Kernel mode code on x86_64 needs to be built with -mno-red-zone as
interrupts would corrupt the red zone if it were in use. However, the
kernel is linked with libsupc++, which was not compiled with
-mno-red-zone. If an interrupt occurred in libsupc++ code the red zone
would get corrupted. This was causing random panics, particularly under
heavy system load. Therefore, on x86_64 a separate build of libsupc++
with -mno-red-zone is now done for the kernel to use. Note: this commit
will require a rerun of configure and rebuild of cross tools.
Added the necessary build flags for modules, and added a module (dpc)
to the floppy image for x86_64 builds for testing purposes. The module
gets loaded correctly and its code runs without issue. Only non-trivial
addition is the different method for generating kernel.so, this is
explained in the kernel Jamfile.
* Use gcc and g++ rather than cc and c++, as the latter now point to
clang with recent Xcode versions and compilation of the host tools
fail for various reasons with it.
* Replace the case-sensitive filesystem check with a more basic one,
as diskutil no longer supports the behaviour of getting info for the
volume that any path is on.
* Updated ReadMe with a correct list of prerequisites for OS X.
* GCC 2 builds are still broken due to a strange error that only
occurs with a GCC 2 built on OS X 10.7
Will be merged with the x86 one later on. Requires -fno-omit-frame-pointer on
the kernel build flags, GCC defaults to not generating stack frames on x86_64.
Since x86 and x86_64 share a lot of common code, x86_64 kernel sources/headers
are going to reside under headers/private/kernel/arch/x86 and
src/system/kernel/arch/x86 along with the existing x86 code. This commit
changes the build system to handle this. A new variable, TARGET_KERNEL_ARCH,
has been added. This is the name of the kernel/boot architecture directory
name, set to x86 on both x86 and x86_64. This is now used in all places where
TARGET_ARCH was used to get to kernel arch sources/headers (I've changed
everything necessary as far as I can tell). Kernel won't build for x86_64
at the moment as the sources have not been merged, loader does.
* platform_allocate_elf_region() is removed, it is implemented in platform-
independent code now (ELF*Class::AllocateRegion). For ELF64 it is now
assumed that 64-bit addresses are mapped in the loader's 32-bit address space
as (address - KERNEL_BASE_64BIT + KERNEL_BASE).
* mapped_delta field from preloaded_*_image removed, now handled compile-time
using the ELF*Class::Map method.
* Also link the kernel with -z max-page-size=0x1000, removes the need for
2MB alignment on the data segment (not going to map the kernel with large
pages for the time being).
The red zone is a 128-byte area below the stack pointer specified by the
AMD64 ABI that can be used by leaf functions for their stack frame without
modifying the stack pointer. It is guaranteed not to be modified by signal
handlers. This cannot be used in kernel mode code, as an interrupt handler
could overwrite it, so stop GCC from generating code that uses it.
* x86_64 is using the existing *_ia32 boot platforms.
* Special flags are required when compiling the loader to get GCC to compile
32-bit code. This adds a new set of rules for compiling boot code rather
than using the kernel rules, which compile using the necessary flags.
* Some x86_64 private headers have been stubbed by #include'ing the x86
versions. These will be replaced later.
* while it seemed to work before, BuildSetup was actually adding includes from the default boot platform, before BoardSetup had a chance of changing it.
* Make the locale kit a part of libbe.
* Drop the LocaleBackend kludge used from within libbe (and from
other places, too) in order to access system catalog strings.
This is now done via gSystemCatalog, which is provided and initialized
by libbe.
* Drop all references to liblocale.so from all Jamfiles.
* Add legacy symlink liblocale.so in order to keep optional packages
that rely on it in a working state.
TODO: the documentation hasn't been updated.
* uncomment the building of libroot_build.a again
* add function remapper to HOST_STATIC_LIBROOT
* drop TODO about the function remapper not working with the static
libroot
Ingo: please review - I think this should work, but I'm not so sure
where HOST_STATIC_LIBROOT should be in the list of libraries of its
only user (<build>bfs_fuse): where it is now or right at the end?
As it is now, the resulting binary still contains references to
host-libc-implementations of close() & others, which are triggered by
the other libs (like libfuse.so). If I put HOST_STATIC_LIBROOT right at
the end, those references are gone, though. But which is correct?
This makes opening symlinks work universally in the build system tools.
Two mechanisms have been implemented, both of which don't always work.
The first is remapping via preprocessor macros. This fails where equally
named methods are used (e.g. STL fstream::open()). The other is using
hidden functions in the new libroot_build_function_remapper.a that is
linked into everything that is linked against libroot_build.so. This one
fails for functions that are defined inline in headers (Linux/glibc does
that). Together they seem to cover our build system needs ATM.
As PulkoMandy pointed out on IRC, darwin10 and 11 (10.6 and 10.7) are at least partially 64bit, so
the test only applies there. When darwin12 comes out it'll have to be fixed.
* Resolve TODO: HOST_GCC_BASE_FLAGS should not be included in
HOST_LDFLAGS. Enable adding "-fno-strict-aliasing -fno-tree-vrp"
accordingly.
* Fix handling of HOST_PLATFORM_IS_64_BIT and HAIKU_HOST_USE_32BIT: The
former does now state whether the platform is effectively treated as
64 bit platform, i.e. it actually is 64 bit and the 32 bit mode is
not enforced. HAIKU_HOST_USE_32BIT is now only set when the platform
is actually 64 bit, but 32 bit mode is enforced.
* Map build variables HOST_CPU and HOST_ARCH to x86_64, if it they are
* x86 and
64 bit and define the __x86_64__ C macro instead of __INTEL__ in that
case.
* <OS.h>: Also handle __x86_64__.
* ensure host isn't darwin as Apple doesn't use binutils
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@43202 a95241bf-73f2-0310-859d-f6bbb57e9c96
* fix build due to changes in DSO Linking the right way
as per bonefish.
* we may need to check host GCC version here... can't find
when this option was introduced
* fixes#8031
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@43191 a95241bf-73f2-0310-859d-f6bbb57e9c96
The flags given to configure are kept and the build just adds its own flags onto these.
Also the kernel flags are also based on the flags given at configure.
I suspect this will be needed for llvm.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@43169 a95241bf-73f2-0310-859d-f6bbb57e9c96
A bit out of my comfortzone with, so please tell me if I'm wrong.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@42949 a95241bf-73f2-0310-859d-f6bbb57e9c96
* libbe_build: Where possible we directly use the actual Haiku headers and
sources, now. In the headers/build headers we just include the respective
Haiku headers as needed. That still allows overrides where necessary. The
intention is to make it easier to keep the build stuff in sync.
* Fixed a few printf() format and signed/unsigned comparison warnings.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@42179 a95241bf-73f2-0310-859d-f6bbb57e9c96
