[PATCH 0/3] net: macb: fix close races (and RX refill error handling)

[PATCH 0/3] net: macb: fix close races (and RX refill error handling)

[Théo Lebrun]

The context-swapping series [0] has been ongoing for a while. The most
interesting part is a proper hardware shutdown sequence, which is truly
lacking in other parts of the MACB driver: at close, at suspend and in
the HRESP error task.

Instead of introducing that sequence for a new feature, apply it now to
fix the main offender, the close path. That fixes the races listed in
patch 3 and will allow the sequence to be reused later.

The first two patches are also fixes, but not related to the shutdown
sequence: they address allocation-failure codepaths. They are sent
alongside patch 3 because they touch the same code and would cause
merge conflicts if applied separately.

Now, let me list issues I'm aware of that we do *not* fix here, to make
the scope explicit:
 - HRESP task should sync with all other contexts. It frees buffers
   under the feet of the whole driver. The hardware shutdown sequence
   will help.
 - Suspend callback is also racy with BH primitives. We'll be able to
   reuse the hardware shutdown sequence.
 - Even if we assume tasks are frozen, the phylink ops aren't and might
   trigger between suspend and resume callbacks. Here we need to
   (1) early return in phylink ops and (2) at resume put the HW in its
   proper state according to phylink ops that occured.
 - Alloc failure codepaths aren't perfect outside open. Hardware and
   ring buffers are left in a sad state. We can probably do better.

[0]: https://lore.kernel.org/all/20260812-macb-context-v9-0-7ddbf5f715e0@bootlin.com/

Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com>
---
Théo Lebrun (3):
      net: macb: never give hardware a NULL RX buffer
      net: macb: propagate RX ring refill errors
      net: macb: quiesce IRQs and drain BH on interface close

 drivers/net/ethernet/cadence/macb.h      |   7 +-
 drivers/net/ethernet/cadence/macb_main.c | 119 ++++++++++++++++++++++++-------
 2 files changed, 98 insertions(+), 28 deletions(-)
---
base-commit: 994db8ab9d90c64dd641b7ead6efe2eaea7a50dc
change-id: 20260918-macb-close-dbbd23b4b8f2

Best regards,
--  
Théo Lebrun <theo.lebrun@bootlin.com>

[PATCH 1/3] net: macb: never give hardware a NULL RX buffer

[Théo Lebrun]

The refill logic is simple: iterate over all pending rx slots, allocate
SKB (& DMA map) if needed and hand it off the to hardware by clearing
the RX_USED flag.

If the refill operation fails mid-way, it early returns leaving the
following slots untouched. In the normal case that is fine, because all
slots have been properly initialised (and might have been already used
by HW meaning they won't be reused). When slots have not been
initialised however, we are in trouble.

After dma_alloc_coherent() of the rx ring buffer, all slots have NULL
pointers and RX_USED cleared meaning HW will try using them. Ensure
this does not happen by setting the RX_USED flag on all slots before
calling refill at buffer alloc, in gem_init_rx_ring(). That way even if
refill fails on an alloc/dma_map, the HW won't try using NULL pointers
as buffers.

Theoretical bugfix, never encountered in practice. To reproduce,
introduce memory pressure (less than 512 SKBs of free memory) and open
the interface.

Note that this codepath also hits at resume, on HRESP errors and on
set_ringparam (while interface is running).

Fixes: 4df95131ea80 ("net/macb: change RX path for GEM")
Cc: stable@vger.kernel.org
Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com>
---
 drivers/net/ethernet/cadence/macb_main.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/drivers/net/ethernet/cadence/macb_main.c b/drivers/net/ethernet/cadence/macb_main.c
index b8234ac4b602..751fa9e68099 100644
--- a/drivers/net/ethernet/cadence/macb_main.c
+++ b/drivers/net/ethernet/cadence/macb_main.c
@@ -2785,9 +2785,14 @@ static int macb_alloc(struct macb *bp)
 
 static void gem_init_rx_ring(struct macb_queue *queue)
 {
+	unsigned int i;
+
 	queue->rx_tail = 0;
 	queue->rx_prepared_head = 0;
 
+	for (i = 0; i < queue->bp->rx_ring_size; i++)
+		macb_rx_desc(queue, i)->addr |= MACB_BIT(RX_USED);
+
 	gem_rx_refill(queue);
 }
 

-- 
2.55.0

[PATCH 2/3] net: macb: propagate RX ring refill errors

[Théo Lebrun]

gem_rx_refill() is responsible for Rx SKB allocation, including at open,
but its prototype indicates a void return value.

