- don''t provide cmpxchg8b emulation where not needed (i.e. page table
ops on 64-bit hv)
- properly deal with stack operands (push/pop)
- synchronize prefix handling with hvm''s instrlen determination and
about to be committed
privileged op decoder (changes coming with the 32on64 patches)
- support cmpxchg16b if the CPU supports it
- a couple of minor adjustments
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Index: 2006-12-18/xen/arch/x86/mm.c
==================================================================---
2006-12-18.orig/xen/arch/x86/mm.c 2006-12-15 16:33:59.000000000 +0100
+++ 2006-12-18/xen/arch/x86/mm.c 2006-12-18 09:36:41.000000000 +0100
@@ -3240,6 +3240,7 @@ static int ptwr_emulated_cmpxchg(
container_of(ctxt, struct ptwr_emulate_ctxt, ctxt));
}
+#ifdef __i386__
static int ptwr_emulated_cmpxchg8b(
enum x86_segment seg,
unsigned long offset,
@@ -3255,13 +3256,16 @@ static int ptwr_emulated_cmpxchg8b(
offset, ((u64)old_hi << 32) | old, ((u64)new_hi << 32) |
new, 8, 1,
container_of(ctxt, struct ptwr_emulate_ctxt, ctxt));
}
+#endif
static struct x86_emulate_ops ptwr_emulate_ops = {
.read = ptwr_emulated_read,
.insn_fetch = ptwr_emulated_read,
.write = ptwr_emulated_write,
.cmpxchg = ptwr_emulated_cmpxchg,
- .cmpxchg8b = ptwr_emulated_cmpxchg8b
+#ifdef __i386__
+ .cmpxchg2 = ptwr_emulated_cmpxchg8b
+#endif
};
/* Write page fault handler: check if guest is trying to modify a PTE. */
Index: 2006-12-18/xen/arch/x86/mm/shadow/common.c
==================================================================---
2006-12-18.orig/xen/arch/x86/mm/shadow/common.c 2006-12-15 16:33:59.000000000
+0100
+++ 2006-12-18/xen/arch/x86/mm/shadow/common.c 2006-12-18 09:36:41.000000000
+0100
@@ -291,6 +291,7 @@ hvm_emulate_cmpxchg(enum x86_segment seg
v, addr, old, new, bytes, sh_ctxt);
}
+#ifdef __i386__
static int
hvm_emulate_cmpxchg8b(enum x86_segment seg,
unsigned long offset,
@@ -314,13 +315,31 @@ hvm_emulate_cmpxchg8b(enum x86_segment s
return v->arch.shadow.mode->x86_emulate_cmpxchg8b(
v, addr, old_lo, old_hi, new_lo, new_hi, sh_ctxt);
}
+#endif
+
+static unsigned int
+hvm_stack_word_size(struct x86_emulate_ctxt *ctxt)
+{
+#ifdef __x86_64__
+ if ( ctxt->mode == X86EMUL_MODE_PROT64 )
+ return 8;
+#endif
+
+ return hvm_get_seg_reg(x86_seg_ss,
+ container_of(ctxt,
+ struct sh_emulate_ctxt,
+ ctxt))->attr.fields.db ? 4 : 2;
+}
static struct x86_emulate_ops hvm_shadow_emulator_ops = {
.read = hvm_emulate_read,
.insn_fetch = hvm_emulate_insn_fetch,
.write = hvm_emulate_write,
.cmpxchg = hvm_emulate_cmpxchg,
- .cmpxchg8b = hvm_emulate_cmpxchg8b,
+#ifdef __i386__
+ .cmpxchg2 = hvm_emulate_cmpxchg8b,
+#endif
+ .stksz = hvm_stack_word_size
};
static int
@@ -371,6 +390,7 @@ pv_emulate_cmpxchg(enum x86_segment seg,
v, offset, old, new, bytes, sh_ctxt);
}
+#ifdef __i386__
static int
pv_emulate_cmpxchg8b(enum x86_segment seg,
