Handle Phosh crash gracefully

When Phosh crash, the phone gets unusable: you have to long press on the power button to forcibly shutdown.

after a crash:

$ ps ax | grep phosh
 2979 antoine   0:00 dbus-run-session /usr/bin/phosh
 2981 antoine   0:17 /usr/bin/phoc -C /usr/share/phosh/rootston.ini -E bash -lc 'gnome-session --builtin --disable-acceleration-check --session=phosh'

If I kill the process manually, lightdm restarts phosh and the phone is usable again. I would be nice for it to be automatic.

added help wanted type::feature ui-phosh labels

I'm wondering why phosh is hanging, maybe there is a way to properly resolve this?

In any case, it is probably possible to create a small watchdog program, that checks if phosh/phoc is still working properly (maybe there's some api call via d-bus or other method that we can check), and if it is not, kill it.

By Oliver Smith on 2020-04-11T00:06:52

I think when this happens for me, the phosh is crashed given that it doesn't show up in ps ax. I was thinking would rather be a patch in phoc instead.

But it is maybe dbus-run-session's fault? I find it quite weird that dbus-run-session doesn't exit. That would be the first thing to figure out IMO.

By Antoine Fontaine on 2020-04-11T08:01:48

phoc -E st (in a regular terminal) then closing phoc doesn't make phoc exit. On the other hand, phoc complains as such:

(phoc:9489): phoc-wlroots-CRITICAL **: 19:36:55.306: [xwayland/xwm.c:1286] xcb error: op 18:0, code 3, sequence 134, value 4194309

(phoc:9489): phoc-wlroots-CRITICAL **: 19:36:57.775: [EGL] command: eglSwapInterval, error: 0x300d, message: "eglSwapInterval"

(phoc:9489): phoc-wlroots-CRITICAL **: 19:36:57.775: [EGL] command: eglSwapBuffers, error: 0x300d, message: "eglSwapBuffers"

(phoc:9489): phoc-wlroots-CRITICAL **: 19:36:57.775: [render/egl.c:440] eglSwapBuffers failed

Which makes me think phoc really should exit.

By Antoine Fontaine on 2020-04-20T23:12:51

https://source.puri.sm/Librem5/phoc/issues/131

By Antoine Fontaine on 2020-04-20T23:12:51

should be fixed in next phoc release (though we would maybe need to wait for gnome-session too). keeping open not to forget to check.

By Antoine Fontaine on 2020-04-21T08:36:24

I just tried master but when the interface crashes there are still no reboot/ui restarts (phosh-git package used).

Did you find a workaround to restart UI? rc-service lightdm restart does not work for me.

PS: linux-postmarketos-qcom-msm8974-5.6.0_rc6-r3 breaks wifi.

By tallero on 2020-05-17T17:23:06

I can't access the URL above (404), is it a private repo? Note that (if I'm not mistaken) phoc is responsible for that and gnome-session also need to be recent (but the latest package could be recent enough?)

By Antoine Fontaine on 2020-05-12T11:15:53

(for the wifi issue, you may want to open an issue for that)

By Antoine Fontaine on 2020-05-12T11:16:27

I forgot to make it public, now it is.

In any case I think this is not related to my problem. Phosh seems running smoothly in background, it's just the screen that is freezed, so maybe it's not phosh' fault.

This is the dmesg when hanging.

dmesg-hammerhead

Related bug: #424 (closed).

By tallero on 2020-05-12T15:22:00

@afontain browser sound keeps playing, so I think phosh still runs normally, otherwise it should have killed the browser since it spawned it, so it has to be #424 (closed).

About the package it should install, the only change I have found is a non-functioning shutdown/reboot button in the overlay.

By tallero on 2020-05-12T15:42:57

I disagree with the conclusion: you can kill phosh without killing the programs it spawned, just like you can kill the window manager on X without killing the X clients. Phoc is the Wayland compositor ("X server"), not Phosh ("window manager + interface").

ps ax | grep phosh
pkill -9 phosh
ps ax | grep phosh
# notice phoc is still running

By Antoine Fontaine on 2020-05-13T07:31:04

your link to your repo above still 404, though :-/

By Antoine Fontaine on 2020-05-13T07:31:29

This happens on 5.6.0-rc6. After happening, restarting lightdm does not unfreeze it.

On the other hand, this suggest it's not phosh or phoc's fault

By Antoine Fontaine on 2020-05-13T08:48:02

your link to your repo above still 404, though :-/

forgot the pmos- prefix

https://gitlab.com/tallero/pmos-phosh-git

By tallero on 2020-05-17T17:23:32

mentioned in issue #424 (closed)

By tallero on 2020-05-12T15:20:22

added 1 deleted label and removed help wanted label

I've tested with phoc's master in qemu, pkill [-9] phosh restarts the session. The next phoc release should resolve this issue.

By Antoine Fontaine on 2020-05-13T23:04:04

The new phosh release indeed fixes the issue, closing.

By Antoine Fontaine on 2020-06-14T10:15:56

closed

By Antoine Fontaine on 2020-06-14T10:15:56

mentioned in merge request !1313 (merged)

By Antoine Fontaine on 2020-06-19T11:30:16

mentioned in issue #712 (closed)

By Antoine Fontaine on 2020-08-13T04:09:12