sys/kern/subr_bus.c
6094	You could simplify this a fair bit by just donating the allocated path pointer and then just malloc'ing the `dln` here. It does mean the fini routine above has to free `dln->dln_path` explicitly, but that seems like a lot simpler in implementation.
6123	You can make `path` a `char *` now that `device_get_path` returns that and then remove the need for the __DECONST below.

imp added inline comments.Nov 29 2021, 6:17 PM

sys/kern/subr_bus.c
6094	That assumes that all the callers will have allocated this string and are cool with donating it. I don't think that's the case, especially for a cache miss. Then we'd have to pass back to the caller whether or not to free this thing.
6123	Sure.

jhb added inline comments.Nov 30 2021, 6:17 PM

sys/kern/subr_bus.c
6094	The only caller is the one call below that uses device_get_path which always has an allocated string (this is a static function, it's not one of the public APIs).
6141	This is the sole call, and `path` is always allocated.

Updating with these suggestions.

I also think that before I add FDT, I'll need to shift to creating a locator object and using it to make decisions rather than wiring different behaviors to string compares and stuff (noted here since all the other reviews are accept with reservations and I think this will address them in the future).

sys/kern/subr_bus.c
6094	The only caller is the one call below that uses device_get_path which always has an allocated string (this is a static function, it's not one of the public APIs). You're correct. I was confusing this with a different interface. Those are great suggestions for improvement. Will roll them into next revision.