On Sun, Dec 29, 2024 at 02:10:11PM -0800, Eric Biggers wrote: > > This seems strictly worse than my version, so I don't plan to do this. It's > more lines of code, and it causes an extra push and pop to be needed in > skcipher_walk_virt() to free up enough registers to hold all values at once. It > may be intended that API users are supposed to use the helper functions instead > of accessing the algorithm struct directly, but this code is not a user; it's > part of the API implementation in crypto/skcipher.c. There are already lots of > other direct accesses to the algorithm struct in the same file, and even another > in skcipher_walk_virt() already. The helper functions are pointless in this > context and just cause problems like the one this patch is fixing. OK, but please add a comment where the accesses occur so that people won't try to turn this back. Thanks, -- Email: Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx> Home Page: http://gondor.apana.org.au/~herbert/ PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
