Question about aarch64 memory access about rust-raspberrypi-os-tutorials HOT 4 CLOSED

The asm output for the 1st variant (loop over values) generates 32-bit load/stores

   801d8:   ldr   w8, [x1]
   801dc:   movk  w21, #16128, lsl #16
   801e0:   str   w8, [x0, #20]
   801e4:   ldr   w8, [x1, #4]
   801e8:   stp   w8, wzr, [x0, #24]

The second variant (copy_from_slice) seems to use one 64-bit load/store

   801d8:   ldr   x8, [x1]
   801dc:   movk  w21, #16128, lsl #16
   801e0:   str   wzr, [x0, #28]
   801e4:   stur  x8, [x0, #20]

I'm not sure why one would work and not the other.

from rust-raspberrypi-os-tutorials.

Hi,

the cluprit is 801e4: stur x8, [x0, #20], resulting from mbox.buffer[5..7].copy_from_slice(&data[0..2]);, which is indeed an unaligned data access. It tries to store a 64bit word to a non-64bit-aligned address (20%8 != 0).

This causes a synchronous exception in AArch64, that's why your RPi3 hardware just halts.
If you port your code to chapter 0E and insert it after vector tables have been set up, you would get prints from the exception handler pointing you directly to this instruction address. You need to disable the MMU setup though (I opened #11 to remind myself to investigate this).

from rust-raspberrypi-os-tutorials.

Ah, thank you for the pointer and the explanation. That makes perfect sense now. Exceptions are next on my list so I guess now will be as good of a time as ever to jump into them.

Big thanks for the work on this repo and other embedded crates. They made the experience of porting my C64 emulator to bare-metal RPI3 quite enjoyable!

Cheers,

from rust-raspberrypi-os-tutorials.

Glad to hear, thanks 👍

I'll go ahead and close this, and try to get +strict-align upstream so this doesn't happen anymore.

from rust-raspberrypi-os-tutorials.