| Message ID | 20220810144536.389237-1-u.kleine-koenig@pengutronix.de (mailing list archive) |
|---|---|
| State | Awaiting Upstream |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | can: rx-offload: Break loop on queue full | expand |
| Context | Check | Description |
|---|---|---|
| netdev/tree_selection | success | Series ignored based on subject |
On 10.08.2022 16:45:36, Uwe Kleine-König wrote: > The following happend on an i.MX25 using flexcan with many packets on > the bus: > > The rx-offload queue reached a length more than skb_queue_len_max. So in > can_rx_offload_offload_one() the drop variable was set to true which > made the call to .mailbox_read() (here: flexcan_mailbox_read()) just > return ERR_PTR(-ENOBUFS) (plus some irrelevant hardware interaction) and > so can_rx_offload_offload_one() returned ERR_PTR(-ENOBUFS), too. > > Now can_rx_offload_irq_offload_fifo() looks as follows: > > while (1) { > skb = can_rx_offload_offload_one(offload, 0); > if (IS_ERR(skb)) > continue; > ... > } > > As the i.MX25 is a single core CPU while the rx-offload processing is > active there is no thread to process packets from the offload queue and > so it doesn't get shorter. > > The result is a tight loop: can_rx_offload_offload_one() does nothing > relevant and returns an error code and so > can_rx_offload_irq_offload_fifo() calls can_rx_offload_offload_one() > again. > > To break that loop don't continue calling can_rx_offload_offload_one() > after it reported an error. > > Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de> > --- > Hello, > > this patch just implements the obvious change to break the loop. I'm not > 100% certain that there is no corner case where the break is wrong. The > problem exists at least since v5.6, didn't go back further to check. > > This fixes a hard hang on said i.MX25. As Uwe suggested in an IRC conversation, the correct fix for the flexcan driver is to return NULL if there is no CAN frame pending. I'll send a -v2. Marc
On 11.08.2022 10:30:39, Marc Kleine-Budde wrote: > On 10.08.2022 16:45:36, Uwe Kleine-König wrote: > > The following happend on an i.MX25 using flexcan with many packets on > > the bus: > > > > The rx-offload queue reached a length more than skb_queue_len_max. So in > > can_rx_offload_offload_one() the drop variable was set to true which > > made the call to .mailbox_read() (here: flexcan_mailbox_read()) just > > return ERR_PTR(-ENOBUFS) (plus some irrelevant hardware interaction) and > > so can_rx_offload_offload_one() returned ERR_PTR(-ENOBUFS), too. > > > > Now can_rx_offload_irq_offload_fifo() looks as follows: > > > > while (1) { > > skb = can_rx_offload_offload_one(offload, 0); > > if (IS_ERR(skb)) > > continue; > > ... > > } > > > > As the i.MX25 is a single core CPU while the rx-offload processing is > > active there is no thread to process packets from the offload queue and > > so it doesn't get shorter. > > > > The result is a tight loop: can_rx_offload_offload_one() does nothing > > relevant and returns an error code and so > > can_rx_offload_irq_offload_fifo() calls can_rx_offload_offload_one() > > again. > > > > To break that loop don't continue calling can_rx_offload_offload_one() > > after it reported an error. > > > > Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de> > > --- > > Hello, > > > > this patch just implements the obvious change to break the loop. I'm not > > 100% certain that there is no corner case where the break is wrong. The > > problem exists at least since v5.6, didn't go back further to check. > > > > This fixes a hard hang on said i.MX25. > > As Uwe suggested in an IRC conversation, the correct fix for the flexcan > driver is to return NULL if there is no CAN frame pending. > > I'll send a -v2. https://lore.kernel.org/all/20220811094254.1864367-1-mkl@pengutronix.de regards, Marc
diff --git a/drivers/net/can/dev/rx-offload.c b/drivers/net/can/dev/rx-offload.c index a32a01c172d4..d5d33692bb6a 100644 --- a/drivers/net/can/dev/rx-offload.c +++ b/drivers/net/can/dev/rx-offload.c @@ -207,7 +207,7 @@ int can_rx_offload_irq_offload_fifo(struct can_rx_offload *offload) while (1) { skb = can_rx_offload_offload_one(offload, 0); if (IS_ERR(skb)) - continue; + break; if (!skb) break;
The following happend on an i.MX25 using flexcan with many packets on the bus: The rx-offload queue reached a length more than skb_queue_len_max. So in can_rx_offload_offload_one() the drop variable was set to true which made the call to .mailbox_read() (here: flexcan_mailbox_read()) just return ERR_PTR(-ENOBUFS) (plus some irrelevant hardware interaction) and so can_rx_offload_offload_one() returned ERR_PTR(-ENOBUFS), too. Now can_rx_offload_irq_offload_fifo() looks as follows: while (1) { skb = can_rx_offload_offload_one(offload, 0); if (IS_ERR(skb)) continue; ... } As the i.MX25 is a single core CPU while the rx-offload processing is active there is no thread to process packets from the offload queue and so it doesn't get shorter. The result is a tight loop: can_rx_offload_offload_one() does nothing relevant and returns an error code and so can_rx_offload_irq_offload_fifo() calls can_rx_offload_offload_one() again. To break that loop don't continue calling can_rx_offload_offload_one() after it reported an error. Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de> --- Hello, this patch just implements the obvious change to break the loop. I'm not 100% certain that there is no corner case where the break is wrong. The problem exists at least since v5.6, didn't go back further to check. This fixes a hard hang on said i.MX25. Best regards Uwe drivers/net/can/dev/rx-offload.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)