!! NVIDIA WORKS !!!

[Bruno Wolff III]

On Fri, Oct 15, 2010 at 19:29:15 -0600,
  "Christopher A. Williams" <chriswfedora at cawllc.com> wrote:
> 
> I just love the "tainted kernel" part almost as much as the "then it
> might get looked at" statement if the problem really is theirs. If it's
> the kernel maintainers problem, then the kernel maintainers had better
> not only look at it, but fix it. The sole reason the nVidia driver

Well I wished it worked that way as well, but there seems to be manpower
problems, at least with some areas of the kernel.

> For example: If I put an aftermarket engine part on my car and my car
> has an engine problem, the warranty on my car isn't instantly voided. If
> the engine had a problem not involving that part, the warranty must
> cover that repair. If the aftermarket part caused the problem, then the
> manufacturer of that part is the one on the hook for the repair
> according to whatever warranty they have (that's the law of the land in
> the US).

The law for allowing third party parts was to keep car owners from being
locked into buying parts from the car manufacturers at greatly inflated
prices. With computer controlled engines we are seeing that kind of problem
again and the laws are likely to adjust for that. (To make sure nondealer
mechanics can work on cars.)

Open source doesn't have that kind of lockin though. Getting changes you
want, in upstream may be a problem. Getting bugs you want looked at for
no dollar cost (you still put in some time) can also be a problem. But it
isn't the same as the lockin problem car repairs can have that need laws
to allow for 3rd party support.

> In all the time I have run Fedora and nVidia, kernel issues from linking
> this module were far from the biggest problem I've had, and nVidia seems
> to do a reasonably good job handling their problems when they do come
> up. At least most of the time...

Most of the complaints seem to be when rpmfusion is slow to get out an
update and people using kmod as opposed to amod. rpmfusion could actually
do some stuff with their packages to prevent that. (At the expense of
delaying relatively important kernel updates.) At major kernel updates there
can be other problems, but it doesn't seem to typically be a problem.
Down the road, user mode setting is going away. There is probably a year
left (but maybe longer) and then only kernel mode setting will be supported.
(At least actively. I expect that legacy user mode support will disabled in
Fedora kernels, though other distros may support it longer.) nVidia hasn't
appeared to have done anything to move in that direction.
 
> In the meantime, if I have a problem with the kernel that doesn't deal
> with the nVidia module, I will rightly expect that the kernel
> maintainers will do their best to troubleshoot and resolve my bugzilla
> reports. They have in the past. Why would that change now?

That's better luck than I have had. I own old hardware and have had kernel
bugs open (in Fedora) for a long time without resolution. The people working
this stuff have limited time and some bugs are more important to fix than
others. Working around issues with old hardware tends to be in the low
priority category.

> As to the lock-in, we'll see if they succeed. I have a feeling that they
> might indeed be trying, but they will probably fail in the end. They
> don't drive enough of the market to make good on that kind of strategy
> from what I can see.

They do have a first mover advantage and people seem to know about CUDA
more than openCL. But hopefully things will move to an open standard.

> No arguments from me with the first part. But the onus actually needs to
> be on whoever, after a reasonable investigation, is most likely to have
> the problem. Even then, all parties need to remain actively involved
> with the effort of fixing the problem. That's why people need to
> cooperate and collaborate like adults. An attitude of "I'm not doing
> squat until you prove it's my bug" helps nobody.

It helps the person who is short on time and whose is getting asked to help
support stuff outside of their primary responsibilities. (For example
non-Fedora software.) Especially when the other side of the inferface is
closed and it takes extra effort to figure out what is going on. At least
some interested users can help do some of the bug isolation. But probably
not a lot and it again takes time that could be spent doing other stuff.
There are things that I am pretty sure I could eventaully do myself, that
I would really like other people to do because of the initial ramp up in
training that would take away from my work in other areas where I already
pretty much know what I am doing.