On Sat, 1 Dec 2012 13:53:39 +0100, Daniel Vetter <daniel.vetter at ffwll.ch> wrote: > Hi all, > > Irq-drive gmbus/dp aux transfer, simply because we can (and at least in the case > of gmbus, it's quite a bit faster than the msleep(1) loop - we now reliably > transfer at full wire speed insteaf of sometimes 2-3x slower). > > Compared to the older version I've floated on irc way back and which Chris has > carried around in his wip branchs (and inflicted upon tons of unsuspecting bug > reporters) there are a few differences: > - handle the hpd vs. setup race - even the current code enables hpd processing > before the fbdev is set up, which is too early. > - fix the gen4 gmbus support, was totally busted. > - tested on hsw (although all the relevant bits are a 100% match with cpt/ivb). > - disable dp aux irq on vlv - too complicated to get at the docs (moved again, > old access revoked) and I don't have the hw. > > While reviewing this patch series I've also noticed two small things in the irq > handling code in general, patches for that at the beginning of the series. > > Comment&review highgly welcome. gm45 appears to be functioning big time, good start. The init sequence looks less convoluted and better understood now, so I am reasonably happy with this series. Will report back later after more testing. -Chris -- Chris Wilson, Intel Open Source Technology Centre
