Orange Pi Zero 3 | 1.5 GB RAM variant looses Ethernet adapter on soft reboot about dietpi

[lalakii]

I think I may need to add a configuration menu to the mainline u-boot, but I don't know much about that, would anyone do that.

[lalakii]

This u-boot seems to have a lot of problems with "dram_init".

In the "mctl_calc_size" method body.

both values have the potential to occur.

        u8 width = config->bus_full_width ? 4 : 2;

        // Sometimes it's 2, sometimes it's 4

When the memory size is incorrectly recognized, the system may crash if the memory usage exceeds the actual amount available.

[lalakii]

I have the 1.5GiB board

I found out the difference by comparing the source code.

source code: u-boot/arch/arm/mach-sunxi/dram_sun50i_h616.c

static unsigned long mctl_calc_size(const struct dram_config *config)
{
        u8 width = config->bus_full_width ? 4 : 2;

        /* 8 banks */
        unsigned long size;
        size = (1ULL << (config->cols + config->rows + 3)) * width * config->ranks;
        //  -----   CONFIG_DRAM_SUN50I_H616_TRIM_SIZE=y   -----
        size = (size * 3) / 4;
        //  ------------------------ END -----------------------
        return size;
} 

[MichaIng]

Ah nice, you just nees to add size = (size * 3) / 4; and this works for U-Boot and also the final Linux system booted?

This looks so easy and obvious, I wonder whether iuncuim did not find that.

Also not sure how to implement into mainline U-Boot. Since this is not done in the device tree, I also have no other idea than basically adding this very same flag. At least it can potentially serve other H616/H618 boards as well.

EDIT: Ah, I think he does the very same for H616 boards like Zero 2. I am currently in China where gist.github.com is blocked, but based on snippets, he edits the same function. The problem IIRC was that the physical RAM size could not be checked the same way on H618, hence he aimed to do it without flag but dynamically bases on hardware. I could imagine that U-Boot is not happy to accept such SoC specific flags in the main config, but not sure whether anyone asked.

EDIT2: Not the SoC is the difference, but LPDDR3 vs LPDDR4. However, this means we can use mainline U-Boot, only applying a one-line patch.

I will ask at Armbian whether they would accept an additional build target for this, otherwise we can do with a fork, hosting the build on our server, so the issue is solved for the 1.5G variant as well.

[lalakii]

Hello!

I've tried different kernels and found that they all have the wrong temperature information, which makes them unbootable

One problem has been identified so far:

[    2.143164] thermal thermal_zone0: gpu-thermal: critical temperature reached, shutting down
[    2.151581] reboot: HARDWARE PROTECTION shutdown (Temperature too high)
[    2.185438] reboot: Power down

My old kernel could boot fine because I removed the temperature module.

[lalakii]

However there doesn't seem to be anything else I can do about this problem. :(

[MichaIng]

Next week I can generate a U-Boot build which matches the one we use for the 1/2/4 GB variants, just with this patch added. That one (using the Armbian build system), does have the right temperatures.

[MichaIng]

Hmm, one would expect that config->bus_full_width is the correct condition for width to result in the correct memory size?