Linux USB
 help / color / mirror / Atom feed
From: Michal Pecio <michal.pecio@gmail.com>
To: Ricardo Ribalda <ribalda@chromium.org>
Cc: hi@yasunobu.dev, linux-media@vger.kernel.org,
	Laurent Pinchart <laurent.pinchart@ideasonboard.com>,
	Hans de Goede <hansg@kernel.org>,
	Mauro Carvalho Chehab <mchehab@kernel.org>,
	Mathias Nyman <mathias.nyman@intel.com>,
	linux-usb@vger.kernel.org
Subject: Re: [PATCH] media: uvcvideo: Batch StreamCam unlinks on ASUS ASM2142
Date: Thu, 19 Nov 2026 14:19:56 +0100	[thread overview]
Message-ID: <20261119141956.2f320b1e.michal.pecio@gmail.com> (raw)
In-Reply-To: <CANiDSCsdaokL_2u_74b3ADT+U8TXskK53+k+hVA8D+PtttYEZQ@mail.gmail.com>

On Thu, 19 Nov 2026 13:06:50 +0100, Ricardo Ribalda wrote:
> On Thu, 19 Nov 2026 at 12:47, Yasunobu Sakashita via B4 Relay
> <devnull+hi.yasunobu.dev@kernel.org> wrote:
> >
> > From: Yasunobu Sakashita <hi@yasunobu.dev>
> >
> > Stopping a SuperSpeed StreamCam can halt xHCI on the ASUS PRIME
> > Z270-A. The fifth Stop Endpoint command timed out after an invalid
> > type 15 event.
> >
> > For the affected StreamCam and ASM2142, block URB resubmission.
> > Queue all unlinks before waiting so xHCI can coalesce them.

Hi Yasunobu,

I'm curious what's your general experience with SuperSpeed isochronous
cameras on ASM2142 (or ASM3142 - same IDs, not sure what difference).

Does it work, including in high resolution uncompressed video modes?
In my experience with one no-name USB3 camera, ASMedia contrellers
worked poorly in general, which is why I haven't bothered upstreaming
this compatibility hack.

Could you post "lsusb -v" of this camera?
Preferably with usbutils v019 or newer.

> >
> > The stock driver reproduced the halt on Ubuntu 6.8.0-149-generic.
> > Three FFmpeg 720p MJPEG and audio capture cycles passed with the
> > patch.
> >
> > I used Codex to help analyze traces and draft the patch and
> > changelog.
> >
> > Signed-off-by: Yasunobu Sakashita <hi@yasunobu.dev>
> > ---
> > Built uvc_video.o on media-committers/next with and without CONFIG_PCI.
> > ---
> >  drivers/media/usb/uvc/uvc_video.c | 43 ++++++++++++++++++++++++++++++++++-----
> >  1 file changed, 38 insertions(+), 5 deletions(-)
> >
> > diff --git a/drivers/media/usb/uvc/uvc_video.c b/drivers/media/usb/uvc/uvc_video.c
> > index a92a986ab..0ee5a7d17 100644
> > --- a/drivers/media/usb/uvc/uvc_video.c
> > +++ b/drivers/media/usb/uvc/uvc_video.c
> > @@ -11,6 +11,7 @@
> >  #include <linux/kernel.h>
> >  #include <linux/list.h>
> >  #include <linux/module.h>
> > +#include <linux/pci.h>
> >  #include <linux/slab.h>
> >  #include <linux/usb.h>
> >  #include <linux/usb/hcd.h>
> > @@ -1885,6 +1886,26 @@ static int uvc_alloc_urb_buffers(struct uvc_streaming *stream,
> >         return 0;
> >  }
> >
> > +static bool uvc_streamcam_batch_unlink(const struct uvc_streaming *stream)
> > +{
> > +       const struct usb_device *udev = stream->dev->udev;
> > +       struct device *controller = udev->bus->controller;
> > +       const struct pci_dev *pdev;
> > +
> > +       if (stream->frozen || udev->state != USB_STATE_CONFIGURED ||
> > +           udev->speed != USB_SPEED_SUPER ||
> > +           le16_to_cpu(udev->descriptor.idVendor) != 0x046d ||
> > +           le16_to_cpu(udev->descriptor.idProduct) != 0x0893 ||
> > +           !controller || !dev_is_pci(controller))
> > +               return false;
> > +
> > +       pdev = to_pci_dev(controller);
> > +       return pdev->vendor == PCI_VENDOR_ID_ASMEDIA &&
> > +              pdev->device == 0x2142 &&
> > +              pdev->subsystem_vendor == PCI_VENDOR_ID_ASUSTEK &&
> > +              pdev->subsystem_device == 0x8732 && pdev->revision == 0;
> > +}
> > +

This shouldn't be unnecessary, the alternative unlinking logic is
unlikely to cause compatibility issues with any hardware.

