Had major problems with 295.20-1 with suspend/shutdown/logout and even stopping gdm or the X server. After downgrading back to 290.10-1 suspend/resume work fine again.
I recently (26 Feb) upgraded from 290.10-1 to 295.20-1 on my laptop (HP 8440p with a NVS 3100M graphics system) using the packages for 295.20-1 from debian testing. My system is mostly debian stable, but with some packages from debian testing. On 27 Feb I stopped X, removed the module and then restarted X when I recalled that I had not done so immediately after the upgrade. Between the upgrade and loading the new module - the system was able to suspend and resume properly! Since then I have had multiple problems, all which seem related to the Nvidia 295.20-1 driver.
- Suspend to Ram stopped working. Would look like it was about to suspend, but the system logs did not show a suspend starting and then CPU/fan would keep running. Only hard power off helped.
- On some occasions shutdown (from Gnome menu when logged in) worked and on other cases it failed. When shutdown failed - only hard power off worked.
- Trying to stop gdm3 caused the same type of freeze.
- Killing X when started using "startx" via control-c, caused the same freeze.
- This version of the driver does not allow changing to a Virtual Terminal from the X screen. This made some debugging and testing more complex than it should be.
In all cases of the "crash" the laptop power would stay ON, the fan was pushing warm air out, and the system would not respond to keyboard at all. Network connections also failed or were dropped - even when the connection was live and working a few seconds before the change. The system would not respond to "ping" anymore either. After every crash and hard-reset, my /home partition was "unclean" and in quite a few cases there were messed up files (multiply claimed blocks, unused inodes, etc.)
On the attempt to suspend - daemon.log shows that NetworkManager received the request to sleep, but kern.log did not get the messages "PM: Syncing filesystems ... done." or "PM: Preparing system for mem sleep" which show up as the suspend process is being done.
I tried to "sleep 5; nvidia-bug-report.sh" in one remote session, and then run "/etc/init.d/gdm3 stop" to see if I could get debugging data while the "crash" was occuring. That was not successful. Instead, I could only attach a "nvidia-bug-report.log.gz" file which was generated after starting x from a remote console with "startx -- -logverbose 6" and before initiating a crash.
Bottom line - at least on my machine the 295.20-1 driver makes a total mess of things on my machine.
Ref: files on Debian BTS also: http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=661822