Re: [PATCH 2/2] zram: user per-cpu compression streams

[Minchan Kim]

On Mon, May 02, 2016 at 06:21:57PM +0900, Sergey Senozhatsky wrote:
> On (05/02/16 17:28), Minchan Kim wrote:
> [..]
> > > aha... I misunderstood you. I thought you talked about the test results
> > > in particular, not about zram stats in general. hm, given that we don't
> > > close the door for 'idle compression streams list' yet, and may revert
> > > per-cpu streams, may be we can introduce this stat in 4.8? otherwise, in
> > > the worst case, we will have to trim a user visible stat file once again
> > > IF we, for some reason, will decide to return idle list back (hopefully
> > > we will not). does it sound OK to you to not touch the stat file now?
> > 
> > My concern is how we can capture such regression without introducing
> > the stat of recompression? Do you have an idea? :)
> 
> ...hm...  inc ->failed_writes?
> 
> ... or as a dirty and ugly and illegal (read "undocumented") hack, we
> probably can use ->failed_reads for that purpose. simply because I don't
> think any one has ever seen ->failed_reads != 0.

:(

How about adding new debugfs for zram? And let's put unstable stats or
things we need to debug to there.

I got lessson from zs_create_pool discussion a few days ago,
"debugfs is just optioinal feature so it should be okay to not support
sometime e.g., -ENOMEM, -EEXIST and so on " although I'm not sure.

Logically, it might be right and easy to say but I don't believe it
makes sense practically. Once *userspace* relies on it, we should keep
the rule. No matter we said it thorugh document

I hope it's not only one person, fortunately one more people.

                https://lwn.net/Articles/309298/

Okay, let's add the knob to the existing sysfs(There is no different
between sysfs and debugfs with point of userspace once they start to
use it) because no need to add new code to avoid such mess.

Any thoughts?