Hi Greg,
On Mon, Jan 29, 2018 at 7:47 PM, Greg Kurz <gr...@kaod.org> wrote:
> On Mon, 29 Jan 2018 16:20:55 +0200
> Gal Hammer <gham...@redhat.com> wrote:
>
>> Using a cleanup callback function (Version 2 of this patches) in order to
>> support transactions fails when the same event notifier fd was used by more
>> than one Memory Region.
>>
>> This time I ask the caller to do the event notifier cleanup by himself.
>>
>
> Much better ! QEMU no longer aborts. Also I could observe a speed-up when
> booting a fedora27 guest with 5 virtio-serial devices (max_ports=511), from
> 1m40s down to 1m10s. Impressive. :)
Thanks for the quick review, testing, and the kudos ;-).
After spending some time trying to understand why the previous patch
fails I was able to figure it out. I was using
"disable-legacy=on,disable-modern=off", so I missed the fact that a
"transition" device registers two different memory regions with the
same event fd.
> For all the patches in this series:
>
> Reviewed-by: Greg Kurz <gr...@kaod.org>
>
> and
>
> Tested-by: Greg Kurz <gr...@kaod.org>
>
>> Gal Hammer (2):
>> virtio: remove event notifier cleanup call on de-assign
>> virtio: improve virtio devices initialization time
>>
>> hw/block/dataplane/virtio-blk.c | 2 ++
>> hw/scsi/virtio-scsi-dataplane.c | 2 ++
>> hw/virtio/vhost.c | 2 ++
>> hw/virtio/virtio-bus.c | 14 ++++++++++----
>> hw/virtio/virtio.c | 22 +++++++++++++++++++++-
>> include/hw/virtio/virtio-bus.h | 2 ++
>> 6 files changed, 39 insertions(+), 5 deletions(-)
>>
>
BTW: There is a pending patch for kvm that will trim few more seconds
from the boot time...
Gal.
