Merge from Linus' tree.

4 files changed, 81 insertions, 31 deletions

diff --git a/arch/sparc64/kernel/entry.S b/arch/sparc64/kernel/entry.S
index 3e0badb820c5..b48349527853 100644
--- a/arch/sparc64/kernel/entry.S
+++ b/arch/sparc64/kernel/entry.S
@@ -42,19 +42,15 @@
  * executing (see inherit_locked_prom_mappings() rant).
  */
 sparc64_vpte_nucleus:
-        /* Load 0xf0000000, which is LOW_OBP_ADDRESS.  */
-        mov             0xf, %g5
-        sllx            %g5, 28, %g5
-
-        /* Is addr >= LOW_OBP_ADDRESS?  */
+        /* Note that kvmap below has verified that the address is
+         * in the range MODULES_VADDR --> VMALLOC_END already.  So
+         * here we need only check if it is an OBP address or not.
+         */
+        sethi           %hi(LOW_OBP_ADDRESS), %g5
         cmp             %g4, %g5
         blu,pn          %xcc, sparc64_vpte_patchme1
          mov            0x1, %g5
-
-        /* Load 0x100000000, which is HI_OBP_ADDRESS.  */
         sllx            %g5, 32, %g5
-
-        /* Is addr < HI_OBP_ADDRESS?  */
         cmp             %g4, %g5
         blu,pn          %xcc, obp_iaddr_patch
          nop
@@ -156,26 +152,29 @@ obp_daddr_patch:
  * rather, use information saved during inherit_prom_mappings() using 8k
  * pagesize.
  */
+        .align          32
 kvmap:
-        /* Load 0xf0000000, which is LOW_OBP_ADDRESS.  */
-        mov             0xf, %g5
-        sllx            %g5, 28, %g5
+        sethi           %hi(MODULES_VADDR), %g5
+        cmp             %g4, %g5
+        blu,pn          %xcc, longpath
+         mov            (VMALLOC_END >> 24), %g5
+        sllx            %g5, 24, %g5
+        cmp             %g4, %g5
+        bgeu,pn         %xcc, longpath
+         nop
 
-        /* Is addr >= LOW_OBP_ADDRESS?  */
+kvmap_check_obp:
+        sethi           %hi(LOW_OBP_ADDRESS), %g5
         cmp             %g4, %g5
-        blu,pn          %xcc, vmalloc_addr
+        blu,pn          %xcc, kvmap_vmalloc_addr
          mov            0x1, %g5
-
-        /* Load 0x100000000, which is HI_OBP_ADDRESS.  */
         sllx            %g5, 32, %g5
-
-        /* Is addr < HI_OBP_ADDRESS?  */
         cmp             %g4, %g5
         blu,pn          %xcc, obp_daddr_patch
          nop
 
-vmalloc_addr:
-        /* If we get here, a vmalloc addr accessed, load kernel VPTE.  */
+kvmap_vmalloc_addr:
+        /* If we get here, a vmalloc addr was accessed, load kernel VPTE.  */
         ldxa            [%g3 + %g6] ASI_N, %g5
         brgez,pn        %g5, longpath
          nop
diff --git a/arch/sparc64/kernel/ptrace.c b/arch/sparc64/kernel/ptrace.c
index 23ad839d113f..5efbff90d668 100644
--- a/arch/sparc64/kernel/ptrace.c
+++ b/arch/sparc64/kernel/ptrace.c
@@ -30,6 +30,7 @@
 #include <asm/psrcompat.h>
 #include <asm/visasm.h>
 #include <asm/spitfire.h>
+#include <asm/page.h>
 
 /* Returning from ptrace is a bit tricky because the syscall return
  * low level code assumes any value returned which is negative and
@@ -128,20 +129,20 @@ void flush_ptrace_access(struct vm_area_struct *vma, struct page *page,
          * is mapped to in the user's address space, we can skip the
          * D-cache flush.
          */
-        if ((uaddr ^ kaddr) & (1UL << 13)) {
+        if ((uaddr ^ (unsigned long) kaddr) & (1UL << 13)) {
                 unsigned long start = __pa(kaddr);
                 unsigned long end = start + len;
 
