> On Fri, 7 Feb 2025 15:21:45 -0800
> lon...@linuxonhyperv.com wrote:
>
> > + /* At least one eth_dev is probed, init shared data */
> > + if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
> > + rte_spinlock_lock(&mana_shared_data_lock);
> > + mana_local_data.primary_cnt++;
> > + rte_spinlock_unlock(&mana_shared_data_lock);
> > + } else {
> > + rte_spinlock_lock(&mana_shared_data_lock);
> > + mana_local_data.secondary_cnt++;
> > + rte_spinlock_unlock(&mana_shared_data_lock);
> > +
> > + rte_spinlock_lock(&mana_shared_data->lock);
> > + mana_shared_data->secondary_cnt++;
> > + rte_spinlock_unlock(&mana_shared_data->lock);
>
> If all you are doing is wanting a MP safe counter, use atomic operations
> instead of
> the overhead of a spin lock.
mana_shared_data_lock is also used to protect init_done, mana_shared_mz and
mana_shared_data. That's why I'm reusing it for primary_cnt and secondary_cnt
as those values are initialized in locked context at the beginning for the
local process. I think this will make the code look clean.
I can remove mana_shared_data->lock and use atomic for
mana_shared_data->secondary_cnt.
Will send v2.
Thanks,
Long
