cpuid-rs

About

This is an incomplete Rust port of my CPUID tool written in C. My primary goal of this project is just to learn to write Rust and to come up with a decent library for decoding CPUID information.

The project is currently has three components:

• cpuid library/crate, with ability to import CPUID data from a text file or from CPUs on the local system
• decode binary, which prints human-readable decoding of the more interesting features exposed in CPUID
• dump binary, which can create text-based dump files of all known CPUID leaves on the local system.

Dumps can be imported with the library via cpuid::System::from_file, or with the dump/decode binaries using the -f argument.

Building

Install rust (ideally via rustup, which is what I use), and then do a cargo build in the project root and test out the target/debug/dump and target/debug/decode binaries.

Current State

The library (and dump binary) can dump all known valid CPUID leaves, even those with weird subleaf indexing (i.e. nonzero input ecx values).

The library and decode tools can decode some of the more interesting features, including:

• Feature flags from various leaves:

  • Leaf 0x0000_0001, feature identifiers
  • Leaf 0x8000_0001, extended feature identifiers
  • Leaf 0x0000_0006, thermal and power management
  • Leaf 0x0000_0007, structured extended feature identifiers
  • Leaf 0x0000_0014, Intel processor trace enumeration
  • Leaf 0x8000_0007, RAS and advanced power management information
  • Leaf 0x8000_0008, extended feature extensions ID
  • Leaf 0x8000_000A, SVM feature identifiers
  • Leaf 0x8000_001A, performance optimization identifiers
  • Leaf 0x8000_001B, instruction based sampling identifiers
  • Leaf 0xC000_0001, Centaur feature identifiers

• Cache descriptors from multiple different leaves, with preference for the most detailed sources:

  • Leaf 0x0000_0004, Deterministic Cache Parameters. Intel-specific leaf.
  • Leaf 0x0000_0018, Deterministic Address Translation Parameters. Intel-specific leaf.
  • Leaf 0x8000_001D, Cache Topology Information. AMD-specific leaf, similar in detail to the Intel-specific "deterministic" leaves above.
  • Leaf 0x8000_0005, L1 TLB and L1 cache information. AMD-specific leaf.
  • Leaf 0x8000_0006, L2 TLB and L2/L3 cache information. AMD-specific leaf.
  • Leaf 0x8000_0019, 1GB page TLB information. AMD-specific leaf.
  • Leaf 0x0000_0002, Cache and TLB information. This leaf has the lowest priority over the others, because it's the least detailed and Intel has deprecated its use in favor of the "Deterministic" leaves above. It also requires maintaining an ugly mapping of magic byte identifiers to structured information about the cache features.

• CPU topology from leaf 0x0000_000B (x2APIC)

Future Work

• Implement support for Alder Lake-like architectures which have mixed feature flags depending on the CPU index. Would need to move the feature and cache vectors from System -> Processor for this, most likely. I'm going to wait until I have an Alder Lake system to test this on, though.
• Implement more CPUID leaves in a library-friendly way. I don't want the interface to become clunky, disorganized, or filled with redundant information.
• Consider different API designs. I cobbled this together and I'm not quite happy with the API the crate currently exposes.
• Consider how to implement feature detection slightly differently -- I currently just generate a big vector of all the features it discovers, but it's not exactly searchable. The design really reflects how I use it right now, which is to just iterate over all of the features and print them out.
• Investigate and fix any bad practices I've used in my Rust code. I am a complete Rust newbie, so I have no doubt done some stupid things.
• Release a crate on crates.io at some point, once the design isn't shifting as much.
• Implement support for reading multiple different formats from CPUID dump files. For example, Todd Allen's CPUID tool, the dumps from InstLatx64, etc.
• Keep dependencies as minimal as possible. Early on I tried using a crate called clap for command-line parsing. It's a wonderful and powerful tool, but it also added like 500KB to the binary sizes. I didn't really need all the flexibility it provided, so I ended up migrating to the much smaller and less capable getopt crate.
• Implement a test/benchmark suite
• Keep the rustdoc information up to date and readable.