Re: [PATCH 0/4] x86/hyper-v: optimize PV IPIs

From: Vitaly Kuznetsov
Date: Wed Jun 27 2018 - 05:25:26 EST


Wanpeng Li <kernellwp@xxxxxxxxx> writes:

> Hi Vitaly, (fix my reply mess this time)
> On Sat, 23 Jun 2018 at 01:09, Vitaly Kuznetsov <vkuznets@xxxxxxxxxx> wrote:
>>
>> When reviewing my "x86/hyper-v: use cheaper HVCALL_FLUSH_VIRTUAL_ADDRESS_
>> {LIST,SPACE} hypercalls when possible" patch Michael suggested to apply the
>> same idea to PV IPIs. Here we go!
>>
>> Despite what Hyper-V TLFS says about HVCALL_SEND_IPI hypercall, it can
>> actually be 'fast' (passing parameters through registers). Use that too.
>>
>> This series can collide with my "KVM: x86: hyperv: PV IPI support for
>> Windows guests" series as I rename ipi_arg_non_ex/ipi_arg_ex structures
>> there. Depending on which one gets in first we may need to do tiny
>> adjustments.
>
> As hyperv PV TLB flush has already been merged, is there any other
> obvious multicast IPIs scenarios? qemu supports interrupt remapping
> since two years ago, I think windows guest can switch to cluster mode
> after entering x2APIC, so sending IPI per cluster.

When we manifest ourselves as Hyper-V Windows 'forgets' about x2apic
mode: Hyper-V has a concept of 'Synthetic interrupt controller' - an
xapic extension which we also support in KVM. I don't really know any
obvious scenarios for mass IPIs in Windows besides TLB flush but I'm
worried they may exist. Without PV IPIs any such attempt will likely
lead to a crash.

In general, I do care more about completeness and correctness of our
Hyper-V emulation at this point: Windows is only being tested on 'real'
Hyper-Vs so when we emulate a subset of enlightenments we're on our own
when something is not working. It is also very helpfult for
Linux-on-Hyper-V depelopment as we can see how Windows-on-Hyper-v
behaves :-)

> In addition, you
> can also post the benchmark result for this PV IPI optimization,
> although it also fixes the bug which you mentioned above.

I'd love to get to know how to trigger mass IPIs in Windows so a
benchmark can be performed...

> I can post one variant for Linux guest PV IPI if it also makes
> sense. :)

With x2apic support I'm actually not sure. Maybe configurations with
a very large number of vCPUs and IPIs going to > 256 vCPUs can benefit
from a 'single hypercall' solution.

--
Vitaly