From: Christoph Lameter <cl@linux.com>
Subject: Re: [patch 1/5] mm: add nofail variants of kmalloc kcalloc and
 kzalloc
Date: Wed, 25 Aug 2010 16:35:04 -0500 (CDT)
Message-ID: <alpine.DEB.2.00.1008251630270.25660@router.home>
References: <alpine.DEB.2.00.1008240340010.18908@chino.kir.corp.google.com> <1282656558.2605.2742.camel@laptop> <4C73CA24.3060707@fusionio.com> <alpine.DEB.2.00.1008241309170.21242@chino.kir.corp.google.com> <20100825112433.GB4453@thunk.org>
 <1282736132.2605.3563.camel@laptop> <20100825115709.GD4453@thunk.org> <1282740516.2605.3644.camel@laptop> <20100825132417.GQ31488@dastard> <1282743342.2605.3707.camel@laptop> <alpine.DEB.2.00.1008251347550.16484@chino.kir.corp.google.com>
 <alpine.DEB.2.00.1008251604120.24931@router.home> <alpine.DEB.2.00.1008251418590.20253@chino.kir.corp.google.com>
Mime-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Cc: Peter Zijlstra <peterz@infradead.org>,
	Dave Chinner <david@fromorbit.com>, Ted Ts'o <tytso@mit.edu>,
	Jens Axboe <jaxboe@fusionio.com>,
	Andrew Morton <akpm@linux-foundation.org>,
	Neil Brown <neilb@suse.de>, Alasdair G Kergon <agk@redhat.com>,
	Chris Mason <chris.mason@oracle.com>,
	Steven Whitehouse <swhiteho@redhat.com>,
	Jan Kara <jack@suse.cz>,
	Frederic Weisbecker <fweisbec@gmail.com>,
	"linux-raid@vger.kernel.org" <linux-raid@vger.kernel.org>,
	"linux-btrfs@vger.kernel.org" <linux-btrfs@vger.kernel.org>,
	"cluster-devel@redhat.com" <cluster-devel@redhat.com>,
	"linux-ext4@vger.kernel.org" <linux-ext4@vger.kernel.org>,
	"reiserfs-devel@vger.kernel.org" <reiserfs-devel@vger.kernel.org>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
To: David Rientjes <rientjes@google.com>
Return-path: <linux-btrfs-owner@vger.kernel.org>
In-Reply-To: <alpine.DEB.2.00.1008251418590.20253@chino.kir.corp.google.com>
Sender: linux-btrfs-owner@vger.kernel.org
List-Id: linux-ext4.vger.kernel.org

On Wed, 25 Aug 2010, David Rientjes wrote:

> It all depends on what flags are passed to kmalloc(), slab nor slub
> enforce __GFP_NOFAIL behavior themselves.  In slab, cache_grow() will
> return NULL depending on whether the page allocator returns NULL, and that
> would only happen for __GFP_NORETRY or
> cachep->gfp->gfporder >= PAGE_ALLOC_COSTLY_ORDER.  In slub, the default
> order is tried with __GFP_NORETRY and if it returns NULL, the higher order
> alloc will fail under the same circumstances.  So the nofail behavior for
> slab depends only on the flags passed from the caller.

If the higher order fails in slub then an order 0 alloc is attempted
without __GFP_NORETRY. In both cases the nofail behavior of the page
allocator determines the outcode.

True if the caller mixes in __GFP_NORETRY then you may still get NULL. But
that is an issue that can be resolved by the caller.