Question - should the default be changed from KMS to FKMS for the
impish release?  It seems that freezing is severe enough either not to
ship / or ship with a less than perfect default until more time can be
devoted to investigate and resolve.

On Mon, 11 Oct 2021 at 15:50, Dave Jones <email address hidden> wrote:
>
> To summarize the current findings after several days of testing:
>
> * In all monitors the freeze manifests the same, with the same dmesg output
> * Some monitors, in particular it seems higher-frame-rate (100+Hz) monitors (but by no means exclusive to these) trigger the freeze quite easily, often before the login screen even appears
> * Other monitors, in particular lower-frame-rate (60Hz) monitors do trigger the freeze, but only after some activity
> * A few monitor(s) (so far only a 60Hz one) don't trigger the freeze at all (thanks to brian-murray for additional testing on this)
>
> Over the weekend I tested the 5.11.0-1019 and 5.11.0-1021 linux-raspi
> kernels from hirsute, installing them onto the impish image. With my
> (60Hz) monitor, I was unable to reproduce the crash with either of
> these, but jawn-smith *did* reproduce the issue with the hirsute kernel
> on his high-frame-rate monitor.
>
> In other words, even rolling all the way back to the hirsute kernel
> won't eliminate the freeze for everyone. The good news is that fkms does
> appear to reliably work around the issue (thanks for knoedelfan and
> fprietog for the reports in #3 and #4), although that in itself comes
> with its own issues (though none as severe as the display freezing).
>
> --
> You received this bug notification because you are subscribed to the bug
> report.
> https://bugs.launchpad.net/bugs/1946368
>
> Title:
>   HDMI output freezes under current/proposed impish kernels
>
> Status in linux-raspi package in Ubuntu:
>   Confirmed
> Status in linux-raspi source package in Impish:
>   Confirmed
>
> Bug description:
>   Under the current (5.13.0-1007.8) or proposed (5.13.0-1008.9) kernels
>   for the Ubuntu Pi pre-installed desktop impish release, the HDMI
>   output occasionally freezes. A known workaround at this time is to
>   change the following line in /boot/firmware/config.txt:
>
>   dtoverlay=vc4-kms-v3d
>
>   To the following:
>
>   dtoverlay=vc4-fkms-v3d
>
>   In other words, to use the "fake" KMS overlay (fkms) instead of the
>   "full" KMS overlay (kms).
>
>   I've been unable to determine a reliable method of guaranteeing a
>   freeze, but it appears to occur much more readily when video playback
>   is occurring, and when other interactions (especially moving windows
>   around, minimizing, restoring) occurs simultaneously.  Display suspend
>   also periodically causes the same hang, which made me suspect this
>   might be related to #1944397 but it appears that had a separate cause
>   (now resolved).
>
>   The following dmesg outputs have been observed immediately after the
>   display hang; this one from 1007.8:
>
>   [  513.762138] [drm:drm_crtc_commit_wait [drm]] *ERROR* flip_done timed out
>   [  513.762288] [drm:drm_atomic_helper_wait_for_dependencies [drm_kms_helper]] *ERROR* [CRTC:76:crtc-3] co
>   mmit wait timed out
>   [  513.762381] [drm:drm_atomic_helper_wait_for_flip_done [drm_kms_helper]] *ERROR* [CRTC:76:crtc-3] flip_
>   done timed out
>   [  524.002211] [drm:drm_crtc_commit_wait [drm]] *ERROR* flip_done timed out
>   [  524.002404] [drm:drm_atomic_helper_wait_for_dependencies [drm_kms_helper]] *ERROR* [PLANE:205:plane-25
>   ] commit wait timed out
>   [  534.242499] [drm:drm_crtc_commit_wait [drm]] *ERROR* flip_done timed out
>   [  534.242657] vc4-drm gpu: [drm] *ERROR* Timed out waiting for commit
>   [  534.250685] ------------[ cut here ]------------
>   [  534.250701] refcount_t: underflow; use-after-free.
>   [  534.250735] WARNING: CPU: 1 PID: 120 at lib/refcount.c:87 refcount_dec_not_one+0xa0/0xbc
>   [  534.250758] Modules linked in: rfcomm cmac algif_hash algif_skcipher af_alg hci_uart btqca btrtl btbcm
>    btintel bnep snd_soc_hdmi_codec vc4 btsdio snd_soc_core input_leds bluetooth snd_compress snd_bcm2835(C)
>    snd_pcm_dmaengine ecdh_generic ecc snd_pcm brcmfmac snd_seq_midi snd_seq_midi_event bcm2835_codec(C) bcm
>   2835_isp(C) bcm2835_v4l2(C) brcmutil snd_rawmidi v4l2_mem2mem bcm2835_mmal_vchiq(C) videobuf2_dma_contig
>   videobuf2_vmalloc cfg80211 videobuf2_memops videobuf2_v4l2 snd_seq videobuf2_common videodev snd_seq_devi
>   ce mc snd_timer vc_sm_cma(C) raspberrypi_hwmon snd bcm2835_gpiomem rpivid_mem uio_pdrv_genirq uio sch_fq_
>   codel ip_tables x_tables autofs4 btrfs blake2b_generic xor xor_neon zstd_compress hid_generic usbhid raid
>   6_pq libcrc32c dm_mirror dm_region_hash dm_log spidev dwc2 v3d roles udc_core gpu_sched crct10dif_ce i2c_
>   brcmstb i2c_bcm2835 spi_bcm2835 drm_kms_helper syscopyarea xhci_pci xhci_pci_renesas sysfillrect sysimgbl
>   t fb_sys_fops cec drm phy_generic ac97_bus aes_arm64
>   [  534.251066] CPU: 1 PID: 120 Comm: kworker/1:2 Tainted: G        WC        5.13.0-1007-raspi #8-Ubuntu
>   [  534.251076] Hardware name: Raspberry Pi 400 Rev 1.1 (DT)
>   [  534.251083] Workqueue: events drm_mode_rmfb_work_fn [drm]
>   [  534.251239] pstate: 60400005 (nZCv daif +PAN -UAO -TCO BTYPE=--)
>   [  534.251248] pc : refcount_dec_not_one+0xa0/0xbc
>   [  534.251257] lr : refcount_dec_not_one+0xa0/0xbc
>   [  534.251265] sp : ffff8000118cbb50
>   [  534.251269] x29: ffff8000118cbb50 x28: ffff6cf7ee894400 x27: ffff6cf80502e000
>   [  534.251285] x26: ffff6cf80502e000 x25: 0000000000000006 x24: ffff6cf7ee94c500
>   [  534.251300] x23: ffffa94dff246018 x22: ffff6cf833068880 x21: ffff6cf805027c80
>   [  534.251314] x20: ffff6cf88ead75ac x19: ffff6cf88ead7400 x18: 0000000000000000
>   [  534.251328] x17: 0000000000000000 x16: ffffa94e0a243314 x15: 0000000000000000
>   [  534.251342] x14: 0000000000000000 x13: 0000000000000030 x12: ffff800010035000
>   [  534.251356] x11: ffffa94e0b30dfd0 x10: 00000000fffff000 x9 : ffffa94e09d09f54
>   [  534.251370] x8 : 00000000ffffefff x7 : ffffa94e0b30dfd0 x6 : 0000000000000000
>   [  534.251384] x5 : ffff6cf8b799f948 x4 : 0000000000000000 x3 : 0000000000000027
>   [  534.251397] x2 : 0000000000000000 x1 : 0000000000000000 x0 : ffff6cf800abec80
>   [  534.251411] Call trace:
>   [  534.251416]  refcount_dec_not_one+0xa0/0xbc
>   [  534.251424]  vc4_bo_dec_usecnt+0x2c/0x120 [vc4]
>   [  534.251473]  vc4_cleanup_fb+0x3c/0x4c [vc4]
>   [  534.251518]  drm_atomic_helper_cleanup_planes+0x74/0xa0 [drm_kms_helper]
>   [  534.251604]  vc4_atomic_commit_tail+0x24c/0x36c [vc4]
>   [  534.251648]  commit_tail+0xac/0x190 [drm_kms_helper]
>   [  534.251723]  drm_atomic_helper_commit+0x168/0x380 [drm_kms_helper]
>   [  534.251796]  drm_atomic_commit+0x58/0x70 [drm]
>   [  534.251936]  atomic_remove_fb+0x2a8/0x2f4 [drm]
>   [  534.252073]  drm_framebuffer_remove+0x164/0x18c [drm]
>   [  534.252209]  drm_mode_rmfb_work_fn+0x50/0x70 [drm]
>   [  534.252346]  process_one_work+0x200/0x4d0
>   [  534.252359]  worker_thread+0x2c8/0x470
>   [  534.252367]  kthread+0x12c/0x140
>   [  534.252374]  ret_from_fork+0x10/0x3c
>   [  534.252386] ---[ end trace a97341262fc57e44 ]---
>
>   And a similar one from 1008.9 (note that most of the time, the stack
>   trace *doesn't* appear hence I'm not sure if it's related to the
>   display freeze itself, or something auxiliary):
>
>   [  221.914617] [drm:drm_atomic_helper_wait_for_flip_done [drm_kms_helper]] *ERROR* [CRTC:76:crtc-3] flip_done timed out
>   [  221.914617] [drm:drm_crtc_commit_wait [drm]] *ERROR* flip_done timed out
>   [  221.914795] [drm:drm_atomic_helper_wait_for_dependencies [drm_kms_helper]] *ERROR* [CRTC:76:crtc-3] commit wait timed out
>   [  232.154711] [drm:drm_crtc_commit_wait [drm]] *ERROR* flip_done timed out
>   [  232.154898] vc4-drm gpu: [drm] *ERROR* Timed out waiting for commit
>
>   I can produce this same stack trace, but *without* a corresponding
>   freeze by manually locking the  desktop (Super+L) and waiting for the
>   display fade. However, after the stack trace appears, one can press a
>   key to bring the display back and login again happily. Here are
>   several repeated traces from such activity under the 1008.9 proposed
>   kernel:
>
>   [ 1043.431061] [drm:drm_atomic_helper_wait_for_flip_done [drm_kms_helper]] *ERROR* [CRTC:76:crtc-3] flip_done timed out
>   [ 1043.431136] [drm:drm_crtc_commit_wait [drm]] *ERROR* flip_done timed out
>   [ 1043.431384] [drm:drm_atomic_helper_wait_for_dependencies [drm_kms_helper]] *ERROR* [CRTC:76:crtc-3] commit wait timed out
>   [ 1053.671415] [drm:drm_crtc_commit_wait [drm]] *ERROR* flip_done timed out
>   [ 1053.671705] [drm:drm_atomic_helper_wait_for_dependencies [drm_kms_helper]] *ERROR* [PLANE:70:plane-3] commit wait timed out
>   [ 1063.911800] [drm:drm_crtc_commit_wait [drm]] *ERROR* flip_done timed out
>   [ 1063.912147] vc4-drm gpu: [drm] *ERROR* Timed out waiting for commit
>   [ 1181.162739] [drm:drm_atomic_helper_wait_for_flip_done [drm_kms_helper]] *ERROR* [CRTC:76:crtc-3] flip_done timed out
>   [ 1181.418774] [drm:drm_crtc_commit_wait [drm]] *ERROR* flip_done timed out
>   [ 1181.419072] [drm:drm_atomic_helper_wait_for_dependencies [drm_kms_helper]] *ERROR* [CRTC:76:crtc-3] commit wait timed out
>   [ 1191.658996] [drm:drm_crtc_commit_wait [drm]] *ERROR* flip_done timed out
>   [ 1191.659289] [drm:drm_atomic_helper_wait_for_dependencies [drm_kms_helper]] *ERROR* [PLANE:70:plane-3] commit wait timed out
>   [ 1201.899168] [drm:drm_crtc_commit_wait [drm]] *ERROR* flip_done timed out
>   [ 1201.899438] vc4-drm gpu: [drm] *ERROR* Timed out waiting for commit
>   [ 1332.461450] [drm:drm_atomic_helper_wait_for_flip_done [drm_kms_helper]] *ERROR* [CRTC:76:crtc-3] flip_done timed out
>   [ 1332.461460] [drm:drm_crtc_commit_wait [drm]] *ERROR* flip_done timed out
>   [ 1332.461717] [drm:drm_atomic_helper_wait_for_dependencies [drm_kms_helper]] *ERROR* [CRTC:76:crtc-3] commit wait timed out
>   [ 1342.701602] [drm:drm_crtc_commit_wait [drm]] *ERROR* flip_done timed out
>   [ 1342.701890] [drm:drm_atomic_helper_wait_for_dependencies [drm_kms_helper]] *ERROR* [PLANE:70:plane-3] commit wait timed out
>   [ 1352.941798] [drm:drm_crtc_commit_wait [drm]] *ERROR* flip_done timed out
>   [ 1352.942067] vc4-drm gpu: [drm] *ERROR* Timed out waiting for commit
>
>   Note that even when the display is frozen (and cannot be resurrected),
>   only the display driver appears to crash; the system remains
>   operational (at least for some time) happily responding to SSH login
>   requests or, if video is playing, continuing audio output.
>
>   The same occurs with the current linux-firmware-raspi2 release (which
>   the above stack traces were taken from), or with the latest firmware
>   (installed from an experimental package in my
>   https://launchpad.net/~waveform/+archive/ubuntu/firmware PPA).
>
> To manage notifications about this bug go to:
> https://bugs.launchpad.net/ubuntu/+source/linux-raspi/+bug/1946368/+subscriptions
>