Lockdep warns about a possible deadlock resulting from the use of regular
spin_locks:
=================================
[ INFO: inconsistent lock state ]
4.4.0-rc2+ #2724 Not tainted
---------------------------------
inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
ksoftirqd/0/3 [HC0[0]:SC1[1]:HE1:SE0] takes:
(&(&crypto_info->lock)->rlock){+.?...}, at: [<bf14a65c>] rk_crypto_tasklet_cb+0x24/0xb4 [rk_crypto]
{SOFTIRQ-ON-W} state was registered at:
[<c007f4ac>] lock_acquire+0x178/0x218
[<c0759bac>] _raw_spin_lock+0x54/0x64
[<bf14af88>] rk_handle_req+0x7c/0xbc [rk_crypto]
[<bf14b040>] rk_des_ecb_encrypt+0x2c/0x30 [rk_crypto]
[<bf14b05c>] rk_aes_ecb_encrypt+0x18/0x1c [rk_crypto]
[<c028c820>] skcipher_encrypt_ablkcipher+0x64/0x68
[<c0290770>] __test_skcipher+0x2a8/0x8dc
[<c0292e94>] test_skcipher+0x38/0xc4
[<c0292fb0>] alg_test_skcipher+0x90/0xb0
[<c0292158>] alg_test+0x1e8/0x280
[<c028f6f4>] cryptomgr_test+0x34/0x54
[<c004bbe8>] kthread+0xf4/0x10c
[<c0010010>] ret_from_fork+0x14/0x24
irq event stamp: 10672
hardirqs last enabled at (10672): [<c002fac8>] tasklet_action+0x48/0x104
hardirqs last disabled at (10671): [<c002faa0>] tasklet_action+0x20/0x104
softirqs last enabled at (10658): [<c002ef84>] __do_softirq+0x358/0x49c
softirqs last disabled at (10669): [<c002f108>] run_ksoftirqd+0x40/0x80
other info that might help us debug this:
Possible unsafe locking scenario:
CPU0
----
lock(&(&crypto_info->lock)->rlock);
<Interrupt>
lock(&(&crypto_info->lock)->rlock);
*** DEADLOCK ***
Fix this by moving to irq-disabling spinlocks.
Signed-off-by: Heiko Stuebner <heiko@xxxxxxxxx>
---
changes in v2:
- don't use irqsave spinlock in the irq-handler itself
drivers/crypto/rockchip/rk3288_crypto.c | 5 +++--
drivers/crypto/rockchip/rk3288_crypto_ablkcipher.c | 10 ++++++----
2 files changed, 9 insertions(+), 6 deletions(-)
diff --git a/drivers/crypto/rockchip/rk3288_crypto.c b/drivers/crypto/rockchip/rk3288_crypto.c
index 6b72f8d..da9c73d 100644
--- a/drivers/crypto/rockchip/rk3288_crypto.c
+++ b/drivers/crypto/rockchip/rk3288_crypto.c
@@ -190,12 +190,13 @@ static void rk_crypto_tasklet_cb(unsigned long data)
{
struct rk_crypto_info *dev = (struct rk_crypto_info *)data;
struct crypto_async_request *async_req, *backlog;
+ unsigned long flags;
int err = 0;
- spin_lock(&dev->lock);
+ spin_lock_irqsave(&dev->lock, flags);
backlog = crypto_get_backlog(&dev->queue);
async_req = crypto_dequeue_request(&dev->queue);
- spin_unlock(&dev->lock);
+ spin_unlock_irqrestore(&dev->lock, flags);
if (!async_req) {
dev_err(dev->dev, "async_req is NULL !!\n");
return;
diff --git a/drivers/crypto/rockchip/rk3288_crypto_ablkcipher.c b/drivers/crypto/rockchip/rk3288_crypto_ablkcipher.c
index 4a8f9de..d98b681 100644
--- a/drivers/crypto/rockchip/rk3288_crypto_ablkcipher.c
+++ b/drivers/crypto/rockchip/rk3288_crypto_ablkcipher.c
@@ -24,6 +24,7 @@ static void rk_crypto_complete(struct rk_crypto_info *dev, int err)
static int rk_handle_req(struct rk_crypto_info *dev,
struct ablkcipher_request *req)
{
+ unsigned long flags;
int err;
if (!IS_ALIGNED(req->nbytes, dev->align_size))
@@ -38,9 +39,9 @@ static int rk_handle_req(struct rk_crypto_info *dev,
dev->aligned = 1;
dev->ablk_req = req;
- spin_lock(&dev->lock);
+ spin_lock_irqsave(&dev->lock, flags);
err = ablkcipher_enqueue_request(&dev->queue, req);
- spin_unlock(&dev->lock);
+ spin_unlock_irqrestore(&dev->lock, flags);
tasklet_schedule(&dev->crypto_tasklet);
return err;
}
@@ -267,12 +268,13 @@ static int rk_set_data_start(struct rk_crypto_info *dev)
static int rk_ablk_start(struct rk_crypto_info *dev)
{
+ unsigned long flags;
int err;
- spin_lock(&dev->lock);
+ spin_lock_irqsave(&dev->lock, flags);
rk_ablk_hw_init(dev);
err = rk_set_data_start(dev);
- spin_unlock(&dev->lock);
+ spin_unlock_irqrestore(&dev->lock, flags);
return err;
}
--
2.6.1
--
To unsubscribe from this list: send the line "unsubscribe linux-crypto" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
