From: Like Xu <[email protected]>
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 <[email protected]>
Closes: https://bugzilla.kernel.org/show_bug.cgi?id=217423
Suggested-by: Sean Christopherson <[email protected]>
Suggested-by: Oliver Upton <[email protected]>
Signed-off-by: Like Xu <[email protected]>
---
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/[email protected]/
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
On 7/24/23 09:35, Like Xu wrote:
> From: Like Xu <[email protected]>
>
> 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 <[email protected]>
> Closes: https://bugzilla.kernel.org/show_bug.cgi?id=217423
> Suggested-by: Sean Christopherson <[email protected]>
> Suggested-by: Oliver Upton <[email protected]>
> Signed-off-by: Like Xu <[email protected]>
> ---
> 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/[email protected]/
>
> 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;
>
I am not sure, when will kvm->arch.nr_vcpus_matched_tsc ever be nonzero
again once a new generation starts? If this patch ever causes
synchronizing to be false, matched will also be false when calling
__kvm_synchronize_tsc.
What was wrong with Oliver's proposed patch?
Paolo