                 if (tlb_type == spitfire) {
                         for (; start < end; start += 32)
-                                spitfire_put_dcache_tag(va & 0x3fe0, 0x0);
+                                spitfire_put_dcache_tag(start & 0x3fe0, 0x0);
                 } else {
                         for (; start < end; start += 32)
                                 __asm__ __volatile__(
                                         "stxa %%g0, [%0] %1\n\t"
                                         "membar #Sync"
                                         : /* no outputs */
-                                        : "r" (va),
+                                        : "r" (start),
                                         "i" (ASI_DCACHE_INVALIDATE));
                 }
         }
diff --git a/arch/sparc64/kernel/una_asm.S b/arch/sparc64/kernel/una_asm.S
index cbb40585253c..da48400bcc95 100644
--- a/arch/sparc64/kernel/una_asm.S
+++ b/arch/sparc64/kernel/una_asm.S
@@ -17,7 +17,7 @@ kernel_unaligned_trap_fault:
 __do_int_store:
         rd      %asi, %o4
         wr      %o3, 0, %asi
-        ldx     [%o2], %g3
+        mov     %o2, %g3
         cmp     %o1, 2
         be,pn   %icc, 2f
          cmp    %o1, 4
diff --git a/arch/sparc64/kernel/unaligned.c b/arch/sparc64/kernel/unaligned.c
index da9739f0d437..42718f6a7d36 100644
--- a/arch/sparc64/kernel/unaligned.c
+++ b/arch/sparc64/kernel/unaligned.c
@@ -184,13 +184,14 @@ extern void do_int_load(unsigned long *dest_reg, int size,
                         unsigned long *saddr, int is_signed, int asi);
         
 extern void __do_int_store(unsigned long *dst_addr, int size,
-                           unsigned long *src_val, int asi);
+                           unsigned long src_val, int asi);
 
 static inline void do_int_store(int reg_num, int size, unsigned long *dst_addr,
-                                struct pt_regs *regs, int asi)
+                                struct pt_regs *regs, int asi, int orig_asi)
 {
         unsigned long zero = 0;
-        unsigned long *src_val = &zero;
+        unsigned long *src_val_p = &zero;
+        unsigned long src_val;
 
         if (size == 16) {
                 size = 8;
@@ -198,7 +199,25 @@ static inline void do_int_store(int reg_num, int size, unsigned long *dst_addr,
                         (unsigned)fetch_reg(reg_num, regs) : 0)) << 32) |
                         (unsigned)fetch_reg(reg_num + 1, regs);
         } else if (reg_num) {
-                src_val = fetch_reg_addr(reg_num, regs);
+                src_val_p = fetch_reg_addr(reg_num, regs);
+        }
+        src_val = *src_val_p;
+        if (unlikely(asi != orig_asi)) {
+                switch (size) {
+                case 2:
+                        src_val = swab16(src_val);
+                        break;
+                case 4:
+                        src_val = swab32(src_val);
+                        break;
+                case 8:
+                        src_val = swab64(src_val);
+                        break;
+                case 16:
+                default:
+                        BUG();
+                        break;
+                };
         }
         __do_int_store(dst_addr, size, src_val, asi);
 }
@@ -276,6 +295,7 @@ asmlinkage void kernel_unaligned_trap(struct pt_regs *regs, unsigned int insn, u
                 kernel_mna_trap_fault();
         } else {
                 unsigned long addr;
+                int orig_asi, asi;
 
                 addr = compute_effective_address(regs, insn,
                                                  ((insn >> 25) & 0x1f));
@@ -285,18 +305,48 @@ asmlinkage void kernel_unaligned_trap(struct pt_regs *regs, unsigned int insn, u
                        regs->tpc, dirstrings[dir], addr, size,
                        regs->u_regs[UREG_RETPC]);
 #endif
+                orig_asi = asi = decode_asi(insn, regs);
+                switch (asi) {
+                case ASI_NL:
+                case ASI_AIUPL:
+                case ASI_AIUSL:
+                case ASI_PL:
+                case ASI_SL:
+                case ASI_PNFL:
+                case ASI_SNFL:
+                        asi &= ~0x08;
+                        break;
+                };
                 switch (dir) {
                 case load:
                         do_int_load(fetch_reg_addr(((insn>>25)&0x1f), regs),
                                     size, (unsigned long *) addr,
-                                    decode_signedness(insn),
-                                    decode_asi(insn, regs));
+                                    decode_signedness(insn), asi);
+                        if (unlikely(asi != orig_asi)) {
+                                unsigned long val_in = *(unsigned long *) addr;
+                                switch (size) {
+                                case 2:
+                                        val_in = swab16(val_in);
+                                        break;
+                                case 4:
+                                        val_in = swab32(val_in);
+                                        break;
+                                case 8:
+                                        val_in = swab64(val_in);
+                                        break;
+                                case 16:
+                                default:
+                                        BUG();
+                                        break;
+                                };
+                                *(unsigned long *) addr = val_in;
+                        }
                         break;
 
                 case store:
                         do_int_store(((insn>>25)&0x1f), size,
                                      (unsigned long *) addr, regs,
-                                     decode_asi(insn, regs));
+                                     asi, orig_asi);
                         break;
 
                 default: