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

Re: [Xen-devel] performace issue when turn on apicv



On Thu, Jun 18, 2015 at 10:35:37AM +0100, Jan Beulich wrote:
> >>> On 18.06.15 at 11:18, <yang.z.zhang@xxxxxxxxx> wrote:
> > Jan Beulich wrote on 2015-06-18:
> >>>>> On 18.06.15 at 10:53, <yang.z.zhang@xxxxxxxxx> wrote:
> >>> Jan Beulich wrote on 2015-06-18:
> >>>>>>> On 18.06.15 at 10:20, <JBeulich@xxxxxxxx> wrote:
> >>>>> Apart from that I notice that the EXIT_REASON_EOI_INDUCED handling
> >>>>> also adds about the same number of ticks...
> >>>> 
> >>>> Are there any other devices in the guest causing meaningful amounts
> >>>> of interrupts (I suppose the SSD itself is using just one)? I ask
> >>>> since I wonder whether e.g. there may be lock contention in
> >> vioapic_update_EOI().
> >>> 
> >>> The EXIT_REASON_EOI_INDUCED is for local apic time.
> >> 
> >> ??? (I.e. I don't see the connection between the context and your reply.
> >> Perhaps I'm missing something obvious...)
> > 
> > I mean the lots of EXIT_REASON_EOI_INDUCEDis for local apic time which is 
> > edge-trigger. So it will not call vioapic_update_EOI.
> 
> Okay, this then suggests the problem being hvm_dpci_msi_eoi(),
> albeit looking at it I can't immediately see why it would take long.

We do take two spin-locks (d->event_lock) and (desc->lock) and also
can IPI set_eoi_ready across different CPUs.

Perhaps it might be worth adding some TRACE with the time taken
to get the locks and also doing the IPIs?

> 
> Jan
> 
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@xxxxxxxxxxxxx
> http://lists.xen.org/xen-devel

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel


 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.