cyblafb in kernel 2.6.16: corrupted display in Xfree86

cyblafb in kernel 2.6.16: corrupted display in Xfree86

[Thomas Arendsen Hein]

Hi!

The cyblafb driver workes fine for me up to Linux kernel 2.6.15.6,
but with 2.6.16 I get the following problem:

After booting the system, the framebuffer console comes up fine, but
when xdm starts up the X server, I get this diplay:
http://ftp.intevation.de/users/thomas/cyblafb/cyblafb-1.jpg

After switching to the console with Ctrl-Alt-F1 and back to X:
http://ftp.intevation.de/users/thomas/cyblafb/cyblafb-2.jpg

When I use Ctrl-Alt-Backspace to restart xdm everything looks fine:
http://ftp.intevation.de/users/thomas/cyblafb/cyblafb-3.jpg

I use Debian sarge on a Lex Light (CV860A board):
0000:01:00.0 VGA compatible controller: Trident Microsystems CyberBlade/i1 (rev 6a) (prog-if 00 [VGA])
        Subsystem: Trident Microsystems CyberBlade/i1
        Flags: bus master, 66MHz, medium devsel, latency 32, IRQ 5
        Memory at e4000000 (32-bit, non-prefetchable) [size=8M]
        Memory at e5000000 (32-bit, non-prefetchable) [size=128K]
        Memory at e4800000 (32-bit, non-prefetchable) [size=8M]
        Expansion ROM at 10000000 [disabled] [size=64K]
        Capabilities: <available only to root>


Resolution 1152x864 pixels (16 bpp) with fbdev driver in XF86Config-4.

/etc/fb.modes:
mode "1152x864-70"
    # D: 100.0 MHz, H: 66.138 kHz, V: 69.99 Hz
    geometry 1152 864 1152 864 8
    timings 10000 168 40 57 13 152 11
    hsync high
    vsync high
endmode

To enable this resolution I use:

/etc/rcS.d/S42cyblafb:
/sbin/modprobe cyblafb bpp=16 mode=1024x768 ref=85 && /usr/sbin/fbset -a 1152x864-70
(because the cyblafb driver doesn't understand mode=1152x864 as parameter)

XFree86 is standard Debian sarge: 4.3.0.dfsg.1-14sarge1

One further thing to note is that I compiled 2.6.15.6 with gcc
2.95.4 and 2.6.16 with gcc 3.4.3, can this make a difference?

Any idea what's happening here?

Regards,
Thomas

-- 
Email: thomas@intevation.de
http://intevation.de/~thomas/


-------------------------------------------------------
This SF.Net email is sponsored by xPML, a groundbreaking scripting language
that extends applications into web and mobile media. Attend the live webcast
and join the prime developer group breaking into this new coding territory!
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=121642

Re: cyblafb in kernel 2.6.16: corrupted display in Xfree86

[Knut Petersen]

Hi Thomas!

>/etc/fb.modes:
>mode "1152x864-70"
>    # D: 100.0 MHz, H: 66.138 kHz, V: 69.99 Hz
>    geometry 1152 864 1152 864 8
>  
>
That is very slow. Use geometry 1152 864 2048 4096 8

>    timings 10000 168 40 57 13 152 11
>    hsync high
>    vsync high
>endmode
>
>To enable this resolution I use:
>
>/etc/rcS.d/S42cyblafb:
>/sbin/modprobe cyblafb bpp=16 mode=1024x768 ref=85 && /usr/sbin/fbset -a 1152x864-70
>  
>

/sbin/modprobe cyblafb && /usr/sbin/fbset -a 1152x864-70 should be 
sufficient.

>(because the cyblafb driver doesn't understand mode=1152x864 as parameter)
>
>XFree86 is standard Debian sarge: 4.3.0.dfsg.1-14sarge1
>
>One further thing to note is that I compiled 2.6.15.6 with gcc
>2.95.4 and 2.6.16 with gcc 3.4.3, can this make a difference?
>
>Any idea what's happening here?
>
>Regards,
>Thomas
>
>  
>

I cannot test 1152x864 on my system because my monitor does not understand
that resolution. But if it worked previously it should work now ;-)

Are there any problems on the framebuffer console? You wrote that the fb 
console
comes up fine, but please confirm that scrolling is ok.

