Hi Mathias, On Thu, Mar 5, 2015 at 4:25 PM, Mathias Nyman <mathias.nyman@xxxxxxxxxxxxxxx> wrote: > On 04.03.2015 15:27, Alistair Grant wrote: >> On Tue, Mar 3, 2015 at 8:40 PM, Alistair Grant <akgrant0710@xxxxxxxxx> wrote: >>> On Tue, Mar 3, 2015 at 2:21 PM, Mathias Nyman >>> <mathias.nyman@xxxxxxxxxxxxxxx> wrote: >>>> On 28.02.2015 09:16, Alistair Grant wrote: >>>>> ... >>>>> * 3.19.0 with the following patches: >>>>> * xhci: Allocate correct amount of scratchpad buffers >>>>> * xhci: Don't touch TRBs memory if those are no longer on the endpoint ring >>>>> * xhci: fix invalid pointer in reset device debugging >>>>> * xhci: add debugging for reset device and stop endpoint commands >>>>> * xhci: add command ring stop and restart debug messages >>>>> >>>> >>>> Does increasing the TRB count per segment help? >>> >>> Success! >>> >>> Increasing TRBS_PER_SEGMENT from 64 to 256 allowed me to successfully >>> record two 30 second segments of video, i.e. start recording with >>> mythffmpeg, Ctrl-C after 30 seconds, then repeat (this is on top of the >>> patched kernel I reported in my last message). >>> >> ... >> I assume that this is a workaround, and that the core problem of ring >> expansion & cancelled URBs is still to be resolved. Let me know if >> you would like that tested when it is ready. >> > > Hi > > yes, this is a workaround. > > The latest theory for the cause is that we fill up the event ring. This > would be possible because we pick events from the event ring only on interrupt. > > Isoc transfers are set to only interrupt at the last TD, with several isoc transfer > going on simultaneously, and especially with isoc transfer containing so many TDs we need > to increase the transfer ring (hence the ring expansion befora failure in the log) we fill > up the event ring and won't receive any stop endpoint event -> timeout -> kill HC > > Increasing the TRBS_PER_SEGMENTS helps as it also increases the event ring. > > If you have time could you try forcing interrupts on every isoc TRB with the following change: > > diff --git a/drivers/usb/host/xhci.c b/drivers/usb/host/xhci.c > index 151484e..dfad305 100644 > --- a/drivers/usb/host/xhci.c > +++ b/drivers/usb/host/xhci.c > @@ -539,6 +539,8 @@ int xhci_init(struct usb_hcd *hcd) > xhci_dbg_trace(xhci, trace_xhci_dbg_init, > "xHCI doesn't need link TRB QUIRK"); > } > + xhci->quirks |= XHCI_AVOID_BEI; > + > retval = xhci_mem_init(xhci, GFP_KERNEL); > xhci_dbg_trace(xhci, trace_xhci_dbg_init, "Finished xhci_init > > with the old TRBS_PER_SEGMENT size, and see if it helps This worked without any problems also - I only recorded about 3 minutes of video, but I doubt that the length matters. FYI, the following messages still appear in syslog: Mar 5 19:56:19 alistair-XPS13 kernel: [ 68.725309] xhci_hcd 0000:00:14.0: WARN Set TR Deq Ptr cmd failed due to incorrect slot or ep state. Mar 5 19:56:19 alistair-XPS13 kernel: [ 68.725411] xhci_hcd 0000:00:14.0: Cancelled TD not on stopped ring (multiple times) Mar 5 19:56:19 alistair-XPS13 kernel: [ 68.725420] xhci_hcd 0000:00:14.0: Cancel URB NOT on current ring (multiple times) Mar 5 20:00:17 alistair-XPS13 kernel: [ 306.192925] xhci_hcd 0000:00:14.0: WARN Set TR Deq Ptr cmd failed due to incorrect slot or ep state. >From memory, the warning only appeared once with TRBS_PER_SEGMENT=256. HTH, Alistair -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
