shmem_get_pages: bad page non-NULL mapping (DMA remap failed) i915
Please note: Google has reported this issue. We are unable to reproduce the issue hence, we are unable to make trials with upstream kernel. Hence we dont have full logs with upstream kernel.
While we are trying for reproduction, request you to kindly suggest any pointers. @vsyrjala can you please suggest anything around this.
Some of the chromebooks are giving these crashes. Initially their suspect was that the issue probably is due to DPT restricted for shrinker. However, @vsyrjala fixed the issue with https://patchwork.freedesktop.org/series/141815/. Even then, the crash is observed by Google randomly mostly with low memory devices. These are seen on ADLP/ADLN devices more.
<4>[ 311.965533] i915 0000:00:02.0: Failed to DMA remap 65536 pages <1>[ 311.972444] BUG: Bad page state in process vkr-ring-9 pfn:17400 <4>[ 311.972450] page:000000005d4502c8 refcount:0 mapcount:1 mapping:00000000d5aba85b index:0x800 pfn:0x17400 <4>[ 311.972453] head:000000005d4502c8 order:0 entire_mapcount:1 nr_pages_mapped:0 pincount:0 <4>[ 311.972456] aops:0xffffffff9e23e770 ino:aef dentry name:"i915"
<4>[ 311.972522] <4>[ 311.972524] dump_stack_lvl+0x69/0xa0 <4>[ 311.972529] bad_page+0xea/0x100 <4>[ 311.972532] free_page_is_bad+0x2f/0x40 <4>[ 311.972534] __free_pages_ok+0x24a/0x3e0 <4>[ 311.972536] shmem_get_pages+0x117/0x2a0 <4>[ 311.972538] __i915_gem_object_get_pages+0x49/0x80 <4>[ 311.972541] i915_vma_pin_ww+0x8f8/0x990 <4>[ 311.972543] eb_validate_vmas+0x3e7/0x7f0 <4>[ 311.972545] i915_gem_execbuffer2_ioctl+0x111c/0x29a0 <4>[ 311.972549] drm_ioctl+0x342/0x680
Activity
<4>[ 311.972450] page:000000005d4502c8 refcount:0 mapcount:1 mapping:00000000d5aba85b index:0x800 pfn:0x17400 <4>[ 311.972453] head:000000005d4502c8 order:0 entire_mapcount:1 nr_pages_mapped:0 pincount:0 <4>[ 311.972456] aops:0xffffffff9e23e770 ino:aef dentry name:"i915"
Whenever the issue occurs, the mapcount is 1 and refcount is 0. Hence it might be a i915 issue and not generic out of memory issue.
@vsrini4 always get full logs from boot to issue seen if possible to help debug. https://drm.pages.freedesktop.org/intel-docs/how-to-file-i915-bugs.html
Please try on latest drm-tip too.
added Community feature: GEM labels
@jani.saarinen Thank you.
We are trying to reproduce the issue, once we are able to reproduce, we will get full logs with upstream kernel. As of now, we only have the report of crash from Google.
Hi, I'm at Google on ChromeOS and I'm now able to reproduce this issue. After several hours of testing, I can confirm the following patch fixes it. This patch is based off of 6.6, just wanted to get comments first. I can remail a version against 6.13 if it looks good.
Edited by Brian GeffonA small request. Could you kindly copy chris.p.wilson@intel.com when you mail the patch. We have seen generally he reviews GEM memory management side. Thanks a lot again for the fix Brian and Tomasz
Edited by vsrini4Hello @jani.saarinen Could you kindly help with review? Thank you.
Hello Brian, We have submitted the patch to trybot https://patchwork.freedesktop.org/patch/632671/ For some reason it wouldnt apply on drmtip. Could you kindly check if the patch submitted to trybot is okay? Thank you so much.
@tmistat , more to core to review.
Thanks a lot Brian. We had submitted v2 version of patch to trybot. https://patchwork.freedesktop.org/series/143602/#rev2
We dont see major failures. Some failures seen may be false alarms (unrelated).
@vsyrjala Hello Ville, many thanks for the review of v1. Could you kindly help to check v2 version from Brian? Thank you so much.
Hello @tmistat @jani.saarinen @vsyrjala
Apologies to bother you. Kindly can you help to review v2 patch from Brian. https://patchwork.freedesktop.org/patch/632697/
Thank you so much. I have also provided my tested-by.
CC: @briang_google
Edited by vsrini4follow-up on review of v2 of https://patchwork.freedesktop.org/patch/632697/ at respective mailing list, that is intel-gfx@lists.freedesktop.org
mentioned in commit 9e304a18
