This series applies onto Tero's series [0], and addresses a number of
additional issues that exist in the omap-aes driver that aren't being
addresses by Tero's fixes.
Note that the resulting code is still not 100% correct: an issue remains
where inputs into GCM consisting solely of assocdata are not being processed
correctly, e.g.,
alg: aead: gcm-aes-omap encryption test failed (wrong result) on test vector
"random: alen=38 plen=0 authsize=16 klen=32",
cfg="random: inplace may_sleep use_digest src_divs=[100.0%@+19] iv_offset=31"
I have no idea how to fix this, so I'll leave this to people that know this
hardware and have access to the Sitara TRM.
Note that I also spotted some issues in the SHAM driver, i.e.,
alg: ahash: omap-sha1 test failed (wrong result) on test vector
"random: psize=7928 ksize=0", cfg="random: inplace use_final
src_divs=[5.64%@+13, 59.70%@+18, <flush>31.53%@+4072,
<flush,nosimd>3.13%@alignmask+263]"
alg: ahash: omap-hmac-sha256 test failed (wrong result) on test vector
"random: psize=960 ksize=37", cfg="random: inplace use_final
src_divs=[32.54%@+2449, 17.18%@+4, <flush>50.28%@+1] iv_offset=31"
All of these failures are triggered by CONFIG_CRYPTO_MANAGER_EXTRA_TESTS,
so they will not show up when using the standard set of test vectors.
[0] https://lore.kernel.org/linux-crypto/20191017122549.4634-1-t-kristo@xxxxxx/
Cc: linux-omap@xxxxxxxxxxxxxxx
Cc: Tero Kristo <t-kristo@xxxxxx>
Ard Biesheuvel (6):
crypto: omap-aes - reject invalid input sizes for block modes
crypto: omap-aes-ctr - set blocksize to 1
crypto: omap-aes-gcm - deal with memory allocation failure
crypto: omap-aes-gcm - add missing .setauthsize hooks
crypto: omap-aes-gcm - check length of assocdata in RFC4106 mode
crypto: omap-aes-gcm - use the AES library to encrypt the tag
drivers/crypto/omap-aes-gcm.c | 119 ++++++++------------
drivers/crypto/omap-aes.c | 33 ++----
drivers/crypto/omap-aes.h | 10 +-
3 files changed, 61 insertions(+), 101 deletions(-)
--
2.17.1
