Re: [RFC PATCH 3/5] mm, memory_hotplug: allocate memmap from the added memory range for sparse-vmemmap

From: Gerald Schaefer
Date: Wed Jul 26 2017 - 13:20:59 EST


On Wed, 26 Jul 2017 14:30:41 +0200
Michal Hocko <mhocko@xxxxxxxxxx> wrote:

> On Wed 26-07-17 13:45:39, Heiko Carstens wrote:
> [...]
> > In general I do like your idea, however if I understand your patches
> > correctly we might have an ordering problem on s390: it is not possible to
> > access hot-added memory on s390 before it is online (MEM_GOING_ONLINE
> > succeeded).
>
> Could you point me to the code please? I cannot seem to find the
> notifier which implements that.

It is in drivers/s390/char/sclp_cmd.c: sclp_mem_notifier().

>
> > On MEM_GOING_ONLINE we ask the hypervisor to back the potential available
> > hot-added memory region with physical pages. Accessing those ranges before
> > that will result in an exception.
>
> Can we make the range which backs the memmap range available? E.g from
> s390 specific __vmemmap_populate path?

No, only the complete range of a storage increment can be made available.
The size of those increments may vary between z/VM and LPAR, but at least
with LPAR it will always be minimum 256 MB, IIRC.

>
> > However with your approach the memory is still allocated when add_memory()
> > is being called, correct? That wouldn't be a change to the current
> > behaviour; except for the ordering problem outlined above.
>
> Could you be more specific please? I do not change when the memmap is
> allocated.

I guess this is about the difference between s390 and others, wrt when
we call add_memory(). It is also in drivers/s390/char/sclp_cmd.c, early
during memory detection, as opposed to other archs, where I guess this
could be triggered by an ACPI event during runtime, at least for newly
added and to-be-onlined memory.

This probably means that any approach that tries to allocate memmap
memory during add_memory(), out of the "to-be-onlined but still offline"
memory, will be difficult for s390, because we call add_memory() only once
during memory detection for the complete range of (hypervisor) defined
online and offline memory. The offline parts are then made available in
the MEM_GOING_ONLINE notifier called from online_pages(). Only after
this point the memory would then be available to allocate a memmap in it.

Nevertheless, we have great interest in such a "allocate memmap from
the added memory range" solution. I guess we would need some way to
separate the memmap allocation from add_memory(), which sounds odd,
or provide some way to have add_memory() only allocate a memmap for
online memory, and a mechanism to add the memmaps for offline memory
blocks later when they are being set online.

Regards,
Gerald