Re: [PATCH 0/8] zcache: page cache compression support

[Nitin Gupta <ngupta@vflare.org>]

On 01/20/2011 08:16 AM, Pekka Enberg wrote:
> Hi Christoph,
>
> On Thu, Jan 20, 2011 at 07:33:29AM -0500, Nitin Gupta wrote:
>>> I just started looking into kztmem (weird name!) but on
>>> the high level it seems so much similar to zcache with some
>>> dynamic resizing added (callback for shrinker interface).
>>>
>>> Now, I'll try rebuilding zcache according to new cleancache
>>> API as provided by these set of patches. This will help refresh
>>> whatever issues I was having back then with pagecache
>>> compression and maybe pick useful bits/directions from
>>> new kztmem work.
> On Thu, Jan 20, 2011 at 2:47 PM, Christoph Hellwig<hch@infradead.org>  wrote:
>> Yes, we shouldn't have two drivers doing almost the same in the
>> tree.  Also adding core hooks for staging drivers really is against
>> the idea of staging of having a separate crap tree.  So it would be
>> good to get zcache into a state where we can merge it into the
>> proper tree first.  And then we can discuss if adding an abstraction
>> layer between it and the core VM really makes sense, and if it does
>> how.   But I'm pretty sure there's now need for multiple layers of
>> abstraction for something that's relatively core VM functionality.
>>
>> E.g. the abstraction should involve because of it's users, not the
>> compressed caching code should involve because it's needed to present
>> a user for otherwise useless code.
> I'm not sure which hooks you're referring to but for zcache we did this:
>
> http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff;h=b3a27d0529c6e5206f1b60f60263e3ecfd0d77cb
>
> I completely agree with getting zcache merged properly before going
> for the cleancache stuff.
>

These hooks are for zram (generic, in-memory compressed block devices)
which can also be used as swap disks. Without that swap notify hook, we
could not free [compressed] swap pages as soon as they are marked free.

For zcache (which does pagecache compression), we need separate set
of hooks, currently known as "cleancache" [1]. These hooks are very
minimal but not sure if they are accepted yet (they are present in
linux-next tree only, see: mm/cleancache.c, include/linux/cleancache.h

[1] cleancache: http://lwn.net/Articles/393013/

Nitin