Are there any problems left after you reach the state of your third 
picture? Switching
between the fb console and X works flawlessly then?

Are there any problems if you use the standard resolutions, eg 1024x768 
or 1280x1024?

Try a fbset -vxres 2048 -vyres 4096  after your first fbset. Does that 
change anything?

Please send dmesg logs of kernel 2.6.15.6 and 2.6.16 after adding 
verbosity=2
to the modprobe of cyblafb - everything from the point of loading 
cyblafb to your third
picture. Also include the X server logs. Maybe it´s a good idea to send 
those logs to my
private address Knut_Petersen@t-online.de.

cu,
 Knut


-------------------------------------------------------
This SF.Net email is sponsored by xPML, a groundbreaking scripting language
that extends applications into web and mobile media. Attend the live webcast
and join the prime developer group breaking into this new coding territory!
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=121642

Re: cyblafb in kernel 2.6.16: corrupted display in Xfree86

[Thomas Arendsen Hein]

* Knut Petersen <Knut_Petersen@t-online.de> [20060323 09:44]:
> >/etc/fb.modes:
> >mode "1152x864-70"
> >   # D: 100.0 MHz, H: 66.138 kHz, V: 69.99 Hz
> >   geometry 1152 864 1152 864 8
>
> That is very slow. Use geometry 1152 864 2048 4096 8

This only works with kernel 2.6.16, not with 2.6.15.6

> >To enable this resolution I use:
> >
> >/etc/rcS.d/S42cyblafb:
> >/sbin/modprobe cyblafb bpp=16 mode=1024x768 ref=85 && /usr/sbin/fbset -a 
> >1152x864-70
> 
> /sbin/modprobe cyblafb && /usr/sbin/fbset -a 1152x864-70 should be 
> sufficient.

Yes, it is.

> >One further thing to note is that I compiled 2.6.15.6 with gcc
> >2.95.4 and 2.6.16 with gcc 3.4.3, can this make a difference?

I compiled 2.6.15.6 with gcc 3.4.3, too, just to make sure that this
makes no difference.

> I cannot test 1152x864 on my system because my monitor does not understand
> that resolution. But if it worked previously it should work now ;-)

Maybe you can test it with your monitor, too, see below.

> Are there any problems on the framebuffer console? You wrote that the fb 
> console
> comes up fine, but please confirm that scrolling is ok.

Scrolling in fbcon works perfectly, with or without xdm started.

> Are there any problems left after you reach the state of your third 
> picture?

No problems then.

> Switching between the fb console and X works flawlessly then?

Yes, no problem.

> Are there any problems if you use the standard resolutions, eg 1024x768 
> or 1280x1024?

No and yes:
Everything works fine when using
/sbin/modprobe cyblafb bpp=16 mode=1280x1024 ref=85
but I get the same problems if I do:
/sbin/modprobe cyblafb && /usr/sbin/fbset -a 1280x1024-85

> Try a fbset -vxres 2048 -vyres 4096  after your first fbset. Does that 
> change anything?

Nothing changed.

> Please send dmesg logs of kernel 2.6.15.6 and 2.6.16 after adding 
> verbosity=2
> to the modprobe of cyblafb - everything from the point of loading 
> cyblafb to your third
> picture. Also include the X server logs. Maybe it?s a good idea to send 
> those logs to my
> private address Knut_Petersen@t-online.de.

I'll do so.

Thomas

-- 
Email: thomas@intevation.de
http://intevation.de/~thomas/


-------------------------------------------------------
This SF.Net email is sponsored by xPML, a groundbreaking scripting language
that extends applications into web and mobile media. Attend the live webcast
and join the prime developer group breaking into this new coding territory!
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=121642

Re: Re: cyblafb in kernel 2.6.16: corrupted display in Xfree86

[Knut Petersen]

Thomas Arendsen Hein schrieb:

>>That is very slow. Use geometry 1152 864 2048 4096 8
>>    
>>
>
>This only works with kernel 2.6.16, not with 2.6.15.6
>
>  
>
ACK

>>Are there any problems if you use the standard resolutions, eg 1024x768 
>>or 1280x1024?
>>    
>>
>
>No and yes:
>Everything works fine when using
>/sbin/modprobe cyblafb bpp=16 mode=1280x1024 ref=85
>but I get the same problems if I do:
>/sbin/modprobe cyblafb && /usr/sbin/fbset -a 1280x1024-85
>
>  
>

Please Try:

/sbin/modprobe cyblafb && /usr/sbin/fbset -a 1280x1024-85 -depth 16

/sbin/modprobe cyblafb && /usr/sbin/fbset -g 1152 864 1152 864 8 -t 10000 168 40 57 13 152 11

/sbin/modprobe cyblafb && /usr/sbin/fbset -g 1152 864 1152 864 16 -t 10000 168 40 57 13 152 11

/sbin/modprobe cyblafb && /usr/sbin/fbset -g 1152 864 2048 4096 8 -t 10000 168 40 57 13 152 11

/sbin/modprobe cyblafb && /usr/sbin/fbset -g 1152 864 2048 2048 16 -t 10000 168 40 57 13 152 11


Is there a special reason not to use the X* trident driver?

cu,
 Knut



-------------------------------------------------------
This SF.Net email is sponsored by xPML, a groundbreaking scripting language
that extends applications into web and mobile media. Attend the live webcast
and join the prime developer group breaking into this new coding territory!
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=121642

Re: cyblafb in kernel 2.6.16: corrupted display in Xfree86

[Thomas Arendsen Hein]

* Knut Petersen <Knut_Petersen@t-online.de> [20060323 16:07]:
> Thomas Arendsen Hein schrieb:
> >>That is very slow. Use geometry 1152 864 2048 4096 8
> >
> >This only works with kernel 2.6.16, not with 2.6.15.6
>
> ACK

Wow, scrolling in the console is definitely _much_ faster with this:

time for "find /dev" on kernel 2.6.16:

       1152/864  2048/4096
 8bpp    18s       0.56s
16bpp    39s       0.65s

> >>Are there any problems if you use the standard resolutions, eg 1024x768 
> >>or 1280x1024?
> >
> >No and yes:
> >Everything works fine when using
> >/sbin/modprobe cyblafb bpp=16 mode=1280x1024 ref=85
> >but I get the same problems if I do:
> >/sbin/modprobe cyblafb && /usr/sbin/fbset -a 1280x1024-85
> 
> Please Try:
> 
> /sbin/modprobe cyblafb && /usr/sbin/fbset -a 1280x1024-85 -depth 16

works

> /sbin/modprobe cyblafb && /usr/sbin/fbset -g 1152 864 1152 864 8 -t 10000 
> 168 40 57 13 152 11

doesn't work

> /sbin/modprobe cyblafb && /usr/sbin/fbset -g 1152 864 1152 864 16 -t 10000 
> 168 40 57 13 152 11

works

> /sbin/modprobe cyblafb && /usr/sbin/fbset -g 1152 864 2048 4096 8 -t 10000 
> 168 40 57 13 152 11

doesn't work

> /sbin/modprobe cyblafb && /usr/sbin/fbset -g 1152 864 2048 2048 16 -t 10000 
> 168 40 57 13 152 11

works

With /sbin/modprobe cyblafb && /usr/sbin/fbset -a 1152x864-70 -depth 16
(changing fb.modes would probably help, too) everything is back to
normal for me.

I think the extra colors on console won't hurt, now that I know
about the 2048/4096 setting.

> Is there a special reason not to use the X* trident driver?

Yes, I had crashes when using more than one X display (e.g. one
local xdm and one "X -query other :1"), and switching between X and
console is _much_ faster this way.

