Fixes issue with memory alignment in STM32 #3045
Conversation
When compiling with optimizations on, an issue occurs where the claimed_pins/never_reset_pins memory location is shared with another variable. This causes some bad memory read, so the USB pins ended up being reset. Setting these to have an alignment of 4 bytes resolves this. Tested on nucleo_f746zg
k0d
force-pushed
the
stm32_alignment_issue
branch
from
f61bc30 to
23da0fa
Compare
|
Why does it matter whether these two variables share a word? It shouldn't. |
|
It's not that they shouldn't share a word, but that they shouldn't cross a word boundary. I don't know why it affects that variable specifically...but some way that it's used is causing issues because it's crossing the word boundary. |
|
Some code might be accidentally treating them as |
|
I'd love to get to the real 100% cause of this, and I will, but I feel that this is a suitable fix for now. I'll keep looking at it, but I need to move on to the DAPIO issue asap. I tried switching everything to uint32_t, this didn't fix it either. |
|
@tannewt You told me once to ping you if PRs are not looked at for a few days...so here's your ping |
|
@k0d Yup! Thank you! I thought you were going to follow up with an update. |
|
Haha...no...that's a good enough fix for now. I'll do some deep dive into it at some point...maybe bring it up at 'in the weeds' on a Monday. |
When compiling with optimizations on, an issue occurs where the claimed_pins/never_reset_pins memory location is shared with another variable. This causes some bad memory read, so the USB pins ended up being reset. Setting these to have an alignment of 4 bytes resolves this.
Tested on nucleo_f746zg