Re: [RFC PATCH] mm: introduce kv[mz]alloc helpers

From: Michal Hocko <mhocko@kernel.org>
To: Andrew Morton <akpm@linux-foundation.org>
Cc: linux-mm@kvack.org, Vlastimil Babka <vbabka@suse.cz>,
	David Rientjes <rientjes@google.com>,
	Mel Gorman <mgorman@suse.de>,
	Johannes Weiner <hannes@cmpxchg.org>,
	Anatoly Stepanov <astepanov@cloudlinux.com>,
	LKML <linux-kernel@vger.kernel.org>,
	Paolo Bonzini <pbonzini@redhat.com>,
	Mike Snitzer <snitzer@redhat.com>,
	dm-devel@redhat.com, "Michael S. Tsirkin" <mst@redhat.com>,
	"Theodore Ts'o" <tytso@mit.edu>,
	kvm@vger.kernel.org, linux-ext4@vger.kernel.org,
	linux-f2fs-devel@lists.sourceforge.net,
	linux-security-module@vger.kernel.org,
	Dave Chinner <david@fromorbit.com>,
	Al Viro <viro@zeniv.linux.org.uk>,
	Mikulas Patocka <mpatocka@redhat.com>,
	Joe Perches <joe@perches.com>
Subject: Re: [RFC PATCH] mm: introduce kv[mz]alloc helpers
Date: Wed, 14 Dec 2016 09:59:16 +0100	[thread overview]
Message-ID: <20161214085916.GB25573@dhcp22.suse.cz> (raw)
In-Reply-To: <1481666853.29291.33.camel@perches.com>

On Tue 13-12-16 14:07:33, Joe Perches wrote:
> On Tue, 2016-12-13 at 11:14 +0100, Michal Hocko wrote:
> > Are there any more comments or objections to this patch? Is this a good
> > start or kv[mz]alloc has to provide a way to cover GFP_NOFS users as
> > well in the initial version.
> 
> Did Andrew Morton ever comment on this?
> I believe he was the primary objector in the past.
> 
> Last I recollect was over a year ago:
> 
> https://lkml.org/lkml/2015/7/7/1050

Let me quote:
: Sigh.  We've resisted doing this because vmalloc() is somewhat of a bad
: thing, and we don't want to make it easy for people to do bad things.
: 
: And vmalloc is bad because a) it's slow and b) it does GFP_KERNEL
: allocations for page tables and c) it is susceptible to arena
: fragmentation.
: 
: We'd prefer that people fix their junk so it doesn't depend upon large
: contiguous allocations.  This isn't userspace - kernel space is hostile
: and kernel code should be robust.
: 
: So I dunno.  Should we continue to make it a bit more awkward to use
: vmalloc()?  Probably that tactic isn't being very successful - people
: will just go ahead and open-code it.  And given the surprising amount
: of stuff you've placed in kvmalloc_node(), they'll implement it
: incorrectly...
: 
: How about we compromise: add kvmalloc_node(), but include a BUG_ON("you
: suck") to it?

While I agree with some of those points, the reality really sucks,
though. We have tried the same tactic with __GFP_NOFAIL and failed as
well. I guess we should just bite the bullet and provide an api which is
so common that people keep reinventing their own ways around that, many
times wrongly or suboptimally. BUG_ON("you suck") is just not going to
help much I am afraid.

What do you think Andrew?
-- 
Michal Hocko
SUSE Labs