[PATCH 0/6] add sysenter/syscall emulation for 32bit compat mode

[PATCH 0/6] add sysenter/syscall emulation for 32bit compat mode

[Andre Przywara]

sysenter/sysexit are not supported on AMD's 32bit compat mode, whereas
syscall is not supported on Intel's 32bit compat mode. To allow cross
vendor migration we emulate the missing instructions by setting up the
processor state accordingly.
The sysenter code was originally sketched by Amit Shah, it was completed,
debugged,  syscall added and made-to-work by Christoph Egger and polished
up by Andre Przywara.
Please note that sysret does not need to be emulated, because it will be
exectued in 64bit mode and returning to 32bit compat mode works on Intel.

This has been tested with GETPIDs in a tight loop in compat mode on both
Intel and AMD boxes. Additionally a 32-bit userland was booted under a
64-bit kernel and then cross-vendor migrated.

Please apply or comment ;-) 

Regards,
Andre.

Signed-off-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Christoph Egger <christoph.egger@amd.com>
Signed-off-by: Andre Przywara <andre.przywara@amd.com>

[PATCH 1/6] allow emulation of syscalls instructions on #UD

[Andre Przywara]

Add the opcodes for syscall, sysenter and sysexit to the list of instructions
handled by the undefined opcode handler.

Signed-off-by: Christoph Egger <christoph.egger@amd.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
---
 arch/x86/kvm/x86.c |   33 ++++++++++++++++++++++++++-------
 1 files changed, 26 insertions(+), 7 deletions(-)

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 6025e5b..88e159c 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -2632,14 +2632,33 @@ int emulate_instruction(struct kvm_vcpu *vcpu,
 
 		r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
 
-		/* Reject the instructions other than VMCALL/VMMCALL when
-		 * try to emulate invalid opcode */
+		/* Only allow emulation of specific instructions on #UD
+		 * (namely VMMCALL, sysenter, sysexit, syscall)*/
 		c = &vcpu->arch.emulate_ctxt.decode;
-		if ((emulation_type & EMULTYPE_TRAP_UD) &&
-		    (!(c->twobyte && c->b == 0x01 &&
-		      (c->modrm_reg == 0 || c->modrm_reg == 3) &&
-		       c->modrm_mod == 3 && c->modrm_rm == 1)))
-			return EMULATE_FAIL;
+		if (emulation_type & EMULTYPE_TRAP_UD) {
+			if (!c->twobyte)
+				return EMULATE_FAIL;
+			switch (c->b) {
+			case 0x01: /* VMMCALL */
+				if (c->modrm_mod != 3 || c->modrm_rm != 1)
+					return EMULATE_FAIL;
+				break;
+			case 0x34: /* sysenter */
+			case 0x35: /* sysexit */
+				if (c->modrm_mod != 0 || c->modrm_rm != 0)
+					return EMULATE_FAIL;
+				break;
+			case 0x05: /* syscall */
+				if (c->modrm_mod != 0 || c->modrm_rm != 0)
+					return EMULATE_FAIL;
+				break;
+			default:
+				return EMULATE_FAIL;
+			}
+
+			if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
+				return EMULATE_FAIL;
+		}
 
 		++vcpu->stat.insn_emulation;
 		if (r)  {
-- 
1.6.1.3

[PATCH 2/6] add missing EFLAGS bit definitions

[Andre Przywara]

Signed-off-by: Christoph Egger <christoph.egger@amd.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
---
 arch/x86/kvm/x86_emulate.c |    3 +++
 1 files changed, 3 insertions(+), 0 deletions(-)

diff --git a/arch/x86/kvm/x86_emulate.c b/arch/x86/kvm/x86_emulate.c
index 22c765d..e387c83 100644
--- a/arch/x86/kvm/x86_emulate.c
+++ b/arch/x86/kvm/x86_emulate.c
@@ -320,8 +320,11 @@ static u32 group2_table[] = {
 };
 
 /* EFLAGS bit definitions. */
+#define EFLG_VM (1<<17)
+#define EFLG_RF (1<<16)
 #define EFLG_OF (1<<11)
 #define EFLG_DF (1<<10)
+#define EFLG_IF (1<<9)
 #define EFLG_SF (1<<7)
 #define EFLG_ZF (1<<6)
 #define EFLG_AF (1<<4)
-- 
1.6.1.3

[PATCH 3/6] prepare for emulation of syscall instructions

[Andre Przywara]

Add the flags needed for syscall, sysenter and sysexit to the opcode table.
Catch (but for now ignore) the opcodes in the emulation switch/case.

Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Christoph Egger <christoph.egger@amd.com>
---
 arch/x86/kvm/x86_emulate.c |   17 +++++++++++++++--
 1 files changed, 15 insertions(+), 2 deletions(-)

diff --git a/arch/x86/kvm/x86_emulate.c b/arch/x86/kvm/x86_emulate.c
index e387c83..328ccba 100644
--- a/arch/x86/kvm/x86_emulate.c
+++ b/arch/x86/kvm/x86_emulate.c
@@ -32,6 +32,8 @@
 #include <linux/module.h>
 #include <asm/kvm_x86_emulate.h>
 
+#include "mmu.h"		/* for is_long_mode() */
+
 /*
  * Opcode effective-address decode tables.
  * Note that we only emulate instructions that have at least one memory
@@ -209,7 +211,7 @@ static u32 opcode_table[256] = {
 
 static u32 twobyte_table[256] = {
 	/* 0x00 - 0x0F */
-	0, Group | GroupDual | Group7, 0, 0, 0, 0, ImplicitOps, 0,
+	0, Group | GroupDual | Group7, 0, 0, 0, ImplicitOps, ImplicitOps, 0,
 	ImplicitOps, ImplicitOps, 0, 0, 0, ImplicitOps | ModRM, 0, 0,
 	/* 0x10 - 0x1F */
 	0, 0, 0, 0, 0, 0, 0, 0, ImplicitOps | ModRM, 0, 0, 0, 0, 0, 0, 0,
@@ -217,7 +219,9 @@ static u32 twobyte_table[256] = {
 	ModRM | ImplicitOps, ModRM, ModRM | ImplicitOps, ModRM, 0, 0, 0, 0,
 	0, 0, 0, 0, 0, 0, 0, 0,
 	/* 0x30 - 0x3F */
-	ImplicitOps, 0, ImplicitOps, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+	ImplicitOps, 0, ImplicitOps, 0,
+	ImplicitOps, ImplicitOps, 0, 0,
+	0, 0, 0, 0, 0, 0, 0, 0,
 	/* 0x40 - 0x47 */
 	DstReg | SrcMem | ModRM | Mov, DstReg | SrcMem | ModRM | Mov,
 	DstReg | SrcMem | ModRM | Mov, DstReg | SrcMem | ModRM | Mov,
@@ -1988,6 +1992,9 @@ twobyte_insn:
 			goto cannot_emulate;
 		}
 		break;
+	case 0x05: 		/* syscall */
+		goto cannot_emulate;
+		break;
 	case 0x06:
 		emulate_clts(ctxt->vcpu);
 		c->dst.type = OP_NONE;
@@ -2054,6 +2061,12 @@ twobyte_insn:
 		rc = X86EMUL_CONTINUE;
 		c->dst.type = OP_NONE;
 		break;
+	case 0x34:		/* sysenter */
+		goto cannot_emulate;
+		break;
+	case 0x35:		/* sysexit */
+		goto cannot_emulate;
+		break;
 	case 0x40 ... 0x4f:	/* cmov */
 		c->dst.val = c->dst.orig_val = c->src.val;
 		if (!test_cc(c->b, ctxt->eflags))
-- 
1.6.1.3

[PATCH 4/6] add syscall emulation

[Andre Przywara]

Handle #UD intercept of the syscall instruction in 32bit compat mode on
an Intel host.
Setup the segment descriptors for CS and SS and the EIP/ESP registers
according to the manual. Save the RIP and EFLAGS to the correct registers.

Signed-off-by: Christoph Egger <christoph.egger@amd.com>
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
---
 arch/x86/kvm/x86_emulate.c |   89 +++++++++++++++++++++++++++++++++++++++++++-
 1 files changed, 88 insertions(+), 1 deletions(-)

diff --git a/arch/x86/kvm/x86_emulate.c b/arch/x86/kvm/x86_emulate.c
index 328ccba..89bd53e 100644
--- a/arch/x86/kvm/x86_emulate.c
+++ b/arch/x86/kvm/x86_emulate.c
@@ -1397,6 +1397,90 @@ void toggle_interruptibility(struct x86_emulate_ctxt *ctxt, u32 mask)
 		ctxt->interruptibility = mask;
 }
 
+static inline void
+setup_syscalls_segments(struct x86_emulate_ctxt *ctxt,
+	struct kvm_segment *cs, struct kvm_segment *ss)
+{
+	memset(cs, 0, sizeof(struct kvm_segment));
+	kvm_x86_ops->get_segment(ctxt->vcpu, cs, VCPU_SREG_CS);
+	memset(ss, 0, sizeof(struct kvm_segment));
+
+	cs->l = 0;		/* will be adjusted later */
+	cs->base = 0;		/* flat segment */
+	cs->g = 1;		/* 4kb granularity */
+	cs->limit = 0xfffff;	/* 4GB limit */
+	cs->type = 0x0b;	/* Read, Execute, Accessed */
+	cs->s = 1;
+	cs->dpl = 0;		/* will be adjusted later */
+	cs->present = 1;
+	cs->db = 1;
+
+	ss->unusable = 0;
+	ss->base = 0;		/* flat segment */
+	ss->limit = 0xfffff;	/* 4GB limit */
+	ss->g = 1;		/* 4kb granularity */
+	ss->s = 1;
+	ss->type = 0x03;	/* Read/Write, Accessed */
+	ss->db = 1;		/* 32bit stack segment */
+	ss->dpl = 0;
+	ss->present = 1;
+}
+
+static int
+emulate_syscall(struct x86_emulate_ctxt *ctxt)
+{
+	struct decode_cache *c = &ctxt->decode;
+	struct kvm_segment cs, ss;
+	u64 msr_data;
+
+	/* syscall is not available in real mode */
+	if (c->lock_prefix || ctxt->mode == X86EMUL_MODE_REAL
+		|| !(ctxt->vcpu->arch.cr0 & X86_CR0_PE))
+		return -1;
+
+	setup_syscalls_segments(ctxt, &cs, &ss);
+
+	kvm_x86_ops->get_msr(ctxt->vcpu, MSR_STAR, &msr_data);
+	msr_data >>= 32;
+	cs.selector = (u16)(msr_data & 0xfffc);
+	ss.selector = (u16)(msr_data + 8);
+
+	if (is_long_mode(ctxt->vcpu)) {
+		cs.db = 0;
+		cs.l = 1;
+		if (ctxt->mode == X86EMUL_MODE_PROT64) {
+			/* Intel cares about granularity (g bit),
+			 * so we don't set the effective limit.
+			 */
+			cs.g = 1;
+			cs.limit = 0xffffffff;
+		}
+	}
+	kvm_x86_ops->set_segment(ctxt->vcpu, &cs, VCPU_SREG_CS);
+	kvm_x86_ops->set_segment(ctxt->vcpu, &ss, VCPU_SREG_SS);
+
+	c->regs[VCPU_REGS_RCX] = c->eip;
+	if (is_long_mode(ctxt->vcpu)) {
+		c->regs[VCPU_REGS_R11] = ctxt->eflags & ~EFLG_RF;
+
+		kvm_x86_ops->get_msr(ctxt->vcpu,
+			ctxt->mode == X86EMUL_MODE_PROT64 ?
+			MSR_LSTAR : MSR_CSTAR, &msr_data);
+		c->eip = msr_data;
+
+		kvm_x86_ops->get_msr(ctxt->vcpu, MSR_SYSCALL_MASK, &msr_data);
+		ctxt->eflags &= ~(msr_data | EFLG_RF);
+	} else {
+		/* legacy mode */
+		kvm_x86_ops->get_msr(ctxt->vcpu, MSR_STAR, &msr_data);
+		c->eip = (u32)msr_data;
+
+		ctxt->eflags &= ~(EFLG_VM | EFLG_IF | EFLG_RF);
+	}
+
+	return 0;
+}
+
 int
 x86_emulate_insn(struct x86_emulate_ctxt *ctxt, struct x86_emulate_ops *ops)
 {
@@ -1993,7 +2077,10 @@ twobyte_insn:
 		}
 		break;
 	case 0x05: 		/* syscall */
-		goto cannot_emulate;
+		if (emulate_syscall(ctxt) == -1)
+			goto cannot_emulate;
+		else
+			goto writeback;
 		break;
 	case 0x06:
 		emulate_clts(ctxt->vcpu);
-- 
1.6.1.3

Re: [PATCH 4/6] add syscall emulation

[Avi Kivity]

On 06/17/2009 04:50 PM, Andre Przywara wrote:
>
> +static inline void
> +setup_syscalls_segments(struct x86_emulate_ctxt *ctxt,
> +	struct kvm_segment *cs, struct kvm_segment *ss)
> +{
> +	memset(cs, 0, sizeof(struct kvm_segment));
> +	kvm_x86_ops->get_segment(ctxt->vcpu, cs, VCPU_SREG_CS);
> +	memset(ss, 0, sizeof(struct kvm_segment));
> +
> +	cs->l = 0;		/* will be adjusted later */
> +	cs->base = 0;		/* flat segment */
> +	cs->g = 1;		/* 4kb granularity */
> +	cs->limit = 0xfffff;	/* 4GB limit */
>    
...
> +	ss->limit = 0xfffff;	/* 4GB limit */
>    

limit in kvm_segment is expanded, so 4GB is 0xffffffff.

-- 
error compiling committee.c: too many arguments to function

Re: [PATCH 4/6] add syscall emulation

[Andre Przywara]

Avi Kivity wrote:
> On 06/17/2009 04:50 PM, Andre Przywara wrote:
>>
>> +static inline void
>> +setup_syscalls_segments(struct x86_emulate_ctxt *ctxt,
>> +    struct kvm_segment *cs, struct kvm_segment *ss)
>> +{
>> +    memset(cs, 0, sizeof(struct kvm_segment));
>> +    kvm_x86_ops->get_segment(ctxt->vcpu, cs, VCPU_SREG_CS);
>> +    memset(ss, 0, sizeof(struct kvm_segment));
>> +
>> +    cs->l = 0;        /* will be adjusted later */
>> +    cs->base = 0;        /* flat segment */
>> +    cs->g = 1;        /* 4kb granularity */
>> +    cs->limit = 0xfffff;    /* 4GB limit */
>>    
> ...
>> +    ss->limit = 0xfffff;    /* 4GB limit */
>>    
> 
> limit in kvm_segment is expanded, so 4GB is 0xffffffff.
You are right. This was probably a leftover from this cross vendor 
migration g-bit issue.
Beside the fix here this removes some code in the separate paths, so I 
remade patch 4-6/6 and will resend them.

Thanks for spotting this.

Regards,
Andre.

-- 
Andre Przywara
AMD-OSRC (Dresden)
Tel: x29712

Re: [PATCH 4/6] add syscall emulation

[Avi Kivity]

On 06/18/2009 01:27 PM, Andre Przywara wrote:
>> limit in kvm_segment is expanded, so 4GB is 0xffffffff.
>
> You are right. This was probably a leftover from this cross vendor 
> migration g-bit issue.
> Beside the fix here this removes some code in the separate paths, so I 
> remade patch 4-6/6 and will resend them.
>

Applied the new patchset; thanks.

-- 
error compiling committee.c: too many arguments to function

[PATCH 5/6] add sysenter emulation

[Andre Przywara]

Handle #UD intercept of the sysenter instruction in 32bit compat mode on
an AMD host.
Setup the segment descriptors for CS and SS and the EIP/ESP registers
according to the manual.

Signed-off-by: Christoph Egger <christoph.egger@amd.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
---
 arch/x86/kvm/x86_emulate.c |   72 +++++++++++++++++++++++++++++++++++++++++++-
 1 files changed, 71 insertions(+), 1 deletions(-)

diff --git a/arch/x86/kvm/x86_emulate.c b/arch/x86/kvm/x86_emulate.c
index 89bd53e..2f62aaa 100644
--- a/arch/x86/kvm/x86_emulate.c
+++ b/arch/x86/kvm/x86_emulate.c
@@ -1481,6 +1481,73 @@ emulate_syscall(struct x86_emulate_ctxt *ctxt)
 	return 0;
 }
 
+static int
+emulate_sysenter(struct x86_emulate_ctxt *ctxt)
+{
+	struct decode_cache *c = &ctxt->decode;
+	struct kvm_segment cs, ss;
+	u64 msr_data;
+
+	/* inject #UD if LOCK prefix is used */
+	if (c->lock_prefix)
+		return -1;
+
+	/* inject #GP if in real mode or paging is disabled */
+	if (ctxt->mode == X86EMUL_MODE_REAL ||
+		!(ctxt->vcpu->arch.cr0 & X86_CR0_PE)) {
+		kvm_inject_gp(ctxt->vcpu, 0);
+		return -1;
+	}
+
+	/* XXX sysenter/sysexit have not been tested in 64bit mode.
+	* Therefore, we inject an #UD.
+	*/
+	if (ctxt->mode == X86EMUL_MODE_PROT64)
+		return -1;
+
+	setup_syscalls_segments(ctxt, &cs, &ss);
+
+	kvm_x86_ops->get_msr(ctxt->vcpu, MSR_IA32_SYSENTER_CS, &msr_data);
+	switch (ctxt->mode) {
+	case X86EMUL_MODE_PROT32:
+		if ((msr_data & 0xfffc) == 0x0) {
+			kvm_inject_gp(ctxt->vcpu, 0);
+			return -1;
+		}
+		break;
+	case X86EMUL_MODE_PROT64:
+		if (msr_data == 0x0) {
+			kvm_inject_gp(ctxt->vcpu, 0);
+			return -1;
+		}
+		break;
+	}
+
+	ctxt->eflags &= ~(EFLG_VM | EFLG_IF | EFLG_RF);
+	cs.selector = (u16)msr_data;
+	cs.selector &= ~SELECTOR_RPL_MASK;
+	ss.selector = cs.selector + 8;
+	ss.selector &= ~SELECTOR_RPL_MASK;
+	if (ctxt->mode == X86EMUL_MODE_PROT64
+		|| is_long_mode(ctxt->vcpu)) {
+		cs.db = 0;
+		cs.l = 1;
+		cs.limit = 0xffffffff;
+		ss.limit = 0xffffffff;
+	}
+
+	kvm_x86_ops->set_segment(ctxt->vcpu, &cs, VCPU_SREG_CS);
+	kvm_x86_ops->set_segment(ctxt->vcpu, &ss, VCPU_SREG_SS);
+
+	kvm_x86_ops->get_msr(ctxt->vcpu, MSR_IA32_SYSENTER_EIP, &msr_data);
+	c->eip = msr_data;
+
+	kvm_x86_ops->get_msr(ctxt->vcpu, MSR_IA32_SYSENTER_ESP, &msr_data);
+	c->regs[VCPU_REGS_RSP] = msr_data;
+
+	return 0;
+}
+
 int
 x86_emulate_insn(struct x86_emulate_ctxt *ctxt, struct x86_emulate_ops *ops)
 {
@@ -2149,7 +2216,10 @@ twobyte_insn:
 		c->dst.type = OP_NONE;
 		break;
 	case 0x34:		/* sysenter */
-		goto cannot_emulate;
+		if (emulate_sysenter(ctxt) == -1)
+			goto cannot_emulate;
+		else
+			goto writeback;
 		break;
 	case 0x35:		/* sysexit */
 		goto cannot_emulate;
-- 
1.6.1.3

[PATCH 6/6] add sysexit emulation

[Andre Przywara]

Handle #UD intercept of the sysexit instruction in 64bit mode returning to
32bit compat mode on an AMD host.
Setup the segment descriptors for CS and SS and the EIP/ESP registers
according to the manual.