Therefore we change the code to propagate allocation and DMA mapping
errors back up the stack, making sure the open fails if it occurs.
Change all those to return errno-style ints:
 - gem_rx_refill()
 - its parent gem_init_rx_ring()
 - its grand-parent gem_init_rings()
 - the macbgem_ops.mog_init_rings function pointer
 - its grand-uncle macb_init_rings()

Theoretical bugfix, never encountered in practice. To reproduce,
introduce memory pressure (less than 512 SKBs of free memory) and open
the interface. I expect the last queue to be unuseable because it has
zero usable rx buffers. Nothing will ever trigger a refill on that
queue which only happens once a frame has been received.

Note that other callers of refill (resume, HRESP error task, NAPI)
cannot do anything useful with that error and keep their best-effort
refill, hoping it will improve.

Fixes: 4df95131ea80 ("net/macb: change RX path for GEM")
Cc: stable@vger.kernel.org
Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com>
---
 drivers/net/ethernet/cadence/macb.h      |  2 +-
 drivers/net/ethernet/cadence/macb_main.c | 30 +++++++++++++++++++++++-------
 2 files changed, 24 insertions(+), 8 deletions(-)

diff --git a/drivers/net/ethernet/cadence/macb.h b/drivers/net/ethernet/cadence/macb.h
index d6931c41f39d..cfaa0ca49f1a 100644
--- a/drivers/net/ethernet/cadence/macb.h
+++ b/drivers/net/ethernet/cadence/macb.h
@@ -1197,7 +1197,7 @@ struct macb_queue;
 struct macb_or_gem_ops {
 	int	(*mog_alloc_rx_buffers)(struct macb *bp);
 	void	(*mog_free_rx_buffers)(struct macb *bp);
-	void	(*mog_init_rings)(struct macb *bp);
+	int	(*mog_init_rings)(struct macb *bp);
 	int	(*mog_rx)(struct macb_queue *queue, struct napi_struct *napi,
 			  int budget);
 };
diff --git a/drivers/net/ethernet/cadence/macb_main.c b/drivers/net/ethernet/cadence/macb_main.c
index 751fa9e68099..c418f859cc34 100644
--- a/drivers/net/ethernet/cadence/macb_main.c
+++ b/drivers/net/ethernet/cadence/macb_main.c
@@ -1486,13 +1486,14 @@ static int macb_tx_complete(struct macb_queue *queue, int budget)
 	return packets;
 }
 
-static void gem_rx_refill(struct macb_queue *queue)
+static int gem_rx_refill(struct macb_queue *queue)
 {
 	struct macb *bp = queue->bp;
 	struct macb_dma_desc *desc;
 	struct sk_buff *skb;
 	unsigned int entry;
 	dma_addr_t paddr;
+	int err = 0;
 
 	while (CIRC_SPACE(queue->rx_prepared_head, queue->rx_tail,
 			bp->rx_ring_size) > 0) {
@@ -1509,6 +1510,7 @@ static void gem_rx_refill(struct macb_queue *queue)
 			if (unlikely(!skb)) {
 				netdev_err(bp->netdev,
 					   "Unable to allocate sk_buff\n");
+				err = -ENOMEM;
 				break;
 			}
 
@@ -1518,6 +1520,7 @@ static void gem_rx_refill(struct macb_queue *queue)
 					       DMA_FROM_DEVICE);
 			if (dma_mapping_error(&bp->pdev->dev, paddr)) {
 				dev_kfree_skb(skb);
+				err = -ENOMEM;
 				break;
 			}
 
@@ -1558,6 +1561,8 @@ static void gem_rx_refill(struct macb_queue *queue)
 
 	netdev_vdbg(bp->netdev, "rx ring: queue: %p, prepared head %d, tail %d\n",
 		    queue, queue->rx_prepared_head, queue->rx_tail);
+
+	return err;
 }
 
 /* Mark DMA descriptors from begin up to and not including end as unused */
@@ -2783,7 +2788,7 @@ static int macb_alloc(struct macb *bp)
 	return -ENOMEM;
 }
 
-static void gem_init_rx_ring(struct macb_queue *queue)
+static int gem_init_rx_ring(struct macb_queue *queue)
 {
 	unsigned int i;
 
