aboutsummaryrefslogtreecommitdiffstats
path: root/arch/tile
diff options
context:
space:
mode:
Diffstat (limited to 'arch/tile')
-rw-r--r--arch/tile/Kconfig8
-rw-r--r--arch/tile/Makefile5
-rw-r--r--arch/tile/include/arch/spr_def.h4
-rw-r--r--arch/tile/include/asm/atomic.h50
-rw-r--r--arch/tile/include/asm/atomic_32.h2
-rw-r--r--arch/tile/include/asm/bitops_64.h8
-rw-r--r--arch/tile/include/asm/cmpxchg.h73
-rw-r--r--arch/tile/include/asm/irq.h2
-rw-r--r--arch/tile/include/asm/spinlock_64.h2
-rw-r--r--arch/tile/include/asm/stack.h1
-rw-r--r--arch/tile/include/asm/traps.h6
-rw-r--r--arch/tile/kernel/entry.S2
-rw-r--r--arch/tile/kernel/intvec_32.S24
-rw-r--r--arch/tile/kernel/intvec_64.S52
-rw-r--r--arch/tile/kernel/module.c2
-rw-r--r--arch/tile/kernel/process.c3
-rw-r--r--arch/tile/kernel/setup.c36
-rw-r--r--arch/tile/kernel/single_step.c31
-rw-r--r--arch/tile/kernel/smp.c8
-rw-r--r--arch/tile/kernel/stack.c232
-rw-r--r--arch/tile/kernel/traps.c15
-rw-r--r--arch/tile/lib/Makefile1
-rw-r--r--arch/tile/lib/cacheflush.c30
-rw-r--r--arch/tile/lib/memcpy_user_64.c8
-rw-r--r--arch/tile/lib/spinlock_common.h2
-rw-r--r--arch/tile/mm/fault.c22
-rw-r--r--arch/tile/mm/homecache.c1
-rw-r--r--arch/tile/mm/init.c26
-rw-r--r--arch/tile/mm/pgtable.c38
29 files changed, 433 insertions, 261 deletions
diff --git a/arch/tile/Kconfig b/arch/tile/Kconfig
index 11270ca22c0a..96033e2d6845 100644
--- a/arch/tile/Kconfig
+++ b/arch/tile/Kconfig
@@ -12,7 +12,7 @@ config TILE
12 select GENERIC_PENDING_IRQ if SMP 12 select GENERIC_PENDING_IRQ if SMP
13 select GENERIC_IRQ_SHOW 13 select GENERIC_IRQ_SHOW
14 select SYS_HYPERVISOR 14 select SYS_HYPERVISOR
15 select ARCH_HAVE_NMI_SAFE_CMPXCHG if !M386 15 select ARCH_HAVE_NMI_SAFE_CMPXCHG
16 16
17# FIXME: investigate whether we need/want these options. 17# FIXME: investigate whether we need/want these options.
18# select HAVE_IOREMAP_PROT 18# select HAVE_IOREMAP_PROT
@@ -69,6 +69,9 @@ config ARCH_PHYS_ADDR_T_64BIT
69config ARCH_DMA_ADDR_T_64BIT 69config ARCH_DMA_ADDR_T_64BIT
70 def_bool y 70 def_bool y
71 71
72config NEED_DMA_MAP_STATE
73 def_bool y
74
72config LOCKDEP_SUPPORT 75config LOCKDEP_SUPPORT
73 def_bool y 76 def_bool y
74 77
@@ -118,7 +121,7 @@ config 64BIT
118 121
119config ARCH_DEFCONFIG 122config ARCH_DEFCONFIG
120 string 123 string
121 default "arch/tile/configs/tile_defconfig" if !TILEGX 124 default "arch/tile/configs/tilepro_defconfig" if !TILEGX
122 default "arch/tile/configs/tilegx_defconfig" if TILEGX 125 default "arch/tile/configs/tilegx_defconfig" if TILEGX
123 126
124source "init/Kconfig" 127source "init/Kconfig"
@@ -240,6 +243,7 @@ endchoice
240 243
241config PAGE_OFFSET 244config PAGE_OFFSET
242 hex 245 hex
246 depends on !64BIT
243 default 0xF0000000 if VMSPLIT_3_75G 247 default 0xF0000000 if VMSPLIT_3_75G
244 default 0xE0000000 if VMSPLIT_3_5G 248 default 0xE0000000 if VMSPLIT_3_5G
245 default 0xB0000000 if VMSPLIT_2_75G 249 default 0xB0000000 if VMSPLIT_2_75G
diff --git a/arch/tile/Makefile b/arch/tile/Makefile
index 17acce70569b..9520bc5a4b7f 100644
--- a/arch/tile/Makefile
+++ b/arch/tile/Makefile
@@ -30,7 +30,8 @@ ifneq ($(CONFIG_DEBUG_EXTRA_FLAGS),"")
30KBUILD_CFLAGS += $(CONFIG_DEBUG_EXTRA_FLAGS) 30KBUILD_CFLAGS += $(CONFIG_DEBUG_EXTRA_FLAGS)
31endif 31endif
32 32
33LIBGCC_PATH := $(shell $(CC) $(KBUILD_CFLAGS) -print-libgcc-file-name) 33LIBGCC_PATH := \
34 $(shell $(CC) $(KBUILD_CFLAGS) $(KCFLAGS) -print-libgcc-file-name)
34 35
35# Provide the path to use for "make defconfig". 36# Provide the path to use for "make defconfig".
36KBUILD_DEFCONFIG := $(ARCH)_defconfig 37KBUILD_DEFCONFIG := $(ARCH)_defconfig
@@ -53,8 +54,6 @@ libs-y += $(LIBGCC_PATH)
53# See arch/tile/Kbuild for content of core part of the kernel 54# See arch/tile/Kbuild for content of core part of the kernel
54core-y += arch/tile/ 55core-y += arch/tile/
55 56
56core-$(CONFIG_KVM) += arch/tile/kvm/
57
58ifdef TILERA_ROOT 57ifdef TILERA_ROOT
59INSTALL_PATH ?= $(TILERA_ROOT)/tile/boot 58INSTALL_PATH ?= $(TILERA_ROOT)/tile/boot
60endif 59endif
diff --git a/arch/tile/include/arch/spr_def.h b/arch/tile/include/arch/spr_def.h
index f548efeb2de3..d6ba449b5363 100644
--- a/arch/tile/include/arch/spr_def.h
+++ b/arch/tile/include/arch/spr_def.h
@@ -60,8 +60,8 @@
60 _concat4(SPR_IPI_EVENT_, CONFIG_KERNEL_PL,,) 60 _concat4(SPR_IPI_EVENT_, CONFIG_KERNEL_PL,,)
61#define SPR_IPI_EVENT_RESET_K \ 61#define SPR_IPI_EVENT_RESET_K \
62 _concat4(SPR_IPI_EVENT_RESET_, CONFIG_KERNEL_PL,,) 62 _concat4(SPR_IPI_EVENT_RESET_, CONFIG_KERNEL_PL,,)
63#define SPR_IPI_MASK_SET_K \ 63#define SPR_IPI_EVENT_SET_K \
64 _concat4(SPR_IPI_MASK_SET_, CONFIG_KERNEL_PL,,) 64 _concat4(SPR_IPI_EVENT_SET_, CONFIG_KERNEL_PL,,)
65#define INT_IPI_K \ 65#define INT_IPI_K \
66 _concat4(INT_IPI_, CONFIG_KERNEL_PL,,) 66 _concat4(INT_IPI_, CONFIG_KERNEL_PL,,)
67 67
diff --git a/arch/tile/include/asm/atomic.h b/arch/tile/include/asm/atomic.h
index bb696da5d7cd..f2461429a4a4 100644
--- a/arch/tile/include/asm/atomic.h
+++ b/arch/tile/include/asm/atomic.h
@@ -17,6 +17,8 @@
17#ifndef _ASM_TILE_ATOMIC_H 17#ifndef _ASM_TILE_ATOMIC_H
18#define _ASM_TILE_ATOMIC_H 18#define _ASM_TILE_ATOMIC_H
19 19
20#include <asm/cmpxchg.h>
21
20#ifndef __ASSEMBLY__ 22#ifndef __ASSEMBLY__
21 23
22#include <linux/compiler.h> 24#include <linux/compiler.h>
@@ -121,54 +123,6 @@ static inline int atomic_read(const atomic_t *v)
121 */ 123 */
122#define atomic_add_negative(i, v) (atomic_add_return((i), (v)) < 0) 124#define atomic_add_negative(i, v) (atomic_add_return((i), (v)) < 0)
123 125
124/* Nonexistent functions intended to cause link errors. */
125extern unsigned long __xchg_called_with_bad_pointer(void);
126extern unsigned long __cmpxchg_called_with_bad_pointer(void);
127
128#define xchg(ptr, x) \
129 ({ \
130 typeof(*(ptr)) __x; \
131 switch (sizeof(*(ptr))) { \
132 case 4: \
133 __x = (typeof(__x))(typeof(__x-__x))atomic_xchg( \
134 (atomic_t *)(ptr), \
135 (u32)(typeof((x)-(x)))(x)); \
136 break; \
137 case 8: \
138 __x = (typeof(__x))(typeof(__x-__x))atomic64_xchg( \
139 (atomic64_t *)(ptr), \
140 (u64)(typeof((x)-(x)))(x)); \
141 break; \
142 default: \
143 __xchg_called_with_bad_pointer(); \
144 } \
145 __x; \
146 })
147
148#define cmpxchg(ptr, o, n) \
149 ({ \
150 typeof(*(ptr)) __x; \
151 switch (sizeof(*(ptr))) { \
152 case 4: \
153 __x = (typeof(__x))(typeof(__x-__x))atomic_cmpxchg( \
154 (atomic_t *)(ptr), \
155 (u32)(typeof((o)-(o)))(o), \
156 (u32)(typeof((n)-(n)))(n)); \
157 break; \
158 case 8: \
159 __x = (typeof(__x))(typeof(__x-__x))atomic64_cmpxchg( \
160 (atomic64_t *)(ptr), \
161 (u64)(typeof((o)-(o)))(o), \
162 (u64)(typeof((n)-(n)))(n)); \
163 break; \
164 default: \
165 __cmpxchg_called_with_bad_pointer(); \
166 } \
167 __x; \
168 })
169
170#define tas(ptr) (xchg((ptr), 1))
171
172#endif /* __ASSEMBLY__ */ 126#endif /* __ASSEMBLY__ */
173 127
174#ifndef __tilegx__ 128#ifndef __tilegx__
diff --git a/arch/tile/include/asm/atomic_32.h b/arch/tile/include/asm/atomic_32.h
index 466dc4a39a4f..54d1da826f93 100644
--- a/arch/tile/include/asm/atomic_32.h
+++ b/arch/tile/include/asm/atomic_32.h
@@ -200,7 +200,7 @@ static inline u64 atomic64_add_return(u64 i, atomic64_t *v)
200 * @u: ...unless v is equal to u. 200 * @u: ...unless v is equal to u.
201 * 201 *
202 * Atomically adds @a to @v, so long as @v was not already @u. 202 * Atomically adds @a to @v, so long as @v was not already @u.
203 * Returns the old value of @v. 203 * Returns non-zero if @v was not @u, and zero otherwise.
204 */ 204 */
205static inline u64 atomic64_add_unless(atomic64_t *v, u64 a, u64 u) 205static inline u64 atomic64_add_unless(atomic64_t *v, u64 a, u64 u)
206{ 206{
diff --git a/arch/tile/include/asm/bitops_64.h b/arch/tile/include/asm/bitops_64.h
index 58d021a9834f..60b87ee54fb8 100644
--- a/arch/tile/include/asm/bitops_64.h
+++ b/arch/tile/include/asm/bitops_64.h
@@ -38,10 +38,10 @@ static inline void clear_bit(unsigned nr, volatile unsigned long *addr)
38 38
39static inline void change_bit(unsigned nr, volatile unsigned long *addr) 39static inline void change_bit(unsigned nr, volatile unsigned long *addr)
40{ 40{
41 unsigned long old, mask = (1UL << (nr % BITS_PER_LONG)); 41 unsigned long mask = (1UL << (nr % BITS_PER_LONG));
42 long guess, oldval; 42 unsigned long guess, oldval;
43 addr += nr / BITS_PER_LONG; 43 addr += nr / BITS_PER_LONG;
44 old = *addr; 44 oldval = *addr;
45 do { 45 do {
46 guess = oldval; 46 guess = oldval;
47 oldval = atomic64_cmpxchg((atomic64_t *)addr, 47 oldval = atomic64_cmpxchg((atomic64_t *)addr,
@@ -85,7 +85,7 @@ static inline int test_and_change_bit(unsigned nr,
85 volatile unsigned long *addr) 85 volatile unsigned long *addr)
86{ 86{
87 unsigned long mask = (1UL << (nr % BITS_PER_LONG)); 87 unsigned long mask = (1UL << (nr % BITS_PER_LONG));
88 long guess, oldval = *addr; 88 unsigned long guess, oldval;
89 addr += nr / BITS_PER_LONG; 89 addr += nr / BITS_PER_LONG;
90 oldval = *addr; 90 oldval = *addr;
91 do { 91 do {
diff --git a/arch/tile/include/asm/cmpxchg.h b/arch/tile/include/asm/cmpxchg.h
new file mode 100644
index 000000000000..276f067e3640
--- /dev/null
+++ b/arch/tile/include/asm/cmpxchg.h
@@ -0,0 +1,73 @@
1/*
2 * cmpxchg.h -- forked from asm/atomic.h with this copyright:
3 *
4 * Copyright 2010 Tilera Corporation. All Rights Reserved.
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation, version 2.
9 *
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
13 * NON INFRINGEMENT. See the GNU General Public License for
14 * more details.
15 *
16 */
17
18#ifndef _ASM_TILE_CMPXCHG_H
19#define _ASM_TILE_CMPXCHG_H
20
21#ifndef __ASSEMBLY__
22
23/* Nonexistent functions intended to cause link errors. */
24extern unsigned long __xchg_called_with_bad_pointer(void);
25extern unsigned long __cmpxchg_called_with_bad_pointer(void);
26
27#define xchg(ptr, x) \
28 ({ \
29 typeof(*(ptr)) __x; \
30 switch (sizeof(*(ptr))) { \
31 case 4: \
32 __x = (typeof(__x))(typeof(__x-__x))atomic_xchg( \
33 (atomic_t *)(ptr), \
34 (u32)(typeof((x)-(x)))(x)); \
35 break; \
36 case 8: \
37 __x = (typeof(__x))(typeof(__x-__x))atomic64_xchg( \
38 (atomic64_t *)(ptr), \
39 (u64)(typeof((x)-(x)))(x)); \
40 break; \
41 default: \
42 __xchg_called_with_bad_pointer(); \
43 } \
44 __x; \
45 })
46
47#define cmpxchg(ptr, o, n) \
48 ({ \
49 typeof(*(ptr)) __x; \
50 switch (sizeof(*(ptr))) { \
51 case 4: \
52 __x = (typeof(__x))(typeof(__x-__x))atomic_cmpxchg( \
53 (atomic_t *)(ptr), \
54 (u32)(typeof((o)-(o)))(o), \
55 (u32)(typeof((n)-(n)))(n)); \
56 break; \
57 case 8: \
58 __x = (typeof(__x))(typeof(__x-__x))atomic64_cmpxchg( \
59 (atomic64_t *)(ptr), \
60 (u64)(typeof((o)-(o)))(o), \
61 (u64)(typeof((n)-(n)))(n)); \
62 break; \
63 default: \
64 __cmpxchg_called_with_bad_pointer(); \
65 } \
66 __x; \
67 })
68
69#define tas(ptr) (xchg((ptr), 1))
70
71#endif /* __ASSEMBLY__ */
72
73#endif /* _ASM_TILE_CMPXCHG_H */
diff --git a/arch/tile/include/asm/irq.h b/arch/tile/include/asm/irq.h
index f80f8ceabc67..33cff9a3058b 100644
--- a/arch/tile/include/asm/irq.h
+++ b/arch/tile/include/asm/irq.h
@@ -21,7 +21,7 @@
21#define NR_IRQS 32 21#define NR_IRQS 32
22 22
23/* IRQ numbers used for linux IPIs. */ 23/* IRQ numbers used for linux IPIs. */
24#define IRQ_RESCHEDULE 1 24#define IRQ_RESCHEDULE 0
25 25
26#define irq_canonicalize(irq) (irq) 26#define irq_canonicalize(irq) (irq)
27 27
diff --git a/arch/tile/include/asm/spinlock_64.h b/arch/tile/include/asm/spinlock_64.h
index 72be5904e020..5f8b6a095fd8 100644
--- a/arch/tile/include/asm/spinlock_64.h
+++ b/arch/tile/include/asm/spinlock_64.h
@@ -137,7 +137,7 @@ static inline void arch_read_unlock(arch_rwlock_t *rw)
137static inline void arch_write_unlock(arch_rwlock_t *rw) 137static inline void arch_write_unlock(arch_rwlock_t *rw)
138{ 138{
139 __insn_mf(); 139 __insn_mf();
140 rw->lock = 0; 140 __insn_exch4(&rw->lock, 0); /* Avoid waiting in the write buffer. */
141} 141}
142 142
143static inline int arch_read_trylock(arch_rwlock_t *rw) 143static inline int arch_read_trylock(arch_rwlock_t *rw)
diff --git a/arch/tile/include/asm/stack.h b/arch/tile/include/asm/stack.h
index 4d97a2db932e..0e9d382a2d45 100644
--- a/arch/tile/include/asm/stack.h
+++ b/arch/tile/include/asm/stack.h
@@ -25,7 +25,6 @@
25struct KBacktraceIterator { 25struct KBacktraceIterator {
26 BacktraceIterator it; 26 BacktraceIterator it;
27 struct task_struct *task; /* task we are backtracing */ 27 struct task_struct *task; /* task we are backtracing */
28 pte_t *pgtable; /* page table for user space access */
29 int end; /* iteration complete. */ 28 int end; /* iteration complete. */
30 int new_context; /* new context is starting */ 29 int new_context; /* new context is starting */
31 int profile; /* profiling, so stop on async intrpt */ 30 int profile; /* profiling, so stop on async intrpt */
diff --git a/arch/tile/include/asm/traps.h b/arch/tile/include/asm/traps.h
index 5f20f920f932..e28c3df4176a 100644
--- a/arch/tile/include/asm/traps.h
+++ b/arch/tile/include/asm/traps.h
@@ -64,7 +64,11 @@ void do_breakpoint(struct pt_regs *, int fault_num);
64 64
65 65
66#ifdef __tilegx__ 66#ifdef __tilegx__
67/* kernel/single_step.c */
67void gx_singlestep_handle(struct pt_regs *, int fault_num); 68void gx_singlestep_handle(struct pt_regs *, int fault_num);
69
70/* kernel/intvec_64.S */
71void fill_ra_stack(void);
68#endif 72#endif
69 73
70#endif /* _ASM_TILE_SYSCALLS_H */ 74#endif /* _ASM_TILE_TRAPS_H */
diff --git a/arch/tile/kernel/entry.S b/arch/tile/kernel/entry.S
index 431e9ae60488..ec91568df880 100644
--- a/arch/tile/kernel/entry.S
+++ b/arch/tile/kernel/entry.S
@@ -85,6 +85,7 @@ STD_ENTRY(cpu_idle_on_new_stack)
85/* Loop forever on a nap during SMP boot. */ 85/* Loop forever on a nap during SMP boot. */
86STD_ENTRY(smp_nap) 86STD_ENTRY(smp_nap)
87 nap 87 nap
88 nop /* avoid provoking the icache prefetch with a jump */
88 j smp_nap /* we are not architecturally guaranteed not to exit nap */ 89 j smp_nap /* we are not architecturally guaranteed not to exit nap */
89 jrp lr /* clue in the backtracer */ 90 jrp lr /* clue in the backtracer */
90 STD_ENDPROC(smp_nap) 91 STD_ENDPROC(smp_nap)
@@ -105,5 +106,6 @@ STD_ENTRY(_cpu_idle)
105 .global _cpu_idle_nap 106 .global _cpu_idle_nap
106_cpu_idle_nap: 107_cpu_idle_nap:
107 nap 108 nap
109 nop /* avoid provoking the icache prefetch with a jump */
108 jrp lr 110 jrp lr
109 STD_ENDPROC(_cpu_idle) 111 STD_ENDPROC(_cpu_idle)
diff --git a/arch/tile/kernel/intvec_32.S b/arch/tile/kernel/intvec_32.S
index aecc8ed5f39b..5d56a1ef5ba5 100644
--- a/arch/tile/kernel/intvec_32.S
+++ b/arch/tile/kernel/intvec_32.S
@@ -799,6 +799,10 @@ handle_interrupt:
799 * This routine takes a boolean in r30 indicating if this is an NMI. 799 * This routine takes a boolean in r30 indicating if this is an NMI.
800 * If so, we also expect a boolean in r31 indicating whether to 800 * If so, we also expect a boolean in r31 indicating whether to
801 * re-enable the oprofile interrupts. 801 * re-enable the oprofile interrupts.
802 *
803 * Note that .Lresume_userspace is jumped to directly in several
804 * places, and we need to make sure r30 is set correctly in those
805 * callers as well.
