x86/speculation: Mitigate eIBRS PBRSB predictions with WRMSR
Subsystems (5)
PATCH 1/1 x86/speculation: Mitigate eIBRS PBRSB predictions with WRMSR
tl;dr: The existing mitigation for eIBRS PBRSB predictions uses an INT3 to ensure a call instruction retires before a following unbalanced RET. Replace this with a WRMSR serialising instruction which has a lower performance penalty.
== Background ==
eIBRS (enhanced indirect branch restricted speculation) is used to prevent predictor addresses from one privilege domain from being used for prediction in a higher privilege domain.
== Problem ==
On processors with eIBRS protections there can be a case where upon VM exit a guest address may be used as an RSB prediction for an unbalanced RET if a CALL instruction hasn't yet been retired. This is termed PBRSB (Post-Barrier Return Stack Buffer).
A mitigation for this was introduced in:
(2b1299322016731d56807aa49254a5ea3080b6b3 x86/speculation: Add RSB VM Exit protections)
This mitigation [1] has a ~1% performance impact on VM exit compared to without it [2].
== Solution ==
The WRMSR instruction can be used as a speculation barrier and a serialising instruction. Use this on the VM exit path instead to ensure that a CALL instruction (in this case the call to vmx_spec_ctrl_restore_host) has retired before the prediction of a following unbalanced RET.
This mitigation [3] has a negligible performance impact.
== Testing ==
Run the outl_to_kernel kvm-unit-tests test 200 times per configuration which counts the cycles for an exit to kernel mode.
[1] With existing mitigation:
Average: 2026 cycles
[2] With no mitigation:
Average: 2008 cycles
[3] With proposed mitigation:
Average: 2008 cycles
--- arch/x86/include/asm/nospec-branch.h | 7 +++---- arch/x86/kvm/vmx/vmenter.S | 3 +-- arch/x86/kvm/vmx/vmx.c | 5 +++++ 3 files changed, 9 insertions(+), 6 deletions(-)
arch/x86/include/asm/nospec-branch.h +3 -4
@@ -159,10 +159,9 @@* A simpler FILL_RETURN_BUFFER macro. Don't make people use the CPP * monstrosity above, manually. */ -.macro FILL_RETURN_BUFFER reg:req nr:req ftr:req ftr2=ALT_NOT(X86_FEATURE_ALWAYS) -ALTERNATIVE_2 "jmp .Lskip_rsb_\@" , \ - __stringify(__FILL_RETURN_BUFFER(\reg,\nr)), \ftr, \ - __stringify(__FILL_ONE_RETURN), \ftr2 +.macro FILL_RETURN_BUFFER reg:req nr:req ftr:req + ALTERNATIVE "jmp .Lskip_rsb_\@", \ + __stringify(__FILL_RETURN_BUFFER(\reg,\nr)), \ftr .Lskip_rsb_\@: .endm
arch/x86/kvm/vmx/vmenter.S +1 -2
@@ -231,8 +231,7 @@ SYM_INNER_LABEL(vmx_vmexit, SYM_L_GLOBAL) * single call to retire, before the first unbalanced RET. */ - FILL_RETURN_BUFFER %_ASM_CX, RSB_CLEAR_LOOPS, X86_FEATURE_RSB_VMEXIT,\ - X86_FEATURE_RSB_VMEXIT_LITE + FILL_RETURN_BUFFER %_ASM_CX, RSB_CLEAR_LOOPS, X86_FEATURE_RSB_VMEXIT pop %_ASM_ARG2 /* @flags */
arch/x86/kvm/vmx/vmx.c +5 -0
@@ -7049,8 +7049,13 @@ void noinstr vmx_spec_ctrl_restore_host(struct vcpu_vmx *vmx,* For legacy IBRS, the IBRS bit always needs to be written after * transitioning from a less privileged predictor mode, regardless of * whether the guest/host values differ. +* +* For eIBRS affected by Post Barrier RSB Predictions a serialising +* instruction (wrmsr) must be executed to ensure a call instruction has +* retired before the prediction of a following unbalanced ret. */ if (cpu_feature_enabled (X86_FEATURE_KERNEL_IBRS ) || +cpu_feature_enabled (X86_FEATURE_RSB_VMEXIT_LITE ) || vmx->spec_ctrl != hostval)native_wrmsrl (MSR_IA32_SPEC_CTRL , hostval);
Signature · 1 line
2.17.1
Quoted 5 lines · Suraj Jitindar Singh
tl;dr: The existing mitigation for eIBRS PBRSB predictions uses an INT3 to ensure a call instruction retires before a following unbalanced RET. Replace this with a WRMSR serialising instruction which has a lower performance penalty.
The INT3 is only on a speculative path and should not impact performance.
Quoted 19 lines
== Background ==
eIBRS (enhanced indirect branch restricted speculation) is used to prevent predictor addresses from one privilege domain from being used for prediction in a higher privilege domain.
== Problem ==
On processors with eIBRS protections there can be a case where upon VM exit a guest address may be used as an RSB prediction for an unbalanced RET if a CALL instruction hasn't yet been retired. This is termed PBRSB (Post-Barrier Return Stack Buffer).
