5 files changed, 98 insertions, 115 deletions

diff --git a/arch/sparc64/kernel/dtlb_base.S b/arch/sparc64/kernel/dtlb_base.S
index 702d349c1e88..6528786840c0 100644
--- a/arch/sparc64/kernel/dtlb_base.S
+++ b/arch/sparc64/kernel/dtlb_base.S
@@ -53,19 +53,18 @@
  * be guaranteed to be 0 ... mmu_context.h does guarantee this
  * by only using 10 bits in the hwcontext value.
  */
-#define CREATE_VPTE_OFFSET1(r1, r2)
+#define CREATE_VPTE_OFFSET1(r1, r2) nop
 #define CREATE_VPTE_OFFSET2(r1, r2) \
                                 srax    r1, 10, r2
-#define CREATE_VPTE_NOP         nop
 #else
 #define CREATE_VPTE_OFFSET1(r1, r2) \
                                 srax    r1, PAGE_SHIFT, r2
 #define CREATE_VPTE_OFFSET2(r1, r2) \
                                 sllx    r2, 3, r2
-#define CREATE_VPTE_NOP
 #endif
 
 /* DTLB ** ICACHE line 1: Quick user TLB misses         */
+        mov             TLB_SFSR, %g1
         ldxa            [%g1 + %g1] ASI_DMMU, %g4       ! Get TAG_ACCESS
         andcc           %g4, TAG_CONTEXT_BITS, %g0      ! From Nucleus?
 from_tl1_trap:
@@ -74,18 +73,16 @@ from_tl1_trap:
         be,pn           %xcc, kvmap                     ! Yep, special processing
          CREATE_VPTE_OFFSET2(%g4, %g6)                  ! Create VPTE offset
         cmp             %g5, 4                          ! Last trap level?
-        be,pn           %xcc, longpath                  ! Yep, cannot risk VPTE miss
-         nop                                            ! delay slot
 
 /* DTLB ** ICACHE line 2: User finish + quick kernel TLB misses */
+        be,pn           %xcc, longpath                  ! Yep, cannot risk VPTE miss
+         nop                                            ! delay slot
         ldxa            [%g3 + %g6] ASI_S, %g5          ! Load VPTE
 1:      brgez,pn        %g5, longpath                   ! Invalid, branch out
          nop                                            ! Delay-slot
 9:      stxa            %g5, [%g0] ASI_DTLB_DATA_IN     ! Reload TLB
         retry                                           ! Trap return
         nop
-        nop
-        nop
 
 /* DTLB ** ICACHE line 3: winfixups+real_faults         */
 longpath:
@@ -106,8 +103,7 @@ longpath:
         nop
         nop
         nop
-        CREATE_VPTE_NOP
+        nop
 
 #undef CREATE_VPTE_OFFSET1
 #undef CREATE_VPTE_OFFSET2
-#undef CREATE_VPTE_NOP
diff --git a/arch/sparc64/kernel/dtlb_prot.S b/arch/sparc64/kernel/dtlb_prot.S
index d848bb7374bb..e0a920162604 100644
--- a/arch/sparc64/kernel/dtlb_prot.S
+++ b/arch/sparc64/kernel/dtlb_prot.S
@@ -14,14 +14,14 @@
  */
 
 /* PROT ** ICACHE line 1: User DTLB protection trap     */
-        stxa            %g0, [%g1] ASI_DMMU             ! Clear SFSR FaultValid bit
-        membar          #Sync                           ! Synchronize ASI stores
-        rdpr            %pstate, %g5                    ! Move into alternate globals
+        mov             TLB_SFSR, %g1
+        stxa            %g0, [%g1] ASI_DMMU             ! Clear FaultValid bit
+        membar          #Sync                           ! Synchronize stores
+        rdpr            %pstate, %g5                    ! Move into alt-globals
         wrpr            %g5, PSTATE_AG|PSTATE_MG, %pstate
-        rdpr            %tl, %g1                        ! Need to do a winfixup?
+        rdpr            %tl, %g1                        ! Need a winfixup?
         cmp             %g1, 1                          ! Trap level >1?
-        mov             TLB_TAG_ACCESS, %g4             ! Prepare reload of vaddr
-        nop
+        mov             TLB_TAG_ACCESS, %g4             ! For reload of vaddr
 
 /* PROT ** ICACHE line 2: More real fault processing */
         bgu,pn          %xcc, winfix_trampoline         ! Yes, perform winfixup
diff --git a/arch/sparc64/kernel/head.S b/arch/sparc64/kernel/head.S
index f1dcdf8f7433..b49dcd4504b0 100644
--- a/arch/sparc64/kernel/head.S
+++ b/arch/sparc64/kernel/head.S
@@ -28,19 +28,14 @@
 #include <asm/mmu.h>
         
