-----Original Message-----
From: Jason A. Donenfeld <Jason@zx2c4.com>
Sent: Monday, October 10, 2022 8:46 PM
To: Pankaj Gupta <pankaj.gupta@nxp.com>
Cc: 'Herbert Xu' <herbert@gondor.apana.org.au>; jarkko@kernel.org;
a.fatoum@pengutronix.de; gilad@benyossef.com; jejb@linux.ibm.com;
zohar@linux.ibm.com; dhowells@redhat.com; sumit.garg@linaro.org;
david@sigma-star.at; michael@walle.cc; john.ernberg@actia.se;
jmorris@namei.org; serge@hallyn.com; davem@davemloft.net;
j.luebbe@pengutronix.de; ebiggers@kernel.org; richard@nod.at;
keyrings@vger.kernel.org; linux-crypto@vger.kernel.org; linux-
integrity@vger.kernel.org; linux-kernel@vger.kernel.org; linux-security-
module@vger.kernel.org; Sahil Malhotra [off-list ref]; Kshitiz
Varshney [off-list ref]; Horia Geanta
[off-list ref]; Varun Sethi [off-list ref]
Subject: Re: [EXT] Re: [PATCH v0 3/8] crypto: hbk flags & info added to the
tfm

Caution: EXT Email

On Mon, Oct 10, 2022 at 11:15:00AM +0000, Pankaj Gupta wrote:

quoted

quoted

Nack.  You still have not provided a convincing argument why this is
necessary since there are plenty of existing drivers in the kernel
already providing similar features.

CAAM is used as a trusted source for trusted keyring. CAAM can expose
these keys either as plain key or HBK(hardware bound key- managed by
the hardware only and never visible in plain outside of hardware).

Thus, Keys that are inside CAAM-backed-trusted-keyring, can either be
plain key or HBK. So the trusted-key-payload requires additional flag
& info(key-encryption-protocol)  to help differentiate it from each
other. Now when CAAM trusted-key is presented to the kernel crypto
framework, the additional information associated with the key, needs
to be passed to the hardware driver. Currently the kernel keyring and
kernel crypto frameworks are associated for plain key, but completely
dis-associated for HBK. This patch addresses this problem.

Similar capabilities (trusted source), are there in other crypto
accelerators on NXP SoC(s). Having hardware specific crypto algorithm
name, does not seems to be a scalable solution.

Do you mean to say that other drivers that use hardware-backed keys do so
by setting "cra_name" to something particular? 

Yes.

Like instead of "aes"
it'd be "aes-but-special-for-this-driver"?

For example: ARM-Crypto-Cell prepends 'p':
- xts(paes) for xts(aes)
- xts(paes) for xts(aes)...etc.

 > If so, that would seem to break the

design of the crypto API. Which driver did you see that does this?  Or perhaps,
more generally, what are the drivers that Herbert is talking about when he
mentions the "plenty of existing drivers" that already do this?

I could find this driver " drivers/crypto/ccree/".
Reference file name is " drivers/crypto/ccree/cc_cipher.c"

Likewise, CAAM being a trust source, new cra_name would be need to deal with CAAM generated HBKs too.
We need to come up with something unique like: for eg:   p(xts(aes)) for xts(aes)             
   
Another trust source from NXP called DCP(drivers/crypto/mxs-dcp.c),  new cra_name would be needed for that too.
There are work in progress for other trust sources from NXP.

So, our approach is too provide generic solution, which can be extended to any trust source generating HBK.

Jason