Message-ID: <20130218045410.GC25347@verge.net.au>
Date:	Mon, 18 Feb 2013 13:54:10 +0900
From:	Simon Horman <horms@...ge.net.au>
To:	Julian Anastasov <ja@....bg>
Cc:	Andrew Morton <akpm@...ux-foundation.org>,
	Zhang Yanfei <zhangyanfei@...fujitsu.com>, davem@...emloft.net,
	Linux MM <linux-mm@...ck.org>,
	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH v2] net: fix functions and variables related to
 netns_ipvs->sysctl_sync_qlen_max

On Fri, Feb 15, 2013 at 11:39:34PM +0200, Julian Anastasov wrote:
> 
> 	Hello,
> 
> On Thu, 14 Feb 2013, Andrew Morton wrote:
> 
> > On Thu, 7 Feb 2013 10:40:26 +0200 (EET)
> > Julian Anastasov <ja@....bg> wrote:
> > 
> > > > Another question about the sysctl_sync_qlen_max:
> > > > This variable is assigned as:
> > > > 
> > > > ipvs->sysctl_sync_qlen_max = nr_free_buffer_pages() / 32;
> > > > 
> > > > The function nr_free_buffer_pages actually means: counts of pages
> > > > which are beyond high watermark within ZONE_DMA and ZONE_NORMAL.
> > > > 
> > > > is it ok to be called here? Some people misused this function because
> > > > the function name was misleading them. I am sorry I am totally not
> > > > familiar with the ipvs code, so I am just asking you about
> > > > this.
> > > 
> > > 	Using nr_free_buffer_pages should be fine here.
> > > We are using it as rough estimation for the number of sync
> > > buffers we can use in NORMAL zones. We are using dev->mtu
> > > for such buffers, so it can take a PAGE_SIZE for a buffer.
> > > We are not interested in HIGHMEM size. high watermarks
> > > should have negliable effect. I'm even not sure whether
> > > we need to clamp it for systems with TBs of memory.
> > 
> > Using nr_free_buffer_pages() is good-enough-for-now.  There are
> > questions around the name of this thing and its exact functionality and
> > whether callers are using it appropriately.  But if anything is changed
> > there, it will be as part of kernel-wide sweep.
> > 
> > One thing to bear in mind is memory hot[un]plug.  Anything which was
> > sized using nr_free_buffer_pages() (or similar) may become
> > inappropriately sized if memory is added or removed.  So any site which
> > uses nr_free_buffer_pages() really should be associated with a hotplug
> > handler and a great pile of code to resize the structure at runtime. 
> > It's pretty ugly stuff :(  I suspect it usually Just Doesn't Matter.
> 
> 	I'll try to think on this hotplug problem
> and also on the si_meminfo usage in net/netfilter/ipvs/ip_vs_ctl.c
> which is quite wrong for systems with HIGHMEM, may be
> we need there a free+reclaimable+file function for
> NORMAL zone.
> 
> > Redarding this patch:
> > net-change-type-of-netns_ipvs-sysctl_sync_qlen_max.patch and
> > net-fix-functions-and-variables-related-to-netns_ipvs-sysctl_sync_qlen_max.patch
> > are joined at the hip and should be redone as a single patch with a
> > suitable changelog, please.  And with a cc:netdev@...r.kernel.org.
> 
> 	Agreed, Zhang Yanfei and Simon? I'm just not sure,
> may be this combined patch should hit only the
> ipvs->nf->net trees? Or may be net-next, if we don't have
> time for 3.8.

ipvs->nf->net seems to be the most logical approach unless
Andrew wishes to take it.

I assume that at this point its too late for v3.8 and I'm not
entirely clear that the changes is needed there anyway.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists