Top wakeups causers

[Adam Jackson]

On Mon, 2007-05-14 at 19:50 +0300, Aaron Baff wrote:

> Hmm...this sounds like it has something to do with either
> Enlightenment17 or the binary NVidia drivers I have. E17 has some
> nifty features which let you have a background that actually morphes
> (basically a mini graphics program). However when I turn it off and
> just use a static background the setitimer() usage does drop quite a
> bit so it may not be that.

The other thing I'm concerned about with the setitimer accounting is
that, afaict, it's not distinguishing between asking for a timer to
expire at some point in the future, and actually having that timer
expire and the signal delivered.  If the server finishes request
processing before the timer expires, it cancels the timer, so there's
really no wakeups at all.  (And most requests take _way_ less than 20ms
to complete.)  Context switches, yes, but not wakeups.

> My other thought is that its the NVidia drivers. Perhaps those
> requires some care taking every so often no matter what which means
> Xorg has to do that every so often. Its an interesting thought. Anyone
> who can actually talk to one of the engineers over there want to try
> and get in touch with them? I'll try the OSS nv driver sometime in the
> next day or so if I can get the time and see if that makes a
> difference.

>From what I remember from talking to them, their driver will
occasionally get interrupts back from the kernel saying that a context
has gone idle, or that they got vblank, or whatever.  They may just not
be turning them off when the server goes idle, as with the 915 driver
bug already noted.

I'll ask, I'm sure they'll want to know.

- ajax