Fix: Semihosting in BMP

[ALTracer]

Detailed description

• The problem is BMP's implementation of semihosting via GDB File-I/O (not BMDA) has been broken since gdb_main: move main loop into main() #1284. Some attempt was made in Fix: Semihosting #1432 to make it at least not crash the adapter.
• The PR solves the problem by introducing a second loop to service GDB packets before/other than F-reply.

I based the WIP tree on PR1432 single commit v1.9.0-435-g906f8878 backported to just past PR1284 merge, v1.9.0-55-g5cf0dd27. I had to enable DEBUG_GDB, DEBUG_GDB_WIRE and drop a bunch of drivers to fit into swlink, because my adapter of choice (blackpill-f411ce) did not exist until approximately v1.9.0-700 in usable form. The logs from BMP internals were critical for understanding how RSP packets go missing.

Current patch adds a while(true) loop at stack depth of 9 functions to allow for parsing the 1-4 packets GDB has to emit before finishing the semihosting File-I/O syscall and returning its result code. AFAIK this has to be avoided in out-of-tree projects like Farpatch which will reboot due to watchdog timeouts from not going through the superloop often enough. But this variant seems to work, and you all can decide how to improve it, because it provides some starting/reference point.

The less related change replaces the syscall 0lx1 (800289c 0 3 20000054) prints
with more readable syscall SYS_OPEN (800289c 0 3 20000054). Since v1.10.0 release I don't care about flash space much; it can be dropped in case of backporting, but it shouldn't affect release firmware because it's in DEBUG_INFO (and I could gate it on ENABLE_DEBUG for 332 bytes).

Tested on swlink attached to Nucleo-G071RB target running SemihostingTest.ino over SWD (with its onboard JLinkReflash idling), using GDB-8/10/12 (NuttX, ST, GNU-RM).

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) and should not affect it
• 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 #1678.

[dragonmux]

The new/extra debug info for syscalls looks quite handy - nice addition. There's a small problem your patch for hostio_reply() introduces, for which we have a suggested fix, but this is looking good overall and explains in the code comments why what we did before was broken (subtly) which is excellent.

It's looking good and with the review item addressed, we'll be happy to merge this.

[dragonmux]

When we merge this we'll backport it to v1.10 as part of the first point release for that firmware - seeing how broken this is in that release.

It would be helpful if @xobs could please give this patch a spin and let us know how it interacts on Farpatch. We are inclined to merge this fix and then any further work that's needed to make Farpatch happy as a second pass.

[xobs]

No issues on Farpatch -- still works just fine, though I've never used semihosting.

[dragonmux]

LGTM, merging and backporting. Thanks for the contribution and all your hard work making it!

[ALTracer]

No issues on Farpatch -- still works just fine, though I've never used semihosting.

The loop is only reachable from (most of) semihosting calls, and it's harder for me to check that it won't cause issue for your platform/RTOS watchdog. If you don't have your own firmware with semihosting, but you have some targets compatible with Arduino cores like stm32duino, then you could just spin Koen's SemihostingTest.ino like I usually do. Although you're welcome to write a simpler testcase, just verify it accesses desktop GDB (some calls like errno and iserror are handled purely inside BMP)