udev vs nvidia

udev vs nvidia

[Daniel Drake]

Hi,

Since switching to udev, I have had problems when starting X relating to 
nvidia drivers. I know this is a tricky situation, with the nvidia drivers 
being closed-source, nobody really knows how they work. If it seems likely 
that this problem is not udev-related, then please tell me, and I will go to 
nvidia instead.

I'm not sure if this is even anything to do with udev. The only reason I 
suspect udev might be involved, is because of the existance of /dev nodes 
seems to play a part.


I am running nvidia drivers version 5336 with Martin Schlemmer's patch applied 
to make them export info into sysfs.
If I manually "modprobe nvidia", then I can see that the /dev/nvidia0 and 
/dev/nvidiactl nodes are created. So at this stage it seems to be loaded.


Ignoring the above situation, on a normal bootup, the nvidia module is not 
loaded at the point where I "startx":

# ls /dev/nvidia*
no such file or directory

# startx
XFree86 Version 4.3.0
Release Date: 27 February 2003
[snip]
(=) Using config file: "/etc/X11/XF86Config"
Using vt 7
(EE) NVIDIA(0): Failed to initialize the NVIDIA kernel module!
(EE) NVIDIA(0):  *** Aborting ***
(EE) Screen(s) found, but none have a usable configuration.

Fatal server error:
no screens found

If I mknod the /dev/nvidia* nodes first, then X is able to start as normal. 
Also, strangely, if I "modprobe nvidia", and then "startx" then X starts as 
normal.

I believe that this problem comes around because the /dev nodes are not 
created quickly enough.

# modprobe nvidia && ls -l /dev/nv*
ls: /dev/nv*: No such file or directory

# rmmod nvidia

# modprobe nvidia && sleep 2 && ls -l /dev/nv*
crw-rw-rw-    1 root     video    195,   0 Feb 16 19:47 /dev/nvidia0
crw-rw-rw-    1 root     video    195, 255 Feb 16 19:47 /dev/nvidiactl


My theory is that X loads the "nvidia" module, and then immediately tries to 
initialize the display and whatever else. But as the nvidia* nodes arent in 
place, it dies.

I am running udev017 on linux 2.6.3-rc2-mm1

Any ideas?

cheers,

Daniel


-------------------------------------------------------
SF.Net is sponsored by: Speed Start Your Linux Apps Now.
Build and deploy apps & Web services for Linux with
a free DVD software kit from IBM. Click Now!
http://ads.osdn.com/?ad_id\x1356&alloc_id438&op=click
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

Re: udev vs nvidia

[Daniel Drake]

claude wrote:
  > Is :
> 
> alias   char-major-195* nvidia
> 
> present in your /etc/modules.conf ?
> 

Yes, well, almost. Here's the relevant parts:


### modules-update: start processing /etc/modules.d/nvidia
# Nvidia drivers support
alias char-major-195 nvidia
alias /dev/nvidiactl char-major-195

# Old nvidia support ...
alias char-major-195 NVdriver
alias /dev/nvidiactl char-major-195

Is the * on char-major-195* important here?

thanks.

Daniel


-------------------------------------------------------
SF.Net is sponsored by: Speed Start Your Linux Apps Now.
Build and deploy apps & Web services for Linux with
a free DVD software kit from IBM. Click Now!
http://ads.osdn.com/?ad_id\x1356&alloc_id438&op=click
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

Re: udev vs nvidia

[claude]

Daniel Drake wrote:
> Hi,
> 

Hi,

> Since switching to udev, I have had problems when starting X relating to 
> nvidia drivers. I know this is a tricky situation, with the nvidia 
> drivers being closed-source, nobody really knows how they work. If it 
> seems likely that this problem is not udev-related, then please tell me, 
> and I will go to nvidia instead.
> 
> I'm not sure if this is even anything to do with udev. The only reason I 
> suspect udev might be involved, is because of the existance of /dev 
> nodes seems to play a part.
> 
> 
> I am running nvidia drivers version 5336 with Martin Schlemmer's patch 
> applied to make them export info into sysfs.
> If I manually "modprobe nvidia", then I can see that the /dev/nvidia0 
> and /dev/nvidiactl nodes are created. So at this stage it seems to be 
> loaded.
> 
> 
> 
> My theory is that X loads the "nvidia" module, and then immediately 
> tries to initialize the display and whatever else. But as the nvidia* 
> nodes arent in place, it dies.
> 
> I am running udev017 on linux 2.6.3-rc2-mm1
> 
> Any ideas?
> 

Is :

alias   char-major-195* nvidia

present in your /etc/modules.conf ?

> cheers,
> 
> Daniel
> 
> 
> -------------------------------------------------------
> SF.Net is sponsored by: Speed Start Your Linux Apps Now.
> Build and deploy apps & Web services for Linux with
> a free DVD software kit from IBM. Click Now!
> http://ads.osdn.com/?ad_id\x1356&alloc_id438&op=click
> _______________________________________________
> Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
> Linux-hotplug-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel
> 


-------------------------------------------------------
SF.Net is sponsored by: Speed Start Your Linux Apps Now.
Build and deploy apps & Web services for Linux with
a free DVD software kit from IBM. Click Now!
http://ads.osdn.com/?ad_id\x1356&alloc_id438&op=click
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

Re: udev vs nvidia

[Daniel Drake]

If anyone is interested, I have solved the below issue. It turns out that you 
are supposed to load the module before starting X (even nvidia say so), and 
devfs did this automatically on bootup. So, the nvidia module should be added 
to an autoload list, and this issue is not a fault of udev/hotplug.

Daniel

Daniel Drake wrote:
> Hi,
> 
> Since switching to udev, I have had problems when starting X relating to 
> nvidia drivers. I know this is a tricky situation, with the nvidia 
> drivers being closed-source, nobody really knows how they work. If it 
> seems likely that this problem is not udev-related, then please tell me, 
> and I will go to nvidia instead.
> 
> I'm not sure if this is even anything to do with udev. The only reason I 
> suspect udev might be involved, is because of the existance of /dev 
> nodes seems to play a part.
> 
> 
> I am running nvidia drivers version 5336 with Martin Schlemmer's patch 
> applied to make them export info into sysfs.
> If I manually "modprobe nvidia", then I can see that the /dev/nvidia0 
> and /dev/nvidiactl nodes are created. So at this stage it seems to be 
> loaded.
> 
> 
> Ignoring the above situation, on a normal bootup, the nvidia module is 
> not loaded at the point where I "startx":
> 
> # ls /dev/nvidia*
> no such file or directory
> 
> # startx
> XFree86 Version 4.3.0
> Release Date: 27 February 2003
> [snip]
> (=) Using config file: "/etc/X11/XF86Config"
> Using vt 7
> (EE) NVIDIA(0): Failed to initialize the NVIDIA kernel module!
> (EE) NVIDIA(0):  *** Aborting ***
> (EE) Screen(s) found, but none have a usable configuration.
> 
> Fatal server error:
> no screens found
> 
> If I mknod the /dev/nvidia* nodes first, then X is able to start as 
> normal. Also, strangely, if I "modprobe nvidia", and then "startx" then 
> X starts as normal.
> 
> I believe that this problem comes around because the /dev nodes are not 
> created quickly enough.
> 
> # modprobe nvidia && ls -l /dev/nv*
> ls: /dev/nv*: No such file or directory
> 
> # rmmod nvidia
> 
> # modprobe nvidia && sleep 2 && ls -l /dev/nv*
> crw-rw-rw-    1 root     video    195,   0 Feb 16 19:47 /dev/nvidia0
> crw-rw-rw-    1 root     video    195, 255 Feb 16 19:47 /dev/nvidiactl
> 
> 
> My theory is that X loads the "nvidia" module, and then immediately 
> tries to initialize the display and whatever else. But as the nvidia* 
> nodes arent in place, it dies.


-------------------------------------------------------
SF.Net is sponsored by: Speed Start Your Linux Apps Now.
Build and deploy apps & Web services for Linux with
a free DVD software kit from IBM. Click Now!
http://ads.osdn.com/?ad_id\x1356&alloc_id438&op=click
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

Re: udev vs nvidia

[MALET JL]

[-- Attachment #1: Type: text/plain, Size: 3803 bytes --]

Daniel Drake a écrit :

> If anyone is interested, I have solved the below issue. It turns out 
> that you are supposed to load the module before starting X (even 
> nvidia say so), and devfs did this automatically on bootup. So, the 
> nvidia module should be added to an autoload list, and this issue is 
> not a fault of udev/hotplug.
>
> Daniel
>
> Daniel Drake wrote:
>
>> Hi,
>>
>> Since switching to udev, I have had problems when starting X relating 
>> to nvidia drivers. I know this is a tricky situation, with the nvidia 
>> drivers being closed-source, nobody really knows how they work. If it 
>> seems likely that this problem is not udev-related, then please tell 
>> me, and I will go to nvidia instead.
>>
>> I'm not sure if this is even anything to do with udev. The only 
>> reason I suspect udev might be involved, is because of the existance 
>> of /dev nodes seems to play a part.
>>
>>
>> I am running nvidia drivers version 5336 with Martin Schlemmer's 
>> patch applied to make them export info into sysfs.
>> If I manually "modprobe nvidia", then I can see that the /dev/nvidia0 
>> and /dev/nvidiactl nodes are created. So at this stage it seems to be 
>> loaded.
>>
>>
>> Ignoring the above situation, on a normal bootup, the nvidia module 
>> is not loaded at the point where I "startx":
>>
>> # ls /dev/nvidia*
>> no such file or directory
>>
>> # startx
>> XFree86 Version 4.3.0
>> Release Date: 27 February 2003
>> [snip]
>> (==) Using config file: "/etc/X11/XF86Config"
>> Using vt 7
>> (EE) NVIDIA(0): Failed to initialize the NVIDIA kernel module!
>> (EE) NVIDIA(0):  *** Aborting ***
>> (EE) Screen(s) found, but none have a usable configuration.
>>
>> Fatal server error:
>> no screens found
>>
>> If I mknod the /dev/nvidia* nodes first, then X is able to start as 
>> normal. Also, strangely, if I "modprobe nvidia", and then "startx" 
>> then X starts as normal.
>>
>> I believe that this problem comes around because the /dev nodes are 
>> not created quickly enough.
>>
>> # modprobe nvidia && ls -l /dev/nv*
>> ls: /dev/nv*: No such file or directory
>>
>> # rmmod nvidia
>>
>> # modprobe nvidia && sleep 2 && ls -l /dev/nv*
>> crw-rw-rw-    1 root     video    195,   0 Feb 16 19:47 /dev/nvidia0
>> crw-rw-rw-    1 root     video    195, 255 Feb 16 19:47 /dev/nvidiactl
>>
>>
>> My theory is that X loads the "nvidia" module, and then immediately 
>> tries to initialize the display and whatever else. But as the nvidia* 
>> nodes arent in place, it dies.
>
>
>
> -------------------------------------------------------
> SF.Net is sponsored by: Speed Start Your Linux Apps Now.
> Build and deploy apps & Web services for Linux with
> a free DVD software kit from IBM. Click Now!
> http://ads.osdn.com/?ad_id=1356&alloc_id=3438&op=click
> _______________________________________________
> Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
> Linux-hotplug-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

here is a solution :

jlm@Sorcerer:~$ cat /etc/udev/nvidia.sh
#!/bin/sh
mknod /dev/nvidia0 c 195 0 -m a+rw
mknod /dev/nvidia1 c 195 1 -m a+rw
mknod /dev/nvidia2 c 195 2 -m a+rw
mknod /dev/nvidia3 c 195 3 -m a+rw
mknod /dev/nvidia4 c 195 4 -m a+rw
mknod /dev/nvidia5 c 195 5 -m a+rw
mknod /dev/nvidia6 c 195 6 -m a+rw
mknod /dev/nvidia7 c 195 7 -m a+rw
mknod /dev/nvidiactl c 195 255 -m a+rw

jlm@Sorcerer:~$ cat  /etc/modprobe.conf
install ppp_generic     mknod /dev/ppp c 108 0 -m 0666 && modprobe 
--ignore-install ppp_generic;
install         nvidia                  /etc/udev/nvidia.sh && modprobe 
--ignore-install nvidia;
[...]



note that the same issue occurs with pppd

[-- Attachment #2: Type: application/pgp-signature, Size: 250 bytes --]