From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:57908) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1Uyk0n-0001AK-3u for qemu-devel@nongnu.org; Mon, 15 Jul 2013 10:44:28 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1Uyk0k-00028m-9K for qemu-devel@nongnu.org; Mon, 15 Jul 2013 10:44:25 -0400 Received: from mx1.redhat.com ([209.132.183.28]:47600) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1Uyk0j-00027o-5W for qemu-devel@nongnu.org; Mon, 15 Jul 2013 10:44:21 -0400 Received: from int-mx02.intmail.prod.int.phx2.redhat.com (int-mx02.intmail.prod.int.phx2.redhat.com [10.5.11.12]) by mx1.redhat.com (8.14.4/8.14.4) with ESMTP id r6FEiKVQ005392 (version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=OK) for ; Mon, 15 Jul 2013 10:44:20 -0400 From: Stefan Hajnoczi Date: Mon, 15 Jul 2013 22:42:58 +0800 Message-Id: <1373899382-13514-10-git-send-email-stefanha@redhat.com> In-Reply-To: <1373899382-13514-1-git-send-email-stefanha@redhat.com> References: <1373899382-13514-1-git-send-email-stefanha@redhat.com> Subject: [Qemu-devel] [PATCH v2 09/13] linux-aio: bind EventNotifier to current AioContext List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: qemu-devel@nongnu.org Cc: Kevin Wolf , Paolo Bonzini , Stefan Hajnoczi Each block/raw-posix.c BlockDriverState has a struct qemu_laio_state that holds the Linux AIO context. Currently the EventNotifier is assigned to the main loop. In order to use block/linux-aio.c from other threads/event loops, we need to assign the EventNotifier to the current event loop. This patch implements this by assigning the EventNotifier upon the first pending request and deassigning it on the last pending request. It assumes that we flush requests before issuing I/O from another event loop (this is true). This new approach could affect performance since we assign/deassign the EventNotifier while processing I/O. However, it is the simplest way to support Linux AIO from multiple threads/event loops, so I think it's a reasonable start. Signed-off-by: Stefan Hajnoczi --- block/linux-aio.c | 19 ++++++++++++++++--- 1 file changed, 16 insertions(+), 3 deletions(-) diff --git a/block/linux-aio.c b/block/linux-aio.c index ee0f8d1..36633a2 100644 --- a/block/linux-aio.c +++ b/block/linux-aio.c @@ -56,6 +56,9 @@ static void qemu_laio_process_completion(struct qemu_laio_state *s, int ret; s->count--; + if (s->count == 0) { + qemu_aio_set_event_notifier(&s->e, NULL, NULL); + } ret = laiocb->ret; if (ret != -ECANCELED) { @@ -176,6 +179,16 @@ BlockDriverAIOCB *laio_submit(BlockDriverState *bs, void *aio_ctx, int fd, __func__, type); goto out_free_aiocb; } + + /* Assign our EventNotifier to the current thread's AioContext. This + * allows Linux AIO to be used from multiple threads so long as they flush + * before switching threads. + */ + if (s->count == 0) { + qemu_aio_set_event_notifier(&s->e, qemu_laio_completion_cb, + qemu_laio_flush_cb); + } + io_set_eventfd(&laiocb->iocb, event_notifier_get_fd(&s->e)); s->count++; @@ -185,6 +198,9 @@ BlockDriverAIOCB *laio_submit(BlockDriverState *bs, void *aio_ctx, int fd, out_dec_count: s->count--; + if (s->count == 0) { + qemu_aio_set_event_notifier(&s->e, NULL, NULL); + } out_free_aiocb: qemu_aio_release(laiocb); return NULL; @@ -203,9 +219,6 @@ void *laio_init(void) goto out_close_efd; } - qemu_aio_set_event_notifier(&s->e, qemu_laio_completion_cb, - qemu_laio_flush_cb); - return s; out_close_efd: -- 1.8.1.4