Gentoo for Amlogic S9xx (TV box S905\S905X\S912)

[The_Pope]

Ideally a dmesg from a system where the periferals work well and one from gentoo/Armbian the kernel is more or less the same but Armbian has more firmware installed.

Fantastic. On Libreelec, via ssh, you have to type "dmesg | paste" and it outputs the link to an already uploaded log! In my case, it's http://ix.io/1vUQ

I forgot to say: if a cable is plugged in to the ethernet port, it doesn't show up as a network connection. The bogus IP only shows up if I plug it in.

The S812 should work well now with mainline kernel, if you manage to configure all devices and find the right dtb

On S812 under Libreelec, even the wired ethernet works. Didn't manage to figure out if it's actually gigabit, though. The problem is, how do I get it to boot, after assuming I managed to install it all.

Now to convince Armbian to give me the dmesg log. EDIT: Ok, this is it.

[erm67]

The ethernet phy is from ST Micro, we should verify if mDup compiled the driver for it:

Code: Select all

[    0.353478@0] libphy: stmmac: probed
[    0.353490@0] eth0: PHY ID 02430c54 at 0 IRQ POLL (stmmac-0:00) active

Code: Select all

 [    2.411841] libphy: Fixed MDIO Bus: probed
 [    2.414556] tun: Universal TUN/TAP device driver, 1.6
 [    2.420505] meson8b-dwmac c9410000.ethernet: PTP uses main clock
 [    2.425132] meson8b-dwmac c9410000.ethernet: no reset control found
 [    2.431722] meson8b-dwmac c9410000.ethernet: User ID: 0x11, Synopsys ID: 0x37
 [    2.438419] meson8b-dwmac c9410000.ethernet:     DWMAC1000
 [    2.443587] meson8b-dwmac c9410000.ethernet: DMA HW capability register supported
 [    2.451005] meson8b-dwmac c9410000.ethernet: RX Checksum Offload Engine supported
 [    2.458422] meson8b-dwmac c9410000.ethernet: COE Type 2
 [    2.463596] meson8b-dwmac c9410000.ethernet: TX Checksum insertion supported
 [    2.470580] meson8b-dwmac c9410000.ethernet: Wake-Up On Lan supported
 [    2.476988] meson8b-dwmac c9410000.ethernet: Normal descriptors
 [    2.482829] meson8b-dwmac c9410000.ethernet: Ring mode enabled
 [    2.488608] meson8b-dwmac c9410000.ethernet: Enable RX Mitigation via HW Watchdog Timer
 [    3.524413] libphy: stmmac: probed
 [    3.524662] mdio_bus stmmac-0: MDIO device at address 3 is missing.

The last message is not good probably that is why the eth is not working, probably there is no external PHY on board and it returns some grabage data that is falsely detected as an STMicro phy. The board has no external Gbit PHY, so ethernet is the internal 100Mbit. Try the dtb with 100Mbit ethernet.

The wifi card is a 4330

Code: Select all

[   16.306759@2] F1 signature OK, socitype:0x1 chip:0x4330 rev:0x4 pkg:0x0

You need to copy the following files to gentoo:

Code: Select all

[   20.414778@2] Final fw_path=/usr/lib/firmware/brcm/fw_bcm40183b2.bin
[   20.414782@2] Final nv_path=/usr/lib/firmware/brcm/nvram_ap6330.txt
[   20.414787@2] Final conf_path=/usr/lib/firmware/brcm/config.txt

You have the off-tree GPU driver installed and need to recompile the kernel in gentoo (or rip the LE kernel&drivers)

Code: Select all

[   13.816137@0] mali-utgard d00c0000.mali: set min pp to default 1
[   13.816149@0] mali-utgard d00c0000.mali: set min clk default to 0
[   13.816248@0] mali-utgard d00c0000.mali: max clk set 5
[   13.816253@0] mali-utgard d00c0000.mali: turbo clk set to 6

I am a bit busy atm ......

[mDup]

The ethernet phy is from ST Micro, we should verify if mDup compiled the driver for it:

Thanks for detailed analyse.
I have used the balbes150 Armbian 5.64 4.19.0-rc7 .config verbatim for my gentoo arm64 image.
It can be found in /boot or simply /proc/config.gz.
I plan to tune 4.20.x further but I want to stay with balbes150 baseline as start, if only to keep discussion simpler upstream.
Like: if something does (not) work in gentoo, compare if it does (not) work in armbian.
fwiw: I am a bit busy with other things as well.
Have a nice day

[erm67]

fwiw: I am a bit busy with other things as well.
Have a nice day

Don't worry I think we can assume that he realized that he was scammed and the box only supports 100Mbit, he is probably getting over it, or maybe he is looking for the seller that advertised it as supporting Gbit eth

mDup: do you know wether the new Amlogic boxes with 4 Gb RAM and USB3 are already supported by mainline?

https://www.cnx-software.com/2018/11/01 ... promotion/
I think I will finally be able to run an elasticsearch server for nextcloud with a box like this ...
It would probably work well as a desktop as well.

[mDup]

[...]mDup: do you know wether the new Amlogic boxes with 4 Gb RAM and USB3 are already supported by mainline?[...]

Sorry for late reaction.
I have seen these boxes mentioned indeed. I am not a HW guru.
I guess best is to check on Armbian forums where the one and only balbes150 supports many devices.

[balbes150]

You can see the support status here.

http://linux-meson.com/doku.php

So far it is almost in zero condition. Perhaps the General part of the system will be able to run on this chip, but so far these samples (as far as I know) no leading developers associated with Linux. In this regard, the models on RK33xx (rk3328 and rk3399) look very interesting and are rapidly developing in terms of support. From Rockchip has a special unit that specializiruetsya on supporting Linux. According to my observations, this process in RK is much faster than in Amlogic.

For example, the Khadas Edge model (rk3399) already has basic support for adding a simple patch with DTS to the sources. The performance of this model is already quite provides a replacement "household" PC, while providing new functionality, inaccessible products In***l\A***D. Allowing you to get an interesting new device PC and TV box in one device, with low (relative to PC) power consumption and the absence of spy hardware elements, which are now crammed all PCs.

The presence of these spyware hardware bookmarks, puts under full control all products In***l\A***D, no matter what OS they are running. So those who think that using Linux, it is protected from the invasion of their freedom and privacy - very wrong, everything is now done at the level of the hardware of the processor and its component and what OS is running is absolutely not important.

[The_Pope]

Don't worry I think we can assume that he realized that he was scammed and the box only supports 100Mbit, he is probably getting over it, or maybe he is looking for the seller that advertised it as supporting Gbit eth

Hahaha, no, I couldn't care less. The armv7 box appears to work with a gigabit dtb, this one doesn't. Neither is mine, both belong to the Libreelec maniac.

But it seems that no matter how many times I say this, you'll still assume I'm some entitled dork thinking he's some fantastic hacker and tries to pay 5$ for a desktop and have it work as a supercomputer.

I'm busy too, I just haven't had time to try the 100Mbit dtb. I am treated like a god: people acknowledge my existence when they need something.

[erm67]

Don't worry I think we can assume that he realized that he was scammed and the box only supports 100Mbit, he is probably getting over it, or maybe he is looking for the seller that advertised it as supporting Gbit eth

Hahaha, no, I couldn't care less. The armv7 box appears to work with a gigabit dtb, this one doesn't. Neither is mine, both belong to the Libreelec maniac.

But it seems that no matter how many times I say this, you'll still assume I'm some entitled dork thinking he's some fantastic hacker and tries to pay 5$ for a desktop and have it work as a supercomputer.

I'm busy too, I just haven't had time to try the 100Mbit dtb. I am treated like a god: people acknowledge my existence when they need something.

