From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754577AbZBGRCj (ORCPT ); Sat, 7 Feb 2009 12:02:39 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1752779AbZBGRC3 (ORCPT ); Sat, 7 Feb 2009 12:02:29 -0500 Received: from mail-bw0-f161.google.com ([209.85.218.161]:35998 "EHLO mail-bw0-f161.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752519AbZBGRC2 (ORCPT ); Sat, 7 Feb 2009 12:02:28 -0500 DomainKey-Signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=date:from:to:cc:subject:message-id:references:mime-version :content-type:content-disposition:in-reply-to:user-agent; b=D4Fs3rze5/3d28SPuxI/d+p0klGIbBCvGT++vu6MwNIRmLiYPrzk4kdC4tYjVwJHbe jgQJiKPdJngjN2QDSFcQ1Lrrfdz5DUvRZQpwVQetrTFG9cJh2EHHDW8HAv4HtJvg7M+w VYA6ywF7s/MRSL697JxaVoaLsCfCW5cuQtZvE= Date: Sat, 7 Feb 2009 18:02:22 +0100 From: Frederic Weisbecker To: Mandeep Singh Baines Cc: mingo@elte.hu, linux-kernel@vger.kernel.org, rientjes@google.com, mbligh@google.com, thockin@google.com, akpm@linux-foundation.org Subject: Re: [PATCH] softlockup: remove timestamp checking from hung_task Message-ID: <20090207170222.GD5779@nowhere> References: <20090206233747.GA8747@google.com> <20090207162328.GA5779@nowhere> <20090207163440.GB5779@nowhere> <20090207165154.GC5779@nowhere> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20090207165154.GC5779@nowhere> User-Agent: Mutt/1.5.18 (2008-05-17) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Sat, Feb 07, 2009 at 05:51:55PM +0100, Frederic Weisbecker wrote: > On Sat, Feb 07, 2009 at 05:34:40PM +0100, Frederic Weisbecker wrote: > > On Sat, Feb 07, 2009 at 05:23:28PM +0100, Frederic Weisbecker wrote: > > > Hi Mandeep, > > > > > > On Fri, Feb 06, 2009 at 03:37:47PM -0800, Mandeep Singh Baines wrote: > > > > Patch against tip/core/softlockup > > > > > > > > --- > > > > Impact: saves sizeof(long) bytes per task_struct > > > > > > > > By guaranteeing that sysctl_hung_task_timeout_secs have elapsed between > > > > tasklist scans we can avoid using timestamps. > > > > > > > > Signed-off-by: Mandeep Singh Baines > > > > > > > > > Good idea. > > > BTW, why haven't you put your name on top of this file? > > > That would help those who will send patches knowing to whom they have to > > > route their mails. > > > > > > I made some comments below about small things... > > > > > > > --- > > > > include/linux/sched.h | 1 - > > > > kernel/fork.c | 8 +++----- > > > > kernel/hung_task.c | 48 +++++++++--------------------------------------- > > > > 3 files changed, 12 insertions(+), 45 deletions(-) > > > > > > > > diff --git a/include/linux/sched.h b/include/linux/sched.h > > > > index 2a2811c..e0d723f 100644 > > > > --- a/include/linux/sched.h > > > > +++ b/include/linux/sched.h > > > > @@ -1241,7 +1241,6 @@ struct task_struct { > > > > #endif > > > > #ifdef CONFIG_DETECT_HUNG_TASK > > > > /* hung task detection */ > > > > - unsigned long last_switch_timestamp; > > > > unsigned long last_switch_count; > > > > #endif > > > > /* CPU-specific state of this task */ > > > > diff --git a/kernel/fork.c b/kernel/fork.c > > > > index fb94442..bf582f7 100644 > > > > --- a/kernel/fork.c > > > > +++ b/kernel/fork.c > > > > @@ -639,6 +639,9 @@ static int copy_mm(unsigned long clone_flags, struct task_struct * tsk) > > > > > > > > tsk->min_flt = tsk->maj_flt = 0; > > > > tsk->nvcsw = tsk->nivcsw = 0; > > > > +#ifdef CONFIG_DETECT_HUNG_TASK > > > > + tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw; > > > > +#endif > > > > > > > > > I think you can directly assign a zero here :-) > > > Or you want to let it as is to give some sense and explanation > > > about the role of this field? > > > Why not, I guess gcc will optimize it anyway. > > > > > > > > > > tsk->mm = NULL; > > > > tsk->active_mm = NULL; > > > > @@ -1041,11 +1044,6 @@ static struct task_struct *copy_process(unsigned long clone_flags, > > > > > > > > p->default_timer_slack_ns = current->timer_slack_ns; > > > > > > > > -#ifdef CONFIG_DETECT_HUNG_TASK > > > > - p->last_switch_count = 0; > > > > - p->last_switch_timestamp = 0; > > > > -#endif > > > > - > > > > task_io_accounting_init(&p->ioac); > > > > acct_clear_integrals(p); > > > > > > > > diff --git a/kernel/hung_task.c b/kernel/hung_task.c > > > > index 3951a80..4a10756 100644 > > > > --- a/kernel/hung_task.c > > > > +++ b/kernel/hung_task.c > > > > @@ -34,7 +34,6 @@ unsigned long __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT; > > > > * Zero means infinite timeout - no checking done: > > > > */ > > > > unsigned long __read_mostly sysctl_hung_task_timeout_secs = 120; > > > > -static unsigned long __read_mostly hung_task_poll_jiffies; > > > > > > > > unsigned long __read_mostly sysctl_hung_task_warnings = 10; > > > > > > > > @@ -69,33 +68,17 @@ static struct notifier_block panic_block = { > > > > .notifier_call = hung_task_panic, > > > > }; > > > > > > > > -/* > > > > - * Returns seconds, approximately. We don't need nanosecond > > > > - * resolution, and we don't need to waste time with a big divide when > > > > - * 2^30ns == 1.074s. > > > > - */ > > > > -static unsigned long get_timestamp(void) > > > > -{ > > > > - int this_cpu = raw_smp_processor_id(); > > > > - > > > > - return cpu_clock(this_cpu) >> 30LL; /* 2^30 ~= 10^9 */ > > > > -} > > > > - > > > > -static void check_hung_task(struct task_struct *t, unsigned long now, > > > > - unsigned long timeout) > > > > +static void check_hung_task(struct task_struct *t, unsigned long timeout) > > > > { > > > > unsigned long switch_count = t->nvcsw + t->nivcsw; > > > > > > > > if (t->flags & PF_FROZEN) > > > > return; > > > > > > > > - if (switch_count != t->last_switch_count || !t->last_switch_timestamp) { > > > > + if (switch_count != t->last_switch_count) { > > > > t->last_switch_count = switch_count; > > > > - t->last_switch_timestamp = now; > > > > return; > > > > } > > > > > > > > > > > > What happens here if khungtaskd is scheduled in after tsk is inserted on the task_list > > > in copy_process() but before tsk has been scheduled once? > > > > > > tsk->last_switch_count and tsk->nvcsw + tsk->nivcsw will still be equal to zero right? > > > > > > Perhaps you could add another check such as > > > > > > if (!switch_count) > > > return; > > > > > > > > > > - if ((long)(now - t->last_switch_timestamp) < timeout) > > > > - return; > > > > if (!sysctl_hung_task_warnings) > > > > return; > > > > sysctl_hung_task_warnings--; > > > > @@ -111,7 +94,6 @@ static void check_hung_task(struct task_struct *t, unsigned long now, > > > > sched_show_task(t); > > > > __debug_show_held_locks(t); > > > > > > > > - t->last_switch_timestamp = now; > > > > touch_nmi_watchdog(); > > > > > > > > if (sysctl_hung_task_panic) > > > > @@ -145,7 +127,6 @@ static void check_hung_uninterruptible_tasks(unsigned