* libcsupport/include/rtems/libcsupport.h, libcsupport/src/newlibc.c,
sapi/Makefile.am, sapi/include/confdefs.h, sapi/src/exinit.c,
score/Makefile.am, score/preinstall.am,
score/include/rtems/score/userext.h, score/src/chain.c,
score/src/userext.c: Switch to newlib reentrancy extension being
installed in the initial set instead of using rtems_extension_create.
While implementing this, noticed that user extensions and chain code
had multiple functions in a single file which is not desirable in the
SuperCore and API portions of RTEMS, so split these into multiple
files with one function per file. Also noticed that some of user
extension code was inlined for no particular reason so moved that to
C bodies. Split executive shutdown from initialization since not
every application shuts down. Moved __fini call to executive shutdown
to be more symmetrical with where it is called at startup.
* sapi/src/exshutdown.c, score/src/chainappend.c,
score/src/chainextract.c, score/src/chainget.c,
score/src/chaininsert.c, score/src/userextaddapiset.c,
score/src/userextaddset.c, score/src/userextremoveset.c,
score/src/userextthreadbegin.c, score/src/userextthreadcreate.c,
score/src/userextthreaddelete.c, score/src/userextthreadrestart.c,
score/src/userextthreadstart.c, score/src/userextthreadswitch.c: New
files.
* score/inline/rtems/score/userext.inl: Removed.
* ChangeLog, libcsupport/src/malloc.c,
libcsupport/src/mallocfreespace.c, sapi/include/confdefs.h,
score/Makefile.am, score/preinstall.am: malloc never blocks so the
Region Manager is quite heavy for implementing this. This patch
implements the C Program Heap directly in terms of the new Protected
Heap handler. This handler is a direct use of a SuperCore Heap in
conjunction with the Allocator Mutex used internally by RTEMS. This
saves 3184 bytes on most SPARC test executables.
* score/include/rtems/score/protectedheap.h, score/src/pheapallocate.c,
score/src/pheapallocatealigned.c, score/src/pheapextend.c,
score/src/pheapfree.c, score/src/pheapgetblocksize.c,
score/src/pheapgetfreeinfo.c, score/src/pheapgetinfo.c,
score/src/pheapinit.c, score/src/pheapresizeblock.c,
score/src/pheapwalk.c: New files.
* posix/Makefile.am, posix/include/rtems/posix/time.h,
posix/src/adjtime.c, posix/src/alarm.c, posix/src/clockgetres.c,
posix/src/condtimedwait.c, posix/src/mqueuetimedreceive.c,
posix/src/mqueuetimedsend.c, posix/src/mutextimedlock.c,
posix/src/nanosleep.c, posix/src/posixtimespecabsolutetimeout.c,
posix/src/pthread.c, posix/src/pthreadcreate.c,
posix/src/pthreadsetschedparam.c, posix/src/ptimer1.c,
posix/src/sched.c, posix/src/semtimedwait.c,
posix/src/sigtimedwait.c, posix/src/ualarm.c,
rtems/src/clocktodtoseconds.c, score/Makefile.am,
score/preinstall.am, score/include/rtems/score/tod.h,
score/inline/rtems/score/tod.inl, score/src/coretod.c,
score/src/coretodget.c, score/src/coretodgetuptime.c,
score/src/coretodset.c, score/src/coretodtickle.c: Provide timespec
manipulation routines in the SuperCore. Use them everywhere possible.
This lead to significant cleanup in the API routines and eliminated
some of the same code from the POSIX API. At this point, the
SuperCore keeps time in POSIX timespec format properly from 1970. You
just cannot set it before 1988 in keeping with RTEMS traditional
behavior.
* score/include/rtems/score/timespec.h, score/src/timespecaddto.c,
score/src/timespecfromticks.c, score/src/timespecisvalid.c,
score/src/timespeclessthan.c, score/src/timespecsubtract.c,
score/src/timespectoticks.c: New files.
* posix/src/posixintervaltotimespec.c,
posix/src/posixtimespecsubtract.c,
posix/src/posixtimespectointerval.c: Removed.
