Re: [PATCH v3] memory-hotplug: add automatic onlining policy for the newly added memory

[Daniel Kiper]

Hey Vitaly,

Sorry for late reply but I have been on holiday and I was not able
to follow this thread.

I saw that Andrew put this patch in -mm queue but I have still some
concerns. I hope that they could be addressed in one way or another.
Please look below for more details.

On Thu, Jan 07, 2016 at 06:23:41PM +0100, Vitaly Kuznetsov wrote:
> Currently, all newly added memory blocks remain in 'offline' state unless
> someone onlines them, some linux distributions carry special udev rules
> like:
>
> SUBSYSTEM=="memory", ACTION=="add", ATTR{state}=="offline", ATTR{state}="online"
>
> to make this happen automatically. This is not a great solution for virtual
> machines where memory hotplug is being used to address high memory pressure
> situations as such onlining is slow and a userspace process doing this
> (udev) has a chance of being killed by the OOM killer as it will probably
> require to allocate some memory.
>
> Introduce default policy for the newly added memory blocks in
> /sys/devices/system/memory/auto_online_blocks file with two possible
> values: "offline" which preserves the current behavior and "online" which
> causes all newly added memory blocks to go online as soon as they're added.
> The default is "offline".
>
> Cc: Jonathan Corbet <corbet@xxxxxxx>
> Cc: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
> Cc: Daniel Kiper <daniel.kiper@xxxxxxxxxx>
> Cc: Dan Williams <dan.j.williams@xxxxxxxxx>
> Cc: Tang Chen <tangchen@xxxxxxxxxxxxxx>
> Cc: David Vrabel <david.vrabel@xxxxxxxxxx>
> Cc: David Rientjes <rientjes@xxxxxxxxxx>
> Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> Cc: Naoya Horiguchi <n-horiguchi@xxxxxxxxxxxxx>
> Cc: Xishi Qiu <qiuxishi@xxxxxxxxxx>
> Cc: Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx>
> Cc: "K. Y. Srinivasan" <kys@xxxxxxxxxxxxx>
> Cc: Igor Mammedov <imammedo@xxxxxxxxxx>
> Cc: Kay Sievers <kay@xxxxxxxx>
> Cc: Konrad Rzeszutek Wilk <konrad.wilk@xxxxxxxxxx>
> Cc: Boris Ostrovsky <boris.ostrovsky@xxxxxxxxxx>
> Signed-off-by: Vitaly Kuznetsov <vkuznets@xxxxxxxxxx>
> ---
> - Changes since 'v2':
>   - Remove config option, revert to 'offline' by default [Andrew Morton,
>     David Rientjes]
>   - Rename 'hotplug_autoonline' to 'auto_online_blocks' [David Rientjes]
>
> - Changes since 'v1':
>   Add 'online' parameter to add_memory_resource() as it is being used by
>   xen ballon driver and it adds "empty" memory pages [David Vrabel].
>   (I don't completely understand what prevents manual onlining in this
>    case as we still have all newly added blocks in sysfs ... this is the
>    discussion point.)

Sometimes Xen guest needs to share some memory with other guests. This
memory must live in an address space and must be described by struct page.
Original mechanism gets such struct page by ballooning down guest memory a bit.
However, if memory hotplug is available then we do not need to steal real
memory from guest to share memory. We can get address space and struct page
from memory hotplug machinery. However, there is one requirement: every
page shared between guests cannot be allocated (onlined) in normal way.
All of them must be taken from hypervisor and later described by struct page.
Otherwise, if we online whole section/block automatically then we must use
ballooning mechanism described earlier.

David, please correct me if I am wrong.

I hope that helps.

