Re: nvidia crashing info

[Ruben F. <par...@gm...>]

A very refreshing story on multi-seated X! I enjoyed reading it and 
learned a lot about stability issues. Keep up the good work, and thank 
you for this well-written report!
> 
> Message: 2
> Date: Mon, 14 Nov 2005 21:36:46 -0600
> From: Open Sense Solutions <in...@op...>
> To: lin...@li...
> Subject: nvidia crashing info
> 
> Here's some more info on the nvidia crashing problem so many people
> have experienced in one form or another:
> Our old Debian software with XFree86 used the nvidia driver and was
> very stable.  Now with Ubuntu 5.10 and Xorg we are having a lot more
> problems.  Using the nv open source driver instead of the binary
> nvidia, things are very stable.  Restarting the primary x server with
> ctrl-alt-bs repeatedly can crash things, but we have gdm set to not
> restart the x servers.  However, the nvidia binary driver is crashing
> about 1/20 times during logout.  A gdm restart gets things back, but
> the crashing is unacceptable.  We have noticed interrupt conflicts
> that made things even worse, but even with the nvidia cards on a
> different interrupt from the ethernet card we still see crashes.  It
> would be interesting to see if we could get each nvidia card to have
> its own interrupt if that would fix things.  Of course interrupt
> conflicts mean the nvidia driver is not pci spec compliant, but what
> will it take to convince nvidia to take a serious look at this issue?
> 
> We just turned off a third station on one of our boxes to see if it
> was more stable, and it still crashed using the nvidia driver, but
> even more interestingly, the gdm.log had this message:
> 
> NVIDIA: could not open the device file /dev/nvidia2 (Input/output error).
> (EE) NVIDIA(0): Failed to initialize the NVIDIA graphics device!
> 
> and /dev/nvidia2 shouldn't even be used!  Why is the primary card
> trying to open /dev/nvidia2???  We were only running 2 xservers which
> should be using /dev/nvidia0 and /dev/nvidia1.  /dev/nvidia2 was
> probably created during our initial probe (we probe all cards even if
> we don't have enough keyboards and mice in case we want to start
> additional servers later) but it should not be referenced at all.
> 
> This kind of error tells us something about how the cards are
> conflicting with other cards in the system, but only nvidia can make
> sense of it since the drivers are closed.  We are fed up, and have 5
> ATI cards getting delivered tomorrow for experimentation: two 9250
> pci, 1 9250 agp, and an x300 pci-e.  Supposedly you can run two
> independent x servers with one dual head card using the binary ATI
> driver, which would be a cost savings.  The multiXnest approach gets
> you that even with nvidia, but it is slow. Everyone says ATI is not
> linux friendly, but maybe once we figure out all the quirks it will
> not crash like nvidia.  If Matrox was a little more Linux friendly we
> might try them, and their new pci-e 1x cards would be very useful in
> modern systems to allow more cards, but Matrox prices are just insane.