Since the very beginning, the stm32 port (first called stm, then stmhal now
stm32) has had a special keyboard interrupt feature which works by using
PendSV to break out of any running code. This preemptive ctrl-C was added
long ago in commit 01156d510c.
The stm32 port still uses that code, and current does this:
- If ctrl-C is received on UART or USB then `mp_sched_keyboard_interrupt()`
is called (like all other ports) to set a flag for the VM to see, and
then the VM (or any loop calling `mp_handle_pending(true)`) will
eventually handle the `KeyboardInterrupt` exception, raising it via NLR.
- If another ctrl-C is received while the existing scheduled keyboard
interrupt is still pending (ie the VM has not yet processed it) then a
special hard NLR jump will activate, that preempts the calling code.
Within the PendSV interrupt the stack is adjusted and an NLR jump is made
to the most recent `nlr_push()` location. This is like a normal NLR
except it is called from an interrupt context and completely annihilates
the code that was interrupted by the IRQ.
The reason for the preemptive interrupt was to handle ctrl-C before the VM
was able to handle it. Eventually a mechanism (that's in use today by all
ports) was added to the VM and runtime to be able to check for pending
interrupts. Then the stm32 port was updated to use this mechanism, with a
fallback to the old preemptive way if a second ctrl-C was received (without
the first one being processed).
This preemptive NLR jump is problematic because it can interrupt
long-running instructions (eg store multiple, usually used at the end of a
function to restore registers and return). If such an instruction is
interrupted the CPU remembers that with some flags, and can resume the
long-running instruction when the interrupt finishes. But the preemptive
NLR does a long jump to different code at thread level and so the
long-running interrupt is never resumed. This leads to a CPU fault.
This fault has been previously reported in issues #3807 and #3842 (see also
issue #294). It's now possible to easily reproduce this problem, since
commit 69c25ea865. Running the test suite
over and over again on any stm32 board will eventually crash the board (it
can happen on a PYBv1.x, but it happens more regularly on PYBD-SF2/6).
The point is, a skipped test now soft resets the board and so the board
must run `boot.py` again. The test runner may then interrupt the execution
of `boot.py` with the double-ctrl-C that it sends (in `tools/pyboard.py`,
`enter_raw_repl()`) in order to get the board into a known good state for
the next test. If the timing is right, this can trigger the preemptive
PendSV in an unfortunate location and hard fault the board.
The fix in this commit is to just remove the preemptive NLR jump feature.
No other port has this feature and it's not needed, ctrl-C works very well
on those ports. Preemptive NLR jump is a very dangerous thing (eg it may
interrupt and break out of an external SPI flash operation when reading
code from a filesystem) and is obviously buggy.
With this commit, stm32 borads no longer hard fault when running the test
suite (but it does leave an issue, the tests can still interrupt `boot.py`
with a single ctrl-C; that will be fixed separately).
An alternative to this commit would be to clear the CPU state for the
long-running instruction as suggested in issue #3842. But it's much
simpler to just remove this code, which is now unnecessary and can have
other problems as per issue #294.
Signed-off-by: Damien George <damien@micropython.org>
This changes stm32 from using PENDSV to run NimBLE to use the MicroPython
scheduler instead. This allows Python BLE callbacks to be invoked directly
(and therefore synchronously) rather than via the ringbuffer.
The NimBLE UART HCI and event processing now happens in a scheduled task
every 128ms. When RX IRQ idle events arrive, it will also schedule this
task to improve latency.
There is a similar change for the unix port where the background thread now
queues the scheduled task.
Signed-off-by: Jim Mussared <jim.mussared@gmail.com>
This makes a cleaner separation between the: driver, HCI UART and BT stack.
Also updated the naming to be more consistent (mp_bluetooth_hci_*).
Work done in collaboration with Jim Mussared aka @jimmo.
This commit adds an implementation of a "software timer" with a 1ms
resolution, using SysTick. It allows unlimited number of concurrent
timers (limited only by memory needed for each timer entry). They can be
one-shot or periodic, and associated with a Python callback.
There is a very small overhead added to the SysTick IRQ, which could be
further optimised in the future, eg by patching SysTick_Handler code
dynamically.
This is to keep the top-level directory clean, to make it clear what is
core and what is a port, and to allow the repository to grow with new ports
in a sustainable way.
Damien George