On Wed Sep 10 2025, Kohei Enju wrote:
> + Aleksandr
>
> On Wed, 10 Sep 2025 10:28:17 +0300, Lifshits, Vitaly wrote:
>
>>On 9/8/2025 9:26 AM, Kurt Kanzenbach wrote:
>>> On Sat Sep 06 2025, Kohei Enju wrote:
>>>> Currently igc_probe() doesn't unregister netdev when igc_led_setup()
>>>> fails, causing BUG_ON() in free_netdev() and then kernel panics. [1]
>>>>
>>>> This behavior can be tested using fault-injection framework. I used the
>>>> failslab feature to test the issue. [2]
>>>>
>>>> Call unregister_netdev() when igc_led_setup() fails to avoid the kernel
>>>> panic.
>>>>
>>>> [1]
>>>> kernel BUG at net/core/dev.c:12047!
>>>> Oops: invalid opcode: 0000 [#1] SMP NOPTI
>>>> CPU: 0 UID: 0 PID: 937 Comm: repro-igc-led-e Not tainted
>>>> 6.17.0-rc4-enjuk-tnguy-00865-gc4940196ab02 #64 PREEMPT(voluntary)
>>>> Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS
>>>> 1.16.3-debian-1.16.3-2 04/01/2014
>>>> RIP: 0010:free_netdev+0x278/0x2b0
>>>> [...]
>>>> Call Trace:
>>>> <TASK>
>>>> igc_probe+0x370/0x910
>>>> local_pci_probe+0x3a/0x80
>>>> pci_device_probe+0xd1/0x200
>>>> [...]
>>>>
>>>> [2]
>>>> #!/bin/bash -ex
>>>>
>>>> FAILSLAB_PATH=/sys/kernel/debug/failslab/
>>>> DEVICE=0000:00:05.0
>>>> START_ADDR=$(grep " igc_led_setup" /proc/kallsyms \
>>>> | awk '{printf("0x%s", $1)}')
>>>> END_ADDR=$(printf "0x%x" $((START_ADDR + 0x100)))
>>>>
>>>> echo $START_ADDR > $FAILSLAB_PATH/require-start
>>>> echo $END_ADDR > $FAILSLAB_PATH/require-end
>>>> echo 1 > $FAILSLAB_PATH/times
>>>> echo 100 > $FAILSLAB_PATH/probability
>>>> echo N > $FAILSLAB_PATH/ignore-gfp-wait
>>>>
>>>> echo $DEVICE > /sys/bus/pci/drivers/igc/bind
>>>>
>>>> Fixes: ea578703b03d ("igc: Add support for LEDs on i225/i226")
>>>> Signed-off-by: Kohei Enju <[email protected]>
>>>
>>> Reviewed-by: Kurt Kanzenbach <[email protected]>
>>
>>Thank you for the patch and for identifying this issue!
>>
>>I was wondering whether we could avoid failing the probe in cases where
>>igc_led_setup fails. It seems to me that a failure in the LED class
>>functionality shouldn't prevent the device's core functionality from
>>working properly.
>
> Indeed, that also makes sense.
>
> The behavior that igc_probe() succeeds even if igc_led_setup() fails
> also seems good to me, as long as notifying users that igc's led
> functionality is not available.SGTM. The LED code is nice to have, but not mandatory at all. The device has sane LED defaults. > >> >> From what I understand, errors in this function are not due to hardware >>malfunctions. Therefore, I suggest we remove the error propagation. >> >>Alternatively, if feasible, we could consider reordering the function >>calls so that the LED class setup occurs before the netdev registration. >> > > I don't disagree with you, but I would like to hear Kurt and Aleksandr's > opinion. Do you have any preference or suggestions? See above. Thanks, Kurt
signature.asc
Description: PGP signature
