* rtems/include/rtems/rtems/event.h,
rtems/inline/rtems/rtems/eventset.inl, rtems/src/event.c,
rtems/src/eventseize.c, rtems/src/eventsurrender.c,
rtems/src/eventtimeout.c, score/Makefile.am, score/preinstall.am,
score/include/rtems/score/interr.h,
score/include/rtems/score/thread.h,
score/include/rtems/score/threadq.h,
score/include/rtems/score/tqdata.h,
score/inline/rtems/score/threadq.inl,
score/inline/rtems/score/tqdata.inl, score/src/threadq.c,
score/src/threadqdequeue.c, score/src/threadqdequeuefifo.c,
score/src/threadqdequeuepriority.c, score/src/threadqenqueue.c,
score/src/threadqenqueuefifo.c, score/src/threadqenqueuepriority.c,
score/src/threadqextract.c, score/src/threadqextractfifo.c,
score/src/threadqextractpriority.c,
score/src/threadqextractwithproxy.c, score/src/threadqfirst.c,
score/src/threadqfirstfifo.c, score/src/threadqfirstpriority.c,
score/src/threadqflush.c, score/src/threadqrequeue.c,
score/src/threadqtimeout.c: Refactor thread queue enqueue and event
blocking synchronization critical sections. This resulted in three
copies of essentially the same hard to test critical section code
becoming the one shared routine _Thread_blocking_operation_Cancel. In
addition, the thread queue and event code now share a common
synchronization enumerated type. Along the way, switches were
reworked to eliminate dead code generated by gcc and comments and
copyrights were updated.
* score/include/rtems/score/threadsync.h,
score/src/threadblockingoperationcancel.c: New files.
* configure.ac, irq/irq.c, startup/bspclean.c, startup/bspstart.c: Add
ALLOW_IRQ_NESTING option. The MPC5200 has a settle time after
acknowledging the IRQs and currently the BSP does not account for
that. After acknowledging an interrupt, it currently gets a second
spurious IRQ a significant percentage of the time. Rename to
BENCHMARK_IRQ_PROCESSING and get it working again. Under one test
load, not nesting interrupts resulted in a 50% reduction in the
number of IRQs and an ~30% reduction in time spent in IRQs.
* score/include/rtems/score/wkspace.h,
score/inline/rtems/score/wkspace.inl, score/src/wkspace.c: Do not
inline _Workspace_Free or _Workspace_Allocate since they are not
always inlined and actually smaller overall as subroutines. They are
not particularly time critical so inlining is not absolutely
necessary.
* posix/Makefile.am, posix/include/rtems/posix/cond.h,
posix/include/rtems/posix/mutex.h, posix/inline/rtems/posix/cond.inl,
posix/inline/rtems/posix/mutex.inl: Do not include POSIX Mutex or
Condition Variable object get helpers because they are more
complicated than the norm. They can implicitly perform a create. They
cross the line as being too complex and large to inline since they
negatively impact size and binary test coverage.
* posix/src/condget.c, posix/src/mutexget.c: New files.
* libcsupport/Makefile.am, libcsupport/include/rtems/malloc.h,
libcsupport/src/free.c, libcsupport/src/malloc.c,
libcsupport/src/malloc_deferred.c,
libcsupport/src/malloc_initialize.c, libcsupport/src/malloc_p.h,
libcsupport/src/malloc_sbrk_helpers.c,
libcsupport/src/posix_memalign.c: Place all deferred free code and
place it in subroutines. Add plugin for dirtying allocated memory to
assist in debugging. Clean up comments and spacing as needed.
* libcsupport/src/malloc_dirtier.c: New file.
Joel Sherrill
Ralf Corsepius
Jennifer Averett
Till Straumann