[OpenIndiana-discuss] Stable Intel DRM/KMS Sol12 backport diffs and test bins

Мартин Бохниг opensxce at mail.ru
Wed Aug 10 11:10:20 UTC 2016


 Fortunately I'm now permitted to post to Illumos-devel again (under strict moderation, but it works nicely), thanks to Rich Lowe and Illumos for that :)


Here extra infos which are related and which will one day be part of the ReleaseNotes, hence need to be posted also here (just in case anybody will ever want KMS for anything, of which I'm not convinced anymore)

https://www.listbox.com/member/archive/182179/2016/08/sort/time_rev/page/1/entry/3:98/20160809131319:8D2306CC-5E54-11E6-B788-F6CA7ACD9AB7/  




>Вторник, 9 августа 2016, 16:21 UTC от "Albert Lee" <trisk at omniti.com>:
>
>Hi Martin,
> 
>These changes provide a full KMS-based console too? That would be exciting.
> 
>-Albert

Hi Albert,

what do you mean with "KMS based console", vconsole?
You cannot exit X11 and return to a working text console with this on Sol11.0 or lower at the moment (including Illumos, while Sol11.0 is a special case in the middle, thanks to its half-way modernized gfx_private which got finished by Sun/Oracle only in 11.1 and hardly changed since then).
Same problem as on FreeBSD's latest KMS port bits (btw 2 years newer than what we have here, but more on that in the deferred long ReleaseNotes), without vbios support inside the kernel (unix) itself hooked into via gfx_private.
On 11.1++ the 100% identical kms bins do not have this issue.
But here on 11.0 and Illumos the same limitation also affects the use of this backport together with vconsole.
Restarting gdm is fine. Exiting X11 also works, stdin does function (only that you have a black screen until you blindly or via ssh restart Xorg).

Or are you referring to somebody else's effort to add VESA/VBE support to text mode?
If that's the case, then that's also in progress and his github is public, probably well-known - but not performed by me (I'm an Solaris/X11 blooded person since I started with OpenSolaris in 2005/2006). Text mode is nice but I personally don't need VBE  text mode for that so I never cared too much since it's not my focus.


But else:

Yes: You can run modern 2.21.x or 2.99.x intel_drv.so Xorg ddx's now, those are all kms-only since version 2.9 (dated 2010).
This brings you hw accelleration in Xorg and e.g. on Hipster or old OpenSXCE (which btw will get its long expected update asap hopefully next month) - on Hipster which is the best OpenSolaris distribution of all times by now (yes, I take my hat!), you can in vlc now watch raw .MTS files recorded in highest bandwidth 50P even on my main x86 machine (which still happens to be a slow Sandy Celeron G530 2.4GHz).
With the non-accellerated vesa_drv.so this was unimaginable

Also there is no need anymore to fiddle around with xorg.conf on the OI cd/dvd in single user mode and an explicit device line 
BusID       "PCI:0:2:0" to get Xorg up via vesa_drv.so, now autodetection works as designed by Xorg and no xorg.conf is required - Xorg will just start (that is, as soon as Hipster team adds these bits and brings out a new iso).


compiz should work mostly stable on most Intel 8xx, 9xx and internal GPU Ivy, Haswell based systems, no matter if the cpu is a Desktop, Mobile or Server variant (DRM/KMS internally distinguishes them and in a few files in the 915 subdir treats them differently). Sandy (the chips themselves) were known on the intel-drm mailing list to have Errata (reported 5 years ago on LinUX) and while on Sol11.1++ this isn't reproducable on my Celeron G530, on the backport you may run into GPU faults every now and then, especially triggered if you watch videos in FireFox  ***in fullscreen***. It took me since last November, but I got more stability also on Sandy and now the GPU will silently fail and reset the GPU. If it gets really bad than svcadm restart gdm shall be sufficient now, no reboot will be necessary on Sandy in most cases. Ivy and Haswell don't appear to show this specific instability anyway.

And as said - it wasn't easy to find out what triggered the problems on Sandy.
It turned out that avoiding to watch youtube videos (html or flash plugin doesn't make the difference here) in fullscreen, on Sandy.


Well, just test  ;)
I would politely ask users to test and report.
This will be useful for Hipster's official KMS page for this backport:   https://wiki.openindiana.org/oi/Intel+KMS+driver  






More information about the openindiana-discuss mailing list