powerpc: Emulate most Book I instructions in emulate_step()
authorPaul Mackerras <paulus@samba.org>
Tue, 15 Jun 2010 04:48:58 +0000 (14:48 +1000)
committerPaul Mackerras <paulus@samba.org>
Tue, 22 Jun 2010 09:40:29 +0000 (19:40 +1000)
This extends the emulate_step() function to handle a large proportion
of the Book I instructions implemented on current 64-bit server
processors.  The aim is to handle all the load and store instructions
used in the kernel, plus all of the instructions that appear between
l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx
and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7).

The new code can emulate user mode instructions, and checks the
effective address for a load or store if the saved state is for
user mode.  It doesn't handle little-endian mode at present.

For floating-point, Altivec/VMX and VSX instructions, it checks
that the saved MSR has the enable bit for the relevant facility
set, and if so, assumes that the FP/VMX/VSX registers contain
valid state, and does loads or stores directly to/from the
FP/VMX/VSX registers, using assembly helpers in ldstfp.S.

Instructions supported now include:
* Loads and stores, including some but not all VMX and VSX instructions,
  and lmw/stmw
* Atomic loads and stores (l[dw]arx, st[dw]cx.)
* Arithmetic instructions (add, subtract, multiply, divide, etc.)
* Compare instructions
* Rotate and mask instructions
* Shift instructions
* Logical instructions (and, or, xor, etc.)
* Condition register logical instructions
* mtcrf, cntlz[wd], exts[bhw]
* isync, sync, lwsync, ptesync, eieio
* Cache operations (dcbf, dcbst, dcbt, dcbtst)

The overflow-checking arithmetic instructions are not included, but
they appear not to be ever used in C code.

This uses decimal values for the minor opcodes in the switch statements
because that is what appears in the Power ISA specification, thus it is
easier to check that they are correct if they are in decimal.

If this is used to single-step an instruction where a data breakpoint
interrupt occurred, then there is the possibility that the instruction
is a lwarx or ldarx.  In that case we have to be careful not to lose the
reservation until we get to the matching st[wd]cx., or we'll never make
forward progress.  One alternative is to try to arrange that we can
return from interrupts and handle data breakpoint interrupts without
losing the reservation, which means not using any spinlocks, mutexes,
or atomic ops (including bitops).  That seems rather fragile.  The
other alternative is to emulate the larx/stcx and all the instructions
in between.  This is why this commit adds support for a wide range
of integer instructions.

Signed-off-by: Paul Mackerras <paulus@samba.org>
arch/powerpc/include/asm/asm-compat.h
arch/powerpc/include/asm/ppc-opcode.h
arch/powerpc/lib/Makefile
arch/powerpc/lib/ldstfp.S [new file with mode: 0644]
arch/powerpc/lib/sstep.c

index 2048a6aeea9140dc88a0cbb218c2b2c301ca5a1e..decad950f11a8e18e7ad4962343d58d77f7fa363 100644 (file)
@@ -30,6 +30,7 @@
 #define PPC_STLCX      stringify_in_c(stdcx.)
 #define PPC_CNTLZL     stringify_in_c(cntlzd)
 #define PPC_LR_STKOFF  16
+#define PPC_MIN_STKFRM 112
 
 /* Move to CR, single-entry optimized version. Only available
  * on POWER4 and later.
@@ -55,6 +56,7 @@
 #define PPC_CNTLZL     stringify_in_c(cntlzw)
 #define PPC_MTOCRF     stringify_in_c(mtcrf)
 #define PPC_LR_STKOFF  4
+#define PPC_MIN_STKFRM 16
 
 #endif
 
index d553bbeb726c1c0f1c6bda61455bfe3006400f25..43adc8b819edf8cf9368b490adcb5d27423b8874 100644 (file)
 #define PPC_INST_WAIT                  0x7c00007c
 #define PPC_INST_TLBIVAX               0x7c000624
 #define PPC_INST_TLBSRX_DOT            0x7c0006a5
+#define PPC_INST_XXLOR                 0xf0000510
 
 /* macros to insert fields into opcodes */
 #define __PPC_RA(a)    (((a) & 0x1f) << 16)
 #define __PPC_RB(b)    (((b) & 0x1f) << 11)
 #define __PPC_RS(s)    (((s) & 0x1f) << 21)
 #define __PPC_RT(s)    __PPC_RS(s)
+#define __PPC_XA(a)    ((((a) & 0x1f) << 16) | (((a) & 0x20) >> 3))
+#define __PPC_XB(b)    ((((b) & 0x1f) << 11) | (((b) & 0x20) >> 4))
 #define __PPC_XS(s)    ((((s) & 0x1f) << 21) | (((s) & 0x20) >> 5))
+#define __PPC_XT(s)    __PPC_XS(s)
 #define __PPC_T_TLB(t) (((t) & 0x3) << 21)
 #define __PPC_WC(w)    (((w) & 0x3) << 21)
 /*
  * the 128 bit load store instructions based on that.
  */
 #define VSX_XX1(s, a, b)       (__PPC_XS(s) | __PPC_RA(a) | __PPC_RB(b))
+#define VSX_XX3(t, a, b)       (__PPC_XT(t) | __PPC_XA(a) | __PPC_XB(b))
 #define STXVD2X(s, a, b)       stringify_in_c(.long PPC_INST_STXVD2X | \
                                               VSX_XX1((s), (a), (b)))
 #define LXVD2X(s, a, b)                stringify_in_c(.long PPC_INST_LXVD2X | \
                                               VSX_XX1((s), (a), (b)))
+#define XXLOR(t, a, b)         stringify_in_c(.long PPC_INST_XXLOR | \
+                                              VSX_XX3((t), (a), (b)))
 
 #endif /* _ASM_POWERPC_PPC_OPCODE_H */
index 3040dac18a37e2b09287823c96acca5ec5f94863..7581dbffa18e4dbc2e03644553e6ba97f66bcb1e 100644 (file)
@@ -18,8 +18,8 @@ obj-$(CONFIG_HAS_IOMEM)       += devres.o
 
 obj-$(CONFIG_PPC64)    += copypage_64.o copyuser_64.o \
                           memcpy_64.o usercopy_64.o mem_64.o string.o
-obj-$(CONFIG_XMON)     += sstep.o
-obj-$(CONFIG_KPROBES)  += sstep.o
+obj-$(CONFIG_XMON)     += sstep.o ldstfp.o
+obj-$(CONFIG_KPROBES)  += sstep.o ldstfp.o
 
