From: Sean Christopherson <firstname.lastname@example.org> To: David Matlack <email@example.com> Cc: Venkatesh Srinivas <firstname.lastname@example.org>, kvm list <email@example.com>, Paolo Bonzini <firstname.lastname@example.org> Subject: Re: [PATCH] kvm: Cap halt polling at kvm->max_halt_poll_ns Date: Thu, 6 May 2021 17:41:34 +0000 [thread overview] Message-ID: <YJQqTuduLxwTDOSa@google.com> (raw) In-Reply-To: <CALzav=e+6mtPUHTHFLbw1Q=1kgstPbAp=2mg9mi+_fc+iWELGA@mail.gmail.com> On Thu, May 06, 2021, David Matlack wrote: > On Thu, May 6, 2021 at 9:13 AM Sean Christopherson <email@example.com> wrote: > > > > Prefer capitalizing KVM in the shortlog, if only because I'm lazy with grep :-) > > > > On Thu, May 06, 2021, Venkatesh Srinivas wrote: > > > From: David Matlack <firstname.lastname@example.org> > > > > > > When growing halt-polling, there is no check that the poll time exceeds > > > the per-VM limit. It's possible for vcpu->halt_poll_ns to grow past > > > kvm->max_halt_poll_ns and stay there until a halt which takes longer > > > than kvm->halt_poll_ns. > > > > > > > Fixes: acd05785e48c ("kvm: add capability for halt polling") > > > > and probably Cc: stable@ too. > > > > > > > Signed-off-by: David Matlack <email@example.com> > > > Signed-off-by: Venkatesh Srinivas <firstname.lastname@example.org> > > > --- > > > virt/kvm/kvm_main.c | 4 ++-- > > > 1 file changed, 2 insertions(+), 2 deletions(-) > > > > > > diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c > > > index 2799c6660cce..120817c5f271 100644 > > > --- a/virt/kvm/kvm_main.c > > > +++ b/virt/kvm/kvm_main.c > > > @@ -2893,8 +2893,8 @@ static void grow_halt_poll_ns(struct kvm_vcpu *vcpu) > > > if (val < grow_start) > > > val = grow_start; > > > > > > - if (val > halt_poll_ns) > > > - val = halt_poll_ns; > > > + if (val > vcpu->kvm->max_halt_poll_ns) > > > + val = vcpu->kvm->max_halt_poll_ns; > > > > Hmm, I would argue that the introduction of the capability broke halt_poll_ns. > > The halt_poll_ns module param is writable after KVM is loaded. Prior to the > > capability, that meant the admin could adjust the param on the fly and all vCPUs > > would honor the new value as it was changed. > > > > By snapshotting the module param at VM creation, those semantics were lost. > > That's not necessarily wrong/bad, but I don't see anything in the changelog for > > the capability that suggests killing the old behavior was intentional/desirable. > > api.rst does say the capability overrides halt_poll_ns. Ya, I'm more concerned about old userspace that isn't aware of the capability, e.g. an old userspace that relies on tuning halt_poll_ns while a VM is running would break when upgrading to a version of KVM that supports the capability. > see value in changing the semantics to something like: > > - halt_poll_ns sets machine-wide maximum halt poll time. > - kvm->max_halt_poll_ns sets VM-wide maximum halt poll time. > - A vCPU will poll for at most min(halt_poll_ns, > kvm->max_halt_poll_ns) (aside from an in-progress poll when either > parameter is changed). Agreed. That would also provide a good opportunity to clean up the benign races in kvm_vcpu_block(); since both the module param and the per-VM variable can be modified at will, they really should only be read once per instance of kvm_vcpu_block(). > On a related note, the capability and these subtle details should be > documented in Documentation/virtual/kvm/halt-polling.txt. > > > > > > > > > > vcpu->halt_poll_ns = val; > > > out: > > > -- > > > 188.8.131.527.g51e8a6a459-goog > > >
prev parent reply other threads:[~2021-05-06 17:41 UTC|newest] Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top 2021-05-06 15:24 Venkatesh Srinivas 2021-05-06 16:13 ` Sean Christopherson 2021-05-06 16:30 ` David Matlack 2021-05-06 17:41 ` Sean Christopherson [this message]
Reply instructions: You may reply publicly to this message via plain-text email using any one of the following methods: * Save the following mbox file, import it into your mail client, and reply-to-all from there: mbox Avoid top-posting and favor interleaved quoting: https://en.wikipedia.org/wiki/Posting_style#Interleaved_style * Reply using the --to, --cc, and --in-reply-to switches of git-send-email(1): git send-email \ --in-reply-to=YJQqTuduLxwTDOSa@google.com \ --email@example.com \ --firstname.lastname@example.org \ --email@example.com \ --firstname.lastname@example.org \ --email@example.com \ --subject='Re: [PATCH] kvm: Cap halt polling at kvm->max_halt_poll_ns' \ /path/to/YOUR_REPLY https://kernel.org/pub/software/scm/git/docs/git-send-email.html * If your mail client supports setting the In-Reply-To header via mailto: links, try the mailto: link
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for NNTP newsgroup(s).