long timeout) > > > > { > > > > int max_count = sysctl_hung_task_check_count; > > > > int batch_count = HUNG_TASK_BATCHING; > > > > - unsigned long now = get_timestamp(); > > > > struct task_struct *g, *t; > > > > > > > > /* > > > > @@ -168,19 +149,16 @@ static void check_hung_uninterruptible_tasks(unsigned long timeout) > > > > } > > > > /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */ > > > > if (t->state == TASK_UNINTERRUPTIBLE) > > > > - check_hung_task(t, now, timeout); > > > > + check_hung_task(t, timeout); > > > > } while_each_thread(g, t); > > > > unlock: > > > > rcu_read_unlock(); > > > > } > > > > > > > > -static void update_poll_jiffies(void) > > > > +static unsigned long timeout_jiffies(unsigned long timeout) > > > > { > > > > /* timeout of 0 will disable the watchdog */ > > > > - if (sysctl_hung_task_timeout_secs == 0) > > > > - hung_task_poll_jiffies = MAX_SCHEDULE_TIMEOUT; > > > > - else > > > > - hung_task_poll_jiffies = sysctl_hung_task_timeout_secs * HZ / 2; > > > > + return (timeout ? timeout * HZ : MAX_SCHEDULE_TIMEOUT); > > > > } > > > > > > > > /* > > > > @@ -197,8 +175,6 @@ int proc_dohung_task_timeout_secs(struct ctl_table *table, int write, > > > > if (ret || !write) > > > > goto out; > > > > > > > > - update_poll_jiffies(); > > > > - > > > > wake_up_process(watchdog_task); > > > > > > > > > I'm not sure what does this function now that you dropped update_poll_jiffies() > > > So if the user sets up a new timeout value, the only effect will be that khungtaskd will > > > be awakened? > > > > > > But actually the /sys file doesn't seem to be set up. > > > > > > Oops, I should have grep on proc_dohung_task_timeout_secs which is set on kernel/sysctl. > > Sorry. > > But still, sysctl_hung_task_timeout_secs doesn't seem to be set :-) > > > And once again I'm wrong, I shoud read sysctl.c or at least read the sysfs documentation. > Sorry for the noise. > > BTW, here is a small fixlet on top of your patch about what I commented concerning > the tasks than weren't yet scheduled once: But it is not supposed to be TASK_UNINTERRUPTIBLE. Forget this fixlet. So, again sorry for the noise, I'm going to hide somewhere... > -- > From e7120e424b031978e482b5fe311d90916ffb8b7e Mon Sep 17 00:00:00 2001 > From: Frederic Weisbecker > Date: Sat, 7 Feb 2009 17:45:12 +0100 > Subject: [PATCH] softlockup: ensure the task has been scheduled once > > When we check if the task has been scheduled since the last scan, we might > have a race condition if the task has been inserted on the task list but not > yet scheduled once. So we just add a small check to ensure it has been switched > in at least one time to avoid false positive. > > Signed-off-by: Frederic Weisbecker > --- > kernel/hung_task.c | 6 +++++- > 1 files changed, 5 insertions(+), 1 deletions(-) > > diff --git a/kernel/hung_task.c b/kernel/hung_task.c > index 4a10756..7f57a71 100644 > --- a/kernel/hung_task.c > +++ b/kernel/hung_task.c > @@ -72,7 +72,11 @@ static void check_hung_task(struct task_struct *t, unsigned long timeout) > { > unsigned long switch_count = t->nvcsw + t->nivcsw; > > - if (t->flags & PF_FROZEN) > + /* > + * Ensure the task is not frozen and that it has been scheduled > + * at least once. > + */ > + if (t->flags & PF_FROZEN || !switch_count) > return; > > if (switch_count != t->last_switch_count) { > -- > 1.6.1 >