Re: DRM lock ordering fix series

[Eric Anholt]

On Sat, 2009-03-28 at 02:29 +0100, Peter Zijlstra wrote:
> On Sat, 2009-03-28 at 01:58 +0100, Peter Zijlstra wrote:
> 
> > > OK.  I'm not too excited here -- 10% of 2% of the CPU time doesn't get
> > > me to the 10% loss that the slow path added up to.  Most of the cost is
> > > in k{un,}map_atomic of the returned pages.  
> > 
> > Also note that doing large gup() with gup_fast() will be undesirable due
> > to it disabling IRQs. So iterating say several MB worth of pages will
> > hurt like crazy. Currently all gup_fast() users do a single page lookup.
> 
> Also, what's this weird facination with 32bit, can you even buy a 32bit
> only cpu these days?

I work on OpenGL.  Many people using OpenGL want to play commercial
games.  Commercial games are 32-bit.  sysprof doesn't work for 32-on-64,
so I'd lose a critical tool.  Thus, 32-only.

keithp runs 32-on-64, and just about every day we're working together,
we lament that he can't run sysprof on his box.  Getting ~10% of my CPU
back by going 32-on-64 would be nice, but it's not worth not being able
to usefully profile.

-- 
Eric Anholt
eric@xxxxxxxxxx                         eric.anholt@xxxxxxxxx

Attachment: signature.asc
Description: This is a digitally signed message part