--- On Wed, 6/9/10, P F <[email protected]> wrote:
> From: P F <[email protected]>
> Subject: [Linux-uvc-devel] [RFC] Discard overrun (corrupted) RX images
> To: [email protected]
> Date: Wednesday, June 9, 2010, 12:59 AM
> Hi,
>
> First of all, thank you for providing such great Linux
> support for webcams. I have encountered an issue in this
> area that I hope you experts can help with.
>
> I am using UVC/V4L on an embedded system, and I am
> experiencing a problem with corrupted camera images. The
> system consists of a custom ASIC with an ARM926EJ-S and OHCI
> USB host controller connected to a UVC camera which supports
> only MJPEG.
>
> The problem I have is that buffers dequeued in user space
> occasionally contain corrupted JPEGs. This corruption
> ranges from slight, with a row of macroblocks shifted, to
> severe, with the JPEG being totally undecompressible.
>
> If I enable UVC_TRACE_FRAME, I see several instances of:
> "uvcvideo: USB isochronous frame lost (-70)."
> in dmesg. I have noticed a correlation between the
> presence of this message and visual corruption in captured
> JPEGs.
>
> I see that this error code corresponds to -ECOMM, which in
> turn corresponds to TD_BUFFEROVERRUN in the isochronous
> packet status. I assume that this is caused by CPU
> starvation, which should just manifest as dropped frames.
>
> However, I think these overrun frames are not being
> dropped, and instead are being passed to user space in an
> indeterminate state, leading to the visual corruption that
> I've witnessed.
>
> The attached patch (against 2.6.31) is my naïve attempt at
> fixing the problem. It does appear to resolve the
> corruption, but I am not entirely sure it's the correct
> thing to do. Does it look reasonable? I would appreciate
> any feedback.
>
> Thank you,
> Filter
>
>
>
> -----Inline Attachment Follows-----
>
> _______________________________________________
> Linux-uvc-devel mailing list
> [email protected]
> https://lists.berlios.de/mailman/listinfo/linux-uvc-devel
>
I have discovered that the previous patch works only when some
significant amount of tracing is active. e.g., UVC_TRACE_FRAME or
UVC_TRACE_IOCTL. With tracing completely disabled, the patch has no
effect. Activating tracing without the patch also does not resolve the
corruption.
Clearly, I have more investigation to do. If anybody has ideas about what
might be occurring, I would like to hear them.
Thank you,
Filter
_______________________________________________
Linux-uvc-devel mailing list
[email protected]
https://lists.berlios.de/mailman/listinfo/linux-uvc-devel