A mitigation for this was introduced in:
(2b1299322016731d56807aa49254a5ea3080b6b3x86/speculation: Add RSB VM Exit protections)This mitigation [1] has a ~1% performance impact on VM exit compared to without it [2].
Quoted 6 lines
== Solution ==
The WRMSR instruction can be used as a speculation barrier and a serialising instruction. Use this on the VM exit path instead to ensure that a CALL instruction (in this case the call to
vmx_spec_ctrl_restore_host) has retired before the prediction of a following unbalanced RET.
I don't buy this solution. According to
https://www.intel.com/content/www/us/en/developer/articles/technical/software-security-guidance/advisory-guidance/post-barrier-return-stack-buffer-predictions.html:
"Note that a WRMSR instruction (used to set IBRS, for example), could also serve as a speculation barrier for such a sequence in place of LFENCE."
This says only that you can replace the LFENCE with a WRMSR. It doesn't say that you can drop the rest of the sequence.
Quoted 7 lines
This mitigation [3] has a negligible performance impact.
== Testing ==
Run the
outl_to_kernelkvm-unit-tests test 200 times per configuration which counts the cycles for an exit to kernel mode.
Quoted 6 lines
[1] With existing mitigation:
Average: 2026 cycles
[2] With no mitigation:
Average: 2008 cycles
[3] With proposed mitigation:
Average: 2008 cycles
What testing did you do to see that this is an effective mitigation? Improved timings are irrelevant if it doesn't work.
Quoted 93 lines · Suraj Jitindar Singh
tl;dr: The existing mitigation for eIBRS PBRSB predictions uses an INT3 to ensure a call instruction retires before a following unbalanced RET. Replace this with a WRMSR serialising instruction which has a lower performance penalty.
== Background ==
eIBRS (enhanced indirect branch restricted speculation) is used to prevent predictor addresses from one privilege domain from being used for prediction in a higher privilege domain.
== Problem ==
On processors with eIBRS protections there can be a case where upon VM exit a guest address may be used as an RSB prediction for an unbalanced RET if a CALL instruction hasn't yet been retired. This is termed PBRSB (Post-Barrier Return Stack Buffer).
A mitigation for this was introduced in:
(2b1299322016731d56807aa49254a5ea3080b6b3x86/speculation: Add RSB VM Exit protections)This mitigation [1] has a ~1% performance impact on VM exit compared to without it [2].
== Solution ==
The WRMSR instruction can be used as a speculation barrier and a serialising instruction. Use this on the VM exit path instead to ensure that a CALL instruction (in this case the call to
vmx_spec_ctrl_restore_host) has retired before the prediction of a following unbalanced RET.This mitigation [3] has a negligible performance impact.
== Testing ==
Run the
outl_to_kernelkvm-unit-tests test 200 times per configuration which counts the cycles for an exit to kernel mode.[1] With existing mitigation:
Average: 2026 cycles
[2] With no mitigation:
Average: 2008 cycles
[3] With proposed mitigation:
Average: 2008 cyclesSigned-off-by Suraj Jitindar Singh <[email protected]>--- arch/x86/include/asm/nospec-branch.h | 7 +++---- arch/x86/kvm/vmx/vmenter.S | 3 +-- arch/x86/kvm/vmx/vmx.c | 5 +++++ 3 files changed, 9 insertions(+), 6 deletions(-)arch/x86/include/asm/nospec-branch.h +3
-4@@ -159,10 +159,9 @@* A simpler FILL_RETURN_BUFFER macro. Don't make people use the CPP * monstrosity above, manually. */ -.macro FILL_RETURN_BUFFER reg:req nr:req ftr:req ftr2=ALT_NOT(X86_FEATURE_ALWAYS) -ALTERNATIVE_2 "jmp .Lskip_rsb_\@" , \ - __stringify(__FILL_RETURN_BUFFER(\reg,\nr)), \ftr, \ - __stringify(__FILL_ONE_RETURN), \ftr2 +.macro FILL_RETURN_BUFFER reg:req nr:req ftr:req + ALTERNATIVE "jmp .Lskip_rsb_\@", \ + __stringify(__FILL_RETURN_BUFFER(\reg,\nr)), \ftr .Lskip_rsb_\@: .endmarch/x86/kvm/vmx/vmenter.S +1
-2@@ -231,8 +231,7 @@ SYM_INNER_LABEL(vmx_vmexit, SYM_L_GLOBAL) * single call to retire, before the first unbalanced RET. */ - FILL_RETURN_BUFFER %_ASM_CX, RSB_CLEAR_LOOPS, X86_FEATURE_RSB_VMEXIT,\ - X86_FEATURE_RSB_VMEXIT_LITE + FILL_RETURN_BUFFER %_ASM_CX, RSB_CLEAR_LOOPS, X86_FEATURE_RSB_VMEXIT pop %_ASM_ARG2 /* @flags */arch/x86/kvm/vmx/vmx.c +0
-0@@ -7049,8 +7049,13 @@ void noinstr vmx_spec_ctrl_restore_host(struct vcpu_vmx *vmx,* For legacy IBRS, the IBRS bit always needs to be written after * transitioning from a less privileged predictor mode, regardless of
arch/x86/kvm/vmx/vmx.c void noinstr vmx_spec_ctrl_restore_host(struct vcpu_vmx *vmx, · line 7060
Quoted 10 lines
arch/x86/kvm/vmx/vmx.c +5
-0* whether the guest/host values differ. +* +* For eIBRS affected by Post Barrier RSB Predictions a serialising +* instruction (wrmsr) must be executed to ensure a call instruction has +* retired before the prediction of a following unbalanced ret. */ if (cpu_feature_enabled (X86_FEATURE_KERNEL_IBRS ) || +cpu_feature_enabled (X86_FEATURE_RSB_VMEXIT_LITE ) || vmx->spec_ctrl != hostval)native_wrmsrl (MSR_IA32_SPEC_CTRL , hostval);
Okay. I see how this almost meets the requirements. But this WRMSR is conditional, which means that there's a speculative path through this code that ends up at the unbalanced RET without executing the WRMSR.
