Messages in this thread | | | Date | Thu, 20 Mar 2008 16:43:13 +0530 (IST) | From | Arun Raghavan <> | Subject | Re: [PATCH] [keys] Always use key_get() to increment key refcount |
| |
On Wed, 19 Mar 2008, David Howells wrote:
> Arun Raghavan <arunsr@cse.iitk.ac.in> wrote: > > > Patch to use key_get() wherever the keys code manually increments the > > key refcount. > > > > This should make debugging a little simpler for clients, since it > > becomes easier to track where a key's refcount changes. > > The problem with this is that key_get() is not simply an atomic_inc(). You > end up introducing an extra conditional into each of these places where one is > not required. Now it's possible that the compiler's optimiser is sufficiently > clever to get rid of them all, but do you guarantee that?
Is this really a significant performance penalty? If yes, wouldn't a likely() be sufficient to mitigate the penalty?
The same arguments should also be applied to key_put(), I guess.
Regards, Arun
| |