From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751861Ab3LLT3g (ORCPT ); Thu, 12 Dec 2013 14:29:36 -0500 Received: from www.sr71.net ([198.145.64.142]:44821 "EHLO blackbird.sr71.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751079Ab3LLT3f (ORCPT ); Thu, 12 Dec 2013 14:29:35 -0500 Message-ID: <52AA0E5D.30903@sr71.net> Date: Thu, 12 Dec 2013 11:28:29 -0800 From: Dave Hansen User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:24.0) Gecko/20100101 Thunderbird/24.2.0 MIME-Version: 1.0 To: Christoph Lameter CC: linux-kernel@vger.kernel.org, linux-mm@kvack.org, kirill.shutemov@linux.intel.com, Andi Kleen Subject: Re: [PATCH] [RFC] mm: slab: separate slab_page from 'struct page' References: <20131210204641.3CB515AE@viggo.jf.intel.com> <00000142de5634af-f92870a7-efe2-45cd-b50d-a6fbdf3b353c-000000@email.amazonses.com> <52A78B55.8050500@sr71.net> <00000142de866123-cf1406b5-b7a3-4688-b46f-80e338a622a1-000000@email.amazonses.com> <52A793D0.4020306@sr71.net> <00000142e7e23135-20f346b1-a880-47b0-946c-122323669ec1-000000@email.amazonses.com> In-Reply-To: <00000142e7e23135-20f346b1-a880-47b0-946c-122323669ec1-000000@email.amazonses.com> Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On 12/12/2013 09:37 AM, Christoph Lameter wrote: > On Tue, 10 Dec 2013, Dave Hansen wrote: >> See? *EVERYTHING* is overridden by at least one of the sl?b allocators >> except ->flags. In other words, there *ARE* no relationships when it >> comes to the sl?bs, except for page->flags. > > Slab objects can be used for I/O and then the page fields become > important. OK, which fields? How are they important? Looking at 'struct page', I don't see any fields other than ->flags that the slab allocators leave alone. I do see some refcounting (page->_count) done on request_queue->dma_drain_buffer which is kmalloc()'d. Although, I'm a bit skeptical that this is correct or has been audited because the slabs do seem to write over this storage. Anything needing to do per-page could probably be converted over to a raw alloc_pages() anyway because it couldn't possibly be doing sub-page-size allocations. What other cases were you thinking of?