On 10/15/20 8:37 AM, Oleg Nesterov wrote:
> On 10/15, Jens Axboe wrote:
>>
>> On 10/15/20 8:31 AM, Oleg Nesterov wrote:
>>> On 10/15, Jens Axboe wrote:
>>>>
>>>> static inline int signal_pending(struct task_struct *p)
>>>> {
>>>> +#if defined(CONFIG_GENERIC_ENTRY) && defined(TIF_NOTIFY_SIGNAL)
>>>> + /*
>>>> + * TIF_NOTIFY_SIGNAL isn't really a signal, but it requires the same
>>>> + * behavior in terms of ensuring that we break out of wait loops
>>>> + * so that notify signal callbacks can be processed.
>>>> + */
>>>> + if (unlikely(test_tsk_thread_flag(p, TIF_NOTIFY_SIGNAL)))
>>>> + return 1;
>>>> +#endif
>>>> return task_sigpending(p);
>>>> }
>>>
>>> I don't understand why does this version requires CONFIG_GENERIC_ENTRY.
>>>
>>> Afaics, it is very easy to change all the non-x86 arches to support
>>> TIF_NOTIFY_SIGNAL, but it is not trivial to change them all to use
>>> kernel/entry/common.c ?
>>
>> I think that Thomas wants to gate TIF_NOTIFY_SIGNAL on conversion to
>> the generic entry code?
>
> Then I think TIF_NOTIFY_SIGNAL will be never fully supported ;)
That is indeed a worry. From a functionality point of view, with the
major archs supporting it, I'm not too worried about that side. But it
does mean that we'll be stuck with the ifdeffery forever, which isn't
great.
Thomas, are you fine with decoupling this from CONFIG_GENERIC_ENTRY,
based on the above concerns?
--
Jens Axboe
