unwanted disk access by the kernel?

I've been trying to set up my laptop for mobile use. I'm having a problem with unwanted disk activity - even when the system is completely idle, there is still an occasional trickle of disk writes (which prevents the poor hard drive from ever spinning down).

Yes, I thought this was a user-space issue too - but even booting into a bare-bones root environment does not stop the occasional disk access! Here is everything that's left:

PID USER VSZ RSS TIME STAT COMMAND WCHAN

    7 root         0    0 00:00:00 SW   [kupdated]       kupdate
    6 root         0    0 00:00:00 SW   [bdflush]        bdflush
    5 root         0    0 00:00:00 SW   [kswapd]         kswapd
    4 root         0    0 00:00:00 SWN  [ksoftirqd_CPU0] ksoftirqd
    1 root      1316  524 00:00:05 S    init [S]         select
    2 root         0    0 00:00:00 SW   [keventd]        context_thread
    3 root         0    0 00:00:00 SW   [kapmd]          apm_mainloop
    8 root         0    0 00:00:00 Z    [khubd <defunct> exit
  879 root      1316  524 00:00:00 S    init [S]         wait4

880 root 2556 1576 00:00:00 S \_ bash wait4
927 root 3524 1512 00:00:00 R \_ ps afx - -

If I manually spin down the disk, it always wakes up within 30 seconds or so. During the spin-up, kupdated goes into the 'D' state and blocks in wait_on_buffer(). This means it's writing dirty filesystem buffers, right? So who is doing the dirtying? I've eliminated all possible user-space sources of I/O! (strace confirms that NO user-space processes are doing I/O; they're all sleeping...)

(I'm running a stock Linus 2.4.18 kernel, with APM enabled. The system is Debian woody. All filesystems are ext2.)

Regards,
Dan

Are all filesystems mounted with the `noatime' mount option?

eww. Does khubd always do that? Does this patch make it behave?

@@ -908,6 +908,7 @@ static int usb_hub_thread(void *__hub)
 	 */
 
 	daemonize();
+	reparent_to_init();
 
 	/* Setup a nice name */
 	strcpy(current->comm, "khubd");

-

You may also want to mount your (root) filesystem(s) with the "noatime" option... check the linux laptops site for other tips.

----- Original Message -----
From: "Dan Maas" <[email protected]>
To: <[email protected]>
Sent: Thursday, March 14, 2002 10:36 PM
Subject: unwanted disk access by the kernel?

-SNIP-

Mounted with or without noatime ?

Thanks Jeremy and Alan - mounting the filesystems 'noatime,nodiratime' cleared up the last bits of disk activity. (I'm still curious why atime updates would be happening even though the system is as idle as I can make it without cutting the power... =)

By the way, if I enable 'APM makes CPU idle calls when idle,' I get a constant stream of 'apm_do_idle failed (3)' messages. APM also doesn't seem to be able to power the machine down... This is a Dell Inspiron 7500... Maybe I should try ACPI?

Regards,
Dan

ACPI is a bit experimental right now but if you want some fun then obviously the more people who break the ACPI code the better.

Alan

For my uses (basically just power down) ACPI has worked on 99% of my machines (mix of pii & piii).

Same here (I'd rather say 100%). Temperature reading seems to work in all machines as well. That's just machines I'm actually touching, though; so they tend not to be supermarket bargains.

On a halfway related note APM fails to power down my Siemens PCD-5ND (Pentium 75 laptop) (which obviously doesn't support ACPI).

The interesting thing is that the kernel shipped with RH 7.2 does power down the machine, just the one from kernel.org doesn't (2.4.16). (APM compiled as a module).

Is there some patch I could try?

So long,
Joe

"I use emacs, which might be thought of as a thermonuclear
 word processor."
-- Neal Stephenson, "In the beginning... was the command line"

Hi!

Get noflushd.

Philips Velo 1: 1"x4"x8", 300gram, 60, 12MB, 40bogomips, linux, mutt,
details at http://atrey.karlin.mff.cuni.cz/~pavel/velo/index.html.

///////