jkohen: I can't blame you :) Any suggestions? Is there any way I can help?
jkohen: Should I at least file a bug?
airlied: jkohen: yeah file a bug on bugs.freedesktop.org
airlied: jkohen: well I mostly wrote that code :)
jkohen: Yeah, I know :)
jkohen: Mind you, there is no Radeon component at b.f.o, contrary to the post in your blog (nor is the mailing list linked from the global list, nor searcheable, for what is worth).
airlied: its xf86-video-ati component
airlied: the mailing list is at lists.x.org and for some reason the vhost is busted
airlied: shows it
jkohen: Filed #13613 under product xorg, component Driver/Radeon. Playing around with xrandr helped me find some weird behavior that I described in the report.
jkohen: Thanks for your help and research!
ethana3: alright, now tell me if I'm getting this right:
ethana3: fifo is a command buffer for the GPU
ethana3: and mmio is for transferring the /data/ that those commands operate on?
ethana3: Ok, so if someone has a 128MB card
ethana3: can the kernel use it for anything?
ethana3: or is it only accessible to the radeon driver?
tilman: mmio: memory mapped io
tilman: ie you have a chunk of memory that's mapped to hw registers
tilman: on the radeon, registers can be either programmed via MMIO, or via the command processor (CP)
tilman: mmio = slow, cp = fast
agd5f: airlied: what do you think of the patch in bug 13610?
airlied: hmm seems wrong but it might be okay.. I'll see if I can work out the logic..
airlied: agd5f: we should try with just ripping out the call on entervt
agd5f: That's what I was thinking
airlied: ah or we can copy what inte ldoes..
airlied: they added an check since I stole the code.
airlied: have a look at want_vblank_interrupts and I830DRISetVBlankInterrupt
agd5f: airlied: we just need to add a check in RADEONDRISetVBlankInterrupt() for info->want_vblank_interrupts
airlied: yup that sounds like the right answer..
agd5f: airlied: shouldn't we also check if only crtc1 is enabled and set crtc1 interrupt bit?
agd5f: or will that break drm compat?
agd5f: although I think it's probably broken now if that's the case and you need vblanks and crtc1 is the only active crtc
airlied: hmm.. I'd need to review the code again..
airlied: we rarely on radeon end up with that case..
airlied: on intel is was common due to lvds being on their crtc1
agd5f: pushed the basic fix
agd5f: airlied: looks like radeon should be ok with drm master, if we do specify crtc1 vs crtc2, but I don't know how far back we need to go