STACKALIGN: Reimplement in terms of __align_down
AcceptedPublic
Actions

Authored by jhb on Tue, Jan 27, 10:29 PM.

Details

Reviewers

Summary

This changes STACKALIGN to be type-preserving when operating on
pointers.

Repository

rG FreeBSD src repository

Lint

Lint Skipped

Unit

Tests Skipped

Build Status

Buildable 70173
Build 67056: arc lint + arc unit

jhb requested review of this revision.Tue, Jan 27, 10:29 PM

sys/arm/include/param.h
50	Why STACKALIGN needs to be MD then? It is enough to have STACKALIGNBYTES defined in machine/param.h. And please add a comment that it is type-preserving.

Effort: CHERI upstreaming or whatever it is seems like it would apply?

brooks added inline comments.

sys/arm/include/param.h
50	It's only defined on arm/riscv. STACKALIGNBYTES should surely be STACKALIGN_MASK or the like, but that's a different change.

This revision is now accepted and ready to land.Wed, Jan 28, 10:12 AM

sys/arm/include/param.h
50	It's only defined on arm/riscv. Yes, this is why I said that it is enough to have STACKALIGNBYTES in machine/param.h. sys/param.h could then do #ifdef STACKALIGNBYTES #define STACKALIGN ... #endif and it would be immediately have uses at least on amd64 where we align stack after the signal frame. Might be for i386 too, but I do not remember.

sys/arm/include/param.h
50	I would be happy to move it. I do think it would be nicer on other arches as well. I do wish the STACKALIGNBYTES wasn't the biased value. I'm not sure if we expose this to userland though. :( Let me see if it's feasible to change that to the unbiased value.