And than you ask in some forum on their behalf you could be an übergod if you only learned to google things like "brcmfmac driver config files"

[erm67]

You can see the support status here.

http://linux-meson.com/doku.php

So far it is almost in zero condition. Perhaps the General part of the system will be able to run on this chip, but so far these samples (as far as I know) no leading developers associated with Linux. In this regard, the models on RK33xx (rk3328 and rk3399) look very interesting and are rapidly developing in terms of support. From Rockchip has a special unit that specializiruetsya on supporting Linux. According to my observations, this process in RK is much faster than in Amlogic.

For example, the Khadas Edge model (rk3399) already has basic support for adding a simple patch with DTS to the sources. The performance of this model is already quite provides a replacement "household" PC, while providing new functionality, inaccessible products In***l\A***D. Allowing you to get an interesting new device PC and TV box in one device, with low (relative to PC) power consumption and the absence of spy hardware elements, which are now crammed all PCs.

The presence of these spyware hardware bookmarks, puts under full control all products In***l\A***D, no matter what OS they are running. So those who think that using Linux, it is protected from the invasion of their freedom and privacy - very wrong, everything is now done at the level of the hardware of the processor and its component and what OS is running is absolutely not important.

Both Amlogic and Rockchip are fabless, when you buy one of their products you don't really know what's inside the box or who produced it ... fabless companies cannot be responsible if some factory puts a spy chip in their design in the production line.

[mDup]

Both Amlogic and Rockchip are fabless, when you buy one of their products you don't really know what's inside the box or who produced it ... fabless companies cannot be responsible if some factory puts a spy chip in their design in the production line.

I may be wrong but I think it was suggested breaking protection is
currently also, or even more, focused on hardware than software.

If so then, for someone with bad intentions, it does not matter so
much where the hardware was made and if it was in or out house. After
all: in house nothing is perfect either. There is hole or spy in there too.

If I were a bad person? : I would go for results in terms of
quantity. This is why, I always assumed, but I may be wrong, I am not
into cracking, Windows attacks were more interesting for bad persons
than Apple or Linux. Not because cracking Linux or Apple is harder.

From what I observed in the past: -all- people I know with Windows
software have once been victim, I repeat: -all-. But -nobody- I know
with Linux software. I happen to not know any Apple victim
either.

I want to say : 100% versus 0% is a clear balance.

But: the past is the past.

Perhaps in the future, or already today, victims are indeed OS
independent.

But then: how to have most success as a cracker?

If I were a bad person? : I would go for vast majority "Intel and
AMD", only later for minority "Amlogic or Rockchip".

OK, this goes way Off Topic.

Let's have fun running gentoo on ARM 32 or 64.

happy newyear everyone

[erm67]

I was able to 'emerge -e @world' using lto and graphite optimization (not extreme) thank to swap over iscsi, there were 4Gb swap used linking mariadb for erxample. graphite optimization works better if the compiler knows the cache size and unfortunately -mcpu=native on aarch64 doesn't automatically set the cache size, so according to Amlogic specifications the level 1 cache is 32k for all products while S905 have 512k level 2 cache and S912 has 2 256k l2 caches. Find the cache line size was a bit more difficult but using a little asm from ARM documentation I found out it is 64.

Code: Select all

#include <stdio.h>

void main(int argc, char**argv)
{
  unsigned icache_lsize;
  unsigned dcache_lsize;
  static unsigned int cache_info = 0;
  if (! cache_info)
    /* CTR_EL0 [3:0] contains log2 of icache line size in words.
       CTR_EL0 [19:16] contains log2 of dcache line size in words.  */
    asm volatile ("mrs\t%0, ctr_el0":"=r" (cache_info));
  icache_lsize = 4 << (cache_info & 0xF);
  dcache_lsize = 4 << ((cache_info >> 16) & 0xF);
  printf("log2 of icache line size in words: %d\n",icache_lsize);
  printf("log2 of dcache line size in words: %d\n",dcache_lsize);
}

