openssl: Fix detection of ARMv7 and ARM64 CPU features

[allanjude]

Summary

OpenSSL assumes the same value for AT_HWCAP at Linux
So it ends up calling elf_auxv_info() with AT_CANARY which
returns ENOENT, and all acceleration features are disabled.

With this, my ARM64 test machine, the benchmark
openssl speed -evp aes-256-gcm is nearly 20x faster
going from 100 MB/sec to 2000 MB/sec

PR: 259937
Sponsored by: Klara Inc.

[manu]

Fixes: 88e852c0b5c872b1a ("OpenSSL: Merge OpenSSL 1.1.1j")

[manu]

This would need to be an EN as 13.0 is affected too

[imp]

Do you need to do the same for AT_hWCAP2? That's a frequent source of problems in bsd-user, so much so we've had to implement a faithful version there. Don't know if that's for crypto or for $OTHER_THINGS though.

crypto/openssl/crypto/armcap.c

118

I think this should be AT_HWCAP2 on FreeBSD

[allanjude]

Update to the cleaner style patch

[jhb]

Do you have a pointer to an upstream OpenSSL PR? I'm happy to comment on it if so (if only to provide a LGTM)

crypto/openssl/crypto/armcap.c

109

I would be tempted to stick these fallbacks under an #ifdef __linux.