Stream: git-wasmtime

Topic: wasmtime / issue #5550 aarch64: Support GOT Relative relo...

view this post on Zulip Wasmtime GitHub notifications bot (Jan 09 2023 at 20:14):

acw commented on issue #5550:

Interesting. Looks like it runs and generates output of the sort I'd expect:

% cargo run
...
% objdump -dr output.o

output.o:       file format mach-o arm64

Disassembly of section __TEXT,__text:

0000000000000020 <_gogogo>:
      20: fd 7b bf a9   stp     x29, x30, [sp, #-16]!
      24: fd 03 00 91   mov     x29, sp
      28: 00 00 00 90   adrp    x0, 0x0 <_gogogo+0x8>
                0000000000000028:  ARM64_RELOC_GOT_LOAD_PAGE21  _variable-name-x
      2c: 00 00 40 f9   ldr     x0, [x0]
                000000000000002c:  ARM64_RELOC_GOT_LOAD_PAGEOFF12       _variable-name-x
      30: 81 00 80 d2   mov     x1, #4
      34: 03 00 00 90   adrp    x3, 0x0 <_gogogo+0x14>
                0000000000000034:  ARM64_RELOC_GOT_LOAD_PAGE21  _print
      38: 63 00 40 f9   ldr     x3, [x3]
                0000000000000038:  ARM64_RELOC_GOT_LOAD_PAGEOFF12       _print
      3c: 60 00 3f d6   blr     x3
      40: fd 7b c1 a8   ldp     x29, x30, [sp], #16
      44: c0 03 5f d6   ret

But that particular relocation type isn't supported on arm64:

% gcc -o test rts.c output.o
ld: in section __TEXT,__text reloc 1: pcrel and ARM64_RELOC_GOT_LOAD_PAGEOFF12 not supported file 'output.o' for architecture arm64
clang: error: linker command failed with exit code 1 (use -v to see invocation)

Although, I did earlier discover that my Xcode suite is a version behind, so I'm going to save this comment for reference, and then see if this magically gets better after that.

view this post on Zulip Wasmtime GitHub notifications bot (Jan 09 2023 at 21:04):

acw commented on issue #5550:

Although, I did earlier discover that my Xcode suite is a version behind, so I'm going to save this comment for reference, and then see if this magically gets better after that.

Sad to say: that wasn't it. Same error, after making sure everything was up-to-date.

view this post on Zulip Wasmtime GitHub notifications bot (Jan 09 2023 at 21:28):

afonso360 commented on issue #5550:

I think I got it, the Page Offset relocation is marked as being PC relative when writing the output file. And I think that was causing some confusion in the linker. Can you try with this latest commit?

view this post on Zulip Wasmtime GitHub notifications bot (Jan 09 2023 at 21:31):

afonso360 edited a comment on issue #5550:

I think I got it, the Page Offset relocation was marked as being PC relative when writing the output file. And I think that was causing some confusion in the linker. Can you try with this latest commit?

view this post on Zulip Wasmtime GitHub notifications bot (Jan 09 2023 at 21:34):

acw commented on issue #5550:

It works!

% cargo run
    Finished dev [unoptimized + debuginfo] target(s) in 0.03s
     Running `target/debug/macos-example`
% objdump -dr output.o

output.o:       file format mach-o arm64

Disassembly of section __TEXT,__text:

0000000000000020 <_gogogo>:
      20: fd 7b bf a9   stp     x29, x30, [sp, #-16]!
      24: fd 03 00 91   mov     x29, sp
      28: 00 00 00 90   adrp    x0, 0x0 <_gogogo+0x8>
                0000000000000028:  ARM64_RELOC_GOT_LOAD_PAGE21  _variable-name-x
      2c: 00 00 40 f9   ldr     x0, [x0]
                000000000000002c:  ARM64_RELOC_GOT_LOAD_PAGEOFF12       _variable-name-x
      30: 81 00 80 d2   mov     x1, #4
      34: 03 00 00 90   adrp    x3, 0x0 <_gogogo+0x14>
                0000000000000034:  ARM64_RELOC_GOT_LOAD_PAGE21  _print
      38: 63 00 40 f9   ldr     x3, [x3]
                0000000000000038:  ARM64_RELOC_GOT_LOAD_PAGEOFF12       _print
      3c: 60 00 3f d6   blr     x3
      40: fd 7b c1 a8   ldp     x29, x30, [sp], #16
      44: c0 03 5f d6   ret
% gcc -fPIC -g3 -o test rts.c output.o
% ./test
x = 4

Thank you!

view this post on Zulip Wasmtime GitHub notifications bot (Jan 23 2023 at 22:58):

cfallin commented on issue #5550:

I worked through the audit backlog and pushed a commit to this branch with audit results. Someone else will need to review the PR since it has a commit from me now.

view this post on Zulip Wasmtime GitHub notifications bot (Jan 26 2023 at 11:46):

afonso360 commented on issue #5550:

Thanks for looking into this! I had to rebase since we had a conflict in Cargo.lock but should be ready for review now.

view this post on Zulip Wasmtime GitHub notifications bot (Jan 26 2023 at 12:19):

afonso360 edited a comment on issue #5550:

Thanks for looking into this! I had to rebase since we had a conflict in Cargo.lock but should be ready for review now.

Edit: It looks like we have a similar situation to #5619 and its failing in CI due to duplicate deps.

I can de-duplicate dependencies the same way I did in #5619 but I'm not sure I can de-duplicate hashbrown without updating a bunch of other stuff.

view this post on Zulip Wasmtime GitHub notifications bot (Jan 26 2023 at 12:48):

afonso360 edited a comment on issue #5550:

Thanks for looking into this! I had to rebase since we had a conflict in Cargo.lock but should be ready for review now.

Edit: It looks like we have a similar situation to #5619 and its failing in CI due to duplicate deps.

I can de-duplicate dependencies the same way I did in #5619 but I'm not sure I can de-duplicate hashbrown without updating a bunch of other stuff.

Edit2: I can't de-duplicate hashbrown, version 0.12 is depended by indexmap and even the latest published version 1.9.2 still depends on hashbrown 0.12. Not entirely sure what to do now.

view this post on Zulip Wasmtime GitHub notifications bot (Jan 26 2023 at 18:58):

jameysharp commented on issue #5550:

If I'm reading the cargo-deny documentation correctly, it looks like we should be able to add something like this to deny.toml:

[[bans.skip]]
name = "hashbrown"
version = "=0.12.3"

Unlike cargo-vet exceptions, I think it's reasonable to add this kind of targeted exception to the "no duplicate dependencies" rule.

view this post on Zulip Wasmtime GitHub notifications bot (Jan 26 2023 at 19:48):

afonso360 commented on issue #5550:

Yeah, that seemed to have worked locally. I've applied the same updates that I did in the other PR (except object which is still at 0.30.1 instead of .3).

view this post on Zulip Wasmtime GitHub notifications bot (Feb 05 2023 at 10:32):

afonso360 commented on issue #5550:

@cfallin would it be possible to vet the remaining dependencies? Once i updated our version of gimli it started saying that we need a safe to deploy validation, but weirdly it didn't before when it was just a transitive dependency of object 0.30.

view this post on Zulip Wasmtime GitHub notifications bot (Feb 06 2023 at 10:51):

afonso360 edited a comment on issue #5550:

@cfallin would it be possible to vet the remaining dependencies? Once i updated our version of gimli it started saying that we need a safe to deploy validation, but weirdly it didn't before when it was just a transitive dependency of object 0.30.

Edit: It looks like some of these might get vetted via #5513

view this post on Zulip Wasmtime GitHub notifications bot (Feb 08 2023 at 23:57):

cfallin commented on issue #5550:

@afonso360 would you be willing to rebase? I think this is good to go otherwise!

view this post on Zulip Wasmtime GitHub notifications bot (Feb 15 2023 at 12:14):

afonso360 commented on issue #5550:

Rebased on top of main, and removed the hashbrown exception. #5513 includes sort of the same exception, but ignores the tree of indexmap which solves the same issue I was having.

Last updated: Dec 23 2024 at 12:05 UTC