Signed-off-by: Christoph Egger <christoph.egger@amd.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
---
 arch/x86/kvm/x86_emulate.c |   79 +++++++++++++++++++++++++++++++++++++++++++-
 1 files changed, 78 insertions(+), 1 deletions(-)

diff --git a/arch/x86/kvm/x86_emulate.c b/arch/x86/kvm/x86_emulate.c
index 2f62aaa..7df05cc 100644
--- a/arch/x86/kvm/x86_emulate.c
+++ b/arch/x86/kvm/x86_emulate.c
@@ -1548,6 +1548,80 @@ emulate_sysenter(struct x86_emulate_ctxt *ctxt)
 	return 0;
 }
 
+static int
+emulate_sysexit(struct x86_emulate_ctxt *ctxt)
+{
+	struct decode_cache *c = &ctxt->decode;
+	struct kvm_segment cs, ss;
+	u64 msr_data;
+	int usermode;
+
+	/* inject #UD if LOCK prefix is used */
+	if (c->lock_prefix)
+		return -1;
+
+	/* inject #GP if in real mode or paging is disabled */
+	if (ctxt->mode == X86EMUL_MODE_REAL
+		|| !(ctxt->vcpu->arch.cr0 & X86_CR0_PE)) {
+		kvm_inject_gp(ctxt->vcpu, 0);
+		return -1;
+	}
+
+	/* sysexit must be called from CPL 0 */
+	if (kvm_x86_ops->get_cpl(ctxt->vcpu) != 0) {
+		kvm_inject_gp(ctxt->vcpu, 0);
+		return -1;
+	}
+
+	setup_syscalls_segments(ctxt, &cs, &ss);
+
+	if ((c->rex_prefix & 0x8) != 0x0)
+		usermode = X86EMUL_MODE_PROT64;
+	else
+		usermode = X86EMUL_MODE_PROT32;
+
+	/* We don't care about cs.g/ss.g bits
+	 * (= 4kb granularity) so we have to set the effective
+	 * limit here or we get a #GP in the guest, otherwise.
+	 */
+	cs.limit = 0xffffffff;
+	ss.limit = 0xffffffff;
+
+	cs.dpl = 3;
+	ss.dpl = 3;
+	kvm_x86_ops->get_msr(ctxt->vcpu, MSR_IA32_SYSENTER_CS, &msr_data);
+	switch (usermode) {
+	case X86EMUL_MODE_PROT32:
+		cs.selector = (u16)(msr_data + 16);
+		if ((msr_data & 0xfffc) == 0x0) {
+			kvm_inject_gp(ctxt->vcpu, 0);
+			return -1;
+		}
+		ss.selector = (u16)(msr_data + 24);
+		break;
+	case X86EMUL_MODE_PROT64:
+		cs.selector = (u16)(msr_data + 32);
+		if (msr_data == 0x0) {
+			kvm_inject_gp(ctxt->vcpu, 0);
+			return -1;
+		}
+		ss.selector = cs.selector + 8;
+		cs.db = 0;
+		cs.l = 1;
+		break;
+	}
+	cs.selector |= SELECTOR_RPL_MASK;
+	ss.selector |= SELECTOR_RPL_MASK;
+
+	kvm_x86_ops->set_segment(ctxt->vcpu, &cs, VCPU_SREG_CS);
+	kvm_x86_ops->set_segment(ctxt->vcpu, &ss, VCPU_SREG_SS);
+
+	c->eip = ctxt->vcpu->arch.regs[VCPU_REGS_RDX];
+	c->regs[VCPU_REGS_RSP] = ctxt->vcpu->arch.regs[VCPU_REGS_RCX];
+
+	return 0;
+}
+
 int
 x86_emulate_insn(struct x86_emulate_ctxt *ctxt, struct x86_emulate_ops *ops)
 {
@@ -2222,7 +2296,10 @@ twobyte_insn:
 			goto writeback;
 		break;
 	case 0x35:		/* sysexit */
-		goto cannot_emulate;
+		if (emulate_sysexit(ctxt) == -1)
+			goto cannot_emulate;
+		else
+			goto writeback;
 		break;
 	case 0x40 ... 0x4f:	/* cmov */
 		c->dst.val = c->dst.orig_val = c->src.val;
-- 
1.6.1.3

Re: [PATCH 0/6] add sysenter/syscall emulation for 32bit compat mode

[Avi Kivity]

On 06/17/2009 04:50 PM, Andre Przywara wrote:
> sysenter/sysexit are not supported on AMD's 32bit compat mode, whereas
> syscall is not supported on Intel's 32bit compat mode. To allow cross
> vendor migration we emulate the missing instructions by setting up the
> processor state accordingly.
> The sysenter code was originally sketched by Amit Shah, it was completed,
> debugged,  syscall added and made-to-work by Christoph Egger and polished
> up by Andre Przywara.
> Please note that sysret does not need to be emulated, because it will be
> exectued in 64bit mode and returning to 32bit compat mode works on Intel.
>
> This has been tested with GETPIDs in a tight loop in compat mode on both
> Intel and AMD boxes. Additionally a 32-bit userland was booted under a
> 64-bit kernel and then cross-vendor migrated.
>
> Please apply or comment ;-)
>    

Looks good except for the limit thing I pointed out.  If the correct 
thing is to change it to 0xffffffff, let me know and I'll do it in 
place, instead of resending the whole thing.

-- 
error compiling committee.c: too many arguments to function

[PATCH 4/6 v2] add syscall emulation

[Andre Przywara]

Handle #UD intercept of the syscall instruction in 32bit compat mode on
an Intel host.
Setup the segment descriptors for CS and SS and the EIP/ESP registers
according to the manual. Save the RIP and EFLAGS to the correct registers.

Signed-off-by: Christoph Egger <christoph.egger@amd.com>
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
---
 arch/x86/kvm/x86_emulate.c |   82 +++++++++++++++++++++++++++++++++++++++++++-
 1 files changed, 81 insertions(+), 1 deletions(-)

Avi,
these are the new versions of patch 4-6/6. If you need delta patches, tell me.

Thanks,
Andre.

diff --git a/arch/x86/kvm/x86_emulate.c b/arch/x86/kvm/x86_emulate.c
index 328ccba..d0a51c4 100644
--- a/arch/x86/kvm/x86_emulate.c
+++ b/arch/x86/kvm/x86_emulate.c
@@ -1397,6 +1397,83 @@ void toggle_interruptibility(struct x86_emulate_ctxt *ctxt, u32 mask)
 		ctxt->interruptibility = mask;
 }
 
+static inline void
+setup_syscalls_segments(struct x86_emulate_ctxt *ctxt,
+	struct kvm_segment *cs, struct kvm_segment *ss)
+{
+	memset(cs, 0, sizeof(struct kvm_segment));
+	kvm_x86_ops->get_segment(ctxt->vcpu, cs, VCPU_SREG_CS);
+	memset(ss, 0, sizeof(struct kvm_segment));
+
+	cs->l = 0;		/* will be adjusted later */
+	cs->base = 0;		/* flat segment */
+	cs->g = 1;		/* 4kb granularity */
+	cs->limit = 0xffffffff;	/* 4GB limit */
+	cs->type = 0x0b;	/* Read, Execute, Accessed */
+	cs->s = 1;
+	cs->dpl = 0;		/* will be adjusted later */
+	cs->present = 1;
+	cs->db = 1;
+
+	ss->unusable = 0;
+	ss->base = 0;		/* flat segment */
+	ss->limit = 0xffffffff;	/* 4GB limit */
+	ss->g = 1;		/* 4kb granularity */
+	ss->s = 1;
+	ss->type = 0x03;	/* Read/Write, Accessed */
+	ss->db = 1;		/* 32bit stack segment */
+	ss->dpl = 0;
+	ss->present = 1;
+}
+
+static int
+emulate_syscall(struct x86_emulate_ctxt *ctxt)
+{
+	struct decode_cache *c = &ctxt->decode;
+	struct kvm_segment cs, ss;
+	u64 msr_data;
+
+	/* syscall is not available in real mode */
+	if (c->lock_prefix || ctxt->mode == X86EMUL_MODE_REAL
+		|| !(ctxt->vcpu->arch.cr0 & X86_CR0_PE))
+		return -1;
+
+	setup_syscalls_segments(ctxt, &cs, &ss);
+
+	kvm_x86_ops->get_msr(ctxt->vcpu, MSR_STAR, &msr_data);
+	msr_data >>= 32;
+	cs.selector = (u16)(msr_data & 0xfffc);
+	ss.selector = (u16)(msr_data + 8);
+
+	if (is_long_mode(ctxt->vcpu)) {
+		cs.db = 0;
+		cs.l = 1;
+	}
+	kvm_x86_ops->set_segment(ctxt->vcpu, &cs, VCPU_SREG_CS);
+	kvm_x86_ops->set_segment(ctxt->vcpu, &ss, VCPU_SREG_SS);
+
+	c->regs[VCPU_REGS_RCX] = c->eip;
+	if (is_long_mode(ctxt->vcpu)) {
+		c->regs[VCPU_REGS_R11] = ctxt->eflags & ~EFLG_RF;
+
+		kvm_x86_ops->get_msr(ctxt->vcpu,
+			ctxt->mode == X86EMUL_MODE_PROT64 ?
+			MSR_LSTAR : MSR_CSTAR, &msr_data);
+		c->eip = msr_data;
+
+		kvm_x86_ops->get_msr(ctxt->vcpu, MSR_SYSCALL_MASK, &msr_data);
+		ctxt->eflags &= ~(msr_data | EFLG_RF);
+	} else {
+		/* legacy mode */
+		kvm_x86_ops->get_msr(ctxt->vcpu, MSR_STAR, &msr_data);
+		c->eip = (u32)msr_data;
+
+		ctxt->eflags &= ~(EFLG_VM | EFLG_IF | EFLG_RF);
+	}
+
+	return 0;
+}
+
 int
 x86_emulate_insn(struct x86_emulate_ctxt *ctxt, struct x86_emulate_ops *ops)
 {
@@ -1993,7 +2070,10 @@ twobyte_insn:
 		}
 		break;
 	case 0x05: 		/* syscall */
-		goto cannot_emulate;
+		if (emulate_syscall(ctxt) == -1)
+			goto cannot_emulate;
+		else
+			goto writeback;
 		break;
 	case 0x06:
 		emulate_clts(ctxt->vcpu);
-- 
1.6.1.3

