[Intel-gfx] [PATCHv2] lib/ratelimit: Lockless ratelimiting

Dmitry Safonov dima at arista.com
Tue Jul 3 20:59:14 UTC 2018


On Tue, 2018-06-26 at 21:41 +0300, Andy Shevchenko wrote:
> > > > @@ -42,9 +41,10 @@ static inline void
> > > > ratelimit_state_init(struct
> > > > ratelimit_state *rs,
> > > >  {
> > > >         memset(rs, 0, sizeof(*rs));
> > > > 
> > > > -       raw_spin_lock_init(&rs->lock);
> > > >         rs->interval    = interval;
> > > >         rs->burst       = burst;
> > > > +       atomic_set(&rs->printed, 0);
> > > > +       atomic_set(&rs->missed, 0);
> > > 
> > > Can it be
> > > 
> > > *rs = RATELIMIT_STATE_INIT(interval, burst);
> > > 
> > > ?
> > > 
> > > (Yes, the '(struct ratelimit_state)' has to be added to macro to
> > > allow this)
> > 
> > Sure.
> 
> This part, by the way, potentially can be split into preparatory
> patch. Please, double check if it possible to do this way.

Hmm, I tried this way:
:#define RATELIMIT_STATE_INIT(name, interval_init, burst_init) ({ \
:               struct ratelimit_state name = {                   \
:                       .lock = __RAW_SPIN_LOCK_UNLOCKED(name.lock), \
:                       .interval       = interval_init,          \
:                       .burst          = burst_init,             \
:               };                                                \
:               name;                                             \
:       })

but the expression becomes non-constant, so it fails to compile in
definitions of globals.

I think I'll change it to
struct ratelimit_state tmp = RATELIMIT_STATE_INIT(...);
*rs = tmp;

Not perfect, but we did memset() and set elements after, so it's kinda
the same.

-- 
Thanks,
             Dmitry


More information about the Intel-gfx mailing list