lkml.org 
[lkml]   [2021]   [Nov]   [23]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Date
SubjectRe: [PATCH] KVM: LAPIC: Per vCPU control over kvm_can_post_timer_interrupt
On Tue, 23 Nov 2021 at 15:04, Aili Yao <yaoaili126@gmail.com> wrote:
>
> On Tue, 23 Nov 2021 14:24:19 +0800
> Wanpeng Li <kernellwp@gmail.com> wrote:
>
> > On Tue, 23 Nov 2021 at 12:11, yaoaili [么爱利] <yaoaili@kingsoft.com>
> > wrote:
> > >
> > > > On Tue, 23 Nov 2021 at 03:14, Sean Christopherson
> > > > <seanjc@google.com> wrote:
> > > > >
> > > > > On Mon, Nov 22, 2021, Aili Yao wrote:
> > > > > > From: Aili Yao <yaoaili@kingsoft.com>
> > > > > >
> > > > > > When we isolate some pyhiscal cores, We may not use them for
> > > > > > kvm guests, We may use them for other purposes like DPDK, or
> > > > > > we can make some kvm guests isolated and some not, the global
> > > > > > judgement pi_inject_timer is not enough; We may make wrong
> > > > > > decisions:
> > > > > >
> > > > > > In such a scenario, the guests without isolated cores will
> > > > > > not be permitted to use vmx preemption timer, and tscdeadline
> > > > > > fastpath also be disabled, both will lead to performance
> > > > > > penalty.
> > > > > >
> > > > > > So check whether the vcpu->cpu is isolated, if not, don't
> > > > > > post timer interrupt.
> > > > > >
> > > > > > Signed-off-by: Aili Yao <yaoaili@kingsoft.com>
> > > > > > ---
> > > > > > arch/x86/kvm/lapic.c | 4 +++-
> > > > > > 1 file changed, 3 insertions(+), 1 deletion(-)
> > > > > >
> > > > > > diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c index
> > > > > > 759952dd1222..72dde5532101 100644
> > > > > > --- a/arch/x86/kvm/lapic.c
> > > > > > +++ b/arch/x86/kvm/lapic.c
> > > > > > @@ -34,6 +34,7 @@
> > > > > > #include <asm/delay.h>
> > > > > > #include <linux/atomic.h>
> > > > > > #include <linux/jump_label.h>
> > > > > > +#include <linux/sched/isolation.h>
> > > > > > #include "kvm_cache_regs.h"
> > > > > > #include "irq.h"
> > > > > > #include "ioapic.h"
> > > > > > @@ -113,7 +114,8 @@ static inline u32 kvm_x2apic_id(struct
> > > > > > kvm_lapic *apic)
> > > > > >
> > > > > > static bool kvm_can_post_timer_interrupt(struct kvm_vcpu
> > > > > > *vcpu) {
> > > > > > - return pi_inject_timer && kvm_vcpu_apicv_active(vcpu);
> > > > > > + return pi_inject_timer && kvm_vcpu_apicv_active(vcpu) &&
> > > > > > + !housekeeping_cpu(vcpu->cpu, HK_FLAG_TIMER);
> > > > >
> > > > > I don't think this is safe, vcpu->cpu will be -1 if the vCPU
> > > > > isn't scheduled in.
> > >
> > > Yes, vcpu->cpu is -1 before vcpu create, but in my environments,
> > > it didn't trigger this issue. I need to dig more, Thanks!
> > > Maybe I need one valid check here.
> > >
> > > > > This also doesn't play nice with the admin forcing
> > > > > pi_inject_timer=1. Not saying there's a reasonable use case for
> > > > > doing that, but it's supported today and this would break that
> > > > > behavior. It would also lead to weird behavior if a vCPU were
> > > > > migrated on/off a housekeeping vCPU. Again, probably not a
> > > > > reasonable use case, but I don't see anything
> > > > that would outright prevent that behavior.
> > >
> > > Yes, this is not one common operation, But I did do test some
> > > scenarios: 1. isolated cpu --> housekeeping cpu;
> > > isolated guest timer is in housekeeping CPU, for migration,
> > > kvm_can_post_timer_interrupt will return false, so the timer may be
> > > migrated to vcpu->cpu; This seems works in my test;
> > > 2. isolated --> isolated
> > > Isolated guest timer is in housekeeping cpu, for
> > > migration,kvm_can_post_timer_interrupt return true, timer is not
> > > migrated 3. housekeeping CPU --> isolated CPU
> > > non-isolated CPU timer is usually in vcpu->cpu, for migration
> > > to isolated, kvm_can_post_timer_interrupt will be true, the timer
> > > remain on the same CPU; This seems works in my test;
> > > 4. housekeeping CPU --> housekeeping CPU
> > > timer migrated;
> > > It seems this is not an affecting problem;
> > >
> > > > >
