Hi,
Thinh Nguyen <Thinh.Nguyen@xxxxxxxxxxxx> writes:
> If an active transfer is dequeued, then the endpoint is freed to start a
> new transfer. Make sure to clear the endpoint's transfer wait flag for
> this case.
>
> Cc: stable@xxxxxxxxxxxxxxx
> Fixes: e0d19563eb6c ("usb: dwc3: gadget: Wait for transfer completion")
> Signed-off-by: Thinh Nguyen <Thinh.Nguyen@xxxxxxxxxxxx>
> ---
> Changes in v2:
> - Only clear the wait flag if the selected request is of an active transfer.
> Otherwise, any dequeue will change the endpoint's state even if it's for
> some random request.
>
> drivers/usb/dwc3/gadget.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/drivers/usb/dwc3/gadget.c b/drivers/usb/dwc3/gadget.c
> index 78cb4db8a6e4..9a00dcaca010 100644
> --- a/drivers/usb/dwc3/gadget.c
> +++ b/drivers/usb/dwc3/gadget.c
> @@ -1763,6 +1763,8 @@ static int dwc3_gadget_ep_dequeue(struct usb_ep *ep,
> list_for_each_entry_safe(r, t, &dep->started_list, list)
> dwc3_gadget_move_cancelled_request(r);
>
> + dep->flags &= ~DWC3_EP_WAIT_TRANSFER_COMPLETE;
I'm not sure this is correct. This could create a race condition between
clearing this bit and getting the transfer complete interrupt. It also
seems to break the assumptions made by
dwc3_gadget_endpoint_trbs_complete() (actually its users), specially
regarding ISOC endpoints.
Have you verified all transfer types with this commit?
--
balbi
