migration: avoid recursive AioContext locking in save_vmstate()
commit17e2a4a47d46dc9c33d5946cbdc1ceb15e34b5ac
authorStefan Hajnoczi <stefanha@redhat.com>
Mon, 22 May 2017 13:57:03 +0000 (22 14:57 +0100)
committerKevin Wolf <kwolf@redhat.com>
Mon, 26 Jun 2017 12:51:13 +0000 (26 14:51 +0200)
tree0cee8694b26584f5a184ac32e49576b57810454a
parentea17c9d20d7396351be5e14317354519ff53721d
migration: avoid recursive AioContext locking in save_vmstate()

AioContext was designed to allow nested acquire/release calls.  It uses
a recursive mutex so callers don't need to worry about nesting...or so
we thought.

BDRV_POLL_WHILE() is used to wait for block I/O requests.  It releases
the AioContext temporarily around aio_poll().  This gives IOThreads a
chance to acquire the AioContext to process I/O completions.

It turns out that recursive locking and BDRV_POLL_WHILE() don't mix.
BDRV_POLL_WHILE() only releases the AioContext once, so the IOThread
will not be able to acquire the AioContext if it was acquired
multiple times.

Instead of trying to release AioContext n times in BDRV_POLL_WHILE(),
this patch simply avoids nested locking in save_vmstate().  It's the
simplest fix and we should step back to consider the big picture with
all the recent changes to block layer threading.

This patch is the final fix to solve 'savevm' hanging with -object
iothread.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
migration/savevm.c