Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx> writes: > On Mon, Jun 10, 2024 at 08:13:14PM -0700, Eric Biggers wrote: >> >> I thought the rule is that there needs to be an in-kernel user to add algorithms >> to the crypto API? Is there any precedent for adding new algorithms purely so >> that accelerators that implement them can be accessed from userspace via AF_ALG? > > I agree. Perhaps this driver could instead be added as a simple > char device that is then used directly by the intended user without > going through the Crypto API at all. > > That would make it a lot simpler. Thanks for all the support Herbert, Eric. Just wanted to confirm, if this is being rejected primarily because 1. there is no in-kernel user for crc64-iso3309 2. or poor performance benefit of using it from userspace The context for asking is that we have another superset IP known as MCRC (this one is MCRC64), which supports crc8/16/32/64(iso-3309). That IP has working DMA and will give good offloading numbers. We are planning to send drivers for crc8/16/32 for MCRC 1.should I put efforts for crc64-iso3309 as well or 2.drop the crc64-iso3309 and send only for remaining crc8/16/32(standard algorithms with already in-kernel user). All our devices either have MCRC or MCRC64. Thanks, Kamlesh
