01:41 <wens> juanesf91: thanks, so it is either really a different die (which I doubt) or it is fused off.

01:45 <wens> hmm, I thought at least the a527 and t527 are pin compatible drop-ins, but looks like they aren't?

01:46 <wens> juanesf91: could you use devmem to poke the npu clock registers and see if it exists or not on the a527?

02:11 <juanesf91> wens: Please give me references as a guide to deliver what you need.

02:12 <juanesf91> With this https://github.com/apritzel/peekpoke ?

06:45 <wens> juanesf91: run this cmd to toggle the bits # busybox devmem 0x20016e0 32 0xffffffff

06:45 <wens> juanesf91: then run this to read back and give me the results? # busybox devmem 0x20016e0

10:57 <apritzel> wens: both the A527 on my Radxa (plastic package, M000000H), and the H728 in the X96QPro+ TV box have the NPU clock register @0x20016e0, it reads back as 0x8700001f, so matching the T527 manual

10:57 <apritzel> will tests my other boards later

11:24 <MoeIcenowy> but maybe the bus select pin of npu is fused off?

11:26 <MoeIcenowy> dlan, apritzel: btw I heard (at radxa) that someone is hired to do u-boot mainlining for A733

12:07 <apritzel> MoeIcenowy: yeah, Tom Cubie mentioned something like this in the CNX forum a few weeks ago, but I was hoping we would be referring to dlan

12:10 <apritzel> (well, not weeks, but just five days ago)

12:11 <wens> it's possible

12:13 <wens> probably check the bus gate / reset register, 0x07102070?

12:20 <wens> "Allwinner is also exploring kernel upstreaming contracts and has already been in touch with Collabora and Baylibre."

12:21 <wens> interesting

12:21 <apritzel> does anyone have a pointer to the BSP code determining the exact chip (A523 vs. A527 vs. T527)? Is that SID based?

12:22 <apritzel> wens: I was contacted by someone from Baylibre like early May, but haven't heard back since then. And I think this was more about T527

12:26 <wens> apritzel: I think so # SID based

12:26 <wens> I think I spotted some related code in the downstream NPU driver

12:31 <wens> check drivers/npu/aw_nna_vip/linux/allwinner/gc_vip_kernel_drv_platform.c ?

12:45 <apritzel> ah, found something in Syterkit: https://github.com/YuzukiHD/SyterKit/blob/0c9adf11193fb5304085d6f9d82888a16acfbc31/board/avaota-a1/board.c#L360-L381

13:16 <wens> maybe we should do an socinfo driver?

15:14 <lschmid> apritzel: I see. How would I go about configuring the gpio-mux then to select the correct port when loaded, and is it supported in u-boot too?

15:17 <lschmid> Oh wait. I guess "idle-state" would be the thing then?

15:38 <apritzel> lschmid: I have no idea, I was just looking for something more architected than a gpio-hog

15:41 <juanesf91> wens: the same as apritzel

15:41 <juanesf91> https://usercontent.irccloud-cdn.com/file/XHPCUEX4/1000148178.jpg

15:54 <lschmid> apritzel: Would the gpio-hog even be upstreamable or will I get the patch thrown back to me because of it anyways?

15:56 <apritzel> wens: SMCCC v1.6 adds a SoC *name* to the SOCID interface, where TF-A can transport a string encoded in registers (similar to x86's CPUID)

16:01 <lschmid> apritzel: As for the gpio-mux. It would work in the kernel with the idle-state simply set to "1". However gpio-mux doesn't seem to be implemented in U-Boot...

17:07 <wens> lschmid: what are you trying to model?

17:08 <wens> ok I see it

17:45 <apritzel> lschmid: I just found gpio-mux while looking for prior art, other (more hackish) tricks would be some fake regulator or a fake reset line

17:46 <apritzel> if you can get the mux to work, that would be ideal. Since it's in the official DT bindings, that should meet little resistance, and this switch can be looked at as some special mux (between on and off)

17:47 <apritzel> if the kernel supports it, and there are existing DT bindings, it's typically a no-brainer to get that into U-Boot, and far easier than doing that for the kernel

18:02 <apritzel> I see, you don't actually reference it ...

