Re: [Patch v2] mm/hotplug: Only respect mem= parameter during boot stage

From: Baoquan He
Date: Tue Dec 10 2019 - 05:11:13 EST


On 12/10/19 at 10:50am, Michal Hocko wrote:
> On Tue 10-12-19 10:36:19, David Hildenbrand wrote:
> > On 10.12.19 10:24, Balbir Singh wrote:
> > >
> > >
> > > On 10/12/19 7:44 pm, Baoquan He wrote:
> > >> In commit 357b4da50a62 ("x86: respect memory size limiting via mem=
> > >> parameter") a global varialbe global max_mem_size is added to store
> > > typo ^^^
> > >> the value parsed from 'mem= ', then checked when memory region is
> > >> added. This truly stops those DIMM from being added into system memory
> > >> during boot-time.
> > >>
> > >> However, it also limits the later memory hotplug functionality. Any
> > >> memory board can't be hot added any more if its region is beyond the
> > >> max_mem_size. System will print error like below:
> > >>
> > >> [ 216.387164] acpi PNP0C80:02: add_memory failed
> > >> [ 216.389301] acpi PNP0C80:02: acpi_memory_enable_device() error
> > >> [ 216.392187] acpi PNP0C80:02: Enumeration failure
> > >>
> > >> From document of 'mem= ' parameter, it should be a restriction during
> > >> boot, but not impact the system memory adding/removing after booting.
> > >>
> > >> mem=nn[KMG] [KNL,BOOT] Force usage of a specific amount of memory
> > >> ...
> > >>
> > >> So fix it by also checking if it's during boot-time when restrict memory
> > >> adding. Otherwise, skip the restriction.
> > >>
> > >
> > > The fix looks reasonable, but I don't get the use case. Booting with mem= is
> > > generally a debug option, is this for debugging memory hotplug + limited memory?
> >
> > Some people/companies use "mem=" along with KVM e.g., to avoid
> > allocating memmaps for guest backing memory and to not expose it to the
> > buddy across kexec's. The excluded physical memory is then memmap into
> > the hypervisor process and KVM can deal with that. I can imagine that
> > hotplug might be desirable as well for such use cases.
>
> If this is really the usecase (it makes some sense to me) then it should
> be folded into the changelog. Because the real semantic is not really
> clear as I've pointed out in the previous version of this patch [1].
> The restriction to BOOT is documented since ever long before the memory
> hotplug was a thing.

I will hold a while and post v3 to take this into log.

>
> [1] Btw. it would have been much better if you posted the version 2 only
> after all the feedback got discussed properly.

You could have replied to wrong person.

I don't know the use case David told. My motivation was to make memory hotplug
not impacted after boot. I thought I have got your question and answered
it. I will wait a little longer when post next time.

Thanks
Baoquan