Re: git: 7ae879b14a20 - main - kern_procctl(): convert the function to be table-driven

On 10/19/21 1:51 PM, Cy Schubert wrote:
> In message <91ebf9d8-5547-8570-18cb-26a58baf89ba@FreeBSD.org>, John Baldwin
> wri
> tes:
>> On 10/19/21 1:35 PM, Cy Schubert wrote:
>>> In message <202110192004.19JK4jN3069844@gitrepo.freebsd.org>, Konstantin
>>> Belous
>>> ov writes:
>>>> The branch main has been updated by kib:
>>>>
>>>> URL: https://cgit.FreeBSD.org/src/commit/?id=7ae879b14a2086df521c59c4a379d
>> 3a0
>>>> 72e08bc6
>>>>
>>>> commit 7ae879b14a2086df521c59c4a379d3a072e08bc6
>>>> Author:     Konstantin Belousov <kib@FreeBSD.org>
>>>> AuthorDate: 2021-10-15 18:57:17 +0000
>>>> Commit:     Konstantin Belousov <kib@FreeBSD.org>
>>>> CommitDate: 2021-10-19 20:04:34 +0000
>>>>
>>>>       kern_procctl(): convert the function to be table-driven
>>>>       
>>>>       Reviewed by:    emaste, markj
>>>>       Sponsored by:   The FreeBSD Foundation
>>>>       MFC after:      1 week
>>>>       Differential revision:  https://reviews.freebsd.org/D32513
>>>> ---
>>>>    sys/kern/kern_procctl.c | 123 +++++++++++++++++++++++++++---------------
>> ----
>>>> --
>>>>    1 file changed, 69 insertions(+), 54 deletions(-)
>>>>
>>>> diff --git a/sys/kern/kern_procctl.c b/sys/kern/kern_procctl.c
>>>> index eb36f0822938..90c5e63c7219 100644
>>>> --- a/sys/kern/kern_procctl.c
>>>> +++ b/sys/kern/kern_procctl.c
>>>> @@ -949,7 +957,14 @@ kern_procctl(struct thread *td, idtype_t idtype, id_t
>>   id
>>>> , int com, void *data)
>>>>    		error = EINVAL;
>>>>    		break;
>>>>    	}
>>>> -	if (tree_locked)
>>>> -		sx_unlock(&proctree_lock);
>>>> +
>>>> +	switch (cmd_info->lock_tree) {
>>>> +	case SA_XLOCKED:
>>>> +		sx_xunlock(&proctree_lock);
>>>> +		break;
>>>> +	case SA_SLOCKED:
>>>> +		sx_sunlock(&proctree_lock);
>>>> +		break;
>>>> +	}
>>>>    	return (error);
>>>>    }
>>>>
>>>
>>> Should SA_* in fact be LA_*? SA_* in sys/sx.h assumes INVARIANTS whereas
>>> LA_* in sys/lock.h has no such requirement.
>>
>> Both are for "assertions".  The LA_* constants aren't really public but are
>> the values used for witness_assert() that various foo_assert() routines in
>> locking APIs (mtx_assert/sx_assert, etc.) can use.  For locking APIs, the
>> type-specific macros are the ones you use, e.g. SA_* with sx_assert().
>>
>> Given that, SA_* is the closest match here.
> 
> We'll need some #ifdefs for non-INVARIANTS built kernels, as it stands
> buildkernel is broken.

So it is helpeful if your e-mail starts with "the build is broken". :)

That said, I think the issue is that SA_* (and LA_*) have to date only been
used for assertions and are thus only relevant when INVARIANTS is defined.
It's probably simplest to just expose SA_* always if that is what is
needed.

-- 
John Baldwin