Return-Path: Received: from userp1040.oracle.com ([156.151.31.81]:31466 "EHLO userp1040.oracle.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750905AbbCTA7W convert rfc822-to-8bit (ORCPT ); Thu, 19 Mar 2015 20:59:22 -0400 Content-Type: text/plain; charset=us-ascii Mime-Version: 1.0 (Mac OS X Mail 7.3 \(1878.6\)) Subject: Re: [PATCH] nfs: don't call blocking operations while !TASK_RUNNING From: Chuck Lever In-Reply-To: <20150114133204.0ea6140f@tlielax.poochiereds.net> Date: Thu, 19 Mar 2015 17:59:15 -0700 Cc: Trond Myklebust , "J. Bruce Fields" , Linux NFS Mailing List Message-Id: <20C39C1C-7697-486C-A4B2-2379158232E6@oracle.com> References: <1421249572-12038-1-git-send-email-jlayton@primarydata.com> <20150114122751.0e1ac594@tlielax.poochiereds.net> <20150114133204.0ea6140f@tlielax.poochiereds.net> To: Jeff Layton Sender: linux-nfs-owner@vger.kernel.org List-ID: On Jan 14, 2015, at 10:32 AM, Jeff Layton wrote: > On Wed, 14 Jan 2015 13:06:17 -0500 > Trond Myklebust wrote: > >> On Wed, Jan 14, 2015 at 12:27 PM, Jeff Layton >> wrote: >>> On Wed, 14 Jan 2015 12:13:43 -0500 >>> Trond Myklebust wrote: >>> >>>> On Wed, Jan 14, 2015 at 10:32 AM, Jeff Layton wrote: >>>>> Bruce reported seeing this warning pop when mounting using v4.1: >>>>> >>>>> ------------[ cut here ]------------ >>>>> WARNING: CPU: 1 PID: 1121 at kernel/sched/core.c:7300 __might_sleep+0xbd/0xd0() >>>>> do not call blocking ops when !TASK_RUNNING; state=1 set at [] prepare_to_wait+0x2f/0x90 >>>>> Modules linked in: rpcsec_gss_krb5 auth_rpcgss nfsv4 dns_resolver nfs lockd grace sunrpc fscache ip6t_rpfilter ip6t_REJECT nf_reject_ipv6 xt_conntrack ebtable_nat ebtable_broute bridge stp llc ebtable_filter ebtables ip6table_nat nf_conntrack_ipv6 nf_defrag_ipv6 nf_nat_ipv6 ip6table_mangle ip6table_security ip6table_raw ip6table_filter ip6_tables iptable_nat nf_conntrack_ipv4 nf_defrag_ipv4 nf_nat_ipv4 nf_nat nf_conntrack iptable_mangle iptable_security iptable_raw snd_hda_codec_generic snd_hda_intel snd_hda_controller snd_hda_codec snd_hwdep snd_pcm snd_timer ppdev joydev snd virtio_console virtio_balloon pcspkr serio_raw parport_pc parport pvpanic floppy soundcore i2c_piix4 virtio_blk virtio_net qxl drm_kms_helper ttm drm virtio_pci virtio_ring ata_generic virtio pata_acpi >>>>> CPU: 1 PID: 1121 Comm: nfsv4.1-svc Not tainted 3.19.0-rc4+ #25 >>>>> Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.7.5-20140709_153950- 04/01/2014 >>>>> 0000000000000000 000000004e5e3f73 ffff8800b998fb48 ffffffff8186ac78 >>>>> 0000000000000000 ffff8800b998fba0 ffff8800b998fb88 ffffffff810ac9da >>>>> ffff8800b998fb68 ffffffff81c923e7 00000000000004d9 0000000000000000 >>>>> Call Trace: >>>>> [] dump_stack+0x4c/0x65 >>>>> [] warn_slowpath_common+0x8a/0xc0 >>>>> [] warn_slowpath_fmt+0x55/0x70 >>>>> [] ? prepare_to_wait+0x2f/0x90 >>>>> [] ? prepare_to_wait+0x2f/0x90 >>>>> [] __might_sleep+0xbd/0xd0 >>>>> [] kmem_cache_alloc_trace+0x243/0x430 >>>>> [] ? groups_alloc+0x3e/0x130 >>>>> [] groups_alloc+0x3e/0x130 >>>>> [] svcauth_unix_accept+0x16e/0x290 [sunrpc] >>>>> [] svc_authenticate+0xe1/0xf0 [sunrpc] >>>>> [] svc_process_common+0x244/0x6a0 [sunrpc] >>>>> [] bc_svc_process+0x1c4/0x260 [sunrpc] >>>>> [] nfs41_callback_svc+0x128/0x1f0 [nfsv4] >>>>> [] ? wait_woken+0xc0/0xc0 >>>>> [] ? nfs4_callback_svc+0x60/0x60 [nfsv4] >>>>> [] kthread+0x11f/0x140 >>>>> [] ? local_clock+0x15/0x30 >>>>> [] ? kthread_create_on_node+0x250/0x250 >>>>> [] ret_from_fork+0x7c/0xb0 >>>>> [] ? kthread_create_on_node+0x250/0x250 >>>>> ---[ end trace 675220a11e30f4f2 ]--- >>>>> >>>>> nfs41_callback_svc does most of its work while in TASK_INTERRUPTIBLE, >>>>> which is just wrong. Fix that by finishing the wait immediately if we've >>>>> found that the list has something on it. >>>> >>>> ACK. >>>> >>>>> Also, we don't expect this kthread to accept signals, so we should be >>>>> using a TASK_UNINTERRUPTIBLE sleep instead. >>>> >>>> Umm... Won't that end up triggering the hung task watchdog for every >>>> 120seconds with no callback activity? >>>> >>> >>> Doh! You're correct. >>> >>> What's the right way to do this then? Do we need to use >>> schedule_timeout and wake up every 100s or so? Using TASK_INTERRUPTIBLE >>> to work around the watchdog seems wrong but I guess we can live with >>> that in the short term if it's the only way. >> >> The alternative is to use TASK_KILLABLE. That's a little more >> restrictive, but still a PITA. Note the both TASK_INTERRUPTIBLE and >> TASK_KILLABLE will need something to handle that signalled() case, >> otherwise we end up with a permanent busy-waiting loop. >> > > Heh, our emails appear to have crossed. I went with a schedule_timeout > for v2 so we can sidestep the whole issue of signals. > > It's simple enough to add a flush_signals() call at the end of the > loop, but I'm a little leery of what effects we might see if the thread > catches a signal while in the middle of doing real work. > > If you really think that's the better approach, I can do a v3 patch > that does that however. After applying this patch to a 3.19.0 test system, the WARNINGS are gone, but the system load average is pinned at 1.00. root 12105 0.1 0.0 0 0 ? D Mar18 1:19 [nfsv4.1-svc] -- Chuck Lever chuck[dot]lever[at]oracle[dot]com