Re: [RFC PATCHC 2/3] idle: store the idle state the cpu is

[Peter Zijlstra]

On Tue, Apr 15, 2014 at 04:17:36PM +0200, Daniel Lezcano wrote:
> On 04/15/2014 02:44 PM, Peter Zijlstra wrote:
> >On Tue, Apr 15, 2014 at 02:43:30PM +0200, Peter Zijlstra wrote:
> >>On Fri, Mar 28, 2014 at 01:29:55PM +0100, Daniel Lezcano wrote:
> >>>@@ -143,6 +145,10 @@ static int cpuidle_idle_call(void)
> >>>  			if (!ret) {
> >>>  				trace_cpu_idle_rcuidle(next_state, dev->cpu);
> >>>
> >>>+				*power = &drv->states[next_state].power;
> >>>+
> >>>+				wmb();
> >>>+
> >>
> >>I very much suspect you meant: smp_wmb(), as I don't see the hardware
> >>reading that pointer, therefore UP wouldn't care. Also, any and all
> >>barriers should come with a comment that describes the data ordering and
> >>points to the matchin barriers.
> >
> >Furthermore, this patch fails to describe the life-time rules of the
> >object placed there. Can the objected pointed to ever disappear?
> 
> Hi Peter,
> 
> thanks for reviewing the patches.
> 
> There are a couple of situations where a cpuidle state can disappear:
> 
> 1. For x86/acpi with dynamic c-states, when a laptop switches from battery
> to AC that could result on removing the deeper idle state. The acpi driver
> triggers:
> 
> 'acpi_processor_cst_has_changed' which will call 'cpuidle_pause_and_lock'.
> This one will call 'cpuidle_uninstall_idle_handler' which in turn calls
> 'kick_all_cpus_sync'.
> 
> All cpus will exit their idle state and the pointed object will be set to
> NULL again.
> 
> 2. The cpuidle driver is unloaded. Logically that could happen but not in
> practice because the drivers are always compiled in and 95% of the drivers
> are not coded to unregister the driver. Anyway ...
> 
> The unloading code must call 'cpuidle_unregister_device', that calls
> 'cpuidle_pause_and_lock' leading to 'kick_all_cpus_sync'.
> 
> IIUC, the race can happen if we take the pointer and then one of these two
> situation occurs at the same moment.
> 
> As the function 'find_idlest_cpu' is inside a rcu_read_lock may be a
> rcu_barrier in 'cpuidle_pause_and_lock' or 'cpuidle_uninstall_idle_handler'
> should suffice, no ?

Indeed. But be sure to document this.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/