From: Like Xu <like.xu.linux@gmail.com>
To: Sean Christopherson <seanjc@google.com>
Cc: Oliver Upton <oliver.upton@linux.dev>,
Paolo Bonzini <pbonzini@redhat.com>,
kvm@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: [PATCH v2] KVM: x86/tsc: Use calculated tsc_offset before matching the fist vcpu's tsc
Date: Mon, 24 Jul 2023 15:35:16 +0800 [thread overview]
Message-ID: <20230724073516.45394-1-likexu@tencent.com> (raw)
From: Like Xu <likexu@tencent.com>
Avoid using kvm->arch.cur_tsc_offset until tsc sync is actually needed.
When the vcpu is created for the first time, its tsc is 0, and KVM
calculates its tsc_offset according to kvm_compute_l1_tsc_offset().
The userspace will then set the first non-zero tsc expected value for the
first guest vcpu, at this time there is no need to play the tsc synchronize
mechanism, the KVM should continue to write tsc_offset based on previously
used kvm_compute_l1_tsc_offset().
If the tsc synchronization mechanism is incorrectly applied at this point,
KVM will use the rewritten offset of the kvm->arch.cur_tsc_offset (on
tsc_stable machines) to write tsc_offset, which is not in line with the
expected tsc of the user space.
Based on the current code, the vcpu's tsc_offset is not configured as
expected, resulting in significant guest service response latency, which
is observed in our production environment.
Applying the tsc synchronization logic after the vcpu's tsc_generation
and KVM's cur_tsc_generation have completed their first match and started
keeping tracked helps to fix this issue, which also does not break any
existing guest tsc test cases.
Reported-by: Yong He <alexyonghe@tencent.com>
Closes: https://bugzilla.kernel.org/show_bug.cgi?id=217423
Suggested-by: Sean Christopherson <seanjc@google.com>
Suggested-by: Oliver Upton <oliver.upton@linux.dev>
Signed-off-by: Like Xu <likexu@tencent.com>
---
V1 -> V2 Changelog:
- Test the 'kvm_vcpu_has_run(vcpu)' proposal; (Sean)
- Test the kvm->arch.user_changed_tsc proposal; (Oliver)
V1: https://lore.kernel.org/kvm/20230629164838.66847-1-likexu@tencent.com/
arch/x86/kvm/x86.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index a6b9bea62fb8..4724dacea2df 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -2721,7 +2721,7 @@ static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
* kvm_clock stable after CPU hotplug
*/
synchronizing = true;
- } else {
+ } else if (kvm->arch.nr_vcpus_matched_tsc) {
u64 tsc_exp = kvm->arch.last_tsc_write +
nsec_to_cycles(vcpu, elapsed);
u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
base-commit: fdf0eaf11452d72945af31804e2a1048ee1b574c
--
2.41.0
next reply other threads:[~2023-07-24 7:35 UTC|newest]
Thread overview: 2+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-07-24 7:35 Like Xu [this message]
2023-07-29 14:53 ` [PATCH v2] KVM: x86/tsc: Use calculated tsc_offset before matching the fist vcpu's tsc Paolo Bonzini
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=20230724073516.45394-1-likexu@tencent.com \
--to=like.xu.linux@gmail.com \
--cc=kvm@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=oliver.upton@linux.dev \
--cc=pbonzini@redhat.com \
--cc=seanjc@google.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).