The critical section check never succeeded since watchdogs are ordered
with respect to the insertion time. Now we call the watchdog routine if
appropriate and bypass the normal watchdog mechanic.
The FAT32 FS info sector contains hints for the free cluster count and
the next free cluster. The previous code read these values during mount
and replaced them with invalid values. The shutdown operation updated
them with the current values. These values are only hints. Every FAT
implementation must cope with arbitrary values. They are intended to
speed up certain operations.
Now we update the free cluster count and next free culster in the FAT32
FS info sector only during unmount or sync operations and only if the
values have changed. This avoids writes to the FS info sector and
conforms to the behaviour of Linux and Windows.
The application can force an update of these values now with the fsync()
and fdatasync() operations. Applications that only read will perform
not write operations to the FAT32 FS info sector.
The new fat_sync() function performs all non-file specific
synchronizations.
Offset calculation was wrong for 16-bit and 64-bit pointer targets.
Remove unused offsets. Move Per_CPU_Control::dispatch_necessary after
Per_CPU_Control::isr_nest_level. Move SMP members to end of structure.
All assembler relevant members are now at the structure beginning.
The initial extensions remain now in a read-only table and will not be
copied to work space memory. The extension chains are statically
initialized. This makes it possible to call _User_extensions_Iterate()
independent of the system state. It is now guaranteed that the fatal
callout of the initial extensions will be called provided the stack
pointer, the read-only data, and code memory are valid.
Replace the separate user extension iterations with a single iteration
function. This reduces code size and improves maintainability since the
iteration logic is only in one function. The runtime overhead is
insignificant.
The comment in confdefs.h shows how to do it:
An application can define its own scheduling policy by defining
CONFIGURE_SCHEDULER_USER and the following:
- CONFIGURE_SCHEDULER_ENTRY_POINTS
- CONFIGURE_MEMORY_FOR_SCHEDULER - base memory
- CONFIGURE_MEMORY_PER_TASK_FOR_SCHEDULER - per task memory
Since the allocator mutex allows nesting now we can use it for the
generic IRQ support (recursive obtain in a potential malloc()). This
simplifies the code and reduces memory usage.
