Hi,
Roger Quadros <rogerq@xxxxxx> writes:
> Felipe,
>
> On 30/10/2019 13:26, Felipe Balbi wrote:
>>
>> Hi,
>>
>> Roger Quadros <rogerq@xxxxxx> writes:
>>>>>> irq/170-dwc3-1240 [000] d... 85.450564: dwc3_gadget_giveback: ep1out: req ee75d6ac length 0/256 zsI ==> -108
>>>>>> irq/170-dwc3-1240 [000] d... 85.450566: dwc3_gadget_giveback: ep1out: req c26c10a3 length 0/256 zsI ==> -108
>>>>>> irq/170-dwc3-1240 [000] d... 85.450567: dwc3_gadget_giveback: : req d4301893 length 0/0 zsI ==> 0
>>>>>
>>>>> Giveback happens on above two reqs after they are already freed. Could
>>>>> it be due to f_uac2 / u_audio.c performing usb_ep_free_request()
>>>>> immediately after usb_ep_dequeue() without waiting for completion?
>>>>
>>>> this could be a race condition. Add a trace_printk() at the cancelation,
>>>> see if the request is moved to cancelled list.
>>>>
>>>> diff --git a/drivers/usb/dwc3/gadget.c b/drivers/usb/dwc3/gadget.c
>>>> index 86dc1db788a9..affc2f7a929d 100644
>>>> --- a/drivers/usb/dwc3/gadget.c
>>>> +++ b/drivers/usb/dwc3/gadget.c
>>>> @@ -1513,6 +1513,7 @@ static void dwc3_gadget_ep_cleanup_cancelled_requests(struct dwc3_ep *dep)
>>>> struct dwc3_request *tmp;
>>>>
>>>> list_for_each_entry_safe(req, tmp, &dep->cancelled_list, list) {
>>>> + trace_printk("---> completing cancellation of req %p\n", req);
>>>> dwc3_gadget_ep_skip_trbs(dep, req);
>>>> dwc3_gadget_giveback(dep, req, -ECONNRESET);
>>>> }
>>>> @@ -1551,6 +1552,7 @@ static int dwc3_gadget_ep_dequeue(struct usb_ep *ep,
>>>> if (!r->trb)
>>>> goto out0;
>>>>
>>>> + trace_printk("---> req %p moved to cancelled\n", req);
>>>> dwc3_gadget_move_cancelled_request(req);
>>>> if (dep->flags & DWC3_EP_TRANSFER_STARTED)
>>>> goto out0;
>>>>
>>>>
>>>
>>> Below is trace with above debug patch applied.
>>
>> Did it fail the same way? It doesn't look like in the trace.
>>
>> Does this help?
>>
>> diff --git a/drivers/usb/dwc3/gadget.c b/drivers/usb/dwc3/gadget.c
>> index 86dc1db788a9..a9aba716bf80 100644
>> --- a/drivers/usb/dwc3/gadget.c
>> +++ b/drivers/usb/dwc3/gadget.c
>> @@ -707,6 +707,12 @@ static void dwc3_remove_requests(struct dwc3 *dwc, struct dwc3_ep *dep)
>>
>> dwc3_gadget_giveback(dep, req, -ESHUTDOWN);
>> }
>> +
>> + while (!list_empty(&dep->cancelled_list)) {
>> + req = next_request(&dep->cancelled_list);
>> +
>> + dwc3_gadget_giveback(dep, req, -ESHUTDOWN);
>> + }
>> }
>>
>> /**
>>
>
> Yes, this fixes the issue.
> I'm able to use aplay/arecord as well. Thanks!
Great, I'll send as a fix
--
balbi