Also, for your timings of "no mitigation" and this proposed mitigation to be the same, I assume that the guest in your timing test has a different IA32_SPEC_CTRL value than the host, which isn't always going to be the case in practice. How much does this WRMSR cost if the guest and the host have the same IA32_SPEC_CTRL value?
Quoted 3 lines
arch/x86/kvm/vmx/vmx.c +0
-1--2.17.1
Pawan Gupta
reviewerQuoted 100 lines · Jim Mattson
Quoted 99 lines · Suraj Jitindar Singh
tl;dr: The existing mitigation for eIBRS PBRSB predictions uses an INT3 to ensure a call instruction retires before a following unbalanced RET. Replace this with a WRMSR serialising instruction which has a lower performance penalty.
== Background ==
eIBRS (enhanced indirect branch restricted speculation) is used to prevent predictor addresses from one privilege domain from being used for prediction in a higher privilege domain.
== Problem ==
On processors with eIBRS protections there can be a case where upon VM exit a guest address may be used as an RSB prediction for an unbalanced RET if a CALL instruction hasn't yet been retired. This is termed PBRSB (Post-Barrier Return Stack Buffer).
A mitigation for this was introduced in:
(2b1299322016731d56807aa49254a5ea3080b6b3x86/speculation: Add RSB VM Exit protections)This mitigation [1] has a ~1% performance impact on VM exit compared to without it [2].
== Solution ==
The WRMSR instruction can be used as a speculation barrier and a serialising instruction. Use this on the VM exit path instead to ensure that a CALL instruction (in this case the call to
vmx_spec_ctrl_restore_host) has retired before the prediction of a following unbalanced RET.This mitigation [3] has a negligible performance impact.
== Testing ==
Run the
outl_to_kernelkvm-unit-tests test 200 times per configuration which counts the cycles for an exit to kernel mode.[1] With existing mitigation:
Average: 2026 cycles
[2] With no mitigation:
Average: 2008 cycles
[3] With proposed mitigation:
Average: 2008 cyclesSigned-off-by Suraj Jitindar Singh <[email protected]>--- arch/x86/include/asm/nospec-branch.h | 7 +++---- arch/x86/kvm/vmx/vmenter.S | 3 +-- arch/x86/kvm/vmx/vmx.c | 5 +++++ 3 files changed, 9 insertions(+), 6 deletions(-)arch/x86/include/asm/nospec-branch.h +3
-4@@ -159,10 +159,9 @@* A simpler FILL_RETURN_BUFFER macro. Don't make people use the CPP * monstrosity above, manually. */ -.macro FILL_RETURN_BUFFER reg:req nr:req ftr:req ftr2=ALT_NOT(X86_FEATURE_ALWAYS) -ALTERNATIVE_2 "jmp .Lskip_rsb_\@" , \ - __stringify(__FILL_RETURN_BUFFER(\reg,\nr)), \ftr, \ - __stringify(__FILL_ONE_RETURN), \ftr2 +.macro FILL_RETURN_BUFFER reg:req nr:req ftr:req + ALTERNATIVE "jmp .Lskip_rsb_\@", \ + __stringify(__FILL_RETURN_BUFFER(\reg,\nr)), \ftr .Lskip_rsb_\@: .endmarch/x86/kvm/vmx/vmenter.S +1
-2@@ -231,8 +231,7 @@ SYM_INNER_LABEL(vmx_vmexit, SYM_L_GLOBAL) * single call to retire, before the first unbalanced RET. */ - FILL_RETURN_BUFFER %_ASM_CX, RSB_CLEAR_LOOPS, X86_FEATURE_RSB_VMEXIT,\ - X86_FEATURE_RSB_VMEXIT_LITE + FILL_RETURN_BUFFER %_ASM_CX, RSB_CLEAR_LOOPS, X86_FEATURE_RSB_VMEXIT pop %_ASM_ARG2 /* @flags */arch/x86/kvm/vmx/vmx.c +4
-0@@ -7049,8 +7049,13 @@ void noinstr vmx_spec_ctrl_restore_host(struct vcpu_vmx *vmx,* For legacy IBRS, the IBRS bit always needs to be written after * transitioning from a less privileged predictor mode, regardless of * whether the guest/host values differ. +* +* For eIBRS affected by Post Barrier RSB Predictions a serialising +* instruction (wrmsr) must be executed to ensure a call instruction has +* retired before the prediction of a following unbalanced ret. */
Quoted 8 lines
arch/x86/kvm/vmx/vmx.c void noinstr vmx_spec_ctrl_restore_host(struct vcpu_vmx *vmx, · line 7060
Quoted 4 lines
arch/x86/kvm/vmx/vmx.c +1
-0if (cpu_feature_enabled (X86_FEATURE_KERNEL_IBRS ) || +cpu_feature_enabled (X86_FEATURE_RSB_VMEXIT_LITE ) || vmx->spec_ctrl != hostval)native_wrmsrl (MSR_IA32_SPEC_CTRL , hostval);Okay. I see how this almost meets the requirements. But this WRMSR is conditional, which means that there's a speculative path through this code that ends up at the unbalanced RET without executing the WRMSR.
