Re: Firefox and the broken about:login, a temporary workaround

[maya%NetBSD.org@localhost]

On Mon, Apr 13, 2020 at 03:54:56PM +0200, Rhialto wrote:
> from Xorg.0.log:
> 
> [    27.951] (**) modeset(0): Option "AccelMethod" "XAA"
> 
> [    28.367] (II) Initializing extension GLX
> [    28.391] (II) AIGLX: Screen 0 is not DRI2 capable
> [    35.846] (II) IGLX: Loaded and initialized swrast
> [    35.846] (II) GLX: Initialized DRISWRAST GL provider for screen 0
> [    35.846] (II) Initializing extension XFree86-VidModeExtension
> [    35.847] (II) Initializing extension XFree86-DGA
> [    35.848] (II) Initializing extension XFree86-DRI
> [    35.868] (II) Initializing extension DRI2
> 
> Note that I used to have XVideo support in the past, but no longer:
> GL screensavers are also falling back to being slow (I usually test this
> with the screensaver /usr/pkg/libexec/xscreensaver/carousel -fps which
> is much slower (~20 fps) than it could be (~50 fps), with much higher
> cpu usage).
> 
> vargaz:/var/log$ xvinfo 
> X-Video Extension version 2.2
> screen #0
>  no adaptors present
> 
> Fortunately my other machine with Radeon graphics still has xv, and
> mpv also works.
> 

So, for clarification, xf86-video-modesetting is supposed to be much
better if glamor is enabled. With base xsrc we have some bug that we
haven't found yet, but I don't remember, it might work fine with pkgsrc
xorg.

Note that pkgsrc xorg has its own problem: x11/xf86-video-intel points
at the latest release, which is (???) outdated and doesn't work for
newer cards. NetBSD xsrc uses a git snapshot, and we should do the same
or even rip it out for modesetting if that one has glamor.

Certainly people have mentioned fewer graphical glitches with it, but I
feel like switching to modesetting is papering over the issue. My
current theory for why the graphical glitches happen is that memory
which is mapped write-combining doesn't get synced often enough, or
perhaps we made an error and it should have been mapped write-back.

For modesetting with xsrc, I feel like that libdrm-style drmGetBusid code
in the modesetting driver might need some adaptations that we had to do
in libdrm.