virtio/s390: fix race on airq_areas[]
authorHalil Pasic <pasic@linux.ibm.com>
Tue, 23 Jul 2019 15:11:01 +0000 (17:11 +0200)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Mon, 16 Sep 2019 06:22:23 +0000 (08:22 +0200)
[ Upstream commit 4f419eb14272e0698e8c55bb5f3f266cc2a21c81 ]

The access to airq_areas was racy ever since the adapter interrupts got
introduced to virtio-ccw, but since commit 39c7dcb15892 ("virtio/s390:
make airq summary indicators DMA") this became an issue in practice as
well. Namely before that commit the airq_info that got overwritten was
still functional. After that commit however the two infos share a
summary_indicator, which aggravates the situation. Which means
auto-online mechanism occasionally hangs the boot with virtio_blk.

Signed-off-by: Halil Pasic <pasic@linux.ibm.com>
Reported-by: Marc Hartmayer <mhartmay@linux.ibm.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Cc: stable@vger.kernel.org
Fixes: 96b14536d935 ("virtio-ccw: virtio-ccw adapter interrupt support.")
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
drivers/s390/virtio/virtio_ccw.c

index ec54538..67efdf2 100644 (file)
@@ -132,6 +132,7 @@ struct airq_info {
        struct airq_iv *aiv;
 };
 static struct airq_info *airq_areas[MAX_AIRQ_AREAS];
+static DEFINE_MUTEX(airq_areas_lock);
 
 #define CCW_CMD_SET_VQ 0x13
 #define CCW_CMD_VDEV_RESET 0x33
@@ -244,9 +245,11 @@ static unsigned long get_airq_indicator(struct virtqueue *vqs[], int nvqs,
        unsigned long bit, flags;
 
        for (i = 0; i < MAX_AIRQ_AREAS && !indicator_addr; i++) {
+               mutex_lock(&airq_areas_lock);
                if (!airq_areas[i])
                        airq_areas[i] = new_airq_info();
                info = airq_areas[i];
+               mutex_unlock(&airq_areas_lock);
                if (!info)
                        return 0;
                write_lock_irqsave(&info->lock, flags);