share/mk: Fix a heuristic in bsd.cpu.mk
ClosedPublic
Actions

Authored by markj on Sep 2 2025, 3:40 PM.

Details

Reviewers

brooks
jrtc27
khorben
ngie
emaste

Commits

rGe338aab61374: share/mk: Fix a heuristic in bsd.cpu.mk
rGa24fa22bf03c: share/mk: Fix a heuristic in bsd.cpu.mk
rG4341f92251ae: share/mk: Fix a heuristic in bsd.cpu.mk

Summary

When cross-building from MacOS we have MACHINE=arm64 MACHINE_ARCH=arm,
so bsd.cpu.mk infers that sizeof(long) == 4, but of course it isn't.
This breaks the bootstrap build of openssl, which tests
MACHINE_ABI:Mlong64 to decide whether to build ecp_nistp224.c. It
doesn't, and crypto/openssl/freebsd/include/openssl/configuration.h
undefines OPENSSL_NO_EC_NISTP_64_GCC_128, so we end up with a link
error.

Diff Detail

Repository

rG FreeBSD src repository

Lint

Lint Not Applicable

Unit

Tests Not Applicable

Event Timeline

markj created this revision.Sep 2 2025, 3:40 PM

Herald added subscribers: emaste, imp, bdrewery. · View Herald TranscriptSep 2 2025, 3:40 PM

markj requested review of this revision.Sep 2 2025, 3:40 PM

Harbormaster completed remote builds in B66761: Diff 161376.Sep 2 2025, 3:40 PM

jrtc27 added inline comments.Sep 2 2025, 3:51 PM

share/mk/bsd.cpu.mk
395
397	Should probably limit this to BOOTSTRAPPING? It doesn't really matter, but I don't like having OS-dependent code that isn't guarded by BOOTSTRAPPING.

imp added inline comments.Sep 2 2025, 3:55 PM

share/mk/bsd.cpu.mk

397

${MACHINE} == "arm64"

covers it too. No reason to have Darwin be special here.

markj marked an inline comment as done.Sep 2 2025, 3:55 PM

markj added inline comments.

share/mk/bsd.cpu.mk
397	That's fine for my immediate purposes but isn't it logically wrong? MACHINE/MACHINE_ARCH describe host architecture, so should be consistent across the whole build. Or is my my mental model wrong?

Fix typos

Harbormaster completed remote builds in B66763: Diff 161378.Sep 2 2025, 3:58 PM

markj added inline comments.Sep 2 2025, 3:59 PM

share/mk/bsd.cpu.mk
397	Well, in principle arm64/aarch32 is a valid ABI, no?

jrtc27 added inline comments.Sep 2 2025, 4:15 PM

share/mk/bsd.cpu.mk
397	MACHINE/MACHINE_ARCH describes the architecture that you are currently generating binaries for. Initially that's the build machine, but during various buildworld/kernel steps MACHINE=TARGET MACHINE=TARGET_ARCH is done to switch what MACHINE* mean. If BOOTSTRAPPING is set that means MACHINE corresponds to the machine you're building on, because that switch has not yet happened. If BOOTSTRAPPING is not set that means MACHiNE corresponds to what was TARGET at the top level. But .MAKE.OS is always the build machine's OS.

imp added inline comments.Sep 2 2025, 4:20 PM

share/mk/bsd.cpu.mk
397	No. No more valid than big endian riscv :) But I'm now going to agree with @jrtc27. It's a pedantic point, but one worth remembering. We're only doing this because this file is now used during bootstrapping AND apple is an odd duck out.

Check whether BOOTSTRAPPING is defined

Harbormaster completed remote builds in B66764: Diff 161379.Sep 2 2025, 4:29 PM

I kind of wonder if it makes sense to provide multiple host OS share/mk/bsd.cpu.mk files (or something similar) to avoid cluttering this file with support for a number of other OS hacks.
@sjg : do you have any thoughts on doing something like this? Has Juniper solved a similar problem in the past?