I'm seeing what I believe is the same issue on my NV40 card (built-in graphics on an ASUS M2NPV-VM), and continue to see it even in 2.6.38-rc2. I've git bisected and the issue started with kernel git revision 72d7c3b: "x86: Use memblock to replace early_res" -- that's when the GPU lockup first appeared.
After that patch, my dmesg output shows various memory ranges being mapped to different locations: -kernel direct mapping tables up to bfef0000 @ 100000-704000 +kernel direct mapping tables up to bfef0000 @ 1f9fc000-20000000 ... -kernel direct mapping tables up to 130000000 @ 1a31000-23b7000 +kernel direct mapping tables up to 130000000 @ bf56a000-bfef0000 ... - NODE_DATA [0000000100000000 - 0000000100004fff] - [ffffea0000000000-ffffea0004bfffff] PMD -> [ffff880100200000-ffff880103dfffff] on node 0 + NODE_DATA [000000012fffb000 - 000000012fffffff] + [ffffea0000000000-ffffea0004bfffff] PMD -> [ffff88012c200000-ffff88012fdfffff] on node 0 ... -PCI-DMA: aperture base @ 20000000 size 65536 KB +PCI-DMA: aperture base @ b8000000 size 65536 KB And this interesting message is no longer present: -early_res array is doubled to 64 at [16000 - 167ff] And, finally, the following lines are new post-72d7c3b: +[drm] nouveau 0000:00:05.0: PFIFO_DMA_PUSHER - Ch 0 +[drm] nouveau 0000:00:05.0: GPU lockup - switching to software fbcon I haven't managed to work out if any of this means anything yet. Any guidance, even a suggestion of who to chase, would be greatly appreciated. Thanks, -Ted -- To UNSUBSCRIBE, email to debian-x-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/23408.1295870...@orpheus.gnusto.com
