C6/H2: flip-link feature #1008
C6/H2: flip-link feature #1008
Conversation
There was a problem hiding this comment.
This is awesome! Really cool that we can achieve this without double linking too, neat!
I have a few comments, mostly around docs just because linker scripts are insanely hard to reason about, so the more docs and info on what they're doing and why the better.
| "ld/riscv/hal-defaults.x", | ||
| out.join("hal-defaults.x"), | ||
| )?; | ||
| preprocess_file(&config_symbols, "ld/riscv/asserts.x", out.join("asserts.x"))?; |
There was a problem hiding this comment.
I don't think these two files need preprocessing?
There was a problem hiding this comment.
True. I just thought applying it everywhere is easier than having some files pre-processed and some not 🤔 but not sure what is better
|
I also wonder if we can improve the error message, if we are in the trap handler and we know SP is out of bounds, we can reason that a stack overflow was detected 🤔. |
We might re-use e.g. exception code 14 to signal that - probably worth a try. Then esp-backtrace could give a definitive hint that it was a stack overflow. Need to try if that would work |
|
That works! Now we get |
|
In the meantime, I tested with esp-wifi .... looks good so far |
bjoernQ
force-pushed
the
esp32c6-esp32h2-flip-link-alike
branch
from
0d97101 to
71a9f9a
Compare
|
Can we please try to remember to update the CI workflow when new features are added, especially when linker scripts are involved. |
Ah true ... |
* C6/H2: flip-link feature * CHANGELOG.md entry * Include .wifiextrairam in .rwtext.wifi * Set exception code 14 if SP was out of bounds
Since on C6/H2 the caches are not residing in the start of RAM we can move the stack to start of RAM to get zero-cost stack overflow protection (inspired by flip-link)
I will probably add some code in esp-backtrace to give a hint when there is an access fault slightly below
_stack_endto make things a bit more obvious.Currently a stack overflow with the flip-link feature looks like this
An access fault trying to access memory near the start of RAM indicates a stack overflow
This needs a lot of testing before it's fine to merge.