CODA MJPG Decoder Problems after Wakeup/Resume

CODA MJPG Decoder Problems after Wakeup/Resume

[Hager Stefan]

Hi.

I observed a problem within the Coda MJPEG Decoder:

An MJPG Stream over Ethernet is decoded - and displayed to a screen afterwards - with 30 frames per second (as sent by the external source) on am i.MX6D 2 Core Processor.
If the System was then sent so sleep ('$ echo mem > /sys/power/state') and woken up again, the stream is only decoded with a framerate of about 1 frame per second. This happens on each wakeup.
Before going to sleep there were 30 Interrupts per Second on the Decoder, afterwards only 1 per Second. I also observed a kthread worker taking nearly 100% of one Core.

What could be the issue? How to resolve it?

Best regards
Stefan

Re: CODA MJPG Decoder Problems after Wakeup/Resume

[Philipp Zabel]

Hi Stefan,

On Di, 2022-04-05 at 07:53 +0000, Hager Stefan wrote:
> Hi.
> 
> I observed a problem within the Coda MJPEG Decoder:
> 
> An MJPG Stream over Ethernet is decoded - and displayed to a screen
> afterwards - with 30 frames per second (as sent by the external
> source) on am i.MX6D 2 Core Processor.
> If the System was then sent so sleep ('$ echo mem >
> /sys/power/state') and woken up again, the stream is only decoded
> with a framerate of about 1 frame per second. This happens on each
> wakeup.

Which kernel version are you using and how are you driving coda-vpu
from userspace? Can you reproduce the issue without the network
streaming or display parts?

> Before going to sleep there were 30 Interrupts per Second on the
> Decoder, afterwards only 1 per Second. I also observed a kthread
> worker taking nearly 100% of one Core.

Could you use perf top to find out where the CPU time is spent?

> What could be the issue? How to resolve it?

Is the decoder still fed frames at 30 fps after resume? You could use
ftrace to trace the coda:*, v4l2:*, and vb2:* trace events to verify
this, and to check whether the hardware still is processing frames at
the expected speed.

regards
Philipp