Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1757064AbcCURH1 (ORCPT ); Mon, 21 Mar 2016 13:07:27 -0400 Received: from mga04.intel.com ([192.55.52.120]:44004 "EHLO mga04.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755732AbcCURH0 (ORCPT ); Mon, 21 Mar 2016 13:07:26 -0400 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.24,372,1455004800"; d="scan'208";a="941957990" Date: Mon, 21 Mar 2016 20:06:55 +0300 From: "Kirill A. Shutemov" To: Christoph Lameter Cc: Andrew Morton , Alexander Viro , Linus Torvalds , Matthew Wilcox , Hugh Dickins , linux-kernel@vger.kernel.org, linux-mm@kvack.org, linux-fsdevel@vger.kernel.org Subject: Re: [PATCH 0/3] fs, mm: get rid of PAGE_CACHE_* and page_cache_{get,release} macros Message-ID: <20160321170655.GA141158@black.fi.intel.com> References: <1458561998-126622-1-git-send-email-kirill.shutemov@linux.intel.com> <20160321163404.GA141069@black.fi.intel.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: User-Agent: Mutt/1.5.23.1 (2014-03-12) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 1235 Lines: 33 On Mon, Mar 21, 2016 at 11:59:25AM -0500, Christoph Lameter wrote: > On Mon, 21 Mar 2016, Kirill A. Shutemov wrote: > > > We do have anon-THP pages on LRU. My huge tmpfs patchset also put > > file-THPs on LRU list. > > So they are on the LRU as 4k units? Tried to look it up. One entry on LRU per huge page. > > > Will this actually work if we have really huge memory (100s of TB) where > > > almost everything is a huge page? Guess we have to use hugetlbfs and we > > > need to think about this as being exempt from paging. > > > > Sorry, I failed to understand your message. > > > > Look on huge tmpfs patchset. It allows both small and huge pages in page > > cache. > > Thus my wonder about this patchset. It seems then that the huge pages are > treated as 4k pages? Otherwise we would have two sizes for pages in the > page cache. Sorry I did not follow that too closely. Will try finding that > patchset. We do have two page sizes in the page cache. It's the only option to get transparent huge pages transparent. We have 512 (on x86-64) entries on radix-tree per huge page, but we can opt to Matthew's multi-order entries later. See e61452365372 "radix_tree: add support for multi-order entries". -- Kirill A. Shutemov