* score/Makefile.am, score/include/rtems/score/tod.h,
score/inline/rtems/score/tod.inl: Make _TOD_Tickle_ticks a real
non-inlined routine. It should only be used once so there is little
advantage to inlining it.
* score/src/coretodtickle.c: New file.
* score/src/heapallocatealigned.c, score/src/threadqrequeue.c: Correct
license URL and/or fix mistake in copyright notice. Both of these
mistakes appear to be from code submitted after these changes were
made previously.
PR 1222/cpukit
* score/Makefile.am, score/include/rtems/score/coremutex.h,
score/include/rtems/score/threadq.h,
score/inline/rtems/score/coremutex.inl, score/src/coremsgsubmit.c,
score/src/coremutexsurrender.c, score/src/threadchangepriority.c,
score/src/threadclearstate.c, score/src/threadhandler.c,
score/src/threadinitialize.c, score/src/threadqdequeuefifo.c,
score/src/threadqdequeuepriority.c, score/src/threadqenqueue.c,
score/src/threadqenqueuefifo.c, score/src/threadqenqueuepriority.c,
score/src/threadqextractfifo.c, score/src/threadqextractpriority.c,
score/src/threadsetstate.c: Enhance so that when the prioirity of a
thread that is blocked on a priority based thread queue is changed,
that its placement in the queue is reevaluated based upon the new
priority. This enhancement includes modifications to the SuperCore as
well as new test cases.
* score/src/threadqrequeue.c: New file.
* posix/src/prwlockunlock.c, rtems/Makefile.am,
rtems/src/barrierrelease.c, rtems/src/barrierwait.c,
score/src/corerwlockobtainread.c, score/src/corerwlockrelease.c:
Complete implementation of barrier as the rest of the test code is
now implemented. Also add blocking code to rwlock.
* score/src/heapwalk.c: Do not use printk yet since many BSPs do not
prove the hooks for printk yet. This will require enough BSP changes
to delay until 4.8.
PR 841/rtems
* itron/inline/rtems/itron/semaphore.inl, itron/src/twai_sem.c,
posix/include/rtems/posix/semaphore.h,
posix/inline/rtems/posix/semaphore.inl,
posix/src/semaphorewaitsupp.c, posix/src/semtimedwait.c,
posix/src/semwait.c, rtems/src/semobtain.c,
rtems/src/semtranslatereturncode.c,
score/include/rtems/score/coresem.h, score/src/coresemseize.c: Make
sem_timedwait more conformant to Open Group specification.
PR 1124/rtems
* score/include/rtems/score/threadq.h, score/src/coremutexseize.c,
score/src/coremutexsurrender.c, score/src/threadqenqueue.c,
score/src/threadqenqueuefifo.c, score/src/threadqenqueuepriority.c:
The placement of the changing a thread's priority when using priority
ceiling should be on the successful transfer of the mutex -- not when
the thread tries to acquire. Plus the lack of a dispatch disable
point lead to the potential for a thread timing out and already
having inherited the ceiling priority.
Victor V. Vengerov <Victor.Vengerov@oktetlabs.ru>
* score/include/rtems/score/mpci.h,
score/include/rtems/score/threadmp.h,
score/inline/rtems/score/threadmp.inl,
score/macros/rtems/score/threadmp.inl, score/src/threadmp.c,
score/src/threadqenqueue.c: Victor spotted the problem that
_MPCI_Receive_server_tcb and _Thread_MP_Receive were duplicate
variables and needed to be set to the same value. I took that
idea and just removed _Thread_MP_Receive. All uses are now
_MPCI_Receive_server_tcb.
PR 761/rtems
* score/src/threadhandler.c: Add volatile cast so test is weak function
is present will do something. Otherwise, it can be assume by gcc to
always be a true condition.
PR 850/rtems
* score/src/watchdogtickle.c: A Watchdog (used to timeout an event)
with a delay of 1 sometimes does not seem to timeout. The problem
occurs, because for whatever reason when the watchdog tickle function
executes, the watchdog->delta_interval is 0. it is then decremented
before being tested, becomes huge and so doesnt time out. It is
thought there is a race condition where the watchdog->delta_interval
is calculated by reference to a head (also with a delay of 1). But
before it can be added after the head, the head is removed, so the
new head now has a delay of 0.
PR 820/rtems
* score/inline/rtems/score/coremsg.inl,
score/macros/rtems/score/coremsg.inl, score/src/coremsginsert.c:
Increment of pending message count should be atomic with insertion on
the pending message chain. Determination of the need to call the
notification handler should also be in this atomic section of code.
PR 807/rtems
* rtems/src/timerfireafter.c, rtems/src/timerserverfireafter.c,
score/src/watchdoginsert.c: Tighten critical section checks on an ISR
using the same timer being inserted by a lower priority ISR or
interupt task.
PR 746/rtems
Optimize realloc(). The problem is that realloc() can neither grow
nor shrink efficiently the current memory region without support
from underlying heap/region modules. The patch introduces one new
routine for each of heap and region modules, _Heap_Resize_block(),
and rtems_region_resize_segment(), respectively, and uses the
latter to optimize realloc().
The implementation of _Heap_Resize_block() lead to changing of the
heap allocation strategy: now the heap manager, when splits larger
free block into used and new free parts, makes the first part of
the block used, not the last one as it was before. Due to this new
strategy, _Heap_Resize_block() never needs to change the user
pointer.
Caveat: unlike previous heap implementation, first few bytes of
the contents of the memory allocated from the heap are now almost
never all zero. This can trigger bugs in client code that have not
been visible before this patch.
* libcsupport/src/malloc.c (realloc): try to resize segment in
place using new rtems_region_resize_segment() routine before
falling back to the malloc()/free() method.
* score/src/heap.c:
(_Heap_Initialize): change initial heap layout to reflect new
allocation strategy of using of the lower part of a previously
free block when splitting it for the purpose of allocation.
(_Heap_Block_allocate): when split, make the lower part used, and
leave the upper part free. Return type changed from Heap_Block* to
uint32_t.
* score/include/rtems/score/heap.h:
(Heap_Statistics): added 'resizes' field.
(Heap_Resize_status): new enum.
(_Heap_Resize_block): new routine.
(_Heap_Block_allocate): return type changed from Heap_Block* to
uint32_t.
* score/src/heapwalk.c: reflect new heap layout in checks.
* score/src/heapsizeofuserarea.c: more assertions added.
* score/src/heapresizeblock.c: new file.
(_Heap_Resize_block): new routine.
* score/src/heapfree.c: reverse the checks _Heap_Is_block_in() and
_Heap_Is_prev_used() on entry to be in this order.
* score/src/heapallocate.c, score/src/heapallocatealigned.c:
ignore return value of _Heap_Block_allocate().
* score/Makefile.am (HEAP_C_FILES): added src/heapresizeblock.c.
* rtems/include/rtems/rtems/region.h:
(rtems_region_resize_segment): new interface routine.
(_Region_Process_queue): new internal routine called from
rtems_region_resize_segment() and rtems_region_return_segment().
* rtems/src/regionreturnsegment.c: move queue management code into
the new internal routine _Region_Process_queue() and call it.
* rtems/src/regionresizesegment.c: new file.
(rtems_region_resize_segment): new interface routine.
* rtems/src/regionprocessqueue.c: new file.
(_Region_Process_queue): new internal routine containing queue
management code factored out from 'regionreturnsegment.c'.
* rtems/Makefile.am (REGION_C_FILES): Added
src/regionresizesegment.c, and src/regionprocessqueue.c.
* ada/rtems.adb, ada/rtems.ads: Added Region_Resize_Segment.
PR 753/rtems
* score/include/rtems/score/object.h,
score/src/objectinitializeinformation.c: Configured number of objects
needs to be of a type with more bits than Objects_Maximum to support
configuring unlimited objects. The type for configuring the maximum
number of objects should probably always be the same fundamental type
as Object Id. For 16-bit Id configurations, the constant
OBJECTS_UNLIMITED_OBJECTS was changed to 0x8000 to move in this
direction.
