linux-kernel - Re: [PATCH v5 1/9] KVM: VMX: Use on-stack copy of @flags in __vmx_vcpu

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <aR0PXEyP_OKuiQOO@google.com>
Date: Tue, 18 Nov 2025 16:29:16 -0800
From: Sean Christopherson <seanjc@...gle.com>
To: Uros Bizjak <ubizjak@...il.com>
Cc: Paolo Bonzini <pbonzini@...hat.com>, Thomas Gleixner <tglx@...utronix.de>, 
	Borislav Petkov <bp@...en8.de>, Peter Zijlstra <peterz@...radead.org>, 
	Josh Poimboeuf <jpoimboe@...nel.org>, kvm@...r.kernel.org, linux-kernel@...r.kernel.org, 
	Pawan Gupta <pawan.kumar.gupta@...ux.intel.com>, Brendan Jackman <jackmanb@...gle.com>
Subject: Re: [PATCH v5 1/9] KVM: VMX: Use on-stack copy of @flags in __vmx_vcpu_run()

On Fri, Nov 14, 2025, Uros Bizjak wrote:
> On 11/14/25 00:37, Sean Christopherson wrote:
> > diff --git a/arch/x86/kvm/vmx/vmenter.S b/arch/x86/kvm/vmx/vmenter.S
> > index 574159a84ee9..93cf2ca7919a 100644
> > --- a/arch/x86/kvm/vmx/vmenter.S
> > +++ b/arch/x86/kvm/vmx/vmenter.S
> > @@ -92,7 +92,7 @@ SYM_FUNC_START(__vmx_vcpu_run)
> >   	/* Save @vmx for SPEC_CTRL handling */
> >   	push %_ASM_ARG1
> > -	/* Save @flags for SPEC_CTRL handling */
> > +	/* Save @flags (used for VMLAUNCH vs. VMRESUME and mitigations). */
> >   	push %_ASM_ARG3
> >   	/*
> > @@ -101,9 +101,6 @@ SYM_FUNC_START(__vmx_vcpu_run)
> >   	 */
> >   	push %_ASM_ARG2
> > -	/* Copy @flags to EBX, _ASM_ARG3 is volatile. */
> > -	mov %_ASM_ARG3L, %ebx
> > -
> >   	lea (%_ASM_SP), %_ASM_ARG2
> >   	call vmx_update_host_rsp
> > @@ -147,9 +144,6 @@ SYM_FUNC_START(__vmx_vcpu_run)
> >   	/* Load @regs to RAX. */
> >   	mov (%_ASM_SP), %_ASM_AX
> > -	/* Check if vmlaunch or vmresume is needed */
> > -	bt   $VMX_RUN_VMRESUME_SHIFT, %ebx
> > -
> >   	/* Load guest registers.  Don't clobber flags. */
> >   	mov VCPU_RCX(%_ASM_AX), %_ASM_CX
> >   	mov VCPU_RDX(%_ASM_AX), %_ASM_DX
> > @@ -173,8 +167,9 @@ SYM_FUNC_START(__vmx_vcpu_run)
> >   	/* Clobbers EFLAGS.ZF */
> >   	CLEAR_CPU_BUFFERS
> > -	/* Check EFLAGS.CF from the VMX_RUN_VMRESUME bit test above. */
> > -	jnc .Lvmlaunch
> > +	/* Check @flags to see if vmlaunch or vmresume is needed. */
> > +	testl $VMX_RUN_VMRESUME, WORD_SIZE(%_ASM_SP)
> > +	jz .Lvmlaunch
> 
> 
> You could use TESTB instead of TESTL in the above code to save 3 bytes
> of code and some memory bandwidth.
> 
> Assembler will report unwanted truncation if VMX_RUN_VRESUME ever
> becomes larger than 255.

Unfortunately, the warning with gcc isn't escalated to an error with -Werror,
e.g. with KVM_WERROR=y.  And AFAICT clang's assembler doesn't warn at all and
happily generates garbage.  E.g. with VMX_RUN_VMRESUME relocated to bit 10, clang
generates this without a warning:

 33c:   f6 44 24 08 00          testb  $0x0,0x8(%rsp)
 341:   74 08                   je     34b <__vmx_vcpu_run+0x9b>
 343:   0f 01 c3                vmresume 

versus the expected:

 33c:   f7 44 24 08 00 04 00    testl  $0x400,0x8(%rsp)
 343:   00 
 344:   74 08                   je     34e <__vmx_vcpu_run+0x9e>
 346:   0f 01 c3                vmresume 

So for now at least, I'll stick with testl.