From mboxrd@z Thu Jan 1 00:00:00 1970 From: Nick Piggin Subject: Re: [PATCH 00/46] rcu-walk and dcache scaling Date: Wed, 8 Dec 2010 18:09:09 +1100 Message-ID: <20101208070909.GB14846@amd> References: <20101207215653.GA25864@dastard> <20101208033212.GF29333@dastard> <20101208042816.GA32766@dastard> Mime-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: Nick Piggin , Nick Piggin , linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org To: Dave Chinner Return-path: Received: from ipmail07.adl2.internode.on.net ([150.101.137.131]:3566 "EHLO ipmail07.adl2.internode.on.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753607Ab0LHHJ1 (ORCPT ); Wed, 8 Dec 2010 02:09:27 -0500 Content-Disposition: inline In-Reply-To: <20101208042816.GA32766@dastard> Sender: linux-fsdevel-owner@vger.kernel.org List-ID: On Wed, Dec 08, 2010 at 03:28:16PM +1100, Dave Chinner wrote: > On Wed, Dec 08, 2010 at 02:32:12PM +1100, Dave Chinner wrote: > > On Wed, Dec 08, 2010 at 12:47:42PM +1100, Nick Piggin wrote: > > > On Wed, Dec 8, 2010 at 8:56 AM, Dave Chinner wrote: > > > > On Sat, Nov 27, 2010 at 09:15:58PM +1100, Nick Piggin wrote: > > > >> > > > >> git://git.kernel.org/pub/scm/linux/kernel/git/npiggin/linux-np= iggin.git vfs-scale-working > > > >> > > > >> Here is an new set of vfs patches for review, not that there w= as much interest > > > >> last time they were posted. It is structured like: > > > >> > > > >> * preparation patches > > > >> * introduce new locks to take over dcache_lock, then remove it > > > >> * cleaning up and reworking things for new locks > > > >> * rcu-walk path walking > > > >> * start on some fine grained locking steps > > > > > > > > Stress test doing: > > > > > > > > =A0 =A0 =A0 =A0single thread 50M inode create > > > > =A0 =A0 =A0 =A0single thread rm -rf > > > > =A0 =A0 =A0 =A02-way 50M inode create > > > > =A0 =A0 =A0 =A02-way rm -rf > > > > =A0 =A0 =A0 =A04-way 50M inode create > > > > =A0 =A0 =A0 =A04-way rm -rf > > > > =A0 =A0 =A0 =A08-way 50M inode create > > > > =A0 =A0 =A0 =A08-way rm -rf > > > > =A0 =A0 =A0 =A08-way 250M inode create > > > > =A0 =A0 =A0 =A08-way rm -rf > > > > > > > > Failed about 5 minutes into the "4-way rm -rf" (~3 hours into t= he test) > > > > with a CPU stuck spinning on here: > > > > > > > > [37372.084012] NMI backtrace for cpu 5 > > > > [37372.084012] CPU 5 > > > > [37372.084012] Modules linked in: > > > > [37372.084012] > > > > [37372.084012] Pid: 15214, comm: rm Not tainted 2.6.37-rc4-dgc+= #797 /Bochs > > > > [37372.084012] RIP: 0010:[] =A0[] __ticket_spin_lock+0x14/0x20 > > > > [37372.084012] RSP: 0018:ffff880114643c98 =A0EFLAGS: 00000213 > > > > [37372.084012] RAX: 0000000000008801 RBX: ffff8800687be6c0 RCX:= ffff8800c4eb2688 > > > > [37372.084012] RDX: ffff880114643d38 RSI: ffff8800dfd4ea80 RDI:= ffff880114643d14 > > > > [37372.084012] RBP: ffff880114643c98 R08: 0000000000000003 R09:= 0000000000000000 > > > > [37372.084012] R10: 0000000000000000 R11: dead000000200200 R12:= ffff880114643d14 > > > > [37372.084012] R13: ffff880114643cb8 R14: ffff880114643d38 R15:= ffff8800687be71c > > > > [37372.084012] FS: =A000007fd6d7c93700(0000) GS:ffff8800dfd4000= 0(0000) knlGS:0000000000000000 > > > > [37372.084012] CS: =A00010 DS: 0000 ES: 0000 CR0: 0000000080050= 03b > > > > [37372.084012] CR2: 0000000000bbd108 CR3: 0000000107146000 CR4:= 00000000000006e0 > > > > [37372.084012] DR0: 0000000000000000 DR1: 0000000000000000 DR2:= 0000000000000000 > > > > [37372.084012] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7:= 0000000000000400 > > > > [37372.084012] Process rm (pid: 15214, threadinfo ffff880114642= 000, task ffff88011b16f890) > > > > [37372.084012] Stack: > > > > [37372.084012] =A0ffff880114643ca8 ffffffff81ad044e ffff8801146= 43cf8 ffffffff81167ae7 > > > > [37372.084012] =A00000000000000000 ffff880114643d38 00000000000= 0000e ffff88011901d800 > > > > [37372.084012] =A0ffff8800cdb7cf5c ffff88011901d8e0 00000000000= 00000 0000000000000000 > > > > [37372.084012] Call Trace: > > > > [37372.084012] =A0[] _raw_spin_lock+0xe/0x20 > > > > [37372.084012] =A0[] shrink_dentry_list+0x47/= 0x370 > > > > [37372.084012] =A0[] __shrink_dcache_sb+0x14e= /0x1e0 > > > > [37372.084012] =A0[] shrink_dcache_parent+0x2= 76/0x2d0 > > > > [37372.084012] =A0[] ? _raw_spin_lock+0xe/0x2= 0 > > > > [37372.084012] =A0[] dentry_unhash+0x42/0x80 > > > > [37372.084012] =A0[] vfs_rmdir+0x68/0x100 > > > > [37372.084012] =A0[] do_rmdir+0x113/0x130 > > > > [37372.084012] =A0[] ? filp_close+0x5d/0x90 > > > > [37372.084012] =A0[] sys_unlinkat+0x35/0x40 > > > > [37372.084012] =A0[] system_call_fastpath+0x1= 6/0x1b > > >=20 > > > OK good, with any luck, that's the same bug. > > >=20 > > > Is this XFS? > >=20 > > Yes. > >=20 > > > Is there any concurrent activity happening on the same dentries? > >=20 > > Not from an application perspective. > >=20 > > > Ie. are the rm -rf threads running on the same directories, > >=20 > > No, each thread operating on a different directory. This is probably fixed by the same patch as the lockdep splat trace. > > > or is there any reclaim happening in the background? > >=20 > > IIRC, kswapd was consuming about 5-10% of a CPU during parallel > > unlink tests. Mainly reclaiming XFS inodes, I think, but there may > > be dentry cache reclaim going as well. >=20 > Turns out that the kswapd peaks are upwards of 50% of a CPU for a > few seconds, then idle for 10-15s. Typical perf top output of kswapd > while it is active during unlinks is: >=20 > samples pcnt function DSO > _______ _____ ___________________________ ______________= ___ >=20 > 17168.00 10.2% __call_rcu [kernel.kallsy= ms] > 13223.00 7.8% kmem_cache_free [kernel.kallsy= ms] > 12917.00 7.6% down_write [kernel.kallsy= ms] > 12665.00 7.5% xfs_iunlock [kernel.kallsy= ms] > 10493.00 6.2% xfs_reclaim_inode_grab [kernel.kallsy= ms] > 9314.00 5.5% __lookup_tag [kernel.kallsy= ms] > 9040.00 5.4% radix_tree_delete [kernel.kallsy= ms] > 8694.00 5.1% is_bad_inode [kernel.kallsy= ms] > 7639.00 4.5% __ticket_spin_lock [kernel.kallsy= ms] > 6821.00 4.0% _raw_spin_unlock_irqrestore [kernel.kallsy= ms] > 5484.00 3.2% __d_drop [kernel.kallsy= ms] > 5114.00 3.0% xfs_reclaim_inode [kernel.kallsy= ms] > 4626.00 2.7% __rcu_process_callbacks [kernel.kallsy= ms] > 3556.00 2.1% up_write [kernel.kallsy= ms] > 3206.00 1.9% _cond_resched [kernel.kallsy= ms] > 3129.00 1.9% xfs_qm_dqdetach [kernel.kallsy= ms] > 2327.00 1.4% radix_tree_tag_clear [kernel.kallsy= ms] > 2327.00 1.4% call_rcu_sched [kernel.kallsy= ms] > 2262.00 1.3% __ticket_spin_unlock [kernel.kallsy= ms] > 2215.00 1.3% xfs_ilock [kernel.kallsy= ms] > 2200.00 1.3% radix_tree_gang_lookup_tag [kernel.kallsy= ms] > 1982.00 1.2% xfs_reclaim_inodes_ag [kernel.kallsy= ms] > 1736.00 1.0% xfs_trans_unlocked_item [kernel.kallsy= ms] > 1707.00 1.0% __ticket_spin_trylock [kernel.kallsy= ms] > 1688.00 1.0% xfs_perag_get_tag [kernel.kallsy= ms] > 1660.00 1.0% flat_send_IPI_mask [kernel.kallsy= ms] > 1538.00 0.9% xfs_inode_item_destroy [kernel.kallsy= ms] > 1312.00 0.8% __shrink_dcache_sb [kernel.kallsy= ms] > 940.00 0.6% xfs_perag_put [kernel.kallsy= ms] >=20 > So there is some dentry cache reclaim going on.=20 >=20 > FWIW, it appears there is quite a lot of RCU freeing overhead (~15% > more CPU time) in the work kswapd is doing during these unlinks, too. > I just had a look at kswapd when a 8-way create is running - it's run= ning at > 50-60% of a cpu for seconds at a time. I caught this while it was doi= ng pure > XFS inode cache reclaim (~10s sample, kswapd reclaimed ~1M inodes): >=20 > samples pcnt function DSO > _______ _____ ___________________________ ______________= ___ >=20 > 27171.00 9.0% __call_rcu [kernel.kallsy= ms] > 21491.00 7.1% down_write [kernel.kallsy= ms] > 20916.00 6.9% xfs_reclaim_inode [kernel.kallsy= ms] > 20313.00 6.7% radix_tree_delete [kernel.kallsy= ms] > 15828.00 5.3% kmem_cache_free [kernel.kallsy= ms] > 15819.00 5.2% xfs_idestroy_fork [kernel.kallsy= ms] > 14893.00 4.9% is_bad_inode [kernel.kallsy= ms] > 14666.00 4.9% _raw_spin_unlock_irqrestore [kernel.kallsy= ms] > 14191.00 4.7% xfs_reclaim_inode_grab [kernel.kallsy= ms] > 14105.00 4.7% xfs_iunlock [kernel.kallsy= ms] > 10916.00 3.6% __ticket_spin_lock [kernel.kallsy= ms] > 10125.00 3.4% xfs_iflush_cluster [kernel.kallsy= ms] > 8221.00 2.7% xfs_qm_dqdetach [kernel.kallsy= ms] > 7639.00 2.5% xfs_trans_unlocked_item [kernel.kallsy= ms] > 7028.00 2.3% xfs_synchronize_times [kernel.kallsy= ms] > 6974.00 2.3% up_write [kernel.kallsy= ms] > 5870.00 1.9% call_rcu_sched [kernel.kallsy= ms] > 5634.00 1.9% _cond_resched [kernel.kallsy= ms] >=20 > Which is showing a similar amount of RCU overhead as the unlink as ab= ove. > And this while it was doing dentry cache reclaim (~10s sample): >=20 > 35921.00 15.7% __d_drop [kernel.kall= syms] > 30056.00 13.1% __ticket_spin_trylock [kernel.kall= syms] > 29066.00 12.7% __ticket_spin_lock [kernel.kall= syms] > 19043.00 8.3% __call_rcu [kernel.kall= syms] > 10098.00 4.4% iput [kernel.kall= syms] > 7013.00 3.1% __shrink_dcache_sb [kernel.kall= syms] > 6774.00 3.0% __percpu_counter_add [kernel.kall= syms] > 6708.00 2.9% radix_tree_tag_set [kernel.kall= syms] > 5362.00 2.3% xfs_inactive [kernel.kall= syms] > 5130.00 2.2% __ticket_spin_unlock [kernel.kall= syms] > 4884.00 2.1% call_rcu_sched [kernel.kall= syms] > 4621.00 2.0% dentry_lru_del [kernel.kall= syms] > 3735.00 1.6% bit_waitqueue [kernel.kall= syms] > 3727.00 1.6% dentry_iput [kernel.kall= syms] > 3473.00 1.5% shrink_icache_memory [kernel.kall= syms] > 3279.00 1.4% kfree [kernel.kall= syms] > 3101.00 1.4% xfs_perag_get [kernel.kall= syms] > 2516.00 1.1% kmem_cache_free [kernel.kall= syms] > 2272.00 1.0% shrink_dentry_list [kernel.kall= syms] >=20 > I've never really seen any signficant dentry cache reclaim overhead > in profiles of these workloads before, so this was a bit of a > surprise.... call_rcu shouldn't be doing much, except for disabling irqs and linking the object into the list. I have a patch somewhere to reduce the irq disable overhead a bit, but it really shouldn't be doing a lot of work. Sometimes you find that touching the rcu head field needs to get a cacheline exclusive, so a bit of work gets transferred there.... But it may also be something going a bit wrong in RCU. I blew it up once already, after the files_lock splitup that enabled all CPUs to create and destroy files :) -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel= " in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html