[REPORT] ext4 deadlock possibilities by DEPT

* [REPORT] ext4 deadlock possibilities by DEPT
@ 2022-02-16  4:11 Byungchul Park
  2022-02-16  4:15 ` Report in ext4_buffered_write_iter() Byungchul Park
  2022-02-16  6:51 ` [REPORT] ext4 deadlock possibilities by DEPT Byungchul Park
  0 siblings, 2 replies; 8+ messages in thread
From: Byungchul Park @ 2022-02-16  4:11 UTC (permalink / raw)
  To: tytso, adilger.kernel, linux-ext4
  Cc: torvalds, mingo, linux-kernel, peterz, will, tglx, rostedt, joel,
	sashal, daniel.vetter, chris, duyuyang, johannes.berg, tj, willy,
	david, amir73il, bfields, gregkh, kernel-team, linux-mm, akpm,
	mhocko, minchan, hannes, vdavydov.dev, sj, jglisse, dennis, cl,
	penberg, rientjes, vbabka, ngupta, linux-block, axboe,
	paolo.valente, josef, linux-fsdevel, viro, jack, jlayton,
	dan.j.williams, hch, djwong, dri-devel, airlied,
	rodrigosiqueiramelo, melissa.srw, hamohammed.sa

Hi Theodore, Andreas and ext4 folks,

I've been developing a tool for detecting deadlock possibilities by
tracking wait/event rather than lock(?) acquisition order to try to
cover all synchonization machanisms. It's done on v5.17-rc1 tag.

https://github.com/lgebyungchulpark/linux-dept/commits/dept1.11_on_v5.17-rc1

Benifit:

	0. Works with all lock primitives.
	1. Works with wait_for_completion()/complete().
	2. Works with 'wait' on PG_locked.
	3. Works with 'wait' on PG_writeback.
	4. Works with swait/wakeup.
	5. Works with waitqueue.
	6. Multiple reports are allowed.
	7. Deduplication control on multiple reports.
	8. Withstand false positives thanks to 6.
	9. Easy to tag any wait/event.

Future work:

	0. To make it more stable.
	1. To separates Dept from Lockdep.
	2. To improves performance in terms of time and space.
	3. To use Dept as a dependency engine for Lockdep.
	4. To add any missing tags of wait/event in the kernel.
	5. To deduplicate stack trace.

I've got several reports from the tool. Some of them look like false
alarms caused by Lockdep's fake annotations added for better detection.
However, some others look like real deadlock possibility. Because of my
unfamiliarity of the domain, it's hard to confirm if it's a real one.
I'd like to ask for your opinion on it and it'd be appreciated.

How to interpret the report is:

	1. E(event) in each context cannot be triggered because of the
	   W(wait) that cannot be woken.
	2. The stack trace helping find the problematic code is located
	   in each conext's detail.

Let me add the reports on this email thread.

---
Thanks,
Byungchul

^ permalink raw reply	[flat|nested] 8+ messages in thread