packet on which to operate.
+ */
+static __inline int
+NETIO_PKT_PREPEND_AVAIL(netio_pkt_t* pkt)
+{
+  if (NETIO_PKT_IS_MINIMAL(pkt))
+  {
+    netio_pkt_minimal_metadata_t* mmd = NETIO_PKT_MINIMAL_METADATA(pkt);
+
+    return NETIO_PKT_PREPEND_AVAIL_MM(mmd, pkt);
+  }
+  else
+  {
+    netio_pkt_metadata_t* mda = NETIO_PKT_METADATA(pkt);
+
+    return NETIO_PKT_PREPEND_AVAIL_M(mda, pkt);
+  }
+}
+
+
+/** Flush a packet's minimal metadata from the cache, using a metadata pointer
+ *  to speed the operation.
+ * @ingroup egress
+ *
+ * @param[in] mmd Pointer to packet's minimal metadata.
+ * @param[in] pkt Packet on which to operate.
+ */
+static __inline void
+NETIO_PKT_FLUSH_MINIMAL_METADATA_MM(netio_pkt_minimal_metadata_t* mmd,
+                                    netio_pkt_t* pkt)
+{
+}
+
+
+/** Invalidate a packet's minimal metadata from the cache, using a metadata
+ *  pointer to speed the operation.
+ * @ingroup egress
+ *
+ * @param[in] mmd Pointer to packet's minimal metadata.
+ * @param[in] pkt Packet on which to operate.
+ */
+static __inline void
+NETIO_PKT_INV_MINIMAL_METADATA_MM(netio_pkt_minimal_metadata_t* mmd,
+                                  netio_pkt_t* pkt)
+{
+}
+
+
+/** Flush and then invalidate a packet's minimal metadata from the cache,
+ *  using a metadata pointer to speed the operation.
+ * @ingroup egress
+ *
+ * @param[in] mmd Pointer to packet's minimal metadata.
+ * @param[in] pkt Packet on which to operate.
+ */
+static __inline void
+NETIO_PKT_FLUSH_INV_MINIMAL_METADATA_MM(netio_pkt_minimal_metadata_t* mmd,
+                                        netio_pkt_t* pkt)
+{
+}
+
+
+/** Flush a packet's metadata from the cache, using a metadata pointer
+ *  to speed the operation.
+ * @ingroup ingress
+ *
+ * @param[in] mda Pointer to packet's minimal metadata.
+ * @param[in] pkt Packet on which to operate.
+ */
+static __inline void
+NETIO_PKT_FLUSH_METADATA_M(netio_pkt_metadata_t* mda, netio_pkt_t* pkt)
+{
+}
+
+
+/** Invalidate a packet's metadata from the cache, using a metadata
+ *  pointer to speed the operation.
+ * @ingroup ingress
+ *
+ * @param[in] mda Pointer to packet's metadata.
+ * @param[in] pkt Packet on which to operate.
+ */
+static __inline void
+NETIO_PKT_INV_METADATA_M(netio_pkt_metadata_t* mda, netio_pkt_t* pkt)
+{
+}
+
+
+/** Flush and then invalidate a packet's metadata from the cache,
+ *  using a metadata pointer to speed the operation.
+ * @ingroup ingress
+ *
+ * @param[in] mda Pointer to packet's metadata.
+ * @param[in] pkt Packet on which to operate.
+ */
+static __inline void
+NETIO_PKT_FLUSH_INV_METADATA_M(netio_pkt_metadata_t* mda, netio_pkt_t* pkt)
+{
+}
+
+
+/** Flush a packet's minimal metadata from the cache.
+ * @ingroup egress
+ *
+ * @param[in] pkt Packet on which to operate.
+ */
+static __inline void
+NETIO_PKT_FLUSH_MINIMAL_METADATA(netio_pkt_t* pkt)
+{
+}
+
+
+/** Invalidate a packet's minimal metadata from the cache.
+ * @ingroup egress
+ *
+ * @param[in] pkt Packet on which to operate.
+ */
+static __inline void
+NETIO_PKT_INV_MINIMAL_METADATA(netio_pkt_t* pkt)
+{
+}
+
+
+/** Flush and then invalidate a packet's minimal metadata from the cache.
+ * @ingroup egress
+ *
+ * @param[in] pkt Packet on which to operate.
+ */
+static __inline void
+NETIO_PKT_FLUSH_INV_MINIMAL_METADATA(netio_pkt_t* pkt)
+{
+}
+
+
+/** Flush a packet's metadata from the cache.
+ * @ingroup ingress
+ *
+ * @param[in] pkt Packet on which to operate.
+ */
+static __inline void
+NETIO_PKT_FLUSH_METADATA(netio_pkt_t* pkt)
+{
+}
+
+
+/** Invalidate a packet's metadata from the cache.
+ * @ingroup ingress
+ *
+ * @param[in] pkt Packet on which to operate.
+ */
+static __inline void
+NETIO_PKT_INV_METADATA(netio_pkt_t* pkt)
+{
+}
+
+
+/** Flush and then invalidate a packet's metadata from the cache.
+ * @ingroup ingress
+ *
+ * @param[in] pkt Packet on which to operate.
+ */
+static __inline void
+NETIO_PKT_FLUSH_INV_METADATA(netio_pkt_t* pkt)
+{
+}
+
+/** Number of NUMA nodes we can distribute buffers to.
+ * @ingroup setup */
+#define NETIO_NUM_NODE_WEIGHTS  16
+
+/**
+ * @brief An object for specifying the characteristics of NetIO communication
+ * endpoint.
+ *
+ * @ingroup setup
+ *
+ * The @ref netio_input_register() function uses this structure to define
+ * how an application tile will communicate with an IPP.
+ *
+ *
+ * Future updates to NetIO may add new members to this structure,
+ * which can affect the success of the registration operation.  Thus,
+ * if dynamically initializing the structure, applications are urged to
+ * zero it out first, for example:
+ *
+ * @code
+ * netio_input_config_t config;
+ * memset(&config, 0, sizeof (config));
+ * config.flags = NETIO_RECV | NETIO_XMIT_CSUM | NETIO_TAG_NONE;
+ * config.num_receive_packets = NETIO_MAX_RECEIVE_PKTS;
+ * config.queue_id = 0;
+ *     .
+ *     .
+ *     .
+ * @endcode
+ *
+ * since that guarantees that any unused structure members, including
+ * members which did not exist when the application was first developed,
+ * will not have unexpected values.
+ *
+ * If statically initializing the structure, we strongly recommend use of
+ * C99-style named initializers, for example:
+ *
+ * @code
+ * netio_input_config_t config = {
+ *    .flags = NETIO_RECV | NETIO_XMIT_CSUM | NETIO_TAG_NONE,
+ *    .num_receive_packets = NETIO_MAX_RECEIVE_PKTS,
+ *    .queue_id = 0,
+ * },
+ * @endcode
+ *
+ * instead of the old-style structure initialization:
+ *
+ * @code
+ * // Bad example! Currently equivalent to the above, but don't do this.
+ * netio_input_config_t config = {
+ *    NETIO_RECV | NETIO_XMIT_CSUM | NETIO_TAG_NONE, NETIO_MAX_RECEIVE_PKTS, 0
+ * },
+ * @endcode
+ *
+ * since the C99 style requires no changes to the code if elements of the
+ * config structure are rearranged.  (It also makes the initialization much
+ * easier to understand.)
+ *
+ * Except for items which address a particular tile's transmit or receive
+ * characteristics, such as the ::NETIO_RECV flag, applications are advised
+ * to specify the same set of configuration data on all registrations.
+ * This prevents differing results if multiple tiles happen to do their
+ * registration operations in a different order on different invocations of
+ * the application.  This is particularly important for things like link
+ * management flags, and buffer size and homing specifications.
+ *
+ * Unless the ::NETIO_FIXED_BUFFER_VA flag is specified in flags, the NetIO
+ * buffer pool is automatically created and mapped into the application's
+ * virtual address space at an address chosen by the operating system,
+ * using the common memory (cmem) facility in the Tilera Multicore
+ * Components library.  The cmem facility allows multiple processes to gain
+ * access to shared memory which is mapped into each process at an
+ * identical virtual address.  In order for this to work, the processes
+ * must have a common ancestor, which must create the common memory using
+ * tmc_cmem_init().
+ *
+ * In programs using the iLib process creation API, or in programs which use
+ * only one process (which include programs using the pthreads library),
+ * tmc_cmem_init() is called automatically.  All other applications
+ * must call it explicitly, before any child processes which might call
+ * netio_input_register() are created.
+ */
+typedef struct
+{
+  /** Registration characteristics.
+
+      This value determines several characteristics of the registration;
+      flags for different types of behavior are ORed together to make the
+      final flag value.  Generally applications should specify exactly
+      one flag from each of the following categories:
+
+      - Whether the application will be receiving packets on this queue
+        (::NETIO_RECV or ::NETIO_NO_RECV).
+
+      - Whether the application will be transmitting packets on this queue,
+        and if so, whether it will request egress checksum calculation
+        (::NETIO_XMIT, ::NETIO_XMIT_CSUM, or ::NETIO_NO_XMIT).  It is
+        legal to call netio_get_buffer() without one of the XMIT flags,
+        as long as ::NETIO_RECV is specified; in this case, the retrieved
+        buffers must be passed to another tile for transmission.
+
+      - Whether the application expects any vendor-specific tags in
+        its packets' L2 headers (::NETIO_TAG_NONE, ::NETIO_TAG_BRCM,
+        or ::NETIO_TAG_MRVL).  This must match the configuration of the
+        target IPP.
+
+      To accommodate applications written to previous versions of the NetIO
+      interface, none of the flags above are currently required; if omitted,
+      NetIO behaves more or less as if ::NETIO_RECV | ::NETIO_XMIT_CSUM |
+      ::NETIO_TAG_NONE were used.  However, explicit specification of
+      the relevant flags allows NetIO to do a better job of resource
+      allocation, allows earlier detection of certain configuration errors,
+      and may enable advanced features or higher performance in the future,
+      so their use is strongly recommended.
+
+      Note that specifying ::NETIO_NO_RECV along with ::NETIO_NO_XMIT
+      is a special case, intended primarily for use by programs which
+      retrieve network statistics or do link management operations.
+      When these flags are both specified, the resulting queue may not
+      be used with NetIO routines other than netio_get(), netio_set(),
+      and netio_input_unregister().  See @ref link for more information
+      on link management.
+
+      Other flags are optional; their use is described below.
+  */
+  int flags;
+
+  /** Interface name.  This is a string which identifies the specific
+      Ethernet controller hardware to be used.  The format of the string
+      is a device type and a device index, separated by a slash; so,
+      the first 10 Gigabit Ethernet controller is named "xgbe/0", while
+      the second 10/100/1000 Megabit Ethernet controller is named "gbe/1".
+   */
+  const char* interface;
+
+  /** Receive packet queue size.  This specifies the maximum number
+      of ingress packets that can be received on this queue without
+      being retrieved by @ref netio_get_packet().  If the IPP's distribution
+      algorithm calls for a packet to be sent to this queue, and this
+      number of packets are already pending there, the new packet
+      will either be discarded, or sent to another tile registered
+      for the same queue_id (see @ref drops).  This value must
+      be at least ::NETIO_MIN_RECEIVE_PKTS, can always be at least
+      ::NETIO_MAX_RECEIVE_PKTS, and may be larger than that on certain
+      interfaces.
+   */
+  int num_receive_packets;
+
+  /** The queue ID being requested.  Legal values for this range from 0
+      to ::NETIO_MAX_QUEUE_ID, inclusive.  ::NETIO_MAX_QUEUE_ID is always
+      greater than or equal to the number of tiles; this allows one queue
+      for each tile, plus at least one additional queue.  Some applications
+      may wish to use the additional queue as a destination for unwanted
+      packets, since packets delivered to queues for which no tiles have
+      registered are discarded.
+   */
+  unsigned int queue_id;
+
+  /** Maximum number of small send buffers to be held in the local empty
+      buffer cache.  This specifies the size of the area which holds
+      empty small egress buffers requested from the IPP but not yet
+      retrieved via @ref netio_get_buffer().  This value must be greater
+      than zero if the application will ever use @ref netio_get_buffer()
+      to allocate empty small egress buffers; it may be no larger than
+      ::NETIO_MAX_SEND_BUFFERS.  See @ref epp for more details on empty
+      buffer caching.
+   */
+  int num_send_buffers_small_total;
+
+  /** Number of small send buffers to be preallocated at registration.
+      If this value is nonzero, the specified number of empty small egress
+      buffers will be requested from the IPP during the netio_input_register
+      operation; this may speed the execution of @ref netio_get_buffer().
+      This may be no larger than @ref num_send_buffers_small_total.  See @ref
+      epp for more details on empty buffer caching.
+   */
+  int num_send_buffers_small_prealloc;
+
+  /** Maximum number of large send buffers to be held in the local empty
+      buffer cache.  This specifies the size of the area which holds empty
+      large egress buffers requested from the IPP but not yet retrieved via
+      @ref netio_get_buffer().  This value must be greater than zero if the
+      application will ever use @ref netio_get_buffer() to allocate empty
+      large egress buffers; it may be no larger than ::NETIO_MAX_SEND_BUFFERS.
+      See @ref epp for more details on empty buffer caching.
+   */
+  int num_send_buffers_large_total;
+
+  /** Number of large send buffers to be preallocated at registration.
+      If this value is nonzero, the specified number of empty large egress
+      buffers will be requested from the IPP during the netio_input_register
+      operation; this may speed the execution of @ref netio_get_buffer().
+      This may be no larger than @ref num_send_buffers_large_total.  See @ref
+      epp for more details on empty buffer caching.
+   */
+  int num_send_buffers_large_prealloc;
+
+  /** Maximum number of jumbo send buffers to be held in the local empty
+      buffer cache.  This specifies the size of the area which holds empty
+      jumbo egress buffers requested from the IPP but not yet retrieved via
+      @ref netio_get_buffer().  This value must be greater than zero if the
+      application will ever use @ref netio_get_buffer() to allocate empty
+      jumbo egress buffers; it may be no larger than ::NETIO_MAX_SEND_BUFFERS.
+      See @ref epp for more details on empty buffer caching.
+   */
+  int num_send_buffers_jumbo_total;
+
+  /** Number of jumbo send buffers to be preallocated at registration.
+      If this value is nonzero, the specified number of empty jumbo egress
+      buffers will be requested from the IPP during the netio_input_register
+      operation; this may speed the execution of @ref netio_get_buffer().
+      This may be no larger than @ref num_send_buffers_jumbo_total.  See @ref
+      epp for more details on empty buffer caching.
+   */
+  int num_send_buffers_jumbo_prealloc;
+
+  /** Total packet buffer size.  This determines the total size, in bytes,
+      of the NetIO buffer pool.  Note that the maximum number of available
+      buffers of each size is determined during hypervisor configuration
+      (see the <em>System Programmer's Guide</em> for details); this just
+      influences how much host memory is allocated for those buffers.
+
+      The buffer pool is allocated from common memory, which will be
+      automatically initialized if needed.  If your buffer pool is larger
+      than 240 MB, you might need to explicitly call @c tmc_cmem_init(),
+      as described in the Application Libraries Reference Manual (UG227).
+
+      Packet buffers are currently allocated in chunks of 16 MB; this
+      value will be rounded up to the next larger multiple of 16 MB.
+      If this value is zero, a default of 32 MB will be used; this was
+      the value used by previous versions of NetIO.  Note that taking this
+      default also affects the placement of buffers on Linux NUMA nodes.
+      See @ref buffer_node_weights for an explanation of buffer placement.
+
+      In order to successfully allocate packet buffers, Linux must have
+      available huge pages on the relevant Linux NUMA nodes.  See the
+      <em>System Programmer's Guide</em> for information on configuring
+      huge page support in Linux.
+   */
+  uint64_t total_buffer_size;
+
+  /** Buffer placement weighting factors.
+
+      This array specifies the relative amount of buffering to place
+      on each of the available Linux NUMA nodes.  This array is
+      indexed by the NUMA node, and the values in the array are
+      proportional to the amount of buffer space to allocate on that
+      node.
