Re: [PATCH] x86: mark hypercall argument regs clobbering for intended fall-through

[Jan Beulich <jbeulich@xxxxxxxx>]

On 09.06.2021 14:49, Andrew Cooper wrote:
> On 09/06/2021 11:34, Jan Beulich wrote:
>> The CIDs below are all for the PV side of things, but also take care of
>> the HVM side.
>>
>> Coverity-ID: 1485896, 1485901, 1485906, 1485910, 1485911, 
>> Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>
>> ---
>> Let's see whether Coverity actually understands the (relatively) new
>> pseudo-keyword.
> 
> This is exceedingly disappointing.  Coverity used to have the only
> sensible rule for not causing spurious fallthrough warnings, but this
> has apparently regressed.
> 
> Coverity works on the AST, so ought to be after GCC has interpreted
> __attribute__((__fallthrough__)) if applicable.
> 
> However, I doubt it will work in the fallback case, because #define
> fallthrough looks dubious.  To trigger the older logic, the /*
> fallthrough */ comment needs to be the final thing before the next case
> label, and it isn't with the added semicolon.
> 
> Given that this pseudo-keyword is restricted to the SMMU driver for now,
> we don't actually know if Coverity likes it or not.

When it was introduced, I did specifically ask whether it pleases
Coverity. I was told it would, and I had no proof to the contrary, so
I had to accept what I was told. My asking at the time was precisely
to avoid having to have two forms of annotation on every single
legitimate / intentional fall-through case.

Jan