Re: [PATCH 1/2] mm: Export vm_committed_as

[Michal Hocko]

On Mon 05-11-12 22:12:25, KY Srinivasan wrote:
> 
> 
> > -----Original Message-----
> > From: Andrew Morton [mailto:akpm@xxxxxxxxxxxxxxxxxxxx]
> > Sent: Monday, November 05, 2012 4:45 PM
> > To: KY Srinivasan
> > Cc: Greg KH; olaf@xxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx; andi@xxxxxxxxxxxxxx;
> > apw@xxxxxxxxxxxxx; devel@xxxxxxxxxxxxxxxxxxxxxx; linux-mm@xxxxxxxxx;
> > Hiroyuki Kamezawa; Michal Hocko; Johannes Weiner; Ying Han
> > Subject: Re: [PATCH 1/2] mm: Export vm_committed_as
> > 
> > On Sat, 3 Nov 2012 14:09:38 +0000
> > KY Srinivasan <kys@xxxxxxxxxxxxx> wrote:
> > 
> > >
> > >
> > > > >
> > > > > Ok, but you're going to have to get the -mm developers to agree that
> > > > > this is ok before I can accept it.
> > > >
> > > > Well I guess it won't kill us.
> > >
> > > Andrew,
> > >
> > > I presumed this was an Ack from you with regards to exporting the
> > > symbol. Looks like Greg is waiting to hear from you before he can check
> > > these patches in. Could you provide an explicit Ack.
> > >
> > 
> > Well, I do have some qualms about exporting vm_committed_as to modules.
> > 
> > vm_committed_as is a global thing and only really makes sense in a
> > non-containerised system.  If the application is running within a
> > memory cgroup then vm_enough_memory() and the global overcommit policy
> > are at best irrelevant and misleading.
> > 
> > If use of vm_committed_as is indeed a bad thing, then exporting it to
> > modules might increase the amount of badness in the kernel.
> > 
> > 
> > I don't think these qualms are serious enough to stand in the way of
> > this patch, but I'd be interested in hearing the memcg developers'
> > thoughts on the matter?
> > 
> > 
> > Perhaps you could provide a detailed description of why your module
> > actually needs this?  Precisely what information is it looking for
> > and why?  If we know that then perhaps a more comfortable alternative
> > can be found.
> 
> The Hyper-V host has a policy engine for managing available physical
> memory across competing virtual machines. This policy decision
> is based on a number of parameters including the memory pressure
> reported by the guest. Currently, the pressure calculation is based
> on the memory commitment made by the guest. From what I can tell, the
> ratio of currently allocated physical memory to the current memory
> commitment made by the guest (vm_committed_as) is used as one of the
> parameters in making the memory balancing decision on the host. This
> is what Windows guests report to the host. So, I need some measure of
> memory commitments made by the Linux guest. This is the reason I want
> export vm_committed_as.

So IIUC it will be guest who reports the value and the guest runs in the
ring-0 so it is not in any user process context, right?
If this is correct then memcg doesn't play any role here.
-- 
Michal Hocko
SUSE Labs
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/