+
+      If memory striping is enabled in the Hypervisor, then there is
+      only one logical NUMA node (node 0). In that case, NetIO will by
+      default ignore the suggested buffer node weights, and buffers
+      will be striped across the physical memory controllers. See
+      UG209 System Programmer's Guide for a description of the
+      hypervisor option that controls memory striping.
+
+      If memory striping is disabled, then there are up to four NUMA
+      nodes, corresponding to the four DDRAM controllers in the TILE
+      processor architecture.  See UG100 Tile Processor Architecture
+      Overview for a diagram showing the location of each of the DDRAM
+      controllers relative to the tile array.
+
+      For instance, if memory striping is disabled, the following
+      configuration strucure:
+
+      @code
+      netio_input_config_t config = {
+            .
+            .
+            .
+        .total_buffer_size = 4 * 16 * 1024 * 1024;
+        .buffer_node_weights = { 1, 0, 1, 0 },
+      },
+      @endcode
+
+      would result in 32 MB of buffers being placed on controller 0, and
+      32 MB on controller 2.  (Since buffers are allocated in units of
+      16 MB, some sets of weights will not be able to be matched exactly.)
+
+      For the weights to be effective, @ref total_buffer_size must be
+      nonzero.  If @ref total_buffer_size is zero, causing the default
+      32 MB of buffer space to be used, then any specified weights will
+      be ignored, and buffers will positioned as they were in previous
+      versions of NetIO:
+
+      - For xgbe/0 and gbe/0, 16 MB of buffers will be placed on controller 1,
+        and the other 16 MB will be placed on controller 2.
+
+      - For xgbe/1 and gbe/1, 16 MB of buffers will be placed on controller 2,
+        and the other 16 MB will be placed on controller 3.
+
+      If @ref total_buffer_size is nonzero, but all weights are zero,
+      then all buffer space will be allocated on Linux NUMA node zero.
+
+      By default, the specified buffer placement is treated as a hint;
+      if sufficient free memory is not available on the specified
+      controllers, the buffers will be allocated elsewhere.  However,
+      if the ::NETIO_STRICT_HOMING flag is specified in @ref flags, then a
+      failure to allocate buffer space exactly as requested will cause the
+      registration operation to fail with an error of ::NETIO_CANNOT_HOME.
+
+      Note that maximal network performance cannot be achieved with
+      only one memory controller.
+   */
+  uint8_t buffer_node_weights[NETIO_NUM_NODE_WEIGHTS];
+
+  /** Fixed virtual address for packet buffers.  Only valid when
+      ::NETIO_FIXED_BUFFER_VA is specified in @ref flags; see the
+      description of that flag for details.
+   */
+  void* fixed_buffer_va;
+
+  /**
+      Maximum number of outstanding send packet requests.  This value is
+      only relevant when an EPP is in use; it determines the number of
+      slots in the EPP's outgoing packet queue which this tile is allowed
+      to consume, and thus the number of packets which may be sent before
+      the sending tile must wait for an acknowledgment from the EPP.
+      Modifying this value is generally only helpful when using @ref
+      netio_send_packet_vector(), where it can help improve performance by
+      allowing a single vector send operation to process more packets.
+      Typically it is not specified, and the default, which divides the
+      outgoing packet slots evenly between all tiles on the chip, is used.
+
+      If a registration asks for more outgoing packet queue slots than are
+      available, ::NETIO_TOOMANY_XMIT will be returned.  The total number
+      of packet queue slots which are available for all tiles for each EPP
+      is subject to change, but is currently ::NETIO_TOTAL_SENDS_OUTSTANDING.
+
+
+      This value is ignored if ::NETIO_XMIT is not specified in flags.
+      If you want to specify a large value here for a specific tile, you are
+      advised to specify NETIO_NO_XMIT on other, non-transmitting tiles so
+      that they do not consume a default number of packet slots.  Any tile
+      transmitting is required to have at least ::NETIO_MIN_SENDS_OUTSTANDING
+      slots allocated to it; values less than that will be silently
+      increased by the NetIO library.
+   */
+  int num_sends_outstanding;
+}
+netio_input_config_t;
+
+
+/** Registration flags; used in the @ref netio_input_config_t structure.
+ * @addtogroup setup
+ */
+/** @{ */
+
+/** Fail a registration request if we can't put packet buffers
+    on the specified memory controllers. */
+#define NETIO_STRICT_HOMING   0x00000002
+
+/** This application expects no tags on its L2 headers. */
+#define NETIO_TAG_NONE        0x00000004
+
+/** This application expects Marvell extended tags on its L2 headers. */
+#define NETIO_TAG_MRVL        0x00000008
+
+/** This application expects Broadcom tags on its L2 headers. */
+#define NETIO_TAG_BRCM        0x00000010
+
+/** This registration may call routines which receive packets. */
+#define NETIO_RECV            0x00000020
+
+/** This registration may not call routines which receive packets. */
+#define NETIO_NO_RECV         0x00000040
+
+/** This registration may call routines which transmit packets. */
+#define NETIO_XMIT            0x00000080
+
+/** This registration may call routines which transmit packets with
+    checksum acceleration. */
+#define NETIO_XMIT_CSUM       0x00000100
+
+/** This registration may not call routines which transmit packets. */
+#define NETIO_NO_XMIT         0x00000200
+
+/** This registration wants NetIO buffers mapped at an application-specified
+    virtual address.
+
+    NetIO buffers are by default created by the TMC common memory facility,
+    which must be configured by a common ancestor of all processes sharing
+    a network interface.  When this flag is specified, NetIO buffers are
+    instead mapped at an address chosen by the application (and specified
+    in @ref netio_input_config_t::fixed_buffer_va).  This allows multiple
+    unrelated but cooperating processes to share a NetIO interface.
+    All processes sharing the same interface must specify this flag,
+    and all must specify the same fixed virtual address.
+
+    @ref netio_input_config_t::fixed_buffer_va must be a
+    multiple of 16 MB, and the packet buffers will occupy @ref
+    netio_input_config_t::total_buffer_size bytes of virtual address
+    space, beginning at that address.  If any of those virtual addresses
+    are currently occupied by other memory objects, like application or
+    shared library code or data, @ref netio_input_register() will return
+    ::NETIO_FAULT.  While it is impossible to provide a fixed_buffer_va
+    which will work for all applications, a good first guess might be to
+    use 0xb0000000 minus @ref netio_input_config_t::total_buffer_size.
+    If that fails, it might be helpful to consult the running application's
+    virtual address description file (/proc/<em>pid</em>/maps) to see
+    which regions of virtual address space are available.
+ */
+#define NETIO_FIXED_BUFFER_VA 0x00000400
+
+/** This registration call will not complete unless the network link
+    is up.  The process will wait several seconds for this to happen (the
+    precise interval is link-dependent), but if the link does not come up,
+    ::NETIO_LINK_DOWN will be returned.  This flag is the default if
+    ::NETIO_NOREQUIRE_LINK_UP is not specified.  Note that this flag by
+    itself does not request that the link be brought up; that can be done
+    with the ::NETIO_AUTO_LINK_UPDN or ::NETIO_AUTO_LINK_UP flags (the
+    latter is the default if no NETIO_AUTO_LINK_xxx flags are specified),
+    or by explicitly setting the link's desired state via netio_set().
+    If the link is not brought up by one of those methods, and this flag
+    is specified, the registration operation will return ::NETIO_LINK_DOWN.
+    This flag is ignored if it is specified along with ::NETIO_NO_XMIT and
+    ::NETIO_NO_RECV.  See @ref link for more information on link
+    management.
+ */
+#define NETIO_REQUIRE_LINK_UP    0x00000800
+
+/** This registration call will complete even if the network link is not up.
+    Whenever the link is not up, packets will not be sent or received:
+    netio_get_packet() will return ::NETIO_NOPKT once all queued packets
+    have been drained, and netio_send_packet() and similar routines will
+    return NETIO_QUEUE_FULL once the outgoing packet queue in the EPP
+    or the I/O shim is full.  See @ref link for more information on link
+    management.
+ */
+#define NETIO_NOREQUIRE_LINK_UP  0x00001000
+
+#ifndef __DOXYGEN__
+/*
+ * These are part of the implementation of the NETIO_AUTO_LINK_xxx flags,
+ * but should not be used directly by applications, and are thus not
+ * documented.
+ */
+#define _NETIO_AUTO_UP        0x00002000
+#define _NETIO_AUTO_DN        0x00004000
+#define _NETIO_AUTO_PRESENT   0x00008000
+#endif
+
+/** Set the desired state of the link to up, allowing any speeds which are
+    supported by the link hardware, as part of this registration operation.
+    Do not take down the link automatically.  This is the default if
+    no other NETIO_AUTO_LINK_xxx flags are specified.  This flag is ignored
+    if it is specified along with ::NETIO_NO_XMIT and ::NETIO_NO_RECV.
+    See @ref link for more information on link management.
+ */
+#define NETIO_AUTO_LINK_UP     (_NETIO_AUTO_PRESENT | _NETIO_AUTO_UP)
+
+/** Set the desired state of the link to up, allowing any speeds which are
+    supported by the link hardware, as part of this registration operation.
+    Set the desired state of the link to down the next time no tiles are
+    registered for packet reception or transmission.  This flag is ignored
+    if it is specified along with ::NETIO_NO_XMIT and ::NETIO_NO_RECV.
+    See @ref link for more information on link management.
+ */
+#define NETIO_AUTO_LINK_UPDN   (_NETIO_AUTO_PRESENT | _NETIO_AUTO_UP | \
+                                _NETIO_AUTO_DN)
+
+/** Set the desired state of the link to down the next time no tiles are
+    registered for packet reception or transmission.  This flag is ignored
+    if it is specified along with ::NETIO_NO_XMIT and ::NETIO_NO_RECV.
+    See @ref link for more information on link management.
+ */
+#define NETIO_AUTO_LINK_DN     (_NETIO_AUTO_PRESENT | _NETIO_AUTO_DN)
+
+/** Do not bring up the link automatically as part of this registration
+    operation.  Do not take down the link automatically.  This flag
+    is ignored if it is specified along with ::NETIO_NO_XMIT and
+    ::NETIO_NO_RECV.  See @ref link for more information on link management.
+  */
+#define NETIO_AUTO_LINK_NONE   _NETIO_AUTO_PRESENT
+
+
+/** Minimum number of receive packets. */
+#define NETIO_MIN_RECEIVE_PKTS            16
+
+/** Lower bound on the maximum number of receive packets; may be higher
+    than this on some interfaces. */
+#define NETIO_MAX_RECEIVE_PKTS           128
+
+/** Maximum number of send buffers, per packet size. */
+#define NETIO_MAX_SEND_BUFFERS            16
+
+/** Number of EPP queue slots, and thus outstanding sends, per EPP. */
+#define NETIO_TOTAL_SENDS_OUTSTANDING   2015
+
+/** Minimum number of EPP queue slots, and thus outstanding sends, per
+ *  transmitting tile. */
+#define NETIO_MIN_SENDS_OUTSTANDING       16
+
+
+/**@}*/
+
+#ifndef __DOXYGEN__
+
+/**
+ * An object for providing Ethernet packets to a process.
