Re: [Linux-fbdev-users] matroxfb, 2.6 and 2.4

[Petr V. <VAN...@vc...>]

On  3 Aug 04 at 2:01, Jonas Meurer wrote:
> sorry? i guess they belong to the same bug, but i'm no kernel developer.

I did not found any bug descriptions in these articles. They are just
vaguely saying that 2.6.x is crap, without any explanation what
should be changed and why.
 
> i'll try to get an oops trace and some further debugging information.

It would help a lot.
 
> > IRQ support is in both stock kernel and in my patches. Other two
> > patches are not going to the driver.
> 
> so you mean that you patch doesn't fix the bug? on irc in #kernelnewbies
> on irc.kernelnewbies.org i got the information, that your backport fixes
> at least this bug behaviour. but it might have been a liar, don't know.
> he also said, that matroxfb in 2.6 is very unstable while being
> distributed as stable. in 2.4 it was tagged as experimental.

In 2.4.x it was marked as EXPERIMENTAL only because nobody cared.
As driver is more than 5 years old, I think that I can say that it
is well tested now.

I'm not aware about any problems with matroxfb in stock 2.6.x kernel.
There are four problems with matroxfb driver in 2.6.x kernel:

(1) 2.6.x fbdev does not allow for native text mode; thus matroxfb
    in 2.6.x kernel does not offer text mode (fbset ... -depth 0).

(2) 2.6.x fbdev does not store fb mode per-VT, but only one globally.
    So you may have troubles with setting each VT to different resolution.
    
(3) 2.6.x fbcon's scrollback (and 2.4.x too, for that matter) crashes
    if you have more than one visible console, and you did not
    disable scrollback (recent 2.6.x "fixed" it indirectly by
    supporting only one visible console).

(4) mga-dri reprograms accelerator from time to time even if X are not
    visible. This causes massive screen corruption if display resolution
    or color depth you use in X is different from one you use on the
    console, and occassional screen corruption if resolutions match.
    
Neither of these bugs are in the matroxfb - they are features of fbdev/fbcon
layer.

Only difference between matroxfb from platan.vc.cvut.cz, and one in
the kernel is that it supports text mode (-depth 0), and that some
PINS values are interpreted differently. Neither of these changes
can cause/fix crashes/lockups/... 

But matroxfb from platan.vc.cvut.cz also comes with different fbcon, 
which is more simillar to 2.4.x than to what 2.6.x offers (and which
fixes problems (2) and (3) above; (4) above is not kernel problem), 
and while this may fix some other bugs I do not know about, it may 
also introduce some other bugs you did not find yet.

But as I said, I'm not aware about any bugs in the matroxfb (last
famous words).
                                                    Best regards,
                                                        Petr Vandrovec