[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [PATCH v3 01/17] block: stop relying on io_flush() in b
From: |
Stefan Hajnoczi |
Subject: |
Re: [Qemu-devel] [PATCH v3 01/17] block: stop relying on io_flush() in bdrv_drain_all() |
Date: |
Fri, 14 Jun 2013 12:58:24 +0200 |
User-agent: |
Mutt/1.5.21 (2010-09-15) |
On Thu, Jun 13, 2013 at 10:33:58AM -0400, Paolo Bonzini wrote:
> Il 10/06/2013 10:38, Stefan Hajnoczi ha scritto:
> > On Mon, Jun 10, 2013 at 02:25:57PM +0200, Stefan Hajnoczi wrote:
> >> @@ -1427,26 +1456,18 @@ void bdrv_close_all(void)
> >> void bdrv_drain_all(void)
> >> {
> >> BlockDriverState *bs;
> >> - bool busy;
> >> -
> >> - do {
> >> - busy = qemu_aio_wait();
> >>
> >> + while (bdrv_requests_pending_all()) {
> >> /* FIXME: We do not have timer support here, so this is
> >> effectively
> >> * a busy wait.
> >> */
> >> QTAILQ_FOREACH(bs, &bdrv_states, list) {
> >> if (!qemu_co_queue_empty(&bs->throttled_reqs)) {
> >> qemu_co_queue_restart_all(&bs->throttled_reqs);
> >> - busy = true;
> >> }
> >> }
> >> - } while (busy);
> >>
> >> - /* If requests are still pending there is a bug somewhere */
> >> - QTAILQ_FOREACH(bs, &bdrv_states, list) {
> >> - assert(QLIST_EMPTY(&bs->tracked_requests));
> >> - assert(qemu_co_queue_empty(&bs->throttled_reqs));
> >> + qemu_aio_wait();
> >> }
> >> }
> >
> > tests/ide-test found an issue here: block.c invokes callbacks from a BH
> > so we may not yet have completed the request when this loop terminates.
> >
> > Kevin: can you fold in this patch?
> >
> > diff --git a/block.c b/block.c
> > index 31f7231..e176215 100644
> > --- a/block.c
> > +++ b/block.c
> > @@ -1469,6 +1469,9 @@ void bdrv_drain_all(void)
> >
> > qemu_aio_wait();
> > }
> > +
> > + /* Process pending completion BHs */
> > + aio_poll(qemu_get_aio_context(), false);
> > }
>
> aio_poll could require multiple iterations, this is why the old code was
> starting with "busy = qemu_aio_wait()". You can add a while loop around
> the new call, or do something more similar to the old code, along the
> lines of "do { QTAILQ_FOREACH(...) ...; busy =
> bdrv_request_pending_all(); busy |= aio_poll(qemu_get_aio_context(),
> busy); } while(busy);".
Good idea, the trick is to use busy as the blocking flag.
I will resend with the fix to avoid making this email thread too messy.
Stefan
- [Qemu-devel] [PATCH v3 00/17] aio: drop io_flush(), Stefan Hajnoczi, 2013/06/10
- [Qemu-devel] [PATCH v3 03/17] tests: adjust test-aio to new aio_poll() semantics, Stefan Hajnoczi, 2013/06/10
- [Qemu-devel] [PATCH v3 02/17] dataplane/virtio-blk: check exit conditions before aio_poll(), Stefan Hajnoczi, 2013/06/10
- [Qemu-devel] [PATCH v3 05/17] aio: stop using .io_flush(), Stefan Hajnoczi, 2013/06/10
- [Qemu-devel] [PATCH v3 06/17] block/curl: drop curl_aio_flush(), Stefan Hajnoczi, 2013/06/10
- [Qemu-devel] [PATCH v3 07/17] block/gluster: drop qemu_gluster_aio_flush_cb(), Stefan Hajnoczi, 2013/06/10
- [Qemu-devel] [PATCH v3 08/17] block/iscsi: drop iscsi_process_flush(), Stefan Hajnoczi, 2013/06/10
- [Qemu-devel] [PATCH v3 09/17] block/linux-aio: drop qemu_laio_completion_cb(), Stefan Hajnoczi, 2013/06/10
- [Qemu-devel] [PATCH v3 10/17] block/nbd: drop nbd_have_request(), Stefan Hajnoczi, 2013/06/10
- [Qemu-devel] [PATCH v3 12/17] block/sheepdog: drop have_co_req() and aio_flush_request(), Stefan Hajnoczi, 2013/06/10
- [Qemu-devel] [PATCH v3 11/17] block/rbd: drop qemu_rbd_aio_flush_cb(), Stefan Hajnoczi, 2013/06/10