+ */
+struct __netio_queue_impl_t;
+
+/**
+ * An object for managing the user end of a NetIO queue.
+ */
+struct __netio_queue_user_impl_t;
+
+#endif /* !__DOXYGEN__ */
+
+
+/** A netio_queue_t describes a NetIO communications endpoint.
+ * @ingroup setup
+ */
+typedef struct
+{
+#ifdef __DOXYGEN__
+  uint8_t opaque[8];                 /**< This is an opaque structure. */
+#else
+  struct __netio_queue_impl_t* __system_part;    /**< The system part. */
+  struct __netio_queue_user_impl_t* __user_part; /**< The user part. */
+#ifdef _NETIO_PTHREAD
+  _netio_percpu_mutex_t lock;                    /**< Queue lock. */
+#endif
+#endif
+}
+netio_queue_t;
+
+
+/**
+ * @brief Packet send context.
+ *
+ * @ingroup egress
+ *
+ * Packet send context for use with netio_send_packet_prepare and _commit.
+ */
+typedef struct
+{
+#ifdef __DOXYGEN__
+  uint8_t opaque[44];   /**< This is an opaque structure. */
+#else
+  uint8_t flags;        /**< Defined below */
+  uint8_t datalen;      /**< Number of valid words pointed to by data. */
+  uint32_t request[9];  /**< Request to be sent to the EPP or shim.  Note
+                             that this is smaller than the 11-word maximum
+                             request size, since some constant values are
+                             not saved in the context. */
+  uint32_t *data;       /**< Data to be sent to the EPP or shim via IDN. */
+#endif
+}
+netio_send_pkt_context_t;
+
+
+#ifndef __DOXYGEN__
+#define SEND_PKT_CTX_USE_EPP   1  /**< We're sending to an EPP. */
+#define SEND_PKT_CTX_SEND_CSUM 2  /**< Request includes a checksum. */
+#endif
+
+/**
+ * @brief Packet vector entry.
+ *
+ * @ingroup egress
+ *
+ * This data structure is used with netio_send_packet_vector() to send multiple
+ * packets with one NetIO call.  The structure should be initialized by
+ * calling netio_pkt_vector_set(), rather than by setting the fields
+ * directly.
+ *
+ * This structure is guaranteed to be a power of two in size, no
+ * bigger than one L2 cache line, and to be aligned modulo its size.
+ */
+typedef struct
+#ifndef __DOXYGEN__
+__attribute__((aligned(8)))
+#endif
+{
+  /** Reserved for use by the user application.  When initialized with
+   *  the netio_set_pkt_vector_entry() function, this field is guaranteed
+   *  to be visible to readers only after all other fields are already
+   *  visible.  This way it can be used as a valid flag or generation
+   *  counter. */
+  uint8_t user_data;
+
+  /* Structure members below this point should not be accessed directly by
+   * applications, as they may change in the future. */
+
+  /** Low 8 bits of the packet address to send.  The high bits are
+   *  acquired from the 'handle' field. */
+  uint8_t buffer_address_low;
+
+  /** Number of bytes to transmit. */
+  uint16_t size;
+
+  /** The raw handle from a netio_pkt_t.  If this is NETIO_PKT_HANDLE_NONE,
+   *  this vector entry will be skipped and no packet will be transmitted. */
+  netio_pkt_handle_t handle;
+}
+netio_pkt_vector_entry_t;
+
+
+/**
+ * @brief Initialize fields in a packet vector entry.
+ *
+ * @ingroup egress
+ *
+ * @param[out] v Pointer to the vector entry to be initialized.
+ * @param[in] pkt Packet to be transmitted when the vector entry is passed to
+ *        netio_send_packet_vector().  Note that the packet's attributes
+ *        (e.g., its L2 offset and length) are captured at the time this
+ *        routine is called; subsequent changes in those attributes will not
+ *        be reflected in the packet which is actually transmitted.
+ *        Changes in the packet's contents, however, will be so reflected.
+ *        If this is NULL, no packet will be transmitted.
+ * @param[in] user_data User data to be set in the vector entry.
+ *        This function guarantees that the "user_data" field will become
+ *        visible to a reader only after all other fields have become visible.
+ *        This allows a structure in a ring buffer to be written and read
+ *        by a polling reader without any locks or other synchronization.
+ */
+static __inline void
+netio_pkt_vector_set(volatile netio_pkt_vector_entry_t* v, netio_pkt_t* pkt,
+                     uint8_t user_data)
+{
+  if (pkt)
+  {
+    if (NETIO_PKT_IS_MINIMAL(pkt))
+    {
+      netio_pkt_minimal_metadata_t* mmd =
+        (netio_pkt_minimal_metadata_t*) &pkt->__metadata;
+      v->buffer_address_low = (uintptr_t) NETIO_PKT_L2_DATA_MM(mmd, pkt) & 0xFF;
+      v->size = NETIO_PKT_L2_LENGTH_MM(mmd, pkt);
+    }
+    else
+    {
+      netio_pkt_metadata_t* mda = &pkt->__metadata;
+      v->buffer_address_low = (uintptr_t) NETIO_PKT_L2_DATA_M(mda, pkt) & 0xFF;
+      v->size = NETIO_PKT_L2_LENGTH_M(mda, pkt);
+    }
+    v->handle.word = pkt->__packet.word;
+  }
+  else
+  {
+    v->handle.word = 0;   /* Set handle to NETIO_PKT_HANDLE_NONE. */
+  }
+
+  __asm__("" : : : "memory");
+
+  v->user_data = user_data;
+}
+
+
+/**
+ * Flags and structures for @ref netio_get() and @ref netio_set().
