Skip to contentSkip to conversation

swapper: page allocation failure. order:1, mode:0x20

DISCUSSION Benjamin L. Shi 6 participants 6 messages Feb 28, 2005 1 message not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw

We've seen these, by adding the following tueables resolved the problem. More specifically, the lower zone protection made the difference.

vm.vfs_cache_pressure=1000
vm.lower_zone_protection=100
vm.max_map_count = 32668
vm.min_free_kbytes = 10000

Quoted 17 lines · Bernd Schubert

Oh no, not this page allocation problems again. In summer I already pos= ted problems with page allocation errors with 2.6.7, but to me it seemed th= at nobody cared. That time we got those problems every morning during the = cron jobs and our main file server always completely crashed. This time its our cluster master system and first happend after an upti= me of 89 days, kernel is 2.6.9. Besides of those messages, the system stil= l seems to run stable

I really beg for help here, so please please please help me solving thi= s probem. What can I do to solve it?

First a (dumb) question, what does 'page allocation failure' really m= ean? Is it some out of memory case?

Thanks a lot in advance for any help,
Bernd

#2 lore raw

Signals/ Communication from kernel to user!

Hello,
We wanted to establish a communication from kernel module (possibly a
driver) to a user level process.

Wanted to know whether signals can be used for this purpose OR there any
other (better) methods of communication??

Regards,
Ravindra N.

#3 lore raw

Re: Signals/ Communication from kernel to user!

Quoted 8 lines · Ravindra Nadgauda

Hello,
We wanted to establish a communication from kernel module (possibly a
driver) to a user level process.

Wanted to know whether signals can be used for this purpose OR there any
other (better) methods of communication??

Perhaps netlink? Here's an introduction: http://qos.ittc.ku.edu/netlink/html/

Quoted 3 lines

Regards,
Ravindra N.

Signature · 1 line
- Timothy R. Chavez
#4 lore raw

Re: Signals/ Communication from kernel to user!

In reply to Ravindra Nadgauda
Quoted 9 lines

Hello,
We wanted to establish a communication from kernel module (possibly a
driver) to a user level process.

Wanted to know whether signals can be used for this purpose OR there any
other (better) methods of communication??

a bit brute force but you can simply run through the task list and kick the pid of your user-space app (example for 2.4 kernel):

hofrat

---snip---
/*
* Copywrite 2002 Der Herr Hofrat
* License GPL V2
* Author [email protected]
*/
/*
* run through the task list of linux search for the passed pid and send it
* a SIGKILL . run as insmod pid=# to send process with pid # a kill signal */

#include <bits/signum.h> /* signal number macros SIGHUP etc. */ #include <linux/kernel.h> /* printk level */
#include <linux/module.h> /* kernel version etc. */ #include <linux/sched.h> /* task_struct */

MODULE_LICENSE("GPL v2");
MODULE_AUTHOR("Der Herr Hofrat"); MODULE_DESCRIPTION("Signal to a user-space app from a kernel module");

int pid=0;
MODULE_PARM(pid,"i");

int
ksignal(int pid,int signum)

{
	struct task_struct *p;

	/* run through the task list of linux until we find our pid */
	for_each_task(p){
		if(p->pid == pid){
			printk("sending signal %d for pid %d\n",signum,(int)p->pid);
			/* don't have a sig_info struct to send along - pass 0 */
			return send_sig(signum,p,0);
		}
	}
	/* did not find the requested pid */
	return -1;
}

int
init_module(void)

{
	/* send pid a SIGKILL */
	ksignal(pid,SIGKILL);
	return 0;
}

void
cleanup_module(void)

{
	printk("out of here\n");
}
#5 lore raw

Re: Signals/ Communication from kernel to user!

In reply to Ravindra Nadgauda
Quoted 7 lines · Ravindra Nadgauda

Hello,
We wanted to establish a communication from kernel module (possibly a
driver) to a user level process.

Wanted to know whether signals can be used for this purpose OR there any
other (better) methods of communication??

If you need fast IPC forget about signals. They are way too slow.

The traditional UNIX way for userspace to talk to the kernel has been ioctls. For various reasons ioctls are not highly regarded in the Linux kernel community.

I believe the currently favored method is to have the driver create sysfs entries which userspace read()s and write()s.

Really, your question is too vague. It would help if you said what exactly you are trying to accomplish.

Lee

#6 lore raw
In reply to Benjamin L. Shi

Hello Benjamin,

Quoted 8 lines · Benjamin L. Shi

We've seen these, by adding the following tueables resolved the problem. More specifically, the lower zone protection made the difference.

vm.vfs_cache_pressure=1000
vm.lower_zone_protection=100
vm.max_map_count = 32668
vm.min_free_kbytes = 10000

many thanks, we will test this now and set those values on all of our 2.6. systems.

Thanks a lot again,
Bernd