[Intel-gfx] [PATCH] drm/i915: Implement wait-time accounting for i915
eric at anholt.net
Wed Jun 24 00:46:23 CEST 2009
On Sun, 2009-06-21 at 14:49 -0400, Ben Gamari wrote:
> Implement a mechanism for tracking CPU waits. This will hopefully aid in
> identifying GPU stalls. Likely wait-points are surrounded with a pair
> of I915_BEGIN_WAIT/I915_END_WAIT macros which time the enclosed region,
> crediting the wait to a given wait source. Thanks to ickle for the
> guidance in creating this patch.
So, I'm definitely interested, but I'd like a little more info: Has this
patch been very useful in fixing up lockstepping with the GPU?
It seems like if I was trying to do this (And I've had to debug this
stuff before, where I built a userland implementation of approximately
the same thing), I'd want to know the call stack all the way up into
userland to figure out who was at fault. So I was thinking that we
might want to be using ftrace or something for capturing forced-wait
information, that could get us stack traces rather than just "waited on
a domain transition for somebody."
eric at anholt.net eric.anholt at intel.com
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 197 bytes
Desc: This is a digitally signed message part
More information about the Intel-gfx