802 */ 806 */
803STD_ENTRY(interrupt_return) 807STD_ENTRY(interrupt_return)
804 /* If we're resuming to kernel space, don't check thread flags. */ 808 /* If we're resuming to kernel space, don't check thread flags. */
@@ -1237,7 +1241,10 @@ handle_syscall:
1237 bzt r30, 1f 1241 bzt r30, 1f
1238 jal do_syscall_trace 1242 jal do_syscall_trace
1239 FEEDBACK_REENTER(handle_syscall) 1243 FEEDBACK_REENTER(handle_syscall)
12401: j .Lresume_userspace /* jump into middle of interrupt_return */ 12441: {
1245 movei r30, 0 /* not an NMI */
1246 j .Lresume_userspace /* jump into middle of interrupt_return */
1247 }
1241 1248
1242.Linvalid_syscall: 1249.Linvalid_syscall:
1243 /* Report an invalid syscall back to the user program */ 1250 /* Report an invalid syscall back to the user program */
@@ -1246,7 +1253,10 @@ handle_syscall:
1246 movei r28, -ENOSYS 1253 movei r28, -ENOSYS
1247 } 1254 }
1248 sw r29, r28 1255 sw r29, r28
1249 j .Lresume_userspace /* jump into middle of interrupt_return */ 1256 {
1257 movei r30, 0 /* not an NMI */
1258 j .Lresume_userspace /* jump into middle of interrupt_return */
1259 }
1250 STD_ENDPROC(handle_syscall) 1260 STD_ENDPROC(handle_syscall)
1251 1261
1252 /* Return the address for oprofile to suppress in backtraces. */ 1262 /* Return the address for oprofile to suppress in backtraces. */
@@ -1262,7 +1272,10 @@ STD_ENTRY(ret_from_fork)
1262 jal sim_notify_fork 1272 jal sim_notify_fork
1263 jal schedule_tail 1273 jal schedule_tail
1264 FEEDBACK_REENTER(ret_from_fork) 1274 FEEDBACK_REENTER(ret_from_fork)
1265 j .Lresume_userspace /* jump into middle of interrupt_return */ 1275 {
1276 movei r30, 0 /* not an NMI */
1277 j .Lresume_userspace /* jump into middle of interrupt_return */
1278 }
1266 STD_ENDPROC(ret_from_fork) 1279 STD_ENDPROC(ret_from_fork)
1267 1280
1268 /* 1281 /*
@@ -1376,7 +1389,10 @@ handle_ill:
1376 1389
1377 jal send_sigtrap /* issue a SIGTRAP */ 1390 jal send_sigtrap /* issue a SIGTRAP */
1378 FEEDBACK_REENTER(handle_ill) 1391 FEEDBACK_REENTER(handle_ill)
1379 j .Lresume_userspace /* jump into middle of interrupt_return */ 1392 {
1393 movei r30, 0 /* not an NMI */
1394 j .Lresume_userspace /* jump into middle of interrupt_return */
1395 }
1380 1396
1381.Ldispatch_normal_ill: 1397.Ldispatch_normal_ill:
1382 { 1398 {
diff --git a/arch/tile/kernel/intvec_64.S b/arch/tile/kernel/intvec_64.S
index 79c93e10ba27..49d9d6621682 100644
--- a/arch/tile/kernel/intvec_64.S
+++ b/arch/tile/kernel/intvec_64.S
@@ -22,6 +22,7 @@
22#include <asm/irqflags.h> 22#include <asm/irqflags.h>
23#include <asm/asm-offsets.h> 23#include <asm/asm-offsets.h>
24#include <asm/types.h> 24#include <asm/types.h>
25#include <asm/signal.h>
25#include <hv/hypervisor.h> 26#include <hv/hypervisor.h>
26#include <arch/abi.h> 27#include <arch/abi.h>
27#include <arch/interrupts.h> 28#include <arch/interrupts.h>
@@ -605,6 +606,10 @@ handle_interrupt:
605 * This routine takes a boolean in r30 indicating if this is an NMI. 606 * This routine takes a boolean in r30 indicating if this is an NMI.
606 * If so, we also expect a boolean in r31 indicating whether to 607 * If so, we also expect a boolean in r31 indicating whether to
607 * re-enable the oprofile interrupts. 608 * re-enable the oprofile interrupts.
609 *
610 * Note that .Lresume_userspace is jumped to directly in several
611 * places, and we need to make sure r30 is set correctly in those
612 * callers as well.
608 */ 613 */
609STD_ENTRY(interrupt_return) 614STD_ENTRY(interrupt_return)
610 /* If we're resuming to kernel space, don't check thread flags. */ 615 /* If we're resuming to kernel space, don't check thread flags. */
@@ -1039,11 +1044,28 @@ handle_syscall:
1039 1044
1040 /* Do syscall trace again, if requested. */ 1045 /* Do syscall trace again, if requested. */
1041 ld r30, r31 1046 ld r30, r31
1042 andi r30, r30, _TIF_SYSCALL_TRACE 1047 andi r0, r30, _TIF_SYSCALL_TRACE
1043 beqzt r30, 1f 1048 {
1049 andi r0, r30, _TIF_SINGLESTEP
1050 beqzt r0, 1f
1051 }
1044 jal do_syscall_trace 1052 jal do_syscall_trace
1045 FEEDBACK_REENTER(handle_syscall) 1053 FEEDBACK_REENTER(handle_syscall)
10461: j .Lresume_userspace /* jump into middle of interrupt_return */ 1054 andi r0, r30, _TIF_SINGLESTEP
1055
10561: beqzt r0, 2f
1057
1058 /* Single stepping -- notify ptrace. */
1059 {
1060 movei r0, SIGTRAP
1061 jal ptrace_notify
1062 }
1063 FEEDBACK_REENTER(handle_syscall)
1064
10652: {
1066 movei r30, 0 /* not an NMI */
1067 j .Lresume_userspace /* jump into middle of interrupt_return */
1068 }
1047 1069
1048.Lcompat_syscall: 1070.Lcompat_syscall:
1049 /* 1071 /*
@@ -1077,7 +1099,10 @@ handle_syscall:
1077 movei r28, -ENOSYS 1099 movei r28, -ENOSYS
1078 } 1100 }
1079 st r29, r28 1101 st r29, r28
1080 j .Lresume_userspace /* jump into middle of interrupt_return */ 1102 {
1103 movei r30, 0 /* not an NMI */
1104 j .Lresume_userspace /* jump into middle of interrupt_return */
1105 }
1081 STD_ENDPROC(handle_syscall) 1106 STD_ENDPROC(handle_syscall)
1082 1107
1083 /* Return the address for oprofile to suppress in backtraces. */ 1108 /* Return the address for oprofile to suppress in backtraces. */
@@ -1093,7 +1118,10 @@ STD_ENTRY(ret_from_fork)
1093 jal sim_notify_fork 1118 jal sim_notify_fork
1094 jal schedule_tail 1119 jal schedule_tail
1095 FEEDBACK_REENTER(ret_from_fork) 1120 FEEDBACK_REENTER(ret_from_fork)
1096 j .Lresume_userspace 1121 {
1122 movei r30, 0 /* not an NMI */
1123 j .Lresume_userspace /* jump into middle of interrupt_return */
1124 }
1097 STD_ENDPROC(ret_from_fork) 1125 STD_ENDPROC(ret_from_fork)
1098 1126
1099/* Various stub interrupt handlers and syscall handlers */ 1127/* Various stub interrupt handlers and syscall handlers */
@@ -1156,6 +1184,18 @@ int_unalign:
1156 push_extra_callee_saves r0 1184 push_extra_callee_saves r0
1157 j do_trap 1185 j do_trap
1158 1186
1187/* Fill the return address stack with nonzero entries. */
1188STD_ENTRY(fill_ra_stack)
1189 {
1190 move r0, lr
1191 jal 1f
1192 }
11931: jal 2f
11942: jal 3f
11953: jal 4f
11964: jrp r0
1197 STD_ENDPROC(fill_ra_stack)
1198
1159/* Include .intrpt1 array of interrupt vectors */ 1199/* Include .intrpt1 array of interrupt vectors */
1160 .section ".intrpt1", "ax" 1200 .section ".intrpt1", "ax"
1161 1201
@@ -1166,7 +1206,7 @@ int_unalign:
1166#define do_hardwall_trap bad_intr 1206#define do_hardwall_trap bad_intr
1167#endif 1207#endif
1168 1208
1169 int_hand INT_MEM_ERROR, MEM_ERROR, bad_intr 1209 int_hand INT_MEM_ERROR, MEM_ERROR, do_trap
1170 int_hand INT_SINGLE_STEP_3, SINGLE_STEP_3, bad_intr 1210 int_hand INT_SINGLE_STEP_3, SINGLE_STEP_3, bad_intr
1171#if CONFIG_KERNEL_PL == 2 1211#if CONFIG_KERNEL_PL == 2
1172 int_hand INT_SINGLE_STEP_2, SINGLE_STEP_2, gx_singlestep_handle 1212 int_hand INT_SINGLE_STEP_2, SINGLE_STEP_2, gx_singlestep_handle
diff --git a/arch/tile/kernel/module.c b/arch/tile/kernel/module.c
index b90ab9925674..98d476920106 100644
--- a/arch/tile/kernel/module.c
+++ b/arch/tile/kernel/module.c
@@ -67,6 +67,8 @@ void *module_alloc(unsigned long size)
67 area = __get_vm_area(size, VM_ALLOC, MEM_MODULE_START, MEM_MODULE_END); 67 area = __get_vm_area(size, VM_ALLOC, MEM_MODULE_START, MEM_MODULE_END);
68 if (!area) 68 if (!area)
69 goto error; 69 goto error;
70 area->nr_pages = npages;
71 area->pages = pages;
70 72
71 if (map_vm_area(area, prot_rwx, &pages)) { 73 if (map_vm_area(area, prot_rwx, &pages)) {
72 vunmap(area->addr); 74 vunmap(area->addr);
diff --git a/arch/tile/kernel/process.c b/arch/tile/kernel/process.c
index 30caecac94dc..2d5ef617bb39 100644
--- a/arch/tile/kernel/process.c
+++ b/arch/tile/kernel/process.c
@@ -28,6 +28,7 @@
28#include <linux/tracehook.h> 28#include <linux/tracehook.h>
29#include <linux/signal.h> 29#include <linux/signal.h>
30#include <asm/stack.h> 30#include <asm/stack.h>
31#include <asm/switch_to.h>
31#include <asm/homecache.h> 32#include <asm/homecache.h>
32#include <asm/syscalls.h> 33#include <asm/syscalls.h>
33#include <asm/traps.h> 34#include <asm/traps.h>
@@ -285,7 +286,7 @@ struct task_struct *validate_current(void)
285 static struct task_struct corrupt = { .comm = "<corrupt>" }; 286 static struct task_struct corrupt = { .comm = "<corrupt>" };
286 struct task_struct *tsk = current; 287 struct task_struct *tsk = current;
287 if (unlikely((unsigned long)tsk < PAGE_OFFSET || 288 if (unlikely((unsigned long)tsk < PAGE_OFFSET ||
288 (void *)tsk > high_memory || 289 (high_memory && (void *)tsk > high_memory) ||
289 ((unsigned long)tsk & (__alignof__(*tsk) - 1)) != 0)) { 290 ((unsigned long)tsk & (__alignof__(*tsk) - 1)) != 0)) {
290 pr_err("Corrupt 'current' %p (sp %#lx)\n", tsk, stack_pointer); 291 pr_err("Corrupt 'current' %p (sp %#lx)\n", tsk, stack_pointer);
291 tsk = &corrupt; 292 tsk = &corrupt;
diff --git a/arch/tile/kernel/setup.c b/arch/tile/kernel/setup.c
index 5f85d8b34dbb..bff23f476110 100644
--- a/arch/tile/kernel/setup.c
+++ b/arch/tile/kernel/setup.c
@@ -103,13 +103,11 @@ unsigned long __initdata pci_reserve_end_pfn = -1U;
103 103
104static int __init setup_maxmem(char *str) 104static int __init setup_maxmem(char *str)
105{ 105{
106 long maxmem_mb; 106 unsigned long long maxmem;
107 if (str == NULL || strict_strtol(str, 0, &maxmem_mb) != 0 || 107 if (str == NULL || (maxmem = memparse(str, NULL)) == 0)
108 maxmem_mb == 0)
109 return -EINVAL; 108 return -EINVAL;
110 109
111 maxmem_pfn = (maxmem_mb >> (HPAGE_SHIFT - 20)) << 110 maxmem_pfn = (maxmem >> HPAGE_SHIFT) << (HPAGE_SHIFT - PAGE_SHIFT);
112 (HPAGE_SHIFT - PAGE_SHIFT);
113 pr_info("Forcing RAM used to no more than %dMB\n", 111 pr_info("Forcing RAM used to no more than %dMB\n",
114 maxmem_pfn >> (20 - PAGE_SHIFT)); 112 maxmem_pfn >> (20 - PAGE_SHIFT));
115 return 0; 113 return 0;
@@ -119,14 +117,15 @@ early_param("maxmem", setup_maxmem);
119static int __init setup_maxnodemem(char *str) 117static int __init setup_maxnodemem(char *str)
120{ 118{
121 char *endp; 119 char *endp;
122 long maxnodemem_mb, node; 120 unsigned long long maxnodemem;
121 long node;
123 122
124 node = str ? simple_strtoul(str, &endp, 0) : INT_MAX; 123 node = str ? simple_strtoul(str, &endp, 0) : INT_MAX;
125 if (node >= MAX_NUMNODES || *endp != ':' || 124 if (node >= MAX_NUMNODES || *endp != ':')
126 strict_strtol(endp+1, 0, &maxnodemem_mb) != 0)
127 return -EINVAL; 125 return -EINVAL;
128 126
129 maxnodemem_pfn[node] = (maxnodemem_mb >> (HPAGE_SHIFT - 20)) << 127 maxnodemem = memparse(endp+1, NULL);
128 maxnodemem_pfn[node] = (maxnodemem >> HPAGE_SHIFT) <<
130 (HPAGE_SHIFT - PAGE_SHIFT); 129 (HPAGE_SHIFT - PAGE_SHIFT);
131 pr_info("Forcing RAM used on node %ld to no more than %dMB\n", 130 pr_info("Forcing RAM used on node %ld to no more than %dMB\n",
132 node, maxnodemem_pfn[node] >> (20 - PAGE_SHIFT)); 131 node, maxnodemem_pfn[node] >> (20 - PAGE_SHIFT));
@@ -913,6 +912,13 @@ void __cpuinit setup_cpu(int boot)
913 912
914#ifdef CONFIG_BLK_DEV_INITRD 913#ifdef CONFIG_BLK_DEV_INITRD
915 914
915/*
916 * Note that the kernel can potentially support other compression
917 * techniques than gz, though we don't do so by default. If we ever
918 * decide to do so we can either look for other filename extensions,
919 * or just allow a file with this name to be compressed with an
920 * arbitrary compressor (somewhat counterintuitively).
921 */
916static int __initdata set_initramfs_file; 922static int __initdata set_initramfs_file;
917static char __initdata initramfs_file[128] = "initramfs.cpio.gz"; 923static char __initdata initramfs_file[128] = "initramfs.cpio.gz";
918 924
@@ -928,9 +934,9 @@ static int __init setup_initramfs_file(char *str)
928early_param("initramfs_file", setup_initramfs_file); 934early_param("initramfs_file", setup_initramfs_file);
929 935
930/* 936/*
931 * We look for an additional "initramfs.cpio.gz" file in the hvfs. 937 * We look for an "initramfs.cpio.gz" file in the hvfs.
932 * If there is one, we allocate some memory for it and it will be 938 * If there is one, we allocate some memory for it and it will be
933 * unpacked to the initramfs after any built-in initramfs_data. 939 * unpacked to the initramfs.
934 */ 940 */
935static void __init load_hv_initrd(void) 941static void __init load_hv_initrd(void)
936{ 942{
@@ -1100,7 +1106,7 @@ EXPORT_SYMBOL(hash_for_home_map);
1100 1106
1101/* 1107/*
1102 * cpu_cacheable_map lists all the cpus whose caches the hypervisor can 1108 * cpu_cacheable_map lists all the cpus whose caches the hypervisor can
1103 * flush on our behalf. It is set to cpu_possible_map OR'ed with 1109 * flush on our behalf. It is set to cpu_possible_mask OR'ed with
1104 * hash_for_home_map, and it is what should be passed to 1110 * hash_for_home_map, and it is what should be passed to
1105 * hv_flush_remote() to flush all caches. Note that if there are 1111 * hv_flush_remote() to flush all caches. Note that if there are
1106 * dedicated hypervisor driver tiles that have authorized use of their 1112 * dedicated hypervisor driver tiles that have authorized use of their
@@ -1186,7 +1192,7 @@ static void __init setup_cpu_maps(void)
1186 sizeof(cpu_lotar_map)); 1192 sizeof(cpu_lotar_map));
1187 if (rc < 0) { 1193 if (rc < 0) {
1188 pr_err("warning: no HV_INQ_TILES_LOTAR; using AVAIL\n"); 1194 pr_err("warning: no HV_INQ_TILES_LOTAR; using AVAIL\n");
1189 cpu_lotar_map = cpu_possible_map; 1195 cpu_lotar_map = *cpu_possible_mask;
1190 } 1196 }
1191 1197
1192#if CHIP_HAS_CBOX_HOME_MAP() 1198#if CHIP_HAS_CBOX_HOME_MAP()
@@ -1196,9 +1202,9 @@ static void __init setup_cpu_maps(void)
1196 sizeof(hash_for_home_map)); 1202 sizeof(hash_for_home_map));
1197 if (rc < 0) 1203 if (rc < 0)
1198 early_panic("hv_inquire_tiles(HFH_CACHE) failed: rc %d\n", rc); 1204 early_panic("hv_inquire_tiles(HFH_CACHE) failed: rc %d\n", rc);
1199 cpumask_or(&cpu_cacheable_map, &cpu_possible_map, &hash_for_home_map); 1205 cpumask_or(&cpu_cacheable_map, cpu_possible_mask, &hash_for_home_map);
1200#else 1206#else
1201 cpu_cacheable_map = cpu_possible_map; 1207 cpu_cacheable_map = *cpu_possible_mask;
1202#endif 1208#endif
1203} 1209}
1204 1210
diff --git a/arch/tile/kernel/single_step.c b/arch/tile/kernel/single_step.c
index bc1eb586e24d..9efbc1391b3c 100644
--- a/arch/tile/kernel/single_step.c
+++ b/arch/tile/kernel/single_step.c
@@ -153,6 +153,25 @@ static tile_bundle_bits rewrite_load_store_unaligned(
153 if (((unsigned long)addr % size) == 0) 153 if (((unsigned long)addr % size) == 0)
154 return bundle; 154 return bundle;
155 155
156 /*
157 * Return SIGBUS with the unaligned address, if requested.
158 * Note that we return SIGBUS even for completely invalid addresses
159 * as long as they are in fact unaligned; this matches what the
160 * tilepro hardware would be doing, if it could provide us with the
161 * actual bad address in an SPR, which it doesn't.
162 */
163 if (unaligned_fixup == 0) {
164 siginfo_t info = {
165 .si_signo = SIGBUS,
166 .si_code = BUS_ADRALN,
167 .si_addr = addr
168 };
169 trace_unhandled_signal("unaligned trap", regs,
170 (unsigned long)addr, SIGBUS);
171 force_sig_info(info.si_signo, &info, current);
172 return (tilepro_bundle_bits) 0;
173 }
174
156#ifndef __LITTLE_ENDIAN 175#ifndef __LITTLE_ENDIAN
157# error We assume little-endian representation with copy_xx_user size 2 here 176# error We assume little-endian representation with copy_xx_user size 2 here
158#endif 177#endif
@@ -192,18 +211,6 @@ static tile_bundle_bits rewrite_load_store_unaligned(
192 return (tile_bundle_bits) 0; 211 return (tile_bundle_bits) 0;
193 } 212 }
194 213
195 if (unaligned_fixup == 0) {
196 siginfo_t info = {
197 .si_signo = SIGBUS,
198 .si_code = BUS_ADRALN,
199 .si_addr = addr
200 };
201 trace_unhandled_signal("unaligned trap", regs,
202 (unsigned long)addr, SIGBUS);
203 force_sig_info(info.si_signo, &info, current);
204 return (tile_bundle_bits) 0;
205 }
206
207 if (unaligned_printk || unaligned_fixup_count == 0) { 214 if (unaligned_printk || unaligned_fixup_count == 0) {
208 pr_info("Process %d/%s: PC %#lx: Fixup of" 215 pr_info("Process %d/%s: PC %#lx: Fixup of"
209 " unaligned %s at %#lx.\n", 216 " unaligned %s at %#lx.\n",
diff --git a/arch/tile/kernel/smp.c b/arch/tile/kernel/smp.c
index a44e103c5a63..91da0f721958 100644
--- a/arch/tile/kernel/smp.c
+++ b/arch/tile/kernel/smp.c
@@ -103,7 +103,7 @@ static void smp_stop_cpu_interrupt(void)
103 set_cpu_online(smp_processor_id(), 0); 103 set_cpu_online(smp_processor_id(), 0);
104 arch_local_irq_disable_all(); 104 arch_local_irq_disable_all();
105 for (;;) 105 for (;;)
106 asm("nap"); 106 asm("nap; nop");
107} 107}
108 108
109/* This function calls the 'stop' function on all other CPUs in the system. */ 109/* This function calls the 'stop' function on all other CPUs in the system. */
@@ -113,6 +113,12 @@ void smp_send_stop(void)
113 send_IPI_allbutself(MSG_TAG_STOP_CPU); 113 send_IPI_allbutself(MSG_TAG_STOP_CPU);
114} 114}
115 115
116/* On panic, just wait; we may get an smp_send_stop() later on. */
117void panic_smp_self_stop(void)
118{
119 while (1)
120 asm("nap; nop");
121}
116 122
117/* 123/*
118 * Dispatch code called from hv_message_intr() for HV_MSG_TILE hv messages. 124 * Dispatch code called from hv_message_intr() for HV_MSG_TILE hv messages.
diff --git a/arch/tile/kernel/stack.c b/arch/tile/kernel/stack.c
index 37ee4d037e0b..b2f44c28dda6 100644
--- a/arch/tile/kernel/stack.c
+++ b/arch/tile/kernel/stack.c
@@ -21,10 +21,12 @@
21#include <linux/stacktrace.h> 21#include <linux/stacktrace.h>
22#include <linux/uaccess.h> 22#include <linux/uaccess.h>
23#include <linux/mmzone.h> 23#include <linux/mmzone.h>
24#include <linux/dcache.h>
25#include <linux/fs.h>
24#include <asm/backtrace.h> 26#include <asm/backtrace.h>
25#include <asm/page.h> 27#include <asm/page.h>
26#include <asm/tlbflush.h>
27#include <asm/ucontext.h> 28#include <asm/ucontext.h>
29#include <asm/switch_to.h>
28#include <asm/sigframe.h> 30#include <asm/sigframe.h>
29#include <asm/stack.h> 31#include <asm/stack.h>
30#include <arch/abi.h> 32#include <arch/abi.h>
@@ -44,72 +46,23 @@ static int in_kernel_stack(struct KBacktraceIterator *kbt, unsigned long sp)
44 return sp >= kstack_base && sp < kstack_base + THREAD_SIZE; 46 return sp >= kstack_base && sp < kstack_base + THREAD_SIZE;
45} 47}
46 48
47/* Is address valid for reading? */
48static int valid_address(struct KBacktraceIterator *kbt, unsigned long address)
49{
50 HV_PTE *l1_pgtable = kbt->pgtable;
51 HV_PTE *l2_pgtable;
52 unsigned long pfn;
53 HV_PTE pte;
54 struct page *page;
55
56 if (l1_pgtable == NULL)
57 return 0; /* can't read user space in other tasks */
58
59#ifdef CONFIG_64BIT
60 /* Find the real l1_pgtable by looking in the l0_pgtable. */
61 pte = l1_pgtable[HV_L0_INDEX(address)];
62 if (!hv_pte_get_present(pte))
63 return 0;
64 pfn = hv_pte_get_pfn(pte);
65 if (pte_huge(pte)) {
66 if (!pfn_valid(pfn)) {
67 pr_err("L0 huge page has bad pfn %#lx\n", pfn);
68 return 0;
69 }
70 return hv_pte_get_present(pte) && hv_pte_get_readable(pte);
71 }
72 page = pfn_to_page(pfn);
73 BUG_ON(PageHighMem(page)); /* No HIGHMEM on 64-bit. */
74 l1_pgtable = (HV_PTE *)pfn_to_kaddr(pfn);
75#endif
76 pte = l1_pgtable[HV_L1_INDEX(address)];
77 if (!hv_pte_get_present(pte))
78 return 0;
79 pfn = hv_pte_get_pfn(pte);
80 if (pte_huge(pte)) {
81 if (!pfn_valid(pfn)) {
82 pr_err("huge page has bad pfn %#lx\n", pfn);
83 return 0;
84 }
85 return hv_pte_get_present(pte) && hv_pte_get_readable(pte);
86 }
87
88 page = pfn_to_page(pfn);
89 if (PageHighMem(page)) {
90 pr_err("L2 page table not in LOWMEM (%#llx)\n",
91 HV_PFN_TO_CPA(pfn));
92 return 0;
93 }
94 l2_pgtable = (HV_PTE *)pfn_to_kaddr(pfn);
95 pte = l2_pgtable[HV_L2_INDEX(address)];
96 return hv_pte_get_present(pte) && hv_pte_get_readable(pte);
97}
98
99/* Callback for backtracer; basically a glorified memcpy */ 49/* Callback for backtracer; basically a glorified memcpy */
100static bool read_memory_func(void *result, unsigned long address, 50static bool read_memory_func(void *result, unsigned long address,
101 unsigned int size, void *vkbt) 51 unsigned int size, void *vkbt)
102{ 52{
103 int retval; 53 int retval;
104 struct KBacktraceIterator *kbt = (struct KBacktraceIterator *)vkbt; 54 struct KBacktraceIterator *kbt = (struct KBacktraceIterator *)vkbt;
55
56 if (address == 0)
57 return 0;
105 if (__kernel_text_address(address)) { 58 if (__kernel_text_address(address)) {
106 /* OK to read kernel code. */ 59 /* OK to read kernel code. */
107 } else if (address >= PAGE_OFFSET) { 60 } else if (address >= PAGE_OFFSET) {
108 /* We only tolerate kernel-space reads of this task's stack */ 61 /* We only tolerate kernel-space reads of this task's stack */
109 if (!in_kernel_stack(kbt, address)) 62 if (!in_kernel_stack(kbt, address))
110 return 0; 63 return 0;
111 } else if (!valid_address(kbt, address)) { 64 } else if (!kbt->is_current) {
112 return 0; /* invalid user-space address */ 65 return 0; /* can't read from other user address spaces */
113 } 66 }
114 pagefault_disable(); 67 pagefault_disable();
115 retval = __copy_from_user_inatomic(result, 68 retval = __copy_from_user_inatomic(result,
@@ -127,6 +80,8 @@ static struct pt_regs *valid_fault_handler(struct KBacktraceIterator* kbt)
127 unsigned long sp = kbt->it.sp; 80 unsigned long sp = kbt->it.sp;
128 struct pt_regs *p; 81 struct pt_regs *p;
129 82
83 if (sp % sizeof(long) != 0)
84 return NULL;
130 if (!in_kernel_stack(kbt, sp)) 85 if (!in_kernel_stack(kbt, sp))
131 return NULL; 86 return NULL;
132 if (!in_kernel_stack(kbt, sp + C_ABI_SAVE_AREA_SIZE + PTREGS_SIZE-1)) 87 if (!in_kernel_stack(kbt, sp + C_ABI_SAVE_AREA_SIZE + PTREGS_SIZE-1))
@@ -169,27 +124,27 @@ static int is_sigreturn(unsigned long pc)
169} 124}
170 125
171/* Return a pt_regs pointer for a valid signal handler frame */ 126/* Return a pt_regs pointer for a valid signal handler frame */
172static struct pt_regs *valid_sigframe(struct KBacktraceIterator* kbt) 127static struct pt_regs *valid_sigframe(struct KBacktraceIterator* kbt,
128 struct rt_sigframe* kframe)
173{ 129{
174 BacktraceIterator *b = &kbt->it; 130 BacktraceIterator *b = &kbt->it;
175 131
176 if (b->pc == VDSO_BASE) { 132 if (b->pc == VDSO_BASE && b->sp < PAGE_OFFSET &&
177 struct rt_sigframe *frame; 133 b->sp % sizeof(long) == 0) {
178 unsigned long sigframe_top = 134 int retval;
179 b->sp + sizeof(struct rt_sigframe) - 1; 135 pagefault_disable();
180 if (!valid_address(kbt, b->sp) || 136 retval = __copy_from_user_inatomic(
181 !valid_address(kbt, sigframe_top)) { 137 kframe, (void __user __force *)b->sp,
182 if (kbt->verbose) 138 sizeof(*kframe));
183 pr_err(" (odd signal: sp %#lx?)\n", 139 pagefault_enable();
184 (unsigned long)(b->sp)); 140 if (retval != 0 ||
141 (unsigned int)(kframe->info.si_signo) >= _NSIG)
185 return NULL; 142 return NULL;
186 }
187 frame = (struct rt_sigframe *)b->sp;
188 if (kbt->verbose) { 143 if (kbt->verbose) {
189 pr_err(" <received signal %d>\n", 144 pr_err(" <received signal %d>\n",
190 frame->info.si_signo); 145 kframe->info.si_signo);
191 } 146 }
192 return (struct pt_regs *)&frame->uc.uc_mcontext; 147 return (struct pt_regs *)&kframe->uc.uc_mcontext;
193 } 148 }
194 return NULL; 149 return NULL;
195} 150}
@@ -202,10 +157,11 @@ static int KBacktraceIterator_is_sigreturn(struct KBacktraceIterator *kbt)
202static int KBacktraceIterator_restart(struct KBacktraceIterator *kbt) 157static int KBacktraceIterator_restart(struct KBacktraceIterator *kbt)
203{ 158{
204 struct pt_regs *p; 159 struct pt_regs *p;
160 struct rt_sigframe kframe;
205 161
206 p = valid_fault_handler(kbt); 162 p = valid_fault_handler(kbt);
207 if (p == NULL) 163 if (p == NULL)
208 p = valid_sigframe(kbt); 164 p = valid_sigframe(kbt, &kframe);
209 if (p == NULL) 165 if (p == NULL)
210 return 0; 166 return 0;
211 backtrace_init(&kbt->it, read_memory_func, kbt, 167 backtrace_init(&kbt->it, read_memory_func, kbt,
@@ -265,41 +221,19 @@ void KBacktraceIterator_init(struct KBacktraceIterator *kbt,
265 221
266 /* 222 /*
267 * Set up callback information. We grab the kernel stack base 223 * Set up callback information. We grab the kernel stack base
268 * so we will allow reads of that address range, and if we're 224 * so we will allow reads of that address range.
269 * asking about the current process we grab the page table
270 * so we can check user accesses before trying to read them.
271 * We flush the TLB to avoid any weird skew issues.
272 */ 225 */
273 is_current = (t == NULL); 226 is_current = (t == NULL || t == current);
274 kbt->is_current = is_current; 227 kbt->is_current = is_current;
275 if (is_current) 228 if (is_current)
276 t = validate_current(); 229 t = validate_current();
277 kbt->task = t; 230 kbt->task = t;
278 kbt->pgtable = NULL;
279 kbt->verbose = 0; /* override in caller if desired */ 231 kbt->verbose = 0; /* override in caller if desired */
280 kbt->profile = 0; /* override in caller if desired */ 232 kbt->profile = 0; /* override in caller if desired */
281 kbt->end = KBT_ONGOING; 233 kbt->end = KBT_ONGOING;
282 kbt->new_context = 0; 234 kbt->new_context = 1;
283 if (is_current) { 235 if (is_current)
284 HV_PhysAddr pgdir_pa = hv_inquire_context().page_table;
285 if (pgdir_pa == (unsigned long)swapper_pg_dir - PAGE_OFFSET) {
286 /*
287 * Not just an optimization: this also allows
288 * this to work at all before va/pa mappings
289 * are set up.
290 */
291 kbt->pgtable = swapper_pg_dir;
292 } else {
293 struct page *page = pfn_to_page(PFN_DOWN(pgdir_pa));
294 if (!PageHighMem(page))
295 kbt->pgtable = __va(pgdir_pa);
296 else
297 pr_err("page table not in LOWMEM"
298 " (%#llx)\n", pgdir_pa);
299 }
300 local_flush_tlb_all();
301 validate_stack(regs); 236 validate_stack(regs);
302 }
303 237
304 if (regs == NULL) { 238 if (regs == NULL) {
305 if (is_current || t->state == TASK_RUNNING) { 239 if (is_current || t->state == TASK_RUNNING) {
@@ -345,6 +279,78 @@ void KBacktraceIterator_next(struct KBacktraceIterator *kbt)
345} 279}
346EXPORT_SYMBOL(KBacktraceIterator_next); 280EXPORT_SYMBOL(KBacktraceIterator_next);
347 281
282static void describe_addr(struct KBacktraceIterator *kbt,
283 unsigned long address,
284 int have_mmap_sem, char *buf, size_t bufsize)
285{
286 struct vm_area_struct *vma;
287 size_t namelen, remaining;
288 unsigned long size, offset, adjust;
289 char *p, *modname;
290 const char *name;
291 int rc;
292
293 /*
294 * Look one byte back for every caller frame (i.e. those that
295 * aren't a new context) so we look up symbol data for the
296 * call itself, not the following instruction, which may be on
297 * a different line (or in a different function).
298 */
299 adjust = !kbt->new_context;
300 address -= adjust;
301
302 if (address >= PAGE_OFFSET) {
303 /* Handle kernel symbols. */
304 BUG_ON(bufsize < KSYM_NAME_LEN);
305 name = kallsyms_lookup(address, &size, &offset,
306 &modname, buf);
307 if (name == NULL) {
308 buf[0] = '\0';
309 return;
310 }
311 namelen = strlen(buf);
312 remaining = (bufsize - 1) - namelen;
313 p = buf + namelen;
314 rc = snprintf(p, remaining, "+%#lx/%#lx ",
315 offset + adjust, size);
316 if (modname && rc < remaining)
317 snprintf(p + rc, remaining - rc, "[%s] ", modname);
318 buf[bufsize-1] = '\0';
319 return;
320 }
321
322 /* If we don't have the mmap_sem, we can't show any more info. */
323 buf[0] = '\0';
324 if (!have_mmap_sem)
325 return;
326
327 /* Find vma info. */
328 vma = find_vma(kbt->task->mm, address);
329 if (vma == NULL || address < vma->vm_start) {
330 snprintf(buf, bufsize, "[unmapped address] ");
331 return;
332 }
333
334 if (vma->vm_file) {
335 char *s;
336 p = d_path(&vma->vm_file->f_path, buf, bufsize);
337 if (IS_ERR(p))
338 p = "?";
339 s = strrchr(p, '/');
340 if (s)
341 p = s+1;
342 } else {
343 p = "anon";
344 }
345
346 /* Generate a string description of the vma info. */
347 namelen = strlen(p);
348 remaining = (bufsize - 1) - namelen;
349 memmove(buf, p, namelen);
350 snprintf(buf + namelen, remaining, "[%lx+%lx] ",
351 vma->vm_start, vma->vm_end - vma->vm_start);
352}
353
348/* 354/*
349 * This method wraps the backtracer's more generic support. 355 * This method wraps the backtracer's more generic support.
350 * It is only invoked from the architecture-specific code; show_stack() 356 * It is only invoked from the architecture-specific code; show_stack()
@@ -353,6 +359,7 @@ EXPORT_SYMBOL(KBacktraceIterator_next);
353void tile_show_stack(struct KBacktraceIterator *kbt, int headers) 359void tile_show_stack(struct KBacktraceIterator *kbt, int headers)
354{ 360{
355 int i; 361 int i;
362 int have_mmap_sem = 0;
356 363
357 if (headers) { 364 if (headers) {
358 /* 365 /*
@@ -369,31 +376,16 @@ void tile_show_stack(struct KBacktraceIterator *kbt, int headers)
369 kbt->verbose = 1; 376 kbt->verbose = 1;
370 i = 0; 377 i = 0;
371 for (; !KBacktraceIterator_end(kbt); KBacktraceIterator_next(kbt)) { 378 for (; !KBacktraceIterator_end(kbt); KBacktraceIterator_next(kbt)) {
372 char *modname;
373 const char *name;
374 unsigned long address = kbt->it.pc;
375 unsigned long offset, size;
376 char namebuf[KSYM_NAME_LEN+100]; 379 char namebuf[KSYM_NAME_LEN+100];
380 unsigned long address = kbt->it.pc;
377 381
378 if (address >= PAGE_OFFSET) 382 /* Try to acquire the mmap_sem as we pass into userspace. */
379 name = kallsyms_lookup(address, &size, &offset, 383 if (address < PAGE_OFFSET && !have_mmap_sem && kbt->task->mm)
380 &modname, namebuf); 384 have_mmap_sem =
381 else 385 down_read_trylock(&kbt->task->mm->mmap_sem);
382 name = NULL; 386
383 387 describe_addr(kbt, address, have_mmap_sem,
384 if (!name) 388 namebuf, sizeof(namebuf));
385 namebuf[0] = '\0';
386 else {
387 size_t namelen = strlen(namebuf);
388 size_t remaining = (sizeof(namebuf) - 1) - namelen;
389 char *p = namebuf + namelen;
390 int rc = snprintf(p, remaining, "+%#lx/%#lx ",
391 offset, size);
392 if (modname && rc < remaining)
393 snprintf(p + rc, remaining - rc,
394 "[%s] ", modname);
395 namebuf[sizeof(namebuf)-1] = '\0';
396 }
397 389
398 pr_err(" frame %d: 0x%lx %s(sp 0x%lx)\n", 390 pr_err(" frame %d: 0x%lx %s(sp 0x%lx)\n",
399 i++, address, namebuf, (unsigned long)(kbt->it.sp)); 391 i++, address, namebuf, (unsigned long)(kbt->it.sp));
@@ -408,6 +400,8 @@ void tile_show_stack(struct KBacktraceIterator *kbt, int headers)
408 pr_err("Stack dump stopped; next frame identical to this one\n"); 400 pr_err("Stack dump stopped; next frame identical to this one\n");
409 if (headers) 401 if (headers)
410 pr_err("Stack dump complete\n"); 402 pr_err("Stack dump complete\n");
403 if (have_mmap_sem)
404 up_read(&kbt->task->mm->mmap_sem);
411} 405}
412EXPORT_SYMBOL(tile_show_stack); 406EXPORT_SYMBOL(tile_show_stack);
413 407
diff --git a/arch/tile/kernel/traps.c b/arch/tile/kernel/traps.c
index 2bb6602a1ee7..73cff814ac57 100644
--- a/arch/tile/kernel/traps.c
+++ b/arch/tile/kernel/traps.c
@@ -200,7 +200,7 @@ void __kprobes do_trap(struct pt_regs *regs, int fault_num,
200{ 200{
201 siginfo_t info = { 0 }; 201 siginfo_t info = { 0 };
202 int signo, code; 202 int signo, code;
203 unsigned long address; 203 unsigned long address = 0;
204 bundle_bits instr; 204 bundle_bits instr;
205 205
206 /* Re-enable interrupts. */ 206 /* Re-enable interrupts. */
@@ -223,6 +223,10 @@ void __kprobes do_trap(struct pt_regs *regs, int fault_num,
223 } 223 }
224 224
225 switch (fault_num) { 225 switch (fault_num) {
226 case INT_MEM_ERROR:
227 signo = SIGBUS;
228 code = BUS_OBJERR;
229 break;
226 case INT_ILL: 230 case INT_ILL:
227 if (copy_from_user(&instr, (void __user *)regs->pc, 231 if (copy_from_user(&instr, (void __user *)regs->pc,
228 sizeof(instr))) { 232 sizeof(instr))) {
@@ -289,7 +293,10 @@ void __kprobes do_trap(struct pt_regs *regs, int fault_num,
289 address = regs->pc; 293 address = regs->pc;
290 break; 294 break;
291#ifdef __tilegx__ 295#ifdef __tilegx__
292 case INT_ILL_TRANS: 296 case INT_ILL_TRANS: {
297 /* Avoid a hardware erratum with the return address stack. */
298 fill_ra_stack();
299
293 signo = SIGSEGV; 300 signo = SIGSEGV;
294 code = SEGV_MAPERR; 301 code = SEGV_MAPERR;
295 if (reason & SPR_ILL_TRANS_REASON__I_STREAM_VA_RMASK) 302 if (reason & SPR_ILL_TRANS_REASON__I_STREAM_VA_RMASK)
@@ -297,6 +304,7 @@ void __kprobes do_trap(struct pt_regs *regs, int fault_num,
297 else 304 else
298 address = 0; /* FIXME: GX: single-step for address */ 305 address = 0; /* FIXME: GX: single-step for address */
299 break; 306 break;
307 }
300#endif 308#endif
301 default: 309 default:
302 panic("Unexpected do_trap interrupt number %d", fault_num); 310 panic("Unexpected do_trap interrupt number %d", fault_num);
@@ -308,7 +316,8 @@ void __kprobes do_trap(struct pt_regs *regs, int fault_num,
308 info.si_addr = (void __user *)address; 316 info.si_addr = (void __user *)address;
309 if (signo == SIGILL) 317 if (signo == SIGILL)
310 info.si_trapno = fault_num; 318 info.si_trapno = fault_num;
311 trace_unhandled_signal("trap", regs, address, signo); 319 if (signo != SIGTRAP)
320 trace_unhandled_signal("trap", regs, address, signo);
312 force_sig_info(signo, &info, current); 321 force_sig_info(signo, &info, current);
313} 322}
314 323
diff --git a/arch/tile/lib/Makefile b/arch/tile/lib/Makefile
index 0c26086ecbef..985f59858234 100644
--- a/arch/tile/lib/Makefile
+++ b/arch/tile/lib/Makefile
@@ -7,6 +7,7 @@ lib-y = cacheflush.o checksum.o cpumask.o delay.o uaccess.o \
7 strchr_$(BITS).o strlen_$(BITS).o 7 strchr_$(BITS).o strlen_$(BITS).o
8 8
9ifeq ($(CONFIG_TILEGX),y) 9ifeq ($(CONFIG_TILEGX),y)
10CFLAGS_REMOVE_memcpy_user_64.o = -fno-omit-frame-pointer
10lib-y += memcpy_user_64.o 11lib-y += memcpy_user_64.o
11else 12else
12lib-y += atomic_32.o atomic_asm_32.o memcpy_tile64.o 13lib-y += atomic_32.o atomic_asm_32.o memcpy_tile64.o
diff --git a/arch/tile/lib/cacheflush.c b/arch/tile/lib/cacheflush.c
index 8928aace7a64..db4fb89e12d8 100644
--- a/arch/tile/lib/cacheflush.c
+++ b/arch/tile/lib/cacheflush.c
@@ -39,7 +39,21 @@ void finv_buffer_remote(void *buffer, size_t size, int hfh)
39{ 39{
40 char *p, *base; 40 char *p, *base;
41 size_t step_size, load_count; 41 size_t step_size, load_count;
42
43 /*
44 * On TILEPro the striping granularity is a fixed 8KB; on
45 * TILE-Gx it is configurable, and we rely on the fact that
46 * the hypervisor always configures maximum striping, so that
47 * bits 9 and 10 of the PA are part of the stripe function, so
48 * every 512 bytes we hit a striping boundary.
