add -fIBT switch and emit ENDBR for function and thunk prologs

[WalterBright]

See https://news.ycombinator.com/item?id=36722823 for more information on IBT.

For the ENDBR32/64 instructions:

https://www.felixcloutier.com/x86/endbr32
https://www.felixcloutier.com/x86/endbr64

I still need to add support for switch jump tables.

I have no way to test this, as my machines don't use IBT. All I can say is disassembly shows the instruction is generated, and the programs still run.

[dlang-bot]

Thanks for your pull request, @WalterBright!

Bugzilla references

Your PR doesn't reference any Bugzilla issue.

If your PR contains non-trivial changes, please reference a Bugzilla issue or create a manual changelog.

Testing this PR locally

If you don't have a local development environment setup, you can use Digger to test this PR:

dub run digger -- build "master + dmd#15415"

[WalterBright]

https://issues.dlang.org/show_bug.cgi?id=20933

[thewilsonator]

is this the same or similar to ldc-developers/ldc#4437 ?
I feel we should try to have a consistent interface for this stuff.

cc @ibuclaw @JohanEngelen

[ErnyTech]

is this the same or similar to ldc-developers/ldc#4437 ? I feel we should try to have a consistent interface for this stuff.

cc @ibuclaw @JohanEngelen

This is the same as -fcf-protection=branch in GDC and LDC (after the PR).
Note that the current PR for DMD does not implement all Intel CET but only IBT (-fcf-protection=branch) which is sufficient for OpenBSD but is not a complete implementation of CET as it lacks the shadow stack (-fcf-protection=return).

It would be good to adopt the -fcf-protection flags in DMD as well

[ibuclaw]

Wouldn't this be a -gxxx option to be consistent with other dmd options?

@ErnyTech I'd expect dub to be a common front for all command-line differences.

[JohanEngelen]

In the LDC PR I have also implemented __traits(getTargetInfo, "CET") == 0 normally == 1 when IBT is turned on.

[dkorpel]

I'd expect dub to be a common front for all command-line differences.

It's still convenient to have consistent naming of flags across compilers, and there's no advantage of naming it differently.

[WalterBright]

-fcf-protection=branch

It's just too long and clunky and un-memorable.

[WalterBright]

I fixed switch code generation so it'll work with IBT.

[ErnyTech]

-fcf-protection=branch

It's just too long and clunky and un-memorable.

Maybe, but that's what the market has chosen and what all programmers expect, GCC and Clang use -fcf-protection=branch, Rustc use cf-protection=branch

Ok to have a different flag to keep consistency with the other flags, but I don't think the memory issues are applicable here as anyone who needs Intel CET IBT already knows -fcf-protection=branch

[ibuclaw]

I'd expect dub to be a common front for all command-line differences.

It's still convenient to have consistent naming of flags across compilers, and there's no advantage of naming it differently.

I'd rather just let dmd be consistent with itself, as it has no incentive to copy from others in a really inconsistent way. (ie: the gcc-style prefixing letters -[fmW] refer to specific parts of the compiler which dmd doesn't really have the same separation).

[ibara]

Hi Walter --

Thanks for this. It works for me.

[WalterBright]

@ibara wonderful!