Re: [RFC PATCH 2/4] xen/arm64: bitops: justify uninitialized variable inside a macro

[Julien Grall <julien@xxxxxxx>]

Hi,

On 17/07/2023 13:16, Nicola Vetrini wrote:
> On 16/07/23 18:50, Julien Grall wrote:
> > Hi,
> >
> > On 16/07/2023 10:20, Luca Fancellu wrote:
> > > > On 14 Jul 2023, at 14:05, Julien Grall <julien@xxxxxxx> wrote:
> > > >
> > > > Hi,
> > > >
> > > > On 14/07/2023 12:49, Nicola Vetrini wrote:
> > > > > The macro 'testop' expands to a function that declares the local
> > > > > variable 'oldbit', which is written before being set, but is such a
> > > > > way that is not amenable to automatic checking.
> > > > The code is pretty straightforward. So I am not entirely sure why 
> > > > Eclair is not happy. Is it because the value is set by assembly code?
> Exactly.
I am a bit puzzled. There are other values read from assembly (e.g. 
READ_SYSREG(), read_atomic()) within Xen. This is including 'res' in the 
same function. So can you elaborate, why 'oldbit' is a problem but not 
the others?
Cheers,

--
Julien Grall