49 *
50 */
51#ifdef __tilegx__
52 const unsigned long STRIPE_WIDTH = 512;
53#else
42 const unsigned long STRIPE_WIDTH = 8192; 54 const unsigned long STRIPE_WIDTH = 8192;
55#endif
56
43#ifdef __tilegx__ 57#ifdef __tilegx__
44 /* 58 /*
45 * On TILE-Gx, we must disable the dstream prefetcher before doing 59 * On TILE-Gx, we must disable the dstream prefetcher before doing
@@ -74,7 +88,7 @@ void finv_buffer_remote(void *buffer, size_t size, int hfh)
74 * memory, that one load would be sufficient, but since we may 88 * memory, that one load would be sufficient, but since we may
75 * be, we also need to back up to the last load issued to 89 * be, we also need to back up to the last load issued to
76 * another memory controller, which would be the point where 90 * another memory controller, which would be the point where
77 * we crossed an 8KB boundary (the granularity of striping 91 * we crossed a "striping" boundary (the granularity of striping
78 * across memory controllers). Keep backing up and doing this 92 * across memory controllers). Keep backing up and doing this
79 * until we are before the beginning of the buffer, or have 93 * until we are before the beginning of the buffer, or have
80 * hit all the controllers. 94 * hit all the controllers.
@@ -88,12 +102,22 @@ void finv_buffer_remote(void *buffer, size_t size, int hfh)
88 * every cache line on a full memory stripe on each 102 * every cache line on a full memory stripe on each
89 * controller" that we simply do that, to simplify the logic. 103 * controller" that we simply do that, to simplify the logic.
90 * 104 *
91 * FIXME: See bug 9535 for some issues with this code. 105 * On TILE-Gx the hash-for-home function is much more complex,
106 * with the upshot being we can't readily guarantee we have
107 * hit both entries in the 128-entry AMT that were hit by any
108 * load in the entire range, so we just re-load them all.
109 * With larger buffers, we may want to consider using a hypervisor
110 * trap to issue loads directly to each hash-for-home tile for
111 * each controller (doing it from Linux would trash the TLB).
92 */ 112 */
93 if (hfh) { 113 if (hfh) {
94 step_size = L2_CACHE_BYTES; 114 step_size = L2_CACHE_BYTES;
115#ifdef __tilegx__
116 load_count = (size + L2_CACHE_BYTES - 1) / L2_CACHE_BYTES;
117#else
95 load_count = (STRIPE_WIDTH / L2_CACHE_BYTES) * 118 load_count = (STRIPE_WIDTH / L2_CACHE_BYTES) *
96 (1 << CHIP_LOG_NUM_MSHIMS()); 119 (1 << CHIP_LOG_NUM_MSHIMS());
120#endif
97 } else { 121 } else {
98 step_size = STRIPE_WIDTH; 122 step_size = STRIPE_WIDTH;
99 load_count = (1 << CHIP_LOG_NUM_MSHIMS()); 123 load_count = (1 << CHIP_LOG_NUM_MSHIMS());
@@ -109,7 +133,7 @@ void finv_buffer_remote(void *buffer, size_t size, int hfh)
109 133
110 /* Figure out how far back we need to go. */ 134 /* Figure out how far back we need to go. */
111 base = p - (step_size * (load_count - 2)); 135 base = p - (step_size * (load_count - 2));
112 if ((long)base < (long)buffer) 136 if ((unsigned long)base < (unsigned long)buffer)
113 base = buffer; 137 base = buffer;
114 138
115 /* 139 /*
diff --git a/arch/tile/lib/memcpy_user_64.c b/arch/tile/lib/memcpy_user_64.c
index 4763b3aff1cc..37440caa7370 100644
--- a/arch/tile/lib/memcpy_user_64.c
+++ b/arch/tile/lib/memcpy_user_64.c
@@ -14,7 +14,13 @@
14 * Do memcpy(), but trap and return "n" when a load or store faults. 14 * Do memcpy(), but trap and return "n" when a load or store faults.
15 * 15 *
16 * Note: this idiom only works when memcpy() compiles to a leaf function. 16 * Note: this idiom only works when memcpy() compiles to a leaf function.
17 * If "sp" is updated during memcpy, the "jrp lr" will be incorrect. 17 * Here leaf function not only means it does not have calls, but also
18 * requires no stack operations (sp, stack frame pointer) and no
19 * use of callee-saved registers, else "jrp lr" will be incorrect since
20 * unwinding stack frame is bypassed. Since memcpy() is not complex so
21 * these conditions are satisfied here, but we need to be careful when
22 * modifying this file. This is not a clean solution but is the best
23 * one so far.
18 * 24 *
19 * Also note that we are capturing "n" from the containing scope here. 25 * Also note that we are capturing "n" from the containing scope here.
20 */ 26 */
diff --git a/arch/tile/lib/spinlock_common.h b/arch/tile/lib/spinlock_common.h
index c10109809132..6ac37509faca 100644
--- a/arch/tile/lib/spinlock_common.h
+++ b/arch/tile/lib/spinlock_common.h
@@ -60,5 +60,5 @@ static void delay_backoff(int iterations)
60 loops += __insn_crc32_32(stack_pointer, get_cycles_low()) & 60 loops += __insn_crc32_32(stack_pointer, get_cycles_low()) &
61 (loops - 1); 61 (loops - 1);
62 62
63 relax(1 << exponent); 63 relax(loops);
64} 64}
diff --git a/arch/tile/mm/fault.c b/arch/tile/mm/fault.c
index cba30e9547b4..22e58f51ed23 100644
--- a/arch/tile/mm/fault.c
+++ b/arch/tile/mm/fault.c
@@ -130,7 +130,7 @@ static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
130} 130}
131 131
132/* 132/*
133 * Handle a fault on the vmalloc or module mapping area 133 * Handle a fault on the vmalloc area.
134 */ 134 */
135static inline int vmalloc_fault(pgd_t *pgd, unsigned long address) 135static inline int vmalloc_fault(pgd_t *pgd, unsigned long address)
136{ 136{
@@ -203,9 +203,14 @@ static pgd_t *get_current_pgd(void)
203 * interrupt or a critical region, and must do as little as possible. 203 * interrupt or a critical region, and must do as little as possible.
204 * Similarly, we can't use atomic ops here, since we may be handling a 204 * Similarly, we can't use atomic ops here, since we may be handling a
205 * fault caused by an atomic op access. 205 * fault caused by an atomic op access.
206 *
207 * If we find a migrating PTE while we're in an NMI context, and we're
208 * at a PC that has a registered exception handler, we don't wait,
209 * since this thread may (e.g.) have been interrupted while migrating
210 * its own stack, which would then cause us to self-deadlock.
206 */ 211 */
207static int handle_migrating_pte(pgd_t *pgd, int fault_num, 212static int handle_migrating_pte(pgd_t *pgd, int fault_num,
208 unsigned long address, 213 unsigned long address, unsigned long pc,
209 int is_kernel_mode, int write) 214 int is_kernel_mode, int write)
210{ 215{
211 pud_t *pud; 216 pud_t *pud;
@@ -227,6 +232,8 @@ static int handle_migrating_pte(pgd_t *pgd, int fault_num,
227 pte_offset_kernel(pmd, address); 232 pte_offset_kernel(pmd, address);
228 pteval = *pte; 233 pteval = *pte;
229 if (pte_migrating(pteval)) { 234 if (pte_migrating(pteval)) {
235 if (in_nmi() && search_exception_tables(pc))
236 return 0;
230 wait_for_migration(pte); 237 wait_for_migration(pte);
231 return 1; 238 return 1;
232 } 239 }
@@ -300,7 +307,7 @@ static int handle_page_fault(struct pt_regs *regs,
300 * rather than trying to patch up the existing PTE. 307 * rather than trying to patch up the existing PTE.
301 */ 308 */
302 pgd = get_current_pgd(); 309 pgd = get_current_pgd();
303 if (handle_migrating_pte(pgd, fault_num, address, 310 if (handle_migrating_pte(pgd, fault_num, address, regs->pc,
304 is_kernel_mode, write)) 311 is_kernel_mode, write))
305 return 1; 312 return 1;
306 313
@@ -335,9 +342,12 @@ static int handle_page_fault(struct pt_regs *regs,
335 /* 342 /*
336 * If we're trying to touch user-space addresses, we must 343 * If we're trying to touch user-space addresses, we must
337 * be either at PL0, or else with interrupts enabled in the 344 * be either at PL0, or else with interrupts enabled in the
338 * kernel, so either way we can re-enable interrupts here. 345 * kernel, so either way we can re-enable interrupts here
346 * unless we are doing atomic access to user space with
347 * interrupts disabled.
