Re: [PATCH 1/2] timers: WARN if add_timer_on is used with offlined cpu.

[imran . f . khan]

Hello Thomas,
On 16/1/2025 1:39 am, Thomas Gleixner wrote:
> On Thu, Jan 16 2025 at 00:41, Imran Khan wrote:
>> timer started on an offlined cpu will not fire after
>> its expiry time and may never fire if that cpu remains
>> offline.
>> So add a WARN_ON_ONCE in add_timer_on, to indicate
>> if any of its users are (wrongly) starting a timer
>> on an offlined cpu.
>>
>> Suggested-by: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
>> Signed-off-by: Imran Khan <imran.f.khan@xxxxxxxxxx>
>> ---
>>  kernel/time/timer.c | 6 ++++++
>>  1 file changed, 6 insertions(+)
>>
>> diff --git a/kernel/time/timer.c b/kernel/time/timer.c
>> index a5860bf6d16f9..ec9eb58e45241 100644
>> --- a/kernel/time/timer.c
>> +++ b/kernel/time/timer.c
>> @@ -1370,6 +1370,12 @@ void add_timer_on(struct timer_list *timer, int cpu)
>>  	if (!timer->function)
>>  		goto out_unlock;
>>  
>> +	/*
>> +	 * WARN if specified cpu is offline, because on offlined cpu
>> +	 * timer will not fire even after its expiry.
>> +	 */
>> +	WARN_ON_ONCE(!cpu_online(cpu));
> 
> Then why queueing the timer in the first place?
> 

The queueing is for the case, if the cpu becomes online again.
Warning tells that timer is being put on an offlined cpu so may never fire,
if the cpu remains offline. But if the cpu becomes online, the timer
will fire.

Or should I just return after warning, like currently being done for
timer_pending. In that case I can move the warning before lock_timer_base.

Thanks,
Imran

> Thanks,
> 
>         tglx