On Sun, Oct 09, 2011 at 01:00:08AM +0200, Sverd Johnsen wrote: > Hey, > > how do they work? Should one use them? Does it work with all ciphers? > Do they exclude each other? They don't exclude each other, but using both in an algorithm instance does not make much sense. It depends on your usecase what to choose. If a single cpu sends a lot of crypto requests (like it happens on IPsec networking), it makes sense to parallelize with pcrypt. Usually you don't need to setup a cryptd algorithm expicit. If an algorithm needs cryptd, it cooses it automatically. It is used e.g. to move the crypto requests out of atomic context (AES-NI does this.) > > Do they need module parameters? What if they are built-in? They don't need module parameters. It does not matter much if they are build-in or a module. If they are build-in, they can be just instantiated. If they are build as module, the crypto layer tries to autoload the module if it is not already loaded. If you want to use pcrypt/cryptd on an algorithm that does not coose them automatically, you have to instantiate (load) these algorithms. Right now, using 'tcrypt' is the only way to instantiate such nonstandard algorithms. That's kind of a hack, but things are getting much easier as soon as we have the crypto userspace configuration API merged. > How can I verify that they work? You can look at /proc/crypto, this shows which algorithms are used. You should find cryptd/pcrypt in the algorithm drivers name if they are instantiated. > How do they work? There is absolutely no fucking information > whatsoever, neither in the help text nor in the docs. Am i missing > something? Indeed, there is no documentation. > > Talking mostly about typical disk encryption on SMP machines via > dm-crypt, interested in best performance. (without AES-NI) dm-crypt uses it's own workqueue, so you don't need cryptd here. On recent kernels this workqueue is multithreaded, so if the crypto requests are send from multiple cpus it behaves well parallelized. If only one cpu sends crypto requests, it is serialized to this cpu. pcrypt could help here, but it still lacks ablkcipher support what is needed for dm-crypt. -- To unsubscribe from this list: send the line "unsubscribe linux-crypto" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