+ * @ingroup config
+ */
+
+/** @{ */
+/** Parameter class; addr is a NETIO_PARAM_xxx value. */
+#define NETIO_PARAM       0
+/** Interface MAC address. This address is only valid with @ref netio_get().
+ *  The value is a 6-byte MAC address.  Depending upon the overall system
+ *  design, a MAC address may or may not be available for each interface. */
+#define NETIO_PARAM_MAC        0
+
+/** Determine whether to suspend output on the receipt of pause frames.
+ *  If the value is nonzero, the I/O shim will suspend output when a pause
+ *  frame is received.  If the value is zero, pause frames will be ignored. */
+#define NETIO_PARAM_PAUSE_IN   1
+
+/** Determine whether to send pause frames if the I/O shim packet FIFOs are
+ *  nearly full.  If the value is zero, pause frames are not sent.  If
+ *  the value is nonzero, it is the delay value which will be sent in any
+ *  pause frames which are output, in units of 512 bit times. */
+#define NETIO_PARAM_PAUSE_OUT  2
+
+/** Jumbo frame support.  The value is a 4-byte integer.  If the value is
+ *  nonzero, the MAC will accept frames of up to 10240 bytes.  If the value
+ *  is zero, the MAC will only accept frames of up to 1544 bytes. */
+#define NETIO_PARAM_JUMBO      3
+
+/** I/O shim's overflow statistics register.  The value is two 16-bit integers.
+ *  The first 16-bit value (or the low 16 bits, if the value is treated as a
+ *  32-bit number) is the count of packets which were completely dropped and
+ *  not delivered by the shim.  The second 16-bit value (or the high 16 bits,
+ *  if the value is treated as a 32-bit number) is the count of packets
+ *  which were truncated and thus only partially delivered by the shim.  This
+ *  register is automatically reset to zero after it has been read.
+ */
+#define NETIO_PARAM_OVERFLOW   4
+
+/** IPP statistics.  This address is only valid with @ref netio_get().  The
+ *  value is a netio_stat_t structure.  Unlike the I/O shim statistics, the
+ *  IPP statistics are not all reset to zero on read; see the description
+ *  of the netio_stat_t for details. */
+#define NETIO_PARAM_STAT 5
+
+/** Possible link state.  The value is a combination of "NETIO_LINK_xxx"
+ *  flags.  With @ref netio_get(), this will indicate which flags are
+ *  actually supported by the hardware.
+ *
+ *  For historical reasons, specifying this value to netio_set() will have
+ *  the same behavior as using ::NETIO_PARAM_LINK_CONFIG, but this usage is
+ *  discouraged.
+ */
+#define NETIO_PARAM_LINK_POSSIBLE_STATE 6
+
+/** Link configuration. The value is a combination of "NETIO_LINK_xxx" flags.
+ *  With @ref netio_set(), this will attempt to immediately bring up the
+ *  link using whichever of the requested flags are supported by the
+ *  hardware, or take down the link if the flags are zero; if this is
+ *  not possible, an error will be returned.  Many programs will want
+ *  to use ::NETIO_PARAM_LINK_DESIRED_STATE instead.
+ *
+ *  For historical reasons, specifying this value to netio_get() will
+ *  have the same behavior as using ::NETIO_PARAM_LINK_POSSIBLE_STATE,
+ *  but this usage is discouraged.
+ */
+#define NETIO_PARAM_LINK_CONFIG NETIO_PARAM_LINK_POSSIBLE_STATE
+
+/** Current link state. This address is only valid with @ref netio_get().
+ *  The value is zero or more of the "NETIO_LINK_xxx" flags, ORed together.
+ *  If the link is down, the value ANDed with NETIO_LINK_SPEED will be
+ *  zero; if the link is up, the value ANDed with NETIO_LINK_SPEED will
+ *  result in exactly one of the NETIO_LINK_xxx values, indicating the
+ *  current speed. */
+#define NETIO_PARAM_LINK_CURRENT_STATE 7
+
+/** Variant symbol for current state, retained for compatibility with
+ *  pre-MDE-2.1 programs. */
+#define NETIO_PARAM_LINK_STATUS NETIO_PARAM_LINK_CURRENT_STATE
+
+/** Packet Coherence protocol. This address is only valid with @ref netio_get().
+ *  The value is nonzero if the interface is configured for cache-coherent DMA.
+ */
+#define NETIO_PARAM_COHERENT 8
+
+/** Desired link state. The value is a conbination of "NETIO_LINK_xxx"
+ *  flags, which specify the desired state for the link.  With @ref
+ *  netio_set(), this will, in the background, attempt to bring up the link
+ *  using whichever of the requested flags are reasonable, or take down the
+ *  link if the flags are zero.  The actual link up or down operation may
+ *  happen after this call completes.  If the link state changes in the
+ *  future, the system will continue to try to get back to the desired link
+ *  state; for instance, if the link is brought up successfully, and then
+ *  the network cable is disconnected, the link will go down.  However, the
+ *  desired state of the link is still up, so if the cable is reconnected,
+ *  the link will be brought up again.
