All the work done by Brad except for some fixes from base by mortimer@
and patrick@
Tests: amd64 Brad, jsg@ and me, i386: Brad, mips64: visa@, sparc64 me.
No tests on powerpc but a test build with only the PowerPC backend
packages on amd64.
- don't build clang itself with -fno-ret-protector and -mno-retpoline,
to regain performance
- use -mretpoline by default on amd64
- fix printf %b length specifiers
from Brad
On non-clang archs, ports-clang uses the libestdc++ backend from
ports-gcc, so it needs the headers in the g++ package at runtime.
Testing CHOSEN_COMPILER == "ports-gcc" is nice semantically but since it
happens after including bsd.port.mk, it's too late to change TEST/RUN_DEPENDS.
Use a different test that may not be optimal but works in practice, so that
the dep is correctly registered.
This should fix random build failures due to dpb junking in bulk builds
on non-clang archs.
Problem pointed out by Goerge Koehler, ok sthen@
Using long calls means slower code. It is not useful right now because
the size of the llvm executables has been reduced, as shown by tests by
Charlene Wendling. And it breaks compilation with gcc 6, as pointed out
by a report from George Koehler.
The goal is to provide quick feedback on proposed diffs, and enforce
testing on at least one gcc arch (probably sparc64) before diffs get
committed. All contributions are still welcome.
Maintainer timeout, ok ajacoutot@ visa@ naddy@ sthen@ espie@
build with ports-gcc which isn't yet in shape to build LLDB.
tests and tweaks from various people including jca@, landry@,
Charlene Wendling, and probably others -- ok jca@
- [mips] Prevent PIC to be set to level 2
- [mips] Improve handling of -fno-[pic/PIC] option
- [PowerPC] Set the default PLT mode on OpenBSD/powerpc to Secure PLT.
- Remove PowerPC -nopie workaround
- Add workaround to PowerPC to only build the PowerPC backend as a means
of reducing the code size
- Enable building on PowerPC
from Brad (maintainer)
This allows building webkitgtk4 on OpenBSD/powerpc
- Accept both `--foo bar` and `--foo=bar` styles options.
- Add support for -z initfirst.
- Generate __data_start symbol that marks the start of .data when __data_start
is referenced from code being linked.
From Brad (maintainer)
When merging sections into the output, lld tries to adjust the alignment of
the section to be at least as large as the entry size of the section.
This causes a later check that validates the alignment to fail if the
entry size isn't a power of two. This happens when building some of the
java support code in ports gcc. Fix this by sticking to the original
alignment if the entry size isn't a power of two.
from Brad (maintainer)
Allow preemption of functions with protected visibility. Disallowing this
makes no sense. Yes it breaks function address equality and therefore
the expectations of the standard C language. However declaring symbols
with protected visibility isn't standard C in the first place.
from Brad (maintainer)
dependent ports, this restricts deps to use the current clang version
which is required for some ports using the unversioned libLLVM*.so
library. see https://marc.info/?l=openbsd-ports&m=152377935312657&w=2
- set RUN_DEPENDS on devel/llvm in lang/crystal as it uses libLLVM*.so
breakage reported by James Turner
I don't like this at all, but don't see what other option we have,
if anyone has a better idea please pipe up.
- Add ret protector options as no-ops.
- Add a clang pass that identifies potential ROP gadgets and replaces ROP
friendly instructions with safe alternatives. This initial commit fixes
this framework.
- Add RETGUARD to clang for amd64. This security mechanism uses per-function
random cookies to protect access to function return instructions, with the
effect that the integrity of the return address is protected, and function
return instructions are harder to use in ROP gadgets.
- Put the new retguard symbols in their own section,
'.openbsd.randomdata.retguard', to make them easier to work with in the
kernel hibernate code.
- Pass -nopie to the linker when -pg is specified to make the
profiler(gprof) work properly.
- Work around a bug where discarding the .ARM.exidx section in the armv7 kernel
linker script makes ld.lld(1) crash. This has been fixed in a different
(proper?) way upstream but backporting their fix is a bit too invasive.
- Merge '.openbsd.randomdata.*' sections into a single '.openbsd.randomdata'
section when linking, as we do when using ld from binutils.
from Brad (maintainer)
- [Sparc] Fix addressing mode when using 64-bit values in inline assembly
- [Sparc] Use synthetic instruction clr to zero register instead of sethi
from Brad (maintainer)
This seems to be the way almost all Linux distributions and BSDs ship
LLVM and is what Mesa expects. There are concerns this may cause issues
if ports start linking it, hopefully this won't happen as everything
links libLLVM-X.Y.so where X.Y is the branch version which will have the
same ABI throughout X.Y.Z releases. The rust port explicitly links
statically even if shared libLLVM is available.
Patch cmake files in a similiar way to a patch in NetBSD pkgsrc to
correct symbols not getting added to the shared library on OpenBSD.
Use the documented cmake var for RTTI while here.
ok brad (MAINTAINER) sthen@ bcallah@
Brings us in line with clang in base, and should fix the builds errors
due to membar instructions used in libestdc++ headers. ok kettenis@,
maintainer timeout.
