Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1575908 > unrolled thread

Re: [PATCH v2 0/6] KVM: x86: cleanup and speedup for APICv

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  Re: [PATCH v2 0/6] KVM: x86: cleanup and speedup for APICv Paolo Bonzini <pbonzini@redhat.com> - 2017-02-07 18:30 +0100
    Re: [PATCH v2 0/6] KVM: x86: cleanup and speedup for APICv Radim Krčmář <rkrcmar@redhat.com> - 2017-02-07 23:10 +0100
      Re: [PATCH v2 0/6] KVM: x86: cleanup and speedup for APICv Paolo Bonzini <pbonzini@redhat.com> - 2017-02-08 11:40 +0100
        Re: [PATCH v2 0/6] KVM: x86: cleanup and speedup for APICv Radim Krčmář <rkrcmar@redhat.com> - 2017-02-08 14:40 +0100
          Re: [PATCH v2 0/6] KVM: x86: cleanup and speedup for APICv Paolo Bonzini <pbonzini@redhat.com> - 2017-02-08 16:10 +0100

#1575908 — Re: [PATCH v2 0/6] KVM: x86: cleanup and speedup for APICv

On 19/12/2016 17:17, Paolo Bonzini wrote:
> These are the fixed patches that now pass vmx.flat.  The issue in
> "KVM: x86: do not scan IRR twice on APICv vmentry" was that,
> in v1 of the patches, kvm_cpu_get_interrupt did not scan IRR
> at all if it found PIR.ON=0.  This is now fixed in patches 4 and 5.
> 
> Another latent bug has to be fixed in patch 2.  The patch also
> highlights the maze that check_nested_events has become.  Fixing
> that is left for later.
> 
> The patches are on top of the (non-APICv-related) KVM_REQ_EVENT
> patches from this morning.  According to kvm-unit-tests, patch 6 speeds
> up self-IPIs (if not done through the accelerated self-IPI MSR) by 5-10%.
> 
> Paolo

Ping?

Paolo

[toc] | [next] | [standalone]

#1576105

2017-02-07 18:23+0100, Paolo Bonzini:
> On 19/12/2016 17:17, Paolo Bonzini wrote:
>> These are the fixed patches that now pass vmx.flat.  The issue in
>> "KVM: x86: do not scan IRR twice on APICv vmentry" was that,
>> in v1 of the patches, kvm_cpu_get_interrupt did not scan IRR
>> at all if it found PIR.ON=0.  This is now fixed in patches 4 and 5.
>> 
>> Another latent bug has to be fixed in patch 2.  The patch also
>> highlights the maze that check_nested_events has become.  Fixing
>> that is left for later.
>> 
>> The patches are on top of the (non-APICv-related) KVM_REQ_EVENT
>> patches from this morning.  According to kvm-unit-tests, patch 6 speeds
>> up self-IPIs (if not done through the accelerated self-IPI MSR) by 5-10%.
>> 
>> Paolo
> 
> Ping?

I think the patches are ready to be applied -- there might be some rough
edges with nested, but it was broken even before, and they fix at least
one known bug.

[toc] | [prev] | [next] | [standalone]

#1576423

On 07/02/2017 22:52, Radim Krčmář wrote:
> there might be some rough
> edges with nested, but it was broken even before, and they fix at least
> one known bug.

I don't think so, nested IRQ injection is tested very well.  Why do you
say it was broken even before?

Paolo

[toc] | [prev] | [next] | [standalone]

#1576585

2017-02-08 11:04+0100, Paolo Bonzini:
> On 07/02/2017 22:52, Radim Krčmář wrote:
>> there might be some rough
>> edges with nested, but it was broken even before, and they fix at least
>> one known bug.
> 
> I don't think so, nested IRQ injection is tested very well.  Why do you
> say it was broken even before?

I was basing on recent bug report where we missed IOMMU injection in a
non-nested case [1].  While we do cover the non-nested case with this
series, I think we still miss posted interrupts from IOMMU that should
trigger a nested VM exit from L2 to L1 -- details in review of [6/6].

1: http://www.spinics.net/lists/kvm/msg144355.html

[toc] | [prev] | [next] | [standalone]

#1576640

On 08/02/2017 14:33, Radim Krčmář wrote:
> 2017-02-08 11:04+0100, Paolo Bonzini:
>> On 07/02/2017 22:52, Radim Krčmář wrote:
>>> there might be some rough
>>> edges with nested, but it was broken even before, and they fix at least
>>> one known bug.
>>
>> I don't think so, nested IRQ injection is tested very well.  Why do you
>> say it was broken even before?
> 
> I was basing on recent bug report where we missed IOMMU injection in a
> non-nested case [1].  While we do cover the non-nested case with this
> series, I think we still miss posted interrupts from IOMMU that should
> trigger a nested VM exit from L2 to L1 -- details in review of [6/6].
> 
> 1: http://www.spinics.net/lists/kvm/msg144355.html

Uh, nice.  Totally missed that.

Paolo

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web