 /* This section from from _start to sparc64_boot_end should fit into
- * 0x0000.0000.0040.4000 to 0x0000.0000.0040.8000 and will be sharing space
- * with bootup_user_stack, which is from 0x0000.0000.0040.4000 to
- * 0x0000.0000.0040.6000 and empty_bad_page, which is from
- * 0x0000.0000.0040.6000 to 0x0000.0000.0040.8000. 
+ * 0x0000000000404000 to 0x0000000000408000.
  */
-
         .text
         .globl  start, _start, stext, _stext
 _start:
 start:
 _stext:
 stext:
-bootup_user_stack:
 ! 0x0000000000404000
         b       sparc64_boot
          flushw                                 /* Flush register file.      */
@@ -191,8 +186,9 @@ prom_boot_mapping_phys_low:
         stx     %l3, [%sp + 2047 + 128 + 0x10]  ! num_rets, 5
         stx     %l2, [%sp + 2047 + 128 + 0x18]  ! arg1: "translate"
         stx     %l5, [%sp + 2047 + 128 + 0x20]  ! arg2: prom_mmu_ihandle_cache
-        srlx    %l0, 22, %l3
-        sllx    %l3, 22, %l3
+        /* PAGE align */
+        srlx    %l0, 13, %l3
+        sllx    %l3, 13, %l3
         stx     %l3, [%sp + 2047 + 128 + 0x28]  ! arg3: vaddr, our PC
         stx     %g0, [%sp + 2047 + 128 + 0x30]  ! res1
         stx     %g0, [%sp + 2047 + 128 + 0x38]  ! res2
@@ -211,6 +207,9 @@ prom_boot_mapping_phys_low:
         ldx     [%sp + 2047 + 128 + 0x48], %l2  ! physaddr high
         stx     %l2, [%l4 + 0x0]
         ldx     [%sp + 2047 + 128 + 0x50], %l3  ! physaddr low
+        /* 4MB align */
+        srlx    %l3, 22, %l3
+        sllx    %l3, 22, %l3
         stx     %l3, [%l4 + 0x8]
 
         /* Leave service as-is, "call-method" */
@@ -388,31 +387,30 @@ tlb_fixup_done:
          * former does use this code, the latter does not yet due
          * to some complexities.  That should be fixed up at some
          * point.
+         *
+         * There used to be enormous complexity wrt. transferring
+         * over from the firwmare's trap table to the Linux kernel's.
+         * For example, there was a chicken & egg problem wrt. building
+         * the OBP page tables, yet needing to be on the Linux kernel
+         * trap table (to translate PAGE_OFFSET addresses) in order to
+         * do that.
+         *
+         * We now handle OBP tlb misses differently, via linear lookups
+         * into the prom_trans[] array.  So that specific problem no
+         * longer exists.  Yet, unfortunately there are still some issues
+         * preventing trampoline.S from using this code... ho hum.
          */
         .globl  setup_trap_table
 setup_trap_table:
         save    %sp, -192, %sp
 
-        /* Force interrupts to be disabled.  Transferring over to
-         * the Linux trap table is a very delicate operation.
-         * Until we are actually on the Linux trap table, we cannot
-         * get the PAGE_OFFSET linear mappings translated.  We need
-         * that mapping to be setup in order to initialize the firmware
-         * page tables.
-         *
-         * So there is this window of time, from the return from
-         * prom_set_trap_table() until inherit_prom_mappings_post()
-         * (in arch/sparc64/mm/init.c) completes, during which no
-         * firmware address space accesses can be made.
-         */
+        /* Force interrupts to be disabled. */
         rdpr    %pstate, %o1
         andn    %o1, PSTATE_IE, %o1
         wrpr    %o1, 0x0, %pstate
         wrpr    %g0, 15, %pil
 
-        /* Ok, now make the final valid firmware call to jump over
-         * to the Linux trap table.
-         */
+        /* Make the firmware call to jump over to the Linux trap table.  */
         call    prom_set_trap_table
          sethi  %hi(sparc64_ttable_tl0), %o0
 
@@ -536,15 +534,21 @@ setup_tba:	/* i0 = is_starfire */
 
         ret
          restore
+sparc64_boot_end:
+
+#include "systbls.S"
+#include "ktlb.S"
+#include "etrap.S"
+#include "rtrap.S"
+#include "winfixup.S"
+#include "entry.S"
 
 /*
- * The following skips make sure the trap table in ttable.S is aligned
+ * The following skip makes sure the trap table in ttable.S is aligned
  * on a 32K boundary as required by the v9 specs for TBA register.
  */
-sparc64_boot_end:
-        .skip   0x2000 + _start - sparc64_boot_end
-bootup_user_stack_end:
-        .skip   0x2000
+1:
+        .skip   0x4000 + _start - 1b
 
 #ifdef CONFIG_SBUS
 /* This is just a hack to fool make depend config.h discovering
@@ -556,15 +560,6 @@ bootup_user_stack_end:
 ! 0x0000000000408000
 
 #include "ttable.S"
-#include "systbls.S"
-#include "ktlb.S"
-#include "etrap.S"
-#include "rtrap.S"
-#include "winfixup.S"
-#include "entry.S"
-
-        /* This is just anal retentiveness on my part... */
-        .align  16384
 
         .data
         .align  8
diff --git a/arch/sparc64/kernel/itlb_base.S b/arch/sparc64/kernel/itlb_base.S
index b5e32dfa4fbc..4951ff8f6877 100644
--- a/arch/sparc64/kernel/itlb_base.S
+++ b/arch/sparc64/kernel/itlb_base.S
@@ -15,14 +15,12 @@
  */
 #define CREATE_VPTE_OFFSET1(r1, r2) \
                                 srax    r1, 10, r2
-#define CREATE_VPTE_OFFSET2(r1, r2)
-#define CREATE_VPTE_NOP         nop
+#define CREATE_VPTE_OFFSET2(r1, r2) nop
 #else /* PAGE_SHIFT */
 #define CREATE_VPTE_OFFSET1(r1, r2) \
                                 srax    r1, PAGE_SHIFT, r2
 #define CREATE_VPTE_OFFSET2(r1, r2) \
                                 sllx    r2, 3, r2
-#define CREATE_VPTE_NOP
 #endif /* PAGE_SHIFT */
 
 
@@ -36,6 +34,7 @@
  */
 
 /* ITLB ** ICACHE line 1: Quick user TLB misses         */
+        mov             TLB_SFSR, %g1
         ldxa            [%g1 + %g1] ASI_IMMU, %g4       ! Get TAG_ACCESS
         CREATE_VPTE_OFFSET1(%g4, %g6)                   ! Create VPTE offset
         CREATE_VPTE_OFFSET2(%g4, %g6)                   ! Create VPTE offset
@@ -43,41 +42,38 @@
 1:      brgez,pn        %g5, 3f                         ! Not valid, branch out
          sethi          %hi(_PAGE_EXEC), %g4            ! Delay-slot
         andcc           %g5, %g4, %g0                   ! Executable?
+
+/* ITLB ** ICACHE line 2: Real faults                   */
         be,pn           %xcc, 3f                        ! Nope, branch.
          nop                                            ! Delay-slot
 2:      stxa            %g5, [%g0] ASI_ITLB_DATA_IN     ! Load PTE into TLB
         retry                                           ! Trap return
-3:      rdpr            %pstate, %g4                    ! Move into alternate globals
-
-/* ITLB ** ICACHE line 2: Real faults                   */
+3:      rdpr            %pstate, %g4                    ! Move into alt-globals
         wrpr            %g4, PSTATE_AG|PSTATE_MG, %pstate
         rdpr            %tpc, %g5                       ! And load faulting VA
         mov             FAULT_CODE_ITLB, %g4            ! It was read from ITLB
-sparc64_realfault_common:                               ! Called by TL0 dtlb_miss too
+
+/* ITLB ** ICACHE line 3: Finish faults */
+sparc64_realfault_common:                               ! Called by dtlb_miss
         stb             %g4, [%g6 + TI_FAULT_CODE]
         stx             %g5, [%g6 + TI_FAULT_ADDR]
         ba,pt           %xcc, etrap                     ! Save state
 1:       rd             %pc, %g7                        ! ...
-        nop
-
-/* ITLB ** ICACHE line 3: Finish faults + window fixups */
         call            do_sparc64_fault                ! Call fault handler
          add            %sp, PTREGS_OFF, %o0! Compute pt_regs arg
         ba,pt           %xcc, rtrap_clr_l6              ! Restore cpu state
          nop
+
+/* ITLB ** ICACHE line 4: Window fixups */
 winfix_trampoline:
         rdpr            %tpc, %g3                       ! Prepare winfixup TNPC
-        or              %g3, 0x7c, %g3                  ! Compute offset to branch
+        or              %g3, 0x7c, %g3                  ! Compute branch offset
         wrpr            %g3, %tnpc                      ! Write it into TNPC
         done                                            ! Do it to it
-
-/* ITLB ** ICACHE line 4: Unused...     */
         nop
         nop
         nop
         nop
-        CREATE_VPTE_NOP
 
 #undef CREATE_VPTE_OFFSET1
 #undef CREATE_VPTE_OFFSET2
-#undef CREATE_VPTE_NOP
diff --git a/arch/sparc64/kernel/ktlb.S b/arch/sparc64/kernel/ktlb.S
index 7796b37f478c..d9244d3c9f73 100644
--- a/arch/sparc64/kernel/ktlb.S
+++ b/arch/sparc64/kernel/ktlb.S
@@ -58,9 +58,6 @@ vpte_noent:
         done
 
 vpte_insn_obp:
-        sethi           %hi(prom_pmd_phys), %g5
-        ldx             [%g5 + %lo(prom_pmd_phys)], %g5
-
         /* Behave as if we are at TL0.  */
         wrpr            %g0, 1, %tl
         rdpr            %tpc, %g4       /* Find original faulting iaddr */
@@ -71,58 +68,57 @@ vpte_insn_obp:
         mov             TLB_SFSR, %g1
         stxa            %g4, [%g1 + %g1] ASI_IMMU
 
-        /* Get PMD offset.  */
-        srlx            %g4, 23, %g6
-        and             %g6, 0x7ff, %g6
-        sllx            %g6, 2, %g6
-
-        /* Load PMD, is it valid?  */
-        lduwa           [%g5 + %g6] ASI_PHYS_USE_EC, %g5
-        brz,pn          %g5, longpath
-         sllx           %g5, 11, %g5
-
-        /* Get PTE offset.  */
-        srlx            %g4, 13, %g6
-        and             %g6, 0x3ff, %g6
-        sllx            %g6, 3, %g6
-
-        /* Load PTE.  */
-        ldxa            [%g5 + %g6] ASI_PHYS_USE_EC, %g5
-        brgez,pn        %g5, longpath
-         nop
-
-        /* TLB load and return from trap.  */
+        sethi           %hi(prom_trans), %g5
+        or              %g5, %lo(prom_trans), %g5
+
+1:      ldx             [%g5 + 0x00], %g6       ! base
+        brz,a,pn        %g6, longpath           ! no more entries, fail
+         mov            TLB_SFSR, %g1           ! and restore %g1
+        ldx             [%g5 + 0x08], %g1       ! len
+        add             %g6, %g1, %g1           ! end
+        cmp             %g6, %g4
+        bgu,pt          %xcc, 2f
+         cmp            %g4, %g1
+        bgeu,pt         %xcc, 2f
+         ldx            [%g5 + 0x10], %g1       ! PTE
+
+        /* TLB load, restore %g1, and return from trap.  */
+        sub             %g4, %g6, %g6
+        add             %g1, %g6, %g5
+        mov             TLB_SFSR, %g1
         stxa            %g5, [%g0] ASI_ITLB_DATA_IN
         retry
 
-kvmap_do_obp:
-        sethi           %hi(prom_pmd_phys), %g5
-        ldx             [%g5 + %lo(prom_pmd_phys)], %g5
-
-        /* Get PMD offset.  */
-        srlx            %g4, 23, %g6
-        and             %g6, 0x7ff, %g6
-        sllx            %g6, 2, %g6
-
-        /* Load PMD, is it valid?  */
-        lduwa           [%g5 + %g6] ASI_PHYS_USE_EC, %g5
-        brz,pn          %g5, longpath
-         sllx           %g5, 11, %g5
-
-        /* Get PTE offset.  */
-        srlx            %g4, 13, %g6
-        and             %g6, 0x3ff, %g6
-        sllx            %g6, 3, %g6
-
-        /* Load PTE.  */
-        ldxa            [%g5 + %g6] ASI_PHYS_USE_EC, %g5
-        brgez,pn        %g5, longpath
-         nop
+2:      ba,pt           %xcc, 1b
+         add            %g5, (3 * 8), %g5       ! next entry
 
-        /* TLB load and return from trap.  */
+kvmap_do_obp:
+        sethi           %hi(prom_trans), %g5
+        or              %g5, %lo(prom_trans), %g5
+        srlx            %g4, 13, %g4
+        sllx            %g4, 13, %g4
+
+1:      ldx             [%g5 + 0x00], %g6       ! base
+        brz,a,pn        %g6, longpath           ! no more entries, fail
+         mov            TLB_SFSR, %g1           ! and restore %g1
+        ldx             [%g5 + 0x08], %g1       ! len
+        add             %g6, %g1, %g1           ! end
+        cmp             %g6, %g4
+        bgu,pt          %xcc, 2f
+         cmp            %g4, %g1
+        bgeu,pt         %xcc, 2f
+         ldx            [%g5 + 0x10], %g1       ! PTE
+
+        /* TLB load, restore %g1, and return from trap.  */
+        sub             %g4, %g6, %g6
+        add             %g1, %g6, %g5
+        mov             TLB_SFSR, %g1
         stxa            %g5, [%g0] ASI_DTLB_DATA_IN
         retry
 
+2:      ba,pt           %xcc, 1b
+         add            %g5, (3 * 8), %g5       ! next entry
+
 /*
  * On a first level data miss, check whether this is to the OBP range (note
  * that such accesses can be made by prom, as well as by kernel using