Hi, Lars-Peter Clausen <lars@xxxxxxxxxx> writes: >> Lars-Peter Clausen <lars@xxxxxxxxxx> writes: >>> Some UDC drivers (like the DWC3) expect that the response to a setup() >> >> not some, but *all*. You can only queue a response later IFF you return >> USB_GADGET_DELAYED_STATUS. > > Yeah, but most drivers don't care. DWC3 is one of two which handles > USB_GADGET_DELAYED_STATUS. those drivers probably wouldn't work with g_mass_storage ;-) >>> This fixed random occasional failures that were previously observed on a >>> DWC3 based system under high system load. >> >> I need to see tracepoint capture from the failure ;-) Care to send them >> to me for analysis? > > I've attached the full trace up to the point where the failure occurs. (The > tracing infrastructure in the DWC3 driver was very helpful in tracking this > down btw.). yeah, that was the idea. I see, however, that you're using an older kernel based on the your trace format. Just to make sure, did you try with v4.14? I see the problem is pretty clear and we will take the patch, but just to make sure, can you try v4.14? > The important parts are around: > > 125.504109: dwc3_complete_trb: ep0out: ... > 125.504111: dwc3_prepare_trb: ep0out: ... > > Then a few moments later > > 125.508529: dwc3_ep0: queueing request ffffffc8787b6d00 to ep0out ... > > Usually the queuing would happen before the dwc3_complete_trb and this > function would remove the URB from the queue. But now it will just sit on correct :-) > the pending list waiting for something to happen since the original transfer > is already over. > > When the next control request comes in and we try to queue something for EP0 > the 'list_empty(&dep->pending_list)' check will fail, we return an error and > the endpoint is stalled and the host will observe an error. right :-) btw, you can enable only some events from dwc3 if you want. Usually I enable all (echo 1 > /.../events/dwc3/enable) and disable read/writel (echo 0 > /.../events/dwc3/dwc3_readl/enable && echo 0 > /.../events/dwc3/dwc3_writel/enable), that helps a lot :-) -- balbi
Attachment:
signature.asc
Description: PGP signature
