On 10.4.2024 10.59, Paul Menzel wrote:
Dear Mathias, dear Michał,
Am 09.04.24 um 13:22 schrieb Mathias Nyman:
On 8.4.2024 22.05, Michał Pecio wrote:
It's also possible this TD/TRB was cancelled due to the disconnect.
Could be that even if driver removes the TD from the list and cleans
out the TRB from the ring buffer (turns TRB to no-op) hardware may
have read ahead and cached the TRB, and process it anyway.
I thought about it, but my debug patch says that the missing TD was
freed by finish_td(), which is called on TDs considered completed by
hardware. A cancelled TD would show giveback_invalidated_tds().
Anyway, we now have new information from the reporter. My v2 patch
keeps a log of the last five events processed on each transfer ring
and dumps the log on TRB mismatch errors.
Unfortunately, it looks like the host controller is broken and signals
completion of those transfers twice. The log below shows two distinct
events for TRB 32959a1c0 and that the coresponding TD has just been
freed by finish_td().
The trace confirms this, we get double completion events for several
Isoc TRBs. These double completions are seen after a transaction
error on the same device (different endpoint). >
Transfer events for TRB ..a1c0 twice, with a transaction error in between:
<idle>-0 [000] d.h2. 33819.709897: xhci_handle_event: EVENT: TRB 000000032959a1c0 status 'Success' len 0 slot 6 ep 2 type 'Transfer Event' flags e:c
<idle>-0 [000] d.h2. 33819.709904: xhci_handle_event: EVENT: TRB 000000041547d010 status 'USB Transaction Error' len 4 slot 6 ep 15 type 'Transfer Event' flags e:c
systemd-journal-395 [000] d.H1. 33819.711886: xhci_handle_event: EVENT: TRB 000000032959a1c0 status 'Success' len 0 slot 6 ep 2 type 'Transfer Event' flags e:c
Transfer events for TRB ..a1d0 twice (the next TRB)
systemd-journal-395 [000] d.H1. 33819.712001: xhci_handle_event: EVENT: TRB 000000032959a1d0 status 'Success' len 0 slot 6 ep 2 type 'Transfer Event' flags e:c
systemd-journal-395 [000] d.H1. 33819.712059: xhci_handle_event: EVENT: TRB 000000032959a1d0 status 'Success' len 0 slot 6 ep 2 type 'Transfer Event' flags e:c
Transfer events for TRB ..a1e0 twice
systemd-journal-395 [000] d.H1. 33819.712139: xhci_handle_event: EVENT: TRB 000000032959a1e0 status 'Success' len 0 slot 6 ep 2 type 'Transfer Event' flags e:c
systemd-journal-395 [000] d.h1. 33819.712871: xhci_handle_event: EVENT: TRB 000000032959a1e0 status 'Success' len 0 slot 6 ep 2 type 'Transfer Event' flags e:c
etc..
Driver can cope with these extra events, but if this is common we should
probably handle it silently and not concern users with that ERROR message.
Thank you for the detailed analysis. Excuse my ignorance, but do you have an idea, what this Sennheiser USB headset does differently than other USB devices? Additionally, is this a known problem with this Intel xHCI controller, meaning, is there an errata about this problem?
There are a few related erratas in older 9 series chipsets that possibly
could explain this, but those issues are no longer listed for newer chipsets.
The Sennheiser headset is a full-speed (FS) device that use 192 byte Isoch transfers.
Series 9 chipset xHC has issues with exactly those FS Isoch transfers over 189 bytes, see
" 1. USB Isoch In Transfer Error Issue"
There are some issues related to FS device removal:
" 13. USB Full-/low-speed Device Removal Issue"
And some related to resending transfer events for "chached" TRBs after FS device
disconnect/reconnect.
"25. USB xHCI may Execute a Stale Transfer Request Block (TRB)"
https://www.intel.co.jp/content/dam/www/public/us/en/documents/specification-updates/9-series-chipset-pch-spec-update.pdf
Thanks
Mathias
