Re: [PATCH v3 15/27] KVM: x86: hyper-v: Allow purging all TLB flush FIFOs

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Low] x86/kvm: redundant guest TLB flush due to uncleared KVM_REQ_HV_TLB_FLUSH
--

commit fd7f6324f98d1c559f015e78c148b1f3ca58d78f
Author: Yosry Ahmed <yosry@kernel.org>

KVM: x86: hyper-v: Allow purging all TLB flush FIFOs

This commit refactors kvm_hv_purge_tlb_flush_fifo() by introducing an
inner helper __kvm_hv_purge_tlb_flush_fifo() to allow purging a specific
FIFO or all FIFOs. The refactoring delegates the KVM_REQ_HV_TLB_FLUSH
request check to the inner helper, which is now skipped if no FIFO is found.

> diff --git a/arch/x86/kvm/hyperv.h b/arch/x86/kvm/hyperv.h
> index 03cc65f8cc52d..a23ef05c1075c 100644
> --- a/arch/x86/kvm/hyperv.h
> +++ b/arch/x86/kvm/hyperv.h
[ ... ]
> @@ -312,18 +312,29 @@ static inline struct kvm_vcpu_hv_tlb_flush_fifo *kvm_hv_get_tlb_flush_fifo(struc
>  	return &hv_vcpu->tlb_flush_fifo[i];
>  }
>  
> -static inline void kvm_hv_purge_tlb_flush_fifo(struct kvm_vcpu *vcpu)
> +/* Purge pending TLB flushes in @fifo, or in all FIFOs if @fifo is NULL */
> +static inline void __kvm_hv_purge_tlb_flush_fifo(struct kvm_vcpu *vcpu,
> +						 struct kvm_vcpu_hv_tlb_flush_fifo *fifo)
>  {
[ ... ]
> +static inline void kvm_hv_purge_tlb_flush_fifo(struct kvm_vcpu *vcpu)
> +{
> +	struct kvm_vcpu_hv_tlb_flush_fifo *fifo;
>  
> -	kfifo_reset_out(&tlb_flush_fifo->entries);
> +	fifo = kvm_hv_get_tlb_flush_fifo(vcpu, is_guest_mode(vcpu));
> +	if (fifo)
> +		__kvm_hv_purge_tlb_flush_fifo(vcpu, fifo);
>  }

[Severity: Low]
Does this change inadvertently leave the KVM_REQ_HV_TLB_FLUSH bit uncleared
when fifo is NULL?

Prior to this patch, kvm_hv_purge_tlb_flush_fifo() unconditionally evaluated
and cleared kvm_check_request(KVM_REQ_HV_TLB_FLUSH, vcpu) before returning
early if the vCPU lacked an initialized Hyper-V state.

With this update, if a Hyper-V VM-wide TLB flush sets the request bit on a
vCPU without initialized Hyper-V state, kvm_hv_get_tlb_flush_fifo() returns
NULL. The helper __kvm_hv_purge_tlb_flush_fifo() is skipped entirely,
meaning the request bit remains uncleared.

If a standard guest TLB flush request arrives and triggers the purge path
in kvm_service_local_tlb_flush_requests(), the uncleared Hyper-V request bit
will still be pending when execution reaches vcpu_enter_guest():

arch/x86/kvm/x86.c:vcpu_enter_guest() {
        ...
        if (kvm_check_request(KVM_REQ_HV_TLB_FLUSH, vcpu) &&
            kvm_hv_vcpu_flush_tlb(vcpu))
                kvm_vcpu_flush_tlb_guest(vcpu);
        ...
}

Since kvm_hv_vcpu_flush_tlb() fails (returns -EINVAL) for a vCPU without
Hyper-V state, would this fallback path execute a second, redundant
kvm_vcpu_flush_tlb_guest() operation?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261012233638.2546339-1-yosry@kernel.org?part=15