From: "Xu, Like" <like.xu@intel.com>
To: Peter Zijlstra <peterz@infradead.org>, Like Xu <like.xu@linux.intel.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>,
linux-kernel@vger.kernel.org, kvm@vger.kernel.org,
Sean Christopherson <sean.j.christopherson@intel.com>,
Vitaly Kuznetsov <vkuznets@redhat.com>,
Wanpeng Li <wanpengli@tencent.com>,
Jim Mattson <jmattson@google.com>, Joerg Roedel <joro@8bytes.org>,
Thomas Gleixner <tglx@linutronix.de>,
ak@linux.intel.com, wei.w.wang@intel.com
Subject: Re: [PATCH v11 10/11] KVM: x86/pmu: Check guest LBR availability in case host reclaims them
Date: Tue, 19 May 2020 21:10:58 +0800 [thread overview]
Message-ID: <3a234754-e103-907f-9b06-44b5e7ae12d3@intel.com> (raw)
In-Reply-To: <20200519111559.GJ279861@hirez.programming.kicks-ass.net>
On 2020/5/19 19:15, Peter Zijlstra wrote:
> On Thu, May 14, 2020 at 04:30:53PM +0800, Like Xu wrote:
>
>> diff --git a/arch/x86/kvm/vmx/pmu_intel.c b/arch/x86/kvm/vmx/pmu_intel.c
>> index ea4faae56473..db185dca903d 100644
>> --- a/arch/x86/kvm/vmx/pmu_intel.c
>> +++ b/arch/x86/kvm/vmx/pmu_intel.c
>> @@ -646,6 +646,43 @@ static void intel_pmu_lbr_cleanup(struct kvm_vcpu *vcpu)
>> intel_pmu_free_lbr_event(vcpu);
>> }
>>
>> +static bool intel_pmu_lbr_is_availabile(struct kvm_vcpu *vcpu)
>> +{
>> + struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
>> +
>> + if (!pmu->lbr_event)
>> + return false;
>> +
>> + if (event_is_oncpu(pmu->lbr_event)) {
>> + intel_pmu_intercept_lbr_msrs(vcpu, false);
>> + } else {
>> + intel_pmu_intercept_lbr_msrs(vcpu, true);
>> + return false;
>> + }
>> +
>> + return true;
>> +}
> This is unreadable gunk, what?
Abstractly, it is saying "KVM would passthrough the LBR satck MSRs if
event_is_oncpu() is true, otherwise cancel the passthrough state if any."
I'm using 'event->oncpu != -1' to represent the guest LBR event
is scheduled on rather than 'event->state == PERF_EVENT_STATE_ERROR'.
For intel_pmu_intercept_lbr_msrs(), false means to passthrough the LBR stack
MSRs to the vCPU, and true means to cancel the passthrough state and make
LBR MSR accesses trapped by the KVM.
>
>> +/*
>> + * Higher priority host perf events (e.g. cpu pinned) could reclaim the
>> + * pmu resources (e.g. LBR) that were assigned to the guest. This is
>> + * usually done via ipi calls (more details in perf_install_in_context).
>> + *
>> + * Before entering the non-root mode (with irq disabled here), double
>> + * confirm that the pmu features enabled to the guest are not reclaimed
>> + * by higher priority host events. Otherwise, disallow vcpu's access to
>> + * the reclaimed features.
>> + */
>> +static void intel_pmu_availability_check(struct kvm_vcpu *vcpu)
>> +{
>> + lockdep_assert_irqs_disabled();
>> +
>> + if (lbr_is_enabled(vcpu) && !intel_pmu_lbr_is_availabile(vcpu) &&
>> + (vmcs_read64(GUEST_IA32_DEBUGCTL) & DEBUGCTLMSR_LBR))
>> + pr_warn_ratelimited("kvm: vcpu-%d: LBR is temporarily unavailable.\n",
>> + vcpu->vcpu_id);
> More unreadable nonsense; when the events go into ERROR state, it's a
> permanent fail, they'll not come back.
It's not true. The guest LBR event with 'ERROR state' or 'oncpu != -1'
would be
lazy released and re-created in the next time the
intel_pmu_create_lbr_event() is
called and it's supposed to be re-scheduled and re-do availability_check()
as well.
From the perspective of the guest user, the guest LBR is only temporarily
unavailable
until the host no longer reclaims the LBR.
>
>> +}
>> +
>> struct kvm_pmu_ops intel_pmu_ops = {
>> .find_arch_event = intel_find_arch_event,
>> .find_fixed_event = intel_find_fixed_event,
>> @@ -662,4 +699,5 @@ struct kvm_pmu_ops intel_pmu_ops = {
>> .reset = intel_pmu_reset,
>> .deliver_pmi = intel_pmu_deliver_pmi,
>> .lbr_cleanup = intel_pmu_lbr_cleanup,
>> + .availability_check = intel_pmu_availability_check,
>> };
>> diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
>> index 9969d663826a..80d036c5f64a 100644
>> --- a/arch/x86/kvm/vmx/vmx.c
>> +++ b/arch/x86/kvm/vmx/vmx.c
>> @@ -6696,8 +6696,10 @@ static fastpath_t vmx_vcpu_run(struct kvm_vcpu *vcpu)
>>
>> pt_guest_enter(vmx);
>>
>> - if (vcpu_to_pmu(vcpu)->version)
>> + if (vcpu_to_pmu(vcpu)->version) {
>> atomic_switch_perf_msrs(vmx);
>> + kvm_x86_ops.pmu_ops->availability_check(vcpu);
>> + }
> AFAICT you just did a call out to the kvm_pmu crud in
> atomic_switch_perf_msrs(), why do another call?
In fact, availability_check() is only called here for just one time.
The callchain looks like:
- vmx_vcpu_run()
- kvm_x86_ops.pmu_ops->availability_check();
- intel_pmu_availability_check()
- intel_pmu_lbr_is_availabile()
- event_is_oncpu() ...
Thanks,
Like Xu
>
>
next prev parent reply other threads:[~2020-05-19 13:11 UTC|newest]
Thread overview: 35+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-05-14 8:30 [PATCH v11 00/11] Guest Last Branch Recording Enabling Like Xu
2020-05-14 8:30 ` [PATCH v11 01/11] perf/x86: Fix variable types for LBR registers Like Xu
2020-05-14 8:30 ` [PATCH v11 02/11] perf/x86/core: Refactor hw->idx checks and cleanup Like Xu
2020-05-14 8:30 ` [PATCH v11 03/11] perf/x86/lbr: Add interface to get basic information about LBR stack Like Xu
2020-05-14 8:30 ` [PATCH v11 04/11] perf/x86: Add constraint to create guest LBR event without hw counter Like Xu
2020-05-18 11:43 ` Peter Zijlstra
2020-07-03 8:01 ` [tip: perf/core] " tip-bot2 for Like Xu
2020-05-14 8:30 ` [PATCH v11 05/11] perf/x86: Keep LBR stack unchanged in host context for guest LBR event Like Xu
2020-05-18 11:59 ` Peter Zijlstra
2020-05-18 12:02 ` Peter Zijlstra
2020-05-19 3:08 ` Like Xu
2020-05-19 10:45 ` Peter Zijlstra
2020-05-19 13:25 ` Xu, Like
2020-07-03 8:01 ` [tip: perf/core] perf/x86: Keep LBR records unchanged in host context for guest usage tip-bot2 for Like Xu
2020-05-14 8:30 ` [PATCH v11 06/11] KVM: x86/pmu: Tweak kvm_pmu_get_msr to pass 'struct msr_data' in Like Xu
2020-05-14 8:30 ` [PATCH v11 07/11] KVM: x86: Expose MSR_IA32_PERF_CAPABILITIES for LBR record format Like Xu
2020-05-19 10:53 ` Peter Zijlstra
2020-05-19 12:19 ` Xu, Like
2020-05-19 15:12 ` Sean Christopherson
2020-05-14 8:30 ` [PATCH v11 08/11] KVM: x86/pmu: Emulate LBR feature via guest LBR event Like Xu
2020-05-19 11:00 ` Peter Zijlstra
2020-05-19 12:24 ` Xu, Like
2020-05-19 11:01 ` Peter Zijlstra
2020-05-19 12:28 ` Xu, Like
2020-05-19 11:03 ` Peter Zijlstra
2020-05-19 12:40 ` Xu, Like
2020-05-14 8:30 ` [PATCH v11 09/11] KVM: x86/pmu: Release guest LBR event via vPMU lazy release mechanism Like Xu
2020-05-14 8:30 ` [PATCH v11 10/11] KVM: x86/pmu: Check guest LBR availability in case host reclaims them Like Xu
2020-05-19 11:15 ` Peter Zijlstra
2020-05-19 13:10 ` Xu, Like [this message]
2020-05-19 14:57 ` Peter Zijlstra
2020-05-20 2:01 ` Xu, Like
2020-05-27 8:17 ` Like Xu
2020-05-14 8:30 ` [PATCH v11 11/11] KVM: x86/pmu: Reduce the overhead of LBR passthrough or cancellation Like Xu
2020-05-27 8:28 ` [PATCH v11 00/11] Guest Last Branch Recording Enabling Xu, Like
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=3a234754-e103-907f-9b06-44b5e7ae12d3@intel.com \
--to=like.xu@intel.com \
--cc=ak@linux.intel.com \
--cc=jmattson@google.com \
--cc=joro@8bytes.org \
--cc=kvm@vger.kernel.org \
--cc=like.xu@linux.intel.com \
--cc=linux-kernel@vger.kernel.org \
--cc=pbonzini@redhat.com \
--cc=peterz@infradead.org \
--cc=sean.j.christopherson@intel.com \
--cc=tglx@linutronix.de \
--cc=vkuznets@redhat.com \
--cc=wanpengli@tencent.com \
--cc=wei.w.wang@intel.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).