 ifeq ($(CONFIG_PPC64),y)
 obj-$(CONFIG_SMP)      += locks.o
diff --git a/arch/powerpc/lib/ldstfp.S b/arch/powerpc/lib/ldstfp.S
new file mode 100644 (file)
index 0000000..f644863
--- /dev/null
@@ -0,0 +1,375 @@
+/*
+ * Floating-point, VMX/Altivec and VSX loads and stores
+ * for use in instruction emulation.
+ *
+ * Copyright 2010 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
+ *
+ *  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; either version
+ *  2 of the License, or (at your option) any later version.
+ */
+
+#include <asm/processor.h>
+#include <asm/ppc_asm.h>
+#include <asm/ppc-opcode.h>
+#include <asm/reg.h>
+#include <asm/asm-offsets.h>
+#include <linux/errno.h>
+
+#define STKFRM (PPC_MIN_STKFRM + 16)
+
+       .macro  extab   instr,handler
+       .section __ex_table,"a"
+       PPC_LONG \instr,\handler
+       .previous
+       .endm
+
+       .macro  inst32  op
+reg = 0
+       .rept   32
+20:    \op     reg,0,r4
+       b       3f
+       extab   20b,99f
+reg = reg + 1
+       .endr
+       .endm
+
+/* Get the contents of frN into fr0; N is in r3. */
+_GLOBAL(get_fpr)
+       mflr    r0
+       rlwinm  r3,r3,3,0xf8
+       bcl     20,31,1f
+       blr                     /* fr0 is already in fr0 */
+       nop
+reg = 1
+       .rept   31
+       fmr     fr0,reg
+       blr
+reg = reg + 1
+       .endr
+1:     mflr    r5
+       add     r5,r3,r5
+       mtctr   r5
+       mtlr    r0
+       bctr
+
+/* Put the contents of fr0 into frN; N is in r3. */
+_GLOBAL(put_fpr)
+       mflr    r0
+       rlwinm  r3,r3,3,0xf8
+       bcl     20,31,1f
+       blr                     /* fr0 is already in fr0 */
+       nop
+reg = 1
+       .rept   31
+       fmr     reg,fr0
+       blr
+reg = reg + 1
+       .endr
+1:     mflr    r5
+       add     r5,r3,r5
+       mtctr   r5
+       mtlr    r0
+       bctr
+
+/* Load FP reg N from float at *p.  N is in r3, p in r4. */
+_GLOBAL(do_lfs)
+       PPC_STLU r1,-STKFRM(r1)
+       mflr    r0
+       PPC_STL r0,STKFRM+PPC_LR_STKOFF(r1)
+       mfmsr   r6
+       ori     r7,r6,MSR_FP
+       cmpwi   cr7,r3,0
+       mtmsrd  r7
+       isync
+       beq     cr7,1f
+       stfd    fr0,STKFRM-16(r1)
+1:     li      r9,-EFAULT
+2:     lfs     fr0,0(r4)
+       li      r9,0
+3:     bl      put_fpr
+       beq     cr7,4f
+       lfd     fr0,STKFRM-16(r1)
+4:     PPC_LL  r0,STKFRM+PPC_LR_STKOFF(r1)
+       mtlr    r0
+       mtmsrd  r6
+       isync
+       mr      r3,r9
+       addi    r1,r1,STKFRM
+       blr
+       extab   2b,3b
+
+/* Load FP reg N from double at *p.  N is in r3, p in r4. */
+_GLOBAL(do_lfd)
+       PPC_STLU r1,-STKFRM(r1)
+       mflr    r0
+       PPC_STL r0,STKFRM+PPC_LR_STKOFF(r1)
+       mfmsr   r6
+       ori     r7,r6,MSR_FP
+       cmpwi   cr7,r3,0
+       mtmsrd  r7
+       isync
+       beq     cr7,1f
+       stfd    fr0,STKFRM-16(r1)
+1:     li      r9,-EFAULT
+2:     lfd     fr0,0(r4)
+       li      r9,0
+3:     beq     cr7,4f
+       bl      put_fpr
+       lfd     fr0,STKFRM-16(r1)
+4:     PPC_LL  r0,STKFRM+PPC_LR_STKOFF(r1)
+       mtlr    r0
+       mtmsrd  r6
+       isync
+       mr      r3,r9
+       addi    r1,r1,STKFRM
+       blr
+       extab   2b,3b
+
+/* Store FP reg N to float at *p.  N is in r3, p in r4. */
+_GLOBAL(do_stfs)
+       PPC_STLU r1,-STKFRM(r1)
+       mflr    r0
+       PPC_STL r0,STKFRM+PPC_LR_STKOFF(r1)
+       mfmsr   r6
+       ori     r7,r6,MSR_FP
+       cmpwi   cr7,r3,0
+       mtmsrd  r7
+       isync
+       beq     cr7,1f
+       stfd    fr0,STKFRM-16(r1)
+       bl      get_fpr
+1:     li      r9,-EFAULT
+2:     stfs    fr0,0(r4)
+       li      r9,0
+3:     beq     cr7,4f
+       lfd     fr0,STKFRM-16(r1)
+4:     PPC_LL  r0,STKFRM+PPC_LR_STKOFF(r1)
+       mtlr    r0
+       mtmsrd  r6
+       isync
+       mr      r3,r9
+       addi    r1,r1,STKFRM
+       blr
+       extab   2b,3b
+
+/* Store FP reg N to double at *p.  N is in r3, p in r4. */
+_GLOBAL(do_stfd)
+       PPC_STLU r1,-STKFRM(r1)
+       mflr    r0
+       PPC_STL r0,STKFRM+PPC_LR_STKOFF(r1)
+       mfmsr   r6
+       ori     r7,r6,MSR_FP
+       cmpwi   cr7,r3,0
+       mtmsrd  r7
+       isync
+       beq     cr7,1f
+       stfd    fr0,STKFRM-16(r1)
+       bl      get_fpr
+1:     li      r9,-EFAULT
+2:     stfd    fr0,0(r4)
+       li      r9,0
+3:     beq     cr7,4f
+       lfd     fr0,STKFRM-16(r1)
+4:     PPC_LL  r0,STKFRM+PPC_LR_STKOFF(r1)
+       mtlr    r0
+       mtmsrd  r6
+       isync
+       mr      r3,r9
+       addi    r1,r1,STKFRM
+       blr
+       extab   2b,3b
+
+#ifdef CONFIG_ALTIVEC
+/* Get the contents of vrN into vr0; N is in r3. */
+_GLOBAL(get_vr)
+       mflr    r0
+       rlwinm  r3,r3,3,0xf8
+       bcl     20,31,1f
+       blr                     /* vr0 is already in vr0 */
+       nop
+reg = 1
+       .rept   31
+       vor     vr0,reg,reg     /* assembler doesn't know vmr? */
+       blr
+reg = reg + 1
+       .endr
+1:     mflr    r5
+       add     r5,r3,r5
+       mtctr   r5
+       mtlr    r0
+       bctr
+
+/* Put the contents of vr0 into vrN; N is in r3. */
+_GLOBAL(put_vr)
+       mflr    r0
+       rlwinm  r3,r3,3,0xf8
+       bcl     20,31,1f
+       blr                     /* vr0 is already in vr0 */
+       nop
+reg = 1
+       .rept   31
+       vor     reg,vr0,vr0
+       blr
+reg = reg + 1
+       .endr
+1:     mflr    r5
+       add     r5,r3,r5
+       mtctr   r5
+       mtlr    r0
+       bctr
+
+/* Load vector reg N from *p.  N is in r3, p in r4. */
+_GLOBAL(do_lvx)
+       PPC_STLU r1,-STKFRM(r1)
+       mflr    r0
+       PPC_STL r0,STKFRM+PPC_LR_STKOFF(r1)
+       mfmsr   r6
+       oris    r7,r6,MSR_VEC@h
+       cmpwi   cr7,r3,0
+       li      r8,STKFRM-16
+       mtmsrd  r7
+       isync
+       beq     cr7,1f
+       stvx    vr0,r1,r8
+1:     li      r9,-EFAULT
+2:     lvx     vr0,0,r4
+       li      r9,0
+3:     beq     cr7,4f
+       bl      put_vr
+       lvx     vr0,r1,r8
+4:     PPC_LL  r0,STKFRM+PPC_LR_STKOFF(r1)
+       mtlr    r0
+       mtmsrd  r6
+       isync
+       mr      r3,r9
+       addi    r1,r1,STKFRM
+       blr
+       extab   2b,3b
+
+/* Store vector reg N to *p.  N is in r3, p in r4. */
+_GLOBAL(do_stvx)
+       PPC_STLU r1,-STKFRM(r1)
+       mflr    r0
+       PPC_STL r0,STKFRM+PPC_LR_STKOFF(r1)
+       mfmsr   r6
+       oris    r7,r6,MSR_VEC@h
+       cmpwi   cr7,r3,0
+       li      r8,STKFRM-16
+       mtmsrd  r7
+       isync
+       beq     cr7,1f
+       stvx    vr0,r1,r8
+       bl      get_vr
+1:     li      r9,-EFAULT
+2:     stvx    vr0,0,r4
+       li      r9,0
+3:     beq     cr7,4f
+       lvx     vr0,r1,r8
+4:     PPC_LL  r0,STKFRM+PPC_LR_STKOFF(r1)
+       mtlr    r0
+       mtmsrd  r6
+       isync
+       mr      r3,r9
+       addi    r1,r1,STKFRM
+       blr
+       extab   2b,3b
+#endif /* CONFIG_ALTIVEC */
+
+#ifdef CONFIG_VSX
+/* Get the contents of vsrN into vsr0; N is in r3. */
+_GLOBAL(get_vsr)
+       mflr    r0
+       rlwinm  r3,r3,3,0x1f8
+       bcl     20,31,1f
+       blr                     /* vsr0 is already in vsr0 */
+       nop
+reg = 1
+       .rept   63
+       XXLOR(0,reg,reg)
+       blr
+reg = reg + 1
+       .endr
+1:     mflr    r5
+       add     r5,r3,r5
+       mtctr   r5
+       mtlr    r0
+       bctr
+
+/* Put the contents of vsr0 into vsrN; N is in r3. */
+_GLOBAL(put_vsr)
+       mflr    r0
+       rlwinm  r3,r3,3,0x1f8
+       bcl     20,31,1f
+       blr                     /* vr0 is already in vr0 */
+       nop
+reg = 1
+       .rept   63
+       XXLOR(reg,0,0)
+       blr
+reg = reg + 1
+       .endr
+1:     mflr    r5
+       add     r5,r3,r5
+       mtctr   r5
+       mtlr    r0
+       bctr
+
+/* Load VSX reg N from vector doubleword *p.  N is in r3, p in r4. */
+_GLOBAL(do_lxvd2x)
+       PPC_STLU r1,-STKFRM(r1)
+       mflr    r0
+       PPC_STL r0,STKFRM+PPC_LR_STKOFF(r1)
+       mfmsr   r6
+       oris    r7,r6,MSR_VSX@h
+       cmpwi   cr7,r3,0
+       li      r8,STKFRM-16
+       mtmsrd  r7
+       isync
+       beq     cr7,1f
+       STXVD2X(0,r1,r8)
+1:     li      r9,-EFAULT
+2:     LXVD2X(0,0,r4)
+       li      r9,0
+3:     beq     cr7,4f
+       bl      put_vsr
+       LXVD2X(0,r1,r8)
+4:     PPC_LL  r0,STKFRM+PPC_LR_STKOFF(r1)
+       mtlr    r0
+       mtmsrd  r6
+       isync
+       mr      r3,r9
+       addi    r1,r1,STKFRM
+       blr
+       extab   2b,3b
+
+/* Store VSX reg N to vector doubleword *p.  N is in r3, p in r4. */
+_GLOBAL(do_stxvd2x)
+       PPC_STLU r1,-STKFRM(r1)
+       mflr    r0
+       PPC_STL r0,STKFRM+PPC_LR_STKOFF(r1)
+       mfmsr   r6
+       oris    r7,r6,MSR_VSX@h
+       cmpwi   cr7,r3,0
+       li      r8,STKFRM-16
+       mtmsrd  r7
+       isync
+       beq     cr7,1f
+       STXVD2X(0,r1,r8)
+       bl      get_vsr
+1:     li      r9,-EFAULT
+2:     STXVD2X(0,0,r4)
+       li      r9,0
+3:     beq     cr7,4f
+       LXVD2X(0,r1,r8)
+4:     PPC_LL  r0,STKFRM+PPC_LR_STKOFF(r1)
+       mtlr    r0
+       mtmsrd  r6
+       isync
+       mr      r3,r9
+       addi    r1,r1,STKFRM
+       blr
+       extab   2b,3b
+
+#endif /* CONFIG_VSX */
index 13b7d54f185bba4013ff736cfaa54035c5072d38..e0a9858d537eaa624246aed33738c0d7240f3bf3 100644 (file)
@@ -13,6 +13,8 @@
 #include <linux/ptrace.h>
 #include <asm/sstep.h>
 #include <asm/processor.h>
+#include <asm/uaccess.h>
+#include <asm/cputable.h>
 
 extern char system_call_common[];
 
@@ -23,6 +25,23 @@ extern char system_call_common[];
 #define MSR_MASK       0x87c0ffff
 #endif
 
+/* Bits in XER */
+#define XER_SO         0x80000000U
+#define XER_OV         0x40000000U
+#define XER_CA         0x20000000U
+
+/*
+ * Functions in ldstfp.S
+ */
+extern int do_lfs(int rn, unsigned long ea);
+extern int do_lfd(int rn, unsigned long ea);
+extern int do_stfs(int rn, unsigned long ea);
+extern int do_stfd(int rn, unsigned long ea);
+extern int do_lvx(int rn, unsigned long ea);
+extern int do_stvx(int rn, unsigned long ea);
+extern int do_lxvd2x(int rn, unsigned long ea);
+extern int do_stxvd2x(int rn, unsigned long ea);
+
 /*
  * Determine whether a conditional branch instruction would branch.
  */
@@ -46,16 +65,499 @@ static int __kprobes branch_taken(unsigned int instr, struct pt_regs *regs)
        return 1;
 }
 
+
+static long __kprobes address_ok(struct pt_regs *regs, unsigned long ea, int nb)
+{
+       if (!user_mode(regs))
+               return 1;
+       return __access_ok(ea, nb, USER_DS);
+}
+
+/*
+ * Calculate effective address for a D-form instruction
+ */
+static unsigned long __kprobes dform_ea(unsigned int instr, struct pt_regs *regs)
+{
+       int ra;
+       unsigned long ea;
+
+       ra = (instr >> 16) & 0x1f;
+       ea = (signed short) instr;              /* sign-extend */
+       if (ra) {
+               ea += regs->gpr[ra];
+               if (instr & 0x04000000)         /* update forms */
+                       regs->gpr[ra] = ea;
+       }
+#ifdef __powerpc64__
+       if (!(regs->msr & MSR_SF))
+               ea &= 0xffffffffUL;
+#endif
+       return ea;
+}
+
+#ifdef __powerpc64__
+/*
+ * Calculate effective address for a DS-form instruction
+ */
+static unsigned long __kprobes dsform_ea(unsigned int instr, struct pt_regs *regs)
+{
+       int ra;
+       unsigned long ea;
+
+       ra = (instr >> 16) & 0x1f;
+       ea = (signed short) (instr & ~3);       /* sign-extend */
+       if (ra) {
+               ea += regs->gpr[ra];
+               if ((instr & 3) == 1)           /* update forms */
+                       regs->gpr[ra] = ea;
+       }
+       if (!(regs->msr & MSR_SF))
+               ea &= 0xffffffffUL;
+       return ea;
+}
+#endif /* __powerpc64 */
+
+/*
+ * Calculate effective address for an X-form instruction
+ */
+static unsigned long __kprobes xform_ea(unsigned int instr, struct pt_regs *regs,
+                                    int do_update)
+{
+       int ra, rb;
+       unsigned long ea;
+
+       ra = (instr >> 16) & 0x1f;
+       rb = (instr >> 11) & 0x1f;
+       ea = regs->gpr[rb];
+       if (ra) {
+               ea += regs->gpr[ra];
+               if (do_update)          /* update forms */
+                       regs->gpr[ra] = ea;
+       }
+#ifdef __powerpc64__
+       if (!(regs->msr & MSR_SF))
+               ea &= 0xffffffffUL;
+#endif
+       return ea;
+}
+
+/*
+ * Return the largest power of 2, not greater than sizeof(unsigned long),
+ * such that x is a multiple of it.
+ */
+static inline unsigned long max_align(unsigned long x)
+{
+       x |= sizeof(unsigned long);
+       return x & -x;          /* isolates rightmost bit */
+}
+
+
+static inline unsigned long byterev_2(unsigned long x)
+{
+       return ((x >> 8) & 0xff) | ((x & 0xff) << 8);
+}
+
+static inline unsigned long byterev_4(unsigned long x)
+{
+       return ((x >> 24) & 0xff) | ((x >> 8) & 0xff00) |
+               ((x & 0xff00) << 8) | ((x & 0xff) << 24);
+}
+
+#ifdef __powerpc64__
+static inline unsigned long byterev_8(unsigned long x)
+{
+       return (byterev_4(x) << 32) | byterev_4(x >> 32);
+}
+#endif
+
+static int __kprobes read_mem_aligned(unsigned long *dest, unsigned long ea,
+                                     int nb)
+{
+       int err = 0;
+       unsigned long x = 0;
+
+       switch (nb) {
+       case 1:
+               err = __get_user(x, (unsigned char __user *) ea);
+               break;
+       case 2:
+               err = __get_user(x, (unsigned short __user *) ea);
+               break;
+       case 4:
+               err = __get_user(x, (unsigned int __user *) ea);
+               break;
+#ifdef __powerpc64__
+       case 8:
+               err = __get_user(x, (unsigned long __user *) ea);
+               break;
+#endif
+       }
+       if (!err)
+               *dest = x;
+       return err;
+}
+
+static int __kprobes read_mem_unaligned(unsigned long *dest, unsigned long ea,
+                                       int nb, struct pt_regs *regs)
+{
+       int err;
+       unsigned long x, b, c;
+
+       /* unaligned, do this in pieces */
+       x = 0;
+       for (; nb > 0; nb -= c) {
+               c = max_align(ea);
+               if (c > nb)
+                       c = max_align(nb);
+               err = read_mem_aligned(&b, ea, c);
+               if (err)
+                       return err;
+               x = (x << (8 * c)) + b;
+               ea += c;
+       }
+       *dest = x;
+       return 0;
+}
+
+/*
+ * Read memory at address ea for nb bytes, return 0 for success
+ * or -EFAULT if an error occurred.
+ */
+static int __kprobes read_mem(unsigned long *dest, unsigned long ea, int nb,
+                             struct pt_regs *regs)
+{
+       if (!address_ok(regs, ea, nb))
+               return -EFAULT;
+       if ((ea & (nb - 1)) == 0)
+               return read_mem_aligned(dest, ea, nb);
+       return read_mem_unaligned(dest, ea, nb, regs);
+}
+
+static int __kprobes write_mem_aligned(unsigned long val, unsigned long ea,
+                                      int nb)
+{
+       int err = 0;
+
+       switch (nb) {
+       case 1:
+               err = __put_user(val, (unsigned char __user *) ea);
+               break;
+       case 2:
+               err = __put_user(val, (unsigned short __user *) ea);
+               break;
+       case 4:
+               err = __put_user(val, (unsigned int __user *) ea);
+               break;
+#ifdef __powerpc64__
+       case 8:
+               err = __put_user(val, (unsigned long __user *) ea);
+               break;
+#endif
+       }
+       return err;
+}
+
+static int __kprobes write_mem_unaligned(unsigned long val, unsigned long ea,
+                                        int nb, struct pt_regs *regs)
+{
+       int err;
+       unsigned long c;
+
+       /* unaligned or little-endian, do this in pieces */
+       for (; nb > 0; nb -= c) {
+               c = max_align(ea);
+               if (c > nb)
+                       c = max_align(nb);
+               err = write_mem_aligned(val >> (nb - c) * 8, ea, c);
+               if (err)
+                       return err;
+               ++ea;
+       }
+       return 0;
+}
+
+/*
+ * Write memory at address ea for nb bytes, return 0 for success
+ * or -EFAULT if an error occurred.
+ */
+static int __kprobes write_mem(unsigned long val, unsigned long ea, int nb,
+                              struct pt_regs *regs)
+{
+       if (!address_ok(regs, ea, nb))
+               return -EFAULT;
+       if ((ea & (nb - 1)) == 0)
+               return write_mem_aligned(val, ea, nb);
+       return write_mem_unaligned(val, ea, nb, regs);
+}
+
 /*
- * Emulate instructions that cause a transfer of control.
+ * Check the address and alignment, and call func to do the actual
+ * load or store.
+ */
+static int __kprobes do_fp_load(int rn, int (*func)(int, unsigned long),
+                               unsigned long ea, int nb,
+                               struct pt_regs *regs)
+{
+       int err;
+       unsigned long val[sizeof(double) / sizeof(long)];
+       unsigned long ptr;
+
+       if (!address_ok(regs, ea, nb))
+               return -EFAULT;
+       if ((ea & 3) == 0)
+               return (*func)(rn, ea);
+       ptr = (unsigned long) &val[0];
+       if (sizeof(unsigned long) == 8 || nb == 4) {
+               err = read_mem_unaligned(&val[0], ea, nb, regs);
+               ptr += sizeof(unsigned long) - nb;
+       } else {
+               /* reading a double on 32-bit */
+               err = read_mem_unaligned(&val[0], ea, 4, regs);
+               if (!err)
+                       err = read_mem_unaligned(&val[1], ea + 4, 4, regs);
+       }
+       if (err)
+               return err;
+       return (*func)(rn, ptr);
+}
+
+static int __kprobes do_fp_store(int rn, int (*func)(int, unsigned long),
+                                unsigned long ea, int nb,
+                                struct pt_regs *regs)
+{
+       int err;
+       unsigned long val[sizeof(double) / sizeof(long)];
+       unsigned long ptr;
+
+       if (!address_ok(regs, ea, nb))
+               return -EFAULT;
+       if ((ea & 3) == 0)
+               return (*func)(rn, ea);
+       ptr = (unsigned long) &val[0];
+       if (sizeof(unsigned long) == 8 || nb == 4) {
+               ptr += sizeof(unsigned long) - nb;
+               err = (*func)(rn, ptr);
+               if (err)
+                       return err;
+               err = write_mem_unaligned(val[0], ea, nb, regs);
+       } else {
+               /* writing a double on 32-bit */
+               err = (*func)(rn, ptr);
+               if (err)
+                       return err;
+               err = write_mem_unaligned(val[0], ea, 4, regs);
+               if (!err)
+                       err = write_mem_unaligned(val[1], ea + 4, 4, regs);
+       }
+       return err;
+}
+
+#ifdef CONFIG_ALTIVEC
+/* For Altivec/VMX, no need to worry about alignment */
+static int __kprobes do_vec_load(int rn, int (*func)(int, unsigned long),
+                                unsigned long ea, struct pt_regs *regs)
+{
+       if (!address_ok(regs, ea & ~0xfUL, 16))
+               return -EFAULT;
+       return (*func)(rn, ea);
+}
+
+static int __kprobes do_vec_store(int rn, int (*func)(int, unsigned long),
+                                 unsigned long ea, struct pt_regs *regs)
+{
+       if (!address_ok(regs, ea & ~0xfUL, 16))
+               return -EFAULT;
+       return (*func)(rn, ea);
+}
+#endif /* CONFIG_ALTIVEC */
+
+#ifdef CONFIG_VSX
+static int __kprobes do_vsx_load(int rn, int (*func)(int, unsigned long),
+                                unsigned long ea, struct pt_regs *regs)
+{
+       int err;
+       unsigned long val[2];
+
+       if (!address_ok(regs, ea, 16))
+               return -EFAULT;
+       if ((ea & 3) == 0)
+               return (*func)(rn, ea);
+       err = read_mem_unaligned(&val[0], ea, 8, regs);
+       if (!err)
+               err = read_mem_unaligned(&val[1], ea + 8, 8, regs);
+       if (!err)
+               err = (*func)(rn, (unsigned long) &val[0]);
+       return err;
+}
+
+static int __kprobes do_vsx_store(int rn, int (*func)(int, unsigned long),
+                                unsigned long ea, struct pt_regs *regs)
+{
+       int err;
+       unsigned long val[2];
+
+       if (!address_ok(regs, ea, 16))
+               return -EFAULT;
+       if ((ea & 3) == 0)
+               return (*func)(rn, ea);
+       err = (*func)(rn, (unsigned long) &val[0]);
+       if (err)
+               return err;
+       err = write_mem_unaligned(val[0], ea, 8, regs);
+       if (!err)
+               err = write_mem_unaligned(val[1], ea + 8, 8, regs);
+       return err;
+}
+#endif /* CONFIG_VSX */
+
+#define __put_user_asmx(x, addr, err, op, cr)          \
+       __asm__ __volatile__(                           \
+               "1:     " op " %2,0,%3\n"               \
+               "       mfcr    %1\n"                   \
+               "2:\n"                                  \
+               ".section .fixup,\"ax\"\n"              \
+               "3:     li      %0,%4\n"                \
+               "       b       2b\n"                   \
+               ".previous\n"                           \
+               ".section __ex_table,\"a\"\n"           \
+                       PPC_LONG_ALIGN "\n"             \
+                       PPC_LONG "1b,3b\n"              \
+               ".previous"                             \
+               : "=r" (err), "=r" (cr)                 \
+               : "r" (x), "r" (addr), "i" (-EFAULT), "0" (err))
+
+#define __get_user_asmx(x, addr, err, op)              \
+       __asm__ __volatile__(                           \
+               "1:     "op" %1,0,%2\n"                 \
+               "2:\n"                                  \
+               ".section .fixup,\"ax\"\n"              \
+               "3:     li      %0,%3\n"                \
+               "       b       2b\n"                   \
+               ".previous\n"                           \
+               ".section __ex_table,\"a\"\n"           \
+                       PPC_LONG_ALIGN "\n"             \
+                       PPC_LONG "1b,3b\n"              \
+               ".previous"                             \
+               : "=r" (err), "=r" (x)                  \
+               : "r" (addr), "i" (-EFAULT), "0" (err))
+
+#define __cacheop_user_asmx(addr, err, op)             \
+       __asm__ __volatile__(                           \
+               "1:     "op" 0,%1\n"                    \
+               "2:\n"                                  \
+               ".section .fixup,\"ax\"\n"              \
+               "3:     li      %0,%3\n"                \
+               "       b       2b\n"                   \
+               ".previous\n"                           \
+               ".section __ex_table,\"a\"\n"           \
+                       PPC_LONG_ALIGN "\n"             \
+                       PPC_LONG "1b,3b\n"              \
+               ".previous"                             \
+               : "=r" (err)                            \
+               : "r" (addr), "i" (-EFAULT), "0" (err))
+
+static void __kprobes set_cr0(struct pt_regs *regs, int rd)
+{
+       long val = regs->gpr[rd];
+
+       regs->ccr = (regs->ccr & 0x0fffffff) | ((regs->xer >> 3) & 0x10000000);
+#ifdef __powerpc64__
+       if (!(regs->msr & MSR_SF))
+               val = (int) val;
+#endif
+       if (val < 0)
+               regs->ccr |= 0x80000000;
+       else if (val > 0)
+               regs->ccr |= 0x40000000;
+       else
+               regs->ccr |= 0x20000000;
+}
+
+static void __kprobes add_with_carry(struct pt_regs *regs, int rd,
+                                    unsigned long val1, unsigned long val2,
+                                    unsigned long carry_in)
+{
+       unsigned long val = val1 + val2;
+
+       if (carry_in)
+               ++val;
+       regs->gpr[rd] = val;
+#ifdef __powerpc64__
+       if (!(regs->msr & MSR_SF)) {
+               val = (unsigned int) val;
+               val1 = (unsigned int) val1;
+       }
+#endif
+       if (val < val1 || (carry_in && val == val1))
+               regs->xer |= XER_CA;
+       else
+               regs->xer &= ~XER_CA;
+}
+
+static void __kprobes do_cmp_signed(struct pt_regs *regs, long v1, long v2,
+                                   int crfld)
+{
+       unsigned int crval, shift;
+
+       crval = (regs->xer >> 31) & 1;          /* get SO bit */
+       if (v1 < v2)
+               crval |= 8;
+       else if (v1 > v2)
+               crval |= 4;
+       else
+               crval |= 2;
+       shift = (7 - crfld) * 4;
+       regs->ccr = (regs->ccr & ~(0xf << shift)) | (crval << shift);
+}
+
+static void __kprobes do_cmp_unsigned(struct pt_regs *regs, unsigned long v1,
+                                     unsigned long v2, int crfld)
+{
+       unsigned int crval, shift;
+
+       crval = (regs->xer >> 31) & 1;          /* get SO bit */
+       if (v1 < v2)
+               crval |= 8;
+       else if (v1 > v2)
+               crval |= 4;
+       else
+               crval |= 2;
+       shift = (7 - crfld) * 4;
+       regs->ccr = (regs->ccr & ~(0xf << shift)) | (crval << shift);
+}
+
+/*
+ * Elements of 32-bit rotate and mask instructions.
+ */
+#define MASK32(mb, me) ((0xffffffffUL >> (mb)) + \
+                        ((signed long)-0x80000000L >> (me)) + ((me) >= (mb)))
+#ifdef __powerpc64__
+#define MASK64_L(mb)   (~0UL >> (mb))
+#define MASK64_R(me)   ((signed long)-0x8000000000000000L >> (me))
+#define MASK64(mb, me) (MASK64_L(mb) + MASK64_R(me) + ((me) >= (mb)))
+#define DATA32(x)      (((x) & 0xffffffffUL) | (((x) & 0xffffffffUL) << 32))
+#else
+#define DATA32(x)      (x)
+#endif
+#define ROTATE(x, n)   ((n) ? (((x) << (n)) | ((x) >> (8 * sizeof(long) - (n)))) : (x))
+
+/*
+ * Emulate instructions that cause a transfer of control,
+ * loads and stores, and a few other instructions.
  * Returns 1 if the step was emulated, 0 if not,
  * or -1 if the instruction is one that should not be stepped,
  * such as an rfid, or a mtmsrd that would clear MSR_RI.
  */
 int __kprobes emulate_step(struct pt_regs *regs, unsigned int instr)
 {
-       unsigned int opcode, rs, rb, rd, spr;
+       unsigned int opcode, ra, rb, rd, spr, u;
        unsigned long int imm;
+       unsigned long int val, val2;
+       unsigned long int ea;
+       unsigned int cr, mb, me, sh;
+       int err;
+       unsigned long old_ra;
+       long ival;
 
        opcode = instr >> 26;
        switch (opcode) {
@@ -78,7 +580,13 @@ int __kprobes emulate_step(struct pt_regs *regs, unsigned int instr)
                 * entry code works.  If that is changed, this will
                 * need to be changed also.
                 */
+               if (regs->gpr[0] == 0x1ebe &&
+                   cpu_has_feature(CPU_FTR_REAL_LE)) {
+                       regs->msr ^= MSR_LE;
+                       goto instr_done;
+               }
                regs->gpr[9] = regs->gpr[13];
+               regs->gpr[10] = MSR_KERNEL;
                regs->gpr[11] = regs->nip + 4;
                regs->gpr[12] = regs->msr & MSR_MASK;
                regs->gpr[13] = (unsigned long) get_paca();
@@ -102,9 +610,9 @@ int __kprobes emulate_step(struct pt_regs *regs, unsigned int instr)
                regs->nip = imm;
                return 1;
        case 19:
-               switch (instr & 0x7fe) {
-               case 0x20:      /* bclr */
-               case 0x420:     /* bcctr */
+               switch ((instr >> 1) & 0x3ff) {
+               case 16:        /* bclr */
+               case 528:       /* bcctr */
                        imm = (instr & 0x400)? regs->ctr: regs->link;
                        regs->nip += 4;
                        if ((regs->msr & MSR_SF) == 0) {
@@ -116,30 +624,233 @@ int __kprobes emulate_step(struct pt_regs *regs, unsigned int instr)
                        if (branch_taken(instr, regs))
                                regs->nip = imm;
                        return 1;
-               case 0x24:      /* rfid, scary */
+
+               case 18:        /* rfid, scary */
                        return -1;
+
+               case 150:       /* isync */
+                       isync();
+                       goto instr_done;
+
+               case 33:        /* crnor */
+               case 129:       /* crandc */
+               case 193:       /* crxor */
+               case 225:       /* crnand */
+               case 257:       /* crand */
+               case 289:       /* creqv */
+               case 417:       /* crorc */
+               case 449:       /* cror */
+                       ra = (instr >> 16) & 0x1f;
+                       rb = (instr >> 11) & 0x1f;
+                       rd = (instr >> 21) & 0x1f;
+                       ra = (regs->ccr >> (31 - ra)) & 1;
+                       rb = (regs->ccr >> (31 - rb)) & 1;
+                       val = (instr >> (6 + ra * 2 + rb)) & 1;
+                       regs->ccr = (regs->ccr & ~(1UL << (31 - rd))) |
+                               (val << (31 - rd));
+                       goto instr_done;
+               }
+               break;
+       case 31:
+               switch ((instr >> 1) & 0x3ff) {
+               case 598:       /* sync */
+#ifdef __powerpc64__
+                       switch ((instr >> 21) & 3) {
+                       case 1:         /* lwsync */
+                               asm volatile("lwsync" : : : "memory");
+                               goto instr_done;
+                       case 2:         /* ptesync */
+                               asm volatile("ptesync" : : : "memory");
+                               goto instr_done;
+                       }
+#endif
+                       mb();
+                       goto instr_done;
+
+               case 854:       /* eieio */
+                       eieio();
+                       goto instr_done;
+               }
+               break;
+       }
+
+       /* Following cases refer to regs->gpr[], so we need all regs */
+       if (!FULL_REGS(regs))
+               return 0;
+
+       rd = (instr >> 21) & 0x1f;
+       ra = (instr >> 16) & 0x1f;
+       rb = (instr >> 11) & 0x1f;
+
+       switch (opcode) {
+       case 7:         /* mulli */
+               regs->gpr[rd] = regs->gpr[ra] * (short) instr;
+               goto instr_done;
+
+       case 8:         /* subfic */
+               imm = (short) instr;
+               add_with_carry(regs, rd, ~regs->gpr[ra], imm, 1);
+               goto instr_done;
+
+       case 10:        /* cmpli */
+               imm = (unsigned short) instr;
+               val = regs->gpr[ra];
+#ifdef __powerpc64__
+               if ((rd & 1) == 0)
+                       val = (unsigned int) val;
+#endif
+               do_cmp_unsigned(regs, val, imm, rd >> 2);
+               goto instr_done;
+
+       case 11:        /* cmpi */
+               imm = (short) instr;
+               val = regs->gpr[ra];
+#ifdef __powerpc64__
+               if ((rd & 1) == 0)
+                       val = (int) val;
+#endif
+               do_cmp_signed(regs, val, imm, rd >> 2);
+               goto instr_done;
+
+       case 12:        /* addic */
+               imm = (short) instr;
+               add_with_carry(regs, rd, regs->gpr[ra], imm, 0);
+               goto instr_done;
+
+       case 13:        /* addic. */
+               imm = (short) instr;
+               add_with_carry(regs, rd, regs->gpr[ra], imm, 0);
+               set_cr0(regs, rd);
+               goto instr_done;
+
+       case 14:        /* addi */
+               imm = (short) instr;
+               if (ra)
+                       imm += regs->gpr[ra];
+               regs->gpr[rd] = imm;
+               goto instr_done;
+
+       case 15:        /* addis */
+               imm = ((short) instr) << 16;
+               if (ra)
+                       imm += regs->gpr[ra];
+               regs->gpr[rd] = imm;
+               goto instr_done;
+
+       case 20:        /* rlwimi */
+               mb = (instr >> 6) & 0x1f;
+               me = (instr >> 1) & 0x1f;
+               val = DATA32(regs->gpr[rd]);
+               imm = MASK32(mb, me);
+               regs->gpr[ra] = (regs->gpr[ra] & ~imm) | (ROTATE(val, rb) & imm);
+               goto logical_done;
+
+       case 21:        /* rlwinm */
+               mb = (instr >> 6) & 0x1f;
+               me = (instr >> 1) & 0x1f;
+               val = DATA32(regs->gpr[rd]);
+               regs->gpr[ra] = ROTATE(val, rb) & MASK32(mb, me);
+               goto logical_done;
+
+       case 23:        /* rlwnm */
+               mb = (instr >> 6) & 0x1f;
+               me = (instr >> 1) & 0x1f;
+               rb = regs->gpr[rb] & 0x1f;
+               val = DATA32(regs->gpr[rd]);
+               regs->gpr[ra] = ROTATE(val, rb) & MASK32(mb, me);
+               goto logical_done;
+
+       case 24:        /* ori */
+               imm = (unsigned short) instr;
+               regs->gpr[ra] = regs->gpr[rd] | imm;
+               goto instr_done;
+
+       case 25:        /* oris */
+               imm = (unsigned short) instr;
+               regs->gpr[ra] = regs->gpr[rd] | (imm << 16);
+               goto instr_done;
+
+       case 26:        /* xori */
+               imm = (unsigned short) instr;
+               regs->gpr[ra] = regs->gpr[rd] ^ imm;
+               goto instr_done;
+
+       case 27:        /* xoris */
+               imm = (unsigned short) instr;
+               regs->gpr[ra] = regs->gpr[rd] ^ (imm << 16);
+               goto instr_done;
+
+       case 28:        /* andi. */
+               imm = (unsigned short) instr;
+               regs->gpr[ra] = regs->gpr[rd] & imm;
+               set_cr0(regs, ra);
+               goto instr_done;
+
+       case 29:        /* andis. */
+               imm = (unsigned short) instr;
+               regs->gpr[ra] = regs->gpr[rd] & (imm << 16);
+               set_cr0(regs, ra);
+               goto instr_done;
+
+#ifdef __powerpc64__
+       case 30:        /* rld* */
+               mb = ((instr >> 6) & 0x1f) | (instr & 0x20);
+               val = regs->gpr[rd];
+               if ((instr & 0x10) == 0) {
+                       sh = rb | ((instr & 2) << 4);
+                       val = ROTATE(val, sh);
+                       switch ((instr >> 2) & 3) {
+                       case 0:         /* rldicl */
+                               regs->gpr[ra] = val & MASK64_L(mb);
+                               goto logical_done;
+                       case 1:         /* rldicr */
+                               regs->gpr[ra] = val & MASK64_R(mb);
+                               goto logical_done;
+                       case 2:         /* rldic */
+                               regs->gpr[ra] = val & MASK64(mb, 63 - sh);
+                               goto logical_done;
+                       case 3:         /* rldimi */
+                               imm = MASK64(mb, 63 - sh);
+                               regs->gpr[ra] = (regs->gpr[ra] & ~imm) |
+                                       (val & imm);
+                               goto logical_done;
+                       }
+               } else {
+                       sh = regs->gpr[rb] & 0x3f;
+                       val = ROTATE(val, sh);
+                       switch ((instr >> 1) & 7) {
+                       case 0:         /* rldcl */
+                               regs->gpr[ra] = val & MASK64_L(mb);
+                               goto logical_done;
+                       case 1:         /* rldcr */
+                               regs->gpr[ra] = val & MASK64_R(mb);
+                               goto logical_done;
+                       }
                }
+#endif
+
        case 31:
-               rd = (instr >> 21) & 0x1f;
-               switch (instr & 0x7fe) {
-               case 0xa6:      /* mfmsr */
+               switch ((instr >> 1) & 0x3ff) {
+               case 83:        /* mfmsr */
+                       if (regs->msr & MSR_PR)
+                               break;
                        regs->gpr[rd] = regs->msr & MSR_MASK;
-                       regs->nip += 4;
-                       if ((regs->msr & MSR_SF) == 0)
-                               regs->nip &= 0xffffffffUL;
-                       return 1;
-               case 0x124:     /* mtmsr */
+                       goto instr_done;
+               case 146:       /* mtmsr */
+                       if (regs->msr & MSR_PR)
+                               break;
                        imm = regs->gpr[rd];
                        if ((imm & MSR_RI) == 0)
                                /* can't step mtmsr that would clear MSR_RI */
                                return -1;
                        regs->msr = imm;
-                       regs->nip += 4;
-                       return 1;
+                       goto instr_done;
 #ifdef CONFIG_PPC64
-               case 0x164:     /* mtmsrd */
+               case 178:       /* mtmsrd */
                        /* only MSR_EE and MSR_RI get changed if bit 15 set */
                        /* mtmsrd doesn't change MSR_HV and MSR_ME */
+                       if (regs->msr & MSR_PR)
