Il 08/10/2013 14:33, Kevin Wolf ha scritto:
>>> > > this converts read, write and flush functions from aio to coroutines.
>> > 
>> > I'm not sure it's already the time for this...  Cancellation sucks in
>> > QEMU, and this is going to make things even worse.
> Not sure what you're referring to. If you mean iscsi_aio_cancel(), isn't
> it dead code anyway since we changed block.c to use coroutines for
> everything? bdrv_co_io_em() even throws the acb away, so even if you
> wanted, there's no way to cancel the request even today.

SCSI tries to use cancellation, and this results in VCPU threads
starving all other threads.  So I would like to introduce cancellation
points for coroutines.

Paolo

Reply via email to