On Thu, Jun 11, 2026 at 6:27 PM Bastian Blank <[email protected]> wrote: > Please switch to q35 and see if it still fails, aka use > "-machine q35,accel=kvm". The only real error in all this is a warning > about a broken legacy interrupt, which are still in use by the legacy pc > type.
It still crashes, but there are some differences. The kwin_wayland process gets stuck in D state only temporarily, and if I kill it, it respawns (this happens until I kill kwin_wayland_wrapper). However, when I run `systemctl stop sddm`, systemd-logind, SDDM and its Xorg get stuck (non-permanently; the first two eventually disappear and Xorg periodically unsuspends, and then gets stuck again) while doing terminal-related ioctls (KDSETMODE, VT_OPENQRY, KDSETMODE, respectively). The stack traces of hung processes are different -- sometimes slightly, sometimes more. After some time, TTY-accessing processes get unstuck, and new ones run without problems. Every 15 seconds, the kernel prints the "Buffer eviction failed" message, which seems to be associated with an increase of Xorg's voluntary_ctxt_switches value. Xorg is still stuck in the same ioctl call, but with a different stacktrace, similar to the one I included below (the lines before drm_fb_helper_pan_display are identical). systemd-logind had an interesting stack trace while stuck: [<0>] qxl_fence_wait+0xf8/0x1a0 [qxl] [<0>] dma_fence_wait_timeout+0x62/0x170 [<0>] dma_resv_wait_timeout+0xce/0x1d0 [<0>] ttm_bo_wait_ctx+0x3d/0x50 [ttm] [<0>] qxl_bo_move+0x4a/0xf0 [qxl] [<0>] ttm_bo_handle_move_mem+0xd0/0x1a0 [ttm] [<0>] ttm_bo_evict+0x115/0x170 [ttm] [<0>] ttm_bo_evict_cb+0x77/0xd0 [ttm] [<0>] ttm_lru_walk_for_evict+0xad/0x1d0 [ttm] [<0>] ttm_bo_alloc_resource+0x181/0x4b0 [ttm] [<0>] ttm_bo_validate+0x9e/0x180 [ttm] [<0>] ttm_bo_init_reserved+0x16d/0x1f0 [ttm] [<0>] qxl_bo_create+0x156/0x210 [qxl] [<0>] qxl_alloc_bo_reserved+0x4a/0xc0 [qxl] [<0>] qxl_image_alloc_objects+0xb3/0x130 [qxl] [<0>] qxl_draw_dirty_fb+0x1a9/0x4b0 [qxl] [<0>] qxl_primary_atomic_update+0xfd/0x130 [qxl] [<0>] drm_atomic_helper_commit_planes+0x100/0x310 [drm_kms_helper] [<0>] drm_atomic_helper_commit_tail+0x5e/0xc0 [drm_kms_helper] [<0>] commit_tail+0xf1/0x150 [drm_kms_helper] [<0>] drm_atomic_helper_commit+0x13c/0x180 [drm_kms_helper] [<0>] drm_atomic_commit+0xb1/0xe0 [drm] [<0>] drm_client_modeset_commit_atomic+0x1ec/0x230 [drm] [<0>] drm_client_modeset_commit_locked+0x58/0x160 [drm] [<0>] drm_fb_helper_pan_display+0x111/0x260 [drm_kms_helper] [<0>] fb_pan_display+0x8e/0x140 [<0>] bit_update_start+0x1d/0x40 [<0>] fbcon_switch+0x467/0x700 [<0>] redraw_screen+0x107/0x2a0 [<0>] fbcon_blank+0x243/0x350 [<0>] do_unblank_screen+0xd8/0x1f0 [<0>] vt_ioctl+0x574/0x1600 [<0>] tty_ioctl+0xe4/0x980 [<0>] __x64_sys_ioctl+0x97/0xe0 [<0>] do_syscall_64+0xcd/0x15f0 [<0>] entry_SYSCALL_64_after_hwframe+0x76/0x7e > X11 is deprecated and might be broken, don't use it. It might leave > hardware in weird states and with the interrupt problem above. I am aware that it is deprecated, but, unfortunately, spice-vdagent doesn't fully support Plasma on Wayland. Without regard to the deprecation, there still is the underlying kernel bug that makes unrelated processes hang. -- Samuel Plavec
