On 11/9/06, Kim Lux <lux@xxxxxxxxxxxxxxxxxxx> wrote:
On Thu, 2006-11-09 at 18:09 -0600, Jonathan Berry wrote:
> > Well, contrary to your argument, the Nvidia drive built under kernel
> > 2835. The problem was that installing the new kernel killed nvidia from
> > working under kernel 2798. As far as I am concerned that is a no no.
> > The second issue in all of this is that we have to BUILD drivers in the
> > first place.
>
> At what point was the old driver "killed"?
>From var/log/nvidia-installer.log
-> There appears to already be a driver installed on your system
(version: 1.0-
8776). As part of installing this driver (version: 1.0-9629), the
existing
driver will be uninstalled. Are you sure you want to continue? ('no'
will a
bort installation) (Answer: Yes)
In other words, when you installed the new driver. This is different
from the kernel update. Installing the new kernel did not kill the
driver, as you purport. Installing the new driver did.
> > > If you are running a testing
> > > kernel, it is assumed that you know what you are doing.
> >
> > Well, I've built a few kernels in my day.
>
> Then why are you complaining, if you know what you are doing?
Because its a pain in the butt and a waste of time. It would be one
thing if the build worked every time, but about half the time it
doesn't.
Agreed. That's why I've switched to the Livna driver. It just works.
> > I understand that, but the real issue is that we should need to depend
> > on a single website to provide nvidia drivers !
>
> I think you left out a "not"... What alternative do you propose?
What is the reason why nvidia can't release source to the kernel
people ?
I mentioned this in a previous email. nVidia does not own all of the
Intellectual Property involved in their graphics cards and their
driver. They cannot give away what is not theirs, so they are legally
prohibited from releasing the driver source. Sorry, but that's the
way it is. Look around for the issue on the web. I've seen this
stated several times by people from nVidia.
What prevents the OS people from putting a wrapper around the Windows
version of the driver ?
That's essentially what the nVidia binary driver is. For the most
part, it is the same code as the Windows driver (as I understand it,
at least). There is a small bit of source code that connects the two
(what you actually build when installing it).
Any chance the NVidia driver could be reverse engineered, especially
since we have examples built for the Linux kernel ?
Sure. Some people have worked on that, even. The problem is, nVidia
has a nasty habit of developing new cards that are faster and fancier
and have new features :-). You cannot keep up. Oh, and that IP issue
above is also why nVidia cannot give out the details needed for
someone else to write an open source driver.
I'm not a kernel developer, so please excuse my naivety.
I'm not either, but I've seen the issue discussed many times (several
on this list...). Graphics cards are about as complex as they come.
> > > That doesn't seem to be what you were claiming here. It seems like
> > > you were fine until you upgraded to the testing kernel...
> >
> > Yes, installing the new kernel disabled the nvidia driver for the old
> > kernel. As far as I know, that happens for ever kernel or at least it
> > has for ever kernel that I have installed for the last two years !
>
> Installing the new kernel should do nothing to the nvidia driver for
> the old kernel. Now, if you go installing the nvidia driver for the
> new kernel, that might do something.
It has been my experience that merely running the new kernel will break
the driver for the old kernel. Like it did this morning. I don't
understand why.
I thought you said it didn't break until you installed the new driver?
Running the new kernel should not break the driver for old kernel, as
far as I know.
Question: why does the old driver have to be uninstalled for the new
driver to be built ? I'm guessing it doesn't reside in the modules
directory like other kernel modules ? Even fixing that would be a step
forward. At least one could step back to old kernels.
If you install a new version of the driver, then the old one has to
go. There are bits that get put somewhere besides the modules
directory and you cannot have two versions of this. Now, by default,
even if you install the same version, I think it will remove the old
kernel module. Look at the -k and -K options for the nvidia
installer. One of them is used to specify the kernel version (if
other than the currently running one) and one says only install the
kernel module, which causes it not to remove the old driver. I'm not
sure why this is not the default. But, when you cross a version
number, you have to remove the old one.
> > First of all, to use a livna driver, you have to know that livna exists
> > and then you have to be able to set livna up as a repository. Now, take
> > your average computer user and tell me that that it is a reasonable
> > expectation that they are able to to do that. It isn't.
>
> Do you think Window's driver model is reasonable?
The Linux driver model is preferable by far.
I agree, at least to some extent. As long as your hardware is
supported, it's great. When it's not and you have to get an out of
tree or binary driver, it's not so great. Which is why we are having
this discussion... :-)
> For nVidia cards,
> you have to go to nvidia.com and download a driver to install.
Yes, but you don't have to build them ! The issue is that the drivers
have to be built, because if you look at the errors, that is where they
are occurring. SELinux gets in the way, the kernel headers have to be
installed, files are missing or renamed in the kernel, gcc is the wrong
version, etc, etc, etc.
You can go to Livna and not have to build them as well. Have I said
that before? :-) Which point are we going to debate where?
> I
> don't see that as being unreasonable, and is about what you have to do
> with Livna.
Its different as explained above.
> And it doesn't take long on this list to find out about
> Livna. If you look a little bit, you can find these things out.
I know about livna, etc. But just because its a solution does it mean
its reasonable ?
Valid point. A solution is not necessarily reasonable just because it
is a solution. Is there a better, more reasonable solution? Or what
is not reasonable about this solution?
