[Crash-utility] [PATCH] fix vmstat handling

[Dave Anderson]

Hi Vinayak,

Thanks for the clarification and secondary fix -- the vmstat patch is queued for crash-7.2.1:
 
  https://github.com/crash-utility/crash/commit/7ac1368cdca0fc2013bb3963456fcd2574c7cdd7

Dave


----- Original Message -----
> Hi Dave,
> 
> On Fri, Nov 3, 2017 at 11:32 PM, Dave Anderson <anderson at redhat.com> wrote:
> >
> >
> > ----- Original Message -----
> >> On Thu, Nov 2, 2017 at 8:58 PM, Dave Anderson <anderson at redhat.com> wrote:
> >> >
> >> >
> >> > ----- Original Message -----
> >> >> From: Vinayak Menon <vinayakm.list at gmail.com>
> >> >>
> >> >> With kernels where LRUs are moved to node and thus vm_stat is
> >> >> split into zone and node, crash utility fails to show the vmstat with
> >> >> "kmem
> >> >> -V", and the "CACHED" of kmem -i is shown as zero.
> >> >>
> >> >> Signed-off-by: Vinayak Menon <vinayakm.list at gmail.com>
> >> >
> >> > Nice -- thanks for catching this!
> >> >
> >> > But can you re-send the patch as an attachment?  It appears that the
> >> > inlined patch below has all tabs replaced with spaces:
> >> >
> >> Sure. Attached.
> >
> > Hi Vinayak,
> >
> > I understand that this kernel change was introduced in 4.8, and
> > I don't have a lot of 4.8 and later kernels to test with.
> >
> > But of the ones I do have, the 4.8, 4.13 and 4.14 kernels tested OK,
> > but both of my 4.11.0-based kernels show the same type of bogus
> > CACHED values:
> >
> >
> > Can you take a look at what happened in those dumpfiles?  I'm
> > guessing there's some negative number involved somewhere.
> >
> That turned out to be a different issue. swapper_space_nrpages was going
> wrong.
> After the following kernel change, the swap cache address space is
> split into many.
> 
> commit 4b3ef9daa4fc0bba742a79faecb17fdaaead083b
> Author: Huang, Ying <ying.huang at intel.com>
> Date:   Wed Feb 22 15:45:26 2017 -0800
> 
>     mm/swap: split swap cache into 64MB trunks
> 
> I have fixed it with the attached patch and with the 2 vmcores you had
> provided it looks okay.
> 
> Thanks,
> Vinayak
>