This tool is better however: https://github.com/pytorch/cpuinfo

For Amlogic CPUs L1 cache is 16kb 4-way set associative data cache + 16kb 2-way set associative instruction cache = 32k but gcc is only interested in data cache, so for a S912 add this to the CFLAGS:

--param l1-cache-size=16 --param l1-cache-line-size=64 --param l2-cache-size=256

and this for a S905:

--param l1-cache-size=16 --param l1-cache-line-size=64 --param l2-cache-size=512

[erm67]

@mDup:
They fixed the ethernet interface for good, it's now possible to remove the 'eee-broken-1000t'
see:
https://patchwork.kernel.org/patch/10712159/

[mDup]

@mDup:
They fixed the ethernet interface for good, it's now possible to remove the 'eee-broken-1000t'
see:
https://patchwork.kernel.org/patch/10712159/

good to know

btw: for my minimx-g, I also need to use 'amlogic,tx-delay-ns = <4>'
I wil check kernel fix when I build next kernel.

[erm67]

@mDup:
They fixed the ethernet interface for good, it's now possible to remove the 'eee-broken-1000t'
see:
https://patchwork.kernel.org/patch/10712159/

good to know

btw: for my minimx-g, I also need to use 'amlogic,tx-delay-ns = <4>'
I wil check kernel fix when I build next kernel.

I patched 4.20.2-gentoo with the patch from the kernel ML, removed 'eee-broken-1000t' recompiled and everything is working
Well I guess you only need to patch the dtb, but it changed recently and patch was not backported yet, better upgrade to 4.20.x and patch the dts.

[mDup]

@mDup:
They fixed the ethernet interface for good, it's now possible to remove the 'eee-broken-1000t'
see:
https://patchwork.kernel.org/patch/10712159/

good to know

btw: for my minimx-g, I also need to use 'amlogic,tx-delay-ns = <4>'
I wil check kernel fix when I build next kernel.

I patched 4.20.2-gentoo with the patch from the kernel ML, removed 'eee-broken-1000t' recompiled and everything is working
Well I guess you only need to patch the dtb, but it changed recently and patch was not backported yet, better upgrade to 4.20.x and patch the dts.

Thanks for prompt test.
I am looking forward to try linux 5 too

[mDup]

My server is headless but there is a driver called panfrost coming for accelerated graphics that could support the S912 as well

linux 5.2.0 with panfrost module and gitlab.freedesktop.org mesa with panfrost gallium seems to work good for me.
That is: a HD video works fine audio/video in sync and smooth (with mvp) whereas it would not with mainline mesa.

[erm67]

You should add S922X to the list of supported chipsets

[mDup]

You should add S922X to the list of supported chipsets

I guess so yes. I have no S922X box yet (I may soon buy Khadas VIM3).
I plan to make a new bootable image based upon mainline 5.2.x with LxQt as tiny GUI.

[alegatorul]

Hi everyone,
this is my first post here
any updates for Amlogic S905Y2 Quad Core?
can i find or use any images for my android tv stick?
it's the model X96S

thanks

[user]

H,
Radxa Zero using Amlogic S905Y2 Quad Core.
You may try recommended/official images from https://wiki.radxa.com/Zero

Using Gentoo for Amlogic S905Y2 Quad Core takes more effort.

1) patched kernel from https://github.com/radxa-repo/bsp (Radxa loses pace to keep patched mainline kernel version up-to-date)
2) patched u-boot bootloader also from https://github.com/radxa-repo/bsp
3) use qemu + binfmt-wrapper + chroot for cross compile
4) doing day long compile session
5) get familiar with Amlogic boot rubbish (no future-proof GPT layout possible as It expects to find the FIP binary in sector 1, 512 bytes offset from the start)
6) also enjoy loveless broadcom wifi device/driver still without WPA3/SAE support in 2024