linux-kernel - Re: [PATCH incremental] cpusets: initialize spread rotor lazily

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <alpine.DEB.2.00.1104181316110.31186@chino.kir.corp.google.com>
Date:	Mon, 18 Apr 2011 13:19:09 -0700 (PDT)
From:	David Rientjes <rientjes@...gle.com>
To:	Michal Hocko <mhocko@...e.cz>
cc:	Andrew Morton <akpm@...ux-foundation.org>,
	KOSAKI Motohiro <kosaki.motohiro@...fujitsu.com>,
	LKML <linux-kernel@...r.kernel.org>,
	Jack Steiner <steiner@....com>,
	Lee Schermerhorn <lee.schermerhorn@...com>,
	Christoph Lameter <cl@...ux-foundation.org>,
	Pekka Enberg <penberg@...helsinki.fi>,
	Paul Menage <menage@...gle.com>, Robin Holt <holt@....com>,
	Linus Torvalds <torvalds@...ux-foundation.org>,
	linux-mm@...ck.org
Subject: Re: [PATCH incremental] cpusets: initialize spread rotor lazily

On Mon, 18 Apr 2011, Michal Hocko wrote:

> > It'd probably be better to just make an incremental patch on top of 
> > mmotm-2011-04-14-15-08 with a new changelog and then propose with with 
> > your list of reviewed-by lines.
> 
> Sure, no problems. Maybe it will be easier for Andrew as well.
> 
> > Andrew could easily drop the earlier version and merge this v2, but I'm 
> > asking for selfish reasons:
> 
> Just out of curiosity. What is the reason? Don't want to wait for new mmotm?
> 

Because lazy initialization is another feature on top of the existing 
patch so it should be done incrementally instead of proposing an entirely 
new patch which is already mostly in -mm.

> > please use NUMA_NO_NODE instead of -1.
> 
> Good idea. I have updated the patch.
> 

Thanks.

> Changes from v2:
>  - use NUMA_NO_NODE rather than hardcoded -1
>  - make the patch incremental to the original one because that one is in
>    -mm tree already.
> Changes from v1:
>  - initialize cpuset_{mem,slab}_spread_rotor lazily}
> 
> [Here is the follow-up patch based on top of
> http://userweb.kernel.org/~akpm/mmotm/broken-out/cpusets-randomize-node-rotor-used-in-cpuset_mem_spread_node.patch]
> ---
> From: Michal Hocko <mhocko@...e.cz>
> Subject: cpusets: initialize spread mem/slab rotor lazily
> 
> Kosaki Motohiro raised a concern that copy_process is hot path and we do
> not want to initialize cpuset_{mem,slab}_spread_rotor if they are not
> used most of the time.
> 
> I think that we should rather intialize it lazily when rotors are used
> for the first time.
> This will also catch the case when we set up spread mem/slab later.
> 
> Also do not use -1 for unitialized nodes and rather use NUMA_NO_NODE
> instead.
> 

Don't need to refer to a previous version that used -1 since it will never 
be committed and nobody will know what you're talking about in the git 
log.

> Signed-off-by: Michal Hocko <mhocko@...e.cz>
> Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@...fujitsu.com>
> ---
>  cpuset.c |    8 ++++++++
>  fork.c   |    4 ++--
>  2 files changed, 10 insertions(+), 2 deletions(-)
> Index: linus_tree/kernel/cpuset.c
> ===================================================================
> --- linus_tree.orig/kernel/cpuset.c	2011-04-18 10:33:15.000000000 +0200
> +++ linus_tree/kernel/cpuset.c	2011-04-18 10:33:56.000000000 +0200
> @@ -2460,11 +2460,19 @@ static int cpuset_spread_node(int *rotor
>  
>  int cpuset_mem_spread_node(void)
>  {
> +	if (current->cpuset_mem_spread_rotor == NUMA_NO_NODE)
> +		current->cpuset_mem_spread_rotor =
> +			node_random(&current->mems_allowed);
> +
>  	return cpuset_spread_node(&current->cpuset_mem_spread_rotor);
>  }
>  
>  int cpuset_slab_spread_node(void)
>  {
> +	if (current->cpuset_slab_spread_rotor == NUMA_NO_NODE)
> +		current->cpuset_slab_spread_rotor
> +			= node_random(&current->mems_allowed);
> +

So one function has the `=' on the line with the assignment (preferred) 
and the other has it on the new value?

>  	return cpuset_spread_node(&current->cpuset_slab_spread_rotor);
>  }
>  
> Index: linus_tree/kernel/fork.c
> ===================================================================
> --- linus_tree.orig/kernel/fork.c	2011-04-18 10:33:15.000000000 +0200
> +++ linus_tree/kernel/fork.c	2011-04-18 10:33:56.000000000 +0200
> @@ -1126,8 +1126,8 @@ static struct task_struct *copy_process(
>  	mpol_fix_fork_child_flag(p);
>  #endif
>  #ifdef CONFIG_CPUSETS
> -	p->cpuset_mem_spread_rotor = node_random(&p->mems_allowed);
> -	p->cpuset_slab_spread_rotor = node_random(&p->mems_allowed);
> +	p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
> +	p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
>  #endif
>  #ifdef CONFIG_TRACE_IRQFLAGS
>  	p->irq_events = 0;
--
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/