Re: git: 80cf427b8dc1 - main - proc: shave a lock trip on exit if possible

On 3/29/23, Dmitry Chagin <dchagin@freebsd.org> wrote:
> On Wed, Mar 29, 2023 at 09:19:09AM +0000, Mateusz Guzik wrote:
>> The branch main has been updated by mjg:
>>
>> URL:
>> https://cgit.FreeBSD.org/src/commit/?id=80cf427b8dc18bc5c26fed6d07bc5c5eda4545b0
>>
>> commit 80cf427b8dc18bc5c26fed6d07bc5c5eda4545b0
>> Author:     Mateusz Guzik <mjg@FreeBSD.org>
>> AuthorDate: 2023-03-29 08:45:46 +0000
>> Commit:     Mateusz Guzik <mjg@FreeBSD.org>
>> CommitDate: 2023-03-29 09:19:03 +0000
>>
>>     proc: shave a lock trip on exit if possible
>>
>>     ... which happens to be vast majority of the time
>> ---
>>  sys/kern/kern_exit.c | 8 +++++---
>>  1 file changed, 5 insertions(+), 3 deletions(-)
>>
>> diff --git a/sys/kern/kern_exit.c b/sys/kern/kern_exit.c
>> index 3f64343aea0e..a92d5cc0f642 100644
>> --- a/sys/kern/kern_exit.c
>> +++ b/sys/kern/kern_exit.c
>> @@ -477,9 +477,11 @@ exit1(struct thread *td, int rval, int signo)
>>  	sx_xunlock(&allproc_lock);
>>
>>  	sx_xlock(&proctree_lock);
>> -	PROC_LOCK(p);
>> -	p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE);
>> -	PROC_UNLOCK(p);
>> +	if ((p->p_flag & (P_TRACED | P_PPWAIT | P_PPTRACE)) != 0) {
>> +		PROC_LOCK(p);
>> +		p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE);
>> +		PROC_UNLOCK(p);
>> +	}
>
> just out of curiosuty, is this can be replaced by a single
> atomic_clear_rel ?
>

Not unless all other times p_flag is modified also start using atomics

>>
>>  	/*
>>  	 * killjobc() might drop and re-acquire proctree_lock to
>


-- 
Mateusz Guzik <mjguzik gmail.com>