This patch started out as an attempt to run linker scripts through the
preprocessor. However, since that required some more infrastructure
changes, the build system is so intertwined, and there are so many other
small issues that turned up and are easier to fix (and get running, and
test thoroughly) in a single go, it turned out a little bigger. In order
of appearance, it:
- wraps direct linker invocations in a macro to avoid the widespread
ifeq($(CONFIG_COMPILER_LLVM_CLANG),y) duplication.
- introduces an $(generic-deps) (equivalent to $(<class>-deps)) variable
for targets that all files depend on
- makes the $(src-to-obj) function usable in multiple places as the
authoritative way to get an output file name (even if the respective
source file is also under build/), and makes it preserve extensions on
everything except %.c and %.S (e.g. %.ld and %.asl)
- replaces the old $(<class>-postprocess) with a single
$(postprocessors) variable. The old ramstage-postprocess was weird
because it contained unescaped $(eval ...)s, meaning it gets executed
as soon as the variable is first substituted (and the other
$(eval ...) in the toplevel Makefile does essentially nothing). The
new mechanism is just $(eval ...)ed directly after the recursive parse
with no further magic. Files can freely append their own (escaped)
content to it and access variables valid in the calling context (like
$(<class>-objs)) directly.
- enhances the existing $(<class>-<type>-ccopts) mechanism with
$(<class>-generic-ccopts) and $(generic-<type>-ccopts) to reduce
duplication.
- makes .ld a type that can be added like a normal class file, causing
it to be preprocessed with the correct #defines for the current class
(needed for a follow-up feature). Migrates all linker scripts to this
mechanism, which allows us to get rid of the weird $(ldoptions)
mechanism (Kconfigs are replaced by preprocessor and no longer need to
be defined as symbols).
- removes duplicate $(INCLUDES) from $(CFLAGS)
- repairs the crazy state of MIPS Makefiles, which seem to have been
copied together from X86 despite having absolutely nothing in common
with that architecture. They were using the same code to paste
assembly pieces and linker scripts together without really needing it
for anything, and even accidentally relied on a Kconfig default set
in the arch/x86 subdirectory (I wish I was kidding). Changed them to
work equivalent to the arm/arm64 Makefiles which are far closer
related (also being SRAM-based platforms).
- moves the x86-specifc $(OPTION_TABLES_H) into the x86 Makefile.inc and
fixes an rule that would've had an empty target if it wasn't defined
- removes the custom ldscript-gathering variables for x86 bootblock and
romstage. The Makefile simply combines all .ld files that have been
added to the respective class now.
- uses the normal class build system to replace some of the custom rules
for autogenerated bootblock/crt0 files on x86, and removes some
hardcoded flags by using the normal $(...-ccopts) variables.
- moves the handling of .asl files from the global Makefile.inc to x86.
Changed to reuse the generic template for the preprocessing and C
compilation steps.
- removed the extra <name>.o linking step before linking an rmodule for
modules that don't require special linker flags (most of them).
- removes the incorrect assumption that there was a global $(LDFLAGS)
from the SMM Makefile.inc (it was named $(LDFLAGFS in one place so it
couldn't have been all that important ;) ).
- allow -j flag for parallel builds to be properly passed through to
vboot child invocation by using special $(MAKE) variable.
BUG=None
TEST=Built for Falco, Nyan_Blaze, Stout, Rush_Ryu and Urara. Binary
diffed old and new coreboot.rom (and some manually uncompressed stages),
confirmed that images/stages are byte-for-byte identical except for some
embedded timestamps and commit hashes. (Addresses in Falco/Stout
ramstages shifting slightly due to different link order for ASL object
files within their directory. Some addresses in Urara ramstage .rodata
and some relocation entries in rmodules moved around due to linking them
in fewer steps.)
Change-Id: I50af7dacf616e0f8ff4c43f4acc679089ad7022b
Signed-off-by: Julius Werner <jwerner@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/219170
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
af6852b98f