Date:	Sat, 26 Mar 2011 20:25:29 -0700
From:	"Paul E. McKenney" <paulmck@...ux.vnet.ibm.com>
To:	sedat.dilek@...il.com
Cc:	Josh Triplett <josh@...htriplett.org>,
	linux-next <linux-next@...r.kernel.org>,
	LKML <linux-kernel@...r.kernel.org>,
	Stephen Rothwell <sfr@...b.auug.org.au>,
	Randy Dunlap <randy.dunlap@...cle.com>,
	"Theodore Ts'o" <tytso@....edu>, Jens Axboe <axboe@...nel.dk>,
	Tejun Heo <tj@...nel.org>, Al Viro <viro@...iv.linux.org.uk>,
	Nick Piggin <npiggin@...nel.dk>
Subject: Re: linux-next: Tree for March 25 (Call trace:
 RCU|workqueues|block|VFS|ext4 related?)

On Sun, Mar 27, 2011 at 03:30:34AM +0200, Sedat Dilek wrote:
> On Sun, Mar 27, 2011 at 1:09 AM, Paul E. McKenney
> <paulmck@...ux.vnet.ibm.com> wrote:
> > On Sat, Mar 26, 2011 at 11:15:22PM +0100, Sedat Dilek wrote:
> >> On Sat, Mar 26, 2011 at 5:02 PM, Paul E. McKenney
> >> <paulmck@...ux.vnet.ibm.com> wrote:
> >> > On Sat, Mar 26, 2011 at 01:34:53PM +0100, Sedat Dilek wrote:
> >> [...]
> >> >> Just FYI: Changed to the following settings:
> >> >>
> >> >> - Enable Preemptible Kernel (Low-Latency Desktop)
> >> >> - Enable Preemptible tree-based hierarchical RCU
> >> >> - Enable RCU priority boosting
> >> >> - Reset RCU CPU stall timeout to default (60 seconds)
> >> >>
> >> >> So far I see no RCU stalls in the logs and my system runs as expected.
> >> >> ( I have noticed here some "stalling" in the webbrowser, but I can do
> >> >> my daily business. )
> >> >
> >> > OK, good to see some progress!
> >> >
> >>
> >> On 1st impression thing went fine, but after a while jobs like opening
> >> several tabs in firefox or doing a simple df command stalled the
> >> machine. No, my system even got frozen and required a brutal reset.
> >>
> >> >From the logs (more see file-attachment):
> >>
> >> Mar 26 19:58:40 tbox kernel: [ 1440.640060] INFO: task systemd:1
> >> blocked for more than 120 seconds.
> >> Mar 26 19:58:40 tbox kernel: [ 1440.640074] "echo 0 >
> >> /proc/sys/kernel/hung_task_timeout_secs" disables this message.
> >>
> >> Following it -> NOPE
> >> $ echo 0 > /proc/sys/kernel/hung_task_timeout_secs
> >
> > These are tasks that are blocked, not tasks that are consuming CPU.
> >
> >> > Is there a runaway process consuming CPU?  The reason that I ask is that
> >> > an infinite loop in the kernel can result in a stall when PREEMPT=n
> >> > but is less likely to if PREEMPT=y.  Could you please check with "top",
> >> > "ps", or whatever?
> >>
> >> Unsure what you mean by this, as you can see from the logs, it's not
> >> only one special task "stalling".
> >> BTW, I have systemd here running.
> >
> > Right.  But I need to know if there are tasks consuming lots of CPU,
> > which is different than tasks that are stalled for a long time.  Look
> > at the message: it says "blocked for more than 120 seconds", not
> > "running for more than 120 seconds".
> >
> > For example, if one of the RCU kthreads is consuming lots of CPU, that
> > would tell me that I should look for an RCU bug (and yank the patches
> > from -next in the meantime).  On the other hand, if some other task is
> > consuming lots of CPU, then that would be a hint as to where to find
> > the bug.
> >
> >> >> I am not sure what the change to PREEMPT exactly mean in the end.
> >> >> ( Let's work with this new kernel and carefully check for possible
> >> >> side-effects. )
> >> >> For example CONFIG_RCU_FAST_NO_HZ=y is now dropped, where the Kconfig
> >> >> descriptive text says some words on better energy saving. For a
> >> >> notebook this is no good.
> >> >
> >> > CONFIG_RCU_FAST_NO_HZ is of no use on a uniprocessor system, so OK
> >> > to disable it.  But are you saying that CONFIG_RCU_FAST_NO_HZ=y
> >> > results in problems that are removed by CONFIG_RCU_FAST_NO_HZ=n?
> >> > That would be a surprise, and I need to know if this is the case.
> >>
> >> In my current setup (PREEMPT and TREE_PREEMPT_RCU)
> >> CONFIG_RCU_FAST_NO_HZ is not considered and set via Kconfig-system
> >> (see excerpt below).
> >> But when you say for UP it is of no use, I need no more info.
> >
> > OK, "of no use" is overstating things a bit.  But CONFIG_RCU_FAST_NO_HZ's
> > main purpose is to get the last CPU into dyntick-idle state
> > (CONFIG_NO_HZ), which is most useful if the system has several CPUs.
> >
> >> Might be good to add some recommended (and/or useless) kernel-config
> >> settings to RCU/UP.txt?
> >>
> >> [ init/Kconfig ]
> >> config RCU_FAST_NO_HZ
> >>         bool "Accelerate last non-dyntick-idle CPU's grace periods"
> >>         depends on TREE_RCU && NO_HZ && SMP
> >>         default n
> >
> > The "depends on TREE_RCU && NO_HZ && SMP" already excludes it from
> > UP kernel builds, so no need to document.
> >
> >> >> I have also questions to some Kconfig dependencies, for example why I
> >> >> can't select TREE_PREEMPT_RCU if CONFIG_PREEMPT_VOLUNTARY=y, etc.
> >> >> Intended?
> >> >
> >> > Yes.  There is no point in TREE_PREEMPT_RCU unless PREEMPT=y.
> >> >
> >>
> >> OK.
> >>
> >> >> Maybe I collect all my askings in a separate email to RCU folks and ML
> >> >> and do not disturb further people from other sub-trees.
> >> >>
> >> >> I enjoyed to read the numerous docs in Documentation/RCU/ (and noticed
> >> >> some typos as well).
> >> >> The RCU folk gave the word "FAQ" a new meaning: Frequenty Asked
> >> >> Questions & Q*uiz* :-).
> >> >>
> >> >> Thanks for the helpful hints and explanations from the RCU folks!
> >> >
> >> > Glad you liked them!  ;-)
> >> >
> >>
> >> Other sub-trees lack of no good or up2date docs.
> >>
> >> >> - Sedat -
> >> >>
> >> >> P.S.: Current RCU and HZ kernel-config settings
> >> >>
> >> >> # grep RCU /boot/config-$(uname -r)
> >> >> # RCU Subsystem
> >> >> CONFIG_TREE_PREEMPT_RCU=y
> >> >> CONFIG_PREEMPT_RCU=y
> >> >> CONFIG_RCU_TRACE=y
> >> >> CONFIG_RCU_FANOUT=32
> >> >> # CONFIG_RCU_FANOUT_EXACT is not set
> >> >> CONFIG_TREE_RCU_TRACE=y
> >> >> CONFIG_RCU_BOOST=y
> >> >> CONFIG_RCU_BOOST_PRIO=1
> >> >> CONFIG_RCU_BOOST_DELAY=500
> >> >> # CONFIG_SPARSE_RCU_POINTER is not set
> >> >> # CONFIG_RCU_TORTURE_TEST is not set
> >> >> CONFIG_RCU_CPU_STALL_TIMEOUT=60
> >> >> CONFIG_RCU_CPU_STALL_VERBOSE=y
> >> >>
> >> >> # grep _HZ /boot/config-$(uname -r)
> >> >> CONFIG_NO_HZ=y
> >> >> # CONFIG_HZ_100 is not set
> >> >> CONFIG_HZ_250=y
> >> >> # CONFIG_HZ_300 is not set
> >> >> # CONFIG_HZ_1000 is not set
> >> >> CONFIG_HZ=250
> >> >
> >> > OK, thank you for the info!
> >> >
> >>
> >> N.P.
> >>
> >> >                                                        Thanx, Paul
> >> >
> >>
> >> I guees I will revert step-by-step the RCU commits in linux-next and report.
> >> This weekend I wanted to enhance Debian's live-cd framework with
> >> overlayfs support and a customized kernel.
> >> But then came RCU :-(.
> >
> > Well, if it turns out to be a problem in RCU I will certainly apologize.
> >
> 
> No, that's not so dramatic.
> Dealing with this RCU issue has nice side-effects: I remembered (and
> finally did) to use a reduced kernel-config set.
> The base for it I created with 'make localmodconfig' and did some
> manual fine-tuning afterwards (throw out media, rc, dvd, unneeded FSs,
> etc.).
> Also, I can use fresh gcc-4.6 (4.6.0-1) from the official Debian repos.
> 
> So, I started building with
> "revert-rcu-patches/0001-Revert-rcu-introduce-kfree_rcu.patch".
> I will let you know.

And please also check for tasks consuming all available CPU.

							Thanx, Paul

> - Sedat -
> 
> >> Can you say some words to kfree_rcu.2011.03.25b (rcu/kfree_rcu) GIT branch(es)?
> >
> > These are just applying Lia Jiangshan's kfree_rcu() to a number of places
> > in the kernel.  You can safely ignore them.
> >
> >                                                                Thanx, Paul
> >
> >> - Sedat -
> >
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists