Re: [RFC] simple system for enable/disable slabs being tracked by memcg.

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi Glauber,

On Wed, Mar 28, 2012 at 9:42 AM, Glauber Costa <glommer@xxxxxxxxxxxxx> wrote:
> Hi.
>
> This is a proposal I've got for how to finally settle down the
> question of which slabs should be tracked. The patch I am providing
> is for discussion only, and should apply ontop of Suleiman's latest
> version posted to the list.
>
> The idea is to create a new file, memory.kmem.slabs_allowed.
> I decided not to overload the slabinfo file for that, but I can,
> if you ultimately want to. I just think it is cleaner this way.
> As a small rationale, I'd like to somehow show which caches are
> available but disabled. And yet, keep the format compatible with
> /proc/slabinfo.
>
> Reading from this file will provide this information
> Writers should write a string:
>  [+-]cache_name
>
> The wild card * is accepted, but only that. I am leaving
> any complex processing to userspace.
>
> The * wildcard, though, is nice. It allows us to do:
>  -* (disable all)
>  +cache1
>  +cache2
>
> and so on.
>
> Part of this patch is actually converting the slab pointers in memcg
> to a complex memcg-specific structure that can hold a disabled pointer.
>
> We could actually store it in a free bit in the address, but that is
> a first version. Let me know if this is how you would like me to tackle
> this.
>
> With a system like this (either this, or something alike), my opposition
> to Suleiman's idea of tracking everything under the sun basically vanishes,
> since I can then selectively disable most of them.
>
> I still prefer a special kmalloc call than a GFP flag, though.

How would something like this interact with slab types that will have
a per-memcg shrinker?
Only do memcg shrinking for a slab type if it's not disabled?

While I like the idea of making it configurable by the user, I wonder
if we should be adding even more complexity to an already large
patchset, at this point.
I am also afraid that we might make this too hard setup correctly and use.

If it's ok, I'd prefer to keep going with a slab flag being passed to
kmem_cache_create, to determine if a slab type should be accounted or
not (opt-in), for now.

-- Suleiman

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href


[Index of Archives]     [Linux ARM Kernel]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [IETF Annouce]     [Bugtraq]     [Linux]     [Linux OMAP]     [Linux MIPS]     [ECOS]     [Asterisk Internet PBX]     [Linux API]