> > > > > The existing behavior also feels a bit unsafe as
> > > > > pi_inject_timer is writable while KVM is running, though I
> > > > > supposed that's orthogonal to this
> > > > discussion.
> > > > >
> > > > > Rather than check vcpu->cpu, is there an existing vCPU flag
> > > > > that can be queried, e.g. KVM_HINTS_REALTIME?
> > > >
> > > > How about something like below:
> > > >
> > > > From 67f605120e212384cb3d5788ba8c83f15659503b Mon Sep 17 00:00:00
> > > > 2001
> > > > From: Wanpeng Li <wanpengli@tencent.com>
> > > > Date: Tue, 23 Nov 2021 10:36:10 +0800
> > > > Subject: [PATCH] KVM: LAPIC: To keep the vCPUs in non-root mode
> > > > for timer- pi
> > > >
> > > > From: Wanpeng Li <wanpengli@tencent.com>
> > > >
> > > > As commit 0c5f81dad46 (KVM: LAPIC: Inject timer interrupt via
> > > > posted interrupt) mentioned that the host admin should well tune
> > > > the guest setup, so that vCPUs are placed on isolated pCPUs, and
> > > > with several pCPUs surplus for
> > > > *busy* housekeeping.
> > > > It is better to disable mwait/hlt/pause vmexits to keep the vCPUs
> > > > in non-root mode. However, we may isolate pCPUs for other purpose
> > > > like DPDK or we can make some guests isolated and others not,
> > > > Let's add the checking kvm_mwait_in_guest() to
> > > > kvm_can_post_timer_interrupt() since we can't benefit from timer
> > > > posted-interrupt w/o keeping the vCPUs in non-root mode.
> > > >
> > > > Reported-by: Aili Yao <yaoaili@kingsoft.com>
> > > > Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
> > > > ---
> > > > arch/x86/kvm/lapic.c | 5 ++---
> > > > 1 file changed, 2 insertions(+), 3 deletions(-)
> > > >
> > > > diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c index
> > > > 759952dd1222..8257566d44c7 100644
> > > > --- a/arch/x86/kvm/lapic.c
> > > > +++ b/arch/x86/kvm/lapic.c
> > > > @@ -113,14 +113,13 @@ static inline u32 kvm_x2apic_id(struct
> > > > kvm_lapic *apic)
> > > >
> > > > static bool kvm_can_post_timer_interrupt(struct kvm_vcpu *vcpu)
> > > > {
> > > > - return pi_inject_timer && kvm_vcpu_apicv_active(vcpu);
> > > > + return pi_inject_timer && kvm_mwait_in_guest(vcpu->kvm) &&
> > > > kvm_vcpu_apicv_active(vcpu);
> > > > }
> > > >
> > > > bool kvm_can_use_hv_timer(struct kvm_vcpu *vcpu) {
> > > > return kvm_x86_ops.set_hv_timer
> > > > - && !(kvm_mwait_in_guest(vcpu->kvm) ||
> > > > - kvm_can_post_timer_interrupt(vcpu));
> > > > + && !kvm_mwait_in_guest(vcpu->kvm);
> > > > }
> > > > EXPORT_SYMBOL_GPL(kvm_can_use_hv_timer);
> > >
> > > This method seems more quick and safe, but I have one question:
> > > Does this kvm_mwait_in_guest can guarantee the CPU isolated, in
> > > some production environments and usually, MWAIT feature is
> > > disabled in host and even guests with isolated CPUs. And also we
> > > can set guests kvm_mwait_in_guest true with CPUs just pinned, not
> > > isolated.
> >
> > You won't benefit from timer posted-interrupt if mwait is not exposed
> > to the guest since you can't keep CPU in non-root mode.
> > kvm_mwait_in_guest() will not guarantee the CPU is isolated, but
> > what's still bothering?
>
> Sorry, Did I miss some thing?
>
> What in my mind: MWait may be disabled in bios, so host will use halt
> instruction as one replacement for idle operation, in such a
> configuration, Mwait in guest will also be disabled even if you try to
> set kvm_mwait_in_guest true; As a result, halt,pause may not exit the
> guest, so the post interrupt still counts?

I prefer to expose mwait/hlt/pause to the guest simultaneously, you
don't need the ultra schedule latency/performance if you aren't
exposing mwait. Then why do you care about latency from the timer?

>
> For current code, We can migrate guest between isolated and
> housekeeping or we can change the cpu pinning on the fly, we allow this
> even the operation is not usually used, right?

My patch will not prevent using vmx preemption timer or tscdeadline fastpath.

Wanpeng

\
 
 \ /
  Last update: 2021-11-23 08:22    [W:2.886 / U:0.008 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site