I was testing dm-crypt performance with a Freescale P1022 board with a recent kernel and was getting IO errors while doing testing with LUKS. Investigation showed that all hardware FIFO slots were filling and the driver was returning EAGAIN to the block layer, which is not an expected response for an async crypto implementation. The following patch series adds a few small fixes, and reworks the submission path to use the crypto_queue mechanism to handle the request backlog. Changes since v1: - Ran checkpatch.pl - Split the path for submitting new requests vs. issuing backlogged requests. - Avoid enqueuing a submitted request to the crypto queue unnecessarily. - Fix return paths where CRYPTO_TFM_REQ_MAY_BACKLOG is not set. Martin Hicks (5): crypto: talitos: Simplify per-channel initialization crypto: talitos: Remove MD5_BLOCK_SIZE crypto: talitos: Fix off-by-one and use all hardware slots crypto: talitos: Reorganize request submission data structures crypto: talitos: Add software backlog queue handling drivers/crypto/talitos.c | 240 +++++++++++++++++++++++++++------------------- drivers/crypto/talitos.h | 44 +++++++-- 2 files changed, 177 insertions(+), 107 deletions(-) -- 1.7.10.4 -- 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
