[PATCH] KVM/VMX: Avoid stack engine synchronization uop in __vmx_vcpu_run

[PATCH] KVM/VMX: Avoid stack engine synchronization uop in __vmx_vcpu_run

[Uros Bizjak]

Avoid instructions with explicit uses of the stack pointer between
instructions that implicitly refer to it. The sequence of
POP %reg; ADD $x, %RSP; POP %reg forces emission of synchronization
uop to synchronize the value of the stack pointer in the stack engine
and the out-of-order core.

Using POP with the dummy register instead of ADD $x, %RSP results in a
smaller code size and faster code.

The patch also fixes the reference to the wrong register in the
nearby comment.

Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Sean Christopherson <seanjc@google.com>
Signed-off-by: Uros Bizjak <ubizjak@gmail.com>
---
 arch/x86/kvm/vmx/vmenter.S | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)

diff --git a/arch/x86/kvm/vmx/vmenter.S b/arch/x86/kvm/vmx/vmenter.S
index 6de96b943804..afcb237e1c17 100644
--- a/arch/x86/kvm/vmx/vmenter.S
+++ b/arch/x86/kvm/vmx/vmenter.S
@@ -189,13 +189,16 @@ SYM_INNER_LABEL(vmx_vmexit, SYM_L_GLOBAL)
 	xor %ebx, %ebx
 
 .Lclear_regs:
+	/* "POP" @regs. */
+	pop %_ASM_AX
+
 	/*
 	 * Clear all general purpose registers except RSP and RBX to prevent
 	 * speculative use of the guest's values, even those that are reloaded
 	 * via the stack.  In theory, an L1 cache miss when restoring registers
 	 * could lead to speculative execution with the guest's values.
 	 * Zeroing XORs are dirt cheap, i.e. the extra paranoia is essentially
-	 * free.  RSP and RAX are exempt as RSP is restored by hardware during
+	 * free.  RSP and RBX are exempt as RSP is restored by hardware during
 	 * VM-Exit and RBX is explicitly loaded with 0 or 1 to hold the return
 	 * value.
 	 */
@@ -216,9 +219,6 @@ SYM_INNER_LABEL(vmx_vmexit, SYM_L_GLOBAL)
 	xor %r15d, %r15d
 #endif
 
-	/* "POP" @regs. */
-	add $WORD_SIZE, %_ASM_SP
-
 	/*
 	 * IMPORTANT: RSB filling and SPEC_CTRL handling must be done before
 	 * the first unbalanced RET after vmexit!
@@ -234,7 +234,6 @@ SYM_INNER_LABEL(vmx_vmexit, SYM_L_GLOBAL)
 	FILL_RETURN_BUFFER %_ASM_CX, RSB_CLEAR_LOOPS, X86_FEATURE_RSB_VMEXIT,\
 			   X86_FEATURE_RSB_VMEXIT_LITE
 
-
 	pop %_ASM_ARG2	/* @flags */
 	pop %_ASM_ARG1	/* @vmx */
 
-- 
2.37.1

Re: [PATCH] KVM/VMX: Avoid stack engine synchronization uop in __vmx_vcpu_run

[Sean Christopherson]

On Tue, Aug 16, 2022, Uros Bizjak wrote:
> Avoid instructions with explicit uses of the stack pointer between
> instructions that implicitly refer to it. The sequence of
> POP %reg; ADD $x, %RSP; POP %reg forces emission of synchronization
> uop to synchronize the value of the stack pointer in the stack engine
> and the out-of-order core.
> 
> Using POP with the dummy register instead of ADD $x, %RSP results in a
> smaller code size and faster code.
> 
> The patch also fixes the reference to the wrong register in the
> nearby comment.
> 
> Cc: Paolo Bonzini <pbonzini@redhat.com>
> Cc: Sean Christopherson <seanjc@google.com>
> Signed-off-by: Uros Bizjak <ubizjak@gmail.com>
> ---

Pushed to branch `for_paolo/6.1` at:

    https://github.com/sean-jc/linux.git

Unless you hear otherwise, it will make its way to kvm/queue "soon".

Note, the commit IDs are not guaranteed to be stable.