Agree. I was just about to post this.
Quoted 93 lines · Suraj Jitindar Singh
tl;dr: The existing mitigation for eIBRS PBRSB predictions uses an INT3 to ensure a call instruction retires before a following unbalanced RET. Replace this with a WRMSR serialising instruction which has a lower performance penalty.
== Background ==
eIBRS (enhanced indirect branch restricted speculation) is used to prevent predictor addresses from one privilege domain from being used for prediction in a higher privilege domain.
== Problem ==
On processors with eIBRS protections there can be a case where upon VM exit a guest address may be used as an RSB prediction for an unbalanced RET if a CALL instruction hasn't yet been retired. This is termed PBRSB (Post-Barrier Return Stack Buffer).
A mitigation for this was introduced in:
(2b1299322016731d56807aa49254a5ea3080b6b3x86/speculation: Add RSB VM Exit protections)This mitigation [1] has a ~1% performance impact on VM exit compared to without it [2].
== Solution ==
The WRMSR instruction can be used as a speculation barrier and a serialising instruction. Use this on the VM exit path instead to ensure that a CALL instruction (in this case the call to
vmx_spec_ctrl_restore_host) has retired before the prediction of a following unbalanced RET.This mitigation [3] has a negligible performance impact.
== Testing ==
Run the
outl_to_kernelkvm-unit-tests test 200 times per configuration which counts the cycles for an exit to kernel mode.[1] With existing mitigation:
Average: 2026 cycles
[2] With no mitigation:
Average: 2008 cycles
[3] With proposed mitigation:
Average: 2008 cyclesSigned-off-by Suraj Jitindar Singh <[email protected]>--- arch/x86/include/asm/nospec-branch.h | 7 +++---- arch/x86/kvm/vmx/vmenter.S | 3 +-- arch/x86/kvm/vmx/vmx.c | 5 +++++ 3 files changed, 9 insertions(+), 6 deletions(-)arch/x86/include/asm/nospec-branch.h +3
-4@@ -159,10 +159,9 @@* A simpler FILL_RETURN_BUFFER macro. Don't make people use the CPP * monstrosity above, manually. */ -.macro FILL_RETURN_BUFFER reg:req nr:req ftr:req ftr2=ALT_NOT(X86_FEATURE_ALWAYS) -ALTERNATIVE_2 "jmp .Lskip_rsb_\@" , \ - __stringify(__FILL_RETURN_BUFFER(\reg,\nr)), \ftr, \ - __stringify(__FILL_ONE_RETURN), \ftr2 +.macro FILL_RETURN_BUFFER reg:req nr:req ftr:req + ALTERNATIVE "jmp .Lskip_rsb_\@", \ + __stringify(__FILL_RETURN_BUFFER(\reg,\nr)), \ftr .Lskip_rsb_\@: .endmarch/x86/kvm/vmx/vmenter.S +1
-2@@ -231,8 +231,7 @@ SYM_INNER_LABEL(vmx_vmexit, SYM_L_GLOBAL) * single call to retire, before the first unbalanced RET. */ - FILL_RETURN_BUFFER %_ASM_CX, RSB_CLEAR_LOOPS, X86_FEATURE_RSB_VMEXIT,\ - X86_FEATURE_RSB_VMEXIT_LITE + FILL_RETURN_BUFFER %_ASM_CX, RSB_CLEAR_LOOPS, X86_FEATURE_RSB_VMEXIT pop %_ASM_ARG2 /* @flags */arch/x86/kvm/vmx/vmx.c +0
-0@@ -7049,8 +7049,13 @@ void noinstr vmx_spec_ctrl_restore_host(struct vcpu_vmx *vmx,* For legacy IBRS, the IBRS bit always needs to be written after * transitioning from a less privileged predictor mode, regardless of
arch/x86/kvm/vmx/vmx.c void noinstr vmx_spec_ctrl_restore_host(struct vcpu_vmx *vmx, · line 7060
Quoted 10 lines
arch/x86/kvm/vmx/vmx.c +5
-0* whether the guest/host values differ. +* +* For eIBRS affected by Post Barrier RSB Predictions a serialising +* instruction (wrmsr) must be executed to ensure a call instruction has +* retired before the prediction of a following unbalanced ret. */ if (cpu_feature_enabled (X86_FEATURE_KERNEL_IBRS ) || +cpu_feature_enabled (X86_FEATURE_RSB_VMEXIT_LITE ) || vmx->spec_ctrl != hostval)native_wrmsrl (MSR_IA32_SPEC_CTRL , hostval);
Better, I think, would be to leave the condition alone and put an LFENCE on the 'else' path:
if (cpu_feature_enabled(X86_FEATURE_KERNEL_IBRS) ||
vmx->spec_ctrl != hostval)
native_wrmsrl(MSR_IA32_SPEC_CTRL, hostval);
else
rmb();When the guest and host have different IA32_SPEC_CTRL values, you get the serialization from the WRMSR. Otherwise, you get it from the cheaper LFENCE.
This is still more convoluted than having the mitigation in one place.
Quoted 114 lines · Jim Mattson
Quoted 93 lines · Suraj Jitindar Singh
tl;dr: The existing mitigation for eIBRS PBRSB predictions uses an INT3 to ensure a call instruction retires before a following unbalanced RET. Replace this with a WRMSR serialising instruction which has a lower performance penalty.
== Background ==
eIBRS (enhanced indirect branch restricted speculation) is used to prevent predictor addresses from one privilege domain from being used for prediction in a higher privilege domain.
== Problem ==
On processors with eIBRS protections there can be a case where upon VM exit a guest address may be used as an RSB prediction for an unbalanced RET if a CALL instruction hasn't yet been retired. This is termed PBRSB (Post-Barrier Return Stack Buffer).
A mitigation for this was introduced in:
(2b1299322016731d56807aa49254a5ea3080b6b3x86/speculation: Add RSB VM Exit protections)This mitigation [1] has a ~1% performance impact on VM exit compared to without it [2].
== Solution ==
The WRMSR instruction can be used as a speculation barrier and a serialising instruction. Use this on the VM exit path instead to ensure that a CALL instruction (in this case the call to
vmx_spec_ctrl_restore_host) has retired before the prediction of a following unbalanced RET.This mitigation [3] has a negligible performance impact.
== Testing ==
Run the
outl_to_kernelkvm-unit-tests test 200 times per configuration which counts the cycles for an exit to kernel mode.[1] With existing mitigation:
Average: 2026 cycles
[2] With no mitigation:
Average: 2008 cycles
[3] With proposed mitigation:
Average: 2008 cyclesSigned-off-by Suraj Jitindar Singh <[email protected]>--- arch/x86/include/asm/nospec-branch.h | 7 +++---- arch/x86/kvm/vmx/vmenter.S | 3 +-- arch/x86/kvm/vmx/vmx.c | 5 +++++ 3 files changed, 9 insertions(+), 6 deletions(-)arch/x86/include/asm/nospec-branch.h +3
-4@@ -159,10 +159,9 @@* A simpler FILL_RETURN_BUFFER macro. Don't make people use the CPP * monstrosity above, manually. */ -.macro FILL_RETURN_BUFFER reg:req nr:req ftr:req ftr2=ALT_NOT(X86_FEATURE_ALWAYS) -ALTERNATIVE_2 "jmp .Lskip_rsb_\@" , \ - __stringify(__FILL_RETURN_BUFFER(\reg,\nr)), \ftr, \ - __stringify(__FILL_ONE_RETURN), \ftr2 +.macro FILL_RETURN_BUFFER reg:req nr:req ftr:req + ALTERNATIVE "jmp .Lskip_rsb_\@", \ + __stringify(__FILL_RETURN_BUFFER(\reg,\nr)), \ftr .Lskip_rsb_\@: .endmarch/x86/kvm/vmx/vmenter.S +1
-2@@ -231,8 +231,7 @@ SYM_INNER_LABEL(vmx_vmexit, SYM_L_GLOBAL) * single call to retire, before the first unbalanced RET. */ - FILL_RETURN_BUFFER %_ASM_CX, RSB_CLEAR_LOOPS, X86_FEATURE_RSB_VMEXIT,\ - X86_FEATURE_RSB_VMEXIT_LITE + FILL_RETURN_BUFFER %_ASM_CX, RSB_CLEAR_LOOPS, X86_FEATURE_RSB_VMEXIT pop %_ASM_ARG2 /* @flags */arch/x86/kvm/vmx/vmx.c +0
-0@@ -7049,8 +7049,13 @@ void noinstr vmx_spec_ctrl_restore_host(struct vcpu_vmx *vmx,* For legacy IBRS, the IBRS bit always needs to be written after * transitioning from a less privileged predictor mode, regardless of arch/x86/kvm/vmx/vmx.c void noinstr vmx_spec_ctrl_restore_host(struct vcpu_vmx *vmx, · line 7060
Quoted 10 lines
arch/x86/kvm/vmx/vmx.c +5
-0* whether the guest/host values differ. +* +* For eIBRS affected by Post Barrier RSB Predictions a serialising +* instruction (wrmsr) must be executed to ensure a call instruction has +* retired before the prediction of a following unbalanced ret. */ if (cpu_feature_enabled (X86_FEATURE_KERNEL_IBRS ) || +cpu_feature_enabled (X86_FEATURE_RSB_VMEXIT_LITE ) || vmx->spec_ctrl != hostval)native_wrmsrl (MSR_IA32_SPEC_CTRL , hostval);Better, I think, would be to leave the condition alone and put an LFENCE on the 'else' path:
if (cpu_feature_enabled(X86_FEATURE_KERNEL_IBRS) || vmx->spec_ctrl != hostval) native_wrmsrl(MSR_IA32_SPEC_CTRL, hostval); else rmb();
Quoted 3 lines
When the guest and host have different
IA32_SPEC_CTRLvalues, you get the serialization from the WRMSR. Otherwise, you get it from the cheaper LFENCE.
In this case systems that don't suffer from PBRSB (i.e. don've have X86_FEATURE_RSB_VMEXIT_LITE set) would be doing a barrier for no reason. We're just trading performance on vulnerable systems for a performance hit on systems that aren't vulnerable.
Quoted 2 lines
This is still more convoluted than having the mitigation in one place.
Agreed.
Quoted 116 lines · Daniel Sneddon
Quoted 114 lines · Jim Mattson
Quoted 93 lines · Suraj Jitindar Singh
tl;dr: The existing mitigation for eIBRS PBRSB predictions uses an INT3 to ensure a call instruction retires before a following unbalanced RET. Replace this with a WRMSR serialising instruction which has a lower performance penalty.
== Background ==
eIBRS (enhanced indirect branch restricted speculation) is used to prevent predictor addresses from one privilege domain from being used for prediction in a higher privilege domain.
== Problem ==
On processors with eIBRS protections there can be a case where upon VM exit a guest address may be used as an RSB prediction for an unbalanced RET if a CALL instruction hasn't yet been retired. This is termed PBRSB (Post-Barrier Return Stack Buffer).
A mitigation for this was introduced in:
(2b1299322016731d56807aa49254a5ea3080b6b3x86/speculation: Add RSB VM Exit protections)This mitigation [1] has a ~1% performance impact on VM exit compared to without it [2].
== Solution ==
The WRMSR instruction can be used as a speculation barrier and a serialising instruction. Use this on the VM exit path instead to ensure that a CALL instruction (in this case the call to
vmx_spec_ctrl_restore_host) has retired before the prediction of a following unbalanced RET.This mitigation [3] has a negligible performance impact.
== Testing ==
Run the
outl_to_kernelkvm-unit-tests test 200 times per configuration which counts the cycles for an exit to kernel mode.[1] With existing mitigation:
Average: 2026 cycles
[2] With no mitigation:
Average: 2008 cycles
[3] With proposed mitigation:
Average: 2008 cyclesSigned-off-by Suraj Jitindar Singh <[email protected]>--- arch/x86/include/asm/nospec-branch.h | 7 +++---- arch/x86/kvm/vmx/vmenter.S | 3 +-- arch/x86/kvm/vmx/vmx.c | 5 +++++ 3 files changed, 9 insertions(+), 6 deletions(-)arch/x86/include/asm/nospec-branch.h +3
-4@@ -159,10 +159,9 @@* A simpler FILL_RETURN_BUFFER macro. Don't make people use the CPP * monstrosity above, manually. */ -.macro FILL_RETURN_BUFFER reg:req nr:req ftr:req ftr2=ALT_NOT(X86_FEATURE_ALWAYS) -ALTERNATIVE_2 "jmp .Lskip_rsb_\@" , \ - __stringify(__FILL_RETURN_BUFFER(\reg,\nr)), \ftr, \ - __stringify(__FILL_ONE_RETURN), \ftr2 +.macro FILL_RETURN_BUFFER reg:req nr:req ftr:req + ALTERNATIVE "jmp .Lskip_rsb_\@", \ + __stringify(__FILL_RETURN_BUFFER(\reg,\nr)), \ftr .Lskip_rsb_\@: .endmarch/x86/kvm/vmx/vmenter.S +1
-2@@ -231,8 +231,7 @@ SYM_INNER_LABEL(vmx_vmexit, SYM_L_GLOBAL) * single call to retire, before the first unbalanced RET. */ - FILL_RETURN_BUFFER %_ASM_CX, RSB_CLEAR_LOOPS, X86_FEATURE_RSB_VMEXIT,\ - X86_FEATURE_RSB_VMEXIT_LITE + FILL_RETURN_BUFFER %_ASM_CX, RSB_CLEAR_LOOPS, X86_FEATURE_RSB_VMEXIT pop %_ASM_ARG2 /* @flags */arch/x86/kvm/vmx/vmx.c +0
-0@@ -7049,8 +7049,13 @@ void noinstr vmx_spec_ctrl_restore_host(struct vcpu_vmx *vmx,* For legacy IBRS, the IBRS bit always needs to be written after * transitioning from a less privileged predictor mode, regardless of arch/x86/kvm/vmx/vmx.c void noinstr vmx_spec_ctrl_restore_host(struct vcpu_vmx *vmx, · line 7060
Quoted 10 lines
arch/x86/kvm/vmx/vmx.c +5
-0* whether the guest/host values differ. +* +* For eIBRS affected by Post Barrier RSB Predictions a serialising +* instruction (wrmsr) must be executed to ensure a call instruction has +* retired before the prediction of a following unbalanced ret. */ if (cpu_feature_enabled (X86_FEATURE_KERNEL_IBRS ) || +cpu_feature_enabled (X86_FEATURE_RSB_VMEXIT_LITE ) || vmx->spec_ctrl != hostval)native_wrmsrl (MSR_IA32_SPEC_CTRL , hostval);Better, I think, would be to leave the condition alone and put an LFENCE on the 'else' path:
if (cpu_feature_enabled(X86_FEATURE_KERNEL_IBRS) || vmx->spec_ctrl != hostval) native_wrmsrl(MSR_IA32_SPEC_CTRL, hostval); else rmb();
Quoted 7 lines
Quoted 3 lines
When the guest and host have different
IA32_SPEC_CTRLvalues, you get the serialization from the WRMSR. Otherwise, you get it from the cheaper LFENCE.In this case systems that don't suffer from PBRSB (i.e. don've have
X86_FEATURE_RSB_VMEXIT_LITEset) would be doing a barrier for no reason. We're just trading performance on vulnerable systems for a performance hit on systems that aren't vulnerable.
