[PULL] hardware latency detector

[PULL] hardware latency detector

[Jon Masters]

git_request_merge.txt

[PULL] hardware latency detector

[Jon Masters]

Linus,

Please pull "hwlat", the hardware latency detector for 2.6.31.

Patches can be pulled from:

  git://git.kernel.org/pub/scm/linux/kernel/git/jcm/linux-2.6-hwlat.git for-linus

Jon Masters (2):
      hwlat_detector: A system hardware latency detector
      hwlat_detector: include linux/delay.h in hwlat_detector

 Documentation/hwlat_detector.txt |   64 ++
 MAINTAINERS                      |    9 +
 drivers/misc/Kconfig             |   28 +
 drivers/misc/Makefile            |    1 +
 drivers/misc/hwlat_detector.c    | 1208 ++++++++++++++++++++++++++++++++++++++
 5 files changed, 1310 insertions(+), 0 deletions(-)
 create mode 100644 Documentation/hwlat_detector.txt
 create mode 100644 drivers/misc/hwlat_detector.c

Re: [PULL] hardware latency detector

[Linus Torvalds]

On Thu, 18 Jun 2009, Jon Masters wrote:
> 
> Please pull "hwlat", the hardware latency detector for 2.6.31.

I really want to know more before I pull. Has this been in -next? 
Discussed on lkml? How does it work? What does it do?

Thanks,

		Linus

Re: [PULL] hardware latency detector

[Jon Masters]

On Thu, 2009-06-18 at 18:25 -0700, Linus Torvalds wrote:
> 
> On Thu, 18 Jun 2009, Jon Masters wrote:
> > 
> > Please pull "hwlat", the hardware latency detector for 2.6.31.
> 
> I really want to know more before I pull.

Ok.

> Has this been in -next?

No, it hasn't. But it has been tested quite extensively by a number of
people and is currently in the -rt tree.

> Discussed on lkml?

Yes. I posted an RFC patch back when it was called the SMI detector,
then a patch last week with the new generic name of hardware latency
detector (at Ingo's suggestion of renaming).

> How does it work?

It started as an SMI detector. Basically, when this module is loaded and
enabled, it will grab all CPUs in periodic calls to stop_machine and sit
in a configurable loop, looking for unexplainable latencies. Doing it
this way allows us to look for something stealing the CPU from Linux
without the kernel knowing - we report the statistics via debugfs.

Hope that helps! Thanks Linus!

Jon.