Many thanks for the help (and for cyblafb in general),
Thomas

-- 
Email: thomas@intevation.de
http://intevation.de/~thomas/


-------------------------------------------------------
This SF.Net email is sponsored by xPML, a groundbreaking scripting language
that extends applications into web and mobile media. Attend the live webcast
and join the prime developer group breaking into this new coding territory!
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=121642

Re: Re: cyblafb in kernel 2.6.16: corrupted display in Xfree86

[Knut Petersen]

Thomas Arendsen Hein schrieb:

>>/sbin/modprobe cyblafb && /usr/sbin/fbset -a 1280x1024-85 -depth 16
>>    
>>
>
>works
>
>  
>
>>/sbin/modprobe cyblafb && /usr/sbin/fbset -g 1152 864 1152 864 8 -t 10000 
>>168 40 57 13 152 11
>>    
>>
>
>doesn't work
>
>  
>
>>/sbin/modprobe cyblafb && /usr/sbin/fbset -g 1152 864 1152 864 16 -t 10000 
>>168 40 57 13 152 11
>>    
>>
>
>works
>
>  
>
>>/sbin/modprobe cyblafb && /usr/sbin/fbset -g 1152 864 2048 4096 8 -t 10000 
>>168 40 57 13 152 11
>>    
>>
>
>doesn't work
>
>  
>
>>/sbin/modprobe cyblafb && /usr/sbin/fbset -g 1152 864 2048 2048 16 -t 10000 
>>168 40 57 13 152 11
>>    
>>
>
>works
>
>With /sbin/modprobe cyblafb && /usr/sbin/fbset -a 1152x864-70 -depth 16
>(changing fb.modes would probably help, too) everything is back to
>normal for me.
>
>I think the extra colors on console won't hurt, now that I know
>about the 2048/4096 setting.
>
>  
>
So the problem is that the 8bpp modes do not work with the
X framebuffer driver. Please report that bug, I think X is responsible.

>>Is there a special reason not to use the X* trident driver?
>>    
>>
>
>Yes, I had crashes when using more than one X display (e.g. one
>local xdm and one "X -query other :1"), 
>
Try it again, the recent cyblafb works around X and xdm bugs
that could have caused that crashes.

>and switching between X and
>console is _much_ faster this way.
>
>  
>
I doubt that ... switching between X trident and cyblafb and back again
takes only a few miliseconds here.

cu,
 knut


-------------------------------------------------------
This SF.Net email is sponsored by xPML, a groundbreaking scripting language
that extends applications into web and mobile media. Attend the live webcast
and join the prime developer group breaking into this new coding territory!
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=121642

Re: cyblafb in kernel 2.6.16: corrupted display in Xfree86

[Thomas Arendsen Hein]

* Knut Petersen <Knut_Petersen@t-online.de> [20060323 18:07]:
> Thomas Arendsen Hein schrieb:
> 
> >With /sbin/modprobe cyblafb && /usr/sbin/fbset -a 1152x864-70 -depth 16
> >(changing fb.modes would probably help, too) everything is back to
> >normal for me.
>
> So the problem is that the 8bpp modes do not work with the
> X framebuffer driver. Please report that bug, I think X is responsible.

But why did it work up to 2.6.15.6 then?
I know for sure that I had 8bpp in fb console and 16bpp in X.

> >>Is there a special reason not to use the X* trident driver?
> >
> >Yes, I had crashes when using more than one X display (e.g. one
> >local xdm and one "X -query other :1"), 
>
> Try it again, the recent cyblafb works around X and xdm bugs
> that could have caused that crashes.

No, I had crashes with non-fb console and trident X driver.

I never tested if cyblafb console and trident X have problems.

> >and switching between X and
> >console is _much_ faster this way.
>
> I doubt that ... switching between X trident and cyblafb and back again
> takes only a few miliseconds here.

Maybe I had little variations in the monitor timings which caused
the monitor to re-sync. With X on cyblafb this is obviously always
identical.


I've found another problem with the new cyblafb under X which didn't
happen before: DPMS no longer turns of the monitor. At the time
where DPMS suspend kicked in before, now the xlock clock simply
stops updating, but is still visible (xlock -mode dclock).

Thomas

-- 
Email: thomas@intevation.de
http://intevation.de/~thomas/


-------------------------------------------------------
This SF.Net email is sponsored by xPML, a groundbreaking scripting language
that extends applications into web and mobile media. Attend the live webcast
and join the prime developer group breaking into this new coding territory!
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=121642

Re: cyblafb in kernel 2.6.16: corrupted display in Xfree86

[Thomas Arendsen Hein]

* Thomas Arendsen Hein <thomas@intevation.de> [20060323 19:40]:
> I've found another problem with the new cyblafb under X which didn't
> happen before: DPMS no longer turns of the monitor. At the time
> where DPMS suspend kicked in before, now the xlock clock simply
> stops updating, but is still visible (xlock -mode dclock).

... and DPMS on the fb console still works fine.

Thomas

-- 
Email: thomas@intevation.de
http://intevation.de/~thomas/


-------------------------------------------------------
This SF.Net email is sponsored by xPML, a groundbreaking scripting language
that extends applications into web and mobile media. Attend the live webcast
and join the prime developer group breaking into this new coding territory!
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=121642

Re: Re: cyblafb in kernel 2.6.16: corrupted display in Xfree86

[Antonino A. Daplas]

Thomas Arendsen Hein wrote:
> * Thomas Arendsen Hein <thomas@intevation.de> [20060323 19:40]:
>> I've found another problem with the new cyblafb under X which didn't
>> happen before: DPMS no longer turns of the monitor. At the time
>> where DPMS suspend kicked in before, now the xlock clock simply
>> stops updating, but is still visible (xlock -mode dclock).
> 
> ... and DPMS on the fb console still works fine.
> 
> Thomas
> 


Try adding 

printk("fb_blank: blank value %i\n", blank);

in drivers/video/fbmem.c:fb_blank(struct fb_info *info, int blank)

just before the call to info->fbops->fb_blank().

Try blanking your screen in X, look at dmesg and let us know what
is the blank value used by X.

Tony


-------------------------------------------------------
This SF.Net email is sponsored by xPML, a groundbreaking scripting language
that extends applications into web and mobile media. Attend the live webcast
and join the prime developer group breaking into this new coding territory!
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=121642

Re: Re: cyblafb in kernel 2.6.16: corrupted display in Xfree86

[Knut Petersen]

Thomas Arendsen Hein schrieb:

>* Knut Petersen <Knut_Petersen@t-online.de> [20060323 18:07]:
>  
>
>>Thomas Arendsen Hein schrieb:
>>
>>    
>>
>>>With /sbin/modprobe cyblafb && /usr/sbin/fbset -a 1152x864-70 -depth 16
>>>(changing fb.modes would probably help, too) everything is back to
>>>normal for me.
>>>      
>>>
>>So the problem is that the 8bpp modes do not work with the
>>X framebuffer driver. Please report that bug, I think X is responsible.
>>    
>>
>
>But why did it work up to 2.6.15.6 then?
>I know for sure that I had 8bpp in fb console and 16bpp in X.
>
>  
>
Well, cyblafb does not use vga style panning registers any longer.
That change is necessary to support ywrap scrolling and to allow
screen rotation. The trident X driver does not know about the
graphics engine registers I use instead, and I had to include
some hacks to allow that to work.

Try to change

    #define CYBLAFB_KD_GRAPHICS_QUIRK 1

to

    #define CYBLAFB_KD_GRAPHICS_QUIRK 0

in cyblafb.c for purely diagnostic purposes. Does that restore the old
behaviour for you?


cu,
 Knut


-------------------------------------------------------
This SF.Net email is sponsored by xPML, a groundbreaking scripting language
that extends applications into web and mobile media. Attend the live webcast
and join the prime developer group breaking into this new coding territory!
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=121642