The lfence could be buried in an ALTERNATIVE keyed to X86_FEATURE_RSB_VMEXIT_LITE.
Quoted 2 lines
Quoted 1 line
This is still more convoluted than having the mitigation in one place.
Agreed.
For a mere 18 cycles, it doesn't really seem worth the obfuscation.
Quoted 2 lines · Suraj Jitindar Singh
tl;dr: The existing mitigation for eIBRS PBRSB predictions uses an INT3 to ensure a call instruction retires before a following unbalanced RET.
No it doesn't. The INT3 is transient. The existing mitigation uses an LFENCE.
Quoted 2 lines
Replace this with a WRMSR serialising instruction which has a lower performance penalty.
What is "this"? The INT3^W LFENCE?
WRMSR is not lower overhead than an LFENCE.
Quoted 6 lines
== Solution ==
The WRMSR instruction can be used as a speculation barrier and a serialising instruction. Use this on the VM exit path instead to ensure that a CALL instruction (in this case the call to
vmx_spec_ctrl_restore_host) has retired before the prediction of a following unbalanced RET.
While both of these sentences are true statements, you've missed the necessary safety property.
One CALL has to retire before *any* RET can execute.
There are several ways the frontend can end up eventually consuming the bad RSB entry; they all stem from an execute (not prediction) of the next RET instruction.
As to the change, ...
arch/x86/kvm/vmx/vmx.c void noinstr vmx_spec_ctrl_restore_host(struct vcpu_vmx *vmx,
Quoted 5 lines
arch/x86/kvm/vmx/vmx.c +0
-0@@ -7049,8 +7049,13 @@ void noinstr vmx_spec_ctrl_restore_host(struct vcpu_vmx *vmx,
... out of context above this hunk is:
if (!cpu_feature_enabled(X86_FEATURE_MSR_SPEC_CTRL)) return;
meaning that there is a return instruction which is programmatically reachable ahead of the WRMSR.
Whether it is speculatively reachable depends on whether the frontend can see through the _static_cpu_has(), as well as X86_FEATURE_MSR_SPEC_CTRL never becoming compile time evaluable.
There is also a second latent bug, to do with the code generation for this_cpu_read(x86_spec_ctrl_current).
It's worth saying that, in principle, this optimisation is safe, but pretty much all the discussion about it is wrong. Here is one I prepared earlier:
https://lore.kernel.org/xen-devel/[email protected]/
OTOH, below the hunk in question, there's a barrier_nospec() which is giving you the actual projection you need (subject to latent and/or code layout bugs), irrespective of the extra WRMSR.
~Andrew
Peter Zijlstra
maintainerQuoted 1 line · Jim Mattson
rmb();
Since we're not doing memory ordering but speculation hardening, this is completely the wrong way to spell LFENCE. Please use barrier_nospec() in those cases.
And as has already been pointed out, there's one of those just two lines down already.
Also, what Andrew said.
From: Jim Mattson
Quoted 10 lines
Sent: 05 October 2022 23:29
Quoted 5 lines · Suraj Jitindar Singh
tl;dr: The existing mitigation for eIBRS PBRSB predictions uses an INT3 to ensure a call instruction retires before a following unbalanced RET. Replace this with a WRMSR serialising instruction which has a lower performance penalty.
The INT3 is only on a speculative path and should not impact performance.
Doesn't that depend on how quickly the cpu can abort the decode and execution of the INT3 instruction?
INT3 is bound to generate a lot of uops and/or be microcoded.
Old cpu couldn't abort fpu instructions.
IIRC the Intel performance guide even suggested not interleaving code and data because the data might get speculatively executed and take a long time to abort.
I actually wonder whether 'JMPS .' (eb fe) shouldn't be used instead of INT3 (cc) because it is fast to decode and execute. But I'm no expect here.