+                               break;
                        imm = (instr & 0x10000)? 0x8002: 0xefffffffffffefffUL;
                        imm = (regs->msr & MSR_MASK & ~imm)
                                | (regs->gpr[rd] & imm);
@@ -147,57 +858,770 @@ int __kprobes emulate_step(struct pt_regs *regs, unsigned int instr)
                                /* can't step mtmsrd that would clear MSR_RI */
                                return -1;
                        regs->msr = imm;
-                       regs->nip += 4;
-                       if ((imm & MSR_SF) == 0)
-                               regs->nip &= 0xffffffffUL;
-                       return 1;
+                       goto instr_done;
 #endif
-               case 0x26:      /* mfcr */
+               case 19:        /* mfcr */
                        regs->gpr[rd] = regs->ccr;
                        regs->gpr[rd] &= 0xffffffffUL;
-                       goto mtspr_out;
-               case 0x2a6:     /* mfspr */
+                       goto instr_done;
+
+               case 144:       /* mtcrf */
+                       imm = 0xf0000000UL;
+                       val = regs->gpr[rd];
+                       for (sh = 0; sh < 8; ++sh) {
+                               if (instr & (0x80000 >> sh))
+                                       regs->ccr = (regs->ccr & ~imm) |
+                                               (val & imm);
+                               imm >>= 4;
+                       }
+                       goto instr_done;
+
+               case 339:       /* mfspr */
                        spr = (instr >> 11) & 0x3ff;
                        switch (spr) {
                        case 0x20:      /* mfxer */
                                regs->gpr[rd] = regs->xer;
                                regs->gpr[rd] &= 0xffffffffUL;
-                               goto mtspr_out;
+                               goto instr_done;
                        case 0x100:     /* mflr */
                                regs->gpr[rd] = regs->link;
-                               goto mtspr_out;
+                               goto instr_done;
                        case 0x120:     /* mfctr */
                                regs->gpr[rd] = regs->ctr;
-                               goto mtspr_out;
-                       }
-                       break;
-               case 0x378:     /* orx */
-                       if (instr & 1)
-                               break;
-                       rs = (instr >> 21) & 0x1f;
-                       rb = (instr >> 11) & 0x1f;
-                       if (rs == rb) {         /* mr */
-                               rd = (instr >> 16) & 0x1f;
-                               regs->gpr[rd] = regs->gpr[rs];
-                               goto mtspr_out;
+                               goto instr_done;
                        }
                        break;
-               case 0x3a6:     /* mtspr */
+
+               case 467:       /* mtspr */
                        spr = (instr >> 11) & 0x3ff;
                        switch (spr) {
                        case 0x20:      /* mtxer */
                                regs->xer = (regs->gpr[rd] & 0xffffffffUL);
-                               goto mtspr_out;
+                               goto instr_done;
                        case 0x100:     /* mtlr */
                                regs->link = regs->gpr[rd];
-                               goto mtspr_out;
+                               goto instr_done;
                        case 0x120:     /* mtctr */
                                regs->ctr = regs->gpr[rd];
-mtspr_out:
-                               regs->nip += 4;
-                               return 1;
+                               goto instr_done;
                        }
+                       break;
+
+/*
+ * Compare instructions
+ */
+               case 0: /* cmp */
+                       val = regs->gpr[ra];
+                       val2 = regs->gpr[rb];
+#ifdef __powerpc64__
+                       if ((rd & 1) == 0) {
+                               /* word (32-bit) compare */
+                               val = (int) val;
+                               val2 = (int) val2;
+                       }
+#endif
+                       do_cmp_signed(regs, val, val2, rd >> 2);
+                       goto instr_done;
+
+               case 32:        /* cmpl */
+                       val = regs->gpr[ra];
+                       val2 = regs->gpr[rb];
+#ifdef __powerpc64__
+                       if ((rd & 1) == 0) {
+                               /* word (32-bit) compare */
+                               val = (unsigned int) val;
+                               val2 = (unsigned int) val2;
+                       }
+#endif
+                       do_cmp_unsigned(regs, val, val2, rd >> 2);
+                       goto instr_done;
+
+/*
+ * Arithmetic instructions
+ */
+               case 8: /* subfc */
+                       add_with_carry(regs, rd, ~regs->gpr[ra],
+                                      regs->gpr[rb], 1);
+                       goto arith_done;
+#ifdef __powerpc64__
+               case 9: /* mulhdu */
+                       asm("mulhdu %0,%1,%2" : "=r" (regs->gpr[rd]) :
+                           "r" (regs->gpr[ra]), "r" (regs->gpr[rb]));
+                       goto arith_done;
+#endif
+               case 10:        /* addc */
+                       add_with_carry(regs, rd, regs->gpr[ra],
+                                      regs->gpr[rb], 0);
+                       goto arith_done;
+
+               case 11:        /* mulhwu */
+                       asm("mulhwu %0,%1,%2" : "=r" (regs->gpr[rd]) :
+                           "r" (regs->gpr[ra]), "r" (regs->gpr[rb]));
+                       goto arith_done;
+
+               case 40:        /* subf */
+                       regs->gpr[rd] = regs->gpr[rb] - regs->gpr[ra];
+                       goto arith_done;
+#ifdef __powerpc64__
+               case 73:        /* mulhd */
+                       asm("mulhd %0,%1,%2" : "=r" (regs->gpr[rd]) :
+                           "r" (regs->gpr[ra]), "r" (regs->gpr[rb]));
+                       goto arith_done;
+#endif
+               case 75:        /* mulhw */
+                       asm("mulhw %0,%1,%2" : "=r" (regs->gpr[rd]) :
+                           "r" (regs->gpr[ra]), "r" (regs->gpr[rb]));
+                       goto arith_done;
+
+               case 104:       /* neg */
+                       regs->gpr[rd] = -regs->gpr[ra];
+                       goto arith_done;
+
+               case 136:       /* subfe */
+                       add_with_carry(regs, rd, ~regs->gpr[ra], regs->gpr[rb],
+                                      regs->xer & XER_CA);
+                       goto arith_done;
+
+               case 138:       /* adde */
+                       add_with_carry(regs, rd, regs->gpr[ra], regs->gpr[rb],
+                                      regs->xer & XER_CA);
+                       goto arith_done;
+
+               case 200:       /* subfze */
+                       add_with_carry(regs, rd, ~regs->gpr[ra], 0L,
+                                      regs->xer & XER_CA);
+                       goto arith_done;
+
+               case 202:       /* addze */
+                       add_with_carry(regs, rd, regs->gpr[ra], 0L,
+                                      regs->xer & XER_CA);
+                       goto arith_done;
+
+               case 232:       /* subfme */
+                       add_with_carry(regs, rd, ~regs->gpr[ra], -1L,
+                                      regs->xer & XER_CA);
+                       goto arith_done;
+#ifdef __powerpc64__
+               case 233:       /* mulld */
+                       regs->gpr[rd] = regs->gpr[ra] * regs->gpr[rb];
+                       goto arith_done;
+#endif
+               case 234:       /* addme */
+                       add_with_carry(regs, rd, regs->gpr[ra], -1L,
+                                      regs->xer & XER_CA);
+                       goto arith_done;
+
+               case 235:       /* mullw */
+                       regs->gpr[rd] = (unsigned int) regs->gpr[ra] *
+                               (unsigned int) regs->gpr[rb];
+                       goto arith_done;
+
+               case 266:       /* add */
+                       regs->gpr[rd] = regs->gpr[ra] + regs->gpr[rb];
+                       goto arith_done;
+#ifdef __powerpc64__
+               case 457:       /* divdu */
+                       regs->gpr[rd] = regs->gpr[ra] / regs->gpr[rb];
+                       goto arith_done;
+#endif
+               case 459:       /* divwu */
+                       regs->gpr[rd] = (unsigned int) regs->gpr[ra] /
+                               (unsigned int) regs->gpr[rb];
+                       goto arith_done;
+#ifdef __powerpc64__
+               case 489:       /* divd */
+                       regs->gpr[rd] = (long int) regs->gpr[ra] /
+                               (long int) regs->gpr[rb];
+                       goto arith_done;
+#endif
+               case 491:       /* divw */
+                       regs->gpr[rd] = (int) regs->gpr[ra] /
+                               (int) regs->gpr[rb];
+                       goto arith_done;
+
+
+/*
+ * Logical instructions
+ */
+               case 26:        /* cntlzw */
+                       asm("cntlzw %0,%1" : "=r" (regs->gpr[ra]) :
+                           "r" (regs->gpr[rd]));
+                       goto logical_done;
+#ifdef __powerpc64__
+               case 58:        /* cntlzd */
+                       asm("cntlzd %0,%1" : "=r" (regs->gpr[ra]) :
+                           "r" (regs->gpr[rd]));
+                       goto logical_done;
+#endif
+               case 28:        /* and */
+                       regs->gpr[ra] = regs->gpr[rd] & regs->gpr[rb];
+                       goto logical_done;
+
+               case 60:        /* andc */
+                       regs->gpr[ra] = regs->gpr[rd] & ~regs->gpr[rb];
+                       goto logical_done;
+
+               case 124:       /* nor */
+                       regs->gpr[ra] = ~(regs->gpr[rd] | regs->gpr[rb]);
+                       goto logical_done;
+
+               case 284:       /* xor */
+                       regs->gpr[ra] = ~(regs->gpr[rd] ^ regs->gpr[rb]);
+                       goto logical_done;
+
+               case 316:       /* xor */
+                       regs->gpr[ra] = regs->gpr[rd] ^ regs->gpr[rb];
+                       goto logical_done;
+
+               case 412:       /* orc */
+                       regs->gpr[ra] = regs->gpr[rd] | ~regs->gpr[rb];
+                       goto logical_done;
