Hi,
okay, from what I understand, this COULD very well be an issue with the timings of the RAM chips.
What we know:
* 2GB RAM boards seem work fine without any freeze on 4.15 and 4.18 kernel * 4GB RAM boards seem to freeze after 5 - 10 minutes on 4.15 and 4.18 kernel * Memtester did work for Tony with the new, optimized 4GB RAM PCB (don't know which kernel, MLO, U-Boot, etc. has been used) * Memtester did work fine with Tonys patches for notaz on the OLD, unoptimized 4GB RAM PCB.
So, it doesn't seem to have much to do with the voltage drop (otherwise the old 4GB RAM PCB wouldn't have survived the memtester). But it can't simply be the new PCB either, as it worked for Tony with memtesting.
Could it be noise? (simply running memtest, maybe even without a screen attached should have less noise than a fully booted Pyra). (I'll test if it crashes as well when I only connect a PCB and also if it crashes if I idle around in U-Boot).
Could it be that the timings are on the edge and therefore it works for some 4GB RAM PCBs but not for others? That would also explain why the bootloader even sometimes fails to detect the RAM properly in my test unit.
In that case, I'd love to play around with some timings.
Could someone help me here? As far as I understand, the EMIF timings are setup in the MLO. I've got a Debian server, I've got an ArchLinux system at home.
Can anyone let me know where to grab the MLO sourcecode and how to compile it for the Pyra? I can provide access to the debian server if someone wants to set it up for me.
I guess the EMIF timings are properly documented in the MLO sourcecode, so I can compare them with the datasheet and change them?