unsigned long offset,
@@ -386,13 +406,16 @@ pv_emulate_cmpxchg8b(enum x86_segment se
return v->arch.shadow.mode->x86_emulate_cmpxchg8b(
v, offset, old_lo, old_hi, new_lo, new_hi, sh_ctxt);
}
+#endif
static struct x86_emulate_ops pv_shadow_emulator_ops = {
.read = pv_emulate_read,
.insn_fetch = pv_emulate_read,
.write = pv_emulate_write,
.cmpxchg = pv_emulate_cmpxchg,
- .cmpxchg8b = pv_emulate_cmpxchg8b,
+#ifdef __i386__
+ .cmpxchg2 = pv_emulate_cmpxchg8b,
+#endif
};
struct x86_emulate_ops *shadow_init_emulation(
Index: 2006-12-18/xen/arch/x86/mm/shadow/multi.c
==================================================================---
2006-12-18.orig/xen/arch/x86/mm/shadow/multi.c 2006-12-15 16:33:59.000000000
+0100
+++ 2006-12-18/xen/arch/x86/mm/shadow/multi.c 2006-12-18 09:36:41.000000000
+0100
@@ -3960,7 +3960,8 @@ sh_x86_emulate_cmpxchg(struct vcpu *v, u
return rv;
}
-int
+#ifdef __i386__
+static int
sh_x86_emulate_cmpxchg8b(struct vcpu *v, unsigned long vaddr,
unsigned long old_lo, unsigned long old_hi,
unsigned long new_lo, unsigned long new_hi,
@@ -3999,6 +4000,7 @@ sh_x86_emulate_cmpxchg8b(struct vcpu *v,
shadow_audit_tables(v);
return rv;
}
+#endif
/**************************************************************************/
@@ -4286,7 +4288,9 @@ struct shadow_paging_mode sh_paging_mode
.detach_old_tables = sh_detach_old_tables,
.x86_emulate_write = sh_x86_emulate_write,
.x86_emulate_cmpxchg = sh_x86_emulate_cmpxchg,
+#ifdef __i386__
.x86_emulate_cmpxchg8b = sh_x86_emulate_cmpxchg8b,
+#endif
.make_monitor_table = sh_make_monitor_table,
.destroy_monitor_table = sh_destroy_monitor_table,
.guest_map_l1e = sh_guest_map_l1e,
Index: 2006-12-18/xen/arch/x86/x86_emulate.c
==================================================================---
2006-12-18.orig/xen/arch/x86/x86_emulate.c 2006-12-13 11:15:54.000000000 +0100
+++ 2006-12-18/xen/arch/x86/x86_emulate.c 2006-12-18 09:36:41.000000000 +0100
@@ -15,6 +15,7 @@
#include <xen/types.h>
#include <xen/lib.h>
#include <asm/regs.h>
+#include <asm/processor.h>
#undef cmpxchg
#endif
#include <asm-x86/x86_emulate.h>
@@ -375,7 +376,10 @@ do{ __asm__ __volatile__ (
/* Fetch next part of the instruction being emulated. */
#define insn_fetch_bytes(_size) \
({ unsigned long _x; \
- rc = ops->insn_fetch(x86_seg_cs, _regs.eip, &_x, (_size), ctxt);
\
+ if ( _regs.eip - ctxt->regs->eip < 16 - (_size) )
\
+ rc = ops->insn_fetch(x86_seg_cs, _regs.eip, &_x, (_size), ctxt);
\
+ else \
+ goto cannot_emulate; \
if ( rc != 0 ) \
goto done; \
_regs.eip += (_size); \
@@ -389,6 +393,14 @@ do{ __asm__ __volatile__ (
(__ea & ((1UL << (ad_bytes << 3)) - 1))); \
})
+#define truncate_sp() \
+({ \
+ if ( !stksz ) \
+ stksz = !ops->stksz ? ad_default : ops->stksz(ctxt); \
+ ((stksz == sizeof(unsigned long)) ? _regs.esp : \
+ (_regs.esp & ((1UL << (stksz << 3)) - 1))); \
+})
+
