On 9/17/2026 2:30 AM, Stephen Hemminger wrote:
On Wed, 16 Sep 2026 13:18:09 +0100 Anatoly Burakov <[email protected]> wrote:+/* + * Simplistic reference counter. Not thread-safe, not atomic. + */ +struct ci_refcount { + uint32_t count; +}; + +/** + * Take a reference. Returns true if this is the first reference. + * + * @param ref + * Reference count to acquire. + * @return + * true if this is the first reference, false otherwise. + */ +static inline bool +ci_refcount_acquire(struct ci_refcount *ref) +{ +#ifdef CI_DEBUG + if (ref->count == UINT32_MAX) + rte_panic("ci_refcount overflow\n"); +#endif + return ref->count++ == 0; +} +Would it help if EAL has a generic refcount? I would make the refcount overflow/underflow still fatal, should never happen.
In the next revision I changed the name to `ci_usage_count_t` which is a more descriptive name than a refcount (less connotations), and made it a direct uint32_t instead of a struct.
It would be nice to have this in EAL, but I'm not sure of the scope of such work and what sort of use cases should this cover. It can be promoted to EAL later if you think it's worth doing, and I'll gladly submit a patch for the next release and we can work both towards defining what sorts of use cases we want to cover and convert the codebase to use those, but I would defer it to next release.
-- Thanks, Anatoly
