Re: Re: [Xpert]Re: But *why* no vblank?

On Sun, 2002-11-24 at 17:42, Michel Dänzer wrote:
> On Mit, 2002-11-06 at 18:04, Keith Packard wrote:
> > Around 16 o'clock on Nov 6, Michel =?ISO-8859-1?Q?D=E4nzer?= wrote:
> > 
> > > Okay, is there anything wrong with turning the struct for the ioctl into
> > > a union of a request and a reply struct? :)
> > 
> > That is the usual way, I believe...  Or, you can just build a larger 
> > struct containing both pieces.
> > 
> > > Yes. The blocking ioctl also returns a timestamp, is that important for
> > > the signal?
> > 
> > Might be nice; there's plenty of space.  Is it expensive to compute?
> > 
> > > Oh, and BTW, is it okay for the ioctl to trigger a single signal, or
> > > would it have to generate signals indefinitely?
> > 
> > Might want a mode that chose between these two options, but if I had to 
> > pick one, I'd ask for a single signal.  That's what SYNC wants.
> 
> http://penguinppc.org/~daenzer/DRI/radeon-vbl-signal.diff
> 
> is an attempt at this, unfortunately not successful - it locks up solid
> when I request a signal to be delivered. Now I'd very much like to get
> this into 4.3.0, so I'd appreciate someone pointing out the stupid
> mistake(s) I'm probably making. :)

Thoughts on it so far: Is it legal to hijack the si_code field for this
use?  Where would the timestamp go that was talked about?  Also,
shouldn't you be using list_add/list_del for those list things done?

-- 
Eric Anholt <[email protected]>
http://people.freebsd.org/~anholt/dri/




-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf