On 08/26/2013 09:20 AM, Marko Vojinovic wrote:
On Sun, 25 Aug 2013 22:12:34 -0700
Joe Zeff <joe@xxxxxxx> wrote:
On 08/25/2013 09:21 PM, Marko Vojinovic wrote:
While I respect Joe Zeff as a reputable member of this list, those
instructions on fedoraforum are a total piece of crap. I don't know
who wrote those instructions, but overengineering a solution for a
common problem is a always a Recipe For Disaster(tm).
Don't forget that some of them were written for newcomers who may or
may not know if their card's even supported. Once you have rpmfusion
set up, all you really need to do is have yum install kmod-nvidia,
the xorg files that it needs, run dracut and reboot. (Using akmod,
of course, gives you a slightly different set of files.) Most of the
complexity comes from the author having to take all sorts of
possibilities into account; as a user, you just pick the set of
instructions that matches you card. I've never yet had them fail,
but I have read posts on that forum where others have; usually,
they've either picked the wrong set of files to install or didn't
follow the instructions correctly. YMMV, and obviously does, but
everything I've seen, both personally and through threads on
fedoraforum lead me to believe that they're about as good a set of
instructions as you're likely to find.
BTW, Marco, do you have a link to a set of instructions you find
better? If so, I'd be interested in looking at them and possibly
pointing others to them instead in the future.
I don't have a link to point you to, aside from my first answer to
Roger above (that link can be found in the list archives).
But it literally boils down to three steps:
(1) activate rpmfusion,
(2) yum install kmod-nvidia,
(3) reboot the machine.
Feel free to substitute akmod in place of kmod if you wish.
With those in mind, if you want to write down an instruction manual, I
can say only this:
* Give the user a link to rpmfusion website, and let him figure out how
to activate it, if he didn't already. Giving him some ugly rpm
--nogpgcheck install http://blabla stuff is a bad idea on several
grounds, but most importantly the rpmfusion website is the
authoritative reference on how to activate it, and there is no need
to reinvent instructions that already exist.
* The yum installation of (a)kmod-nvidia will pull in any dependencies
it needs, including xorg libs, -header and -devel packages, even gcc
if necessary. There is no need to specify any of those manually.
* The nouveau is being disabled and dracut is being run as part of the
post-install scripts for kmod-nvidia. Neither of those should be done
manually. Especially not by a newbie.
* Any of the remaining stuff about PAE kernels, selinux policies, grub
tweaking and manually blacklisting nouveau should be frown upon. Not
only that those things are not necessary for the installation of the
driver, but moreover they can be downright dangerous if handled by
a newbie.
There is one more thing regarding the dracut stuff --- aside from the
fact that it is completely unnecessary since kmod-nvidia will already
do it by itself, doing it like this:
mv /boot/initramfs-$(uname -r).img /boot/initramfs-$(uname
-r)-nouveau.img
dracut /boot/initramfs-$(uname -r).img $(uname -r)
/boot/initramfs-$(uname -r).img is the default value of the first argument and
can be omitted.
$(uname -r) is the default value of the second argument and can be omitted.
So, this results in
# dracut
literally means asking for trouble. The mv command is almost
instantaneous, while the dracut command will take a good several
minutes to complete. In that sense, during the period after you have
renamed a valid initramfs file to some name that grub will not know to
look for, and before dracut has completed the new file, you do not have
a bootable machine, since there is no initramfs file for grub to
fallback on if something goes wrong. What will happen if a power surge
shuts down your computer in the middle of dracut run? Doing a mv before
dracut is a recipe to paint yourself into a corner.
What should be done instead is to first invoke dracut with a different
file name:
dracut /boot/mynewramfs-$(uname -r).img $(uname -r)
$(uname -r) is the default value of the second argument and can be omitted.
So, this results in
# dracut /boot/mynewramfs-$(uname -r).img
so that you don't touch the original file while this is being done.
Then, you want to *copy* the current initramfs into a backup:
cp /boot/initramfs-$(uname -r).img /boot/initramfs-$(uname -r)-nouveau.img
Finally, you want to move the new file into the place of the old one:
mv /boot/mynewramfs-$(uname -r).img /boot/initramfs-$(uname -r)
Since the file already exists, and since the root alias for mv is
interactive, it will ask you do you really want to overwrite the old
file. Answering yes will be fast, painless and a proof that you didn't
make any typos in the above commands.
During the whole procedure, if the machine loses power, it has a valid
initramfs file to fallback on automatically.
Again, all that said, yum install kmod-nvidia will will do all that
automatically, and there is really no reason to invoke dracut manually
in the first place. You just risk to fsck up something, like it just
happened to Roger.
HTH, :-)
Marko
All in all
I would encourage as many people to read and understand the installtion discussion. I think my initial disaster was, indeed, not understanding that rebooting could take a long time while it rebuilds the initrd...imge file.
I ran into issues and this series of posts should, for the future, help others thinking of using nvidia in the interim, until Nouveau is up to speed.
I would take this opportunity to thank Nouveau devs for tackling a huge task in reverse engineering.
Marko, thanks, your discussion has helped me immesuarbly. I am downloading the F19 iso and hope it will linux rescue the problem and install a working .img so I can at least get into the file system.
As for the instructions, I finally recognised them as similar to what I had used back in Fedora 14 days.
Thank you all
Roger
