You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I apologize for opening this issue as I'm not a TinyCore user, but this is my last resort.
For several months, I've been trying to get tinyx to run at full speed without success. Despite extensive research and various attempts, the performance remains slow. I've attached a video that demonstrates the issue, although it's longer than intended due to the slowness.
screencast-20240912-15-44.mp4
I'm using Oasis and compiled tinyx from source. My window manager is a lightly customized version of dwm, but the problem persists with the standard version as well.
In the kernel, I've enabled several framebuffer and Xfbdev related options, including:
for CONFIG_DRM_I915_ALPHA_SUPPORT pass i915.alpha_support=1 on the kernel command line.
also in my notes on kernel config i have the following: "https://wiki.gentoo.org/wiki/Xorg/Configuration KMS and FB drivers conflict". the accompanying commit disables all CONFIG_FB_ items. see 86cba4d4b7de4cf240b8a2e71a93d4657cca4464 in sabotage linux. i recall that before disabling those i had some issues with FB, but i dont recall whether it was performance related.
the config items i posted are 5.4 era. i cant help you anymore than what i stated above. try to disable all _FB_ items and enable the KMS ones. i hope you use ccache for the kernel build...
No problem with filing this issue here, this isn't exclusively TinyCore sw.
[h264 @ 0x11ecfc0]error while decoding MB 41 47, bytestream -47
[h264 @ 0x11ecfc0]concealing 94 DC, 94 AC, 94 MV errors in P frame
Your video has a lot of h264 corruption errors, so I'm not entirely sure what is from the corruption and what from the recording. But it seems to show that your shell took ~8s to show a prompt. While Xfbdev is certainly slow in comparison to Xvesa and accelerated drivers, you're right that this is greatly too slow, especially as you look like you're running on a more modern cpu.
I'm afraid I don't know what could be the main issue. The DRM fbdev emulation is considered terribly slow and bad in general, so that would be my first guess, but I think many TC users use the KMS FB with Xfbdev without such issues. Try profiling? Perf or oprofile's operf should do (with root rights and the options that enable full-system and kernel profiling).
Oprofile's docs are quite good, haven't used perf much. In short, you first configure one or two sysctl options to enable wide profiling, then instead of "Xfbdev -args" you launch "operf -operfargs Xfbdev -args", run it for a bit, and using the tools then look at the results.
maybe @aligrudi has some advice or tuned kernel config, as he's the author of some linux fb software.
Any particular reason why you prefer older versions?
i guess i just cant be bothered to try a new kernel and fix all the new issues that inevitably show up, when the one i use is still LTS-supported and works just fine.
maybe @aligrudi has some advice or tuned kernel config, as he's the author of some linux fb software.
> Any particular reason why you prefer older versions?
i guess i just cant be bothered to try a new kernel and fix all the new issues that inevitably show up, when the one i use is still LTS-supported and works just fine.
It is difficult to guess the cause of the slowness just based
on the posted video.
The CPU usage seem to be above 40%: it may be due to capturing
the screen. Does it remain so without any activity?
It takes some seconds until the shell prompt appears in opened
terminals: it cannot be directly related to the performance
of the framebuffer.
Moving terminals are not rendered fast: this may be related
to the performance of the framebuffer. Because the contents
of the windows need to be copied to the framebuffer after
each movement, that is not unexpected. However, tinyx or
Xfbdev may not be updating the framebuffer as efficiently
as possible.
Ali
Activity
rofl0r commentedon Sep 12, 2024
try
for CONFIG_DRM_I915_ALPHA_SUPPORT pass i915.alpha_support=1 on the kernel command line.
also in my notes on kernel config i have the following: "https://wiki.gentoo.org/wiki/Xorg/Configuration KMS and FB drivers conflict". the accompanying commit disables all CONFIG_FB_ items. see 86cba4d4b7de4cf240b8a2e71a93d4657cca4464 in sabotage linux. i recall that before disabling those i had some issues with FB, but i dont recall whether it was performance related.
rofl0r commentedon Sep 12, 2024
the config items i posted are 5.4 era. i cant help you anymore than what i stated above. try to disable all
_FB_items and enable the KMS ones. i hope you use ccache for the kernel build...clbr commentedon Sep 13, 2024
No problem with filing this issue here, this isn't exclusively TinyCore sw.
Your video has a lot of h264 corruption errors, so I'm not entirely sure what is from the corruption and what from the recording. But it seems to show that your shell took ~8s to show a prompt. While Xfbdev is certainly slow in comparison to Xvesa and accelerated drivers, you're right that this is greatly too slow, especially as you look like you're running on a more modern cpu.
I'm afraid I don't know what could be the main issue. The DRM fbdev emulation is considered terribly slow and bad in general, so that would be my first guess, but I think many TC users use the KMS FB with Xfbdev without such issues. Try profiling? Perf or oprofile's operf should do (with root rights and the options that enable full-system and kernel profiling).
clbr commentedon Sep 13, 2024
Oprofile's docs are quite good, haven't used perf much. In short, you first configure one or two sysctl options to enable wide profiling, then instead of "Xfbdev -args" you launch "operf -operfargs Xfbdev -args", run it for a bit, and using the tools then look at the results.
rofl0r commentedon Sep 13, 2024
maybe @aligrudi has some advice or tuned kernel config, as he's the author of some linux fb software.
i guess i just cant be bothered to try a new kernel and fix all the new issues that inevitably show up, when the one i use is still LTS-supported and works just fine.
aligrudi commentedon Sep 14, 2024