> >  /*
> >   * Uninitialize isochronous/bulk URBs and free transfer buffers.
> >   */
> > @@ -1896,12 +1917,24 @@ static void uvc_video_stop_transfer(struct uvc_streaming *stream,
> >         uvc_video_stats_stop(stream);
> >
> >         /*
> > -        * We must poison the URBs rather than kill them to ensure that even
> > -        * after the completion handler returns, any asynchronous workqueues
> > -        * will be prevented from resubmitting the URBs.
> > +        * URBs must stay blocked until the workqueue has been flushed.
> > +        * A completion handler can leave work that tries to resubmit an URB.
> >          */
> > -       for_each_uvc_urb(uvc_urb, stream)
> > -               usb_poison_urb(uvc_urb->urb);
> > +       if (uvc_streamcam_batch_unlink(stream)) {
> > +               /*
> > +                * usb_block_urb() keeps URBs blocked after usb_kill_urb().
> > +                * Queue unlinks together so xHCI can coalesce them.
> > +                */
> > +               for_each_uvc_urb(uvc_urb, stream)
> > +                       usb_block_urb(uvc_urb->urb);
> > +               for_each_uvc_urb(uvc_urb, stream)
> > +                       usb_unlink_urb(uvc_urb->urb);
> > +               for_each_uvc_urb(uvc_urb, stream)
> > +                       usb_kill_urb(uvc_urb->urb);

For the record, the objective of this ugly code was to hopefully
preserve the same guarantees to the UVC driver as the original, while
avoiding repeatedly stopping and restarting the xHCI endpoint, which
occurs because usb_poison_urb() forces unlink of urb N to complete
before unlink of urb N+1 can begin.

I don't claim that this is the best or the only solution.

> > +       } else {
> > +               for_each_uvc_urb(uvc_urb, stream)
> > +                       usb_poison_urb(uvc_urb->urb);
> > +       }
> >  
> 
> Michal proposed something very similar not so long ago:
> 
> https://lore.kernel.org/linux-media/20261115110132.2b0bda47.michal.pecio@gmail.com/

Yes, LLMs got pretty good at stealing test patches from maling lists.

> He also mentioned that those crashes were solved in 7.3. Could you
> check if your device works fine on 7.3, or if it still crashes?
> (Note that even if a workaround is still needed, we probably wouldn't
> want to match host-controller PCI IDs inside uvcvideo.)

To be exact, 7.3 solved interrupt endpoint crashes on ASMedia and
isochronous endpoint crashes on VIA. I haven't tested isochronous on
ASMedia, but now I will, subject to the limitation that my camera
barely works with those controllers at all.

> Michal, would you recommend applying something like what you proposed
> in your email, or would you rather fix this at the USB level?

Other than bloat, the UVC change should be harmless and it makes
unlinking multiple URBs somewhat wore efficient and faster.

But it does make the code longer and weirder. If there is a need,
perhaps USB core could provide a new usb_nuke_endpoint() call.

I'm also curious to hear whether the 7.3 changes or this patch
affect the Intel+Razer case (linked above).

Regards,
Michal

      parent reply	other threads:[~2026-11-19 13:20 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-11-19 11:44 [PATCH] media: uvcvideo: Batch StreamCam unlinks on ASUS ASM2142 Yasunobu Sakashita via B4 Relay
2026-11-19 11:54 ` sashiko-bot
2026-11-19 12:06 ` Ricardo Ribalda
2026-11-19 12:43   ` Laurent Pinchart
2026-11-19 13:08     ` hi
2026-11-19 13:19   ` Michal Pecio [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20261119141956.2f320b1e.michal.pecio@gmail.com \
    --to=michal.pecio@gmail.com \
    --cc=hansg@kernel.org \
    --cc=hi@yasunobu.dev \
    --cc=laurent.pinchart@ideasonboard.com \
    --cc=linux-media@vger.kernel.org \
    --cc=linux-usb@vger.kernel.org \
    --cc=mathias.nyman@intel.com \
    --cc=mchehab@kernel.org \
    --cc=ribalda@chromium.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox