routines and structures that require CPU model specific information
are now in libcpu. This required significant rework of the
score/cpu header files and the creation of multiple header files
and subdirectories in libcpu/i960.
that decouples exec/ for the sh, m68k and i960 from targopts.h.
NOTE: The change to system.h is a hack to enable cpuopts.h
for some targets, but keep using targopts.h for others - I know it
does *not* work for sparc, mips, i386 and ppc. This will have
to be addressed as work continues on multilibing.
that addresses aspects of the targopts.h multilib related
issues.
Changes:
* Move targopts.h to libbsp/include, because the current targopts.h
actually is a per-BSP-header and therefore can not stay below exec/.
* Introduce an autoheader generated header file
(exec/score/include/rtems/score/cpuopts.h), which shall take per-cpu
configuration options only.
* Move all autoconf-detectable/configure specified per-cpu
option-defines from targopts.h to cpuopts.h.
* Add Makefiles to the libbsp/shared directory hierarchy.
Notes:
* The new per-bsp targopts.h in libbsp includes the per-cpu
cpuopts.h. This way, the new targopts.h is kept backward compatible
to the old targopts.h and existing BSPs which (carelessly) include
targopts.h (i386, ppc) should be kept working when using the
multilib-disabled configuration scheme.
* cpuopts.h is not yet complete, because the per-BSP make-targopts
rules from custom/<BSP>.cfg files can not be applied to files below
exec/ when building multilibs.
* All files below exec/ should not include targopts.h anymore, but
should include cpuopts.h instead. However, eliminating inclusion of
targopts.h currently triggers further structural / header file inclusion
related issues, because several ports apply BSP or CPU_MODEL specific
defines from targopts.h below exec/
that is another small step toward multilibs:
Changes:
* Eliminate RTEMS_CHECK_CUSTOM_BSP from all configure.ins below exec/
* Add RTEMS_CHECK_CUSTOM_BSP to the multilib-disabled branch in
RTEMS_ENV_RTEMSCPU to preserve the old per-BSP building scheme
functional.
Notes:
* --enable-multilibs is still non-functional
* This patch should not have any influence on the old building
scheme (--disable-multilib should silently fall back to the old
building scheme).
* running ./bootstrap is required after applying the patch
from score/cpu to libcpu because the determination of which to use is
based on RTEMS_CPU_MODEL. Thus it can not be determined based solely on
multilib information.
where alarm() did not correctly account for the watchdog start_time
and stop_time fields being based on ticks not seconds. This resulted
in alarm() returning a bogus number of seconds remaining.
that is a somewhat experimental, multilib-related patch:
Changes:
* Use RTEMS_ENV_RTEMSCPU instead of RTEMS_ENV_RTEMSBSP in
configure.ins below exec/.
At the moment, RTEMS_ENV_RTEMSCPU is more or less an optical change
to emphasize that these subdirectories shall not depend on RTEMS_BSP
than a real behavioral change.
* Add AC_DEFINE_* to several aclocal/*.m4 macros to prepare
autoheader/autoconf generated targopts.h and similiar configuration
headers.
* c/src/configure.in: remove exec from cfg_subdirs if multilibs are
enabled (c/src is build per bsp, exec shall be build per cpu in c/
or from the toplevel in future, when multilibs are enabled.)
Notes:
* This patch should not have any impact on the current building scheme.
* --enable-multilib still does not work.
* running bootstrap from the toplevel directory is required.
CVS: ----------------------------------------------------------------------
CVS: Enter Log. Lines beginning with `CVS:' are removed automatically
CVS:
CVS: Committing in .
CVS:
CVS: Modified Files:
CVS: c/src/tests/libtests/termios/init.c
CVS: ----------------------------------------------------------------------
Comments follow:
Building linux/posix fails due to including lib/include/sys/cdefs.h
instead of linux's sys/cdefs.h.
AFAIS, this sys/cdefs.h is specific to newlib and should probably
not be used on any unix host.
Therefore, I changed the relevant Makefile.am/configure.in to
condionally install sys/cdefs.h only if newlib is present.
This triggered another bug inside of the newlib-check for one of
RTEMS target's (Unfortunately I can't remember which - mips/i960 ?
... one of the more exotic ...), for which int func() isn't
identical to void func().
Note: This patch needs to be tested under Cygwin and Solaris. I am
particularily unsure about Cygwin/posix, as it applies a different
version newlib which must not trigger RTEMS newlib check (it is
supposed to fail) and if it may need RTEMS's sys/cdefs.h.
Joel Sherrill