[oi-dev] MAJOR BREAKTHROUGH: On Solaris 11.0 and even Illumos it works now!!! __/__ Re[2]: [OpenIndiana-discuss] Sun/Oracle China's DRM//KMS Sol11.2 port backported to function on old-style gfxp_private from pre-2010 era but still immediatedly PANICS

[Мартин Бохниг]

 This only to let you know, I wasn't taking about vapoware.

Yes, I'm still spending most of my time with it.
Over the last week I had the problem, that everything appeared to work fine at first.

But after 10 minutes of browsing FireFox at some (but always different) point one would get a GPU error and the entire host would hang and needs to be rebooted.

This only happened if DRI2 was active, independently if using compiz or not.
Reducing the GPU clock in the system bios appeared to help a bit, and I'm not even sure if it is related to my backport of Sun's port or Sun's port (in this revision) itself.


So until today the situation was:

2D UXA accelerated Intel 2.99 ddx::: YES:***STABLE***
DRI2::: YES:BUGGY
COMPIZ::: YES:BUGGY(because of DRI2)
Return_to_TextConsole::: NO:Illumos's gfx_private too old (and more modern gfx_private's) calls deep into the kernel are simply not implemented in Illumos (to find this out alone I spent 3 days, info on functions and structs involved later with some RelNotes that still need to be written)

Now I played again with some dma attr finetuning.
And the DRI2/Compiz Gnome session is still happily running after 10 hrs of test-browsing the web and opening as many tabs as possible.
Let's hope this is really the result of the changed burst size. 
On the other hand: Before the change it would never have run for so long while FireFox was in use or unless DRI2 (and with it compiz and its special effects) would have been disabled in a xorg.conf to be created by the enduser.

If I can trust my eyes, then since today the new situation might hopefully be:


2D UXA accelerated Intel 2.99 ddx::: YES:***STABLE***
DRI2::: YES:stable_enough
COMPIZ::: YES:stable_enough
Return_to_TextConsole::: NO:Illumos's gfx_private too old (and more modern gfx_private's) calls deep into the kernel are simply not implemented in Illumos (to find this out alone I spent 3 days, info on functions and structs involved later with some RelNotes that still need to be written)

The burst size change also had one negative effect: Some invalid patterns somewhere, sometimes.


NEXT STEP: Over the next days I finally modernize  agpgart with the latest pciids and generation specific logic.
Also I found out how to update devfsadm to create the correct /dev/dri/controlD64 symlink (rather than /dev/dri/card0).

I won't be able to test it on other hardware (I only have 2nd gen Sandy Celeron G530 and now also 2700K which is also a Sandy).
We need many testers with more current hw after it is ready for testing (on newer hw, not only 2nd gen) in a few days.

I will create a demo screen video about OpenSXCE technologies meeting Openindiana/Hipster.
Then everybody can see how well compiz now works on internal Intel GPU's.
The main problem was that it would hang after 10 minutes. But without promising anything: At the _moment_ it looks like solved (hopefully).

No response required.
This was just an update FYI.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openindiana.org/pipermail/oi-dev/attachments/20151231/c56a462c/attachment.html>