Re: [PATCH for Linux v2] virtio_balloon: Add VIRTIO_BALLOON_VQ_INFLATE_CONT to handle THP split issue

[teawater <teawaterz@linux.alibaba.com>]

> 2020年3月26日 15:17，Michael S. Tsirkin <mst@redhat.com> 写道：
> 
> First, either QEMU or Linux version of any interface changes
> should be copied to the virtio TC.
> 
> :

OK.

> 
> 
> On Mon, Mar 23, 2020 at 12:04:56AM +0800, Hui Zhu wrote:
>> The first version is in [1].
>> According to the comments from Michael and David, I updated the patch.
>> 1. Added a separate vq inflate_cont_vq to transport inflate continuous
>>   pages.
>> 2. Set all the pages in the continuous pages movable then they can be
>>   compaction.
>> 3. Added a new element pages_order to virtio_balloon_config.  It is the
>>   inflate pages order that is set by the QEMU.
>> 4. If the balloon cannot get any continuous pages from the system.
>>   Go back to use the single page to fill the balloon.
>> 5.  Use balloon_pages_alloc to allocate the single page and continuous
>>    pages.  Replace __GFP_NORETRY with __GFP_RETRY_MAYFAIL when allocating
>>    the continuous pages because it can increase the success rate of
>>    allocating large chunks of memory.
>> 
>> Following is the introduction of the function.
>> If the guest kernel has many fragmentation pages, use virtio_balloon
>> will split THP of QEMU when it calls MADV_DONTNEED madvise to release
>> the balloon pages.
>> This is an example in a VM with 1G memory 1CPU:
>> // This is the THP number before VM execution in the host.
>> // None use THP.
>> cat /proc/meminfo | grep AnonHugePages:
>> AnonHugePages:         0 kB
>> 
>> // After VM start, use usemem
>> // (https://git.kernel.org/pub/scm/linux/kernel/git/wfg/vm-scalability.git)
>> // punch-holes function generates 400m fragmentation pages in the guest
>> // kernel.
>> usemem --punch-holes -s -1 800m &
>> 
>> // This is the THP number after this command in the host.
>> // Some THP is used by VM because usemem will access 800M memory
>> // in the guest.
>> cat /proc/meminfo | grep AnonHugePages:
>> AnonHugePages:    976896 kB
>> 
>> // Connect to the QEMU monitor, setup balloon, and set it size to 600M.
>> (qemu) device_add virtio-balloon-pci,id=balloon1
>> (qemu) info balloon
>> balloon: actual=1024
>> (qemu) balloon 600
>> (qemu) info balloon
>> balloon: actual=600
>> 
>> // This is the THP number after inflate the balloon in the host.
>> cat /proc/meminfo | grep AnonHugePages:
>> AnonHugePages:    151552 kB
>> 
>> THP number decreased more than 800M.
>> The reason is usemem with punch-holes option will free every other
>> page after allocation.  Then 400M free memory inside the guest kernel
>> is fragmentation pages.
>> The guest kernel will use them to inflate the balloon.  When these
>> fragmentation pages are freed, THP will be split.
>> 
>> This commit tries to handle this with add a new flag
>> VIRTIO_BALLOON_VQ_INFLATE_CONT.
>> When this flag is set, the balloon will try to use continuous pages
>> inflate the balloon.  And the pages order is set to THP order.
>> Then THP pages will be freed together in the host.
>> This is an example in a VM with 1G memory 1CPU:
>> // This is the THP number before VM execution in the host.
>> // None use THP.
>> cat /proc/meminfo | grep AnonHugePages:
>> AnonHugePages:         0 kB
>> 
>> // After VM start, use usemem punch-holes function generates 400M
>> // fragmentation pages in the guest kernel.
>> usemem --punch-holes -s -1 800m &
>> 
>> // This is the THP number after this command in the host.
>> // Some THP is used by VM because usemem will access 800M memory
>> // in the guest.
>> cat /proc/meminfo | grep AnonHugePages:
>> AnonHugePages:    976896 kB
>> 
>> // Connect to the QEMU monitor, setup balloon, and set it size to 600M.
>> (qemu) device_add virtio-balloon-pci,id=balloon1,cont-pages=on
>> (qemu) info balloon
>> balloon: actual=1024
>> (qemu) balloon 600
>> (qemu) info balloon
>> balloon: actual=600
>> 
>> // This is the THP number after inflate the balloon in the host.
>> cat /proc/meminfo | grep AnonHugePages:
>> AnonHugePages:    610304 kB
>> 
>> The THP number decreases 358M.  This shows that
>> VIRTIO_BALLOON_VQ_INFLATE_CONT can help handle the THP split issue.
>> 
>> [1] https://lkml.org/lkml/2020/3/12/144
>> 
>> Signed-off-by: Hui Zhu <teawaterz@linux.alibaba.com>
> 
> I'd like to repeat my original idea of doing large page allocations
> unconditionally within guest.
> 

If so, should it transfer the large page information using inflate_vq?

