Skip to contentSkip to conversation

FAILED: patch "[PATCH] platform/mellanox: fix potential deadlock in the tmfifo" failed to apply to 5.5-stable tree

PATCH [email protected] 2 participants 2 messages Feb 9, 2020

Subsystems (2)
lore
Share & tools
lore

[email protected]

author maintainer
#0 lore raw

The patch below does not apply to the 5.5-stable tree.
If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <[email protected]>.

thanks,

greg k-h

------------------ original commit in Linus's tree ------------------

From 3454eeeebd115891f34aa2e76eccf08c9b0882bb Mon Sep 17 00:00:00 2001
From: Liming Sun <[email protected]>
Date: Fri, 20 Dec 2019 12:04:33 -0500
Subject: [PATCH] platform/mellanox: fix potential deadlock in the tmfifo
driver

This commit fixes the potential deadlock caused by the console Rx and Tx processing at the same time. Rx and Tx both take the console and tmfifo spinlock but in different order which causes potential deadlock. The fix is to use different tmfifo spinlock for Rx and Tx since they protect different resources and it's safe to split the lock.

Below is the reported call trace when copying/pasting large string in the console.

Rx:
_raw_spin_lock_irqsave (hvc lock)
__hvc_poll
hvc_poll
in_intr
vring_interrupt
mlxbf_tmfifo_rxtx_one_desc (tmfifo lock)
mlxbf_tmfifo_rxtx
mlxbf_tmfifo_work_rxtx
Tx:
_raw_spin_lock_irqsave (tmfifo lock)
mlxbf_tmfifo_virtio_notify
virtqueue_notify
virtqueue_kick
put_chars
hvc_push
hvc_write (hvc lock)
...
do_tty_write
tty_write

Fixes 1357dfd7261f ("platform/mellanox: Add TmFifo driver for Mellanox BlueField Soc")
Cc <[email protected]> # 5.4+
drivers/platform/mellanox/mlxbf-tmfifo.c +10 -9
index 9a5c9fd2dbc6..5739a9669b29 100644
@@ -149,7 +149,7 @@ struct mlxbf_tmfifo_irq_info {
  * @work: work struct for deferred process
  * @timer: background timer
  * @vring: Tx/Rx ring
- * @spin_lock: spin lock
+ * @spin_lock: Tx/Rx spin lock
  * @is_ready: ready flag
  */
 struct mlxbf_tmfifo {
@@ -164,7 +164,7 @@ struct mlxbf_tmfifo {
 	struct work_struct work;
 	struct timer_list timer;
 	struct mlxbf_tmfifo_vring *vring[2];
-	spinlock_t spin_lock;		/* spin lock */
+	spinlock_t spin_lock[2];	/* spin lock */
 	bool is_ready;
 };
 
@@ -525,7 +525,7 @@ static void mlxbf_tmfifo_console_tx(struct mlxbf_tmfifo *fifo, int avail)
 	writeq(*(u64 *)&hdr, fifo->tx_base + MLXBF_TMFIFO_TX_DATA);
 
 	/* Use spin-lock to protect the 'cons->tx_buf'. */
-	spin_lock_irqsave(&fifo->spin_lock, flags);
+	spin_lock_irqsave(&fifo->spin_lock[0], flags);
 
 	while (size > 0) {
 		addr = cons->tx_buf.buf + cons->tx_buf.tail;
@@ -552,7 +552,7 @@ static void mlxbf_tmfifo_console_tx(struct mlxbf_tmfifo *fifo, int avail)
 		}
 	}
 
-	spin_unlock_irqrestore(&fifo->spin_lock, flags);
+	spin_unlock_irqrestore(&fifo->spin_lock[0], flags);
 }
 
 /* Rx/Tx one word in the descriptor buffer. */
@@ -731,9 +731,9 @@ static bool mlxbf_tmfifo_rxtx_one_desc(struct mlxbf_tmfifo_vring *vring,
 		fifo->vring[is_rx] = NULL;
 
 		/* Notify upper layer that packet is done. */
-		spin_lock_irqsave(&fifo->spin_lock, flags);
+		spin_lock_irqsave(&fifo->spin_lock[is_rx], flags);
 		vring_interrupt(0, vring->vq);
-		spin_unlock_irqrestore(&fifo->spin_lock, flags);
+		spin_unlock_irqrestore(&fifo->spin_lock[is_rx], flags);
 	}
 
 mlxbf_tmfifo_desc_done:
@@ -852,10 +852,10 @@ static bool mlxbf_tmfifo_virtio_notify(struct virtqueue *vq)
 		 * worker handler.
 		 */
 		if (vring->vdev_id == VIRTIO_ID_CONSOLE) {
-			spin_lock_irqsave(&fifo->spin_lock, flags);
+			spin_lock_irqsave(&fifo->spin_lock[0], flags);
 			tm_vdev = fifo->vdev[VIRTIO_ID_CONSOLE];
 			mlxbf_tmfifo_console_output(tm_vdev, vring);
-			spin_unlock_irqrestore(&fifo->spin_lock, flags);
+			spin_unlock_irqrestore(&fifo->spin_lock[0], flags);
 		} else if (test_and_set_bit(MLXBF_TM_TX_LWM_IRQ,
 					    &fifo->pend_events)) {
 			return true;
@@ -1189,7 +1189,8 @@ static int mlxbf_tmfifo_probe(struct platform_device *pdev)
 	if (!fifo)
 		return -ENOMEM;
 
-	spin_lock_init(&fifo->spin_lock);
+	spin_lock_init(&fifo->spin_lock[0]);
+	spin_lock_init(&fifo->spin_lock[1]);
 	INIT_WORK(&fifo->work, mlxbf_tmfifo_work_handler);
 	mutex_init(&fifo->lock);
 
maintainer
#1 lore raw
Quoted 49 lines · On Sun, Feb 09, 2020 at 12:39:11PM +0100, [email protected] wrote:

The patch below does not apply to the 5.5-stable tree.
If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <[email protected]>.

thanks,

greg k-h

------------------ original commit in Linus's tree ------------------

From 3454eeeebd115891f34aa2e76eccf08c9b0882bb Mon Sep 17 00:00:00 2001
From: Liming Sun <[email protected]>
Date: Fri, 20 Dec 2019 12:04:33 -0500
Subject: [PATCH] platform/mellanox: fix potential deadlock in the tmfifo driver

This commit fixes the potential deadlock caused by the console Rx and Tx processing at the same time. Rx and Tx both take the console and tmfifo spinlock but in different order which causes potential deadlock. The fix is to use different tmfifo spinlock for Rx and Tx since they protect different resources and it's safe to split the lock.

Below is the reported call trace when copying/pasting large string in the console.

Rx:
_raw_spin_lock_irqsave (hvc lock)
__hvc_poll
hvc_poll
in_intr
vring_interrupt
mlxbf_tmfifo_rxtx_one_desc (tmfifo lock)
mlxbf_tmfifo_rxtx
mlxbf_tmfifo_work_rxtx
Tx:
_raw_spin_lock_irqsave (tmfifo lock)
mlxbf_tmfifo_virtio_notify
virtqueue_notify
virtqueue_kick
put_chars
hvc_push
hvc_write (hvc lock)
...
do_tty_write
tty_write

Quoted 5 lines
Fixes 1357dfd7261f ("platform/mellanox: Add TmFifo driver for Mellanox BlueField Soc")
Cc <[email protected]> # 5.4+

It looks to me like this commit went in through two different trees:

638bc4ca3d28 ("platform/mellanox: fix potential deadlock in the tmfifo driver") 3454eeeebd11 ("platform/mellanox: fix potential deadlock in the tmfifo driver")

They are identical, so one probably got cleaned up in a merge commit.

Anyway, 3454eeeebd11 is in both 5.4 and 5.5, so nothing to do here.

Signature · 2 lines
Thanks,
Sasha