Re: [PATCH v2] mm: Only re-generate demotion targets when a numa node changes its N_CPU state

From: "Huang, Ying" <ying.huang@intel.com>
To: Oscar Salvador <osalvador@suse.de>
Cc: Andrew Morton <akpm@linux-foundation.org>,
	Dave Hansen <dave.hansen@linux.intel.com>,
	Abhishek Goel <huntbag@linux.vnet.ibm.com>,
	Baolin Wang <baolin.wang@linux.alibaba.com>,
	linux-mm@kvack.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH v2] mm: Only re-generate demotion targets when a numa node changes its N_CPU state
Date: Mon, 14 Mar 2022 09:03:59 +0800	[thread overview]
Message-ID: <87czip73b4.fsf@yhuang6-desk2.ccr.corp.intel.com> (raw)
In-Reply-To: <YisK2PEkKAqtZPfp@localhost.localdomain> (Oscar Salvador's message of "Fri, 11 Mar 2022 09:39:52 +0100")

Oscar Salvador <osalvador@suse.de> writes:

> On Fri, Mar 11, 2022 at 10:24:07AM +0800, Huang, Ying wrote:
>> It may be unnecessary to be fixed in this patch.  But I think we need to
>> cleanup the kernel config dependencies of the demotion code at some time.
>
> I am glad you brought this up because it is something I have been
> thinking about.
> I already added it in my to-do list, but I would do it in a separate
> patch if you do not mind.

Thanks!

> Now, let us try to untangle this mess:
>
>> 1. Even if !defined(CONFIG_HOTPLUG_CPU) &&
>>    !defined(CONFIG_MEMORY_HOTPLUG), we still need to allocate
>>    "node_demotion" and call set_migration_target_nodes() during boot time.
>> 
>> 2. If !defined(CONFIG_MEMORY_HOTPLUG), we don't need
>>    migrate_on_reclaim_callback().
>> 
>> 3. We need defined(CONFIG_NUMA) && defined(CONFIG_MIGRATION) for all
>>    these code.
>
> Back in the early versions [1] I asked whether we could have some
> scenario where this feature could be used when !CONFIG_MEMORY_HOTPLUG
> [2].
> The reason I was given is that in order to bind the expose PMEM memory
> as RAM (add_memory_driver_managed()), we need MEMORY_HOTPLUG.
>
> Now, as I said back then, I am not sure whether PMEM memory can be
> exposed as RAM by other means, but as it was pointed out back then,
> it really looks like we, at least, need CONFIG_MEMORY_HOTPLUG.
>
> Ok, so we have our first dependency: CONFIG_MEMORY_HOTPLUG.

On host machine, PMEM is always exposed via memory hotplug.  But later
on, we found that for guest system it's possible for PMEM to be exposed
as normal memory.

Best Regards,
Huang, Ying

> Now, about CONFIG_HOTPLUG_CPU, it seems that that is not a strong dependency,
> as we do not need cpu-hotplug in order to use the feature.
>
> We definitely need CONFIG_MIGRATION and CONFIG_NUMA though.
>
> So, we have something like:
>
> - Depends:
>   * CONFIG_NUMA           (obvius)
>   * CONFIG_MIGRATION      (to migrate between nodes)
>   * CONFIG_MEMORY_HOTPLUG (to expose PMEM as RAM)
>
> Sounds about right?
>
> [1] https://patchwork.kernel.org/project/linux-mm/patch/20210401183221.977831DE@viggo.jf.intel.com/#24099405
> [2] https://patchwork.kernel.org/project/linux-mm/patch/20210401183221.977831DE@viggo.jf.intel.com/#24103467