06:38 <davidlt[m]> aurel32: at some point it should land, it's already in U-boot.

06:38 <davidlt[m]> there are a bunch of other stuff that should probably go upstream.

06:41 <davidlt[m]> I have a branch somewhere with a stuff that should go to upstream.

06:45 <pierce> meanwhile i'm just day dreaming about the imaginary 0005-riscv-sifive-unmatched-add-gpio-reboot-node.patch

06:47 <davidlt[m]> pierce: There is a patchset under review (well more RFC) which kinda provides that.

06:48 <pierce> oh, i thought it was constrained by hardware limitations? (no gpio access to reboot)

06:48 <davidlt[m]> The thing is that the sequence doesn't provide true reset, but from users point of view it is.

06:48 <pierce> ah right, yep

06:49 <davidlt[m]> Not exactly as you can talk to PMIC via I2C, but you cannot assert RESET, just doesn't work.

06:50 <davidlt[m]> The way to deal with it is to use RTC tick wake up signal, but IIRC it could cause problems as discussed in the review of that patchset.

06:50 <davidlt[m]> You kinda could "miss it" if the system is rebooting at the same time as tick is happening.

06:50 <davidlt[m]> Also there are other problems too mentioned in the review.

06:51 <davidlt[m]> Like users might want to use this RTC tick wake up signal for other setups.

06:51 <davidlt[m]> Also doing this within OpenSBI is also complicated as some clearing of events needs to happen after the reboot on the Linux side IIRC.

06:51 <davidlt[m]> Thus I am not sure if you could contain as-is a proper thing within OpenSBI.

06:52 <davidlt[m]> OpenBSD went with tick wake up implementation for Unmatched.

06:52 <davidlt[m]> Basically it's a hardware quirk for Unmatched.

06:53 <davidlt[m]> There is no AUTOBOOT set in the OTP in DA chip and that's what Linux drivers expect by default.

06:53 <pierce> DA chip?

06:53 <davidlt[m]> DA9063 PMIC is what we use for Unmatched.

06:54 <pierce> ah right, yep

06:54 <davidlt[m]> The SoC itself isn't going via DA9063, there is a separate chip for SoC power IIRC.

06:54 <davidlt[m]> But IO, MEM, etc. all go via DA9063.

06:55 <davidlt[m]> You can buy a dongle with RTC for 20 USD or cheaper and use that to power down and powerup :)

06:55 <davidlt[m]> Or go with PiKVM (I want one!)

06:56 <pierce> i honestly haven't touched my unmatched for a little while now (work and things) but i'm hopeful the rough edges do get smoothed out

06:56 <davidlt[m]> or use the existing patchset that's available on the mailing list.

06:56 <pierce> davidlt[m]: oh really?! do you have a link?

06:56 <davidlt[m]> See this thing: https://files.digiowls.org/doc/wdt-pro2-eng.pdf

06:56 <pierce> davidlt[m]: yeah, i did see this, it wasn't available to purchase last time i checked, still crowd funding

06:57 <davidlt[m]> It has watchdog, but it allows to control reset/shutdown buttons via a bash script (untested by me, but I really like this product after reading docs and exchanging some emails with them).

06:58 <davidlt[m]> You basically just echo <command> > /dev/ttyBLAHBLAH

06:59 <davidlt[m]> ~Mx Commands for controlling remote devices: 1 - Reset, 2 - Power.

06:59 <davidlt[m]> ~M1 -- trigger reset button, ~M2 -- trigger power button

07:00 <pierce> davidlt[m]: this one won't fix a likely data centre issue of when an unmatched board loses power, would it?

07:00 <davidlt[m]> IIUC

07:00 <davidlt[m]> It would

07:00 <davidlt[m]> The buttons are line on a normal PC, they are powered by ATX 5V always on line

07:00 <davidlt[m]> So if you track Unmatched (pings or something) you can now when to trigger the power button.

07:01 <pierce> davidlt[m]: of course!

07:01 <davidlt[m]> Like: If (unmatched_ping > 10 minutes) { press power_button }

07:01 <davidlt[m]> Just have a raspberry pi or something that always powers up on power.

07:02 <davidlt[m]> I have something like that for Koji farm for VMs.

07:02 <davidlt[m]> If machine is not responding for an hour it's destroyed and new one launches.

07:02 <davidlt[m]> Koji constantly records each builders last ping time.

07:03 <davidlt[m]> Sadly you cannot short power on button pins

07:03 <davidlt[m]> It has to be shorted and released to power on

07:04 <davidlt[m]> thus you need something "smart" to mimic actual button press

07:04 <pierce> yeah, it would need to be momentary

07:04 <davidlt[m]> These watchdog dongles are usually used for bitcoin miners to keep systems always online

07:05 <davidlt[m]> there are cheap ones <10 USD, but this one is so well documented, etc.

07:05 <davidlt[m]> But again, untested by me.

07:06 <davidlt[m]> I should get one to test.

07:22 <aurel32> something smart that to mimic actual button press can just be a capacitor with possibly a resistor in parallel. It usally works on PC, so it might just work on unmatched

07:23 <pierce> Would that be momentary though?

07:24 <aurel32> yes, when the board is powered, the capacitor is like a wire, than it charges and is like an open circuit

07:26 <aurel32> the resistor is for discharging it when the board is not powered anymore, otherwise it might takes dozen of seconds to discharge, so you need to wait long enough between a power-off and a power-on

07:26 <aurel32> usually 100µF / 10k works well, but I don't know if it works on the unmatched

07:26 <aurel32> and you can put that in parallel to the button, which continues to work

08:38 <rwmjones> morning

08:54 <pierce> <aurel32> "yes, when the board is powered..." <- Goes to show how little I know about electronics! Very cool