7

I can not kill irq/${nnn}-nvidia by kill -9 or pkill -f -9. Does anyone how to kill or stop those process?

the irq using 100% of CPU

(I am using Ubuntu 16.04, if that is relevant.)

1
  • I solved this problem. The point of this topic is can not reboot when 100% CPU of irq. I did to reboot from this command. sudo su echo s > /proc/sysrq-trigger echo u > /proc/sysrq-trigger echo b > /proc/sysrq-trigger
    – TMit
    Dec 19, 2017 at 3:58

3 Answers 3

7

As @hobbs explained, it is a kernel thread. A broader perspective is the following:

IRQ handling is problematic in any OS because interrupts can arrive at any time. Interrupts can arrive even while the kernel is in the middle of working on a complex task and resources are inconsistent (pointers are pointing to invalid addresses and so on). This problem can be solved with locks, i.e. don't allow the interrupt handlers to run until the kernel is in an interruptible, consistent state. The disadvantage of using locks is that too many locks make the system slow and inefficient.

Thus, the optimal solution for the problem is this:

  1. The kernel interrupt handlers are as short as possible.
  2. Their only job is to move all relevant interrupt data into a temporary buffer
  3. Some "background" thread works continuously on this buffer and does the real work on behalf of the interrupt handlers.

These "background" threads are the interrupt handler kernel threads.

  • You see them in top as normal processes.
  • However, they are displayed as if they use zero memory.
  • And yes, this is true, because no real user space memory belongs to them.

They are essentially kernel threads running in the background.

You can't kill kernel threads: they are managed entirely by the kernel. If you could kill it, the irq/142 handler in your nvidia driver wouldn't exist any more: if your video card sends an interrupt, nothing would handle it. The result would be likely a freeze, but your video surely wouldn't work any more.

The problem in your system is that this interrupt handler gets a lot of CPU resource. There are many potential reasons:

  • For some reason, the hardware (your video card) sends so many interrupts that your CPU can't handle all of them.
  • The hardware is buggy.
  • The driver is buggy.

Knowing the quality of the Nvidia drivers, unfortunately a buggy driver is the most likely.

The solution is to somehow reset this driver. Some ideas, ordered ascending by brutality:

  • Is it running some 3D accelerated process in the background? Google Earth, for example? If yes, stop or kill it.
  • From X, switch back to character console (alt/ctrl/f1) and then back (alt/ctrl/f7). Then most of the video will re-initialize.
  • Restart X (exit ordinarily, or type alt/ctrl/backspace to kill the X server).
  • Kill X (killall -9 Xorg). It is better if you do this from the character console.

If you kill X and you still see this kernel thread, you may try to remove the Nvidia kernel module (you can see it in the list given by lsmod, then you can remove it with rmmod). Restarting X will insmod it automatically, resetting the hardware.

If none of these work, you need to reboot. If an ordinary reboot doesn't work you can do this with additional brutality: use alt/printscreen/s followed by alt/printscreen/b.


Extension: as a temporary workaround you could try to give a very low priority to that thread (renice +20 -p 1135). Then it will still run, but it will have less impact on your system performance.

5
  • Any way to reboot when got happened?
    – TMit
    Nov 13, 2017 at 13:11
  • @TMit Having a process running with 100% shouldn't avoid reboot. However, alt/printscreen/s syncs the disk, alt/printscreen/r mounts everything as readonly, and alt/printscreen/b reboots on the spot. If you can't reboot with alt/printscreen/b, your system is nearly surely dead.
    – peterh
    Nov 13, 2017 at 13:16
  • yes, I can Magic SysRq Key to reboot. How to get that by remote?
    – TMit
    Nov 14, 2017 at 14:27
  • @TMit The essence of the magic sysrq key is that it works always, even without network, or without anything. If you have a working keyboard driver, it will work. The same from net is inherently impossible. There is a software named sysrdq which emulates magic sysrq on a tcp port, of course it won't work if you don't have a network.
    – peterh
    Nov 14, 2017 at 15:32
  • @TMit If your problem is only that your X has totally f*cked your screen and you can't even switch back to character console: 1) login remotely with ssh 2) kill the old X server 3) start a new. The new X will reset the video chip.
    – peterh
    Nov 14, 2017 at 15:34
3

You can't. It's not a process, it's a kernel thread. You can't kill it, and if you did manage to you would only make your system (more) unusable.

1
  • I see. can not disable it too? I just need reboot but can not. so, any idea to reboot when it happen?
    – TMit
    Aug 19, 2017 at 3:51
-3

You've probably been hacked and they are running spambots, tradingbots or even mining. Get in firewall the crap out of it, change root password, deinstall any instance of Openssh and upgrade the kernel. I just moved up to the next release.

Be sure to block outgoing ssh. Then watch the ufw.log as the peers try to connect (list of other hacked nodes in the network). Note the list of repositories when running the upgrade. You are likely to note your desktop's 'username'-files as a subdomain listed. Because they tar balled and ex-filtrated your workstation's files when you connected.

Then dig in... I'm looking for the credentials to kill the repository.

1
  • 1
    Welcome to the site. To make your answer more helpful, please indicate why you believe the OP's problem may be the result of being hacked, e.g. by adding links to publications where this kind of symptom has been traced to a malware infestation. Also, if the problem actually is a hacking attack, more severe procedures are likely necessary, depending on how critical the affected system is.
    – AdminBee
    Dec 3, 2019 at 11:57

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .