pkg/stack/unwind: Add unwind tables support for Arm64

[Sylfrena]

What?

Add support for extracting unwind table information from the ".eh-frame"
section of Arm64 binaries. So far, only X86 binaries were supported.

• Arm64 majorly differs from X86 in having a "Link Register(x30)" which saves the Return Address
• The changes are only limited to userspace to generate "compact-unwind-tables"; we do
  not send the content of the Link Register to the BPF side- this will be done in follow-ups.
• Enable DWARF-unwinding for Arm64 in "main.go" to enable building unwind-tables.
• CompactUnwindTableRow has a new field lrOffset which holds the offset for the Link Register.
  This is initiialised to 0 for X86 architectures.

Test Plan

$ go test  ./pkg/stack/unwind -count=1
ok      github.com/parca-dev/parca-agent/pkg/stack/unwind       0.007s

• tested for libc.so.6, basic-cpp-no-fp, systemd on both Arm64 and X86
  using make-test-dwarf-unwind-tables

• tables generated using "dist/eh-frame" verified against readelf -wF output

 1. Run `make build` in `parca-agent/testdata`
    `cd testdata && make build`
 2. Run `make build && make-test-dwarf-unwind-tables` from parca-agent root on the main branch.
    `cd .. && make build && make-test-dwarf-unwind-tables`
 3. Copy the output in `testdata/tables` and `testdata/compact_tables` in a separate untracked folder
    `mkdir ~/testdata-arm64-main && cp -r testdata/tables testdata/compact_tables ~/testdata-arm64-main`
 4. Switch to current `dwarf-unwinding-arm64` branch
 5. Run `make build` in parca-agent root. (Make sure you don't do this in the `testdata` submodule)
 6. Run `make-test-dwarf-unwind-tables`.
 7. Compare the output in `testdata/tables` and `testdata/compact_tables` to the results saved earlier in
    `~/testdata-arm64-main`
 8. Verify all tables against `readelf -wF "required-binary"`
 9. Repeat above for x86 binaries.

Check #1797 for Arm64 support roadmap

[Sylfrena]

This is still a WIP - some work still needs to be done for unit tests and testing a few other binaries. Apart from that, comments and a lot of println debugging need to be cleaned up as well- but they are helpful for testing at,- so please ignore them while reviewing atm.

[kakkoyun]

LGTM ❤️ 💚 💙 💛 💜
Just one readability suggestion.

I think if you could clean it up, we can merge it.

[javierhonduco]

Looking good so far! A bit more work is needed so we can get this landed, but it's getting dangerously close 😄 .

One thing to think about as a follow-up (let's make a note, perhaps an issue or add a line in the tracking issue) about DWARF expressions, not sure if arm64 emits the same 2 PLT expressions we support and definitely the registers will be off, so we need to add special cases for them. Perhaps in the meantime, we can say that we don't support expressions at all and then change this later on?

[javierhonduco]

This error is not handled anywhere, it's used more like a "default error" state, what about having that as a string directly rather than an error?

[javierhonduco]

Also now that I think about it it should be more like "architecture not supported", no?

[Sylfrena]

Good catch! I do feel we should capture an error here and just log it for easier debugging later but open to other suggestions :D