[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <4bf20a74-4534-9074-688f-a8b55708d936@amd.com>
Date: Sun, 4 Oct 2020 06:27:03 +0700
From: Suravee Suthikulpanit <suravee.suthikulpanit@....com>
To: Paolo Bonzini <pbonzini@...hat.com>, linux-kernel@...r.kernel.org,
kvm@...r.kernel.org
Cc: joro@...tes.org
Subject: Re: [PATCH] KVM: SVM: Initialize ir_list and ir_list_lock regardless
of AVIC enablement
Paolo,
On 9/28/20 3:01 PM, Paolo Bonzini wrote:
> On 28/09/20 07:53, Suravee Suthikulpanit wrote:
>> Hi,
>>
>> Are there any issues or concerns about this patch?
>
> Yes, sorry I haven't replied yet. Looks like Linus is doing an -rc8 so
> there's plenty of time to have it in 5.9.
>
> The thing I'm wondering is, why is svm_update_pi_irte doing anything if
> you don't have AVIC enabled? In other word, this might not be the root
> cause of the bug. You always get to the "else" branch of the loop of
> course, and I'm not sure how irq_set_vcpu_affinity returns something
> with pi.prev_ga_tag set.
You are right. pi_prev_ga_tag needs to be initialized before used
(in case AVIC is not enabled). I have already sent out another patch
to properly fix the issue instead with subject
(KVM: SVM: Initialize prev_ga_tag before use).
Thanks,
Suravee
Powered by blists - more mailing lists