Fix segemation errror for ARM64/ Apple siliion

[ghassanmas]

Relevant issues are

• After installing pycryptodome in arm architecture, python -m Crypto.SelfTest Segmentation fault (core damped) #509
• Segmentation fault on macOS 11/arm64 (M1) #506

Description of work:

Solve segmentation error which might occur on apple silicon when using GMP,
the work around was to raise a new exception in Numbers module, if ARM64 is detected as the platform,
so that Numbers would export IntegerCustom or IntegerNative instead of IntegerGMP.

[ghassanmas]

Please note, that the segmentation error doesn't occur if the system/platform failes to load the GMP at,

pycryptodome/lib/Crypto/Math/_IntegerGMP.py

Line 98 in 5dace63

lib = load_lib("gmp", gmp_defs)

, since the ImportError will be raised at:

pycryptodome/lib/Crypto/Math/Numbers.py

Line 36 in 5dace63

except (ImportError, OSError, AttributeError):

.

However if the GMP is in the system/platform, (which was the case for me when python3.9 was installed through brew) the load_lib will succeed, but then an function is called when for example creating a new instance of the Integer at:

pycryptodome/lib/Crypto/Math/_IntegerGMP.py

Line 162 in 5dace63

_gmp.mpz_init(self._mpz_p)

A new instance of Integer is created or used by the RSA.PublicKey, when calling it's generate fucntion, hence this

Also lastly. for reference here is what GMP lastest release note state v6.2.1:

Issues with GMP 6.2.1:
While we added support for Apple's new Arm based computers, our support has a problem. The problem is that Apple reserves CPU register x18, but GMP's mpn/arm64 assembly code uses that register. While GMP runs fine in our tests, we expect things to go awry in some execution situation. (Apple has not been kind enough to specify how they use x18. Therefore, we don't know what the consequences of using x18 might be.)

[regisb]

Could someone from pycryptodome please review this ASAP? I don't mean to be pushy, but (in my understanding) pycryptodome is now broken for all Mac OS users running Python 3.9, which affects our project.

Thanks a lot, and keep up the great work!

[Legrandin]

Is this fix really needed on all ARM64 platforms, or only on the M1 (one of the several types of ARM64)?
I will look into this for the M1, but we should not have regressions on other platforms unless it's needed.

[ghassanmas]

@Legrandin No certainly it's only needed when for ARM and drawin/MAC. And actualy @regisb have had the same rationale, where he added a new condition to check for the OS running, i.e. sys.platform == "darwin".
That being said, I should this commit should be udpated to check for the OS as well.

[Legrandin]

Thanks for this PR.
In parallel, I have chased the root cause and I think I finally fixed with 2e7dad9, so it won't be necessary anymore to disable GMP on ARM (or on M1). The fix will end up in a new release very soon.

[regisb]

Thanks again for your work @Legrandin!