[Mesa-dev] [PATCH] i965: Use the predicate enable bit for conditional rendering without stalling

[Kenneth Graunke]

On Wednesday, November 12, 2014 11:28:15 AM Daniel Vetter wrote:
> On Tue, Nov 11, 2014 at 11:13:28AM -0800, Kenneth Graunke wrote:
> > On Tuesday, November 11, 2014 06:59:51 PM Neil Roberts wrote:
> > > Kenneth Graunke <kenneth at whitecape.org> writes:
> > > 
> > > > drm-intel-next must have the new software checker turned on, which
> > > > disallows non-whitelisted register writes (along with libva, so it
> > > > can't really be enabled upstream yet).
> > > 
> > > For what it's worth, I get the EINVAL error even on the stock Fedora 20
> > > kernel on Haswell (and presumably IvyBridge) so I can only assume the
> > > software checker is already upstream, unless I'm misunderstanding
> > > something.
> > > 
> > > $ uname -r
> > > 3.16.7-200.fc20.x86_64
> > > $ modinfo i915 | grep cmd_parser
> > > parm: enable_cmd_parser:Enable command parsing [...]
> > >       (1=enabled [default], 0=disabled) (int)
> > > $ sudo cat /sys/module/i915/parameters/enable_cmd_parser 
> > > 1
> > > 
> > > If I cat 0 to /sys/module/i915/parameters/enable_cmd_parser then I no
> > > longer get the EINVAL error.
> > > 
> > > - Neil
> > 
> > Huh.  Yeah, I thought they turned it on by default in 3.16, which I don't 
> > understand at all.  AFAIK the libva issue isn't fixed (or wasn't by then), 
so 
> > it sure seems like it would've broken userspace.  Which would be a pretty 
> > clear kernel policy violation...
> 
> We let libva pass. And in the latest patches from Brad if we detect libva
> tricks we'll still let it pass, just not with elevated privs needed for
> writing special registers. And the point of enabling the parser in 3.16
> already was to have as much coverage early as possible to catch any
> userspace issues we've missed.
> -Daniel

Cool.  I'd seen the discussiosn on making it work, and thought it hadn't 
actually happened...but I had my facts wrong.  Sorry for the trouble.

I actually haven't heard anyone complaining about breakage, so apparently it 
works pretty well.  Nice!

--Ken
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: This is a digitally signed message part.
URL: <http://lists.freedesktop.org/archives/mesa-dev/attachments/20141112/d1e927d2/attachment.sig>