On Wed, Jun 26, 2019 at 07:02:13PM +0800, Wanpeng Li wrote:
> On Wed, 26 Jun 2019 at 03:03, Marcelo Tosatti <mtosatti@xxxxxxxxxx> wrote:
> >
> > On Mon, Jun 24, 2019 at 04:53:53PM +0800, Wanpeng Li wrote:
> > > On Sat, 22 Jun 2019 at 06:11, Marcelo Tosatti <mtosatti@xxxxxxxxxx> wrote:
> > > >
> > > > On Fri, Jun 21, 2019 at 09:42:39AM +0800, Wanpeng Li wrote:
> > > > > On Thu, 20 Jun 2019 at 05:04, Marcelo Tosatti <mtosatti@xxxxxxxxxx> wrote:
> > > > > >
> > > > > > Hi Li,
> > > > > >
> > > > > > On Wed, Jun 19, 2019 at 08:36:06AM +0800, Wanpeng Li wrote:
> > > > > > > On Tue, 18 Jun 2019 at 21:36, Marcelo Tosatti <mtosatti@xxxxxxxxxx> wrote:
> > > > > > > >
> > > > > > > > On Mon, Jun 17, 2019 at 07:24:44PM +0800, Wanpeng Li wrote:
> > > > > > > > > From: Wanpeng Li <wanpengli@xxxxxxxxxxx>
> > > > > > > > >
> > > > > > > > > Dedicated instances are currently disturbed by unnecessary jitter due
> > > > > > > > > to the emulated lapic timers fire on the same pCPUs which vCPUs resident.
> > > > > > > > > There is no hardware virtual timer on Intel for guest like ARM. Both
> > > > > > > > > programming timer in guest and the emulated timer fires incur vmexits.
> > > > > > > > > This patch tries to avoid vmexit which is incurred by the emulated
> > > > > > > > > timer fires in dedicated instance scenario.
> > > > > > > > >
> > > > > > > > > When nohz_full is enabled in dedicated instances scenario, the emulated
> > > > > > > > > timers can be offload to the nearest busy housekeeping cpus since APICv
> > > > > > > > > is really common in recent years. The guest timer interrupt is injected
> > > > > > > > > by posted-interrupt which is delivered by housekeeping cpu once the emulated
> > > > > > > > > timer fires.
> > > > > > > > >
> > > > > > > > > The host admin should fine tuned, e.g. dedicated instances scenario w/
> > > > > > > > > nohz_full cover the pCPUs which vCPUs resident, several pCPUs surplus
> > > > > > > > > for busy housekeeping, disable mwait/hlt/pause vmexits to keep in non-root
> > > > > > > > > mode, ~3% redis performance benefit can be observed on Skylake server.
> > > > > > > > >
> > > > > > > > > w/o patch:
> > > > > > > > >
> > > > > > > > > VM-EXIT Samples Samples% Time% Min Time Max Time Avg time
> > > > > > > > >
> > > > > > > > > EXTERNAL_INTERRUPT 42916 49.43% 39.30% 0.47us 106.09us 0.71us ( +- 1.09% )
> > > > > > > > >
> > > > > > > > > w/ patch:
> > > > > > > > >
> > > > > > > > > VM-EXIT Samples Samples% Time% Min Time Max Time Avg time
> > > > > > > > >
> > > > > > > > > EXTERNAL_INTERRUPT 6871 9.29% 2.96% 0.44us 57.88us 0.72us ( +- 4.02% )
> > > > > > > > >
> > > > > > > > > Cc: Paolo Bonzini <pbonzini@xxxxxxxxxx>
> > > > > > > > > Cc: Radim Krčmář <rkrcmar@xxxxxxxxxx>
> > > > > > > > > Cc: Marcelo Tosatti <mtosatti@xxxxxxxxxx>
> > > > > > > > > Signed-off-by: Wanpeng Li <wanpengli@xxxxxxxxxxx>
> > > > > > > > > ---
> > > > > > > > > arch/x86/kvm/lapic.c | 33 ++++++++++++++++++++++++++-------
> > > > > > > > > arch/x86/kvm/lapic.h | 1 +
> > > > > > > > > arch/x86/kvm/vmx/vmx.c | 3 ++-