/* Update address held in a register, based on addressing mode. */
#define register_address_increment(reg, inc) \
do { \
@@ -400,6 +412,18 @@ do {
(((reg) + _inc) & ((1UL << (ad_bytes << 3)) -
1)); \
} while (0)
+#define sp_increment(inc) \
+do { \
+ int _inc = (inc); /* signed type ensures sign extension to long */ \
+ if ( !stksz ) \
+ stksz = !ops->stksz ? ad_default : ops->stksz(ctxt); \
+ if ( stksz == sizeof(unsigned long) ) \
+ _regs.esp += _inc; \
+ else \
+ _regs.esp = (_regs.esp & ~((1UL << (stksz << 3)) - 1))
| \
+ ((_regs.esp + _inc) & ((1UL << (stksz << 3)) -
1)); \
+} while (0)
+
void *
decode_register(
uint8_t modrm_reg, struct cpu_user_regs *regs, int highbyte_regs)
@@ -450,7 +474,8 @@ x86_emulate_memop(
uint8_t b, d, sib, sib_index, sib_base, twobyte = 0, rex_prefix = 0;
uint8_t modrm, modrm_mod = 0, modrm_reg = 0, modrm_rm = 0;
- unsigned int op_bytes, ad_bytes, lock_prefix = 0, rep_prefix = 0, i;
+ unsigned int op_bytes, ad_bytes, lock_prefix = 0, rep_prefix = 0;
+ unsigned int op_default, ad_default, stksz = 0;
int rc = 0;
struct operand src, dst;
int mode = ctxt->mode;
@@ -467,15 +492,15 @@ x86_emulate_memop(
{
case X86EMUL_MODE_REAL:
case X86EMUL_MODE_PROT16:
- op_bytes = ad_bytes = 2;
+ op_default = op_bytes = ad_default = ad_bytes = 2;
break;
case X86EMUL_MODE_PROT32:
- op_bytes = ad_bytes = 4;
+ op_default = op_bytes = ad_default = ad_bytes = 4;
break;
#ifdef __x86_64__
case X86EMUL_MODE_PROT64:
- op_bytes = 4;
- ad_bytes = 8;
+ op_default = op_bytes = 4;
+ ad_default = ad_bytes = 8;
break;
#endif
default:
@@ -483,18 +508,18 @@ x86_emulate_memop(
}
/* Legacy prefixes. */
- for ( i = 0; i < 8; i++ )
+ for ( ; ; )
{
switch ( b = insn_fetch_type(uint8_t) )
{
case 0x66: /* operand-size override */
- op_bytes ^= 6; /* switch between 2/4 bytes */
+ op_bytes = op_default ^ 6; /* switch between 2/4 bytes */
break;
case 0x67: /* address-size override */
if ( mode == X86EMUL_MODE_PROT64 )
- ad_bytes ^= 12; /* switch between 4/8 bytes */
+ ad_bytes = ad_default ^ 12; /* switch between 4/8 bytes */
else
- ad_bytes ^= 6; /* switch between 2/4 bytes */
+ ad_bytes = ad_default ^ 6; /* switch between 2/4 bytes */
break;
case 0x2e: /* CS override */
ea.mem.seg = x86_seg_cs;
@@ -517,25 +542,29 @@ x86_emulate_memop(
case 0xf0: /* LOCK */
lock_prefix = 1;
break;
+ case 0xf2: /* REPNE/REPNZ */
case 0xf3: /* REP/REPE/REPZ */
rep_prefix = 1;
break;
- case 0xf2: /* REPNE/REPNZ */
- break;
+#ifdef __x86_64__
+ case 0x40 ... 0x4f:
+ if ( mode == X86EMUL_MODE_PROT64 )
+ {
+ rex_prefix = b;
+ continue;
+ }
+ /* FALLTHRU */
+#endif
default:
goto done_prefixes;
}
+ rex_prefix = 0;
}
done_prefixes:
/* REX prefix. */
- if ( (mode == X86EMUL_MODE_PROT64) && ((b & 0xf0) == 0x40) )
- {
- rex_prefix = b;
- if ( b & 8 ) /* REX.W */
- op_bytes = 8;
- b = insn_fetch_type(uint8_t);
- }
+ if ( rex_prefix & 8 ) /* REX.W */
+ op_bytes = 8;
/* Opcode byte(s). */
