/proc/meminfo swap counter wraparound in 2.2

/proc/meminfo swap counter wraparound in 2.2

[Sean Hunter]

Hi there

Having put 8Gb of swap on one of our production database servers[1], I found
that "top", "free" et al report the amount of available swap space incorrectly.

This is because they all parse /proc/meminfo, and its counters seem to wrap
when they encounter very high numbers.  The relevant bit is the totalswap
member of "struct sysinfo"

sean@lisa:~$ cat /proc/swaps
Filename                        Type            Size    Used    Priority
/dev/rd/c0d2p1                  partition       530104  41436   10
/usr/SWAPFILE1                  file            1048568 56      2
/usr/local/SWAPFILE1            file            1048568 52      2
/usr/SWAPFILE2                  file            1572856 0       3
/usr/local/SWAPFILE2            file            1572856 0       3
/var/SWAPFILE2                  file            1572856 0       3
/home/SWAPFILE2                 file            1572856 0       -6

sean@lisa:~$ cat /proc/meminfo 
        total:    used:    free:  shared: buffers:  cached:
Mem:  4125683712 2820616192 1305067520        0 82919424 232722432
Swap: 542777344 42541056 500236288
MemTotal:   4028988 kB
MemFree:    1274480 kB
MemShared:        0 kB
Buffers:      80976 kB
Cached:      227268 kB
BigTotal:   3111908 kB
BigFree:    1266988 kB
SwapTotal:   530056 kB
SwapFree:    488512 kB

sean@lisa:~$ uname -a
Linux lisa.sportingbet.com 2.2.19-6.2.7enterprise #1 SMP Thu Jun 14 07:34:12 EDT 2001 i686 unknown

As you can see, we are running redhat's "enterprise" 2.2.19 kernel.

I would be amazed if this bug were not also in the main 2.2.x tree.  Is a fix
likely or even possible in 2.2 ?

Sean

[1]One of several changes to avoid random oracle 600 errors under heavy load.

Re: /proc/meminfo swap counter wraparound in 2.2

[Alan Cox]

> I would be amazed if this bug were not also in the main 2.2.x tree.  Is a fix
> likely or even possible in 2.2 ?

No fix is planned for 2.2

Re: /proc/meminfo swap counter wraparound in 2.2

[Sean Hunter]

On Wed, Sep 12, 2001 at 11:07:55AM +0100, Alan Cox wrote:
> > I would be amazed if this bug were not also in the main 2.2.x tree.  Is a fix
> > likely or even possible in 2.2 ?
> 
> No fix is planned for 2.2
> 

OK, thanks.  I'm a little reluctant to move this box to 2.4 yet (before Monday
we had almost a year's uptime out of it- frequent reboots are frowned upon),
but looking at the 2.4 source, the relevant bits of fs/proc/array.c and
include/linux/kernel.h look exactly the same, so the problem could be there as
well.

Am I being dumb?

Sean

Re: /proc/meminfo swap counter wraparound in 2.2

[Sean Hunter]

On Wed, Sep 12, 2001 at 11:12:29AM +0100, Sean Hunter wrote:
> On Wed, Sep 12, 2001 at 11:07:55AM +0100, Alan Cox wrote:
> > > I would be amazed if this bug were not also in the main 2.2.x tree.  Is a fix
> > > likely or even possible in 2.2 ?
> > 
> > No fix is planned for 2.2
> > 
> 
> OK, thanks.  I'm a little reluctant to move this box to 2.4 yet (before Monday
> we had almost a year's uptime out of it- frequent reboots are frowned upon),
> but looking at the 2.4 source, the relevant bits of fs/proc/array.c and
> include/linux/kernel.h look exactly the same, so the problem could be there as
> well.
> 
> Am I being dumb?

Yes, of course I am.  Having now Actually Tried It Instead Of Just Talking
Bollocks[tm], I can confirm that large amounts of swap space seem to work just
fine in 2.4.

Sorry for wasting people's time.

Sean