> > > > > > > > > arch/x86/kvm/x86.c | 5 +++++
> > > > > > > > > arch/x86/kvm/x86.h | 2 ++
> > > > > > > > > include/linux/sched/isolation.h | 2 ++
> > > > > > > > > kernel/sched/isolation.c | 6 ++++++
> > > > > > > > > 7 files changed, 44 insertions(+), 8 deletions(-)
> > > > > > > > >
> > > > > > > > > diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
> > > > > > > > > index 87ecb56..9ceeee5 100644
> > > > > > > > > --- a/arch/x86/kvm/lapic.c
> > > > > > > > > +++ b/arch/x86/kvm/lapic.c
> > > > > > > > > @@ -122,6 +122,13 @@ static inline u32 kvm_x2apic_id(struct kvm_lapic *apic)
> > > > > > > > > return apic->vcpu->vcpu_id;
> > > > > > > > > }
> > > > > > > > >
> > > > > > > > > +bool posted_interrupt_inject_timer(struct kvm_vcpu *vcpu)
> > > > > > > > > +{
> > > > > > > > > + return pi_inject_timer && kvm_vcpu_apicv_active(vcpu) &&
> > > > > > > > > + kvm_hlt_in_guest(vcpu->kvm);
> > > > > > > > > +}
> > > > > > > > > +EXPORT_SYMBOL_GPL(posted_interrupt_inject_timer);
> > > > > > > >
> > > > > > > > Paolo, can you explain the reasoning behind this?
> > > > > > > >
> > > > > > > > Should not be necessary...
> > > > >
> > > > > https://lkml.org/lkml/2019/6/5/436 "Here you need to check
> > > > > kvm_halt_in_guest, not kvm_mwait_in_guest, because you need to go
> > > > > through kvm_apic_expired if the guest needs to be woken up from
> > > > > kvm_vcpu_block."
> > > >
> > > > Ah, i think he means that a sleeping vcpu (in kvm_vcpu_block) must
> > > > be woken up, if it receives a timer interrupt.
> > > >
> > > > But your patch will go through:
> > > >
> > > > kvm_apic_inject_pending_timer_irqs
> > > > __apic_accept_irq ->
> > > > vmx_deliver_posted_interrupt ->
> > > > kvm_vcpu_trigger_posted_interrupt returns false
> > > > (because vcpu->mode != IN_GUEST_MODE) ->
> > > > kvm_vcpu_kick
> > > >
> > > > Which will wakeup the vcpu.
> > >
> > > Hi Marcelo,
> > >
> > > >
> > > > Apart from this oops, which triggers when running:
> > > > taskset -c 1 ./cyclictest -D 3600 -p 99 -t 1 -h 30 -m -n -i 50000 -b 40
> > >
> > > I try both host and guest use latest kvm/queue w/ CONFIG_PREEMPT
> > > enabled, and expose mwait as your config, however, there is no oops.
> > > Can you reproduce steadily or encounter casually? Can you reproduce
> > > w/o the patchset?
> >
> > Hi Li,
>
> Hi Marcelo,
>
> >
> > Steadily.
> >
> > Do you have this as well:
>
> w/ or w/o below diff, testing on both SKX and HSW servers on hand, I
> didn't see any oops. Could you observe the oops disappear when w/o
> below diff? If the answer is yes, then the oops will not block to
> merge the patchset since Paolo prefers to add the kvm_hlt_in_guest()
> condition to guarantee be woken up from kvm_vcpu_block().
He agreed that its not necessary. Removing the HLT in guest widens
the scope of the patch greatly.
> For the
> exitless injection if the guest is running(DPDK style workloads that
> busy-spin on network card) scenarios, we can find a solution later.
What is the use-case for HLT in guest again?
I'll find the source for the oops (or confirm can't reproduce with
kvm/queue RSN).
