Loic Poulain <loic.poulain@xxxxxxxxxx> writes:
> Hi Kalle,
>
> On Tue, 9 Feb 2021 at 16:45, Loic Poulain <loic.poulain@xxxxxxxxxx> wrote:
>>
>> mhi_deinit_chan_ctxt functionthat takes care of unitializing channel
>> resources, including unmapping coherent MHI areas, can be called
>> from different path in case of controller unregistering/removal:
>> - From a client driver remove callback, via mhi_unprepare_channel
>> - From mhi_driver_remove that unitialize all channels
>>
>> mhi_driver_remove()
>> |-> driver->remove()
>> | |-> mhi_unprepare_channel()
>> | |-> mhi_deinit_chan_ctxt()
>> |...
>> |-> mhi_deinit_chan_ctxt()
>>
>> This leads to double dma freeing...
>>
>> Fix that by preventing deinit for already uninitialized channel.
>>
>> Signed-off-by: Loic Poulain <loic.poulain@xxxxxxxxxx>
>> Reported-by: Kalle Valo <kvalo@xxxxxxxxxxxxxx>
>
> This is a 'blind' fix tentative, can you please check if it resolves
> the issue on your side?
I did a quick test few times and I don't see any crashes anymore, thanks!
Tested-by: Kalle Valo <kvalo@xxxxxxxxxxxxxx>
And like Jeffrey said:
Fixes: a7f422f2f89e ("bus: mhi: Fix channel close issue on driver remove")
Is it too late to push this to v5.11? But this should go to v5.12.
--
https://patchwork.kernel.org/project/linux-wireless/list/
https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches
