From mboxrd@z Thu Jan  1 00:00:00 1970
From: Christoph Lameter <cl@gentwo.org>
Date: Fri, 11 Jul 2014 15:13:57 +0000
Subject: Re: [RFC Patch V1 07/30] mm: Use cpu_to_mem()/numa_mem_id() to support memoryless node
Message-Id: <alpine.DEB.2.11.1407111012210.25527@gentwo.org>
List-Id: <linux-hotplug.vger.kernel.org>
References: <1405064267-11678-1-git-send-email-jiang.liu@linux.intel.com> <1405064267-11678-8-git-send-email-jiang.liu@linux.intel.com> <20140711144205.GA27706@htj.dyndns.org>
In-Reply-To: <20140711144205.GA27706@htj.dyndns.org>
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
To: Tejun Heo <tj@kernel.org>
Cc: Jiang Liu <jiang.liu@linux.intel.com>, Andrew Morton <akpm@linux-foundation.org>, Mel Gorman <mgorman@suse.de>, David Rientjes <rientjes@google.com>, Mike Galbraith <umgwanakikbuti@gmail.com>, Peter Zijlstra <peterz@infradead.org>, "Rafael J . Wysocki" <rafael.j.wysocki@intel.com>, Vladimir Davydov <vdavydov@parallels.com>, Johannes Weiner <hannes@cmpxchg.org>, "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>, Rik van Riel <riel@redhat.com>, Wanpeng Li <liwanp@linux.vnet.ibm.com>, Zhang Yanfei <zhangyanfei@cn.fujitsu.com>, Catalin Marinas <catalin.marinas@arm.com>, Jianyu Zhan <nasa4836@gmail.com>, malc <av1474@comtv.ru>, Joonsoo Kim <iamjoonsoo.kim@lge.com>, Fabian Frederick <fabf@skynet.be>, Tony Luck <tony.luck@intel.com>, linux-mm@kvack.org, linux-hotplug@vger.kernel.org, linux-kernel@vger.kernel.org

On Fri, 11 Jul 2014, Tejun Heo wrote:

> Hello,
>
> On Fri, Jul 11, 2014 at 03:37:24PM +0800, Jiang Liu wrote:
> > When CONFIG_HAVE_MEMORYLESS_NODES is enabled, cpu_to_node()/numa_node_id()
> > may return a node without memory, and later cause system failure/panic
> > when calling kmalloc_node() and friends with returned node id.
>
> The patch itself looks okay to me but is this the right way to handle
> this?  Can't we just let the allocators fall back to the nearest node
> with memory?  Why do we need to impose this awareness of memory-less
> node on all the users?

Allocators typically fall back but they wont in some cases if you say
that you want memory from a particular node. A GFP_THISNODE would force a
failure of the alloc. In other cases it should fall back. I am not sure
that all allocations obey these conventions though.