Subject : Random crashes easily reproducible with make -j5 - intel i915 - kernel 2.6.36 on intel/nvidia hybrid graphics machine Submitter : Giacomo <delleceste@gmail.com> Date : 2010-12-10 8:57 Message-ID : AANLkTimkQM94u9iz7FVVjehB0mwDwfkNwKhF2F2tYq-r@mail.gmail.com References : http://marc.info/?l=linux-kernel&m=129197146619176&w=2 This entry is being used for tracking a regression from 2.6.35. Please don't close it until the problem is fixed in the mainline.
*** Bug 24542 has been marked as a duplicate of this bug. ***
The nouveau module wasn't even compiled when the crash used to happen. Another detail: kms was built as a module and the machine booted with kernel parameter i915 modeset=1. In the kernel 2.6.35 to which I switched back (and where I still haven't run into this issue so far) kms for intel is built in. I did not try to switch to nvidia drivers yet. To do that I would have to blacklist i915 and install nvidia proprietary drivers I think... so far vga switcheroo does not work for me and nvidia proprietary drivers hang the machine if i915 is loaded.
Giacomo, is this still an issue with 2.6.37?
Hi! I use my laptop to work so I could not accept the behaviour I experienced in 2.6.36. If in your opinion changes have been done to the code in order to solve the issue, I will try it with pleasure and let you know. Thanks for your interest. Giacomo.
There were 345 commits to drivers/gpu/drm/i915/*. (and 200+ commits from 2.6.37 to 2.6.38-rc2..) But other than that, no I have no specific reason to expect the issue to be fixed. So test at your own economic valuation. (Without specific (back-) traces, it's hard to know what the issue is.. )
I cannot actually use kernel 2.6.37 because I am unable to switch off the nvidia card due to http://bugs.gentoo.org/show_bug.cgi?id=351276 In fact, the acpi_call module which I use together with kernel 2.6.35 is able to switch off the card without problems. With 2.6.37 something went wrong and when I try to switch off the nvidia card via acpi_call method the machine hangs. There must have been a regression as to this aspect. I'd like to know if there are some good news about vga_switcheroo in 2.6.37, so that I can try to use that method instead of acpi_call. Thanks Giacomo
Can you retest with 2.6.38.y ?
Not sure how this was asigned to gfx, but it's been long enough without an update that either this is fixed or the reporter got a new machine. :) Giacomo, please re-open if this is still a problem. A bisect between 2.6.35 and 2.6.36 identifying the problem would help a lot too as it may not be a gfx issue.