Re: HVR-1600: Skipped encoder MPEG, MDL 63, 62 times - it must have dropped out of rotation

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Mon, 2012-04-23 at 13:07 -0400, Brian J. Murrell wrote:
> On 12-04-22 04:56 PM, Andy Walls wrote:
> > 
> > If, in your system, IRQ service for device A under some circumstances
> > has precendence over IRQ service for the CX23418 and hence holds off its
> > service; and the irq handler in the driver for device A decides to
> > perform some some long I/O operations with device A; then it doesn't
> > matter how fast your CPU is. 
> 
> Yes, quite true.  I was forgetting about how nasty an irq handler can be
> on other hardware.
> 
> > You may wish to use perf or ftrace, or some other tool/method of
> > measuring kernel interrupt handling latency to find out what causes any
> > delays from the CX23418 raising its IRQ line to cx18_irq_handler() being
> > called by the kernel.
> 
> Excellent idea.  I'm afraid I'm quite (read: very) green in the area of
> kernel performance profiling. 

Here's an example of me checking latencies in the ivtv driver:

http://www.spinics.net/lists/linux-media/msg15762.html

Here are some good articles:

http://lwn.net/Articles/322666/
http://lwn.net/Articles/322731/
http://lwn.net/Articles/366796/
http://lwn.net/Articles/365835/
http://lwn.net/Articles/410200/
http://lwn.net/Articles/425583/
http://lwn.net/Articles/370423/

http://people.redhat.com/srostedt/ftrace-tutorial.odp
https://events.linuxfoundation.org/slides/2010/linuxcon_japan/linuxcon_jp2010_rostedt.pdf

An account of an in depth investigation into the maximum interrupts
disabled duration on some Sony of America test systems:
https://events.linuxfoundation.org/slides/2010/linuxcon_japan/linuxcon_jp2010_rowand.pdf


>  But I'm smart.  Looking around, it seems
> that with ftrace, I am looking for the irqsoff tracer, is that correct?

That sounds good as good as any.  (I always just end up learning what I
have to do when I need it.)


>  Unfortunately my kernel doesn't have that one:

It was probably just compiled without it.  On my Fedora 15 system, that
is the case with the stock kernel:

[andy@palomino ~]$  grep IRQSOFF_TRACER /boot/config-`uname -r`
# CONFIG_IRQSOFF_TRACER is not set

> # cat /sys/kernel/debug/tracing/available_tracers
> blk function_graph mmiotrace wakeup_rt wakeup function nop

*sigh*  So much for being convenient and easy.


> I can't seem to find any useful information on using perf to analyze ISR
> latency.  Any pointers?

Not really.  I have never used perf.  I assume you found this page:

https://perf.wiki.kernel.org/

> Cheers,
> b.

Regards,
Andy


--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html


[Index of Archives]     [Linux Input]     [Video for Linux]     [Gstreamer Embedded]     [Mplayer Users]     [Linux USB Devel]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [Yosemite Backpacking]
  Powered by Linux