> - Changes since 'RFC':
>   It seems nobody is strongly opposed to the idea, thus non-RFC.
>   Change memhp_autoonline to bool, we support only MMOP_ONLINE_KEEP
>   and MMOP_OFFLINE for the auto-onlining policy, eliminate 'unknown'
>   from show_memhp_autoonline(). [Daniel Kiper]
>   Put everything under CONFIG_MEMORY_HOTPLUG_AUTOONLINE, enable the
>   feature by default (when the config option is selected) and add
>   kernel parameter (nomemhp_autoonline) to disable the functionality
>   upon boot when needed.
>
> - RFC:
>   I was able to find previous attempts to fix the issue, e.g.:
>   http://marc.info/?l=linux-kernel&m=137425951924598&w=2
>   http://marc.info/?l=linux-acpi&m=127186488905382
>   but I'm not completely sure why it didn't work out and the solution
>   I suggest is not 'smart enough', thus 'RFC'.
> ---
>  Documentation/memory-hotplug.txt | 19 +++++++++++++++----
>  drivers/base/memory.c            | 32 ++++++++++++++++++++++++++++++++
>  drivers/xen/balloon.c            |  2 +-
>  include/linux/memory_hotplug.h   |  4 +++-
>  mm/memory_hotplug.c              | 12 ++++++++++--
>  5 files changed, 61 insertions(+), 8 deletions(-)
>
> diff --git a/Documentation/memory-hotplug.txt b/Documentation/memory-hotplug.txt
> index ce2cfcf..ceaf40c 100644
> --- a/Documentation/memory-hotplug.txt
> +++ b/Documentation/memory-hotplug.txt
> @@ -254,12 +254,23 @@ If the memory block is online, you'll read "online".
>  If the memory block is offline, you'll read "offline".
>
>
> -5.2. How to online memory
> +5.2. Memory onlining
>  ------------
> -Even if the memory is hot-added, it is not at ready-to-use state.
> -For using newly added memory, you have to "online" the memory block.
> +When the memory is hot-added, the kernel decides whether or not to "online"
> +it according to the policy which can be read from "auto_online_blocks" file:
>
> -For onlining, you have to write "online" to the memory block's state file as:
> +% cat /sys/devices/system/memory/auto_online_blocks
> +
> +The default is "offline" which means the newly added memory is not in a
> +ready-to-use state and you have to "online" the newly added memory blocks
> +manually. Automatic onlining can be requested by writing "online" to
> +"auto_online_blocks" file:
> +
> +% echo online > /sys/devices/system/memory/auto_online_blocks
> +
> +If the automatic onlining wasn't requested or some memory block was offlined
> +it is possible to change the individual block's state by writing to the "state"
> +file:
>
>  % echo online > /sys/devices/system/memory/memoryXXX/state

Please say clearly that offlined blocks are not onlined automatically
when /sys/devices/system/memory/auto_online_blocks is set to online.

> diff --git a/drivers/base/memory.c b/drivers/base/memory.c
> index 25425d3..44a618d 100644
> --- a/drivers/base/memory.c
> +++ b/drivers/base/memory.c
> @@ -439,6 +439,37 @@ print_block_size(struct device *dev, struct device_attribute *attr,
>  static DEVICE_ATTR(block_size_bytes, 0444, print_block_size, NULL);
>
>  /*
> + * Memory auto online policy.
> + */
> +
> +static ssize_t
> +show_auto_online_blocks(struct device *dev, struct device_attribute *attr,
> +			char *buf)
> +{
> +	if (memhp_auto_online)
> +		return sprintf(buf, "online\n");
> +	else
> +		return sprintf(buf, "offline\n");
> +}
> +
> +static ssize_t
> +store_auto_online_blocks(struct device *dev, struct device_attribute *attr,
> +			 const char *buf, size_t count)
> +{
> +	if (sysfs_streq(buf, "online"))
> +		memhp_auto_online = true;
> +	else if (sysfs_streq(buf, "offline"))
> +		memhp_auto_online = false;
> +	else
> +		return -EINVAL;
> +
> +	return count;
> +}
> +
> +static DEVICE_ATTR(auto_online_blocks, 0644, show_auto_online_blocks,
> +		   store_auto_online_blocks);
> +
> +/*
>   * Some architectures will have custom drivers to do this, and
>   * will not need to do it from userspace.  The fake hot-add code
>   * as well as ppc64 will do all of their discovery in userspace
> @@ -737,6 +768,7 @@ static struct attribute *memory_root_attrs[] = {
>  #endif
>
>  	&dev_attr_block_size_bytes.attr,
> +	&dev_attr_auto_online_blocks.attr,
>  	NULL
>  };
>
> diff --git a/drivers/xen/balloon.c b/drivers/xen/balloon.c
> index 12eab50..890c3b5 100644
> --- a/drivers/xen/balloon.c
> +++ b/drivers/xen/balloon.c
> @@ -338,7 +338,7 @@ static enum bp_state reserve_additional_memory(void)
>  	}
>  #endif
>
> -	rc = add_memory_resource(nid, resource);
> +	rc = add_memory_resource(nid, resource, false);

