adds fix to generate _armv8_sha512_probe func

[yavtuk]

Instruction should be generated by using .inst directive or
using sha512su0 directly. In another case, symtab section
contains an attribute indicating the presence of data inside
the function, which does not allow disassembly this func correctly

[t8m]

I'm OK with CLA: trivial.

[openssl-machine]

This pull request is ready to merge

[yavtuk]

thanks 😃

[tom-cosgrove-arm]

Why not change all of the .longs to .inst in this PR? If it's needed for one, it's surely needed for all the others...

Obviously a change to the commit message required too, something along the lines of

Prefer .inst rather than .long for probe instructions in arm64cpuid.pl

Fixes an issue disassembling the functions because the symtab contains an
attribute indicating the presence of data within them.

[yavtuk]

No problem, I’ll check and fix other places

[yavtuk]

@tom-cosgrove-arm one more question about data in text for armv8,
what do you think about to move data from text to rodata section?
example,
using assembler for armv8
https://github.com/openssl/openssl/blob/master/crypto/sha/asm/keccak1600-armv8.pl#L85
here we explicitly specify that the data has to be in .text section

using C
https://github.com/openssl/openssl/blob/master/crypto/sha/keccak1600.c#L87
here compiler put it to .rodata section

[tom-cosgrove-arm]

@yavtuk Moving data from one section to another is quite a different change than using .inst instead of .long, even if logically it's just the converse. We would need to be certain that it would work with all reasonable assemblers and systems, and that the Perl xlate scripts handle it correctly. While it's probably okay, it's also probably an unnecessary risk if it's not causing any actual problems in practice. And at the very least it would be a different PR. But others might have a different opinion on this - @zorrorffm what are your thoughts?

[yavtuk]

fixed in 3 more places in the crypto/arm64cpuid.pl file

[zorrorffm]

I would prefer to keep the constant data in text section instead of rodata section in the sense of performance. The data in rodata section maybe is far from the place where it is loaded, so ldr instruction should be used to load rodata data. If the constant data is in text section, adr/adrp instruction can be used to load that data. adr and adrp are much more efficient than ldr instruction.

[yavtuk]

@zorrorffm thanks

[t8m]

Still OK by me also with CLA: trivial. @paulidale please re-approve.

[paulidale]

Good and agreed trivial.

[yavtuk]

thanks

[openssl-machine]

24 hours has passed since 'approval: done' was set, but as this PR has been updated in that time the label 'approval: ready to merge' is not being automatically set. Please review the updates and set the label manually.

[t8m]

Merged to master branch. Thank you for your contribution.