Re: [RFC PATCH v1 1/9] timer: fix circular header dependency

[Christophe LEROY <christophe.leroy@c-s.fr>]

Le 25/09/2018 à 14:11, Mark Rutland a écrit :
> On Tue, Sep 25, 2018 at 07:34:15AM +0200, Christophe LEROY wrote:
>>
>>
>> Le 24/09/2018 à 17:52, Christophe Leroy a écrit :
>>> When switching powerpc to CONFIG_THREAD_INFO_IN_TASK, include/sched.h
>>> has to be included in asm/smp.h for the following change, in order
>>> to avoid uncomplete definition of task_struct:
>>>
>>> -#define raw_smp_processor_id() (current_thread_info()->cpu)
>>> +#define raw_smp_processor_id() (current->cpu)
>>>
>>> But this generates the following compilation error, due to circular
>>> header dependency.
>>>
>>>     CC      kernel/time/alarmtimer.o
>>> In file included from ./arch/powerpc/include/asm/smp.h:31,
>>>                    from ./include/linux/smp.h:64,
>>>                    from ./include/linux/percpu.h:7,
>>>                    from ./include/linux/hrtimer.h:22,
>>>                    from kernel/time/alarmtimer.c:19:
>>> ./include/linux/sched.h:558:19: error: field 'dl_timer' has incomplete type
>>>     struct hrtimer   dl_timer;
>>>                      ^~~~~~~~
>>> ./include/linux/sched.h:567:17: error: field 'inactive_timer' has incomplete type
>>>     struct hrtimer inactive_timer;
>>>                    ^~~~~~~~~~~~~~
>>> make[1]: *** [kernel/time/alarmtimer.o] Error 1
>>> make: *** [kernel/time/alarmtimer.o] Error 2
>>>
>>>     CC      fs/timerfd.o
>>> In file included from ./arch/powerpc/include/asm/smp.h:31,
>>>                    from ./include/linux/smp.h:64,
>>>                    from ./include/linux/percpu.h:7,
>>>                    from ./include/linux/hrtimer.h:22,
>>>                    from ./include/linux/alarmtimer.h:6,
>>>                    from fs/timerfd.c:12:
>>> ./include/linux/sched.h:558:19: error: field 'dl_timer' has incomplete type
>>>     struct hrtimer   dl_timer;
>>>                      ^~~~~~~~
>>> ./include/linux/sched.h:567:17: error: field 'inactive_timer' has incomplete type
>>>     struct hrtimer inactive_timer;
>>>                    ^~~~~~~~~~~~~~
>>> make[1]: *** [fs/timerfd.o] Error 1
>>> make: *** [fs/timerfd.o] Error 2
>>>
>>> This patch fixes it by including linux/hrtimer.h after linux/sched.h
>>>
>>> Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
>>> ---
>>>    Should it be fixed in powerpc instead ? In that case, how ? Any idea ?
>>
>>
>> Looks like there are several other places where the problem occurs, so it
>> has to be fixed in the powerpc headers instead.
>>
>> Seems like RISC arch faced the same issue, and fixed it the following way:
>>
>> /*
>>   * This is particularly ugly: it appears we can't actually get the
>> definition
>>   * of task_struct here, but we need access to the CPU this task is running
>> on.
>>   * Instead of using C we're using asm-offsets.h to get the current processor
>>   * ID.
>>   */
>> #define raw_smp_processor_id() (*((int*)((char*)get_current() +
>> TASK_TI_CPU)))
>>
>>
>> Unless someone has a better idea, I'll fixed it that way.
> 
> To solve this on arm64 we made the cpu number a percpu variable; see
> commit:
> 
>    57c82954e77fa12c ("arm64: make cpu number a percpu variable")

Thanks, I'll have a look

> IIUC, you could do that by placing it in paca_struct.

By the way, it is already in PACA struct for PPC64.
The issue is with PPC32, up to now it was in current_thread_info:

#ifdef CONFIG_PPC64
#define raw_smp_processor_id()	(local_paca->paca_index)
#define hard_smp_processor_id() (get_paca()->hw_cpu_id)
#else
/* 32-bit */
extern int smp_hw_index[];

#define raw_smp_processor_id()	(current_thread_info()->cpu)
#define hard_smp_processor_id() 	(smp_hw_index[smp_processor_id()])

[..]

#endif


Christophe