Bug 16833

Created attachment 17852 [details]
Xorg log with modedebug true

We can easily test any new code to help resolve the issue. Let me know how we can help.

fedora 8 use xorg 7.2, should upgrade first , to xorg 7.3 with xserver-1.4 

Why do I need to update to server 1.4? Is the bug fixed in 1.4?

That is not an easy update for me to do. My understanding is that
the driver is supposed to work against 1.3.

I usually use XAA on my i945GM but I'm playing with EXA... I'm seeing the expected issues (slow scrolling in FF etc.) but I also see:

(II) intel(0): EDID vendor "LPL", prod id 0
(II) intel(0): DDCModeFromDetailedTiming: 1680x1050 Warning: We only handle seperate sync.
(EE) intel(0): underrun on pipe B!
(II) intel(0): EDID vendor "LPL", prod id 0
(II) intel(0): DDCModeFromDetailedTiming: 1680x1050 Warning: We only handle seperate sync.
(II) intel(0): Printing DDC gathered Modelines:
(II) intel(0): Modeline "1680x1050"x0.0  122.00  1680 1712 1776 1904  1050 1051 1054 1066 -hsync -vsync
 (64.1 kHz)
(II) intel(0): Modeline "1680x1050"x0.0  122.00  1680 1712 1776 1904  1050 1051 1054 1066 -hsync -vsync
 (64.1 kHz)


On EXA I don't seem to get the underrun but the other output is the same. It seems to be related to a visible flicker in the display and I'm currently trying to track it down (not sure if it's a regression in the driver or something else (but I will refrain from spamming this bug about it).

Barry, besides the (EE) messages, are you suffering real impact?

Could you try xf86-video-intel 2.4.0 release or master tip? There's a major fix for underrun pipe issue recently.

The impact we're seeing is that the screen goes completely black. Syncs are still being generated, but no video data appears on the VGA output.

Using imlib2_grab to read back the X root window shows that the display is still being drawn.

(In reply to comment #6)
> Barry, besides the (EE) messages, are you suffering real impact?
> 
> Could you try xf86-video-intel 2.4.0 release or master tip? There's a major fix
> for underrun pipe issue recently.
> 

As you can see from the log we are running 2.4.0 release. Have you fixed
underrun issue since 2.4.0 release?

As Simon says we have major problem as the screen is black.

Barry

(In reply to comment #8)
> As you can see from the log we are running 2.4.0 release. Have you fixed
> underrun issue since 2.4.0 release?

No, I meant the fix in 2.4.0.
Then I'll ask Jesse to look into this new issue.

(In reply to comment #4)
> Why do I need to update to server 1.4? Is the bug fixed in 1.4?
> 

I don't know but have may bug fixes .

what version of intel drive  have break yours computer ? 

(In reply to comment #10)
> (In reply to comment #4)
> > Why do I need to update to server 1.4? Is the bug fixed in 1.4?
> > 
> 
> I don't know but have may bug fixes .
> 
> what version of intel drive  have break yours computer ? 
> 

None of the versions we've tried work correctly - given that this appears to be a display pipe underrun, and upgrading X is a serious hassle for us, we'd like to wait until it's confirmed that the X server is at fault.

We've tried 2.3.0 and 2.4.0; if there's a particular version we should try instead, we can do that.

Created attachment 17931 [details] [review]
program FIFO watermark values

Looks like the watermark values on your platform may be incorrect.  Do the underruns still happen with this patch applied?  What about if you run at a lower resolution or pixel clock?

(In reply to comment #11)
> (In reply to comment #10)
> > (In reply to comment #4)
> > > Why do I need to update to server 1.4? Is the bug fixed in 1.4?
> > > 
> > 
> > I don't know but have may bug fixes .
> > 
> > what version of intel drive  have break yours computer ? 
> > 
> 
> None of the versions we've tried work correctly 

So this is not about one regression :)  

> - given that this appears to be a display pipe underrun, and upgrading X is a > serious hassle for us, we'd like
> to wait until it's confirmed that the X server is at fault.

I put here: http://sergiomb.no-ip.org/specs/xstuff/?C=M;O=D
my specs for build Xorg 7.3.1 on F8 .
you don't need change libdrm .
I, usually, use source from gits but should work with released sources on 
http://xorg.freedesktop.org/releases/individual/driver/


> 
> We've tried 2.3.0 and 2.4.0; if there's a particular version we should try
> instead, we can do that.
> 

well , my experience with my i915GM, xserver 1.4 do the difference , 
else you should use the old i810-1.6.5 version . Since Intel 2.0 is not a good idea, use xserver 1.3 . 

(In reply to comment #12)
> Created an attachment (id=17931) [details]
> program FIFO watermark values
> 
> Looks like the watermark values on your platform may be incorrect.  Do the
> underruns still happen with this patch applied?  What about if you run at a
> lower resolution or pixel clock?
> 
It's not consistent - sometimes it's fine, sometimes it's not. Reducing to 640x480@60Hz for testing doesn't fix it.

Changing resolution doesn't appear to affect it on unpatched 2.4.0; we still get the blank screen problem regularly. The patch doesn't appear to solve it - I'll attach the log with ModeDebug 1.

Also, should I switch to the old i810 driver as recommended by Sergio? If not, will a new X server resolve the problem on my GME965?

Created attachment 17955 [details]
X log with Jesse's patch applied

The old driver may or may not work for you, but it won't support dynamic output discovery & reconfiguration.

I don't see how a new X server would help you here, but I may be overlooking some interaction...

Fundamentally, the underrun messages indicate that the pipe in question wasn't able to get the data it needed in time to send it out to the display.  Usually, this is because the FIFO size is too small for the programmed display mode, or the FIFO watermarks are set too low, which would cause the pipe to refill its FIFO too late to serve the current request.  I'll have to check if there's anything special we're missing for 965GME in this area.

Created attachment 18003 [details] [review]
only update fifo watermark

Ok, let's try leaving DSPARB alone and just fixing up the FIFO watermark...

(In reply to comment #16)

> I don't see how a new X server would help you here, but I may be overlooking
> some interaction...
> 

Simply, because xserver 1.3 is too buggy , that what I take for my experience. 

Created attachment 18004 [details]
Log with the patch in attachment 18003 [details] [review] applied

The "only update fifo watermark" patch doesn't fix it, either; I've attached the X log with ModeDebug 1 again, in case it shows something interesting.

http://people.freedesktop.org/~jcristau/i830_dsparb_sooner.diff
fixes the problem for me.

I tested 360 reboots without a failure.

Well, that's a nice bonus!  Thanks for testing Barry.