[0/4] crypto: sun8i-ce: add Allwinner H616 support

Message

Andre Przywara June 16, 2024, 10:07 p.m. UTC

This series adds support for the crypto engine in the Allwinner H616
SoC. The IP and its capabilities are very similar to the H6, with the
major difference of the DMA engine supporting 34 bit wide addresses.
This is achieved by just shifting every address by 2 bits in the DMA
descriptors; Allwinner calls this "word addresses".
Patch 2/4 adds support for this by wrapping every address access in a
function that does the shift as needed. Patch 1/4 adds the new
compatible string to the binding, patch 3/4 adds that string to the
driver and enables the address shift for it. The final patch 4/4 adds
the DT node to the SoC .dtsi. Since this is an internal peripheral,
it's always enabled.

Corentin's cryptotest passed for me, though I haven't checked how fast
it is and if it really brings an advantage performance-wise, but maybe
people find it useful to offload that from the CPU cores.
One immediate advantage is the availability of the TRNG device, which
helps to feed the kernel's entropy pool much faster - typically before
we reach userland. Without the driver this sometimes takes minutes, and
delays workloads that rely on the entropy pool.

Please have a look and comment!

Cheers,
Andre

Andre Przywara (4):
  dt-bindings: crypto: sun8i-ce: Add compatible for H616
  crypto: sun8i-ce - wrap accesses to descriptor address fields
  crypto: sun8i-ce - add Allwinner H616 support
  arm64: dts: allwinner: h616: add crypto engine node

 .../bindings/crypto/allwinner,sun8i-ce.yaml   |  2 ++
 .../arm64/boot/dts/allwinner/sun50i-h616.dtsi | 10 +++++++
 .../allwinner/sun8i-ce/sun8i-ce-cipher.c      |  8 ++---
 .../crypto/allwinner/sun8i-ce/sun8i-ce-core.c | 29 ++++++++++++++++++-
 .../crypto/allwinner/sun8i-ce/sun8i-ce-hash.c |  6 ++--
 .../crypto/allwinner/sun8i-ce/sun8i-ce-prng.c |  6 ++--
 .../crypto/allwinner/sun8i-ce/sun8i-ce-trng.c |  2 +-
 drivers/crypto/allwinner/sun8i-ce/sun8i-ce.h  | 10 +++++++
 8 files changed, 61 insertions(+), 12 deletions(-)

Comments

Ryan Walklin June 22, 2024, 11:37 p.m. UTC | #1

On Mon, 17 Jun 2024, at 10:07 AM, Andre Przywara wrote:

Thanks Andre!

> Corentin's cryptotest passed for me, though I haven't checked how fast
> it is and if it really brings an advantage performance-wise, but maybe
> people find it useful to offload that from the CPU cores.

Running the rngtest gives the following output:

localhost:~# rngtest -c 10000 < /dev/random
rngtest 6.16
Copyright (c) 2004 by Henrique de Moraes Holschuh
This is free software; see the source for copying conditions.  There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

rngtest: starting FIPS tests...
rngtest: bits received from input: 200000032
rngtest: FIPS 140-2 successes: 9991
rngtest: FIPS 140-2 failures: 9
rngtest: FIPS 140-2(2001-10-10) Monobit: 0
rngtest: FIPS 140-2(2001-10-10) Poker: 2
rngtest: FIPS 140-2(2001-10-10) Runs: 2
rngtest: FIPS 140-2(2001-10-10) Long run: 5
rngtest: FIPS 140-2(2001-10-10) Continuous run: 0
rngtest: input channel speed: (min=144.496; avg=808.068; max=866.977)Mibits/s
rngtest: FIPS tests speed: (min=17.199; avg=60.937; max=62.949)Mibits/s
rngtest: Program run time: 3369060 microseconds

So looks like a nice performance boost. 

> One immediate advantage is the availability of the TRNG device, which
> helps to feed the kernel's entropy pool much faster - typically before
> we reach userland. Without the driver this sometimes takes minutes, and
> delays workloads that rely on the entropy pool.

CRNG bringup now also very fast:

[    1.114790] sun8i-ce 1904000.crypto: CryptoEngine Die ID 0
[    1.116253] random: crng init done

Tested-by: Ryan Walklin <[e-mail]>

Regards,

Ryan