Hi, Thinh Nguyen <Thinh.Nguyen@xxxxxxxxxxxx> writes: >> 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? >> > > It shouldn't race. It's protected by the spinlock irq and it doesn't > matter whether dwc3_gadget_endpoint_trbs_complete() or this dequeue > function clears it first. The flag DWC3_EP_WAIT_TRANSFER_COMPLETE is > only applicable to stream transfer as the driver needs to wait for 1 > stream to finish before starting another. > > This is verified with our test environment (which includes UASP CV and > others). Fair enough, I'll take your word for it :-) Acked-by: Felipe Balbi <balbi@xxxxxxxxxx> -- balbi