[PATCH 5/6 v2] add sysenter emulation

[Andre Przywara]

Handle #UD intercept of the sysenter instruction in 32bit compat mode on
an AMD host.
Setup the segment descriptors for CS and SS and the EIP/ESP registers
according to the manual.

Signed-off-by: Christoph Egger <christoph.egger@amd.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
---
 arch/x86/kvm/x86_emulate.c |   70 +++++++++++++++++++++++++++++++++++++++++++-
 1 files changed, 69 insertions(+), 1 deletions(-)

diff --git a/arch/x86/kvm/x86_emulate.c b/arch/x86/kvm/x86_emulate.c
index d0a51c4..fdf75f6 100644
--- a/arch/x86/kvm/x86_emulate.c
+++ b/arch/x86/kvm/x86_emulate.c
@@ -1474,6 +1474,71 @@ emulate_syscall(struct x86_emulate_ctxt *ctxt)
 	return 0;
 }
 
+static int
+emulate_sysenter(struct x86_emulate_ctxt *ctxt)
+{
+	struct decode_cache *c = &ctxt->decode;
+	struct kvm_segment cs, ss;
+	u64 msr_data;
+
+	/* inject #UD if LOCK prefix is used */
+	if (c->lock_prefix)
+		return -1;
+
+	/* inject #GP if in real mode or paging is disabled */
+	if (ctxt->mode == X86EMUL_MODE_REAL ||
+		!(ctxt->vcpu->arch.cr0 & X86_CR0_PE)) {
+		kvm_inject_gp(ctxt->vcpu, 0);
+		return -1;
+	}
+
+	/* XXX sysenter/sysexit have not been tested in 64bit mode.
+	* Therefore, we inject an #UD.
+	*/
+	if (ctxt->mode == X86EMUL_MODE_PROT64)
+		return -1;
+
+	setup_syscalls_segments(ctxt, &cs, &ss);
+
+	kvm_x86_ops->get_msr(ctxt->vcpu, MSR_IA32_SYSENTER_CS, &msr_data);
+	switch (ctxt->mode) {
+	case X86EMUL_MODE_PROT32:
+		if ((msr_data & 0xfffc) == 0x0) {
+			kvm_inject_gp(ctxt->vcpu, 0);
+			return -1;
+		}
+		break;
+	case X86EMUL_MODE_PROT64:
+		if (msr_data == 0x0) {
+			kvm_inject_gp(ctxt->vcpu, 0);
+			return -1;
+		}
+		break;
+	}
+
+	ctxt->eflags &= ~(EFLG_VM | EFLG_IF | EFLG_RF);
+	cs.selector = (u16)msr_data;
+	cs.selector &= ~SELECTOR_RPL_MASK;
+	ss.selector = cs.selector + 8;
+	ss.selector &= ~SELECTOR_RPL_MASK;
+	if (ctxt->mode == X86EMUL_MODE_PROT64
+		|| is_long_mode(ctxt->vcpu)) {
+		cs.db = 0;
+		cs.l = 1;
+	}
+
+	kvm_x86_ops->set_segment(ctxt->vcpu, &cs, VCPU_SREG_CS);
+	kvm_x86_ops->set_segment(ctxt->vcpu, &ss, VCPU_SREG_SS);
+
+	kvm_x86_ops->get_msr(ctxt->vcpu, MSR_IA32_SYSENTER_EIP, &msr_data);
+	c->eip = msr_data;
+
+	kvm_x86_ops->get_msr(ctxt->vcpu, MSR_IA32_SYSENTER_ESP, &msr_data);
+	c->regs[VCPU_REGS_RSP] = msr_data;
+
+	return 0;
+}
+
 int
 x86_emulate_insn(struct x86_emulate_ctxt *ctxt, struct x86_emulate_ops *ops)
 {
@@ -2142,7 +2207,10 @@ twobyte_insn:
 		c->dst.type = OP_NONE;
 		break;
 	case 0x34:		/* sysenter */
-		goto cannot_emulate;
+		if (emulate_sysenter(ctxt) == -1)
+			goto cannot_emulate;
+		else
+			goto writeback;
 		break;
 	case 0x35:		/* sysexit */
 		goto cannot_emulate;
-- 
1.6.1.3

