Graphical Issues on Skylake with 14.2

D1ver · 04-12-2016, 07:37 PM

Grabbed today's upgrade to the intel driver. No issues with the xfce desktop but still some problems in libreoffice. Maybe the problem is with libreoffice?

lazardo · 04-12-2016, 11:43 PM

Quote:

Originally Posted by dmcrtr

...
I just installed xf86-video-intel-2.21.15-x86_64-1.txz from Slackware 14.1 on 14.2 and all the issues I was having went away.

Update: firefox and seamonkey scrolling improved by disabling smooth scroll but still far from "normal". libreoffice 5.1.2 has similar tuning hooks, including OpenCL, but none of these made any noticeable difference re: window refresh.

Update1:tried d167280, no noticeable differences for browser/libreoffice so back to a7526ea (current master). https://cgit.freedesktop.org/xorg/dr...6-video-intel/ is an interesting read

Update2: Added minimal 20-intel.conf (TearFree true, DRI false, AccelMethod sna), browsers behave as normal, libreoffice better but still not normal.

Tried 14.1 kit, X failed to init then hard locked whereas git_20160411_a7526ea is "usable" for firefox and libreoffice 5.1.2 (Alien Bob), eg, slow update/scroll/new window but no tearing. Seamonkey is marginally better than firefox.

skylake w HD530, 4.6-rc3 w intel_idle.max_cstate=6 i915.preliminary_hw_support=1 idle=nomwait pcie_aspm=force, Xorg 1.18.3, no xorg.conf (GTX 970M off in /etc/rc.local w bbswitch 0.8)

Latest git pull next. Cheers,

atelszewski · 04-13-2016, 08:03 AM

Hi,

With a7526ea I have artifacts/corruptions in at least Scilab, QTGZManager and Libreoffice Writer.
With d167280 the problem seems to be with Scilab only.

BTW, below is a rewritten script to fetch and pack the driver's source code.
It does not delete the already downloaded git, but operates on separate copy.

The desired branch (or whatever it is called) should be specified as the first parameter.
If it's empty, the latest source code is packed.

Code:

$ cat xf86-video-intel-mksrctarball.sh 
#!/bin/sh

set -e
umask 022

BRANCH=${1:-master}

if [ -e xf86-video-intel ]; then
  cd xf86-video-intel
  git checkout master
  git pull -f
  cd ..
else
  git clone git://git.freedesktop.org/git/xorg/driver/xf86-video-intel/
fi

cd xf86-video-intel
git checkout $BRANCH
VERSION=$( git log -1 --format=git_%ad_%h --date=format:%Y%m%d )
cd ..

rm -rf xf86-video-intel-${VERSION} || true
cp -a xf86-video-intel xf86-video-intel-${VERSION}
( cd xf86-video-intel-${VERSION} && find . -type d -name ".git*" -exec rm -rf {} \; 2>/dev/null ) || true
tar -Jcf xf86-video-intel-${VERSION}.tar.xz xf86-video-intel-${VERSION}
rm -rf xf86-video-intel-${VERSION}
echo "xf86-video-intel packaged as xf86-video-intel-${VERSION}.tar.xz"

--
Best regards,
Andrzej Telszewski

Panagiotis Nik · 04-14-2016, 09:34 AM

The last update in current make system to freeze under heavy use of Chromium:
GPU HANG: ecode 9:0:0x84dffff8, in Xorg [1706]

Staring Chromium with --disable-gpu seems temporary solve the problem...

lazardo · 04-14-2016, 07:27 PM

Finally a stable Skylake/hybrid (intel HD530 + nvidia) path.

For the record, this is a MSI GS60 6QE that had Skylake issues. The setup is now stable without lockup/libreoffice/browser/repaint issues. 15 Apr slackware64-current (or later) with these changes:

4.6.0-rc2 or higher from kernel.org. Note: I worked incrementally from 4.4.5 to 4.6-rc2 before things got stable. If you are tracking rc releases you will want the 'interdiff' utility from http://slackbuilds.org/repository/14...uti/patchutils. Successive rc patches are cumulative against the previous release (eg, 4.5), not incremental.
specific boot parameters
disabled nvidia https://github.com/Bumblebee-Project/bbswitch
BIOS settings: Speedstep, Hyper-threading, C-states all on, Legacy boot mode.
Minimal xorg.conf.d/20-intel.conf: AccelMethod=sna DRI=2 TearFree=true

The attachment is a patch against slackware64-current/testing/source/linux-4.6-rc1-configs/config-huge-4.6-rc1.x64, remember to walk thru it with xconfig before rebuilding the kernel as it is targeted at GS60 hardware (no initrd).

