[SOLVED]BUG: soft lockup - CPU#0 stuck for 11s! [java:10670]

[gentunian]

Hi there,

I was having some weird problem with the keyboard. The keys stopped working, all of them. The only combination possible was to do a safe restart with ctrl+alt+sysreq. First i suppose a hacker attack (paranoia) but when i started to track the problem, always happens running azureus. So I look at the logs and I found this:

[gentunian]

sun-sdk-1.6.0.03 it's a dependency:

chaplin seba # emerge -av azureus

These are the packages that would be merged, in order:

[irgu]

From http://article.gmane.org/gmane.comp.file-systems.ntfs-3g.devel/418

[gentunian]

thanks! I'll try the kernel module.
_________________
Si un pueblo tiene hambre, no le des un pez. Enséñale a pescar.
http://labombiya.com.ar

[brazso]

Using "fuse" compiled in the kernel solved your lockup problem? I'm using the latest 2.6.24-r3 (stable branch) kernel but I still have complete system lockups during writing to ntfs partition. At first the touched avidemux application freezes then soon I lose the keyboard and/or the mouse. Emerging ntfs3g still includes the emerge of fuse 2.7.2 (latest in test branch) but it says that it is not used as module due to the found fuse in the kernel. Listmod displays fuse, but I cannot specify its version. Is there a way to display versions of loaded modules?

[gentunian]

Yes. Using fuse compiled in the kernel solved my lockup problem. I really forgot this thread, I apologized for that. The problem initiated when azureus was accessing to a ntfs partition using fuse compiled outside the kernel, that is, by the fuse package. The fuse package complains about the module being compiled in the kernel, as you said.

My kernel version is 2.6.24-gentoo-r3 (don't know if that is now stable, when I emerged was masked, but you are saying that it's stable, so maybe it is) and the fuse version is 2.7.2. Nothing to do with, but just for the record my ntfs3g is compiled with the suid flag.

You could use modinfo to gather information about a module, but I think it would not provide the module version information you want though. I've been trying the --dump-modversions option for modprobe, but it seems that it doesn't work.

It's weird that we have the same kernel and the same package version. Did you see the logs? Maybe it's not the same problem.
_________________
Si un pueblo tiene hambre, no le des un pez. Enséñale a pescar.
http://labombiya.com.ar

[brazso]

Thanks for your detailed answer, gentunian! To be frankly I found nothing suspicious in the logs after lockups. I checked everything under /var/log directory, especially the current output file of dmesg. Do I have to activate some extra log in the kernel before? After I had various lockups during the usage of avidemux (output went to ntfs partition), I just tried to copy manually a bigger file into my ntfs partition, and it resulted the same lockup. That is why I think ntfs-3g is responsible for the problem.

[irgu]

The problem is the FUSE kernel module. Either you need to use the FUSE module in the 2.6.24 kernel or the one in the FUSE 2.7.3 package. The FUSE kernel module in the FUSE 2.7.2 package is broken.

[gentunian]

[brazso]

gentunian> I had installed gentoo to my desktop machine at least 2 years ago so I did not remember which log package was active. I checked it, it's the metalog. In its configuration file I think everything is switched on and I have lot of files under /var/log, e.g. "messages" that you mentioned. However I found nothing inside of the log files which might be related with the lockups. Following your advice I shall try to activate all debug facilities in the kernel, moreover I shall try the ntfs writing with my previous kernel (2.6.23). Thanks for the detailed description of using SysRq button, I always learn something new.
I have just noticed irgu's comment. Theoretically I'm using the compiled fuse module from kernel 2.6.24-r3. Command modinfo fuse confirms that fuse.ko is used and loaded from the actual kernel path. However the output of dmesg contains some message about fuse 2.7.2 as loaded one. I will try a later kernel from the test branch, I'm curious to know which fuse version is used there.

[gentunian]

brazso, checking the log for all fuse lines (eg, "cat /var/log/messages | grep fuse") I found that since I compiled the FUSE module in the kernel, the line:

[brazso]

gentunian> after system boot dmesg now displays exactly the same loaded fuse info (7.8, 2.7.2) that you included before activating its kernel usage. I'm afraid you are right and the kernel still uses the fuse package despite it is set in the kernel as module however modinfo says its opposite. Did you remove the emerged fuse package? More packages (one of them is ntfs3g) depend on it, so it cannot be removed in my case. Still I can try to embed it (=y) into the kernel instead of modul setting (=m). You have the same kernel version than mine (2.6.24-r3). Could you tell me the size of the fuse.ko file displayed by modinfo fuse? It cannot be different, but who knows I will check the logs again after setting the kernel debug on, thanks!

[gentunian]

Well, if you have two modules of the same name you can located using slocate. You can check the /lib/modules directory. Also, it's safe to unmerge the fuse package, compile the kernel module, and the emerge the fuse package.

BTW, modinfo tells you the path to the module. That doesn't mean that the module it's from the kernel source. So, if you have your slocate updated, you could check for the fuse module doing:

[brazso]

The proposed slocate command solved the problem, it displayed 2 fuse.ko files under /lib/modules/2.6.24-gentoo-r3.

[gentunian]