What makes a production kernel?

[Eric Hattemer]

Andrew Farris wrote:

>
>In my opinion, just because the driver is closed source does not mean
>there can be no improvement in how nVIDIA and linux developers work
>together... and there should be improvement (in both directions).
>Nevertheless, cutting edge changes cannot be made without breaking the
>old, unchanged drivers -- the changes will not be made pro-actively by
>the vendors because making those changes costs them money.  They play
>catchup, and probably always will (at least until their support of Linux
>becomes a legitimate economic expense in comparison to support of the
>competing OS).
>  
>
>  
>
This is the most solid point to come out of this discussion.  I 
understand that NVIDIA can't patch their drivers for a kernel that isn't 
being used.  I don't know whether its practical, but a better solution 
would be to let NVIDIA know that their drivers will not work on the 
kernels of the future, but stick to the working kernels in the mean 
time.  To alienate the huge percentage of the PC market who have NVIDIA 
hardware for six months for kernel options that may not make a noticable 
performance difference doesn't seem like a popular way to go.  I haven't 
compiled my own kernel in a while, but isn't the problem not the kernel 
itself, but the options and extra patches that are applied? 

>It is interesting to consider the same situation with the release of
>Windows XP... where the nVIDIA driver base for Windows 2000 was
>extremely unstable for use in XP for several months after the public
>release.  Did Microsoft not release XP, nope.
>(there was probably more unilateral cooperation in that situation)
>  
>
There are a few differences between that situation and this one, 
though.  First, the 2000 drivers worked decently well, could accelerate 
video, but might crash on some complex game.  The nvidia X driver 
hardlocks on the newer Red Hat kernels, and the nv driver isn't fast 
enough to play video without skipping.  And you're extremely correct in 
the unilateral cooperation.  NVIDIA doesn't care too much about Fedora 
and doesn't need to.  Linux does not have the leverage to say, "We're 
doing this our way, now fix your stuff".  The only way Linux could gain 
that leverage would be to gain popular support by working with the 
systems the populous have. 

I know that there are some people who live in a 1960's UNIXy cave and 
say that things were better when everything was command line.  But a 
large part of the population wants a fancy colorful system that works 
automatically and plays their video and games.  Linux in general has 
made phenominal progress toward that goal in the last 5 years without 
losing sight of their hardcore UNIX base.  But its always seemed like 
all the other distros have been leading the way.  I know Red Hat is now 
focusing on their RHEL base, so they're allowed to care a little less 
about popularizing Linux for the home crowd, but otherwise what's the 
point of Fedora? 

Lastly, this isn't just a NVIDIA problem.  I brought this up because 
neither of my graphics cards work.  People have been talking about 
random deadlocks and other strange behavior.  I don't know whether the 
savage driver is strictly the problem, or something else.  If this 
problem isn't well understood by now, I will try to research where the 
problem is with my own 2.6.5 kernels.  But the actual reason I asked 
this was not to incite a pro NVIDIA vs. anti NVIDIA war, but to 
understand where the kernel is going.  It seems like as FC2 comes 
closer, the other software freezes and becomes more stable, but the 
kernels keep going and become increasingly unstable and incompatible.  
That's why I wanted to know if FC2 would release with a stable 2.6.3 
with a choice to upgrade to some super modified 2.6.5, or whether 2.6.5 
was the only way to go. 

-Eric Hattemer