media: raspberrypi: rp1-cfe: acquire state_lock in cfe_start_streaming()
Subsystems (3)
PATCH 1/1 media: raspberrypi: rp1-cfe: acquire state_lock in cfe_start_streaming()
cfe_start_streaming() modifies shared device state without holding state_lock. The driver exposes multiple video nodes backed by a single cfe_device. While one node runs cfe_start_streaming(), another node's cfe_buffer_queue() may read the node state via test_all_nodes() under state_lock to decide whether to schedule a job. Another case is when node->fs_count is read by the interrupt handler in cfe_sof_isr(). Modifying this state without state_lock races against those readers.
The counterpart cfe_stop_streaming() already takes state_lock around its state modification. Fix cfe_start_streaming() to do the same.
Found by code inspection.
--- drivers/media/platform/raspberrypi/rp1-cfe/cfe.c | 3 +++ 1 file changed, 3 insertions(+)
drivers/media/platform/raspberrypi/rp1-cfe/cfe.c +3 -0
@@ -1151,6 +1151,7 @@ static int cfe_start_streaming(struct vb2_queue *vq, unsigned int count)struct v4l2_subdev_state *state;struct v4l2_subdev_route *route;s64 link_freq; +unsigned long flags;int ret;cfe_dbg (cfe,"%s: [%s]\n" , __func__, node_desc[node->id].name);
@@ -1184,9 +1185,11 @@ static int cfe_start_streaming(struct vb2_queue *vq, unsigned int count) state =v4l2_subdev_lock_and_get_active_state (&cfe->csi2.sd); +spin_lock_irqsave (&cfe->state_lock, flags);clear_state (cfe,FS_INT |FE_INT , node->id);set_state (cfe,NODE_STREAMING , node->id); node->fs_count =0 ; +spin_unlock_irqrestore (&cfe->state_lock, flags); ret =cfe_start_channel (node);if (ret)
base-commit: 06cb687a5132fcffe624c0070576ab852ac6b568
prerequisite-patch-id: 4c010e20cdeb611d14546bc729b513f959e25afd
Signature · 1 line
2.53.0