> On Nov 23, 2019, at 00:35, Mathias Nyman <mathias.nyman@xxxxxxxxxxxxxxx> wrote: > > On 22.11.2019 17.10, Kai-Heng Feng wrote: >>> On Nov 22, 2019, at 20:17, Mathias Nyman <mathias.nyman@xxxxxxxxxxxxxxx> wrote: >> [snipped] >>> >>> The event ring is already overwritten with new events when dumped. >>> >>> Could you apply the code below, it will dump both the endpoint transfer ring >>> and the event ring at the right moment: >> The log is quite large so I pasted it here: >> https://pastebin.ubuntu.com/p/wWpS8ctNrb/ >> Kai-Heng > > Thanks, ring dumps shows xHC hardware never wrote an event to the event ring for the TRB at 0x487a0fe0. > > Event ring: > [ 61.114263] xhci_hcd 0000:3f:00.0: 0x000000049803a5d0: TRB 0000000487a08fd0 status 'Short Packet' len 20416 slot 4 ep 3 type 'Transfer Event' flags e:c > we got a short packet event for TRB at 0x487a08fd0, which is one TRB before > [ 61.114269] xhci_hcd 0000:3f:00.0: 0x000000049803a5e0: TRB 0000000487a07a30 status 'Success' len 0 slot 4 ep 4 type 'Transfer Event' flags e:c > [ 61.114275] xhci_hcd 0000:3f:00.0: 0x000000049803a5f0: TRB 0000000487a07a40 status 'Success' len 0 slot 4 ep 4 type 'Transfer Event' flags e:c > [ 61.114281] xhci_hcd 0000:3f:00.0: 0x000000049803a600: TRB 0000000487a09000 status 'Short Packet' len 20416 slot 4 ep 3 type 'Transfer Event' flags e:c > we got a short packet event for TRB at 487a09000, which is one TRB after, > but no event for 0x487a0fe0 > > The TRB at 0x487a0fe0 in the endpoint ring looks fine: > [ 61.118683] xhci_hcd 0000:3f:00.0: 0x0000000487a08fc0: Buffer 000000046eef0000 length 32768 TD size 0 intr 0 type 'Normal' flags b:i:I:c:s:I:e:C > [ 61.118689] xhci_hcd 0000:3f:00.0: 0x0000000487a08fd0: Buffer 000000048e740000 length 32768 TD size 0 intr 0 type 'Normal' flags b:i:I:c:s:I:e:C > [ 61.118696] xhci_hcd 0000:3f:00.0: 0x0000000487a08fe0: Buffer 000000046ebe8000 length 32768 TD size 0 intr 0 type 'Normal' flags b:i:I:c:s:I:e:C > [ 61.118702] xhci_hcd 0000:3f:00.0: 0x0000000487a08ff0: LINK 0000000487a09000 intr 0 type 'Link' flags i:c:T:C > > The TRB with missing event is similar to all the other Normal TRBs in the ring. > It has both interrupt on completion, and interrupt on short packet bits set. > Can't see any reason why an event wasn't generated. Thanks for you analysis. The driver for xHCI is a little bit overwhelming, so I didn't read the code through. > > It seems it's always the TRB just before the LINK TRB that is missing an event. > (address ends with 0xfe0). > > Best guess is that this particular xHC (ASmedia) can't correctly generate events > for the TRB right before Link TRB Is it possible to still use the TRB with missing event, based on this behavior? Kai-Heng > > -Mathias
