Thread (22 messages) flat view 22 messages, 1 author, 12h ago
HOTtoday

[PATCH net-next v8 14/17] net: macb: move printk() calls out of bp->lock critical section

From: Théo Lebrun <theo.lebrun@bootlin.com>
Date: 2026-08-05 17:43:58
Also in: lkml
Subsystem: atmel macb ethernet driver, networking drivers, the rest · Maintainers: Théo Lebrun, Andrew Lunn, "David S. Miller", Eric Dumazet, Jakub Kicinski, Paolo Abeni, Linus Torvalds

printk() while bp->lock is acquired is dangerous if netconsole is active
on the interface. In that setup, we might land in macb_poll_controller()
-> macb_interrupt() -> spin_lock(&bp->lock) but bp->lock is already
acquired.

This is not an issue currently because macb_interrupt() first checks IRQ
status, potentially early returns, then grabs lock. This early exit is
expected in netpoll scenario.

However if we came to reading the status inside the bp->lock critical
section (as it should be to avoid races), then this would turn into a
deadlock. And we will.

Solution: move printk() calls out of the critical section, to ensure we
can never netpoll under bp->lock's reign. Added benefit is a smaller
and simpler atomic section.

--

In macb_tx_error_task(), defer netdev_err("halt tx timed out") call to
after the critical section. Update the message to highlight it occurred
in the past. Inherit the buffer exhaustion boolean variable name from
the old code comment.

In macb_tx_error_task(), defer the netdev_err("TX buffers exhausted
mid-frame") call out of the loop. This also means it goes from
1-per-error to 1-per-task-invocation.

In IRQ handling, move HRESP error printing out of macb_interrupt_misc()
into macb_interrupt(). Again, it means we dedup error reporting if
status is read multiple times in a row with HRESP bit set. This is fine
as from past instances I've seen, this message spams our log if it
occurs.

Notice we *ignore* debug printks; if you are debugging MACB maybe don't
use netconsole...

Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com>
---
 drivers/net/ethernet/cadence/macb_main.c | 23 ++++++++++++++++++-----
 1 file changed, 18 insertions(+), 5 deletions(-)
diff --git a/drivers/net/ethernet/cadence/macb_main.c b/drivers/net/ethernet/cadence/macb_main.c
index 5a4eb87f5a97..4bf55994420f 100644
--- a/drivers/net/ethernet/cadence/macb_main.c
+++ b/drivers/net/ethernet/cadence/macb_main.c
@@ -1303,6 +1303,7 @@ static void macb_tx_error_task(struct work_struct *work)
 	struct macb_tx_skb *tx_skb;
 	struct macb_dma_desc *desc;
 	bool halt_timeout = false;
+	bool buggy_driver = false;
 	struct sk_buff *skb;
 	unsigned long flags;
 	unsigned int tail;
@@ -1329,7 +1330,6 @@ static void macb_tx_error_task(struct work_struct *work)
 	 * macb/gem must be halted to write TBQP register
 	 */
 	if (macb_halt_tx(bp)) {
-		netdev_err(bp->netdev, "BUG: halt tx timed out\n");
 		macb_writel(bp, NCR, macb_readl(bp, NCR) & (~MACB_BIT(TE)));
 		halt_timeout = true;
 	}
@@ -1374,8 +1374,7 @@ static void macb_tx_error_task(struct work_struct *work)
 			 * those. Statistics are updated by hardware.
 			 */
 			if (ctrl & MACB_BIT(TX_BUF_EXHAUSTED))
-				netdev_err(bp->netdev,
-					   "BUG: TX buffers exhausted mid-frame\n");
+				buggy_driver = true;
 
 			desc->ctrl = ctrl | MACB_BIT(TX_USED);
 		}
@@ -1412,7 +1411,14 @@ static void macb_tx_error_task(struct work_struct *work)
 	macb_writel(bp, NCR, macb_readl(bp, NCR) | MACB_BIT(TSTART));
 
 	spin_unlock_irqrestore(&bp->lock, flags);
+
 	napi_enable(&queue->napi_tx);
+
+	if (halt_timeout)
+		netdev_err(bp->netdev, "BUG: halt tx timed out, we ignored it\n");
+
+	if (buggy_driver)
+		netdev_err(bp->netdev, "BUG: TX buffers exhausted mid-frame\n");
 }
 
 static bool ptp_one_step_sync(struct sk_buff *skb)
@@ -2165,7 +2171,6 @@ static int macb_interrupt_misc(struct macb_queue *queue, u32 status)
 
 	if (status & MACB_BIT(HRESP)) {
 		queue_work(system_bh_wq, &bp->hresp_err_bh_work);
-		netdev_err(netdev, "DMA bus error: HRESP not OK\n");
 		macb_queue_isr_clear(bp, queue, MACB_BIT(HRESP));
 	}
 
@@ -2185,6 +2190,7 @@ static irqreturn_t macb_interrupt(int irq, void *dev_id)
 	struct macb_queue *queue = dev_id;
 	struct macb *bp = queue->bp;
 	struct net_device *netdev = bp->netdev;
+	bool hresp_err = false;
 	u32 status;
 
 	status = queue_readl(queue, ISR);
@@ -2231,15 +2237,22 @@ static irqreturn_t macb_interrupt(int irq, void *dev_id)
 			napi_schedule_irqoff(&queue->napi_tx);
 		}
 
-		if (unlikely(status & MACB_INT_MISC_FLAGS))
+		if (unlikely(status & MACB_INT_MISC_FLAGS)) {
 			if (macb_interrupt_misc(queue, status))
 				break;
 
+			if (status & MACB_BIT(HRESP))
+				hresp_err = true;
+		}
+
 		status = queue_readl(queue, ISR);
 	}
 
 	spin_unlock(&bp->lock);
 
+	if (hresp_err)
+		netdev_err(netdev, "DMA bus error: HRESP not OK\n");
+
 	return IRQ_HANDLED;
 }
 
-- 
2.55.0
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help