Hi Dan,
On 1/25/22 13:20, Dan Carpenter wrote:
On Tue, Jan 25, 2022 at 01:06:14PM +0300, Pavel Skripkin wrote:
Hi Dan,
On 1/25/22 12:59, Dan Carpenter wrote:
[...]
> > > - service->trace = 0;
> > > atomic_set(&service->poll_flags, 0);
> >
> > Nit: atomic_set(0) can be also removed
> >
>
> No, let's not write code that assumes it knows atomic_t internals.
>
Isn't atomic_set() just (let's say) "magic" write to memory in all cases? Am
I missing some design issue here?
The point is atomic values should only be set using atomic_set/inc/dec()
etc. Anything else is wrong even if it works.
I think it might trigger a kcsan warning these days? I don't know what
that instrumentation does. But even in olden times, the rule was you
always had to use the accessor functions.
Ok, got it, thanks for explanations!
With regards,
Pavel Skripkin
