From mboxrd@z Thu Jan 1 00:00:00 1970 From: Mathieu Desnoyers Subject: Re: [Bug #13475] suspend/hibernate lockdep warning Date: Thu, 11 Jun 2009 11:23:29 -0400 Message-ID: <20090611152329.GB28099@Krystal> References: <84144f020906070621r1f480eaeief026d23662df380@mail.gmail.com> <1244447366.13471.4.camel@penberg-laptop> <20090608124844.GA17588@Krystal> <20090608143220.GC2516@redhat.com> <1244727561.5350.32.camel@odie.local> Mime-Version: 1.0 Content-Transfer-Encoding: QUOTED-PRINTABLE Return-path: Content-Disposition: inline In-Reply-To: <1244727561.5350.32.camel-78RDdhuQolGs1BDpvl8NfQ@public.gmane.org> Sender: kernel-testers-owner-u79uwXL29TY76Z2rM5mHXA@public.gmane.org List-ID: Content-Type: text/plain; charset="iso-8859-1" To: Simon Holm =?iso-8859-1?Q?Th=F8gersen?= Cc: Dave Jones , Pekka Enberg , Dave Young , "Rafael J. Wysocki" , Linux Kernel Mailing List , Kernel Testers List , cpufreq-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, Rusty Russell , trenn-l3A5Bk7waGM@public.gmane.org, sven.wegener-sQQoR7IzGU7R7s880joybQ@public.gmane.org, Venkatesh Pallipadi * Simon Holm Th=F8gersen (odie-t5LvXY1cjzpaa/9Udqfwiw@public.gmane.org) wrote: > man, 08 06 2009 kl. 10:32 -0400, skrev Dave Jones:=20 > > On Mon, Jun 08, 2009 at 08:48:45AM -0400, Mathieu Desnoyers wrote: > > =20 > > > > > >> Bug-Entry : http://bugzilla.kernel.org/show_bug.cgi= ?id=3D13475 > > > > > >> Subject : suspend/hibernate lockdep warning > > > > > >> References : http://marc.info/?l=3Dlinux-kernel&m=3D= 124393723321241&w=3D4 > > > > >=20 > > > > > I suspect the following commit, after revert this patch I te= st 5 times > > > > > without lockdep warnings. > > > > >=20 > > > > > commit b14893a62c73af0eca414cfed505b8c09efc613c > > > > > Author: Mathieu Desnoyers > > > > > Date: Sun May 17 10:30:45 2009 -0400 > > > > >=20 > > > > > [CPUFREQ] fix timer teardown in ondemand governor > > > >=20 > > > > The patch is probably not at fault here. I suspect it's some l= atent bug > > > > that simply got exposed by the change to cancel_delayed_work_s= ync(). In > > > > any case, Mathieu, can you take a look at this please? > > >=20 > > > Yes, it's been looked at and discussed on the cpufreq ML. The sh= ort > > > answer is that they plan to re-engineer cpufreq and remove the p= olicy > > > rwlock taken around almost every operations at the cpufreq level= =2E > > >=20 > > > The short-term solution, which is recognised as ugly, would be d= o to the > > > following before doing the cancel_delayed_work_sync() : > > >=20 > > > unlock policy rwlock write lock > > >=20 > > > lock policy rwlock write lock > > >=20 > > > It basically works because this rwlock is unneeded for teardown,= hence > > > the future re-work planned. > > >=20 > > > I'm sorry I cannot prepare a patch current... I've got quite a f= ew pages > > > of Ph.D. thesis due for the beginning of July. > > =20 > > I'm kinda scared to touch this code at all for .30 due to the numbe= r of > > unexpected gotchas we seem to run into every time we touch somethin= g > > locking related. So I'm inclined to just live with the lockdep war= ning > > for .30, and see how the real fixes look for .31, and push them bac= k > > as -stable updates if they work out. >=20 > Unfortunately I don't think it is just theoretical, I've actually hit > the following (that haven't got anything to do with suspend/hibernate= ) >=20 > INFO: task cpufreqd:4676 blocked for more than 120 seconds. > "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this mes= sage. > cpufreqd D eee2ac60 0 4676 1 > ee01bd68 00000086 eee2aad0 eee2ac60 00000533 eee2aad0 eee2ac60 0002= b16f > 00000000 eee2ac60 7fffffff 7fffffff eee2ac60 7fffffff 7fffffff 0000= 0000 > ee01bd70 c03117ee ee01bdbc c0311c0c eee2aad0 eecf6900 eee2aad0 eecf= 6900 > Call Trace: > [] schedule+0x12/0x24 > [] schedule_timeout+0x17/0x170 > [] ? __wake_up+0x2b/0x51 > [] wait_for_common+0xc4/0x135 > [] ? default_wake_function+0x0/0xd > [] wait_for_completion+0x12/0x14 > [] __cancel_work_timer+0xfe/0x129 > [] ? wq_barrier_func+0x0/0xd > [] cancel_delayed_work_sync+0xb/0xd > [] cpufreq_governor_dbs+0x22e/0x291 [cpufreq_ondemand] > [] __cpufreq_governor+0x65/0x9d > [] __cpufreq_set_policy+0xd1/0x11f > [] store_scaling_governor+0x18a/0x1b2 > [] ? handle_update+0x0/0xd > [] ? store_scaling_governor+0x0/0x1b2 > [] store+0x48/0x61 > [] sysfs_write_file+0xb4/0xdf > [] ? sysfs_write_file+0x0/0xdf > [] vfs_write+0x8a/0x104 > [] sys_write+0x3b/0x60 > [] sysenter_do_call+0x12/0x2c > INFO: task kondemand/0:4956 blocked for more than 120 seconds. > "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this mes= sage. > kondemand/0 D 00000533 0 4956 2 > ee1d9efc 00000046 c011815f 00000533 071148de ee1e0080 ee1e0210 0000= 0000 > c03ff478 9189e633 00000082 c03ff478 ee1e0210 c04159f4 c04159f0 0000= 0000 > ee1d9f04 c03117ee ee1d9f28 c0313104 ee1d9f30 c04159f4 ee1e0080 c011= 83be > Call Trace: > [] ? update_curr+0x6c/0x14b > [] schedule+0x12/0x24 > [] rwsem_down_failed_common+0x150/0x16e > [] ? dequeue_task_fair+0x51/0x56 > [] rwsem_down_write_failed+0x1b/0x23 > [] call_rwsem_down_write_failed+0x6/0x8 > [] ? down_write+0x14/0x16 > [] lock_policy_rwsem_write+0x1d/0x33 > [] do_dbs_timer+0x45/0x266 [cpufreq_ondemand] > [] worker_thread+0x165/0x212 > [] ? do_dbs_timer+0x0/0x266 [cpufreq_ondemand] > [] ? autoremove_wake_function+0x0/0x33 > [] ? worker_thread+0x0/0x212 > [] kthread+0x42/0x67 > [] ? kthread+0x0/0x67 > [] kernel_thread_helper+0x7/0x10 >=20 > I've only seen it once in 5 boots and CONFIG_PROVELOCKING does not gi= ve any > warnings about this, though it does yell when switching governor as r= eported > by others in bug #13493. >=20 > Let's hope Mathieu nails it, though I know he's busy with his thesis. >=20 Thanks for the lockdep reports, I'm currently looking into it, and it's not pretty. Basically we have : A B (means B nested in A) work read rwlock policy dbs_mutex work read rwlock policy write rwlock policy dbs_mutex So the added dbs_mutex <- work <- rwlock policy dependency (for proper teardown) is firing the reverse dependency between policy rwlock and dbs_mutex. The real way to fix this is to do not take the rwlock policy around non-policy-related actions, like governor START/STOP doing worker creation/teardown. One simple short-term solution would be to take a mutex outside of the policy rwlock write lock in cpufreq.c. This mutex would be the equivalent of dbs_mutex "lifted" outside of the rwlock write lock. For teardown, we only need to hold this mutex, not the rwlock write lock. Then we can remove the dbs_mutex from the governors. But looking at cpufreq.c's cpufreq_add_dev() is very much like kicking = a wasp nest: a lot of error paths are not handled properly, and I fear someone will have to go through the code, fix the currently incorrect code paths, and then add the lifted mutex. I currently have no time for implementation due to my thesis, but I'll be happy to review a patch. Mathieu --=20 Mathieu Desnoyers OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F BA06 3F25 A8FE 3BAE = 9A68