Closed rhclayto closed 7 years ago
Also, after some testing, without the hardware acceleration, the scroll lag was just horrendous. Huge CPU spikes on all four cores on the slightest scroll. Somehow I need to use the vc4 driver but without the crashing! It seems I'm in a bind. :-/
would QtWebEngine be a valuable solution ?
It's HW accelerated on Xorg, but it needs node-worker instead of Electron. https://github.com/WebReflection/benja/issues/22#issuecomment-320229837
Investigating now.
Well since switching back to dtoverlay=vc4-fkms-v3d
I have had no crashes! Whereas before it used to be consistent, at least one crash within a 24-hour period. Perhaps I chcaged something in the electron app itself that has corrected this, but if I did, I have no idea what it was. So anyway, I am having no crashes with the following boot settings:
config.txt
gpu_mem=8
dispmanx_offline=1
dtoverlay=vc4-fkms-v3d
avoid_warnings=2
dtparam=audio=on
disable_overscan=1
cmdline.txt:
root=/dev/mmcblk0p2 rootfstype=ext4 rw rootwait quiet loglevel=0 splash console=ttyAMA0,115200 console=tty1 selinux=0 plymouth.enable=0 smsc95xx.turbo_mode=N dwc_otg.lpm_enable=0 kgdboc=ttyAMA0,115200 elevator=noop cma=256MB
So I'll close this.
oh gosh ... I use gpu_mem=256 as minimum, no idea how your Electron even starts with 8MB of GPU ... are you sure you need the GPU at all ?
From what I understand, the open-source vc4 driver used by dtoverlay=vc4-fkms-v3d
doesn't use gpu_mem
at all. Eric Anholt, who is writing the driver, has said things like this on a few occasions in the GitHub issue queue: 'Note: The gpu mem setting in config.txt reserves memory that doesn't get used by the open graphics driver.' (https://github.com/anholt/linux/issues/28), '(Also note that the open driver doesn't want any gpu_mem)' (https://github.com/anholt/linux/issues/47), 'vc4-kms-v3d doesn't care about any of your config.txt display settings (except avoid_warnings=1 which is required). If any of config.txt did affect vc4-kms-v3d, that would actually be a bug. ' (https://github.com/anholt/linux/issues/92).
So it seems to open vc4 driver itself isn't responding to anything in config.txt,. However, I can't boot up at all without something in gpu_mem
, so perhaps it's needed for an early stage before the vc4 driver takes over. I have cma=256MB
in cmdline.txt, so RAM should be assigned dynamically to the GPU as needed. Keeping gpu_mem low frees up memory for the CPU.
I am no expert at all in any of this, but the config I have now has been working for about three days. Fingers crossed it keeps working.
Hi, I'm experiencing GPU crashes with BENJA (on Raspberry Pi 3) when the process has been running for a number of hours. Usually, if I leave it overnight the GPU will have crashed by the morning & the HDMI monitor no longer receives signal from the Raspberry Pi. This is the relevant journalctl logs for when wayland/weston starts:
and this is the log for when it crashes:
All after
pacman -Syu
updating everyuthing, latest Raspberry Pi firmware, etc.It appears to be a problem with the OpenGL drivers set with
dtoverlay=vc4-fkms-v3d
in/boot/config.txt
. I couldn't get weston to run any other way than the way you have it here, so I have had to use SDDM window manager/desktop & script BENJA to start over it.This may not be a BENJA issue per se, but since BENJA ships with this config for Waylad/Weston & the experiemntal vc4 drivers, & since I am coming from a non-graphical *nix background (FreeBSD servers actually) & have spent all day wading through a welter of graphics-related stuff that didn't solve much of this problem, I thought I'd bring it up here.
A couple of questions, then. Is there some way first of all to fix this while still using Weston? For instance, a way to get weston to work on Raspberry Pi without using the experimental drivers?
Secondly, since everything is working OK with SDDM/Xorg, how are you accomplishing the
touch reload
reloading of the app? It no longer works with the SDDm setup.