From: Gavin Shan <gshan@redhat.com>
To: Marc Zyngier <maz@kernel.org>, Peter Xu <peterx@redhat.com>
Cc: kvmarm@lists.cs.columbia.edu, kvm@vger.kernel.org,
linux-doc@vger.kernel.org, catalin.marinas@arm.com,
linux-kselftest@vger.kernel.org, bgardon@google.com,
shuah@kernel.org, corbet@lwn.net, drjones@redhat.com,
will@kernel.org, zhenyzha@redhat.com, dmatlack@google.com,
linux-arm-kernel@lists.infradead.org,
linux-kernel@vger.kernel.org, pbonzini@redhat.com,
oliver.upton@linux.dev, shan.gavin@gmail.com
Subject: Re: [PATCH v2 1/5] KVM: x86: Introduce KVM_REQ_RING_SOFT_FULL
Date: Mon, 19 Sep 2022 09:58:10 +1000 [thread overview]
Message-ID: <a2e0b9bc-2c67-8683-d722-7298bd65058c@redhat.com> (raw)
In-Reply-To: <87illlkqfu.wl-maz@kernel.org>
On 9/18/22 7:00 PM, Marc Zyngier wrote:
> On Fri, 16 Sep 2022 19:09:52 +0100,
> Peter Xu <peterx@redhat.com> wrote:
>>
>> On Fri, Sep 16, 2022 at 12:51:31PM +0800, Gavin Shan wrote:
>>> This adds KVM_REQ_RING_SOFT_FULL, which is raised when the dirty
>>> ring of the specific VCPU becomes softly full in kvm_dirty_ring_push().
>>> The VCPU is enforced to exit when the request is raised and its
>>> dirty ring is softly full on its entrance.
>>>
>>> Suggested-by: Marc Zyngier <maz@kernel.org>
>>> Signed-off-by: Gavin Shan <gshan@redhat.com>
>>> ---
>>> arch/x86/kvm/x86.c | 5 +++--
>>> include/linux/kvm_host.h | 1 +
>>> virt/kvm/dirty_ring.c | 4 ++++
>>> 3 files changed, 8 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
>>> index 43a6a7efc6ec..7f368f59f033 100644
>>> --- a/arch/x86/kvm/x86.c
>>> +++ b/arch/x86/kvm/x86.c
>>> @@ -10265,8 +10265,9 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
>>> bool req_immediate_exit = false;
>>>
>>> /* Forbid vmenter if vcpu dirty ring is soft-full */
>>> - if (unlikely(vcpu->kvm->dirty_ring_size &&
>>> - kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
>>> + if (kvm_check_request(KVM_REQ_RING_SOFT_FULL, vcpu) &&
>>> + kvm_dirty_ring_soft_full(&vcpu->dirty_ring)) {
>>> + kvm_make_request(KVM_REQ_RING_SOFT_FULL, vcpu);
>>> vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
>>> trace_kvm_dirty_ring_exit(vcpu);
>>> r = 0;
>>
>> As commented previously - can we use kvm_test_request() instead? because we
>> don't want to unconditionally clear the bit. Instead of making the request
>> again, we can clear request only if !full.
>
> I have the feeling that this is a micro-optimisation that won't lead
> to much benefit in practice. You already have the cache line, just not
> in exclusive mode, and given that this is per-vcpu, you'd only see the
> cost if someone else is setting a request to this vcpu while
> evaluating the local requests.
>
> And now you need extra barriers...
>
> Also, can we please refrain from changing things without data showing
> that this actually is worse than what we had before? The point below
> makes me think that this is actually beneficial as is.
>
I think Marc's explanation makes sense. It won't make difference in terms
of performance. We need to explicitly handle barrier when kvm_test_request()
is used. So I prefer to keep the code if Peter agrees.
>> We can also safely move this into the block of below kvm_request_pending()
>> as Marc used to suggest.
>
> This, on the other hand, makes sure that we share the cost across all
> requests. Requests should be extremely rare anyway (and if they
> aren't, you have a whole lot of performance issues on your hands
> anyway).
>
Yeah, We shouldn't have too much requests. I missed the comment from Marc
to move this chunk to kvm_request_pending(). I will fix it in v3.
>>
>> To explicitly use kvm_clear_request(), we may need to be careful on the
>> memory barriers. I'm wondering whether we should have moved
>> smp_mb__after_atomic() into kvm_clear_request() because kvm_clear_request()
>> is used outside kvm_check_request() and IIUC all the call sites should
>> better have that barrier too to be safe.
>>
>> Side note: when I read the code around I also see some mis-use of clear
>> request where it can be omitted, e.g.:
>>
>> if (kvm_check_request(KVM_REQ_UNHALT, vcpu)) {
>> kvm_clear_request(KVM_REQ_UNHALT, vcpu);
>> vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
>> }
>>
>> Maybe it's a sign of bad naming, so we should renamed kvm_check_request()
>> to kvm_test_clear_request() too to show that clearing after that is not
>> needed?
>
> Yeah, this kvm_clear_request() is superfluous. But this is rather well
> documented, for once, and I don't think we should repaint it based on
> a sample of one.
>
Yeah, I think Peter is correct that smp_mb__after_atomic() would be
part of kvm_clear_request(). Otherwise, the following two cases aren't
in same order:
// kvm_check_request() // test and clear
kvm_test_request() kvm_test_request()
kvm_clear_request() kvm_clear_request()
smp_mb__after_atomic()
Thanks,
Gavin
next prev parent reply other threads:[~2022-09-18 23:58 UTC|newest]
Thread overview: 13+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-09-16 4:51 [PATCH v2 0/5] KVM: arm64: Enable ring-based dirty memory tracking Gavin Shan
2022-09-16 4:51 ` [PATCH v2 1/5] KVM: x86: Introduce KVM_REQ_RING_SOFT_FULL Gavin Shan
2022-09-16 18:09 ` Peter Xu
2022-09-18 9:00 ` Marc Zyngier
2022-09-18 23:58 ` Gavin Shan [this message]
2022-09-19 9:22 ` Marc Zyngier
2022-09-19 16:17 ` Peter Xu
2022-09-16 4:51 ` [PATCH v2 2/5] KVM: arm64: Enable ring-based dirty memory tracking Gavin Shan
2022-09-17 15:10 ` kernel test robot
2022-09-19 1:27 ` Gavin Shan
2022-09-16 4:51 ` [PATCH v2 3/5] KVM: selftests: Use host page size to map ring buffer in dirty_log_test Gavin Shan
2022-09-16 4:51 ` [PATCH v2 4/5] KVM: selftests: Clear dirty ring states between two modes " Gavin Shan
2022-09-16 4:51 ` [PATCH v2 5/5] KVM: selftests: Automate choosing dirty ring size " Gavin Shan
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=a2e0b9bc-2c67-8683-d722-7298bd65058c@redhat.com \
--to=gshan@redhat.com \
--cc=bgardon@google.com \
--cc=catalin.marinas@arm.com \
--cc=corbet@lwn.net \
--cc=dmatlack@google.com \
--cc=drjones@redhat.com \
--cc=kvm@vger.kernel.org \
--cc=kvmarm@lists.cs.columbia.edu \
--cc=linux-arm-kernel@lists.infradead.org \
--cc=linux-doc@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-kselftest@vger.kernel.org \
--cc=maz@kernel.org \
--cc=oliver.upton@linux.dev \
--cc=pbonzini@redhat.com \
--cc=peterx@redhat.com \
--cc=shan.gavin@gmail.com \
--cc=shuah@kernel.org \
--cc=will@kernel.org \
--cc=zhenyzha@redhat.com \
/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
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
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).