Re: [[PATCH]] mm: khugepaged: recalculate min_free_kbytes after memory hotplug as expected by khugepaged

[Vijay Balakrishna]

On 9/10/2020 3:28 PM, Pavel Tatashin wrote:
> Hi Kirill,
>
> On Thu, Sep 10, 2020 at 6:01 PM Kirill A. Shutemov
> <kirill.shutemov@xxxxxxxxxxxxxxx> wrote:
> > On Thu, Sep 10, 2020 at 01:47:39PM -0700, Vijay Balakrishna wrote:
> > > When memory is hotplug added or removed the min_free_kbytes must be
> > > recalculated based on what is expected by khugepaged.  Currently
> > > after hotplug, min_free_kbytes will be set to a lower default and higher
> > > default set when THP enabled is lost. This leaves the system with small
> > > min_free_kbytes which isn't suitable for systems especially with network
> > > intensive loads.  Typical failure symptoms include HW WATCHDOG reset,
> > > soft lockup hang notices, NETDEVICE WATCHDOG timeouts, and OOM process
> > > kills.
> > >
> > > Fixes: f000565adb77 ("thp: set recommended min free kbytes")
> > >
> > > Signed-off-by: Vijay Balakrishna <vijayb@xxxxxxxxxxxxxxxxxxx>
> > > Cc: stable@xxxxxxxxxxxxxxx
> >
> > NAK. It would override min_free_kbytes set by user.
>
> Hi Kirill,
>
> Thank you for looking into this. How is this different from when
> khugepaged modifies it?
>
> echo always >/sys/kernel/mm/transparent_hugepage/enabled
>
> Which results in:
>
> start_stop_khugepaged
>    set_recommended_min_free_kbytes
>
> Which will also adjust min_free_kbytes according to hugepaged requirement?
>
> This bug that Vijay described is another hot-plug related issue that
> we have found on our system where we perform memory hot add and hot
> remove on every reboot.
>
> Thank you,
> Pasha

Thanks Kirill for taking a look and spotting it.

IIUC, it is an existing issue, we should fix it while we are at it. 
Otherwise with my propsed patch introduces a regression for hotplug 
memory consumers with THP enabled and min_free_kbytes set by user higher 
than calculated by THP.

Vijay