Update: https://bugzilla.kernel.org/show_bug.cgi?id=109081 points at major skylake boot/ACPI/cstate fixes, so for 4.6.0-rc4 the following boot params are no longer needed. They have also been queued for upstream so maybe one more kernel bump cycle should be considered, if my opinion counts

https://bugzilla.kernel.org/show_bug.cgi?id=109081#c92

The boot parameters are:

Code:

modprobe.blacklist=nouveau intel_idle.max_cstate=6 i915.preliminary_hw_support=1 idle=nomwait pcie_aspm=force

An /etc/rc.local entry for delayed nvidia shutdown:

Code:

( sleep 20; modprobe bbswitch load_state=0 unload_state=1 ) &

The above should get you started for similar bleeding edge Skylake hardware ala Dell XPS, Alienware, etc. Active nvidia graphics via http://docs.slackware.com/howtos:har...nvidia_optimus next up.

Cheers,

atelszewski.versades · 04-15-2016, 01:50 AM

Hi,

Maybe this Mesa update will bring something new to the table.

--
Best regards,
Andrzej Telszewski

lazardo · 04-15-2016, 04:48 PM

Fortunately I got the GS60 for photo editing so 3D is not high priority.

atelszewski · 04-15-2016, 05:27 PM

Hi,

I have problem with LibreOffice. Whenever I have graphics in Writer document, everything becomes dog slow.
I don't know if it is related to Intel driver, though.

EDIT:
Nope, it's not related to the Intel driver.

--
Best regards,
Andrzej Telszewski

atelszewski · 04-16-2016, 06:25 AM

Hi,

Checked version git_20160415_c3dc831 and results in artifacts.

--
Best regards,
Andrzej Telszewski

lonestar_italy · 04-16-2016, 06:41 AM

On my XPS 13 9350 (i7 with Iris HD 540), with Slackware-current, I use in /etc/X11/xorg.conf.d/20-intel.conf:

Code:

Section "Device"
        Identifier "Intel Graphics"
        Driver  "intel"
        Option  "AccelMethod" "uxa"
        Option  "TearFree" "true"
        Option  "SwapbuffersWait" "false"
EndSection

and in kernel boot parameters: i915.enable_rc6=0 pcie_aspm=force

No graphics issues nor artifacts.

atelszewski · 04-16-2016, 07:14 AM

Hi,

Sorry if I introduced misunderstanding.
I'm using version 20160224_d167280 that entered -current as of Fri Apr 15 20:37:37 UTC 2016 and it seems to be working fine. I just wanted to let you know that I also tested git_20160415_c3dc831 and it does not work well. I have simple test bed, qtgzmanager, that produces artifacts immediately if there's something wrong. I'm going to be testing latest commits and reporting the results here.

This is my /etc/X11/xorg.conf:

Code:

$ cat /etc/X11/xorg.conf
Section "Device"
  Identifier  "card0"
  Driver      "intel"
  Option      "AccelMethod"  "SNA"
  Option      "TearFree"     "True"
  Option      "DRI"          "3"
EndSection

I'm on SNA, since UXA is unacceptably slow in Senmonkey.

--
Best regards,
Andrzej Telszewski

atelszewski · 04-16-2016, 01:03 PM

Hi,

Could you guys try out git_20160416_f2a4645 ?
It seems to bring an important fix, but what is more important, qtgzmanager does not exhibit artifacts with this version.

And I was able to get the artifacts just by switching back and forth between git_20160416_f2a4645 and 20160415_c3dc831, literally by reinstalling the driver and restarting X.

--
Best regards,
Andrzej Telszewski

lazardo · 04-16-2016, 03:51 PM

f2a4645 up for 33 minutes w sna and DRI=3, can't see any difference from d167280.

Testing is basic: scrolling latency in seamonkey and firefox, window repaints in libreoffice. AlienBob's 5.1.2 -current libreoffice package, firefox and seamonkey are pulled from mozilla unmodified, 45.0.2 and 2.40 respectively.

I also use this check between boots:

Code:

vimdiff <( sed -n 's/\[ *[0-9].\....]//p' /var/log/Xorg.0.log ) <( sed -n 's/\[ *[0-9].\....]//p' /var/log/Xorg.0.log.old )

Interesting that DRI 3 reports 'SNA initialized with Skylake (gen9) backend' whereas other DRI settings report a 'generic' backend.

Also I am seeing artifacts editing this post in seamonkey when backspacing over text.

Update: Ran thru combinations of Tearfree, DRI '1 2 3 none' and all of them produce artifacts when editing linuxquestions posts. Retrying d167280.

Update 1: Sticking w f2a4645 as I'm seeing the linuxquestions backspace artifacts in both. If that's the only issue I'll just have to type more accurately.

Update 2: Bumped to 4.6-rc4 + f2a4546, no more artifacts.

Update last: rc4 unstable, back to rc3 + artifacts w linuxquestions, else ok.

Moving on to Bumblebee for the nvidia side. Thanks to all.

Cheers,

BratPit · 04-16-2016, 04:05 PM

One word

Quote:

Option "TearFree" "true"

is useless with

Quote:

Option "AccelMethod" "uxa"

it work with SNA

atelszewski · 04-16-2016, 04:49 PM

Hi,

Furthermore, f2a4645 is SNA specific.

--
Best regards,
Andrzej Telszewski