Re: [PATCH] KVM: drop the kvm_has_noapic_vcpu optimization

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Fri, Nov 01, 2024 at 07:49:10AM -0700, Sean Christopherson wrote:
> On Fri, Nov 01, 2024, Bernhard Kauer wrote:
> > On Thu, Oct 31, 2024 at 04:29:52PM -0700, Sean Christopherson wrote:
> > > With a userspace APIC, the roundtrip to userspace to emulate the EOI is measured
> > > in tens of thousands of cycles.  IIRC, last I played around with userspace exits
> > > the average turnaround time was ~50k cycles.
> > 
> > 
> > That sound a lot so I did some quick benchmarking.  An exit is around 1400
> > TSC cycles on my AMD laptop, instruction emulation takes 1200 and going
> > to user-level needs at least 6200.  Not terribly slow but still room for
> > optimizations.
> 
> Ah, I suspect my recollection of ~50k cycles is from measuring all exits to
> userspace, i.e. included the reaaaaly slow paths.

I finally found the reason for the slow user-level roundtrip on my Zen3+
machine.

Disabling SRSO with spec_rstack_overflow=off improves the user-level part
by 3x.  The exit as well as the instruction emulation overhead is down by 40%.

Thus without SRSO a roundtrip to user-level needs roughly 2000 cycles.


                        SRSO=off        default         factor
INSTR   CPUID           1008            1394            1.4x
        RDMSR           1072            1550            1.4x
MMIO    APIC            1666            2609            1.6x
        IOAPIC          1783            2800            1.6x
        HPET            3626            9426            2.6x
PIO     PIC             1250            1804            1.4x
        UART            2837            8011            2.8x





[Index of Archives]     [KVM ARM]     [KVM ia64]     [KVM ppc]     [Virtualization Tools]     [Spice Development]     [Libvirt]     [Libvirt Users]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite Questions]     [Linux Kernel]     [Linux SCSI]     [XFree86]

  Powered by Linux