LWN: Comments on "Finding system latency with LatencyTOP"

Audio capture too, or just playback?

arjan — Thu, 24 Jan 2008 21:59:23 +0000

Today.. only a little... (eg it shows at least how long things take etc)

But the plan/hope is that the -rt latency tracer makes it to mainline at some point in the
future, which then really would help latencytop in giving the level of deep details needed for
this kind of analysis.

Audio capture too, or just playback?

johnkarp — Thu, 24 Jan 2008 18:35:39 +0000

Would this patch also help diagnose latency in the case of audio capture, 
not just playback?

Because they seem to be related but different problems; playback latency 
is the time between a userspace process making a syscall to write audio 
and the kernel poking the audio hardware, capture latency the time between 
the audio hardware asserting an interrupt and the userspace process 
receiving the data. Or at least thats my educated guess, I Am Not A Kernel 
Hacker.

Finding system latency with LatencyTOP

jake — Thu, 24 Jan 2008 15:15:41 +0000

> The function is called restore_latency_reason in the code and reset_latency_reason in the
article.

Good catch!  Fixed, thanks.

jake

Finding system latency with LatencyTOP

csali — Thu, 24 Jan 2008 11:35:00 +0000

The function is called restore_latency_reason in the code and reset_latency_reason in the article.

Finding system latency with LatencyTOP

arjan — Thu, 24 Jan 2008 07:12:38 +0000

Btw, in the current version of the LatencyTOP patch, the annotations are all gone, they're no
longer needed.
(and, many aren't needed in the original patch, since most of the Unknown ones actually get
resolved in userspace even in the original, and adding a missing annotation even there was
just adding a line to the userspace program)