processor_rng: Add support for ARMv8.5-A FEAT_RNG extension

[mikatammi]

Add support for ARMv8.5-A FEAT_RNG extension into processor_rng.
Add ISA extension for the RNG extension, and also new flag into cpuid_arm64.

I'm wondering is this enough or should I also do something else for this PR? (I was thinking about maybe adding those instruction probes into cpuid_aarch64?) I have tested this only on aarch64-linux. On aarch64-darwin I tried to force the RNG extension to be used and I got an illegal instruction when running on Macbook Pro with M3 Max.

[Copilot]

Pull request overview

This PR adds support for the ARMv8.5-A FEAT_RNG extension to the processor_rng module, enabling hardware random number generation on ARM64 platforms that support this feature. The implementation adds CPU feature detection via Linux hwcap2, integrates the RNDR instruction into the RNG implementation, and configures the build system with appropriate compiler flags.

Changes:

• Added RNG CPU feature flag and detection for ARM64 via hwcap2
• Implemented RNDR instruction support in processor_rng with both inline assembly and intrinsic paths
• Updated build system configuration to enable ARMv8.5-a+rng compiler flags for GCC and Clang

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
src/lib/utils/cpuid/cpuid_aarch64/cpuid_features.h	Adds RNG bit flag to CPUFeature enum
src/lib/utils/cpuid/cpuid_aarch64/cpuid_features.cpp	Adds string mapping for RNG feature ("armv8rng")
src/lib/utils/cpuid/cpuid_aarch64/cpuid_aarch64.cpp	Implements RNG detection via hwcap2 auxval on Linux
src/lib/rng/processor_rng/processor_rng.cpp	Implements RNDR instruction for ARM64 with success validation
src/lib/rng/processor_rng/info.txt	Adds arm64 architecture support and armv8rng ISA requirement
src/build-data/cc/gcc.txt	Adds GCC compiler flag for armv8rng ISA extension
src/build-data/cc/clang.txt	Adds Clang compiler flag for armv8rng ISA extension
src/build-data/arch/arm64.txt	Registers armv8rng as an ISA extension for arm64

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[randombit]

The CI failure looks relevant. I am not sure exactly why it's happening but it seems like the compiler is rejecting even inline asm that does not have the relevant CPU features enabled via either the command line or function attribute [*]. I think the fix is something like #5229

[*] You are correctly setting the command line flag but in the amalgamation mode that doesn't appear.

I was thinking about maybe adding those instruction probes into cpuid_aarch64?

I would say do not, unless you have personally a specific need for this. Despite being the person who wrote that SIGILL-based instruction probing code, I am not at all enthused about its existence, and inclined to the take that if the operating system just cannot be bothered to expose the relevant APIs we need to detect CPU features (getauxval, sysctlbyname, whatever Windows API that we don't currently support, ...) then it probably deserves to be slow.

[mikatammi]

Okay good to know, I was wondering why the probing was done like that. I'm thinking the proper way would probably be to read the ID_AA64ISAR0_EL1, AArch64 Instruction Set Attribute Register 0, and check the values for existence of different ISA extensions, instead of trying to "brute-force" and probe through all of them.

The ARM architecture reference manual is available as a PDF here: https://developer.arm.com/documentation/ddi0487/latest

[coveralls]

coverage: 90.433% (-0.001%) from 90.434%
when pulling c7b6852 on mikatammi:arm64_rng
into b941fcb on randombit:master.

[randombit]

read the ID_AA64ISAR0_EL1

Isn't this only readable in kernel mode though?

[mikatammi]

read the ID_AA64ISAR0_EL1

Isn't this only readable in kernel mode though?

It is readable at least in Linux userspace, I tested with: asm volatile("mrs %0, id_aa64isar0_el1" : "=r"(output) : "0"(output) : "cc");

[mikatammi]

I rebased this on top of #5229 for now

[mikatammi]

Also dropped the commit which had the legacy module-level armv8rng extension specifications for arm64. If you want I can add those back, I'm not sure is this project trying to migrate away from those?

[mikatammi]

@randombit Could you please trigger the CI tests again?