[Intel-gfx] [PATCH 00/22] .rodata diet

Chris Wilson chris at chris-wilson.co.uk
Wed Oct 5 13:54:34 UTC 2016


On Wed, Oct 05, 2016 at 02:39:16PM +0100, Tvrtko Ursulin wrote:
> 
> On 05/10/2016 14:01, Jani Nikula wrote:
> 
> >On Wed, 05 Oct 2016, Tvrtko Ursulin <tursulin at ursulin.net> wrote:
> >>Dynamic typing in __I915__ (INTEL_INFO) has and unfortuante consequence that
> >>for every time it is called under a WARN it generates a very verbose string
> >>placed into the appropriate .rodata section.
> >AFAIK the idea all along was to use this "dynamic typing" for a
> >transition period. I'm in favor of moving towards only accepting
> >dev_priv. I'm a bit hesitant to make the change in one go, though.
> 
> Oh right, maybe polymorphism would have been a better term to use. I
> know that it was a temporary solution but did not realize until
> yesterday what unbearable log messages it produces for WARNs and the
> bloat it adds.
> 
> Anyway, there are several arguments to be made here.
> 
> First of all, it has probably been some months since the last time
> people tried this (maybe in a slightly different form) so one point
> is that there will probably never be a good time for this amount of
> churn. However you do it it is so sprinkled all over that it is
> probably true.

The challenge with such sweeping changes is coordination of the trees.
David's fell flat because patches against -nightly couldn't be applied
against dinq. So for those we really need a back-merge + application.
 
> Secondly, this series does not go all the way. The issue of
> converting things alike INTEL_INFO(*)->gen to INTEL_GEN remains. It
> would be another good chunk of work on top of this. Only when that
> is done could the __I915__ magic go away (more or less).
> 
> Positive is that every patch in this series helps a bit by locking
> down a particular macro to dev_priv. This is since it seems we are
> not catching and stopping all the new code which does the wrong
> thing during review. So applying a sub part of this series is still
> a good step IMHO.

Yes, it does look gentler, and the .rodata does expain where David was
seeing the huge gains from the conversion.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre


More information about the Intel-gfx mailing list