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 14:28:13 +1100 Message-ID: <20101208032813.GA6812@amd> References: <20101207112555.GE16103@dastard> Mime-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: Dave Chinner , Nick Piggin , linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org, Peter Zijlstra To: Nick Piggin Return-path: Content-Disposition: inline In-Reply-To: Sender: linux-kernel-owner@vger.kernel.org List-Id: linux-fsdevel.vger.kernel.org On Wed, Dec 08, 2010 at 02:24:23AM +1100, Nick Piggin wrote: > On Tue, Dec 7, 2010 at 10:25 PM, Dave Chinner w= rote: > > [ 5954.061633] BUG: sleeping function called from invalid context a= t arch/x86/mm/fault.c:1081 > > [ 5954.062466] in_atomic(): 0, irqs_disabled(): 1, pid: 2927, name:= rm > > [ 5954.063122] 3 locks held by rm/2927: > > [ 5954.063476] =A0#0: =A0(&sb->s_type->i_mutex_key#12/1){+.+.+.}, a= t: [] do_rmdir+0x81/0x130 > > [ 5954.064014] =A0#1: =A0(&sb->s_type->i_mutex_key#12){+.+.+.}, at:= [] vfs_rmdir+0x58/0xe0 > > [ 5954.064014] =A0#2: =A0(rcu_read_lock){.+.+..}, at: [] shrink_dentry_list+0x0/0x430 > > [ 5954.064014] irq event stamp: 1484376719 > > [ 5954.064014] hardirqs last =A0enabled at (1484376719): [] __call_rcu+0xd7/0x1a0 > > [ 5954.064014] hardirqs last disabled at (1484376718): [] __call_rcu+0x4a/0x1a0 > > [ 5954.064014] softirqs last =A0enabled at (1484376586): [] __do_softirq+0x161/0x270 > > [ 5954.064014] softirqs last disabled at (1484376581): [] call_softirq+0x1c/0x50 > > [ 5954.064014] Pid: 2927, comm: rm Not tainted 2.6.37-rc4-dgc+ #794 > > [ 5954.064014] Call Trace: > > [ 5954.064014] =A0[] ? print_irqtrace_events+0xd0= /0xe0 > > [ 5954.064014] =A0[] __might_sleep+0xf5/0x130 > > [ 5954.064014] =A0[] do_page_fault+0x103/0x4f0 > > [ 5954.064014] =A0[] ? pvclock_clocksource_read+0= x58/0xd0 > > [ 5954.064014] =A0[] ? pvclock_clocksource_read+0= x58/0xd0 > > [ 5954.064014] =A0[] ? pvclock_clocksource_read+0= x58/0xd0 > > [ 5954.064014] =A0[] ? trace_hardirqs_off_thunk+0= x3a/0x3c > > [ 5954.064014] =A0[] ? usage_match+0x0/0x20 > > [ 5954.064014] =A0[] page_fault+0x25/0x30 > > [ 5954.064014] =A0[] ? usage_match+0x0/0x20 > > [ 5954.064014] =A0[] ? __bfs+0xc8/0x260 > > [ 5954.064014] =A0[] ? __bfs+0x23/0x260 > > [ 5954.064014] =A0[] find_usage_backwards+0x42/0x= 80 > > [ 5954.064014] =A0[] check_usage_backwards+0x64/0= xf0 > > [ 5954.064014] =A0[] ? save_stack_trace+0x2f/0x50 > > [ 5954.064014] =A0[] ? check_usage_backwards+0x0/= 0xf0 > > [ 5954.064014] =A0[] mark_lock+0x1a9/0x440 > > [ 5954.064014] =A0[] __lock_acquire+0x5a9/0x14b0 > > [ 5954.064014] =A0[] ? __lock_acquire+0x336/0x14b= 0 > > [ 5954.064014] =A0[] ? pvclock_clocksource_read+0= x58/0xd0 > > [ 5954.064014] =A0[] lock_acquire+0xb4/0x140 > > [ 5954.064014] =A0[] ? shrink_dentry_list+0x5c/0x= 430 > > [ 5954.064014] =A0[] _raw_spin_lock+0x36/0x70 > > [ 5954.064014] =A0[] ? shrink_dentry_list+0x5c/0x= 430 > > [ 5954.064014] =A0[] shrink_dentry_list+0x5c/0x43= 0 > > [ 5954.064014] =A0[] ? shrink_dentry_list+0x0/0x4= 30 > > [ 5954.064014] =A0[] ? do_raw_spin_unlock+0x5e/0x= b0 > > [ 5954.064014] =A0[] __shrink_dcache_sb+0x13d/0x1= c0 > > [ 5954.064014] =A0[] shrink_dcache_parent+0x32f/0= x390 > > [ 5954.064014] =A0[] dentry_unhash+0x3d/0x70 > > [ 5954.064014] =A0[] vfs_rmdir+0x60/0xe0 > > [ 5954.064014] =A0[] do_rmdir+0x113/0x130 > > [ 5954.064014] =A0[] ? sysret_check+0x2e/0x69 > > [ 5954.064014] =A0[] ? trace_hardirqs_on_thunk+0x= 3a/0x3f > > [ 5954.064014] =A0[] sys_unlinkat+0x35/0x40 > > [ 5954.064014] =A0[] system_call_fastpath+0x16/0x= 1b >=20 > Seems that lockdep exploded. >=20 > > [ 5954.092916] BUG: unable to handle kernel NULL pointer dereferenc= e at =A0 =A0 =A0 =A0 =A0 (null) > > [ 5954.093806] IP: [] __bfs+0xc8/0x260 > > [ 5954.094331] PGD 1084e5067 PUD 102368067 PMD 0 > > [ 5954.094830] Oops: 0000 [#1] SMP > > [ 5954.095194] last sysfs file: /sys/devices/system/cpu/online > > [ 5954.095760] CPU 6 > > [ 5954.095954] Modules linked in: > > [ 5954.096319] > > [ 5954.096483] Pid: 2927, comm: rm Not tainted 2.6.37-rc4-dgc+ #794= /Bochs > > [ 5954.096665] RIP: 0010:[] =A0[] __bfs+0xc8/0x260 > > [ 5954.096665] RSP: 0018:ffff8801175539a8 =A0EFLAGS: 00010046 > > [ 5954.096665] RAX: ffffffff8267d980 RBX: ffffffff8267d980 RCX: fff= f880117553a48 > > [ 5954.096665] RDX: ffff8801175539d0 RSI: 0000000000000000 RDI: fff= f880117553a48 > > [ 5954.096665] RBP: ffff880117553a08 R08: 0000000000000000 R09: 000= 0000000000000 > > [ 5954.096665] R10: 0000000000000000 R11: 0000000000000001 R12: 000= 0000000000000 > > [ 5954.096665] R13: ffffffff810b8e20 R14: ffff880117553a90 R15: 000= 0000000000000 > > [ 5954.096665] FS: =A000007f4594cf3700(0000) GS:ffff8800dfa00000(00= 00) knlGS:0000000000000000 > > [ 5954.096665] CS: =A00010 DS: 0000 ES: 0000 CR0: 000000008005003b > > [ 5954.096665] CR2: 00007f2f21e89c60 CR3: 0000000110b0f000 CR4: 000= 00000000006e0 > > [ 5954.096665] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 000= 0000000000000 > > [ 5954.096665] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 000= 0000000000400 > > [ 5954.096665] Process rm (pid: 2927, threadinfo ffff880117552000, = task ffff88010ff10b00) > > [ 5954.096665] Stack: > > [ 5954.096665] =A0ffffffff8267d868 00007fffa02d1428 ffff8800fffffff= f ffff880100000000 > > [ 5954.096665] =A0000000000000b720 ffff880117553a48 ffffffff8267d86= 8 ffff880117553a48 > > [ 5954.096665] =A00000000000000000 ffff88010ff10b00 000000000000000= 0 ffffffff81dacba0 > > [ 5954.096665] Call Trace: > > [ 5954.096665] =A0[] find_usage_backwards+0x42/0x= 80 > > [ 5954.096665] =A0[] check_usage_backwards+0x64/0= xf0 > > [ 5954.096665] =A0[] ? save_stack_trace+0x2f/0x50 > > [ 5954.096665] =A0[] ? check_usage_backwards+0x0/= 0xf0 > > [ 5954.096665] =A0[] mark_lock+0x1a9/0x440 > > [ 5954.096665] =A0[] __lock_acquire+0x5a9/0x14b0 > > [ 5954.096665] =A0[] ? __lock_acquire+0x336/0x14b= 0 > > [ 5954.096665] =A0[] ? pvclock_clocksource_read+0= x58/0xd0 > > [ 5954.096665] =A0[] lock_acquire+0xb4/0x140 > > [ 5954.096665] =A0[] ? shrink_dentry_list+0x5c/0x= 430 > > [ 5954.096665] =A0[] _raw_spin_lock+0x36/0x70 > > [ 5954.096665] =A0[] ? shrink_dentry_list+0x5c/0x= 430 > > [ 5954.096665] =A0[] shrink_dentry_list+0x5c/0x43= 0 > > [ 5954.096665] =A0[] ? shrink_dentry_list+0x0/0x4= 30 > > [ 5954.096665] =A0[] ? do_raw_spin_unlock+0x5e/0x= b0 > > [ 5954.096665] =A0[] __shrink_dcache_sb+0x13d/0x1= c0 > > [ 5954.096665] =A0[] shrink_dcache_parent+0x32f/0= x390 > > [ 5954.096665] =A0[] dentry_unhash+0x3d/0x70 > > [ 5954.096665] =A0[] vfs_rmdir+0x60/0xe0 > > [ 5954.096665] =A0[] do_rmdir+0x113/0x130 > > [ 5954.096665] =A0[] ? sysret_check+0x2e/0x69 > > [ 5954.096665] =A0[] ? trace_hardirqs_on_thunk+0x= 3a/0x3f > > [ 5954.096665] =A0[] sys_unlinkat+0x35/0x40 > > [ 5954.096665] =A0[] system_call_fastpath+0x16/0x= 1b > > [ 5954.096665] Code: 0a 89 05 dc 0f a8 01 48 8b 41 10 48 85 c0 0f 8= 4 1f 01 00 00 48 8d 98 70 01 00 00 48 05 80 01 00 00 45 85 c0 48 0f 44 = d8 4c 8b 3b <49> 8b 07 49 39 > > [ 5954.096665] RIP =A0[] __bfs+0xc8/0x260 > > [ 5954.096665] =A0RSP > > [ 5954.096665] CR2: 0000000000000000 > > [ 5954.127991] ---[ end trace 85a6727c2d4e3d90 ]--- >=20 > So vfs-scale-working branch may not be entirely in the clear, seeing Ah, may have been a stupid little bug. The list entry check was being d= one and then the pointer reloaded to be used. What does the asm for shrink_dentry_list look like (before this patch)? Thanks, Nick Index: linux-2.6/fs/dcache.c =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D --- linux-2.6.orig/fs/dcache.c 2010-12-08 12:41:35.000000000 +1100 +++ linux-2.6/fs/dcache.c 2010-12-08 14:23:46.000000000 +1100 @@ -657,10 +657,10 @@ static void shrink_dentry_list(struct li struct dentry *dentry; =20 rcu_read_lock(); - while (!list_empty(list)) { - dentry =3D list_entry(list->prev, struct dentry, d_lru); - - /* Don't need RCU dereference because we recheck under lock */ + for (;;) { + dentry =3D list_entry_rcu(list->prev, struct dentry, d_lru); + if (&dentry->d_lru =3D=3D list) + break; spin_lock(&dentry->d_lock); if (dentry !=3D list_entry(list->prev, struct dentry, d_lru)) { spin_unlock(&dentry->d_lock);