[bug report] rdma_rxe module unload failure with DEBUG_LOCK_ALLOC enabled

[bug report] rdma_rxe module unload failure with DEBUG_LOCK_ALLOC enabled

[Shinichiro Kawasaki]

Hello all,

While I evaluate v6.19-rc1 kernel, I found that rdma_rxe module unload fails.
The failure can be recreated by simple two commands below:

   $ sudo modprobe rdma_rxe
   $ sudo modprobe -r rdma_rxe
   modprobe: FATAL: Module rdma_rxe is in use.

I bisected and found the trigger commit is this:

   80a85a771deb ("RDMA/rxe: reclassify sockets in order to avoid false positives from lockdep")

This commit changes the driver behavior when the kconfig DEBUG_LOCK_ALLOC is
enabled, and my kconfig does so.

Actions for fix will be appreciated.

Re: [bug report] rdma_rxe module unload failure with DEBUG_LOCK_ALLOC enabled

[Stefan Metzmacher]

Hi Shinichiro,

> While I evaluate v6.19-rc1 kernel, I found that rdma_rxe module unload fails.
> The failure can be recreated by simple two commands below:
> 
>     $ sudo modprobe rdma_rxe
>     $ sudo modprobe -r rdma_rxe
>     modprobe: FATAL: Module rdma_rxe is in use.
> 
> I bisected and found the trigger commit is this:
> 
>     80a85a771deb ("RDMA/rxe: reclassify sockets in order to avoid false positives from lockdep")
> 
> This commit changes the driver behavior when the kconfig DEBUG_LOCK_ALLOC is
> enabled, and my kconfig does so.
> 
> Actions for fix will be appreciated.

I have a fix for this, see:
https://git.samba.org/?p=metze/linux/wip.git;a=commitdiff;h=7f55eb3373dca97b706e8521705a06d4bf84b0f0

I'll post it to the list later today.

Sorry for the trouble!
metze

Re: [bug report] rdma_rxe module unload failure with DEBUG_LOCK_ALLOC enabled

[yanjun.zhu]

On 12/19/25 5:45 AM, Stefan Metzmacher wrote:
> Hi Shinichiro,
> 
>> While I evaluate v6.19-rc1 kernel, I found that rdma_rxe module unload 
>> fails.
>> The failure can be recreated by simple two commands below:
>>
>>     $ sudo modprobe rdma_rxe
>>     $ sudo modprobe -r rdma_rxe
>>     modprobe: FATAL: Module rdma_rxe is in use.
>>
>> I bisected and found the trigger commit is this:
>>
>>     80a85a771deb ("RDMA/rxe: reclassify sockets in order to avoid 
>> false positives from lockdep")

Thanks a lot.

https://git.samba.org/?p=metze/linux/wip.git;a=commitdiff;h=7f55eb3373dca97b706e8521705a06d4bf84b0f0

Hi, Shinichiro Kawasaki

Please confirm if the above link can fix the above problem or not.

Best Regards,
Yanjun.Zhu

>>
>> This commit changes the driver behavior when the kconfig 
>> DEBUG_LOCK_ALLOC is
>> enabled, and my kconfig does so.
>>
>> Actions for fix will be appreciated.
> 
> I have a fix for this, see:
> https://git.samba.org/?p=metze/linux/ 
> wip.git;a=commitdiff;h=7f55eb3373dca97b706e8521705a06d4bf84b0f0
> 
> I'll post it to the list later today.
> 
> Sorry for the trouble!
> metze

Re: [bug report] rdma_rxe module unload failure with DEBUG_LOCK_ALLOC enabled

[Shinichiro Kawasaki]

On 12/20/25 2:59 AM, yanjun.zhu wrote:
> On 12/19/25 5:45 AM, Stefan Metzmacher wrote:
>> Hi Shinichiro,
>>
>>> While I evaluate v6.19-rc1 kernel, I found that rdma_rxe module unload
>>> fails.
>>> The failure can be recreated by simple two commands below:
>>>
>>>      $ sudo modprobe rdma_rxe
>>>      $ sudo modprobe -r rdma_rxe
>>>      modprobe: FATAL: Module rdma_rxe is in use.
>>>
>>> I bisected and found the trigger commit is this:
>>>
>>>      80a85a771deb ("RDMA/rxe: reclassify sockets in order to avoid
>>> false positives from lockdep")
> 
> Thanks a lot.
> 
> https://git.samba.org/?p=metze/linux/wip.git;a=commitdiff;h=7f55eb3373dca97b706e8521705a06d4bf84b0f0
> 
> Hi, Shinichiro Kawasaki
> 
> Please confirm if the above link can fix the above problem or not.

Yes, I confirmed that the patch avoids the rdma_rxe module unload failure,
and replied with my Tested-by tag. Stefan, Yenjun, thank you again for the fix.