@@ -2793,14 +2798,16 @@ static void gem_init_rx_ring(struct macb_queue *queue)
 	for (i = 0; i < queue->bp->rx_ring_size; i++)
 		macb_rx_desc(queue, i)->addr |= MACB_BIT(RX_USED);
 
-	gem_rx_refill(queue);
+	return gem_rx_refill(queue);
 }
 
-static void gem_init_rings(struct macb *bp)
+static int gem_init_rings(struct macb *bp)
 {
 	struct macb_queue *queue;
 	struct macb_dma_desc *desc = NULL;
+	int last_err = 0;
 	unsigned int q;
+	int err;
 	int i;
 
 	for (q = 0, queue = bp->queues; q < bp->num_queues; ++q, ++queue) {
@@ -2813,11 +2820,15 @@ static void gem_init_rings(struct macb *bp)
 		queue->tx_head = 0;
 		queue->tx_tail = 0;
 
-		gem_init_rx_ring(queue);
+		err = gem_init_rx_ring(queue);
+		if (err)
+			last_err = err;
 	}
+
+	return last_err;
 }
 
-static void macb_init_rings(struct macb *bp)
+static int macb_init_rings(struct macb *bp)
 {
 	int i;
 	struct macb_dma_desc *desc = NULL;
@@ -2832,6 +2843,8 @@ static void macb_init_rings(struct macb *bp)
 	bp->queues[0].tx_head = 0;
 	bp->queues[0].tx_tail = 0;
 	desc->ctrl |= MACB_BIT(TX_WRAP);
+
+	return 0;
 }
 
 static void macb_reset_hw(struct macb *bp)
@@ -3162,7 +3175,9 @@ static int macb_open(struct net_device *netdev)
 		goto pm_exit;
 	}
 
-	bp->macbgem_ops.mog_init_rings(bp);
+	err = bp->macbgem_ops.mog_init_rings(bp);
+	if (err)
+		goto free_rings;
 	macb_init_buffers(bp);
 
 	for (q = 0, queue = bp->queues; q < bp->num_queues; ++q, ++queue) {
@@ -3200,6 +3215,7 @@ static int macb_open(struct net_device *netdev)
 		napi_disable(&queue->napi_rx);
 		napi_disable(&queue->napi_tx);
 	}
+free_rings:
 	macb_free(bp);
 pm_exit:
 	pm_runtime_put_sync(&bp->pdev->dev);

-- 
2.55.0

[PATCH 3/3] net: macb: quiesce IRQs and drain BH on interface close

[Théo Lebrun]

The macb_close() operation is facing races as it disables IRQs late in
its sequence and keeps BH primitives alive while shutdown.

Non exhaustive list of races that could occur:

 - macb_tx_error_task() could be scheduled and access the buffers freed
   by macb_close().

 - macb_tx_error_task() or macb_hresp_error_task() might re-enable
   interrupts after the IDR write in macb_close().

 - macb_close() calls napi_disable() meaning that if
   macb_tx_error_task() occurs later, it will deadlock on napi_disable()
   that shouldn't be called if NAPI is already disabled.

 - macb_hresp_error_task() might reinit every RX/TX ring under
   macb_close()'s foot.

 - macb_interrupt() might re-enable NAPI just after it has been
   disabled by macb_close().

Instead, disable all our primitives one by one:
 - (1) mask and sync on IRQ handlers,
 - (2) drain any scheduled bp->hresp_err_bh_work,
 - (3) drain any scheduled queue->tx_error_task,
 - (4) drain queue->napi_rx/napi_tx,
 - (5) drain bp->tx_lpi_work.

Careful! Ordering is important because our scheduling primitives can
wake each other up. Recap table:

|               |   enable/disable   |           schedule            |
|               |----|-------|-------|------|----|----|--------|-----|
|               |IRQs|napi_tx|napi_rx|tx_lpi|napi|napi|tx_error|hresp|
| Context       |    |       |       | task | rx | tx |  task  |task |
|===============|====|=======|=======|======|====|====|========|=====|
| open          | X  |   X   |   X   |      |    |    |        |     |
| link_up       | X  |       |       |      |    |    |        |     |
| link_down     | X  |       |       |      |    |    |        |     |
| close         | X  |   X   |   X   |      |    |    |        |     |
| enable_tx_lpi |    |       |       |  X   |    |    |        |     |
| swap          | X  |   X   |   X   |  X   |    |    |        |     |
| suspend       | X  |   X   |   X   |      |    |    |        |     |
| resume        | X  |   X   |   X   |      |    |    |        |     |
|---------------|----|-------|-------|------|----|----|--------|-----|
| irq & netpoll | X  |       |       |      | X  | X  |   X    | X   |
|---------------|----|-------|-------|------|----|----|--------|-----|
| napi_rx       | X  |       |       |      | X  |    |        |     |
| napi_tx       | X  |       |       |  X   |    | X  |        |     |
|---------------|----|-------|-------|------|----|----|--------|-----|
| tx_error_task | X  |   X   |       |      |    |    |        |     |
| hresp task    | X  |       |       |      |    |    |        |     |