339 */ 348 */
340 local_irq_enable(); 349 if (!(regs->flags & PT_FLAGS_DISABLE_IRQ))
350 local_irq_enable();
341 351
342 mm = tsk->mm; 352 mm = tsk->mm;
343 353
@@ -665,7 +675,7 @@ struct intvec_state do_page_fault_ics(struct pt_regs *regs, int fault_num,
665 */ 675 */
666 if (fault_num == INT_DTLB_ACCESS) 676 if (fault_num == INT_DTLB_ACCESS)
667 write = 1; 677 write = 1;
668 if (handle_migrating_pte(pgd, fault_num, address, 1, write)) 678 if (handle_migrating_pte(pgd, fault_num, address, pc, 1, write))
669 return state; 679 return state;
670 680
671 /* Return zero so that we continue on with normal fault handling. */ 681 /* Return zero so that we continue on with normal fault handling. */
diff --git a/arch/tile/mm/homecache.c b/arch/tile/mm/homecache.c
index 1cc6ae477c98..499f73770b05 100644
--- a/arch/tile/mm/homecache.c
+++ b/arch/tile/mm/homecache.c
@@ -394,6 +394,7 @@ int page_home(struct page *page)
394 return pte_to_home(*virt_to_pte(NULL, kva)); 394 return pte_to_home(*virt_to_pte(NULL, kva));
395 } 395 }
396} 396}
397EXPORT_SYMBOL(page_home);
397 398
398void homecache_change_page_home(struct page *page, int order, int home) 399void homecache_change_page_home(struct page *page, int order, int home)
399{ 400{
diff --git a/arch/tile/mm/init.c b/arch/tile/mm/init.c
index 830c4908ea76..6a9d20ddc34f 100644
--- a/arch/tile/mm/init.c
+++ b/arch/tile/mm/init.c
@@ -254,11 +254,6 @@ static pgprot_t __init init_pgprot(ulong address)
254 return construct_pgprot(PAGE_KERNEL_RO, PAGE_HOME_IMMUTABLE); 254 return construct_pgprot(PAGE_KERNEL_RO, PAGE_HOME_IMMUTABLE);
255 } 255 }
256 256
257 /* As a performance optimization, keep the boot init stack here. */
258 if (address >= (ulong)&init_thread_union &&
259 address < (ulong)&init_thread_union + THREAD_SIZE)
260 return construct_pgprot(PAGE_KERNEL, smp_processor_id());
261
262#ifndef __tilegx__ 257#ifndef __tilegx__
263#if !ATOMIC_LOCKS_FOUND_VIA_TABLE() 258#if !ATOMIC_LOCKS_FOUND_VIA_TABLE()
264 /* Force the atomic_locks[] array page to be hash-for-home. */ 259 /* Force the atomic_locks[] array page to be hash-for-home. */
@@ -557,6 +552,7 @@ static void __init kernel_physical_mapping_init(pgd_t *pgd_base)
557 552
558 address = MEM_SV_INTRPT; 553 address = MEM_SV_INTRPT;
559 pmd = get_pmd(pgtables, address); 554 pmd = get_pmd(pgtables, address);
555 pfn = 0; /* code starts at PA 0 */
560 if (ktext_small) { 556 if (ktext_small) {
561 /* Allocate an L2 PTE for the kernel text */ 557 /* Allocate an L2 PTE for the kernel text */
562 int cpu = 0; 558 int cpu = 0;
@@ -579,10 +575,15 @@ static void __init kernel_physical_mapping_init(pgd_t *pgd_base)
579 } 575 }
580 576
581 BUG_ON(address != (unsigned long)_stext); 577 BUG_ON(address != (unsigned long)_stext);
582 pfn = 0; /* code starts at PA 0 */ 578 pte = NULL;
583 pte = alloc_pte(); 579 for (; address < (unsigned long)_einittext;
584 for (pte_ofs = 0; address < (unsigned long)_einittext; 580 pfn++, address += PAGE_SIZE) {
585 pfn++, pte_ofs++, address += PAGE_SIZE) { 581 pte_ofs = pte_index(address);
582 if (pte_ofs == 0) {
583 if (pte)
584 assign_pte(pmd++, pte);
585 pte = alloc_pte();
586 }
586 if (!ktext_local) { 587 if (!ktext_local) {
587 prot = set_remote_cache_cpu(prot, cpu); 588 prot = set_remote_cache_cpu(prot, cpu);
588 cpu = cpumask_next(cpu, &ktext_mask); 589 cpu = cpumask_next(cpu, &ktext_mask);
@@ -591,7 +592,8 @@ static void __init kernel_physical_mapping_init(pgd_t *pgd_base)
591 } 592 }
592 pte[pte_ofs] = pfn_pte(pfn, prot); 593 pte[pte_ofs] = pfn_pte(pfn, prot);
593 } 594 }
594 assign_pte(pmd, pte); 595 if (pte)
596 assign_pte(pmd, pte);
595 } else { 597 } else {
596 pte_t pteval = pfn_pte(0, PAGE_KERNEL_EXEC); 598 pte_t pteval = pfn_pte(0, PAGE_KERNEL_EXEC);
597 pteval = pte_mkhuge(pteval); 599 pteval = pte_mkhuge(pteval);
@@ -614,7 +616,9 @@ static void __init kernel_physical_mapping_init(pgd_t *pgd_base)
614 else 616 else
615 pteval = hv_pte_set_mode(pteval, 617 pteval = hv_pte_set_mode(pteval,
616 HV_PTE_MODE_CACHE_NO_L3); 618 HV_PTE_MODE_CACHE_NO_L3);
617 *(pte_t *)pmd = pteval; 619 for (; address < (unsigned long)_einittext;
620 pfn += PFN_DOWN(HPAGE_SIZE), address += HPAGE_SIZE)
621 *(pte_t *)(pmd++) = pfn_pte(pfn, pteval);
618 } 622 }
619 623
620 /* Set swapper_pgprot here so it is flushed to memory right away. */ 624 /* Set swapper_pgprot here so it is flushed to memory right away. */
diff --git a/arch/tile/mm/pgtable.c b/arch/tile/mm/pgtable.c
index 87303693a072..2410aa899b3e 100644
--- a/arch/tile/mm/pgtable.c
+++ b/arch/tile/mm/pgtable.c
@@ -177,14 +177,10 @@ void shatter_huge_page(unsigned long addr)
177 if (!pmd_huge_page(*pmd)) 177 if (!pmd_huge_page(*pmd))
178 return; 178 return;
179 179
180 /* 180 spin_lock_irqsave(&init_mm.page_table_lock, flags);
181 * Grab the pgd_lock, since we may need it to walk the pgd_list,
182 * and since we need some kind of lock here to avoid races.
183 */
184 spin_lock_irqsave(&pgd_lock, flags);
185 if (!pmd_huge_page(*pmd)) { 181 if (!pmd_huge_page(*pmd)) {
186 /* Lost the race to convert the huge page. */ 182 /* Lost the race to convert the huge page. */
187 spin_unlock_irqrestore(&pgd_lock, flags); 183 spin_unlock_irqrestore(&init_mm.page_table_lock, flags);
188 return; 184 return;
189 } 185 }
190 186
@@ -194,6 +190,7 @@ void shatter_huge_page(unsigned long addr)
194 190
195#ifdef __PAGETABLE_PMD_FOLDED 191#ifdef __PAGETABLE_PMD_FOLDED
196 /* Walk every pgd on the system and update the pmd there. */ 192 /* Walk every pgd on the system and update the pmd there. */
193 spin_lock(&pgd_lock);
197 list_for_each(pos, &pgd_list) { 194 list_for_each(pos, &pgd_list) {
198 pmd_t *copy_pmd; 195 pmd_t *copy_pmd;
199 pgd = list_to_pgd(pos) + pgd_index(addr); 196 pgd = list_to_pgd(pos) + pgd_index(addr);
@@ -201,6 +198,7 @@ void shatter_huge_page(unsigned long addr)
201 copy_pmd = pmd_offset(pud, addr); 198 copy_pmd = pmd_offset(pud, addr);
202 __set_pmd(copy_pmd, *pmd); 199 __set_pmd(copy_pmd, *pmd);
203 } 200 }
201 spin_unlock(&pgd_lock);
204#endif 202#endif
205 203
206 /* Tell every cpu to notice the change. */ 204 /* Tell every cpu to notice the change. */
@@ -208,7 +206,7 @@ void shatter_huge_page(unsigned long addr)
208 cpu_possible_mask, NULL, 0); 206 cpu_possible_mask, NULL, 0);
209 207
210 /* Hold the lock until the TLB flush is finished to avoid races. */ 208 /* Hold the lock until the TLB flush is finished to avoid races. */
211 spin_unlock_irqrestore(&pgd_lock, flags); 209 spin_unlock_irqrestore(&init_mm.page_table_lock, flags);
212} 210}
213 211
214/* 212/*
@@ -217,9 +215,13 @@ void shatter_huge_page(unsigned long addr)
217 * against pageattr.c; it is the unique case in which a valid change 215 * against pageattr.c; it is the unique case in which a valid change
218 * of kernel pagetables can't be lazily synchronized by vmalloc faults. 216 * of kernel pagetables can't be lazily synchronized by vmalloc faults.
219 * vmalloc faults work because attached pagetables are never freed. 217 * vmalloc faults work because attached pagetables are never freed.
220 * The locking scheme was chosen on the basis of manfred's 218 *
221 * recommendations and having no core impact whatsoever. 219 * The lock is always taken with interrupts disabled, unlike on x86
222 * -- wli 220 * and other platforms, because we need to take the lock in
221 * shatter_huge_page(), which may be called from an interrupt context.
222 * We are not at risk from the tlbflush IPI deadlock that was seen on
223 * x86, since we use the flush_remote() API to have the hypervisor do
224 * the TLB flushes regardless of irq disabling.
223 */ 225 */
224DEFINE_SPINLOCK(pgd_lock); 226DEFINE_SPINLOCK(pgd_lock);
225LIST_HEAD(pgd_list); 227LIST_HEAD(pgd_list);
@@ -469,10 +471,18 @@ void __set_pte(pte_t *ptep, pte_t pte)
469 471
470void set_pte(pte_t *ptep, pte_t pte) 472void set_pte(pte_t *ptep, pte_t pte)
471{ 473{
472 struct page *page = pfn_to_page(pte_pfn(pte)); 474 if (pte_present(pte) &&
473 475 (!CHIP_HAS_MMIO() || hv_pte_get_mode(pte) != HV_PTE_MODE_MMIO)) {
474 /* Update the home of a PTE if necessary */ 476 /* The PTE actually references physical memory. */
475 pte = pte_set_home(pte, page_home(page)); 477 unsigned long pfn = pte_pfn(pte);
478 if (pfn_valid(pfn)) {
479 /* Update the home of the PTE from the struct page. */
480 pte = pte_set_home(pte, page_home(pfn_to_page(pfn)));
481 } else if (hv_pte_get_mode(pte) == 0) {
482 /* remap_pfn_range(), etc, must supply PTE mode. */
483 panic("set_pte(): out-of-range PFN and mode 0\n");
484 }
485 }
476 486
477 __set_pte(ptep, pte); 487 __set_pte(ptep, pte);
478} 488}
generated by cgit v1.2.2 (git 2.25.0) at 2025-01-16 14:39:10 -0500