Re: [PATCH] fscrypt: don't use hardware offload Crypto API drivers

[Eric Biggers <ebiggers@kernel.org>]

On Wed, Jun 25, 2025 at 08:44:45AM -0400, Theodore Ts'o wrote:
> On Tue, Jun 24, 2025 at 11:32:52PM -0700, Eric Biggers wrote:
> > 
> > That was the synchronous throughput.  However, submitting multiple requests
> > asynchronously (which again, fscrypt doesn't actually do) barely helps.
> > Apparently the STM32 crypto engine has only one hardware queue.
> > 
> > I already strongly suspected that these non-inline crypto engines
> > aren't worth using.  But I didn't realize they are quite this bad.
> > Even with AES on a Cortex-A7 CPU that lacks AES instructions, the
> > CPU is much faster!
> 
> I wonder if the primary design goal of the STM32 crypto engine is that
> it might reduce power consumption --- after all, one of the primary
> benchmarketing metrics that vendors care about is "hours of You Tube
> watch time" --- and decryptoing a video stream doesn't require high
> performance.
> 
> Given that the typical benchmarketing number which handset vendors
> tend to care about is SQLite transactions per second, maybe they
> wouldn't be all that eager to use the crypto engine.  :-)
> 

My STM32MP157F-DK2 board (with screen removed) is pulling 1.5W regardless of
whether it's running the benchmark with the STM32 crypto engine or with the NEON
bit-sliced code.  However, the NEON bit-sliced code finishes 5 times faster.

- Eric