From: Jake Hamby <Jake.Hamby@Teledyne.com>

If an interrupt (RX-complete or error flag) is set when bringing up
the CAN device, e.g. due to CAN bus traffic before initializing the
device, when m_can_start() is called and interrupts are enabled,
m_can_isr() is called immediately, which disables all CAN interrupts
and calls napi_schedule().

Because napi_enable() isn't called until later in m_can_open(), the
call to napi_schedule() never schedules the m_can_poll() callback and
the device is left with interrupts disabled and can't receive any CAN
packets until rebooted.

This can be verified by running "cansend" from another device before
setting the bitrate and calling "ip link set up can0" on the test
device. Adding debug lines to m_can_isr() shows it's called with flags
(IR_EP | IR_EW | IR_CRCE), which calls m_can_disable_all_interrupts()
and napi_schedule(), and then m_can_poll() is never called.

Move the call to napi_enable() above the call to m_can_start() to
enable any initial interrupt flags to be handled by m_can_poll() so
that interrupts are reenabled. Add a call to napi_disable() in the
error handling section of m_can_open(), to handle the case where later
functions return errors.

Also, in m_can_close(), move the call to napi_disable() below the call
to m_can_stop() to ensure all interrupts are handled when bringing
down the device. This race condition is much less likely to occur.

Tested on a Microchip SAMA7G54 MPU. The fix should be applicable to
any SoC with a Bosch M_CAN controller.

Signed-off-by: Jake Hamby <Jake.Hamby@Teledyne.com>
Fixes: e0d1f4816f2a ("can: m_can: add Bosch M_CAN controller support")
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
---
 drivers/net/can/m_can/m_can.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/drivers/net/can/m_can/m_can.c b/drivers/net/can/m_can/m_can.c
index 012c3d22b01dd3d8558f2a40448770ca1da1aa1e..c1a07013433eb7b863eee072b959f46c1d5b008d 100644
--- a/drivers/net/can/m_can/m_can.c
+++ b/drivers/net/can/m_can/m_can.c
@@ -1763,9 +1763,6 @@ static int m_can_close(struct net_device *dev)
 
 	netif_stop_queue(dev);
 
-	if (!cdev->is_peripheral)
-		napi_disable(&cdev->napi);
-
 	m_can_stop(dev);
 	m_can_clk_stop(cdev);
 	free_irq(dev->irq, dev);
@@ -1776,6 +1773,8 @@ static int m_can_close(struct net_device *dev)
 		destroy_workqueue(cdev->tx_wq);
 		cdev->tx_wq = NULL;
 		can_rx_offload_disable(&cdev->offload);
+	} else {
+		napi_disable(&cdev->napi);
 	}
 
 	close_candev(dev);
@@ -2030,6 +2029,8 @@ static int m_can_open(struct net_device *dev)
 
 	if (cdev->is_peripheral)
 		can_rx_offload_enable(&cdev->offload);
+	else
+		napi_enable(&cdev->napi);
 
 	/* register interrupt handler */
 	if (cdev->is_peripheral) {
@@ -2063,9 +2064,6 @@ static int m_can_open(struct net_device *dev)
 	if (err)
 		goto exit_start_fail;
 
-	if (!cdev->is_peripheral)
-		napi_enable(&cdev->napi);
-
 	netif_start_queue(dev);
 
 	return 0;
@@ -2079,6 +2077,8 @@ static int m_can_open(struct net_device *dev)
 out_wq_fail:
 	if (cdev->is_peripheral)
 		can_rx_offload_disable(&cdev->offload);
+	else
+		napi_disable(&cdev->napi);
 	close_candev(dev);
 exit_disable_clks:
 	m_can_clk_stop(cdev);

-- 
2.45.2

On 10.09.2024 19:15:28, Marc Kleine-Budde wrote:
> From: Jake Hamby <Jake.Hamby@Teledyne.com>
> 
> If an interrupt (RX-complete or error flag) is set when bringing up
> the CAN device, e.g. due to CAN bus traffic before initializing the
> device, when m_can_start() is called and interrupts are enabled,
> m_can_isr() is called immediately, which disables all CAN interrupts
> and calls napi_schedule().
> 
> Because napi_enable() isn't called until later in m_can_open(), the
> call to napi_schedule() never schedules the m_can_poll() callback and
> the device is left with interrupts disabled and can't receive any CAN
> packets until rebooted.
> 
> This can be verified by running "cansend" from another device before
> setting the bitrate and calling "ip link set up can0" on the test
> device. Adding debug lines to m_can_isr() shows it's called with flags
> (IR_EP | IR_EW | IR_CRCE), which calls m_can_disable_all_interrupts()
> and napi_schedule(), and then m_can_poll() is never called.
> 
> Move the call to napi_enable() above the call to m_can_start() to
> enable any initial interrupt flags to be handled by m_can_poll() so
> that interrupts are reenabled. Add a call to napi_disable() in the
> error handling section of m_can_open(), to handle the case where later
> functions return errors.
> 
> Also, in m_can_close(), move the call to napi_disable() below the call
> to m_can_stop() to ensure all interrupts are handled when bringing
> down the device. This race condition is much less likely to occur.
> 
> Tested on a Microchip SAMA7G54 MPU. The fix should be applicable to
> any SoC with a Bosch M_CAN controller.
> 
> Signed-off-by: Jake Hamby <Jake.Hamby@Teledyne.com>
> Fixes: e0d1f4816f2a ("can: m_can: add Bosch M_CAN controller support")
> Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

Tested successfully on a stm32mp1.

Tested-by: Marc Kleine-Budde <mkl@pengutronix.de>

regards,
Mar
c
-- 
Pengutronix e.K.                 | Marc Kleine-Budde          |
Embedded Linux                   | https://www.pengutronix.de |
Vertretung Nürnberg              | Phone: +49-5121-206917-129 |
Amtsgericht Hildesheim, HRA 2686 | Fax:   +49-5121-206917-9   |