Use new assembly syntax

[ghost]

This pull request updates x86_64 to use the new asm! macro. This will break older nightlies, but makes the syntax easier to read.

Fixes #164

[ghost]

This PR should be ready to merge. I have tested it with Blog OS and it works fine, but this does not mean that it will always work

We really need to write a full test suite for this crate. Also can I please be added to the x86_64 team phil?

cc @phil-opp

[phil-opp]

Thanks a lot for the pull request! I have some minor comments, but overall this looks good to me.

[phil-opp]

We really need to write a full test suite for this crate.

Absolutely! This would be very useful for changes like this.

Also can I please be added to the x86_64 team phil?

Done!

[ghost]

All of those have been fixed, feel free to merge when you want

[phil-opp]

Sorry for the late reply! The code looks good to me overall, but the double_fault_stack_overflow test seems to fail on CI:

Running: qemu-system-x86_64 -drive format=raw,file=/home/runner/work/x86_64/x86_64/testing/target/x86_64-bare-metal/debug/deps/bootimage-double_fault_stack_overflow-0254781a0bba0dab.bin -device isa-debug-exit,iobase=0xf4,iosize=0x04 -serial stdio -display none
qemu-system-x86_64: warning: TCG doesn't support requested feature: CPUID.01H:ECX.vmx [bit 5]
qemu-system-x86_64: Trying to execute code outside RAM or ROM at 0x0000000108234489

I'm not sure what's the cause of this, but it seems like something goes wrong with the IDT or TSS loading.

[ghost]

I believe that's caused by the code that reloads cs, but I can't identify why. I'll do some testing today and then merge this when CI passes

[phil-opp]

@rybot666 Any updates on this?

[ghost]

Sorry I totally forgot about this! I'll take another look

[phil-opp]

No worries, thanks!

[ghost]

For some reason cargo xtest doesn't run on my machine (it can't find rust-src despite me having rust-src installed on nightly)

[phil-opp]

For some reason cargo xtest doesn't run on my machine (it can't find rust-src despite me having rust-src installed on nightly)

There was a breaking change in the Rust repository lately (change of the directory layout), so you might need to update your nightly version and/or cargo-xbuild. See the note from the cargo-xbuild readme:

Note: The latest version of cargo-xbuild supports all nightlies after 2020-07-30. If you are on an older nightly, you need to install version 0.5.35: cargo install cargo-xbuild --version 0.5.35.

If you're on a newer nightly, you need cargo-xbuild version 0.6.0.

[josephlr]

I think this might be why we were seeing test failures. We can just directly use the existing implementation with the new asm! syntax:

asm!(
    "push {cs}",
    "lea {tmp}, [1f + rip]",
    "push {tmp}",
    "retfq",
    "1:",
    cs = in(reg) u64::from(sel),
    tmp = lateout(reg) _,
    options(nomem, preserves_flags),
);

I compiled this locally, and got the same assembly output as the existing implementation.

[josephlr]

@phil-opp @rybot666 I rebased this, implemented the missing commands, and fixed the set_cs issue.

PTAL

[phil-opp]

The updates look reasonable to me, but I didn't look at the set_cs assembly in detail. The fact that the tests pass again is a good sign though.

[josephlr]

The updates look reasonable to me, but I didn't look at the set_cs assembly in detail. The fact that the tests pass again is a good sign though.

I disassembled the double_fault_stack_overflow test, and the code that sets cs is identical before and after this change:

movzx  ecx,WORD PTR [rax+0x48]
push   rcx
lea    rax,[rip+0x3]
push   rax
rex.W  retf 
mov    rdi,rbx

So this PR (at least) didn't mess things up.

[phil-opp]

Great, thanks a lot for checking!