Re: [PATCH v2 0/7] crypto: sha256 - Merge 2 separate C implementations into 1, put into separate library

From: Hans de Goede
Date: Mon Aug 19 2019 - 15:38:26 EST


On 19-08-19 17:08, Ard Biesheuvel wrote:
On Sat, 17 Aug 2019 at 17:24, Hans de Goede <hdegoede@xxxxxxxxxx> wrote:

Hi All,

Here is v2 of my patch series refactoring the current 2 separate SHA256
C implementations into 1 and put it into a separate library.

There are 3 reasons for this:

1) Remove the code duplication of having 2 separate implementations

2) Offer a separate library SHA256 implementation which can be used
without having to call crypto_alloc_shash first. This is especially
useful for use during early boot when crypto_alloc_shash does not
work yet.

3) Having the purgatory code using the same code as the crypto subsys means
that the purgratory code will be tested by the crypto subsys selftests.

This has been tested on x86, including checking that kecec still works.

This has NOT been tested on s390, if someone with access to s390 can
test that things still build with this series applied and that
kexec still works, that would be great.

Changes in v2:
- Use put_unaligned_be32 to store the hash to allow callers to use an
unaligned buffer for storing the hash
- Add a comment to include/crypto/sha256.h explaining that these functions
now may be used outside of the purgatory too (and that using the crypto
API instead is preferred)
- Add sha224 support to the lib/crypto/sha256 library code
- Make crypto/sha256_generic.c not only use sha256_transform from
lib/crypto/sha256.c but also switch it to using sha256_init, sha256_update
and sha256_final from there so that the crypto subsys selftests fully test
the lib/crypto/sha256.c implementation

This looks fine to me, although I agree with Eric's feedback regarding
further cleanups.

Ack, as I already told Eric I'm happy to do a follow up series with
the necessary local static function renames so that we can then merge
sha256.h into sha.h .

Also, now that we have a C library, I'd like to drop
the dependency of the mips and x86 sha256 algo implementations up
sha256_generic.c, and use the library directly instead (so that
sha256-generic is no longer needed on x86 or mips)

I assume this is more of a generic remark and not targeted towards me?