> Is
> it ideal? Probably not. But that's where we are. Don't like it?
> Don't use Fedora. Find a distro that ships the nvidia driver.
And which one would that be ?
I'm not sure. Many distros have gone to an OSS only approach. nVidia
only recently modified the license to allow distros to redistribute
their driver if they so choose. Fedora chooses not to, because the
driver is not OSS, because nVidia cannot release the source...
> > Secondly, livna is not always up and its drivers have not always worked.
> > Then what ?
>
> You wait and don't update your system. Or I think there is a way to
> make it not update the kernel if the module is not ready. Do you
> think your "average computer user" will be doing updates anyway?
Yes !
Really? It seems that most of the zombied Windows boxes out there are
cracked because someone has neglected to update the thing, which would
have plugged the hole. You always have the choice not to update. It
may not be the best choice, security-wise, but it is a choice.
> > > During the time I used the package from nVidia, it worked most of the
> > > time.
> >
> > Yeah, most of the time. Great.
>
> This is Fedora. It's cutting edge. Sometimes things break, and those
> things may not be the nvidia driver. If you are running Fedora, you
> are already taking that chance. Deal with it.
I don't buy that.
Why not? If all the distros are the same, what's the point in having
multiple distros. Fedora has some features to it that do not
necessarily make it right for a first-time Linux user. I don't mind
that; I can deal with it. It is getting better, though. A lot of
things just work, and that's great.
> > > The times it did not, I usually found a fix fairly quickly.
> >
> > Just like I did this morning. How how about my mom, could she find a
> > fix quickly ?
>
> Are we talking about your mom?
Yeah, that is part of it. I got my nvidia driver working this morning.
This is a general usability issue.
Okay, well, let's keep things organized here. So your mom cannot
handle updating the nvidia driver. Fine. Let her use the nv driver.
Or the Livna package. Or you do the updates and fix it for her. Or
don't buy an nVidia card or a laptop with an nVidia card. As far as
fixing nVidia, I'm not sure there is much we can do there,
unfortunately.
> > > Then I switched to the Livna package and have had even fewer issues.
> >
> > Even fewer means some ?
>
> Fewer means probably none, but just in case I don't remember.
> Certainly I've had to hold off updating my kernel. I've also rebuild
> the src.rpm for the driver. Sometimes that didn't work the first
> time.
Yeah.
> > > The problem, especially when you throw in test kernels, is that the
> > > kernel changes quickly (and Fedora updates to those changes quickly)
> > > and nVidia cannot keep up with those changes.
> >
> > Even test kernels aside, you are telling us you had issues.
>
> No, I was talking about using nVidia's binary package. It's not
> always up to date. But even then, there are usually patches on the
> nvnews.net forum.
Yeah, visit a forum and muck around and then apply patches and compile.
Fun. Totally unnecessary.
Right. Just use the Livna package ;-).
> You missed my point. I was talking about Livna. Why are you singling
> out the nvidia driver when you have happily delt with the NTFS driver?
> It suffers the same issues with Livna that the nvidia driver does
> that you are complaining about.
Because I haven't had a single issue with the ntfs driver. Never, none,
zero.
Try the Livna package and see if you have any trouble with it. If you
like the Livna NTFS driver, you'll probably like the nvidia driver as
well.
> > I have never been able to boot into a previous kernel. As a matter of
> > fact, the installer script goes like this for me:
> >
> > "-> There appears to already be a driver installed on your system
> > (version: 1.0-
> > 8776). As part of installing this driver (version: 1.0-9629), the
> > existing
> > driver will be uninstalled. Are you sure you want to continue? ('no'
> > will a
> > bort installation) (Answer: Yes)"
>
> Notice the version numbers. You cannot have two different versions
> installed. There is a way to install the same version on multple
> kernels. Try running the nvidia script with the --help option. Lots
> of info there.
The issue isn't that it can't be made to do this and that and work in
general. Because it can. And I am sure we can all become experts on
installing nvidia drivers. Its totally unnecessary ! If nvidia would
give the kernel developers some info, we wouldn't have to put up with
these work arounds.
Yes, it's not ideal. But that's where we are. See above for the IP
issues involved. IP is big business.
> And you have been offered several solutions to this issue. An issue
> is a non-issue when it is solvable with a known solution.
Well then I guess we would never have computers in the first place
because the early computers didn't do much more than we could do in our
head !
Huh? But they could do it faster, more accurately, and for hours on
end. There were advantages to computers when they were first
developed. I'm not saying don't question the status quo; sometimes
that is necessary. But if you do, don't do it in a way that will make
people mad (again, see your subject line). First, see what solutions
are out there. If one works, great! Use it! If you don't like it,
voice your opinion and back it up with facts. Offer other
suggestions. You seem like you bring this up so that the issue can be
addressed and hopefully improved, but for the most part, you have not
gotten people's attention in a good way.
If you don't like nVidia and using the driver, I'd suggest when you
purchase your next computer, don't buy an nVidia card. Personally, I
think nVidia is doing the most for Linux of all the graphics card
companies, so I support them. Yeah, Intel has open source Linux
drivers, but are they as good as the Windows drivers? Is the hardware
as good as nVidia's? Can I use an Intel graphics card with my AMD CPU
that I prefer? Maybe you can find something that works better for
you, but for me, nVidia works.
Jonathan
