System hard locks leaving X - many scenarios tested...

[cleverhandle]

Already sent to linux-bugs@ - any other ideas to try here?

I'm getting total system lockups when I leave X11 (i.e. switch to console or shutdown/reboot from GDM). When this occurs, I'm left with a blinking underscore cursor and a totally dead machine - it does not respond to pings or SSH, only the reset button. This does not happen absolutely every time, but it is still pretty consistent. I'd say that at least 2 out of 3 X11-->text switches locks up. The system runs Debian testing with a Chaintech FX5700 Ultra on an Asus A7V333. At the moment, I'm using the 6106 drivers and the 2.6.6 kernel, though more variations are listed below.

Unfortunately, I'm not positive how long this has been going on - I almost never use a real console and only reboot the box maybe once a month. So I had seen the behavior once or twice, but just passed it off as a bizarre fluke. I know the problem occurs at least in the last two driver revisions (6106 and the 5xxx one before that), spanning several 2.6 kernel versions. I think I saw this in 2.4 kernels as well. I'm fairly certain the problem did not occur with my old Ti4400, only with my new FX5700 Ultra (a few months old).

Googling and searching the nvnews.net Linux forum reveals a number of problem reports (some of them identical, some not-quite-identical) and many proposed solutions. Here are some of the things I've tried that have no effect:

1) AGP - AGPGART, NvAGP, or no AGP behave the same

2) Local APIC - enabled or disabled in both the BIOS and kernel boot parameter

3) "Connected Monitor" Option in XF86Config - seems that a lot of people have seen TV-Out-related problems, so I tried this even though I never use TC-Out

4) Framebuffer - I run a kernel with all framebuffer options disabled, but I tried passing vga=normal to the kernel for kicks. No difference.

5) Kernel version - besides earlier versions, I've tried 4.6.8rc2 as well

6) APM - my kernels usually have only APM, no ACPI. But disabling that with an apm=no boot parameter has no effect.

7) Video BIOS - I hate to mess with BIOS's, so this was the last thing I tried. But a marginally newer version of the Chaintech BIOS showed the same behavior as well.

[ridcully]

Hmm.. same problem here, but other card (Geforce4 Ti4200).
But my problems dissappear, if I deactivate all other outputs (DVI and TV-OUT) in XF86Config with:
Option "IgnoreDisplayDevices" "DFP,TV"

The only difference to your problem is that in most cases I can reboot or poweroff from gdm. Just switching from X to console (Alt-F1) freezes my machine.

But thanks that you've provided all the details of what you've tried already. I wanted to do this in next days on my machine, too.

One thing I've also tried:

8) Enabled or Disabled SideBandAdressing and FastWrites - No differences

[ridcully]

Just for information:

I tried another scenario:

9) Booted up with 2 screens in XF86Config (CRT and TV-OUT) but set both screens to 640x480 with Depth 8 - no difference, hard lock when switching to tty1 or killing X server with Alt-Backspace.

(Using 1.0-6106, with both, NvAGP and agpgart, without acpi or apm, without any framebuffer support in kernel)

[cleverhandle]

I can add another...

10) No effect from enabling/disabling ehci (USB 2).

[whig]

Does the "nv" driver cause grief too?

[cleverhandle]

The nv included in Sarge doesn't support the FX5700 Ultra, so I can't say much there. I suppose I could install something like FC2 on an extra drive to test, though that may introduce other variables into the situation as well.

To be honest, I'm not really keen on spending much more time testing this unless I hear something back from nVidia. I don't expect them to devote an engineer to my particular problem, but an acknowledgement of "Yes, we're listening" would really be nice.

[cleverhandle]

Well, crap... it works fine in Fedora Core 2, but not in Debian. That's a first. (Kidding, kidding...)

But seriously, that kind of stinks because I have no idea where to look now. The kernel version was the same (2.6.6) and the X configuration was the same. My kernel is a custom job, so maybe I'll try the stock one, though I doubt that's it. Which leaves X11, most likely. Debian X is 4.3 with a lot of the X.Org stuff backported in, from what I can glean from the mailing lists. But there are no real X.Org packages available. Hmm...

Any other ideas as to what might make FC2 and Debian behave differently in this regard?

[cleverhandle]

Well, one step closer... the problem is in the kernel somewhere. The stock Debian kernels for both i386 and k7 work fine. So I must have changed something to break things, though I think I've tried the usual suspects already. The output of "diff myconfig debconfig" is attached - anyone see anything that looks significant?