Hi Dmitry, > On Fri, May 28, 2010 at 05:10:32PM +0200, Henrik Rydberg wrote: >> In preparation of larger buffer needs, convert the current per-client >> circular buffer to a single buffer with multiple clients. Since there >> is only one writer but potentially many readers, use the somewhat >> discouraged rwlock_t pattern. > > rwlock is completely unsuitable here: > > 1. You must absiolutely not starve writers since they are quite likely > are interrupt handlers. If anything they should get priority. Thanks, I overlooked the starvation argument here. > 2. The amount of time the thing is locked is very miniscule, I'd say > that regular spinlocks will still give better performance. I still have a feeling it should be possible to do this lock-less, but I can see all docs and your comments pointing towards using the spinlock, so consider it changed. > > Also, use of atomic is funky, I do not think it does whta you think it > would... Could you explain more why you are using it? The use of atomic here serves a declarative purpose, I imagine the assembly code is more or less identical to the current evdev code. I personally think it is good to be able to see which variables are involved in a thread-to-thread exchange. Removing it is not a problem. Is it just me being an old fart, not understanding that the atomic int has been assumed for the last half decade, or why is it considered funky? Cheers, Henrik -- 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
