RE: [linux-mm] Drastic increase in application memory usage with Kernel version upgrade

[PINTU KUMAR]

Hi,
> -----Original Message-----
> From: vinayak menon [mailto:vinayakm.list@xxxxxxxxx]
> Sent: Thursday, August 11, 2016 10:23 AM
> To: PINTU KUMAR
> Cc: Pavel Machek; Konstantin Khlebnikov; Minchan Kim; linux-
> kernel@xxxxxxxxxxxxxxx; linux-mm@xxxxxxxxx; jaejoon.seo@xxxxxxxxxxx;
> jy0.jeon@xxxxxxxxxxx; vishnu.ps@xxxxxxxxxxx; chulspro.kim@xxxxxxxxxxx
> Subject: Re: [linux-mm] Drastic increase in application memory usage with Kernel
> version upgrade
> 
> On Wed, Aug 10, 2016 at 6:56 PM, PINTU KUMAR <pintu.k@xxxxxxxxxxx> wrote:
> > Hi,
> >
> >> -----Original Message-----
> >> From: Pavel Machek [mailto:pavel@xxxxxx]
> >> Sent: Saturday, August 06, 2016 2:20 AM
> >> To: PINTU KUMAR
> >> Cc: 'Minchan Kim'; linux-kernel@xxxxxxxxxxxxxxx; linux-mm@xxxxxxxxx;
> >> jaejoon.seo@xxxxxxxxxxx; jy0.jeon@xxxxxxxxxxx;
> vishnu.ps@xxxxxxxxxxx
> >> Subject: Re: [linux-mm] Drastic increase in application memory usage
> >> with
> > Kernel
> >> version upgrade
> >>
> >> On Fri 2016-08-05 20:17:36, PINTU KUMAR wrote:
> >> > Hi,
> >>
> >> > > On Fri, Aug 05, 2016 at 10:26:37AM +0530, PINTU KUMAR wrote:
> >> > > > Hi All,
> >> > > >
> >> > > > For one of our ARM embedded product, we recently updated the
> >> > > > Kernel version from 3.4 to 3.18 and we noticed that the same
> >> > > > application memory usage  (PSS value) gone up by ~10% and for
> >> > > > some cases it even crossed ~50%. There is no change in platform
> >> > > > part. All platform component was  built with ARM 32-bit toolchain.
> >> > > > However, the Kernel is changed from 32-bit to 64-bit.
> >> > > >
> >> > > > Is upgrading Kernel version and moving from 32-bit to 64-bit is
> >> > > > such a risk?
> >> > > > After the upgrade, what can we do further to reduce the
> >> > > > application memory usage ?
> >> > > > Is there any other factor that will help us to improve without
> >> > > > major modifications in platform ?
> >> > > >
> >> > > > As a proof, we did a small experiment on our Ubuntu-32 bit machine.
> >> > > > We upgraded Ubuntu Kernel version from 3.13 to 4.01 and we
> >> > > > observed the following:
> >> > > > ---------------------------------------------------------------
> >> > > > ---
> >> > > > |UBUNTU-32 bit  |Kernel 3.13    |Kernel 4.03    |DIFF   |
> >> > > > |CALCULATOR PSS |6057 KB        |6466 KB        |409 KB |
> >> > > > ---------------------------------------------------------------
> >> > > > --- So, just by upgrading the Kernel version: PSS value for
> >> > > > calculator is increased by 409KB.
> >> > > >
> >> > > > If anybody knows any in-sight about it please point out more
> >> > > > details about the root cause.
> >> > >
> >> > > One of culprit is [8c6e50b0290c, mm: introduce vm_ops->map_pages()].
> >> > Ok. Thank you for your reply.
> >> > So, if I revert this patch, will the memory usage be decreased for
> >> > the processes with Kernel 3.18 ?
> >>
> >> I guess you should try it...
> >>
> > Thanks for the reply and confirmation.
> > Our exact kernel version is: 3.18.14
> > And, we already have this patch:
> > /*
> > mm: do not call do_fault_around for non-linear fault Ingo Korb
> > reported that "repeated mapping of the same file on tmpfs using
> > remap_file_pages sometimes triggers a BUG at mm/filemap.c:202 when the
> > process exits".
> > He bisected the bug to d7c1755179b8 ("mm: implement ->map_pages for
> > shmem/tmpfs"), although the bug was actually added by commit
> > 8c6e50b0290c ("mm: introduce vm_ops->map_pages()").
> > */
> >
> > So, I guess, reverting this patch (8c6e50b0290c), is not required ?
> > But, still we have memory usage issue.
> >
> I had observed the PSS increase with 3.18, and that was because of the
> faultaround patch which MInchan mentioned.
> Without reverting the patch you can just try reducing fault_around_bytes
> (mm/memory.c) to PAGE_SIZE. That should bring down the PSS.
> 
Thanks for your reply.
I tried changing fault_around_bytes value from 65536 to 4096.
But, still there is no change in PSS.
Please let me know if anything is missing.
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -2776,7 +2776,8 @@ void do_set_pte(struct vm_area_struct *vma, unsigned long address,
 }
 static unsigned long fault_around_bytes __read_mostly =
-       rounddown_pow_of_two(65536);
+       rounddown_pow_of_two(PAGE_SIZE);

> Thanks,
> Vinayak