On Tue, May 22, 2012 at 7:29 PM, Paolo Bonzini <pbonz...@redhat.com> wrote:
> Il 22/05/2012 11:15, ronnie sahlberg ha scritto:
>> Hi,
>>
>> Now that I see what happens, I can easily workaround this in block/iscsi.c
>> by the patch below, but I dont know if this is the right thing to do.
>>
>> It does appear that here, when I use qemu_set_fd_handler() and add a
>> handler for "writeble" it takes 55ms before the event system notices
>> this and reacts.
>>
>>
>>
>> diff --git a/block/iscsi.c b/block/iscsi.c
>> index d37c4ee..1ebff0f 100644
>> --- a/block/iscsi.c
>> +++ b/block/iscsi.c
>> @@ -105,6 +105,10 @@ iscsi_set_events(IscsiLun *iscsilun)
>> {
>> struct iscsi_context *iscsi = iscsilun->iscsi;
>>
>> + if (iscsi_which_events(iscsi) & POLLOUT) {
>> + iscsi_process_write(iscsilun);
>> + }
>> +
>> qemu_aio_set_fd_handler(iscsi_get_fd(iscsi), iscsi_process_read,
>> (iscsi_which_events(iscsi) & POLLOUT)
>> ? iscsi_process_write : NULL,
>
> Doh, now I remember. Whenever you change the aio handlers you need to
> call qemu_notify_event() afterwards, if the handler may fire right away.
Thanks. I just confirmed that qemu_notify_event() fixes the issue.
Ill send a patch that uses qemu_notify_event() and a comment why this is needed.
regards
ronnie sahlberg
