Rename the scheduler ask for help stuff since this will be replaced step
by step with a second generation of the scheduler helping protocol.
Keep the old one for now in parallel to reduce the patch set sizes.
Update #2556.
The single write to memory or ioport output are mostly
atomic operations already. The proper memory synchronization barrier
should be used around them to guarantee ordering (sync or eieio
on PowerPC for example) but because I have not found settable
portable primitive only compiler barrier is used.
It should be enough on x86 because the externally visible order
should be/is guaranteed to be preserved on x86 architecture.
When GCC option -march is not specifies i386-rtems toolchain
defaults to i386 architecture instruction set. It does not
provide atomic instructions which results in really inefficient
atomic_fetch_or even on UP build.
SMP build is broken with i386 set because libatomic and GCC
generate infinite loop for __atomic_fetch_add_4 used
in rtems_interrupt_lock_acquire
__atomic_fetch_add_4:
push %ebp
mov %esp,%ebp
movl $0x5,0x10(%ebp)
pop %ebp
jmp __atomic_fetch_add_4
The global state of enabled and disabled interrupts has to hold
interrupts really disabled by drivers and system. If the state is
combined with interrupts temporarily disabled because they are
processed at given time then it is impossible to maintain state
by interrupt handlers in drivers.
The use of actual cache line max bytes and minimum required alignment
in architecture but not-BSP dependent code could be problematic
because there exists even ARM instruction set implementations
with 128 byte line length and real maximum can be quite problematic
to say. But actually supported ARM BSPs should be OK with these values.
Use of rtems_cache_get_maximal_line_size() is more descriptive
choice. The min/max data/instruction cache line size is not critical
there, value is used for optimization only to use single operation
for directly following sections.
Rename _CORE_mutex_Seize_no_protocol_slow() in _CORE_mutex_Seize_slow().
Remove previous _CORE_mutex_Seize_slow() since the protocol handling is
now done in the thread queue operations.
Only use CPU_Per_CPU_control if it contains at least one filed. In GNU
C empty structures have a size of zero. In C++ structures have a
non-zero size. In case CPU_PER_CPU_CONTROL_SIZE is defined to zero,
then this structure is not used anymore.
Sebastian Huber
James
Alexander Krutwig
Pavel Pisa
Chris Johns