uptime reset after about 45 days

uptime reset after about 45 days

[Dave Brondsema]

After about 45 days or so, my uptime was reset. My idle time is correct.

$ cat /proc/uptime
94245.37 3686026.54

$ cat /proc/version
Linux version 2.4.20-gentoo-r1 (root@dpb2.resnet.calvin.edu) (gcc version 3.2.2)
#6 SMP Thu Apr 17 14:11:34 EDT 2003


-- 
Dave Brondsema 
dave@brondsema.net 
http://www.brondsema.net - personal 
http://www.splike.com - programming 
http://csx.calvin.edu - student org 

Re: uptime reset after about 45 days

[Maciej Zenczykowski]

> After about 45 days or so, my uptime was reset. My idle time is correct.
> 
> $ cat /proc/uptime
> 94245.37 3686026.54
> 
> $ cat /proc/version Linux version 2.4.20-gentoo-r1
> (root@dpb2.resnet.calvin.edu) (gcc version 3.2.2) #6 SMP Thu Apr 17
> 14:11:34 EDT 2003

Uptime is stored in jiffies which is 32bit on your arch, which results in 
an overflow after 2^32 clock ticks. TTTicks were 100 HZ till recently 
(overflow after 470 or so days) now, they're 1000 -> overflows after 45 
days.  Doesn't wreck anything except for uptime display - known problem, 
not worth the trouble fixing it would cause (64 bit values are 
non-atomic, unless MMX/SSE which isn't allowed in kernel) - however there 
is (if I'm not mistaken) a patch available wihich fixes this 'problem'.

However since it is only a matter of uptime display...

Cheers,
MaZe.

Re: uptime reset after about 45 days

[Miquel van Smoorenburg]

In article <Pine.LNX.4.44.0310310005090.11473-100000@gaia.cela.pl>,
Maciej Zenczykowski  <maze@cela.pl> wrote:
>> After about 45 days or so, my uptime was reset. My idle time is correct.
>> 
>> $ cat /proc/uptime
>> 94245.37 3686026.54
>> 
>> $ cat /proc/version Linux version 2.4.20-gentoo-r1
>> (root@dpb2.resnet.calvin.edu) (gcc version 3.2.2) #6 SMP Thu Apr 17
>> 14:11:34 EDT 2003
>
>Uptime is stored in jiffies which is 32bit on your arch, which results in 
>an overflow after 2^32 clock ticks. TTTicks were 100 HZ till recently 
>(overflow after 470 or so days) now, they're 1000 -> overflows after 45 
>days.  Doesn't wreck anything except for uptime display - known problem, 

No, that's only on 2.6, and it has been fixed in 2.6 too.
The 2.4 32 bits kernels run with HZ=100.

Sounds like the gentoo-kernel has just upped HZ to 1000 without
fixing these problems properly. That's .. disappointing.

Mike.

Re: uptime reset after about 45 days

[Robert Love]

On Thu, 2003-10-30 at 18:34, Miquel van Smoorenburg wrote:

> No, that's only on 2.6, and it has been fixed in 2.6 too.
> The 2.4 32 bits kernels run with HZ=100.
> 
> Sounds like the gentoo-kernel has just upped HZ to 1000 without
> fixing these problems properly. That's .. disappointing.

Yup.  Last I heard, they merged the variable-HZ patch, set HZ to 1000,
but did but merge Tim's 64-bit jiffies patch.

So they roll over in 49 days.

	Robert Love

Re: uptime reset after about 45 days

[Jakob Oestergaard]

On Fri, Oct 31, 2003 at 12:09:25AM +0100, Maciej Zenczykowski wrote:
...
> Uptime is stored in jiffies which is 32bit on your arch, which results in 
> an overflow after 2^32 clock ticks. TTTicks were 100 HZ till recently 
> (overflow after 470 or so days) now, they're 1000 -> overflows after 45 
> days.  Doesn't wreck anything except for uptime display - known problem, 
> not worth the trouble fixing it would cause (64 bit values are 
> non-atomic, unless MMX/SSE which isn't allowed in kernel) - however there 
> is (if I'm not mistaken) a patch available wihich fixes this 'problem'.
> 
> However since it is only a matter of uptime display...

For me it would mean that I got disturbed or woken up by an SMS every 45
/ (number_of_servers) = (low_number) days, because the monitoring system
sees that a server suddenly has a 'suspiciously low' uptime.

Fix the monitoring system to detect uptime wraps?

Perhaps.  It would be needed for Windows 95 as well, anyway.

Still, it's pretty darn pathetic to be required to include workarounds
in *Linux* apps that would otherwise only be needed for '95.

All in my humble oppinion of course.

-- 
................................................................
:   jakob@unthought.net   : And I see the elder races,         :
:.........................: putrid forms of man                :
:   Jakob Østergaard      : See him rise and claim the earth,  :
:        OZ9ABN           : his downfall is at hand.           :
:.........................:............{Konkhra}...............:

Re: uptime reset after about 45 days

[Gene Heskett]

On Friday 31 October 2003 05:37, Jakob Oestergaard wrote:
>On Fri, Oct 31, 2003 at 12:09:25AM +0100, Maciej Zenczykowski wrote:
>...
>
>> Uptime is stored in jiffies which is 32bit on your arch, which
>> results in an overflow after 2^32 clock ticks. TTTicks were 100 HZ
>> till recently (overflow after 470 or so days) now, they're 1000 ->
>> overflows after 45 days.  Doesn't wreck anything except for uptime
>> display - known problem, not worth the trouble fixing it would
>> cause (64 bit values are non-atomic, unless MMX/SSE which isn't
>> allowed in kernel) - however there is (if I'm not mistaken) a
>> patch available wihich fixes this 'problem'.
>>
>> However since it is only a matter of uptime display...
>
>For me it would mean that I got disturbed or woken up by an SMS
> every 45 / (number_of_servers) = (low_number) days, because the
> monitoring system sees that a server suddenly has a 'suspiciously
> low' uptime.
>
>Fix the monitoring system to detect uptime wraps?
>
>Perhaps.  It would be needed for Windows 95 as well, anyway.

Win95 needs more than an uptime patch at that point because the 
keyboard is dead and only a reboot fixes it.  We've been rebooting a 
Win95 wire capture machine every 45-46 days on a schedule because of 
that.

>Still, it's pretty darn pathetic to be required to include
> workarounds in *Linux* apps that would otherwise only be needed for
> '95.
>
>All in my humble oppinion of course.

-- 
Cheers, Gene
AMD K6-III@500mhz 320M
Athlon1600XP@1400mhz  512M
99.27% setiathome rank, not too shabby for a WV hillbilly
Yahoo.com attornies please note, additions to this message
by Gene Heskett are:
Copyright 2003 by Maurice Eugene Heskett, all rights reserved.

Re: uptime reset after about 45 days

[Mike Dresser]

On Fri, 31 Oct 2003, Gene Heskett wrote:

> Win95 needs more than an uptime patch at that point because the
> keyboard is dead and only a reboot fixes it.  We've been rebooting a
> Win95 wire capture machine every 45-46 days on a schedule because of
> that.

Have you seen this fix?

http://support.microsoft.com/default.aspx?scid=http://support.microsoft.com:80/support/kb/articles/q216/6/41.asp&NoWebContent=1

Mike

Re: uptime reset after about 45 days

[Tim Schmielau]

On Fri, 31 Oct 2003, Jakob Oestergaard wrote:

> Still, it's pretty darn pathetic to be required to include workarounds
> in *Linux* apps that would otherwise only be needed for '95.

There's nothing wrong with Linux here, it works out of the box.
If you choose to patch with an inappropriate set of patches, it breaks.

So either apply
  http://www.physik3.uni-rostock.de/tim/kernel/2.4/jiffies64-21.patch.gz
as well, or don't patch at all.

Just my two cents.

Tim

Re: uptime reset after about 45 days

[Tim Schmielau]

On Fri, 31 Oct 2003, Tim Schmielau wrote:

> So either apply
>   http://www.physik3.uni-rostock.de/tim/kernel/2.4/jiffies64-21.patch.gz
> as well, or don't patch at all.

... or just apply the combined patch to save you from fixing a few rejects 
by hand.

Robert, would you mind placing the combined patch beside the variable-HZ
patch in your kernel.org directory, to save the cluel^W unaware?

Thanks,
Tim

Re: uptime reset after about 45 days

[Robert Love]

On Fri, 2003-10-31 at 11:40, Tim Schmielau wrote:

> ... or just apply the combined patch to save you from fixing a few rejects 
> by hand.
> 
> Robert, would you mind placing the combined patch beside the variable-HZ
> patch in your kernel.org directory, to save the cluel^W unaware?

Good idea.  I put it up along with a note:

http://www.kernel.org/pub/linux/kernel/people/rml/variable-HZ/v2.4/

	Robert Love

Re: uptime reset after about 45 days

[Tim Schmielau]

On Fri, 31 Oct 2003, Robert Love wrote:
> On Fri, 2003-10-31 at 11:40, Tim Schmielau wrote:
> 
> > ... or just apply the combined patch to save you from fixing a few rejects 
> > by hand.
> > 
> > Robert, would you mind placing the combined patch beside the variable-HZ
> > patch in your kernel.org directory, to save the cluel^W unaware?
> 
> Good idea.  I put it up along with a note:
> 
> http://www.kernel.org/pub/linux/kernel/people/rml/variable-HZ/v2.4/

Oops, of course I intended to include the patch with my mail...
Thanks, and thank you for downloading it yourself.

Tim