13:49 <Jacmet> smaeul: I'm trying to get your WIP branches running on my Nezha D1 board, but get a crash as soon as Linux tries to touch the PLIC registers in plic_toggle(). Same behaviour if I try to dump the registers in u-boot. Presumably I must have forgotten something, any ideas?

14:08 <hlauer> could the driver strenght setting of a20 GPIO registers be used for current limiting, e.g. driving LEDs directly?

18:39 <jernej> apritzel: smaeul: H616 reset doesn't work with latest U-Boot master branch on OrangePi Zero2, but with v2021.10 does

18:41 <jernej> I'm trying to figure out why Linux can't reboot OrangePi Zero2 (hangs at "reboot: Restarting system"), but it works on T95 TV box (also H616 based)

19:06 <jernej> apritzel: for some reason, boot over FEL also doesn't work with v2021.01 on OrangePi Zero2

19:06 <jernej> not sure what broke

20:13 <gnarface> jernej: the real rpi's couldn't reboot either, maybe it's a hardware limitation

20:14 <gnarface> jernej: on a PC the bios relies on the ATX power supply to hit the reboot button for it on command, but a pi has neither an ATX power supply, BIOS, nor reboot button

20:15 <gnarface> it would make sense though that something sold as a "TV box" would have a real power supply

20:17 <jernej> no, h616-v5 branch from apritzel has working reboot on OrangePi Zero2

20:17 <jernej> however, if I rebase same branch on something newer, let say 5.14, it doesn't work anymore

20:17 <jernej> now I'm bisecting which commit causes this issue

20:32 <gnarface> oh hmm

20:32 <gnarface> no idea then

20:32 <gnarface> sorry

21:58 <jernej> smaeul: apritzel: found offending commit: 96be41d74f2e ("net: stmmac: dwmac-sun8i: Add a shutdown callback")

21:59 <jernej> bisecting with adding H616 support every time, I found that this commit prevents OrangePi Zero2 to reboot

22:05 <apritzel> jernej: many thanks for doing this (especially "adding h616 support every time" sounds tedious)

22:06 <jernej> not really, fortunately I found one branch where I needed to skip just one commit

22:06 <jernej> nevertheless it took about an hour

22:08 <jernej> still much faster than cluelessly

22:08 <jernej> looking around

22:09 <jernej> I confirmed that taking 5.14 with H616 patches and reverting aforementioned commit, reboot works

22:09 <apritzel> gnarface: rebooting on Allwinner (and most other ARM based SoCs) works via deliberately letting the watchdog kick in, so that's pretty easy and reliable

22:09 <apritzel> jernej: did it just hang/not reboot, or did you see a kernel crash dump

22:10 <jernej> first

22:10 <apritzel> I remember seeing some splats on reboot with some branches, due to refcounting issues

22:10 <gnarface> apritzel: hmm, interesting

22:11 <apritzel> (ala: clocks disabled twice)

22:11 <jernej> yeah, that's still present

22:11 <gnarface> apritzel: do you think that would work for a rpi1?

22:11 <jernej> r-apb2-rsb already disabled

22:11 <apritzel> yes, that one

22:11 <jernej> well, without revert, there is no this crash

22:12 <apritzel> gnarface: don't know, but watchdog reset is the most common method, so if it's not working upstream, there must be a reason

22:13 <apritzel> jernej: so you reckon it's actually sun8i_dwmac_shutdown() hanging?

22:13 <jernej> yes, somehow

22:13 <apritzel> probably something else (clock, reset) already disabled, so MMIO hangs?

22:13 <jernej> but I'm out for today, I'll research this tomorrow (if you or smaeul don't beat me to it)

22:14 <apritzel> jernej: sure, many thanks for the bisect, will have a look

22:14 <jernej> hm... good idea

22:16 <jernej> note that it works on T95 where emac is also used (but second one)

22:16 <jernej> so clocks/resets are good bet