Fix fifo memory overflow

[hathach]

Describe the PR

While trying to fix an issue reported in #1729 (detected using fuzzing #1716). I am able to fix a couple of issues with current tusb_fifo when using with overwritable mode

• Prevent double overflowed to happen which make fifo count invalid. Though due to the cost of moving memory, the data at read index is not replaced by newer one.
• fix _tu_fifo_remaining() return incorrect value which cause memory overflowed, detected by fuzzing
• minimize tu_fifo size to 16
• simplify and remove _tu_fifo_empty, _tu_fifo_full. Also correct full condition check
• simplify _tu_fifo_count() and _tu_fifo_remaining(), also rename to _ff_count() and _ff_remaining()
• update advance_pointer/backward_pointer to use depth instead of fifo, also rename to advance/backward_index
• simplify _ff_correct_read_index()
• remove _ff_overflowed() due to lack of use
• using clang with ceedling unit-test with -fsanitize=address

Both of above cases can be reproduced/tested with 2 new unit tests. I also take the chance to add fifo implementation note (since I kind of forgot about it) to help future troubleshooting and rename absolute/relative naming to index/pointer and other minor changes. There is still room to improve fifo, thus submitted this as draft first for feedback.

ref reading: https://www.snellman.net/blog/archive/2016-12-13-ring-buffers/9

probably fix #1039 , and also related to adafruit/circuitpython#7041 since this could be the bug of cdc fifo instead of what we thought is Linux with echo mode.

[PanRe]

Looks good! Nice that the buffer would be fuzzing-proof :)

Is there anything else to improve since you mentioned there would be room therefore!

[hathach]

Looks good! Nice that the buffer would be fuzzing-proof :)

Is there anything else to improve since you mentioned there would be room therefore!

just update the PR with your feedback, thank you for reviewing. For room to improve, I feel that we could do better to simplify the current code, but it is not urgent. Like merging get_realitve_poiinter() and _ffmod() to idx2ptr(), mostly for readability and maintenance only. I will try to wrap this up soon enough.

[silvergasp]

Just a suggestion. It might be worth enabling -sanitizer=address which will detect any overflows in your unit test.

However that particular flag does require clang.

[hathach]

Just a suggestion. It might be worth enabling -sanitizer=address which will detect any overflows in your unit test.

However that particular flag does require clang.

yeah, this is great idea. I will check to see we could switch to clang. Unit test like fuzzing run on host, so that wouldn't be an issue at all.

[silvergasp]

Apologies the flag is -fsanitize=address docs here https://clang.llvm.org/docs/AddressSanitizer.html

[hathach]

Apologies the flag is -fsanitize=address docs here https://clang.llvm.org/docs/AddressSanitizer.html

unit-test is now configured to run with clang and -fsanitize=address e885ced , thanks for your suggestion.