d = opcode_table[b];
@@ -611,7 +640,16 @@ x86_emulate_memop(
if ( (modrm_mod == 0) && ((sib_base & 7) == 5) )
ea.mem.off += insn_fetch_type(int32_t);
else
+ {
ea.mem.off += *(long*)decode_register(sib_base, &_regs,
0);
+ if ( sib_base == 4 && !twobyte && b == 0x8f
)
+ {
+ /* POP to an address with esp as base register. */
+ if ( !stksz )
+ stksz = !ops->stksz ? ad_default :
ops->stksz(ctxt);
+ ea.mem.off += stksz;
+ }
+ }
}
else
{
@@ -865,12 +903,12 @@ x86_emulate_memop(
break;
case 0x8f: /* pop (sole member of Grp1a) */
/* 64-bit mode: POP always pops a 64-bit operand. */
- if ( mode == X86EMUL_MODE_PROT64 )
+ if ( mode == X86EMUL_MODE_PROT64 && dst.bytes == 4 )
dst.bytes = 8;
- if ( (rc = ops->read(x86_seg_ss, truncate_ea(_regs.esp),
+ if ( (rc = ops->read(x86_seg_ss, truncate_sp(),
&dst.val, dst.bytes, ctxt)) != 0 )
goto done;
- register_address_increment(_regs.esp, dst.bytes);
+ sp_increment(dst.bytes);
break;
case 0xc0 ... 0xc1: grp2: /* Grp2 */
switch ( modrm_reg & 7 )
@@ -941,15 +979,15 @@ x86_emulate_memop(
break;
case 6: /* push */
/* 64-bit mode: PUSH always pushes a 64-bit operand. */
- if ( mode == X86EMUL_MODE_PROT64 )
+ if ( mode == X86EMUL_MODE_PROT64 && dst.bytes == 4 )
{
dst.bytes = 8;
if ( (rc = ops->read(dst.mem.seg, dst.mem.off,
&dst.val, 8, ctxt)) != 0 )
goto done;
}
- register_address_increment(_regs.esp, -dst.bytes);
- if ( (rc = ops->write(x86_seg_ss, truncate_ea(_regs.esp),
+ sp_increment(-dst.bytes);
+ if ( (rc = ops->write(x86_seg_ss, truncate_sp(),
dst.val, dst.bytes, ctxt)) != 0 )
goto done;
dst.val = dst.orig_val; /* skanky: disable writeback */
@@ -999,7 +1037,7 @@ x86_emulate_memop(
special_insn:
if ( twobyte )
goto twobyte_special_insn;
- if ( rep_prefix )
+ if ( rep_prefix && (b & ~3) != 0xa0 )
{
if ( _regs.ecx == 0 )
{
@@ -1148,6 +1186,7 @@ x86_emulate_memop(
case 1: goto bts;
case 2: goto btr;
case 3: goto btc;
+ default: goto cannot_emulate;
}
break;
case 0xbe: /* movsx rm8,r{16,32,64} */
@@ -1180,57 +1219,66 @@ x86_emulate_memop(
case 0x0d: /* GrpP (prefetch) */
case 0x18: /* Grp16 (prefetch/nop) */
break;
- case 0xc7: /* Grp9 (cmpxchg8b) */
-#if defined(__i386__)
- {
- unsigned long old_lo, old_hi;
- if ( (rc = ops->read(ea.mem.seg, ea.mem.off+0, &old_lo, 4,
ctxt)) ||
- (rc = ops->read(ea.mem.seg, ea.mem.off+4, &old_hi, 4,
ctxt)) )
- goto done;
- if ( (old_lo != _regs.eax) || (old_hi != _regs.edx) )
- {
- _regs.eax = old_lo;
- _regs.edx = old_hi;
- _regs.eflags &= ~EFLG_ZF;
- }
- else if ( ops->cmpxchg8b == NULL )
- {
- rc = X86EMUL_UNHANDLEABLE;
- goto done;
- }
- else
+ case 0xc7: /* Grp9 (cmpxchg{8,16}b) */
+#ifdef __x86_64__
+ if ( op_bytes != 8 )
{
- if ( (rc = ops->cmpxchg8b(ea.mem.seg, ea.mem.off, old_lo,
old_hi,
- _regs.ebx, _regs.ecx, ctxt)) != 0 )
+ unsigned long old, new;
+ if ( (rc = ops->read(ea.mem.seg, ea.mem.off, &old, 8, ctxt))
)
goto done;
- _regs.eflags |= EFLG_ZF;
