XFree86-4.0.2-11.4.0 on Alpha UP1100 with Matrox G450 - failure in direct rendering. This is with 2.4.2-ac12 kernel which supposedly includes various AGP fixes. When agpgart and mga modules are loaded then the following shows up in 'dmesg' buffers: Linux agpgart interface v0.99 (c) Jeff Hartmann agpgart: Maximum main memory to use for agp memory: 203M agpgart: Detected AMD Irongate chipset agpgart: AGP aperture is 32M @ 0x82000000 [drm] AGP 0.99 on AMD Irongate @ 0x82000000 32MB [drm] Initialized mga 2.0.1 20000928 on minor 63 .... (II) MGA(0): [drm] bpp: 16 depth: 16 (II) MGA(0): [drm] Sarea 2196+632: 2828 (0): [drm] created "mga" driver at busid "PCI:1:5:0" (0): [drm] added 8192 byte SAREA at 0xfffffc00082ea000 (0): [drm] mapped SAREA 0xfffffc00082ea000 to 0x2000002a000 (0): [drm] framebuffer handle = 0x80084000000 (0): [drm] added 1 reserved context for kernel (II) MGA(0): [drm] Registers = 0x80086820000 [drm] drmAgpEnabled succeeded for AGP mode 1x (0): [drm] failed to remove DRM signal handler (0): [drm] removed 1 reserved context for kernel DRIUnlock called when not locked (0): [drm] unmapping 8192 bytes of SAREA 0xfffffc00082ea000 at 0x2000002a000 (EE) MGA(0): [drm] drmAddMap failed on AGP aperture ... (II) MGA(0): Using overlay video (II) MGA(0): direct rendering disabled For a situation when a direct rendering works (the same XFree86-4.0.2-11.4.0 on x86 with G400 card and in 16 bit depth) the corresponding output looks like this (dmesg and log fragments): agpgart: Maximum main memory to use for agp memory: 94M agpgart: Detected Via Apollo KX133 chipset agpgart: AGP aperture is 64M @ 0xd0000000 [drm] AGP 0.99 on VIA Apollo KX133 @ 0xd0000000 64MB [drm] Initialized mga 2.0.0 20000910 on minor 63 .... (II) MGA(0): [drm] bpp: 16 depth: 16 (II) MGA(0): [drm] Sarea 2196+632: 2828 (0): [drm] created "mga" driver at busid "PCI:1:0:0" (0): [drm] added 4096 byte SAREA at 0xc515a000 (0): [drm] mapped SAREA 0xc515a000 to 0x4001c000 (0): [drm] framebuffer handle = 0xd8000000 (0): [drm] added 1 reserved context for kernel (II) MGA(0): [drm] Registers = 0xd4000000 [drm] drmAgpEnabled succeeded for AGP mode 1x (II) MGA(0): WARP Microcode Loaded (II) MGA(0): [drm] calced backoffset: 0x53b000 (II) MGA(0): [drm] added 31 65536 byte DMA buffers (II) MGA(0): [drm] Mga Dma Initialization start (II) MGA(0): [drm] Mga Dma Initialization done (II) MGA(0): [drm] Initialized Dma Engine (II) MGA(0): [drm] dma control initialized, using IRQ 4 .... (II) MGA(0): Using overlay video (0): X context handle = 0x00000001 (0): [drm] installed DRM signal handler (0): [DRI] installation complete (II) MGA(0): direct rendering enabled Any ideas how to fix that? A full log from X attached. Michal
Created attachment 11942 [details] full log from a direct rendering failure
I'm not sure DRI has worked at all on alpha before. But I could be wrong.
No. It was even not reasonable to try for different reasons. But this does not mean that it cannot work - eventually.
G450 and DRI don't get along yet on x86 let alone Alpha.. ;o) Heck, G450 doesn't get along period yet. ;o) I'll keep this problem in mind when investigating g450 issues however.
Whoops..
> Heck, G450 doesn't get along period yet. It gets along sufficiently enough that I was able to attach to this report a log file from a picture producing run on Alpha. You want these notes in e-mail instead? BTW - currently without direct rendering 'gears' benchmark on Alpha with G450 produces numbers in 215-230 fps range. Yes, indeed, there are various open issues.
Does rawhide 4.2.0 or any other newer version of X than was last reported tested fix this?
Still present on RHL 7.2 for Alpha? I've got a UP2000 and a DEC 500a, both which only have PCI slots. My Matrox G400, G450, and G550 are all AGP. I have PCI G200 and lower hardware. Do you have a spare UP1100 by chance? ;o) I almost bought one on Ebay a few months ago. Nice machine.
I do not have at this moment any Alpha so I cannot do any direct tests; but my experience with rawhide would indicate that this is the case. A support for AGP seems to be very 32-bit, if not x86, specific.
I'm still unable to test AGP functionality on Alpha. ;o/ I'd love to get a UP1100 or UP1500 for this sort of thing though. I'll ping George France and see if he might be able to play with it.
Defering until I've got an AGP alpha to poke at
I believe that the problem is really rooted in a kernel AGP support or at least is hard to solve on X level without fixing a kernel side first. Somebody eventually started rewrite of that in 2.5 kernels. Jeff Hartmann cannot be reached by anybody for a long time and the original code has numerous problems not only on Alpha. In any case XFree 4.2.0 works reasonably well (IIRC, I am using Radeon 7500 on Alpha for a long time) by switching to PCI mode; so this report, as can be expected, is now long obsolete.
I'm currently reviewing all X related bugs in "DEFERRED" state. As you've mentioned above, this issue is quite obsolete, and likely fixed for ages now in upstream kernel and/or X.Org X11. Setting to "CURRENTRELEASE" even though we don't ship an Alpha release anymore. If someone experiences the issue still, they should report it in the upstream X.Org and/or kernel bugzillas however. Thanks.