+
+               case 444:       /* or */
+                       regs->gpr[ra] = regs->gpr[rd] | regs->gpr[rb];
+                       goto logical_done;
+
+               case 476:       /* nand */
+                       regs->gpr[ra] = ~(regs->gpr[rd] & regs->gpr[rb]);
+                       goto logical_done;
+
+               case 922:       /* extsh */
+                       regs->gpr[ra] = (signed short) regs->gpr[rd];
+                       goto logical_done;
+
+               case 954:       /* extsb */
+                       regs->gpr[ra] = (signed char) regs->gpr[rd];
+                       goto logical_done;
+#ifdef __powerpc64__
+               case 986:       /* extsw */
+                       regs->gpr[ra] = (signed int) regs->gpr[rd];
+                       goto logical_done;
+#endif
+
+/*
+ * Shift instructions
+ */
+               case 24:        /* slw */
+                       sh = regs->gpr[rb] & 0x3f;
+                       if (sh < 32)
+                               regs->gpr[ra] = (regs->gpr[rd] << sh) & 0xffffffffUL;
+                       else
+                               regs->gpr[ra] = 0;
+                       goto logical_done;
+
+               case 536:       /* srw */
+                       sh = regs->gpr[rb] & 0x3f;
+                       if (sh < 32)
+                               regs->gpr[ra] = (regs->gpr[rd] & 0xffffffffUL) >> sh;
+                       else
+                               regs->gpr[ra] = 0;
+                       goto logical_done;
+
+               case 792:       /* sraw */
+                       sh = regs->gpr[rb] & 0x3f;
+                       ival = (signed int) regs->gpr[rd];
+                       regs->gpr[ra] = ival >> (sh < 32 ? sh : 31);
+                       if (ival < 0 && (sh >= 32 || (ival & ((1 << sh) - 1)) != 0))
+                               regs->xer |= XER_CA;
+                       else
+                               regs->xer &= ~XER_CA;
+                       goto logical_done;
+
+               case 824:       /* srawi */
+                       sh = rb;
+                       ival = (signed int) regs->gpr[rd];
+                       regs->gpr[ra] = ival >> sh;
+                       if (ival < 0 && (ival & ((1 << sh) - 1)) != 0)
+                               regs->xer |= XER_CA;
+                       else
+                               regs->xer &= ~XER_CA;
+                       goto logical_done;
+
+#ifdef __powerpc64__
+               case 27:        /* sld */
+                       sh = regs->gpr[rd] & 0x7f;
+                       if (sh < 64)
+                               regs->gpr[ra] = regs->gpr[rd] << sh;
+                       else
+                               regs->gpr[ra] = 0;
+                       goto logical_done;
+
+               case 539:       /* srd */
+                       sh = regs->gpr[rb] & 0x7f;
+                       if (sh < 64)
+                               regs->gpr[ra] = regs->gpr[rd] >> sh;
+                       else
+                               regs->gpr[ra] = 0;
+                       goto logical_done;
+
+               case 794:       /* srad */
+                       sh = regs->gpr[rb] & 0x7f;
+                       ival = (signed long int) regs->gpr[rd];
+                       regs->gpr[ra] = ival >> (sh < 64 ? sh : 63);
+                       if (ival < 0 && (sh >= 64 || (ival & ((1 << sh) - 1)) != 0))
+                               regs->xer |= XER_CA;
+                       else
+                               regs->xer &= ~XER_CA;
+                       goto logical_done;
+
+               case 826:       /* sradi with sh_5 = 0 */
+               case 827:       /* sradi with sh_5 = 1 */
+                       sh = rb | ((instr & 2) << 4);
+                       ival = (signed long int) regs->gpr[rd];
+                       regs->gpr[ra] = ival >> sh;
+                       if (ival < 0 && (ival & ((1 << sh) - 1)) != 0)
+                               regs->xer |= XER_CA;
+                       else
+                               regs->xer &= ~XER_CA;
+                       goto logical_done;
+#endif /* __powerpc64__ */
+
+/*
+ * Cache instructions
+ */
+               case 54:        /* dcbst */
+                       ea = xform_ea(instr, regs, 0);
+                       if (!address_ok(regs, ea, 8))
+                               return 0;
+                       err = 0;
+                       __cacheop_user_asmx(ea, err, "dcbst");
+                       if (err)
+                               return 0;
+                       goto instr_done;
+
+               case 86:        /* dcbf */
+                       ea = xform_ea(instr, regs, 0);
+                       if (!address_ok(regs, ea, 8))
+                               return 0;
+                       err = 0;
+                       __cacheop_user_asmx(ea, err, "dcbf");
+                       if (err)
+                               return 0;
+                       goto instr_done;
+
+               case 246:       /* dcbtst */
+                       if (rd == 0) {
+                               ea = xform_ea(instr, regs, 0);
+                               prefetchw((void *) ea);
+                       }
+                       goto instr_done;
+
+               case 278:       /* dcbt */
+                       if (rd == 0) {
+                               ea = xform_ea(instr, regs, 0);
+                               prefetch((void *) ea);
+                       }
+                       goto instr_done;
+
                }
+               break;
        }
-       return 0;
+
+       /*
+        * Following cases are for loads and stores, so bail out
+        * if we're in little-endian mode.
+        */
+       if (regs->msr & MSR_LE)
+               return 0;
+
+       /*
+        * Save register RA in case it's an update form load or store
+        * and the access faults.
+        */
+       old_ra = regs->gpr[ra];
+
+       switch (opcode) {
+       case 31:
+               u = instr & 0x40;
+               switch ((instr >> 1) & 0x3ff) {
+               case 20:        /* lwarx */
+                       ea = xform_ea(instr, regs, 0);
+                       if (ea & 3)
+                               break;          /* can't handle misaligned */
+                       err = -EFAULT;
+                       if (!address_ok(regs, ea, 4))
+                               goto ldst_done;
+                       err = 0;
+                       __get_user_asmx(val, ea, err, "lwarx");
+                       if (!err)
+                               regs->gpr[rd] = val;
+                       goto ldst_done;
+
+               case 150:       /* stwcx. */
+                       ea = xform_ea(instr, regs, 0);
+                       if (ea & 3)
+                               break;          /* can't handle misaligned */
+                       err = -EFAULT;
+                       if (!address_ok(regs, ea, 4))
+                               goto ldst_done;
+                       err = 0;
+                       __put_user_asmx(regs->gpr[rd], ea, err, "stwcx.", cr);
+                       if (!err)
+                               regs->ccr = (regs->ccr & 0x0fffffff) |
+                                       (cr & 0xe0000000) |
+                                       ((regs->xer >> 3) & 0x10000000);
+                       goto ldst_done;
+
+#ifdef __powerpc64__
+               case 84:        /* ldarx */
+                       ea = xform_ea(instr, regs, 0);
+                       if (ea & 7)
+                               break;          /* can't handle misaligned */
+                       err = -EFAULT;
+                       if (!address_ok(regs, ea, 8))
+                               goto ldst_done;
+                       err = 0;
+                       __get_user_asmx(val, ea, err, "ldarx");
+                       if (!err)
+                               regs->gpr[rd] = val;
+                       goto ldst_done;
+
+               case 214:       /* stdcx. */
+                       ea = xform_ea(instr, regs, 0);
+                       if (ea & 7)
+                               break;          /* can't handle misaligned */
+                       err = -EFAULT;
+                       if (!address_ok(regs, ea, 8))
+                               goto ldst_done;
+                       err = 0;
+                       __put_user_asmx(regs->gpr[rd], ea, err, "stdcx.", cr);
+                       if (!err)
+                               regs->ccr = (regs->ccr & 0x0fffffff) |
+                                       (cr & 0xe0000000) |
+                                       ((regs->xer >> 3) & 0x10000000);
+                       goto ldst_done;
+
+               case 21:        /* ldx */
+               case 53:        /* ldux */
+                       err = read_mem(&regs->gpr[rd], xform_ea(instr, regs, u),
+                                      8, regs);
+                       goto ldst_done;
+#endif
+
+               case 23:        /* lwzx */
+               case 55:        /* lwzux */
+                       err = read_mem(&regs->gpr[rd], xform_ea(instr, regs, u),
+                                      4, regs);
+                       goto ldst_done;
+
+               case 87:        /* lbzx */
+               case 119:       /* lbzux */
+                       err = read_mem(&regs->gpr[rd], xform_ea(instr, regs, u),
+                                      1, regs);
+                       goto ldst_done;
+
+#ifdef CONFIG_ALTIVEC
+               case 103:       /* lvx */
+               case 359:       /* lvxl */
+                       if (!(regs->msr & MSR_VEC))
+                               break;
+                       ea = xform_ea(instr, regs, 0);
+                       err = do_vec_load(rd, do_lvx, ea, regs);
+                       goto ldst_done;
+
+               case 231:       /* stvx */
+               case 487:       /* stvxl */
+                       if (!(regs->msr & MSR_VEC))
+                               break;
+                       ea = xform_ea(instr, regs, 0);
+                       err = do_vec_store(rd, do_stvx, ea, regs);
+                       goto ldst_done;
+#endif /* CONFIG_ALTIVEC */
+
+#ifdef __powerpc64__
+               case 149:       /* stdx */
+               case 181:       /* stdux */
+                       val = regs->gpr[rd];
+                       err = write_mem(val, xform_ea(instr, regs, u), 8, regs);
+                       goto ldst_done;
+#endif
+
+               case 151:       /* stwx */
+               case 183:       /* stwux */
+                       val = regs->gpr[rd];
+                       err = write_mem(val, xform_ea(instr, regs, u), 4, regs);
+                       goto ldst_done;
+
+               case 215:       /* stbx */
+               case 247:       /* stbux */
+                       val = regs->gpr[rd];
+                       err = write_mem(val, xform_ea(instr, regs, u), 1, regs);
+                       goto ldst_done;
+
+               case 279:       /* lhzx */
+               case 311:       /* lhzux */
+                       err = read_mem(&regs->gpr[rd], xform_ea(instr, regs, u),
+                                      2, regs);
+                       goto ldst_done;
+
+#ifdef __powerpc64__
+               case 341:       /* lwax */
+               case 373:       /* lwaux */
+                       err = read_mem(&regs->gpr[rd], xform_ea(instr, regs, u),
+                                      4, regs);
+                       if (!err)
+                               regs->gpr[rd] = (signed int) regs->gpr[rd];
+                       goto ldst_done;
+#endif
+
+               case 343:       /* lhax */
+               case 375:       /* lhaux */
+                       err = read_mem(&regs->gpr[rd], xform_ea(instr, regs, u),
+                                      2, regs);
+                       if (!err)
+                               regs->gpr[rd] = (signed short) regs->gpr[rd];
+                       goto ldst_done;
+
+               case 407:       /* sthx */
+               case 439:       /* sthux */
+                       val = regs->gpr[rd];
+                       err = write_mem(val, xform_ea(instr, regs, u), 2, regs);
+                       goto ldst_done;
+
+#ifdef __powerpc64__
+               case 532:       /* ldbrx */
+                       err = read_mem(&val, xform_ea(instr, regs, 0), 8, regs);
+                       if (!err)
+                               regs->gpr[rd] = byterev_8(val);
+                       goto ldst_done;
+
+#endif
+
+               case 534:       /* lwbrx */
+                       err = read_mem(&val, xform_ea(instr, regs, 0), 4, regs);
+                       if (!err)
+                               regs->gpr[rd] = byterev_4(val);
+                       goto ldst_done;
+
+               case 535:       /* lfsx */
+               case 567:       /* lfsux */
+                       if (!(regs->msr & MSR_FP))
+                               break;
+                       ea = xform_ea(instr, regs, u);
+                       err = do_fp_load(rd, do_lfs, ea, 4, regs);
+                       goto ldst_done;
+
+               case 599:       /* lfdx */
+               case 631:       /* lfdux */
+                       if (!(regs->msr & MSR_FP))
+                               break;
+                       ea = xform_ea(instr, regs, u);
+                       err = do_fp_load(rd, do_lfd, ea, 8, regs);
+                       goto ldst_done;
+
+               case 663:       /* stfsx */
+               case 695:       /* stfsux */
+                       if (!(regs->msr & MSR_FP))
+                               break;
+                       ea = xform_ea(instr, regs, u);
+                       err = do_fp_store(rd, do_stfs, ea, 4, regs);
+                       goto ldst_done;
+
+               case 727:       /* stfdx */
+               case 759:       /* stfdux */
+                       if (!(regs->msr & MSR_FP))
+                               break;
+                       ea = xform_ea(instr, regs, u);
+                       err = do_fp_store(rd, do_stfd, ea, 8, regs);
+                       goto ldst_done;
+
+#ifdef __powerpc64__
+               case 660:       /* stdbrx */
+                       val = byterev_8(regs->gpr[rd]);
+                       err = write_mem(val, xform_ea(instr, regs, 0), 8, regs);
+                       goto ldst_done;
+
+#endif
+               case 662:       /* stwbrx */
+                       val = byterev_4(regs->gpr[rd]);
+                       err = write_mem(val, xform_ea(instr, regs, 0), 4, regs);
+                       goto ldst_done;
+
+               case 790:       /* lhbrx */
+                       err = read_mem(&val, xform_ea(instr, regs, 0), 2, regs);
+                       if (!err)
+                               regs->gpr[rd] = byterev_2(val);
+                       goto ldst_done;
+
+               case 918:       /* sthbrx */
+                       val = byterev_2(regs->gpr[rd]);
+                       err = write_mem(val, xform_ea(instr, regs, 0), 2, regs);
+                       goto ldst_done;
+
+#ifdef CONFIG_VSX
+               case 844:       /* lxvd2x */
+               case 876:       /* lxvd2ux */
+                       if (!(regs->msr & MSR_VSX))
+                               break;
+                       rd |= (instr & 1) << 5;
+                       ea = xform_ea(instr, regs, u);
+                       err = do_vsx_load(rd, do_lxvd2x, ea, regs);
+                       goto ldst_done;
+
+               case 972:       /* stxvd2x */
+               case 1004:      /* stxvd2ux */
+                       if (!(regs->msr & MSR_VSX))
+                               break;
+                       rd |= (instr & 1) << 5;
+                       ea = xform_ea(instr, regs, u);
+                       err = do_vsx_store(rd, do_stxvd2x, ea, regs);
+                       goto ldst_done;
+
+#endif /* CONFIG_VSX */
+               }
+               break;
+
+       case 32:        /* lwz */
+       case 33:        /* lwzu */
+               err = read_mem(&regs->gpr[rd], dform_ea(instr, regs), 4, regs);
+               goto ldst_done;
+
+       case 34:        /* lbz */
+       case 35:        /* lbzu */
+               err = read_mem(&regs->gpr[rd], dform_ea(instr, regs), 1, regs);
+               goto ldst_done;
+
+       case 36:        /* stw */
+       case 37:        /* stwu */
+               val = regs->gpr[rd];
+               err = write_mem(val, dform_ea(instr, regs), 4, regs);
+               goto ldst_done;
+
+       case 38:        /* stb */
+       case 39:        /* stbu */
+               val = regs->gpr[rd];
+               err = write_mem(val, dform_ea(instr, regs), 1, regs);
+               goto ldst_done;
+
+       case 40:        /* lhz */
+       case 41:        /* lhzu */
+               err = read_mem(&regs->gpr[rd], dform_ea(instr, regs), 2, regs);
+               goto ldst_done;
+
+       case 42:        /* lha */
+       case 43:        /* lhau */
+               err = read_mem(&regs->gpr[rd], dform_ea(instr, regs), 2, regs);
+               if (!err)
+                       regs->gpr[rd] = (signed short) regs->gpr[rd];
+               goto ldst_done;
+
+       case 44:        /* sth */
+       case 45:        /* sthu */
+               val = regs->gpr[rd];
+               err = write_mem(val, dform_ea(instr, regs), 2, regs);
+               goto ldst_done;
+
+       case 46:        /* lmw */
+               ra = (instr >> 16) & 0x1f;
+               if (ra >= rd)
+                       break;          /* invalid form, ra in range to load */
+               ea = dform_ea(instr, regs);
+               do {
+                       err = read_mem(&regs->gpr[rd], ea, 4, regs);
+                       if (err)
+                               return 0;
+                       ea += 4;
+               } while (++rd < 32);
+               goto instr_done;
+
+       case 47:        /* stmw */
+               ea = dform_ea(instr, regs);
+               do {
+                       err = write_mem(regs->gpr[rd], ea, 4, regs);
+                       if (err)
+                               return 0;
+                       ea += 4;
+               } while (++rd < 32);
+               goto instr_done;
+
+       case 48:        /* lfs */
+       case 49:        /* lfsu */
+               if (!(regs->msr & MSR_FP))
+                       break;
+               ea = dform_ea(instr, regs);
+               err = do_fp_load(rd, do_lfs, ea, 4, regs);
+               goto ldst_done;
+
+       case 50:        /* lfd */
+       case 51:        /* lfdu */
+               if (!(regs->msr & MSR_FP))
+                       break;
+               ea = dform_ea(instr, regs);
+               err = do_fp_load(rd, do_lfd, ea, 8, regs);
+               goto ldst_done;
+
+       case 52:        /* stfs */
+       case 53:        /* stfsu */
+               if (!(regs->msr & MSR_FP))
+                       break;
+               ea = dform_ea(instr, regs);
+               err = do_fp_store(rd, do_stfs, ea, 4, regs);
+               goto ldst_done;
+
+       case 54:        /* stfd */
+       case 55:        /* stfdu */
+               if (!(regs->msr & MSR_FP))
+                       break;
+               ea = dform_ea(instr, regs);
+               err = do_fp_store(rd, do_stfd, ea, 8, regs);
+               goto ldst_done;
+
+#ifdef __powerpc64__
+       case 58:        /* ld[u], lwa */
+               switch (instr & 3) {
+               case 0:         /* ld */
+                       err = read_mem(&regs->gpr[rd], dsform_ea(instr, regs),
+                                      8, regs);
+                       goto ldst_done;
+               case 1:         /* ldu */
+                       err = read_mem(&regs->gpr[rd], dsform_ea(instr, regs),
+                                      8, regs);
+                       goto ldst_done;
+               case 2:         /* lwa */
+                       err = read_mem(&regs->gpr[rd], dsform_ea(instr, regs),
+                                      4, regs);
+                       if (!err)
+                               regs->gpr[rd] = (signed int) regs->gpr[rd];
+                       goto ldst_done;
+               }
+               break;
+
+       case 62:        /* std[u] */
+               val = regs->gpr[rd];
+               switch (instr & 3) {
+               case 0:         /* std */
+                       err = write_mem(val, dsform_ea(instr, regs), 8, regs);
+                       goto ldst_done;
+               case 1:         /* stdu */
+                       err = write_mem(val, dsform_ea(instr, regs), 8, regs);
+                       goto ldst_done;
+               }
+               break;
+#endif /* __powerpc64__ */
+
+       }
+       err = -EINVAL;
+
+ ldst_done:
+       if (err) {
+               regs->gpr[ra] = old_ra;
+               return 0;       /* invoke DSI if -EFAULT? */
+       }
+ instr_done:
+       regs->nip += 4;
+#ifdef __powerpc64__
+       if ((regs->msr & MSR_SF) == 0)
+               regs->nip &= 0xffffffffUL;
+#endif
+       return 1;
+
+ logical_done:
+       if (instr & 1)
+               set_cr0(regs, ra);
+       goto instr_done;
+
+ arith_done:
+       if (instr & 1)
+               set_cr0(regs, rd);
+       goto instr_done;
 }