From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Paul E. McKenney" Subject: Re: 2.6.39-rc4+: Kernel leaking memory during FS scanning, regression? Date: Thu, 28 Apr 2011 17:42:55 -0700 Message-ID: <20110429004255.GF2191@linux.vnet.ibm.com> References: <20110428222301.0b745a0a@neptune.home> <20110428224444.43107883@neptune.home> <1304027480.2971.121.camel@work-vm> Reply-To: paulmck@linux.vnet.ibm.com Mime-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1 Content-Transfer-Encoding: quoted-printable Cc: Thomas Gleixner , john stultz , Bruno =?iso-8859-1?Q?Pr=E9mont?= , Mike Galbraith , Linus Torvalds , Ingo Molnar , Peter Zijlstra , Mike Frysinger , KOSAKI Motohiro , LKML , linux-mm@kvack.org, linux-fsdevel@vger.kernel.org, "Paul E. McKenney" , Pekka Enberg To: sedat.dilek@gmail.com Return-path: Content-Disposition: inline In-Reply-To: Sender: owner-linux-mm@kvack.org List-Id: linux-fsdevel.vger.kernel.org On Fri, Apr 29, 2011 at 01:35:44AM +0200, Sedat Dilek wrote: > On Fri, Apr 29, 2011 at 1:06 AM, Sedat Dilek wrote: > > On Fri, Apr 29, 2011 at 12:02 AM, Thomas Gleixner wrote: > >> On Thu, 28 Apr 2011, john stultz wrote: > >>> On Thu, 2011-04-28 at 23:04 +0200, Thomas Gleixner wrote: > >>> > /me suspects hrtimer changes to be the real culprit. > >>> > >>> I'm not seeing anything on right off, but it does smell like > >>> e06383db9ec591696a06654257474b85bac1f8cb would be where such an iss= ue > >>> would crop up. > >>> > >>> Bruno, could you try checking out e06383db9ec, confirming it still > >>> occurs (and then maybe seeing if it goes away at e06383db9ec^1)? > >>> > >>> I'll keep digging in the meantime. > >> > >> I found the bug already. The problem is that sched_init() calls > >> init_rt_bandwidth() which calls hrtimer_init() _BEFORE_ > >> hrtimers_init() is called. > >> > >> That was unnoticed so far as the CLOCK id to hrtimer base conversion > >> was hardcoded. Now we use a table which is set up at hrtimers_init()= , > >> so the bandwith hrtimer ends up on CLOCK_REALTIME because the table = is > >> in the bss. > >> > >> The patch below fixes this, by providing the table statically rather > >> than runtime initialized. Though that whole ordering wants to be > >> revisited. > >> > >> Thanks, > >> > >> =A0 =A0 =A0 =A0tglx > >> > >> --- linux-2.6.orig/kernel/hrtimer.c > >> +++ linux-2.6/kernel/hrtimer.c > >> @@ -81,7 +81,11 @@ DEFINE_PER_CPU(struct hrtimer_cpu_base, > >> =A0 =A0 =A0 =A0} > >> =A0}; > >> > >> -static int hrtimer_clock_to_base_table[MAX_CLOCKS]; > >> +static int hrtimer_clock_to_base_table[MAX_CLOCKS] =3D { > >> + =A0 =A0 =A0 [CLOCK_REALTIME] =3D HRTIMER_BASE_REALTIME, > >> + =A0 =A0 =A0 [CLOCK_MONOTONIC] =3D HRTIMER_BASE_MONOTONIC, > >> + =A0 =A0 =A0 [CLOCK_BOOTTIME] =3D HRTIMER_BASE_BOOTTIME, > >> +}; > >> > >> =A0static inline int hrtimer_clockid_to_base(clockid_t clock_id) > >> =A0{ > >> @@ -1722,10 +1726,6 @@ static struct notifier_block __cpuinitda > >> > >> =A0void __init hrtimers_init(void) > >> =A0{ > >> - =A0 =A0 =A0 hrtimer_clock_to_base_table[CLOCK_REALTIME] =3D HRTIME= R_BASE_REALTIME; > >> - =A0 =A0 =A0 hrtimer_clock_to_base_table[CLOCK_MONOTONIC] =3D HRTIM= ER_BASE_MONOTONIC; > >> - =A0 =A0 =A0 hrtimer_clock_to_base_table[CLOCK_BOOTTIME] =3D HRTIME= R_BASE_BOOTTIME; > >> - > >> =A0 =A0 =A0 =A0hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_U= P_PREPARE, > >> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0(void *)(long)smp= _processor_id()); > >> =A0 =A0 =A0 =A0register_cpu_notifier(&hrtimers_nb); > >> > >> > >> > > > > Looks good so far, no stalls or call-traces. > > > > Really stressing with 20+ open tabs in firefox with flash-movie > > running in one of them , tar-job, IRC-client etc. > > I will run some more tests and collect data and send them later. > > > > - Sedat - > > > > P.S.: Patchset against linux-2.6-rcu.git#sedat.2011.04.23a where 0003 > > is from [2] > > > > [1] http://git.us.kernel.org/?p=3Dlinux/kernel/git/paulmck/linux-2.6-= rcu.git;a=3Dshortlog;h=3Drefs/heads/sedat.2011.04.23a > > [2] https://patchwork.kernel.org/patch/739782/ > > > > $ l ../RCU-HOORAY/ > > insgesamt 40 > > drwxr-xr-x =A02 sd sd =A04096 29. Apr 01:02 . > > drwxr-xr-x 35 sd sd 20480 29. Apr 01:01 .. > > -rw-r--r-- =A01 sd sd =A0 726 29. Apr 01:01 > > 0001-Revert-rcu-restrict-TREE_RCU-to-SMP-builds-with-PREE.patch > > -rw-r--r-- =A01 sd sd =A0 735 29. Apr 01:01 > > 0002-sched-Add-warning-when-RT-throttling-is-activated.patch > > -rw-r--r-- =A01 sd sd =A02376 29. Apr 01:01 > > 0003-2.6.39-rc4-Kernel-leaking-memory-during-FS-scanning-.patch > > >=20 > As promised the tarball (at the end of the log I made some XZ compressi= ng). >=20 > Wow! > $ uptime > 01:35:17 up 45 min, 3 users, load average: 0.45, 0.57, 1.27 >=20 > Thanks to all involved people helping to kill that bug (Come on Paul, s= mile!). Woo-hoo!!!! Many thanks to Thomas for tracking this down -- it is fair to say that I never would have thought to look at timer initialization! ;-) Thanx, Paul -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@kvack.org. For more info on Linux MM, see: http://www.linux-mm.org/ . Fight unfair telecom internet charges in Canada: sign http://stopthemeter= .ca/ Don't email: email@kvack.org