Skip to content

Revert "We shouldn't be waiting too long for marking task hung. Set...

Lucas De Marchi requested to merge demarchi/i915-infra:tip-hung-task into master

Revert "We shouldn't be waiting too long for marking task hung. Set DEFAULT_HUNG_TASK_TIMEOUT to 30 seconds"

This reverts commit 92ca4e9b.

The sound driver sets a timeout of 60 seconds to wait i915 to bind to the respective pci device. When testing discrete cards and the integrated card is disabled in the BIOS, the igpu will never appear. Since the hung task timeout is smaller than the timeout on sound driver, we see the following trace:

<3>[   60.945948] INFO: task kworker/1:1:83 blocked for more than 30 seconds.
<3>[   60.946101]       Tainted: G        W         5.17.0-rc5-CI-CI_DRM_11260+ #1
<3>[   60.946191] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
<6>[   60.946288] task:kworker/1:1     state:D stack:14136 pid:   83 ppid:     2 flags:0x00004000
<6>[   60.946311] Workqueue: events azx_probe_work [snd_hda_intel]
<6>[   60.946334] Call Trace:
<6>[   60.946339]  <TASK>
<6>[   60.946355]  __schedule+0x42c/0xa80
<6>[   60.946386]  schedule+0x3f/0xc0
<6>[   60.946398]  schedule_timeout+0x1be/0x2e0
<6>[   60.946414]  ? del_timer_sync+0xb0/0xb0
<6>[   60.946432]  ? 0xffffffff81000000
<6>[   60.946446]  ? wait_for_completion_timeout+0x79/0x120
<6>[   60.946459]  wait_for_completion_timeout+0xab/0x120
<6>[   60.946483]  snd_hdac_i915_init+0xa5/0xb0 [snd_hda_core]
<6>[   60.946515]  azx_probe_work+0x71/0x84c [snd_hda_intel]
<6>[   60.946543]  process_one_work+0x275/0x5c0
<6>[   60.946574]  worker_thread+0x37/0x370
<6>[   60.946590]  ? process_one_work+0x5c0/0x5c0
<6>[   60.946600]  kthread+0xef/0x120
<6>[   60.946608]  ? kthread_complete_and_exit+0x20/0x20
<6>[   60.946623]  ret_from_fork+0x22/0x30
<6>[   60.946664]  </TASK>
<4>[   60.946690]
		  Showing all locks held in the system:
<4>[   60.946702] 1 lock held by khungtaskd/73:
<4>[   60.946709]  #0: ffffffff827372c0 (rcu_read_lock){....}-{1:2}, at: debug_show_all_locks+0x15/0x17a
<4>[   60.946746] 2 locks held by kworker/1:1/83:
<4>[   60.946753]  #0: ffff888100079b38 ((wq_completion)events){+.+.}-{0:0}, at: process_one_work+0x1eb/0x5c0
<4>[   60.946783]  #1: ffffc900003bbe78 ((work_completion)(&(&hda->probe_work)->work)){+.+.}-{0:0}, at: process_one_work+0x1eb/0x5c0
<4>[   60.946874] 1 lock held by in:imklog/401:

<4>[   60.946893] =============================================

<6>[   67.089990] snd_hda_intel 0000:00:1f.3: couldn't bind with audio component

This warn taints CI an then BAT can't proceed. Go back to the previous 60 seconds, which seems sufficient. Alternative would be to reduce the timeout in the sound driver or make it so the task is not eligible for hung task.

Signed-off-by: Lucas De Marchi lucas.demarchi@intel.com

Merge request reports