[Bug 289605] graphics/drm-61-kmod amdgpu screen freeze

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=289605

--- Comment #1 from Patrick Mackinlay <freebsd.68fba@nospam.spacesurfer.com> ---
It should be noted that this problem seems to be somehow related to up time. I
get the lockup after 7-10 days using my desktop.
I tried using 
Option "accel" "false"
with the amdgpu x11 driver. However, the issue I originally described happened
after a week.

I then tried using the modesetting driver (all default options) and a similar
issue happened.

After a week, when using eclipse, some mouseover overlay windows would be
really slow to render. Eventually, they would render nothing. At this point I
noticed a whole load of /usr/local/libexec/webkit2gtk-4.1/WebKitWebProcess
processes running on 100% CPU (I think eclipse spawns these for some dialog
boxes and overlay windows). Initially, it seems these would eventually recover,
sometime taking minutes to recover. However, after an hour or so they stopped
recovering. Again, the processes could not be killed. Eventually opening a
thunderbird window the screen froze again.

Accessing the desktop from ssh I was able to run 
procstat kstack
on a /usr/local/libexec/webkit2gtk-4.1/WebKitWebProcess process stuck on 100%.
I ran it multiple times. This would produce one of the two outputs attached
(radeon2.txt and radeon3.txt). Looks like a thread is stuck in a loop trying to
allocate some kernel memory. The two kstack traces in question are:

vm_page_reclaim_contig_domain+0xe kmem_alloc_contig_pages+0xf5
kmem_alloc_contig_domainset+0x19c kmem_alloc_contig+0x29 contigmalloc+0x33
lkpi_kmalloc_cb+0x89 lkpi_kmalloc+0x27 amdgpu_bo_create+0xf6
amdgpu_bo_create_vm+0x1b amdgpu_vm_pt_create+0x190 amdgpu_vm_init+0x2ca
amdgpu_driver_open_kms+0xad drm_file_alloc+0x1ff drm_open+0x128
drm_stub_open+0x74 linux_dev_fdopen+0x141 devfs_open+0x110 VOP_OPEN_APV+0x1d 

vm_page_reclaim_contig_domain_ext+0x4e6 vm_page_reclaim_contig_domain+0xe
kmem_alloc_contig_pages+0xf5 kmem_alloc_contig_domainset+0x19c
kmem_alloc_contig+0x29 contigmalloc+0x33 lkpi_kmalloc_cb+0x89 lkpi_kmalloc+0x27
amdgpu_bo_create+0xf6 amdgpu_bo_create_vm+0x1b amdgpu_vm_pt_create+0x190
amdgpu_vm_init+0x2ca amdgpu_driver_open_kms+0xad drm_file_alloc+0x1ff
drm_open+0x128 drm_stub_open+0x74 linux_dev_fdopen+0x141 devfs_open+0x110

-- 
You are receiving this mail because:
You are the assignee for the bug.