- }
- break;
- }
-#elif defined(__x86_64__)
- {
- unsigned long old, new;
- if ( (rc = ops->read(ea.mem.seg, ea.mem.off, &old, 8, ctxt)) !=
0 )
- goto done;
- if ( ((uint32_t)(old>>0) != (uint32_t)_regs.eax) ||
- ((uint32_t)(old>>32) != (uint32_t)_regs.edx) )
- {
- _regs.eax = (uint32_t)(old>>0);
- _regs.edx = (uint32_t)(old>>32);
- _regs.eflags &= ~EFLG_ZF;
+ if ( ((uint32_t)(old>>0) != (uint32_t)_regs.eax) ||
+ ((uint32_t)(old>>32) != (uint32_t)_regs.edx) )
+ {
+ _regs.eax = (uint32_t)(old>>0);
+ _regs.edx = (uint32_t)(old>>32);
+ _regs.eflags &= ~EFLG_ZF;
+ }
+ else
+ {
+ new = (_regs.ecx<<32)|(uint32_t)_regs.ebx;
+ if ( (rc = ops->cmpxchg(ea.mem.seg, ea.mem.off, old,
+ new, 8, ctxt)) != 0 )
+ goto done;
+ _regs.eflags |= EFLG_ZF;
+ }
}
else
+#endif
{
- new = (_regs.ecx<<32)|(uint32_t)_regs.ebx;
- if ( (rc = ops->cmpxchg(ea.mem.seg, ea.mem.off, old,
- new, 8, ctxt)) != 0 )
+ unsigned long old_lo, old_hi;
+ if ( (rc = ops->read(ea.mem.seg, ea.mem.off, &old_lo,
+ sizeof(old_lo), ctxt)) ||
+ (rc = ops->read(ea.mem.seg, ea.mem.off+sizeof(old_lo),
+ &old_hi, sizeof(old_hi), ctxt)) )
+ goto done;
+ if ( (old_lo != _regs.eax) || (old_hi != _regs.edx) )
+ {
+ _regs.eax = old_lo;
+ _regs.edx = old_hi;
+ _regs.eflags &= ~EFLG_ZF;
+ }
+ else if ( ops->cmpxchg2 == NULL )
+ {
+ rc = X86EMUL_UNHANDLEABLE;
goto done;
- _regs.eflags |= EFLG_ZF;
+ }
+#ifdef __x86_64__
+ else if ( !cpu_has_cmpxchg16b )
+ {
+ rc = X86EMUL_UNHANDLEABLE;
+ goto done;
+ }
+#endif
+ else
+ {
+ if ( (rc = ops->cmpxchg2(ea.mem.seg, ea.mem.off, old_lo,
old_hi,
+ _regs.ebx, _regs.ecx, ctxt)) != 0 )
+ goto done;
+ _regs.eflags |= EFLG_ZF;
+ }
}
break;
}
-#endif
- }
goto writeback;
cannot_emulate:
Index: 2006-12-18/xen/include/asm-x86/cpufeature.h
==================================================================---
2006-12-18.orig/xen/include/asm-x86/cpufeature.h 2006-12-13 11:15:56.000000000
+0100
+++ 2006-12-18/xen/include/asm-x86/cpufeature.h 2006-12-18 09:36:41.000000000
+0100
@@ -118,6 +118,7 @@
#define cpu_has_cyrix_arr boot_cpu_has(X86_FEATURE_CYRIX_ARR)
#define cpu_has_centaur_mcr boot_cpu_has(X86_FEATURE_CENTAUR_MCR)
#define cpu_has_clflush boot_cpu_has(X86_FEATURE_CLFLSH)
+#define cpu_has_cmpxchg16b 0
#else /* __x86_64__ */
#define cpu_has_vme 0
#define cpu_has_de 1
@@ -140,6 +141,7 @@
#define cpu_has_cyrix_arr 0
#define cpu_has_centaur_mcr 0
#define cpu_has_clflush boot_cpu_has(X86_FEATURE_CLFLSH)
+#define cpu_has_cmpxchg16b boot_cpu_has(X86_FEATURE_CX16)
#endif
#endif /* __ASM_I386_CPUFEATURE_H */
Index: 2006-12-18/xen/include/asm-x86/shadow.h
==================================================================---
2006-12-18.orig/xen/include/asm-x86/shadow.h 2006-12-15 16:33:59.000000000 +0100
+++ 2006-12-18/xen/include/asm-x86/shadow.h 2006-12-18 09:36:41.000000000 +0100
@@ -157,12 +157,14 @@ struct shadow_paging_mode {
unsigned long new,
unsigned int bytes,
struct sh_emulate_ctxt *sh_ctxt);
+#ifdef __i386__
int (*x86_emulate_cmpxchg8b )(struct vcpu *v, unsigned long va,
unsigned long old_lo,
unsigned long old_hi,
unsigned long new_lo,
unsigned long new_hi,
struct sh_emulate_ctxt *sh_ctxt);
+#endif
mfn_t (*make_monitor_table )(struct vcpu *v);
void (*destroy_monitor_table )(struct vcpu *v, mfn_t mmfn);
void * (*guest_map_l1e )(struct vcpu *v, unsigned long va,
Index: 2006-12-18/xen/include/asm-x86/x86_emulate.h
==================================================================---
2006-12-18.orig/xen/include/asm-x86/x86_emulate.h 2006-12-13 11:15:56.000000000
+0100
+++ 2006-12-18/xen/include/asm-x86/x86_emulate.h 2006-12-18 09:36:41.000000000
+0100
@@ -39,8 +39,9 @@ enum x86_segment {
* some out-of-band mechanism, unknown to the emulator. The memop signals
* failure by returning X86EMUL_PROPAGATE_FAULT to the emulator, which will
* then immediately bail.
- * 2. Valid access sizes are 1, 2, 4 and 8 bytes. On x86/32 systems only
- * cmpxchg8b_emulated need support 8-byte accesses.
+ * 2. Valid access sizes are 1, 2, 4, 8, and 16 bytes. On x86/32 systems only
+ * cmpxchg2_emulated need support 8-byte accesses. On x86/64 systems only
+ * cmpxchg2_emulated need support 16-byte accesses.
* 3. The emulator cannot handle 64-bit mode emulation on an x86/32 system.
*/
/* Access completed successfully: continue emulation as normal. */
@@ -110,16 +111,17 @@ struct x86_emulate_ops
struct x86_emulate_ctxt *ctxt);
/*
- * cmpxchg8b: Emulate an atomic (LOCKed) CMPXCHG8B operation.
+ * cmpxchg2: Emulate an atomic (LOCKed) CMPXCHG{8,16}B operation.
* @old: [IN ] Value expected to be current at @addr.
* @new: [IN ] Value to write to @addr.
* NOTES:
- * 1. This function is only ever called when emulating a real CMPXCHG8B.
- * 2. This function is *never* called on x86/64 systems.
- * 2. Not defining this function (i.e., specifying NULL) is equivalent
+ * 1. This function is only ever called when emulating a real
CMPXCHG{8,16}B.
+ * 2. This function is *never* called on x86/64 systems for emulating
+ * CMPXCHG8B.
+ * 3. Not defining this function (i.e., specifying NULL) is equivalent
* to defining a function that always returns X86EMUL_UNHANDLEABLE.
*/
- int (*cmpxchg8b)(
+ int (*cmpxchg2)(
enum x86_segment seg,
unsigned long offset,
unsigned long old_lo,
@@ -127,6 +129,16 @@ struct x86_emulate_ops
unsigned long new_lo,
unsigned long new_hi,
struct x86_emulate_ctxt *ctxt);
+
+ /*
+ * stksz: Determine the item size of the guest stack.
+ * NOTE:
+ * Not defining this function (i.e., specifying NULL) is equivalent
+ * to defining a function that returns the default address size of
+ * the execution mode (X86EMUL_MODE_xxx).
+ */
+ unsigned int (*stksz)(
+ struct x86_emulate_ctxt *ctxt);
};
struct cpu_user_regs;
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel