8MB (or more) Motherboard Fastram with JIT and KMS triggers crash

[solskogen]

trafficstars

With a A4000 or A4000T quickstart, adding 8MB or more Motherboard Fast ram crashes amiberry.

[solskogen]

Same with Processor slot Fast RAM

[midwan]

This is something I cannot recreate, at least on the x86_64 version I'm testing with. We're going to need more specifics

[solskogen]

Raspberry Pi 5, Raspberry Pi OS bookworm aarch64.

[solskogen]

Happens only when JIT is enabled.

[midwan]

amiberry_crash.log

[midwan]

It doesn't happen on my Pi400, but I think I narrowed down when it probably happens. I've made a small change in the commit above (in a separate branch), give it a try when you can and see if that helps or not!

[solskogen]

To make matters worse: It does not happen in sway or X11. Only in KMS.

[midwan]

@solskogen does this happen on the preview branch? That one uses a completely different memory allocation routine, so maybe it helps

[solskogen]

Does that mean we have JIT for aarch64 on preview now? :-)

[midwan]

Doh, I forgot this only happens with JIT - never mind :)

[midwan]

Did some more testing today. This is not related to the memory allocation, but rather to something in the JIT implementation, in combination perhaps with newcpu. When this is triggered, the memory address retrieved in newcpu, seems to be invalid. Not sure why, but I suspect something in the JIT implementation, since it was designed to hit the memory directly at all times.

The preview branch has JIT for x86, but that's the "proper" version with indirect memory access if direct is not possible. I expect that one will work normally with KMSDRM as well.

[cmitu]

Maybe something similar to https://github.com/raspberrypi/bookworm-feedback/issues/107 ? You can try switching to a 4k page kernel by adding kernel=kernel8.img to /boot/firmware/config.txt and rebooting.