Re: [RFC] time: xtime_lock is held too long

[Eric Dumazet <eric.dumazet@gmail.com>]

Le vendredi 06 mai 2011 à 18:59 +0200, Andi Kleen a écrit :

> If you have a better way to make it faster please share it.

Ideally we could use RCU :)

Have whatever state hold in one structure (possibly big, it doesnt
matter) and switch pointer in writer once everything is setup in new
structure.

struct {
	struct timespec xtime;
	struct timespec wall_to_monotonic;
	...
	} time_keep_blob;

struct time_keep_blob __rcu *xtime_cur;

ktime_t ktime_get(void)
{
	const struct time_keep_blob *xp;
	s64 secs, nsecs;

	rcu_read_lock();
	xp = rcu_dereference(xtime_cur);
	secs = xp->xtime.tv_sec + xp->wall_to_monotonic.tv_sec;
	nsecs = xp->xtime.tv_nsec + xp->wall_to_monotonic.tv_nsec;
	nsecs += timekeeping_get_ns(xp);
	rcu_read_unlock();

	return ktime_add_ns(ktime_set(secs, 0), nsecs);
}

I dont know timekeeping details, maybe its necessary to loop if
xtime_cur changes :

ktime_t ktime_get(void)
{
	const struct time_keep_blob *xp;
	s64 secs, nsecs;

	rcu_read_lock();
	do {
		xp = rcu_dereference(xtime_cur);
		secs = xp->xtime.tv_sec + xp->wall_to_monotonic.tv_sec;
		nsecs = xp->xtime.tv_nsec + xp->wall_to_monotonic.tv_nsec;
		nsecs += timekeeping_get_ns(xp);
	} while (rcu_dereference(xtime_cur) != xp);
	rcu_read_unlock();
	return ktime_add_ns(ktime_set(secs, 0), nsecs);
}