Blame SOURCES/kvm-linux-aio-explain-why-max-batch-is-checked-in-laio_i.patch

29b115
From 49d9c9dced7278517105e9cfec34ea4af716432d Mon Sep 17 00:00:00 2001
29b115
From: Stefan Hajnoczi <stefanha@redhat.com>
29b115
Date: Thu, 9 Jun 2022 17:47:12 +0100
29b115
Subject: [PATCH 6/6] linux-aio: explain why max batch is checked in
29b115
 laio_io_unplug()
29b115
29b115
RH-Author: Stefan Hajnoczi <stefanha@redhat.com>
29b115
RH-MergeRequest: 102: linux-aio: fix unbalanced plugged counter in laio_io_unplug()
29b115
RH-Commit: [2/2] b3d6421086bde50d4baad2343b2df89c5f66950e (stefanha/centos-stream-qemu-kvm)
29b115
RH-Bugzilla: 2092788
29b115
RH-Acked-by: Miroslav Rezanina <mrezanin@redhat.com>
29b115
RH-Acked-by: Kevin Wolf <kwolf@redhat.com>
29b115
RH-Acked-by: Stefano Garzarella <sgarzare@redhat.com>
29b115
29b115
It may not be obvious why laio_io_unplug() checks max batch. I discussed
29b115
this with Stefano and have added a comment summarizing the reason.
29b115
29b115
Cc: Stefano Garzarella <sgarzare@redhat.com>
29b115
Cc: Kevin Wolf <kwolf@redhat.com>
29b115
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
29b115
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
29b115
Message-id: 20220609164712.1539045-3-stefanha@redhat.com
29b115
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
29b115
(cherry picked from commit 99b969fbe105117f5af6060d3afef40ca39cc9c1)
29b115
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
29b115
---
29b115
 block/linux-aio.c | 6 ++++++
29b115
 1 file changed, 6 insertions(+)
29b115
29b115
diff --git a/block/linux-aio.c b/block/linux-aio.c
29b115
index 6078da7e42..9c2393a2f7 100644
29b115
--- a/block/linux-aio.c
29b115
+++ b/block/linux-aio.c
29b115
@@ -365,6 +365,12 @@ void laio_io_unplug(BlockDriverState *bs, LinuxAioState *s,
29b115
     assert(s->io_q.plugged);
29b115
     s->io_q.plugged--;
29b115
 
29b115
+    /*
29b115
+     * Why max batch checking is performed here:
29b115
+     * Another BDS may have queued requests with a higher dev_max_batch and
29b115
+     * therefore in_queue could now exceed our dev_max_batch. Re-check the max
29b115
+     * batch so we can honor our device's dev_max_batch.
29b115
+     */
29b115
     if (s->io_q.in_queue >= laio_max_batch(s, dev_max_batch) ||
29b115
         (!s->io_q.plugged &&
29b115
          !s->io_q.blocked && !QSIMPLEQ_EMPTY(&s->io_q.pending))) {
29b115
-- 
29b115
2.31.1
29b115