Re: [PATCH v3 1/6] xen: add reference counter support

[Roger Pau Monné <roger.pau@xxxxxxxxxx>]

On Thu, Mar 16, 2023 at 05:32:38PM +0100, Jan Beulich wrote:
> On 16.03.2023 17:19, Roger Pau Monné wrote:
> > On Tue, Mar 14, 2023 at 08:56:29PM +0000, Volodymyr Babchuk wrote:
> >> +static inline void refcnt_get(refcnt_t *refcnt)
> >> +{
> >> +    int old = atomic_add_unless(&refcnt->refcnt, 1, 0);
> > 
> > Occurred to me while looking at the next patch:
> > 
> > Don't you also need to print a warning (and saturate the counter
> > maybe?) if old == 0, as that would imply the caller is attempting
> > to take a reference of an object that should be destroyed? IOW: it
> > would point to some kind of memory leak.
> 
> Hmm, I notice the function presently returns void. I think what to do
> when the counter is zero needs leaving to the caller. See e.g.
> get_page() which will simply indicate failure to the caller in case
> the refcnt is zero. (There overflow handling also is left to the
> caller ... All that matters is whether a ref can be acquired.)

Hm, likely.  I guess pages never go away even when it's refcount
reaches 0.

For the pdev case attempting to take a refcount on an object that has
0 refcounts implies that the caller is using leaked memory, as the
point an object reaches 0 it supposed to be destroyed.

Returning success would be fine, as then for the pdev use-case we
could print a warning and likely take some action to prevent further
damage if possible.

Thanks, Roger.