Re: [PATCH RFC v7 08/23] dept: Apply sdt_might_sleep_strong() to PG_{locked,writeback} wait

[Byungchul Park]

Hillf wrote:
> On 9 Jan 2023 12:33:36 +0900 Byungchul Park <byungchul.park@xxxxxxx>
>> Makes Dept able to track dependencies by PG_{locked,writeback} waits.
>> 
>> Signed-off-by: Byungchul Park <byungchul.park@xxxxxxx>
>> ---
> 
> Hey Byungchul 

+cc max.byungchul.park@xxxxxxxxx

Hi,

This email never reached to me.

> Is DEPT able to get deadlock reported for the syzbot report [1]?

DEPT can detect the case 100% *IF* the folio_trylock() is released
within the same context since DEPT tracks folio_trylock(), folio_lock()
and folio_unlock(), and it's definitely a deadlock.

But as we know, because folio_trylock() can be released by another
context like irq, it might be either just a severe slowdown of the
context triggering folio_unlock() or a literal deadlock where the
context is involved. I dunno which one is the case.

In short, DEPT can detect this case too *IF* it's a literal deadlock,
but it doesn't if it's just a slowdown. I'm planning to warn it even if
there is a slowdown tho, not for now.

Let me reproduce the following issue. I will share the result.

> Hillf
> 
> [1] https://lore.kernel.org/lkml/6383cde5-cf4b-facf-6e07-1378a485657d@xxxxxxxxxxxxxxxxxxx/

	Byungchul