Am Donnerstag, 16. Juni 2011, 12:29:40 schrieb David Herrmann: > On Thu, Jun 16, 2011 at 12:18 PM, Oliver Neukum <oneukum@xxxxxxx> wrote: > > Am Donnerstag, 16. Juni 2011, 11:20:18 schrieb Jiri Kosina: > >> > Here an smp_wmb() > >> > It is redundant because input_register_device takes a spinlock which > >> > implies a barrier, but you'll make the automated checking tools happier. > >> > >> Yeah, spinlock implies SMP barrier implicitly. > >> > >> I don't think adding barriers just to make some buggy automatic checker > >> happy is worth it. > > > > Fine by me as long as the problem is understood. > > Does that mean that I can omit the smp_rmb(), too? I do not like > relying on input_register_device to use a spinlock. It makes reading > the code hard, so I added both barriers. Should I remove them again or > keep them? The full explanation is in Documentation/atomic-ops.txt Basically, whenever you do if (!descriptor->flag) return -ERR; descriptor->pointer->member = expression; //or anything else with the pointer you need smp_rmb() in between If you do if (!descriptor->flag) return -ERR; // no use of pointer here spin_lock(&lock); descriptor->pointer->member = expression; //or anything else with the pointer You are save and can do without the smp_rmb() But, you still need smp_rmb() if you do: if (!descriptor->flag) return -ERR; // no use of pointer here spin_lock(&descriptor->pointer->lock); All clear? I must say I stopped listing and checking all the places you use the flag, but I think some were buggy. Regards Oliver -- To unsubscribe from this list: send the line "unsubscribe linux-input" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
