Hi folks! 
I have a bit of an interesting bug for you here.
We have an application (STM32L4) that does RMW operations on the flash where it reads contents into a buffer, replaces certain bytes, erases the page and writes the new contents. The application is getting stuck just after the erase (We can see the page is erased afterwards), but the write never seems to happen. We are not seeing any exceptions, so it would appear the MCU gets stuck somewhere. The full RMW operation in enclosed in a critical section mutex. Alignment is already checked and double checked.

The more interesting part is, we did a git bisect to figure out when this was introduced, and found the commit that seemingly broke it (bump of a git dependency). 
Downgrading the git dependency to the rev before seems to fix it. Strange thing is that building with the working git rev, vs checking it out and build with `path = ""` dependency (Exact same rev checked out), yields a binary that does not work?! and it is 350 bytes larger than letting cargo do the clone?!

We have been hunting this for almost 2 weeks now using all kinds of debuggers, and are at a loss now. ANY suggestions are gratefully received!