Re: [RFC PATCH v3 0/13] memory-hotplug : hot-remove physical memory

From: Jiang Liu <liuj97@gmail.com>
To: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: len.brown@intel.com, wency@cn.fujitsu.com,
	linux-acpi@vger.kernel.org, linux-kernel@vger.kernel.org,
	linux-mm@kvack.org, paulus@samba.org, minchan.kim@gmail.com,
	kosaki.motohiro@jp.fujitsu.com, rientjes@google.com,
	Christoph Lameter <cl@linux.com>,
	linuxppc-dev@lists.ozlabs.org, akpm@linux-foundation.org
Subject: Re: [RFC PATCH v3 0/13] memory-hotplug : hot-remove physical memory
Date: Wed, 11 Jul 2012 00:50:11 +0800	[thread overview]
Message-ID: <4FFC5D43.7040206@gmail.com> (raw)
In-Reply-To: <4FFBFCAC.4010007@jp.fujitsu.com>

On 07/10/2012 05:58 PM, Yasuaki Ishimatsu wrote:
> Hi Christoph,
> 
> 2012/07/10 0:18, Christoph Lameter wrote:
>>
>> On Mon, 9 Jul 2012, Yasuaki Ishimatsu wrote:
>>
>>> Even if you apply these patches, you cannot remove the physical memory
>>> completely since these patches are still under development. I want you to
>>> cooperate to improve the physical memory hot-remove. So please review these
>>> patches and give your comment/idea.
>>
>> Could you at least give a method on how you want to do physical memory
>> removal?
> 
> We plan to release a dynamic hardware partitionable system. It will be
> able to hot remove/add a system board which included memory and cpu.
> But as you know, Linux does not support memory hot-remove on x86 box.
> So I try to develop it.
> 
> Current plan to hot remove system board is to use container driver.
> Thus I define the system board in ACPI DSDT table as a container device.
> It have supported hot-add a container device. And if container device
> has _EJ0 ACPI method, "eject" file to remove the container device is
> prepared as follow:
> 
> # ls -l /sys/bus/acpi/devices/ACPI0004\:01/eject
> --w-------. 1 root root 4096 Jul 10 18:19 /sys/bus/acpi/devices/ACPI0004:01/eject
> 
> When I hot-remove the container device, I echo 1 to the file as follow:
> 
> #echo 1 > /sys/bus/acpi/devices/ACPI0004\:02/eject
> 
> Then acpi_bus_trim() is called. And it calls acpi_memory_device_remove()
> for removing memory device. But the code does not do nothing.
> So I developed the continuation of the function.
> 
>> You would have to remove all objects from the range you want to
>> physically remove. That is only possible under special circumstances and
>> with a limited set of objects. Even if you exclusively use ZONE_MOVEABLE
>> you still may get cases where pages are pinned for a long time.
> 
> I know it. So my memory hot-remove plan is as follows:
> 
> 1. hot-added a system board
>    All memory which included the system board is offline.
> 
> 2. online the memory as removable page
>    The function has not supported yet. It is being developed by Lai as follow:
>    http://lkml.indiana.edu/hypermail/linux/kernel/1207.0/01478.html
>    If it is supported, I will be able to create movable memory.
> 
> 3. hot-remove the memory by container device's eject file
We have implemented a prototype to do physical node (mem + CPU + IOH) hotplug
for Itanium and is now porting it to x86. But with currently solution, memory
hotplug functionality may cause 10-20% performance decrease because we concentrate
all DMA/Normal memory to the first NUMA node, and all other NUMA nodes only
hosts ZONE_MOVABLE. We are working on solution to minimize the performance
drop now.

> 
> Thanks,
> Yasuaki Ishimatsu
> 
>>
>> I am not sure that these patches are useful unless we know where you are
>> going with this. If we end up with a situation where we still cannot
>> remove physical memory then this patchset is not helpful.
> 
> 
>