Re: [Qemu-devel] [4341] qemu: generate signals on tap I/O

[Anders]

Anthony Liguori wrote:
> Anders wrote:
>> Anthony Liguori wrote:
>>> You're seeing improvement with normal QEMU?  Can you please post a
>>> description of what you're seeing improve with SIGIO.  SIGIO should
>>> really only slow things down.
>>>
>>
>> I found this issue in KVM. I have not timed in QEMU, but the issue
>> seemed generic, so I posted it upstream.
>
> Right, this was an artifact of how KVM handled select.  It's not generic
> at all.

If that is true, then the SIGALRM IO handler must also be redundant?


> In KVM today, select() is in it's own thread so it will service IO
> immediately.  This makes SIGIO unnecessary (not quite correct yet but
> will be very soon).

Yes, you keep saying that. I think that work on a new implementation is
orthogonal to fixing the current one, though.


Cheers,
Anders.