Support raw image (vmlinux) loading for Aarch64

[michael2012z]

Add support for Aarch64 to load vmlinux in PE format.

Reference to this Firecracker commit.

Note: In unit test, a binary file ("src/loader/test_arm64_pe.bin") is loaded. (For x86 test, it is "test_elf.bin") But as per our company's policy, I can't upload any binary file to a public repository. Please any maintainer help download the file from here and upload. Thanks.

Signed-off-by: Michael Zhao michael.zhao@arm.com

[michael2012z]

Hi, @andreeaflorescu @sameo , no reviewer was assigned automatically. Could you help review this PR?

[alxiord]

Late to the party, here goes round 1: I'm still considering whether we should have separate sources per-arch 🤔 In any case, I have a bunch of nits, including the commit message with what looks like an autogenerated change ID (can you please remove it?)

[alxiord]

@michael2012z you can cherry pick alxiord@d4b9720 for the kernel binary. The unit tests pass with this commit, however there are a bunch of build warnings on aarch64 that would need cleaning up too. Would you like to take a look at those as well?

[michael2012z]

@aghecenco Thank you very much for review and adding the binary. I will come back soon to update this long-waiting PR.

[michael2012z]

Hi, @aghecenco , the file test_pe.bin turns a HTML text file: https://github.com/aghecenco/linux-loader/blob/aarch64pe/src/loader/test_pe.bin

Could you help again?

[alxiord]

Hi, @aghecenco , the file test_pe.bin turns a HTML text file: https://github.com/aghecenco/linux-loader/blob/aarch64pe/src/loader/test_pe.bin

Could you help again?

Oops! Cherry-pick alxiord@2175ff8 instead. I updated it.

[dianpopa]

@aghecenco @michael2012z i strongly encourage you to implement the kernel loading protocols in separate sources (even folders) per architecture.
Right now the hierarchy of the crate looks like this:
root
-> mod.rs (contains kernel loading protocol for all supported archs including unit tests)
-> elf.rs (x86_64 specific)
-> struct_util.rs (x86_64 specific)
-> bootparam.rs (x86_64 specific)
-> test_elf.bin (x86_64 specific)

root
-> x86_64 -> mod.rs, elf.rs, struct_util.rs, bootparam.rs
-> aarhc64 -> mod.rs, test_arm64_pe.bin

I think the second proposal would reflect a much cleaner hierarchy, would make the code much easier to read since it gets rid of lots of cfg syntax and it is also encourages extensibility.

[michael2012z]

I renamed it to from InvalidElfMagicNumber to InvalidMagicNumber, a more generic one. Because an error message is needed when parsing ARM kernel, but defining a new InvalidPeMagicNumber looks too verbose.

[michael2012z]

@aghecenco @michael2012z i strongly encourage you to implement the kernel loading protocols in separate sources (even folders) per architecture.

This echos an earlier comment here by @aghecenco : #9 (review)

That sounds a good idea. Now ARM code has very little in common with X86 staff.

[michael2012z]

@dianpopa , @aghecenco , Thanks for reviewing the code. I handled the comments about code that have been placed so far. And errors and warnings were all cleared.

Next I am going to refactor the repo to divide source code files by architecture. I tried that a bit, looks like some effort is needed to finish. Considering now there are 2 PR's doing the same job (#16 and this), I think it's better to merge one of them at first to avoid investing more effort (to code and review) in a PR but it is finally given up.

[alxiord]

@michael2012z I compared the changes with #16 and to me the image loading part looks identical functionality-wise, but it's slightly compact in the newer PR and I expect it to be easier to rebase #26 on top of it. There's also the added device tree blob loader. Would you mind closing this in favor of #16 and reviewing it, as you're clearly more familiar with the topic?

[michael2012z]

Hi, @aghecenco , no problem, I will close it. Sure, I will continue working in this crate.
And thank you very much for reviewing this PR.