As example, one ordering constraint that can be deduced from the table:
napi_tx can schedule tx_lpi_task meaning napi_tx must be disabled
before tx_lpi_task, else we risk napi_tx re-enabling tx_lpi_task after
it has been disabled by macb_close().

We do *not* use IDR masking to shutdown IRQs because that risks
conflicting with BH primitives we have not disabled yet. For example if
we writel(IDR) in macb_close() and napi_rx is pending then IRQs might
be unmasked by the NAPI poll. As for why we do not use disable_irq():
we will have situations where we are quiesced but want to listen to
some IRQs and (minor reason) we register shared IRQ handlers so we
shouldn't disable the full IRQ line.

Instead we introduce a bool that tells macb_interrupt() to self-disarm.
Its default value is true as we start closed. It gets set to false
while interface is active. Reading into my crystal ball, we'll reuse
that flag in suspend/WOL, set_ringparam and change_mtu (context swap).

Note that old IRQ handler tried preventing a race with close by
self-disarming based on netif_running(). This might work, but it does
not prevent a race with the error codepath of macb_open() which needs
to run with IRQs dis-armed but netif_running() returns true during that
time.

Fixes: e86cd53afc59 ("net/macb: better manage tx errors")
Cc: stable@vger.kernel.org
Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com>
---
 drivers/net/ethernet/cadence/macb.h      |  5 ++
 drivers/net/ethernet/cadence/macb_main.c | 84 ++++++++++++++++++++++++--------
 2 files changed, 69 insertions(+), 20 deletions(-)

diff --git a/drivers/net/ethernet/cadence/macb.h b/drivers/net/ethernet/cadence/macb.h
index cfaa0ca49f1a..1cb2778fe49e 100644
--- a/drivers/net/ethernet/cadence/macb.h
+++ b/drivers/net/ethernet/cadence/macb.h
@@ -1382,6 +1382,11 @@ struct macb {
 	struct delayed_work	tx_lpi_work;
 	u32			tx_lpi_timer;
 
+	/* ISR must not drive NAPI & BH mechanisms. True when the interface
+	 * is closed. Protected by bp->lock.
+	 */
+	bool			irq_quiesced;
+
 	int	rx_bd_rd_prefetch;
 	int	tx_bd_rd_prefetch;
 
diff --git a/drivers/net/ethernet/cadence/macb_main.c b/drivers/net/ethernet/cadence/macb_main.c
index c418f859cc34..75bbde81d62c 100644
--- a/drivers/net/ethernet/cadence/macb_main.c
+++ b/drivers/net/ethernet/cadence/macb_main.c
@@ -1995,6 +1995,53 @@ static int macb_tx_poll(struct napi_struct *napi, int budget)
 	return work_done;
 }
 
+static void macb_quiesce_start(struct macb *bp)
+{
+	struct macb_queue *queue;
+	unsigned long flags;
+	unsigned int q;
+
+	spin_lock_irqsave(&bp->lock, flags);
+	bp->irq_quiesced = true;
+	spin_unlock_irqrestore(&bp->lock, flags);
+
+	for (q = 0, queue = bp->queues; q < bp->num_queues; ++q, ++queue)
+		synchronize_irq(queue->irq);
+
+	cancel_work_sync(&bp->hresp_err_bh_work);
+
+	for (q = 0, queue = bp->queues; q < bp->num_queues; ++q, ++queue) {
+		/* Must be done before NAPI is disabled: the task ends with a
+		 * napi_enable() call.
+		 */
+		cancel_work_sync(&queue->tx_error_task);
+
+		napi_disable(&queue->napi_rx);
+		napi_disable(&queue->napi_tx);
+	}
+
+	/* Must be done after napi_tx is disabled: its completion re-arms
+	 * the LPI timer.
+	 */
+	cancel_delayed_work_sync(&bp->tx_lpi_work);
+}
+
+static void macb_quiesce_end(struct macb *bp)
+{
+	struct macb_queue *queue;
+	unsigned long flags;
+	unsigned int q;
+
+	for (q = 0, queue = bp->queues; q < bp->num_queues; ++q, ++queue) {
+		napi_enable(&queue->napi_rx);
+		napi_enable(&queue->napi_tx);
+	}
+
+	spin_lock_irqsave(&bp->lock, flags);
+	bp->irq_quiesced = false;
+	spin_unlock_irqrestore(&bp->lock, flags);
+}
+
 static void macb_hresp_error_task(struct work_struct *work)
 {
 	struct macb *bp = from_work(bp, work, hresp_err_bh_work);
@@ -2137,8 +2184,8 @@ static irqreturn_t macb_interrupt(int irq, void *dev_id)
 	spin_lock(&bp->lock);
 