+ *
+ *  With @ref netio_get(), this will indicate the desired state for the
+ *  link, as set with a previous netio_set() call, or implicitly by a
+ *  netio_input_register() or netio_input_unregister() operation.  This may
+ *  not reflect the current state of the link; to get that, use
+ *  ::NETIO_PARAM_LINK_CURRENT_STATE. */
+#define NETIO_PARAM_LINK_DESIRED_STATE 9
+
+/** NetIO statistics structure.  Retrieved using the ::NETIO_PARAM_STAT
+ *  address passed to @ref netio_get(). */
+typedef struct
+{
+  /** Number of packets which have been received by the IPP and forwarded
+   *  to a tile's receive queue for processing.  This value wraps at its
+   *  maximum, and is not cleared upon read. */
+  uint32_t packets_received;
+
+  /** Number of packets which have been dropped by the IPP, because they could
+   *  not be received, or could not be forwarded to a tile.  The former happens
+   *  when the IPP does not have a free packet buffer of suitable size for an
+   *  incoming frame.  The latter happens when all potential destination tiles
+   *  for a packet, as defined by the group, bucket, and queue configuration,
+   *  have full receive queues.   This value wraps at its maximum, and is not
+   *  cleared upon read. */
+  uint32_t packets_dropped;
+
+  /*
+   * Note: the #defines after each of the following four one-byte values
+   * denote their location within the third word of the netio_stat_t.  They
+   * are intended for use only by the IPP implementation and are thus omitted
+   * from the Doxygen output.
+   */
+
+  /** Number of packets dropped because no worker was able to accept a new
+   *  packet.  This value saturates at its maximum, and is cleared upon
+   *  read. */
+  uint8_t drops_no_worker;
+#ifndef __DOXYGEN__
+#define NETIO_STAT_DROPS_NO_WORKER   0
+#endif
+
+  /** Number of packets dropped because no small buffers were available.
+   *  This value saturates at its maximum, and is cleared upon read. */
+  uint8_t drops_no_smallbuf;
+#ifndef __DOXYGEN__
+#define NETIO_STAT_DROPS_NO_SMALLBUF 1
+#endif
+
+  /** Number of packets dropped because no large buffers were available.
+   *  This value saturates at its maximum, and is cleared upon read. */
+  uint8_t drops_no_largebuf;
+#ifndef __DOXYGEN__
+#define NETIO_STAT_DROPS_NO_LARGEBUF 2
+#endif
+
+  /** Number of packets dropped because no jumbo buffers were available.
+   *  This value saturates at its maximum, and is cleared upon read. */
+  uint8_t drops_no_jumbobuf;
+#ifndef __DOXYGEN__
+#define NETIO_STAT_DROPS_NO_JUMBOBUF 3
+#endif
+}
+netio_stat_t;
+
+
+/** Link can run, should run, or is running at 10 Mbps. */
+#define NETIO_LINK_10M         0x01
+
+/** Link can run, should run, or is running at 100 Mbps. */
+#define NETIO_LINK_100M        0x02
+
+/** Link can run, should run, or is running at 1 Gbps. */
+#define NETIO_LINK_1G          0x04
+
+/** Link can run, should run, or is running at 10 Gbps. */
+#define NETIO_LINK_10G         0x08
+
+/** Link should run at the highest speed supported by the link and by
+ *  the device connected to the link.  Only usable as a value for
+ *  the link's desired state; never returned as a value for the current
+ *  or possible states. */
+#define NETIO_LINK_ANYSPEED    0x10
+
+/** All legal link speeds. */
+#define NETIO_LINK_SPEED  (NETIO_LINK_10M  | \
+                           NETIO_LINK_100M | \
+                           NETIO_LINK_1G   | \
+                           NETIO_LINK_10G  | \
+                           NETIO_LINK_ANYSPEED)
+
+
+/** MAC register class.  Addr is a register offset within the MAC.
+ *  Registers within the XGbE and GbE MACs are documented in the Tile
+ *  Processor I/O Device Guide (UG104). MAC registers start at address
+ *  0x4000, and do not include the MAC_INTERFACE registers. */
+#define NETIO_MAC             1
+
+/** MDIO register class (IEEE 802.3 clause 22 format).  Addr is the "addr"
+ *  member of a netio_mdio_addr_t structure. */
+#define NETIO_MDIO            2
+
+/** MDIO register class (IEEE 802.3 clause 45 format).  Addr is the "addr"
+ *  member of a netio_mdio_addr_t structure. */
+#define NETIO_MDIO_CLAUSE45   3
+
+/** NetIO MDIO address type.  Retrieved or provided using the ::NETIO_MDIO
+ *  address passed to @ref netio_get() or @ref netio_set(). */
+typedef union
+{
+  struct
+  {
+    unsigned int reg:16;  /**< MDIO register offset.  For clause 22 access,
+                               must be less than 32. */
+    unsigned int phy:5;   /**< Which MDIO PHY to access. */
+    unsigned int dev:5;   /**< Which MDIO device to access within that PHY.
+                               Applicable for clause 45 access only; ignored
+                               for clause 22 access. */
+  }
+  bits;                   /**< Container for bitfields. */
+  uint64_t addr;          /**< Value to pass to @ref netio_get() or
+                           *   @ref netio_set(). */
+}
+netio_mdio_addr_t;
+
+/** @} */
+
+#endif /* __NETIO_INTF_H__ */