From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-pd0-f173.google.com (mail-pd0-f173.google.com [209.85.192.173]) by kanga.kvack.org (Postfix) with ESMTP id 11A4A6B00AC for ; Tue, 3 Dec 2013 19:59:13 -0500 (EST) Received: by mail-pd0-f173.google.com with SMTP id p10so21217853pdj.18 for ; Tue, 03 Dec 2013 16:59:13 -0800 (PST) Received: from mail.linuxfoundation.org (mail.linuxfoundation.org. [140.211.169.12]) by mx.google.com with ESMTP id bo6si30667650pab.85.2013.12.03.16.59.11 for ; Tue, 03 Dec 2013 16:59:12 -0800 (PST) Date: Tue, 3 Dec 2013 16:59:10 -0800 From: Andrew Morton Subject: Re: [patch 2/2] fs: buffer: move allocation failure loop into the allocator Message-Id: <20131203165910.54d6b4724a1f3e329af52ac6@linux-foundation.org> In-Reply-To: <1381265890-11333-2-git-send-email-hannes@cmpxchg.org> References: <1381265890-11333-1-git-send-email-hannes@cmpxchg.org> <1381265890-11333-2-git-send-email-hannes@cmpxchg.org> Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Sender: owner-linux-mm@kvack.org List-ID: To: Johannes Weiner Cc: Michal Hocko , azurIt , linux-mm@kvack.org, cgroups@vger.kernel.org, linux-kernel@vger.kernel.org, Christian Casteyde , Pekka Enberg , Christoph Lameter On Tue, 8 Oct 2013 16:58:10 -0400 Johannes Weiner wrote: > Buffer allocation has a very crude indefinite loop around waking the > flusher threads and performing global NOFS direct reclaim because it > can not handle allocation failures. > > The most immediate problem with this is that the allocation may fail > due to a memory cgroup limit, where flushers + direct reclaim might > not make any progress towards resolving the situation at all. Because > unlike the global case, a memory cgroup may not have any cache at all, > only anonymous pages but no swap. This situation will lead to a > reclaim livelock with insane IO from waking the flushers and thrashing > unrelated filesystem cache in a tight loop. > > Use __GFP_NOFAIL allocations for buffers for now. This makes sure > that any looping happens in the page allocator, which knows how to > orchestrate kswapd, direct reclaim, and the flushers sensibly. It > also allows memory cgroups to detect allocations that can't handle > failure and will allow them to ultimately bypass the limit if reclaim > can not make progress. Problem. > --- a/fs/buffer.c > +++ b/fs/buffer.c > @@ -1005,9 +1005,19 @@ grow_dev_page(struct block_device *bdev, sector_t block, > struct buffer_head *bh; > sector_t end_block; > int ret = 0; /* Will call free_more_memory() */ > + gfp_t gfp_mask; > > - page = find_or_create_page(inode->i_mapping, index, > - (mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS)|__GFP_MOVABLE); > + gfp_mask = mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS; > + gfp_mask |= __GFP_MOVABLE; https://bugzilla.kernel.org/show_bug.cgi?id=65991 WARNING: CPU: 0 PID: 1 at mm/page_alloc.c:1539 get_page_from_freelist+0x8a9/0x8c0() Modules linked in: CPU: 0 PID: 1 Comm: swapper/0 Not tainted 3.13.0-rc1 #42 Hardware name: Acer Aspire 7750G/JE70_HR, BIOS V1.07 03/02/2011 0000000000000009 ffff8801c6121650 ffffffff81898d39 0000000000000000 ffff8801c6121688 ffffffff8107dc43 0000000000000002 0000000000000001 0000000000284850 0000000000000000 ffff8801cec04680 ffff8801c6121698 Call Trace: [] dump_stack+0x4e/0x7a [] warn_slowpath_common+0x73/0x90 [] warn_slowpath_null+0x15/0x20 [] get_page_from_freelist+0x8a9/0x8c0 [] ? trace_hardirqs_off_thunk+0x3a/0x3c [] __alloc_pages_nodemask+0xf0/0x770 [] ? trace_hardirqs_off_thunk+0x3a/0x3c [] kmemcheck_alloc_shadow+0x53/0xf0 [] new_slab+0x345/0x3e0 [] __slab_alloc.isra.57+0x215/0x535 [] ? __radix_tree_preload+0x60/0xf0 [] kmem_cache_alloc+0x118/0x150 [] ? __radix_tree_preload+0x60/0xf0 [] __radix_tree_preload+0x60/0xf0 [] radix_tree_maybe_preload+0x25/0x30 [] add_to_page_cache_locked+0x37/0x100 [] add_to_page_cache_lru+0x15/0x40 [] find_or_create_page+0x57/0x90 [] __getblk+0xf0/0x2f0 That __GFP_NOFAIL is getting down into radix_tree_preload->kmem_cache_alloc() and I expect that in its boundless stupidity, slab has decided to inappropriately go and use an unnecessarily massive page size for radix_tree_node_cachep's underlying memory allocations. So we end up using GFP_NOFAIL for an order=2 (or more) allocation, which is unacceptably risky, methinks. I really really wish slab wouldn't do this. The benefit is surely very small and these unnecessary higher-order allocations are quite abusive of the page allocator. Can we please make slab stop doing this? radix_tree_nodes are 560 bytes and the kernel often allocates them in times of extreme memory stress. We really really want them to be backed by order=0 pages. -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@kvack.org. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: email@kvack.org