Re: [PATCH AUTOSEL 4.9 4/4] thermal: intel_powerclamp: Use get_cpu() instead of smp_processor_id() to avoid crash

[srinivas pandruvada]

On Tue, 2022-10-11 at 21:22 +0800, Chen Yu wrote:
> Hi Pavel,
> On 2022-10-11 at 13:36:46 +0200, Pavel Machek wrote:
> > Hi!
> > 
> > > From: Srinivas Pandruvada <srinivas.pandruvada@xxxxxxxxxxxxxxx>
> > > 
> > > [ Upstream commit 68b99e94a4a2db6ba9b31fe0485e057b9354a640 ]
> > > 
> > > When CPU 0 is offline and intel_powerclamp is used to inject
> > > idle, it generates kernel BUG:
> > > 
> > > BUG: using smp_processor_id() in preemptible [00000000] code:
> > > bash/15687
> > > caller is debug_smp_processor_id+0x17/0x20
> > > CPU: 4 PID: 15687 Comm: bash Not tainted 5.19.0-rc7+ #57
> > > Call Trace:
> > > <TASK>
> > > dump_stack_lvl+0x49/0x63
> > > dump_stack+0x10/0x16
> > > check_preemption_disabled+0xdd/0xe0
> > > debug_smp_processor_id+0x17/0x20
> > > powerclamp_set_cur_state+0x7f/0xf9 [intel_powerclamp]
> > > ...
> > > ...
> > > 
> > > Here CPU 0 is the control CPU by default and changed to the current
> > > CPU,
> > > if CPU 0 offlined. This check has to be performed under
> > > cpus_read_lock(),
> > > hence the above warning.
> > > 
> > > Use get_cpu() instead of smp_processor_id() to avoid this BUG.
> > 
> > This has exactly the same problem as smp_processor_id(), you just
> > worked around the warning. If it is okay that control_cpu contains
> > stale value, could we have a comment explaining why?
> > 
> May I know why does control_cpu have stale value? The control_cpu
> is a random picked online CPU which will be used later to collect
> statistics.
> As long as the control_cpu is online, it is valid IMO.
> 

I am also interested to know why this can be stale. The get_cpu() call
disables preemption. 

#define get_cpu()		({ preempt_disable();
__smp_processor_id(); })


Even if you change it to call debug_smp_processor_id() instead of
__smp_processor_id(), it will still not print warning as
preempt_count() will return 1.

If after the preemption is enabled if the CPU is offlined, there are
hotplug callbacks to handle.


Thanks,
Srinivas
 

> thanks,
> Chenyu
> > Thanks,
> >                                                                 Pavel
> >                                                                 
> > > +++ b/drivers/thermal/intel_powerclamp.c
> > > @@ -519,8 +519,10 @@ static int start_power_clamp(void)
> > >  
> > >         /* prefer BSP */
> > >         control_cpu = 0;
> > > -       if (!cpu_online(control_cpu))
> > > -               control_cpu = smp_processor_id();
> > > +       if (!cpu_online(control_cpu)) {
> > > +               control_cpu = get_cpu();
> > > +               put_cpu();
> > > +       }
> > >  
> > >         clamping = true;
> > >         schedule_delayed_work(&poll_pkg_cstate_work, 0);
> > > -- 
> > > 2.35.1
> > 
> > -- 
> > People of Russia, stop Putin before his war on Ukraine escalates.
> 
>