Fix: Semihosting

[dragonmux]

Detailed description

In the wake of #1284, Koen correctly identified that a private (UB) declaration of gdb_main_loop() had the semihosting Host IO code broken with nobody having noticed. This PR addresses this by removing that declaration, putting one in the gdb_main.h header so mismatches with gdb_main.c result in compilation failure, and updating how the Host IO code calls the main loop.

Your checklist for this pull request

• I've read the Code of Conduct
• I've read the guidelines for contributing to this repository
• It builds for hardware native (make PROBE_HOST=native)
• It builds as BMDA (make PROBE_HOST=hosted)
• I've tested it to the best of my ability
• My commit messages provide a useful short description of what the commits do

Closing issues

Fixes #1385

[dragonmux]

@koendv Please give this a test when you find a minute. It should have solved the bug you reported back in February. We are unfortunately ill-equipped with suitable firmware, etc, to actually test this ourself so please let us know how you get on and if this PR needs to go further.

[esden]

LGTM

[koendv]

ok.
As test programs I use the examples of this Arduino library: https://github.com/koendv/STM32duino-Semihosting
Running the test program "SemihostingTest.ino.elf" in a stm32f103 with openocd, a commercial probe, blackmagic hosted and blackmagic firmware ought to give mostly similar results. (I say similar because not all debuggers implement all semihosting calls. E.g. blackmagic firmware implements semi-hosting using the gdb fileio calls, while blackmagic hosted implements semi-hosting using posix system calls.)
Sample output of "SemihostingTest.ino.elf" is the attachment "semihosting_testbench.txt" in #665
Output of current firmware is semihosting-output.txt
I add a zip with the compiled binary of "SemihostingTest". SemihostingTest.zip
Tests were run with arm-none-eabi-gdb 12.1.90.

My impression is it doesn't quite work.
When using bmp firmware, sometimes speed is an order of magnitude slower than I expect; as if something in bmp is looping. Return values are odd; e.g. the flen() semihosting call for determing file length returns -1, independent of file length.
Edit: Using latest git, semihosting works on bmp hosted.
This feels as if there might be more than one bug.

I'd like a second opinion on this one.

[koendv]

I would suggest adding to the documentation: If using black magic debug app with black magic debug, make sure both have the same major release. (e.g. not app 1.9 with firmware 1.8)

[dragonmux]

Using a newer BMDA with older firmware should work (there are fallbacks in BMDA for this) - if it doesn't, please let us know as that's a bug.

Do note that v1.8 firmware with BMDA from main will fall back right now to the low-level JTAG/SWD remote functions as the protocol rewrite done in #1389 did not extend to providing a complete set of backwards compatible remote protocol functions for firmware with remote protocol versions 1 and 2, so that may be part of where the slowness you were experiencing is coming from