Re: [GIT PULL v2] sched: Make sleep inside atomic detection work on !PREEMPT

[Frederic Weisbecker <fweisbec@gmail.com>]

On Fri, Jul 01, 2011 at 02:59:26PM +0200, Ingo Molnar wrote:
> 
> * Frederic Weisbecker <fweisbec@gmail.com> wrote:
> 
> > On Fri, Jul 01, 2011 at 02:36:29PM +0200, Ingo Molnar wrote:
> > > 
> > > * Frederic Weisbecker <fweisbec@gmail.com> wrote:
> > > 
> > > > On Fri, Jun 10, 2011 at 03:30:18PM +0200, Frederic Weisbecker wrote:
> > > > > Ingo,
> > > > > 
> > > > > Please pull the sched/core branch that can be found at:
> > > > > 
> > > > > git://git.kernel.org/pub/scm/linux/kernel/git/frederic/random-tracing.git
> > > > > 	sched/core
> > > > 
> > > > Hi Ingo,
> > > > 
> > > > I have added Randy's ack on the last patch. To get it, please pull the v2 in
> > > > the following branch:
> > > > 
> > > > git://git.kernel.org/pub/scm/linux/kernel/git/frederic/random-tracing.git
> > > >        sched/core-v2
> > > > 
> > > > There are no other changes.
> > > 
> > > Hm, this triggers such warnings now:
> > > 
> > > Detected 2010.217 MHz processor.
> > > Marking TSC unstable due to TSCs unsynchronized
> > > Calibrating delay loop (skipped), value calculated using timer frequency.. 4022.95 BogoMIPS (lpj=6700723)
> > > pid_max: default: 4096 minimum: 301
> > > BUG: scheduling while atomic: swapper/0/0x10000002
> > > no locks held by swapper/0.
> > > Pid: 0, comm: swapper Not tainted 3.0.0-rc5-tip-01401-ga7adf5f-dirty #141020
> > > Call Trace:
> > >  [<ffffffff81ddb3b0>] __schedule_bug+0x60/0x65
> > >  [<ffffffff81dee803>] schedule+0x953/0x980
> > >  [<ffffffff81571e30>] ? serial8250_console_putchar+0x30/0x40
> > >  [<ffffffff81df177b>] ? _raw_spin_unlock+0x2b/0x50
> > >  [<ffffffff8107882a>] __cond_resched+0x2a/0x40
> > >  [<ffffffff81dee8e2>] _cond_resched+0x32/0x40
> > >  [<ffffffff81110ba8>] __alloc_pages_nodemask+0x1b8/0x880
> > >  [<ffffffff81df23ce>] ? common_interrupt+0xe/0x13
> > >  [<ffffffff81080649>] ? vprintk+0x359/0x530
> > >  [<ffffffff8113b4e7>] slob_new_pages+0x17/0x80
> > >  [<ffffffff8113bd13>] __kmalloc_node+0xa3/0x270
> > >  [<ffffffff81ddb6d5>] ? printk+0x41/0x43
> > >  [<ffffffff82738420>] pidmap_init+0x80/0xbf
> > >  [<ffffffff8272aa54>] start_kernel+0x28b/0x300
> > >  [<ffffffff8272a2ee>] x86_64_start_reservations+0xfe/0x102
> > >  [<ffffffff8272a3e2>] x86_64_start_kernel+0xf0/0xf7
> > > Security Framework initialized
> > > AppArmor: AppArmor initialized
> > > Mount-cache hash table entries: 256
> > > Initializing cgroup subsys cpuacct
> > > 
> > > Not sure we want to warn about schedule during early init, or can 
> > > it cause problems and should thus be fixed? I bet there's more 
> > > such instances.
> > 
> > I believe this is harmless because no other other tasks than init 
> > are queued at that time. But I don't know, may be calling 
> > schedule() involves some things that are not ready yet at that 
> > time.
> 
> So why does it have need_resched set then? (which i presume must be a 
> condition for cond_resched() to call into schedule())

Good point, I'm looking at this.
 
> > Either we add a system_state check in schedule_debug() or we fix 
> > the callers to not call schedule when system_state shows we are 
> > booting..
> 
> Hm, such attempts were rather fragile in the past.
> 
> Thanks,
> 
> 	Ingo