Re: [CONFIG_FB_NVIDIA] Mirrored characters on the console on a G5

[Francois Taiani <f.taiani@computer.org>]

Hi Tony,

Sory for my *very* late reply. I got caught in too much work. Anyway I 
have looked at the "mirrored characters" issue in more details, and the 
conclusion is that commenting out 'reverse_order()' in 
drivers/video/nvidia/nv_accel.c solves the problem (in linux-2.6.12-rc2, 
I haven't try with more recent versions).

I had wrongly compile my first kernel with FB_RIBA on instead FB_NVIDIA 
(stupid me), hence my weird results.

A difference between FB_RIVA and FB_NVIDIA on my G5 (GeForce FX 5200 
Nvidia card) is that the console on FB_NVIDIA has far bigger characters 
(and hence less lines and columns) that with FB_RIVA. I have not counted 
them, but I'd say the RIVA console has 80 lines with the NVIDIA one has 
only 40. This is not really a problem though.

Thanks for your support, and sorry for the high latency!

Francois


 >> I've compiled a new kernel with those lines commented out, and this
 >>  solves the console issue: no more mirrored characters. However the
 >> X display is now garbled in the same way as with the ribafb driver:
 >> it looks like some pixel columns are not used (fine vertical black
 >> lines repeated in a regular pattern (every 4 screen pixel ?)), and
 >> the whole screen is stretched horizontally as if an "X pixel" was
 >> represented as a 3x1 rectangle. If I move my mouse to the right
 >> border is re-appears on the left border and as I move it it
 >> overwrites the previous pixels there with what should be beyond the
 >> right border (as if multiple 'X pixels' were mapped to the same
 >> location on the screen in a kind of modulo operation). I can go
 >> through three and half whole horizontal screens like that.
 >>
 >> I guess this is quite fuzzy, but I don't know how to explain it
 >> better. It looks like reverse_order is not needed for in console
 >> mode but is needed in graphic mode. Could it make sense?
 >
 >
 > No, it does not make sense to me how reverse_order() can affect the
 > output of X.
 >
 > Can you verify that it is really reverse_order() that affects X's
 > output and not if accel is on or off?
 >




Antonino A. Daplas wrote:
> On Thursday 21 April 2005 01:21, Francois wrote:
> 
>> Hi Tony,
>> 
>> 
>>> On Fri, 2005-04-15 at 11:07 +0800, Antonino A. Daplas wrote:
>>> 
>>>>> If it does, this is a problem with nvidia's imageblit
>>>>> function which converts big-endian bitmaps to little endian?
>>>>> Perhaps the help reverse_order() is not needed for your arch.
>>>>> Can you comment out calls to reverse_order() in
>>>>> drivers/video/nvidia/nv_accel.c?
>>>> 
>>>> I'll try this and I'll get back to you this the results.
>>> 
>>> I'll wait, thanks.
>> 
>> I've compiled a new kernel with those lines commented out, and this
>>  solves the console issue: no more mirrored characters. However the
>> X display is now garbled in the same way as with the ribafb driver:
>> it looks like some pixel columns are not used (fine vertical black
>> lines repeated in a regular pattern (every 4 screen pixel ?)), and
>> the whole screen is stretched horizontally as if an "X pixel" was
>> represented as a 3x1 rectangle. If I move my mouse to the right
>> border is re-appears on the left border and as I move it it
>> overwrites the previous pixels there with what should be beyond the
>> right border (as if multiple 'X pixels' were mapped to the same
>> location on the screen in a kind of modulo operation). I can go
>> through three and half whole horizontal screens like that.
>> 
>> I guess this is quite fuzzy, but I don't know how to explain it
>> better. It looks like reverse_order is not needed for in console
>> mode but is needed in graphic mode. Could it make sense?
> 
> 
> No, it does not make sense to me how reverse_order() can affect the
> output of X.
> 
> Can you verify that it is really reverse_order() that affects X's
> output and not if accel is on or off?
> 
> Tony
> 
>> Cheers
>> 
>> Francois
>> 
>> 
>>> On Fri, 2005-04-15 at 11:07 +0800, Antonino A. Daplas wrote:
>>> 
>>> On Friday 15 April 2005 00:52, Francois wrote:
>>> 
>>>> Hi Tony,
>>>> 
>>>> 
>>>>> Only characters are mirrored, and not the entire row (ie, row
>>>>> starts at the left side, and not at the right)?
>>>> 
>>>> Only the characters (so I can't look in a mirror and have a
>>>> "normal" screen
>>>> 
>>>> :-). :
>>>> 
>>>>> Can you try fbset -accel false and see if it helps?
>>>> 
>>>> Right on the spot! I can now switch between mirrored and not
>>>> mirrored.
>>> 
>>> You can also append this when you boot:
>>> 
>>> video=nvidiafb:noaccel
>>> 
>>> so you don't have to do an fbset -accel false each time.
>>> 
>>> 
>>>>> I think rivafb supports this particular chipset, but without 
>>>>> acceleration.
>>>> 
>>>> Actually my X display is garbled with rivafb (with which
>>>> console characters are fine). With fb_nvidia, X seems to work
>>>> fine independently of whether the acceleration is switched on
>>>> or off (some screen size and frequency issues needs tuning
>>>> though).
>>>> 
>>>> 
>>>>> If it does, this is a problem with nvidia's imageblit
>>>>> function which converts big-endian bitmaps to little endian?
>>>>> Perhaps the help reverse_order() is not needed for your arch.
>>>>> Can you comment out calls to reverse_order() in
>>>>> drivers/video/nvidia/nv_accel.c?
>>>> 
>>>> I'll try this and I'll get back to you this the results.
>>> 
>>> I'll wait, thanks.
>>> 
>>> Tony
> 
> 



-------------------------------------------------------
SF.Net email is Sponsored by the Better Software Conference & EXPO
September 19-22, 2005 * San Francisco, CA * Development Lifecycle Practices
Agile & Plan-Driven Development * Managing Projects & Teams * Testing & QA
Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf