linux-kernel - Re: [PATCH v2] kvm: Replace vcpu->swait with rcuwait

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 for Android: free password hash cracker in your pocket

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

Date:   Tue, 21 Apr 2020 11:43:39 +0200
From:   Paolo Bonzini <pbonzini@...hat.com>
To:     Davidlohr Bueso <dave@...olabs.net>
Cc:     Marc Zyngier <maz@...nel.org>, tglx@...utronix.de,
        kvm@...r.kernel.org, Davidlohr Bueso <dbueso@...e.de>,
        peterz@...radead.org, torvalds@...ux-foundation.org,
        bigeasy@...utronix.de, linux-kernel@...r.kernel.org,
        rostedt@...dmis.org, linux-mips@...r.kernel.org,
        Paul Mackerras <paulus@...abs.org>, joel@...lfernandes.org,
        will@...nel.org, kvmarm@...ts.cs.columbia.edu
Subject: Re: [PATCH v2] kvm: Replace vcpu->swait with rcuwait

On 20/04/20 23:50, Davidlohr Bueso wrote:
> On Mon, 20 Apr 2020, Paolo Bonzini wrote:
> 
>> On 20/04/20 22:56, Davidlohr Bueso wrote:
>>> On Mon, 20 Apr 2020, Marc Zyngier wrote:
>>>
>>>> This looks like a change in the semantics of the tracepoint. Before
>>>> this
>>>> change, 'waited' would have been true if the vcpu waited at all. Here,
>>>> you'd
>>>> have false if it has been interrupted by a signal, even if the vcpu
>>>> has waited
>>>> for a period of time.
>>>
>>> Hmm but sleeps are now uninterruptible as we're using TASK_IDLE.
>>
>> Hold on, does that mean that you can't anymore send a signal in order to
>> kick a thread out of KVM_RUN?  Or am I just misunderstanding?
> 
> Considering that the return value of the interruptible wait is not
> checked, I would not think this breaks KVM_RUN.

What return value?  kvm_vcpu_check_block checks signal_pending, so you
could have a case where the signal is injected but you're not woken up.

Admittedly I am not familiar with how TASK_* work under the hood, but it
does seem to be like that.

Paolo