On Fri, Jan 25, 2019 at 10:36:25AM +0100, Ard Biesheuvel wrote: > Both the GHASH and the CRC-T10DIF SIMD routines are based on 64-bit > polynomial multiplication, but fall back to the slower 8-bit polynomial > instructions if the optional 64-bit PMULL instruction is not implemented. > > This means we can only ever test one of the two on a single system, > which is unfortunate. So instead, register both versions on hardware > that supports them, and let the crypto API decide which is the preferred > one based on priorities. That way, both will be tested, but only the > fastest available one will be used. > > Ard Biesheuvel (2): > crypto: arm64/crct10dif - register PMULL variants as separate algos > crypto: arm64/ghash - register PMULL variants as separate algos > > arch/arm64/crypto/crct10dif-ce-glue.c | 54 +++++++-- > arch/arm64/crypto/ghash-ce-glue.c | 118 +++++++++++++++----- > 2 files changed, 132 insertions(+), 40 deletions(-) All applied. Thanks. -- Email: Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx> Home Page: http://gondor.apana.org.au/~herbert/ PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
