On 10/26/20 3:56 PM, Linus Torvalds wrote:
> On Mon, Oct 26, 2020 at 1:48 PM Luc Van Oostenryck
> <luc.vanoostenryck@xxxxxxxxx> wrote:
>>
>> On Mon, Oct 26, 2020 at 10:41:36AM -0700, Linus Torvalds wrote:
>>> But can you point to the actual 5.10-rc1 problem? I haven't seen the
>>> report, and afaik gcc will complain about this too ("label at end of
>>> compound statement") so I'm surprised sparse hits it..
>>
>> Well, I was surprised it wasn't caught and was ready to send a patch
>> but it can only happen when using sparse, see below, quite ironic:
>> drivers/scsi/qla2xxx/qla_tmpl.c:1052.
>> 1050 }
>> 1051
>> 1052 bailout:
>> 1053 #ifndef __CHECKER__
>> 1054 if (!hardware_locked)
>> 1055 spin_unlock_irqrestore(&vha->hw->hardware_lock, flags);
>> 1056 #endif
>> 1057 }
>
> Yeah, that "don't sparse this" pattern has been problematic before.
>
> I removed it.
>
> Bart, Arun - see commit 4525c8781ec0 ("scsi: qla2xxx: remove incorrect
> sparse #ifdef").
Hi Linus,
Thank you for having Cc-ed me. I agree that it's better to make the
kernel code compliant with the C standard than to make sparse accept
non-standard code.
The #ifndef __CHECKER__ / #endif that I added in 2015 in the above code
(commit 8d16366b5f23) was added because at that time I didn't see a
better solution.
BTW, personally I'm neither enthusiast about #ifndef __CHECKER__ /
#endif nor about if (expression) <locking statement>.
Bart.