David
-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)
[email protected]
reviewerQuoted 16 lines · David Laight
From: Jim Mattson
Quoted 10 lines
Sent: 05 October 2022 23:29
Quoted 5 lines · Suraj Jitindar Singh
tl;dr: The existing mitigation for eIBRS PBRSB predictions uses an INT3 to ensure a call instruction retires before a following unbalanced RET. Replace this with a WRMSR serialising instruction which has a lower performance penalty.
The INT3 is only on a speculative path and should not impact performance.
Doesn't that depend on how quickly the cpu can abort the decode and execution of the INT3 instruction?
INT3 is bound to generate a lot of uops and/or be microcoded.
Quoted 8 lines
Old cpu couldn't abort fpu instructions.
IIRC the Intel performance guide even suggested not interleaving code and data because the data might get speculatively executed and take a long time to abort.I actually wonder whether 'JMPS .' (eb fe) shouldn't be used instead of INT3 (cc) because it is fast to decode and execute. But I'm no expect here.
I have been told that INT3 is better in this case because 'JMP .' would waste CPU resources.
[email protected]
reviewerQuoted 30 lines · Andrew Cooper
Quoted 6 lines · Suraj Jitindar Singh
== Solution ==
The WRMSR instruction can be used as a speculation barrier and a serialising instruction. Use this on the VM exit path instead to ensure that a CALL instruction (in this case the call to
vmx_spec_ctrl_restore_host) has retired before the prediction of a following unbalanced RET.While both of these sentences are true statements, you've missed the necessary safety property.
One CALL has to retire before *any* RET can execute.
There are several ways the frontend can end up eventually consuming the bad RSB entry; they all stem from an execute (not prediction) of the next RET instruction.
As to the change, ...
arch/x86/kvm/vmx/vmx.c void noinstr vmx_spec_ctrl_restore_host(struct vcpu_vmx *vmx,
Quoted 5 lines
arch/x86/kvm/vmx/vmx.c +0
-0@@ -7049,8 +7049,13 @@ void noinstr vmx_spec_ctrl_restore_host(struct vcpu_vmx *vmx,... out of context above this hunk is:
if (!cpu_feature_enabled(X86_FEATURE_MSR_SPEC_CTRL)) return;
Quoted 6 lines
meaning that there is a return instruction which is programmatically reachable ahead of the WRMSR.
Whether it is speculatively reachable depends on whether the frontend can see through the
_static_cpu_has(), as well asX86_FEATURE_MSR_SPEC_CTRLnever becoming compile time evaluable.
In this case wouldn't _static_cpu_has() be runtime patched to a JMP (<+8> below) or a NOP? RET (at <+13>) should not be reachable even speculatively. What am I missing?
Dump of assembler code for function vmx_spec_ctrl_restore_host:
arch/x86/kvm/vmx/vmx.c:
u64 hostval = this_cpu_read(x86_spec_ctrl_current);
<+0>: mov %gs:0x7e022e60(%rip),%r8 # 0x1ad48 <x86_spec_ctrl_current> ./arch/x86/include/asm/cpufeature.h:
asm_volatile_goto( <+8>: jmp <vmx_spec_ctrl_restore_host+14> <+10>: nopl (%rax) <+13>: ret
arch/x86/kvm/vmx/vmx.c:
if (flags & VMX_RUN_SAVE_SPEC_CTRL)
<+14>: and $0x2,%esi <+17>: je <vmx_spec_ctrl_restore_host+40>
[...]
Pawan Gupta
reviewerHi Suraj,
Quoted 35 lines · Suraj Jitindar Singh
tl;dr: The existing mitigation for eIBRS PBRSB predictions uses an INT3 to ensure a call instruction retires before a following unbalanced RET. Replace this with a WRMSR serialising instruction which has a lower performance penalty.
== Background ==
eIBRS (enhanced indirect branch restricted speculation) is used to prevent predictor addresses from one privilege domain from being used for prediction in a higher privilege domain.
== Problem ==
On processors with eIBRS protections there can be a case where upon VM exit a guest address may be used as an RSB prediction for an unbalanced RET if a CALL instruction hasn't yet been retired. This is termed PBRSB (Post-Barrier Return Stack Buffer).
A mitigation for this was introduced in:
(2b1299322016731d56807aa49254a5ea3080b6b3x86/speculation: Add RSB VM Exit protections)This mitigation [1] has a ~1% performance impact on VM exit compared to without it [2].
== Solution ==
The WRMSR instruction can be used as a speculation barrier and a serialising instruction. Use this on the VM exit path instead to ensure that a CALL instruction (in this case the call to
vmx_spec_ctrl_restore_host) has retired before the prediction of a following unbalanced RET.This mitigation [3] has a negligible performance impact.
== Testing ==
Quoted 7 lines
Run the
outl_to_kernelkvm-unit-tests test 200 times per configuration which counts the cycles for an exit to kernel mode.[1] With existing mitigation:
Average: 2026 cycles
[2] With no mitigation:
Average: 2008 cycles
During these tests was the value of MSR SPEC_CTRL for host and guest different?
Quoted 2 lines
[3] With proposed mitigation:
Average: 2008 cycles