When the malloc lock and unlock functions are inside a library they
conflict with the standard libraries weak version of these functions.
This is because of the way weak references are handled by the linker.
This patch renames the lock and unlock functions defined inside RTX
so they do not conflict. A thunk inside retarget.cpp then calls the
RTX functions. This problem does not occur with retarget.cpp since
it is always build into an object file rather than a library file.
The define OS_TCB_SIZE does not match the real stack size. This
leaves the system with less than OS_TASKCNT TCBs and prevents
a single thread configuration. This updates OS_TCB_SIZE to the
correct value to fix the problem.
The main stack ends at the start of the heap. In some configurations
this causes the guard word to be overwritten when memory is allocated.
This causes an OS error in RTX since it appears that the main stack
overflowed.
This patch moves the main stack to the upper 1/4th of the heap which
prevents the guard word from getting overwritten.
The rtx SVC_Handler for IAR clobbers r0-r3 despite the number of
arguments. However, in the SVC calls, the __swi function is declared
with fewer arguments. IAR doesn't understand that the other registers are
clobbered and stores variables in r0-r3 when multiple SVCs are
dispatched in a single function.
This bug was noticed in osThreadExit, which hard-faults on IAR,
preventing any threads from exiting.
Now there is no need any more to specify the UVISOR_PRESENT symbol from
the mbed command line tools. By using a target that has UVISOR_SUPPORTED
has a label, the uVisor-internal UVISOR_PRESENT symbol will be set
automatically.
Prevent a switch to a NULL target thread by setting the new task to run
to be the idle task. Otherwise, we get nasty usage fault because we
would be returning from the rt_sys_init SVC with a PSP of 0x20.
This is backwards-compatible with unsupported targets.
If an application is compiled using the K64F target without uVisor
(UVISOR_PRESENT not set or set to 0) then uvisor_init() will just be an
empty function that immediately returns.
This commit includes all sections that are just added to the linker
script. These changes are backwards-compatible, meaning that they will
not affect the existing code.
Targets that do not support uVisor will leave those sections empty.
uvisor-lib has an init function that must be called before the RTOS kernel
is initialized. Call uvisor_lib_init from software_init_hook to accomplish
this.
Wrap software_init_hook so that it can be used or extended from outside the
RTOS. This is desirable so that code can be added to the software_init_hook
without making the RTOS depend on new features or libraries.
Add the OsEventObserver mechanism. A client interested in receiving
notifications on certain OS events can register to receive notifications
with osRegisterForOsEvents. This is useful for clients like the secure
memory allocator, which observes thread switching events in order to swap
in and out different memory allocator objects.
Add the memory management code from uVisor to a shared location that is
suitable for both supported and unsupported use. To do this, we copy all
the RTX-specific files from the RTX folders in uVisor.
Note that the page allocator code is so identical between uVisor supported
and unsupported mode that we can copy the file as-is from uVisor for use on
unsupported targets. This commit performs that copying, to a file called
page_allocator.c_inc, which is included from unsupported_page_allocator.c
only for targets where uVisor is not present.
softdevice.
The call stack of the soft device can be 0x600 (1536) bytes long, by
adjusting the stack to a value of 2048 bytes, their is enough room for
the softdevice and RTX kernel to live together in the main stack.
Random issues due to stack overflow were visible with the previous value.
Russ Butler
Sam Grove
geky
Mihail Stoyanov
Mihail Stoyanov
Yoshihiro TSUBOI
Jimmy Brisson
Milosch Meriac
Alessandro Angelino
Jaeden Amero
Niklas Hauser
Niklas Hauser
Vincent Coubard
bcostm