Eric Boutilier wrote: > I need some help getting my onboard Intel 945G graphics working > properly. I'm using Nevada build 70a. What do these error messages from > Xorg.0.log indicate? > > ... > (WW) I810(0): xf86DeAllocateGARTMemory: deallocation of gart memory with key > 6 failed > (WW) I810(0): xf86UnbindGARTMemory: unbinding of gart memory with key 5 > failed (Not owner) > (WW) I810(0): xf86DeAllocateGARTMemory: deallocation of gart memory with key > 5 failed > (WW) I810(0): xf86AllocateGARTMemory: allocation of 1536 pages failed > ... > (EE) I810(0): Failed to allocate back buffer space. > (II) I810(0): Oh dear, the back buffer failed - badness > ... > > These errors get reported when Xorg starts, and also when I try to > change resolution (via the resolution preferences tool in GNOME). > > I'm just allowing Xorg to configure itself. In other words, I haven't > configured a custom xorg.conf file. > > This is an integrated Intel graphics device with the 82945G/GZ chipset > (aka "Intel 945G", aka "GMA 950"). Also, here's a link to the > motherboard specs page, in case it helps: > > http://h10025.www1.hp.com/ewfrf/wc/genericDocument?docname=c00916910&cc=us
I don't have any testing motherboard of that kind, nor mentionable experience with x86's agpgart lkm (as x86 isn't my main focus at this time). You may have best luck if you ask they author(s) of the i810 driver directly on Xorg's devel list. Here is all I could find via a quick google search (nobody ever reported exactly that funny error message that you got, never). http://lists.freedesktop.org/archives/xorg-commit/2007-March.txt.gz So the support for your chipset is quite new. That archive contains (at least parts) of/to the solution concerning above issue. E.g. the lines 208104 to 208124: if (alloced < size) { + /* Give up on trying to tile */ + tileable = FALSE; + size = ROUND_TO_PAGE(pScrn->displayWidth * height * pI830->cpp); + align = GTT_PAGE_SIZE; + alloced = I830AllocVidMem(pScrn, &(pI830->BackBuffer), + &(pI830->StolenPool), size, align, + flags | FROM_ANYWHERE | ALLOCATE_AT_TOP); + } + if (alloced < size) { + if (!dryrun) { + xf86DrvMsg(pScrn->scrnIndex, X_ERROR, + "Failed to allocate back buffer space.\n"); + } + return FALSE; + } + xf86DrvMsgVerb(pScrn->scrnIndex, X_INFO, verbosity, + "%sAllocated %ld kB for the back buffer at 0x%lx.\n", s, + alloced / 1024, pI830->BackBuffer.Start); - return TRUE; + return TRUE; } It *could* be something like that, for whatever reason. That is, that - for some reason - less memory can be allocated or mmap()'ed, than is available in total, according to the value of "size". Alan Hourihane seems to be the author of that driver (?). Please ask him on the Xorg list. p.s. Try to increase or decrease the Bios setting for shared system mem to be dedicated to video usage (although chances are low that this helps). Your last interim-work-around could be trying to use the (unaccelerated) VESA driver. Or just look into i810's and/or agpgart's sources by yourself, and simply trace back the actual root cause for bringing that chain into action. Sorry that I cannot tell you more than that. --martin
