Re: [PATCH 04/10] memcg: Introduce __GFP_NOACCOUNT.

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

 



On Wed, 29 Feb 2012 11:09:50 -0800
Suleiman Souhlal <suleiman@xxxxxxxxxx> wrote:

> On Tue, Feb 28, 2012 at 10:00 PM, KAMEZAWA Hiroyuki
> <kamezawa.hiroyu@xxxxxxxxxxxxxx> wrote:
> > On Mon, 27 Feb 2012 14:58:47 -0800
> > Suleiman Souhlal <ssouhlal@xxxxxxxxxxx> wrote:
> >
> >> This is used to indicate that we don't want an allocation to be accounted
> >> to the current cgroup.
> >>
> >> Signed-off-by: Suleiman Souhlal <suleiman@xxxxxxxxxx>
> >
> > I don't like this.
> >
> > Please add
> >
> > ___GFP_ACCOUNT  "account this allocation to memcg"
> >
> > Or make this as slab's flag if this work is for slab allocation.
> 
> We would like to account for all the slab allocations that happen in
> process context.
> 
> Manually marking every single allocation or kmem_cache with a GFP flag
> really doesn't seem like the right thing to do..
> 
> Can you explain why you don't like this flag?
> 

For example, tcp buffer limiting has another logic for buffer size controling.
_AND_, most of kernel pages are not reclaimable at all.
I think you should start from reclaimable caches as dcache, icache etc.

If you want to use this wider, you can discuss

+ #define GFP_KERNEL	(.....| ___GFP_ACCOUNT)

in future. I'd like to see small start because memory allocation failure
is always terrible and make the system unstable. Even if you notify
"Ah, kernel memory allocation failed because of memory.limit? and
 many unreclaimable memory usage. Please tweak the limitation or kill tasks!!"

The user can't do anything because he can't create any new task because of OOM.

The system will be being unstable until an admin, who is not under any limit,
tweaks something or reboot the system.

Please do small start until you provide Eco-System to avoid a case that
the admin cannot login and what he can do was only reboot.

Thanks,
-Kame

--
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=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>


[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]