This is partial solution and does not allow us to use new feature in Xen.
Could you add separate patch which fixes this issue?

>  	if (rc) {
>  		pr_warn("Cannot add additional memory (%i)\n", rc);
>  		goto err;
> diff --git a/include/linux/memory_hotplug.h b/include/linux/memory_hotplug.h
> index 2ea574f..4b7949a 100644
> --- a/include/linux/memory_hotplug.h
> +++ b/include/linux/memory_hotplug.h
> @@ -99,6 +99,8 @@ extern void __online_page_free(struct page *page);
>
>  extern int try_online_node(int nid);
>
> +extern bool memhp_auto_online;
> +
>  #ifdef CONFIG_MEMORY_HOTREMOVE
>  extern bool is_pageblock_removable_nolock(struct page *page);
>  extern int arch_remove_memory(u64 start, u64 size);
> @@ -267,7 +269,7 @@ static inline void remove_memory(int nid, u64 start, u64 size) {}
>  extern int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
>  		void *arg, int (*func)(struct memory_block *, void *));
>  extern int add_memory(int nid, u64 start, u64 size);
> -extern int add_memory_resource(int nid, struct resource *resource);
> +extern int add_memory_resource(int nid, struct resource *resource, bool online);
>  extern int zone_for_memory(int nid, u64 start, u64 size, int zone_default,
>  		bool for_device);
>  extern int arch_add_memory(int nid, u64 start, u64 size, bool for_device);
> diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
> index a042a9d..0ecf860 100644
> --- a/mm/memory_hotplug.c
> +++ b/mm/memory_hotplug.c
> @@ -76,6 +76,9 @@ static struct {
>  #define memhp_lock_acquire()      lock_map_acquire(&mem_hotplug.dep_map)
>  #define memhp_lock_release()      lock_map_release(&mem_hotplug.dep_map)
>
> +bool memhp_auto_online;
> +EXPORT_SYMBOL_GPL(memhp_auto_online);
> +
>  void get_online_mems(void)
>  {
>  	might_sleep();
> @@ -1232,7 +1235,7 @@ int zone_for_memory(int nid, u64 start, u64 size, int zone_default,
>  }
>
>  /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
> -int __ref add_memory_resource(int nid, struct resource *res)
> +int __ref add_memory_resource(int nid, struct resource *res, bool online)
>  {
>  	u64 start, size;
>  	pg_data_t *pgdat = NULL;
> @@ -1292,6 +1295,11 @@ int __ref add_memory_resource(int nid, struct resource *res)
>  	/* create new memmap entry */
>  	firmware_map_add_hotplug(start, start + size, "System RAM");
>
> +	/* online pages if requested */
> +	if (online)
> +		online_pages(start >> PAGE_SHIFT, size >> PAGE_SHIFT,
> +			     MMOP_ONLINE_KEEP);
> +

This way we go in deadlock if auto online feature is enabled in Xen (this was
pointed out by David Vrabel). And we want to have it working out of the box.
So, I think that we should find proper solution. I suppose that we can schedule
a task here which auto online attached blocks. Hmmm... Not nice but should work.
Or maybe you have better idea how to fix this issue.

Daniel