[PATCH 6/6 v2] add sysexit emulation

[Andre Przywara]

Handle #UD intercept of the sysexit instruction in 64bit mode returning to
32bit compat mode on an AMD host.
Setup the segment descriptors for CS and SS and the EIP/ESP registers
according to the manual.

Signed-off-by: Christoph Egger <christoph.egger@amd.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
---
 arch/x86/kvm/x86_emulate.c |   72 +++++++++++++++++++++++++++++++++++++++++++-
 1 files changed, 71 insertions(+), 1 deletions(-)

diff --git a/arch/x86/kvm/x86_emulate.c b/arch/x86/kvm/x86_emulate.c
index fdf75f6..6849868 100644
--- a/arch/x86/kvm/x86_emulate.c
+++ b/arch/x86/kvm/x86_emulate.c
@@ -1539,6 +1539,73 @@ emulate_sysenter(struct x86_emulate_ctxt *ctxt)
 	return 0;
 }
 
+static int
+emulate_sysexit(struct x86_emulate_ctxt *ctxt)
+{
+	struct decode_cache *c = &ctxt->decode;
+	struct kvm_segment cs, ss;
+	u64 msr_data;
+	int usermode;
+
+	/* inject #UD if LOCK prefix is used */
+	if (c->lock_prefix)
+		return -1;
+
+	/* inject #GP if in real mode or paging is disabled */
+	if (ctxt->mode == X86EMUL_MODE_REAL
+		|| !(ctxt->vcpu->arch.cr0 & X86_CR0_PE)) {
+		kvm_inject_gp(ctxt->vcpu, 0);
+		return -1;
+	}
+
+	/* sysexit must be called from CPL 0 */
+	if (kvm_x86_ops->get_cpl(ctxt->vcpu) != 0) {
+		kvm_inject_gp(ctxt->vcpu, 0);
+		return -1;
+	}
+
+	setup_syscalls_segments(ctxt, &cs, &ss);
+
+	if ((c->rex_prefix & 0x8) != 0x0)
+		usermode = X86EMUL_MODE_PROT64;
+	else
+		usermode = X86EMUL_MODE_PROT32;
+
+	cs.dpl = 3;
+	ss.dpl = 3;
+	kvm_x86_ops->get_msr(ctxt->vcpu, MSR_IA32_SYSENTER_CS, &msr_data);
+	switch (usermode) {
+	case X86EMUL_MODE_PROT32:
+		cs.selector = (u16)(msr_data + 16);
+		if ((msr_data & 0xfffc) == 0x0) {
+			kvm_inject_gp(ctxt->vcpu, 0);
+			return -1;
+		}
+		ss.selector = (u16)(msr_data + 24);
+		break;
+	case X86EMUL_MODE_PROT64:
+		cs.selector = (u16)(msr_data + 32);
+		if (msr_data == 0x0) {
+			kvm_inject_gp(ctxt->vcpu, 0);
+			return -1;
+		}
+		ss.selector = cs.selector + 8;
+		cs.db = 0;
+		cs.l = 1;
+		break;
+	}
+	cs.selector |= SELECTOR_RPL_MASK;
+	ss.selector |= SELECTOR_RPL_MASK;
+
+	kvm_x86_ops->set_segment(ctxt->vcpu, &cs, VCPU_SREG_CS);
+	kvm_x86_ops->set_segment(ctxt->vcpu, &ss, VCPU_SREG_SS);
+
+	c->eip = ctxt->vcpu->arch.regs[VCPU_REGS_RDX];
+	c->regs[VCPU_REGS_RSP] = ctxt->vcpu->arch.regs[VCPU_REGS_RCX];
+
+	return 0;
+}
+
 int
 x86_emulate_insn(struct x86_emulate_ctxt *ctxt, struct x86_emulate_ops *ops)
 {
@@ -2213,7 +2280,10 @@ twobyte_insn:
 			goto writeback;
 		break;
 	case 0x35:		/* sysexit */
-		goto cannot_emulate;
+		if (emulate_sysexit(ctxt) == -1)
+			goto cannot_emulate;
+		else
+			goto writeback;
 		break;
 	case 0x40 ... 0x4f:	/* cmov */
 		c->dst.val = c->dst.orig_val = c->src.val;
-- 
1.6.1.3