 	while (status) {
-		/* close possible race with dev_close */
-		if (unlikely(!netif_running(netdev))) {
+		/* self-disarm while the netdev is closed */
+		if (unlikely(bp->irq_quiesced)) {
 			queue_writel(queue, IDR, -1);
 			macb_queue_isr_clear(bp, queue, -1);
 			break;
@@ -3155,8 +3202,6 @@ static int macb_open(struct net_device *netdev)
 {
 	size_t bufsz = netdev->mtu + ETH_HLEN + ETH_FCS_LEN + NET_IP_ALIGN;
 	struct macb *bp = netdev_priv(netdev);
-	struct macb_queue *queue;
-	unsigned int q;
 	int err;
 
 	netdev_dbg(bp->netdev, "open\n");
@@ -3180,10 +3225,7 @@ static int macb_open(struct net_device *netdev)
 		goto free_rings;
 	macb_init_buffers(bp);
 
-	for (q = 0, queue = bp->queues; q < bp->num_queues; ++q, ++queue) {
-		napi_enable(&queue->napi_rx);
-		napi_enable(&queue->napi_tx);
-	}
+	macb_quiesce_end(bp);
 
 	macb_init_hw(bp);
 
@@ -3210,11 +3252,10 @@ static int macb_open(struct net_device *netdev)
 	phy_power_off(bp->phy);
 
 reset_hw:
+	/* The netdev stays down: quiesce and drain, as macb_close() does. */
+	macb_quiesce_start(bp);
+
 	macb_reset_hw(bp);
-	for (q = 0, queue = bp->queues; q < bp->num_queues; ++q, ++queue) {
-		napi_disable(&queue->napi_rx);
-		napi_disable(&queue->napi_tx);
-	}
 free_rings:
 	macb_free(bp);
 pm_exit:
@@ -3225,19 +3266,17 @@ static int macb_open(struct net_device *netdev)
 static int macb_close(struct net_device *netdev)
 {
 	struct macb *bp = netdev_priv(netdev);
-	struct macb_queue *queue;
 	unsigned long flags;
 	unsigned int q;
 
+	macb_quiesce_start(bp);
+
+	/* Drain the BH contexts before stopping the queues: NAPI completion
+	 * and tx_error_task wake them up.
+	 */
 	netif_tx_stop_all_queues(netdev);
-
-	for (q = 0, queue = bp->queues; q < bp->num_queues; ++q, ++queue) {
-		napi_disable(&queue->napi_rx);
-		napi_disable(&queue->napi_tx);
+	for (q = 0; q < bp->num_queues; ++q)
 		netdev_tx_reset_queue(netdev_get_tx_queue(netdev, q));
-	}
-
-	cancel_delayed_work_sync(&bp->tx_lpi_work);
 
 	phylink_stop(bp->phylink);
 	phylink_disconnect_phy(bp->phylink);
@@ -4759,6 +4798,11 @@ static int macb_init_dflt(struct platform_device *pdev)
 	bp->tx_ring_size = DEFAULT_TX_RING_SIZE;
 	bp->rx_ring_size = DEFAULT_RX_RING_SIZE;
 
+	/* No locking needed because the IRQs are not requested yet. The
+	 * flag is cleared by macb_open() and re-armed by macb_close().
+	 */
+	bp->irq_quiesced = true;
+
 	/* set the queue register mapping once for all: queue0 has a special
 	 * register mapping but we don't want to test the queue index then
 	 * compute the corresponding register offset at run time.

-- 
2.55.0