From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mx2.suse.de ([195.135.220.15]:35182 "EHLO mx1.suse.de" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1726558AbfK0Nyk (ORCPT ); Wed, 27 Nov 2019 08:54:40 -0500 Date: Wed, 27 Nov 2019 14:54:36 +0100 From: David Sterba Subject: Re: [PATCH 5/5] btrfs: Increase buffer size for zlib functions Message-ID: <20191127135436.GR2734@twin.jikos.cz> Reply-To: dsterba@suse.cz References: <20191126144130.75710-1-zaslonko@linux.ibm.com> <20191126144130.75710-6-zaslonko@linux.ibm.com> <20191126155249.j2dktiggykfoz4iz@MacBook-Pro-91.local> <11377b99-b66c-fdc3-5c8f-0bae34c92c03@linux.ibm.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <11377b99-b66c-fdc3-5c8f-0bae34c92c03@linux.ibm.com> Sender: linux-s390-owner@vger.kernel.org List-ID: To: Zaslonko Mikhail Cc: Josef Bacik , Andrew Morton , Chris Mason , David Sterba , Richard Purdie , Heiko Carstens , Vasily Gorbik , Christian Borntraeger , linux-s390@vger.kernel.org, linux-kernel@vger.kernel.org On Wed, Nov 27, 2019 at 02:42:20PM +0100, Zaslonko Mikhail wrote: > Hello, > > On 26.11.2019 16:52, Josef Bacik wrote: > > On Tue, Nov 26, 2019 at 03:41:30PM +0100, Mikhail Zaslonko wrote: > >> Due to the small size of zlib buffer (1 page) set in btrfs code, s390 > >> hardware compression is rather limited in terms of performance. Increasing > >> the buffer size to 4 pages would bring significant benefit for s390 > >> hardware compression (up to 60% better performance compared to the > >> PAGE_SIZE buffer) and should not bring much overhead in terms of memory > >> consumption due to order 2 allocations. > >> > >> Signed-off-by: Mikhail Zaslonko > > > > We may have to make these allocations under memory pressure in the IO context, > > order 2 allocations here is going to be not awesome. If you really want it then > > you need to at least be able to fall back to single page if you fail to get the > > allocation. Thanks, > > As far as I understand GFP_KERNEL allocations would never fail for the order <= > PAGE_ALLOC_COSTLY_ORDER. There's no guaranteed no-fail semantics for GFP flags (obviously besides __GFP_NOFAIL), GFP_KERNEL can fail and GFP_NOFS is unlikely to fail for order below costly allocations. This depends on the allocator internals and has never been an API-level guarantee AFAIK. There's ongoing to work to relax the allocator constraints and allow to fail in more cases (like for GFP_NOFS). > How else can the memory pressure condition be identified > here? All data write paths must consider what happens under memory pressure, because the reason to write the data could be started by an allocation that can get free memory by writing dirty data. So it's kind of implied here.