crypto: mxs-dcp - Fix wait logic on chan threads
authorLeonard Crestez <leonard.crestez@nxp.com>
Fri, 21 Sep 2018 15:03:18 +0000 (18:03 +0300)
committerHerbert Xu <herbert@gondor.apana.org.au>
Fri, 28 Sep 2018 04:45:37 +0000 (12:45 +0800)
commitd80771c08363ad7fbf0f56f5301e7ca65065c582
tree8f7a8d92e7d942308902084ad75e7fd13593b186
parentadd92a817e60e308a419693413a38d9d1e663aff
crypto: mxs-dcp - Fix wait logic on chan threads

When compiling with CONFIG_DEBUG_ATOMIC_SLEEP=y the mxs-dcp driver
prints warnings such as:

WARNING: CPU: 0 PID: 120 at kernel/sched/core.c:7736 __might_sleep+0x98/0x9c
do not call blocking ops when !TASK_RUNNING; state=1 set at [<8081978c>] dcp_chan_thread_sha+0x3c/0x2ec

The problem is that blocking ops will manipulate current->state
themselves so it is not allowed to call them between
set_current_state(TASK_INTERRUPTIBLE) and schedule().

Fix this by converting the per-chan mutex to a spinlock (it only
protects tiny list ops anyway) and rearranging the wait logic so that
callbacks are called current->state as TASK_RUNNING. Those callbacks
will indeed call blocking ops themselves so this is required.

Cc: <stable@vger.kernel.org>
Signed-off-by: Leonard Crestez <leonard.crestez@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
drivers/crypto/mxs-dcp.c