Kernel Stack Sizes

[Steve Brenneis]

On Thu, 2004-06-17 at 09:47, Mike Hearn wrote:
> On Wed, 16 Jun 2004 23:13:13 -0400, Steve Brenneis wrote:
> > Another point: Linux has made great strides with the hardware vendors. A
> > few short years ago, a company like nVidia would never have even
> > considered producing a Linux driver, binary or otherwise. Does it make
> > sense now to get up in their faces and say, in essence, "you will now do
> > things to suit us?" Remember, they have little or no incentive to do
> > anything for Linux.
> 
> That's incorrect, nVidia have huge incentive to produce drivers for Linux.
> The large commercial graphics houses demand it - or what, did you think
> nVidia made Linux drivers because they are fluffy and nice and sweet?
> 

Not necessarily. I think they made the drivers to test the waters. Their
production of Linux drivers and their assignment of resources to handle
maintenance and upgrades will be dependent on the return on their
investment. That is simple economics. Community versions of Linux
distributions tend to have very short life-cycles. No hardware
manufacturer is going to devote resources to keeping up with these
life-cycles unless the users of these versions represent a reasonable
percentage of their market.

If there was such a huge demand, ATI and Broadcom would both have
developed at least binary drivers for Linux. Both steadfastly refuse to
do so. In fact, the 2.6 kernel isn't exactly new. If the demand existed,
it is likely that nVidia would have already developed a new driver that
lives within the 4K stack limitation.

> This is a classic case of the problems proprietary software
> can cause. Everything is peaches until you need the software to do
> something it doesn't currently do, then you're stuck. I say this as
> somebody who uses the nVidia drivers on FC1 today by the way, but I'm
> still intending to upgrade. I'll use the open source (sort of, iirc they
> are obfuscate) drivers until nVidia release an update.
> 

Good luck, the open source drivers are completely non-functional on the
geForce 4 family. If you boot to run level 5, you will see the red
screen of death somewhere around the time your root file system is being
mounted. If you boot to run level 3 and run startx, your screen will go
blank and your system will happily wait until stars burn out for you to
hit the power button.

I use the open source Radeon drivers on another system and they are
barely functional. I don't think it will be possible for anyone to
satisfactorily reverse-engineer some of these devices. At least not
without risking the ire (and the lawyers) of the manufacturer. That
being said, I would be happy to help if anyone is making the effort to
produce high quality drivers for some of the devices that are stubbornly
Windows-centric. Lawyers don't scare me.

> Unfortunately there are no good answers to this problem ... so far
> nobody has come up with a convincing business strategy that lets nVidia
> open source their drivers, and the kernel developers won't give binary
> driver developers any breaks. Rock, meet hard place.
> 

Maybe the HP model is the one to suggest to companies like nVidia. They
assigned some development resources to create the hpij and hpoj drivers
and then handed it over to the community. That way they get the benefit
of the ever-growing number of Linux users but they don't have to expend
the resources to maintain the drivers for every new flavor of
distribution. I think the relationship of HP to Debian for hppa-linux
was similar.

> thanks -mike
-- 
Steve Brenneis <sbrenneis at surry.net>