Re: [PATCH v2] mm/damon/dbgfs: avoid duplicate context directory creation

[SeongJae Park]

On Fri, 19 Aug 2022 15:44:51 -0700 Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:

> On Fri, 19 Aug 2022 21:16:31 +0000 SeongJae Park <sj@xxxxxxxxxx> wrote:
> 
> > > It would be simpler (and less racy) to check the debugfs_create_dir()
> > > return value for IS_ERR()?
> > 
> > I was merely following Greg's previous advice for ignoring the return value[1]
> > of the function, but I might misunderstanding his intention, so CC-ing Greg.
> > Greg, may I ask your opinion?
> > 
> > [1] https://lore.kernel.org/linux-mm/YB1kZaD%2F7omxXztF@xxxxxxxxx/
> 
> Thing is, the correct functioning of the debugfs interfaces is utterly
> critical to damon.  And that's apart from these memory leak and
> oops-we-killed-damon issues.
> 
> So damon simply cannot ignore the state of its debugfs interfaces and
> keep going along - because if something goes wrong at the debugfs
> layer, damon is dead and useless and the machine needs a reboot.
> 
> Perhaps this means that damon should not be using debugfs for its
> interfaces at all.  Or it means that the debugfs interfaces are
> misdesigned.  I go with the latter, which, alas, also affirms the
> former.

I'd save my word about the latter, but agreed on the former.  Fortunately we
already have an alternative (DAMON sysfs interface), and the debugfs interface
deprecation plan was announced for a while ago.  Not sure if the deprecation
will be well as hoped, though.

> 
> From a quick scan it appears that a significant minority (20%?) of
> drivers are checking the debugfs_create_dir() return value.

Maybe partly owing to Greg's previous efforts for removing the checks[1,2]?
Anyway, based on the previous discussions, I'd expect Greg might prefer not
checking the return code.

Anyway, waiting for his opinion.

[1] https://lore.kernel.org/all/20190122152151.16139-14-gregkh@xxxxxxxxxxxxxxxxxxx/
[2] https://lore.kernel.org/lkml/20190122152151.16139-7-gregkh@xxxxxxxxxxxxxxxxxxx/


Thanks,
SJ