12:56 <DC-IRC> <elias> I'm working on an A64 based board (recore A7) and I'm seeing some dram corruption when switching the DRAM frequency. It appears to be stable when running at full speed, but I'm getting kernel errors that seems to happen when the dram devfreq governor changes speed. If I set the governor to performance, the DRAM frequency is always max, and things work fine, but using sysfsutils to set the governor takes some time and it might cause issues before

12:56 <DC-IRC> <elias> Is there a way to set the dram governor before the kernel has booted? Has anyone else seen issues like this? Here is how I do it today:

12:56 <DC-IRC> <elias> ```root@recore:~# cat /etc/sysfs.d/dram_governor.conf

12:56 <DC-IRC> <elias> class/devfreq/1c62000.dram-controller/governor = performance```

12:56 <DC-IRC> <elias> I can disable the mbus driver init in the kernel device tree and then the frequency is set by ATF or u-boot, but it would be nice to know a bit more about this problem. The DRAM routing is not fly-by-wire, but direct, I think that is different from most of the A64 boards, so that might be relevant.

13:02 <DC-IRC> <Tenkawa> you should be able to modify the dts for that

13:02 <DC-IRC> <Tenkawa> force it to only allow set frequencies and settings

13:04 <DC-IRC> <Tenkawa> Not sure if it would need to be done in both u-boot and the kernel's dts but the kernel for sure

13:30 <DC-IRC> <elias> Thanks for the reply. Looking at the driver (sun8i-mbus.c) It looks to me like the governor is hardcoded to simple-ondemand in kernel 5.15 at least: https://github.com/armbian/build/blob/59363d5a0f1a833bb54efbde0831a2e3f6626c8a/patch/kernel/archive/sunxi-5.16/patches.megous/PM-devfreq-Add-a-driver-for-the-sun8i-sun50i-MBUS.patch#LL482C13-L482C40

13:30 <DC-IRC> <elias> I guess if I don't need that frequency scaling regardless, I might as well disable the driver. This is for a 3D printer, so power saving is not an issue.

13:54 <DC-IRC> <Tenkawa> ah its in a patch.. there might be a "technical" reason that was done though

13:55 <DC-IRC> <Tenkawa> you might need to read and see if there was a bug that that was implemented for

13:55 <DC-IRC> <Tenkawa> (inconsistent behaviour/etc)

14:33 <DC-IRC> <elias> I'm on the legacy kernel, but it looks like it's the same in mainline: https://github.com/torvalds/linux/blob/ae8373a5add4ea39f032563cf12a02946d1e3546/drivers/devfreq/sun8i-a33-mbus.c#L433

14:33 <DC-IRC> <elias> And looking at the implementation in other platforms, it looks similar (hardcoded).