Re: [PATCH v5 3/5] mm,memory_hotplug: Add kernel boot option to enable memmap_on_memory

[Michal Hocko <mhocko@suse.com>]

On Wed 24-03-21 09:45:01, Oscar Salvador wrote:
> On Tue, Mar 23, 2021 at 11:47:53AM +0100, Michal Hocko wrote:
> > On Fri 19-03-21 10:26:33, Oscar Salvador wrote:
> > > Self stored memmap leads to a sparse memory situation which is unsuitable
> > > for workloads that requires large contiguous memory chunks, so make this
> > > an opt-in which needs to be explicitly enabled.
> > > 
> > > To control this, let memory_hotplug have its own memory space, as suggested
> > > by David, so we can add memory_hotplug.memmap_on_memory parameter.
> > > 
> > > Signed-off-by: Oscar Salvador <osalvador@suse.de>
> > > Reviewed-by: David Hildenbrand <david@redhat.com>
> > 
> > Acked-by: Michal Hocko <mhocko@suse.com>
> > 
> > I would just rephrased the help text to be less low level
> ...
> > 			When enabled, runtime hotplugged memory will
> > 			allocate its internal metadata (struct pages)
> > 			from the hotadded memory which will allow to
> > 			hotadd a lot of memory without requiring
> > 			additional memory to do so.
> > 			This feature is disabled by default because it
> > 			has some implication on large (e.g. GB)
> > 			allocations in some configurations (e.g. small
> > 			memory blocks).
> 
> Ok, this sounds good as well, and I guess it might suit best for what admin-guide
> is about.
> 
> > The memmap_on_memory can be dropped from the 1st patch IIUC and only
> > introduce it now.
> 
> It could be done, and I __think__ in some previous persion it was that way, but
> I am leaning to not do it.
> In the 1st patch, memmap_on_memory is false by default, so I see it as a preparatory
> step for later (this patchset) till it might be enabled.
> 
> Moreover, the big comment from mhp_support_memmap_on_memory() should change to not
> mention it, and change here again to reflect it.
> 
> All in all, I think it can stay, but maybe place a comment in the 1st patch above
> the variable saying something like "This is a noop now, it will be enabled later on"

I will leave that up to you. This is likely not worth a larger
discussion but it seems quite pointless to add a variable which never
changes. The resulting code might look different than you expect because
compiler is allowed to simply drop the whole condition.
 
> > > +
> > > +/*
> > > + * memory_hotplug.memmap_on_memory parameter
> > > + */
> > > +static bool memmap_on_memory __ro_after_init;
> > > +#ifdef CONFIG_MHP_MEMMAP_ON_MEMORY
> > > +module_param(memmap_on_memory, bool, 0444);
> > > +MODULE_PARM_DESC(memmap_on_memory, "Enable memmap on memory for memory hotplug");
> > > +#endif
> > 
> > I am not very much familiar with the machinery. Does this expose the
> > state to the userspace?
> 
> Kind of:
> 
> # ls /sys/module/memory_hotplug/parameters
> memmap_on_memory
> # cat /sys/module/memory_hotplug/parameters/memmap_on_memory 
> Y
> 
> But that is not really the state, but rather it shows whether the user
> opted-in the feature by passing "memory_hotplug.memmap_on_memory=yes".
> It might be that the user opted-in the feature, but it cannot be used at
> at runtime (e.g: mhp_support_memmap_on_memory() return false due to size !=
> memory_block_size())

Thanks for the clarification.
-- 
Michal Hocko
SUSE Labs