18:58 <lschmid> apritzel: Sorry for just throwing this last patch out there. I'll be gone for the next two weeks and didn't wan't to leave the last version of the patch how it stands. I'll see what I get back after the two weeks or maybe I get a better Idea. I do however know that the gpio-mux with the idle-state set to 1 does work and does exactly what i wan't it to do

18:59 <lschmid> And yes. The whole point of the mux is to only disconnect it during the bootrom and fel. U-Boot and Linux will just "set and forget" it basically since we're "trusted" at this point

19:00 <lschmid> wens: I've got a usb multiplexer on my System-on-Module that serves to disconnect the usb-otg lines in "untrusted" software state, e.g. in BootROM or FEL to not allow FEL access to bypass the Secure Boot system

23:06 <yang2> How do i load "sun7i-dwmac" ethernet module?

23:12 <apritzel> yang2: if you want Ethernet on the Allwinner A20, you need dwmac-sunxi.ko

23:12 <apritzel> created by CONFIG_DWMAC_SUNXI=m, or you build this in (=y)

23:12 <yang2> hmmmm, I see that it used to be supported on the 6.5.5 gnu-kernel

23:13 <yang2> but right now with the same kernel i cannot make ethernet working

23:13 <yang2> also I don't seem to have /lib/modules/6.5.5 ...

23:14 <yang2> if you look at the dmesg here https://wiki.parabola.nu/Computers/BananaPi_M1

23:14 <yang2> you can see it recognized

23:16 <yang2> hm

23:19 <yang2> https://paste.debian.net/hidden/49100781/

23:19 <yang2> can something from here be temporarily reused?

23:21 <Jookia> for a different kernel? no

23:21 <yang2> hm

23:32 <apritzel> yang2: do you use modules? what is your config? something self made? If you don't know what you are doing, please use sunxi_defconfig

23:35 <apritzel> or even better: use a distro kernel, A20s are well supported by off-the-shelf distros

23:37 <yang2> apritzel: I use the distro kernel, however it seems that I am missing /lib/modules/the-booted-up-kernel, i don't know why

23:37 <yang2> 'lsmod' doesnt' display any modules at all

23:39 <Jookia> you need to reinstall your kernel if thats the case

23:39 <yang2> right

23:39 <yang2> thinking about how to do that with only UART access, no ethernet working

23:40 <apritzel> yang2: do you have an older kernel still installed, a version where you have the modules directory? then boot that, and then install the modules

23:41 <yang2> I only have 5.18 and several 6.5.5

23:41 <yang2> but none of these in /lib/modules

23:42 <yang2> i can boot into 5.18 and 6.5.5

23:43 <apritzel> do you have a USB to Ethernet adapter? This might help for the moment

23:43 <yang2> yes, i have

23:43 <yang2> i tried both

23:43 <apritzel> otherwise you can download the .deb(?) package in your browser, copy it to the SD card, then "dpkg -i" it

23:44 <yang2> RJ-45 ethernet jack and USB/ETH dongle

23:44 <yang2> i can also eject the SD card and upload the package on it and then re-run pacman installer, thatr would do

23:48 <apritzel> yes, don't know how to look up Arch packages, but on Debian the dwmac-sunxi.ko modules is in the normal linux-image-6.1.0-32-armmp package

23:48 <yang2> yes, debian doesn't seem to apply in this case

23:50 <yang2> I guess these dwmac-sunxi.ko.gz can't be reused https://paste.debian.net/hidden/49100781/

23:50 <yang2> because they belong to different kernels

23:50 <apritzel> no, the .ko must exactly match your kernel (normally)

23:51 <yang2> ok

23:51 <yang2> Alright, thanks, I will continue in the morning

23:52 <apritzel> yang2: I guess you want linux-armv7 6.16.4-1?

23:55 <yang2> actually

23:55 <yang2> # uname -a

23:55 <yang2> Linux banana 6.5.5-gnu-1-vanilla #1 SMP PREEMPT Tue, 03 Oct 2023 22:10:37 +0000 armv7l GNU/Linux

23:56 <yang2> these are customized linux-libre kernels, not mainline