]> git.itanic.dy.fi Git - linux-stable/commit
Drivers: hv: vmbus: Fix signaling logic in hv_need_to_signal_on_read()
authorK. Y. Srinivasan <kys@microsoft.com>
Sat, 2 Apr 2016 23:17:38 +0000 (16:17 -0700)
committerSasha Levin <sasha.levin@oracle.com>
Tue, 17 May 2016 17:42:54 +0000 (13:42 -0400)
commitc172113994b7feb3f1d7fa92b9e803e31d03a256
tree3ffbc0f351cffa854d620fdddf391b1c74e54cc4
parent26ac029f2468a529bf5c9c3a5d9ef7a27376218a
Drivers: hv: vmbus: Fix signaling logic in hv_need_to_signal_on_read()

[ Upstream commit 1db488d12894f1936360779d6ab2aede3dd7f06a ]

On the consumer side, we have interrupt driven flow management of the
producer. It is sufficient to base the signaling decision on the
amount of space that is available to write after the read is complete.
The current code samples the previous available space and uses this
in making the signaling decision. This state can be stale and is
unnecessary. Since the state can be stale, we end up not signaling
the host (when we should) and this can result in a hang. Fix this
problem by removing the unnecessary check. I would like to thank
Arseney Romanenko <arseneyr@microsoft.com> for pointing out this issue.

Also, issue a full memory barrier before making the signaling descision
to correctly deal with potential reordering of the write (read index)
followed by the read of pending_sz.

Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Tested-by: Dexuan Cui <decui@microsoft.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
drivers/hv/ring_buffer.c