> 
>> ---
>> drivers/virtio/virtio_balloon.c     | 78 ++++++++++++++++++++++++++++++-------
>> include/linux/balloon_compaction.h  |  9 ++++-
>> include/uapi/linux/virtio_balloon.h |  3 ++
>> mm/balloon_compaction.c             | 40 +++++++++++++++----
>> 4 files changed, 109 insertions(+), 21 deletions(-)
>> 
>> diff --git a/drivers/virtio/virtio_balloon.c b/drivers/virtio/virtio_balloon.c
>> index 341458f..fbd2b02f 100644
>> --- a/drivers/virtio/virtio_balloon.c
>> +++ b/drivers/virtio/virtio_balloon.c
>> @@ -47,6 +47,7 @@ enum virtio_balloon_vq {
>> 	VIRTIO_BALLOON_VQ_DEFLATE,
>> 	VIRTIO_BALLOON_VQ_STATS,
>> 	VIRTIO_BALLOON_VQ_FREE_PAGE,
>> +	VIRTIO_BALLOON_VQ_INFLATE_CONT,
>> 	VIRTIO_BALLOON_VQ_MAX
>> };
>> 
>> @@ -56,7 +57,8 @@ enum virtio_balloon_config_read {
>> 
>> struct virtio_balloon {
>> 	struct virtio_device *vdev;
>> -	struct virtqueue *inflate_vq, *deflate_vq, *stats_vq, *free_page_vq;
>> +	struct virtqueue *inflate_vq, *deflate_vq, *stats_vq, *free_page_vq,
>> +			 *inflate_cont_vq;
>> 
>> 	/* Balloon's own wq for cpu-intensive work items */
>> 	struct workqueue_struct *balloon_wq;
>> @@ -114,6 +116,12 @@ struct virtio_balloon {
>> 
>> 	/* To register a shrinker to shrink memory upon memory pressure */
>> 	struct shrinker shrinker;
>> +
>> +	/* Pages order if VIRTIO_BALLOON_F_CONT_PAGES is set.
>> +	 * if VIRTIO_BALLOON_F_CONT_PAGES is not set,
>> +	 * pages_order will be set to 0.
>> +	 */
>> +	__u32 pages_order;
>> };
>> 
>> static struct virtio_device_id id_table[] = {
>> @@ -169,10 +177,11 @@ static void set_page_pfns(struct virtio_balloon *vb,
>> 					  page_to_balloon_pfn(page) + i);
>> }
>> 
>> -static unsigned fill_balloon(struct virtio_balloon *vb, size_t num)
>> +static unsigned int fill_balloon_order(struct virtio_balloon *vb, size_t num,
>> +				       unsigned int order)
>> {
>> -	unsigned num_allocated_pages;
>> -	unsigned num_pfns;
>> +	unsigned int num_allocated_pages;
>> +	unsigned int num_pfns;
>> 	struct page *page;
>> 	LIST_HEAD(pages);
>> 
>> @@ -181,12 +190,12 @@ static unsigned fill_balloon(struct virtio_balloon *vb, size_t num)
>> 
>> 	for (num_pfns = 0; num_pfns < num;
>> 	     num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
>> -		struct page *page = balloon_page_alloc();
>> +		struct page *page = balloon_pages_alloc(order);
>> 
>> 		if (!page) {
>> 			dev_info_ratelimited(&vb->vdev->dev,
>> -					     "Out of puff! Can't get %u pages\n",
>> -					     VIRTIO_BALLOON_PAGES_PER_PAGE);
>> +				"Out of puff! Can't get %u pages\n",
>> +				VIRTIO_BALLOON_PAGES_PER_PAGE << order);
>> 			/* Sleep for at least 1/5 of a second before retry. */
>> 			msleep(200);
>> 			break;
>> @@ -200,25 +209,48 @@ static unsigned fill_balloon(struct virtio_balloon *vb, size_t num)
>> 	vb->num_pfns = 0;
>> 
>> 	while ((page = balloon_page_pop(&pages))) {
>> -		balloon_page_enqueue(&vb->vb_dev_info, page);
>> +		balloon_pages_enqueue(&vb->vb_dev_info, page, order);
>> 
>> 		set_page_pfns(vb, vb->pfns + vb->num_pfns, page);
>> -		vb->num_pages += VIRTIO_BALLOON_PAGES_PER_PAGE;
>> +		vb->num_pages += VIRTIO_BALLOON_PAGES_PER_PAGE << order;
>> 		if (!virtio_has_feature(vb->vdev,
>> 					VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
>> -			adjust_managed_page_count(page, -1);
>> +			adjust_managed_page_count(page, -(1 << order));
>> 		vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE;
>> 	}
>> 
>> -	num_allocated_pages = vb->num_pfns;
>> +	num_allocated_pages = vb->num_pfns << order;
>> 	/* Did we get any? */
>> -	if (vb->num_pfns != 0)
>> -		tell_host(vb, vb->inflate_vq);
>> +	if (vb->num_pfns != 0) {
>> +		if (order)
>> +			tell_host(vb, vb->inflate_cont_vq);
>> +		else
>> +			tell_host(vb, vb->inflate_vq);
>> +	}
>> 	mutex_unlock(&vb->balloon_lock);
>> 
>> 	return num_allocated_pages;
>> }
>> 
>> +static unsigned int fill_balloon(struct virtio_balloon *vb, size_t num)
>> +{
>> +	unsigned int num_allocated_pages = 0;
>> +
>> +	if (vb->pages_order) {
>> +		num_allocated_pages = fill_balloon_order(vb, num,
>> +							 vb->pages_order);
>> +		if (!num_allocated_pages)
>> +			dev_info_ratelimited(&vb->vdev->dev,
>> +				"Fail get any continuous pages in order %u\n",
>> +				vb->pages_order);
>> +	}
>> +
>> +	if (!num_allocated_pages)
>> +		num_allocated_pages = fill_balloon_order(vb, num, 0);
>> +
>> +	return num_allocated_pages;
>> +}
>> +
>> static void release_pages_balloon(struct virtio_balloon *vb,
>> 				 struct list_head *pages)
>> {
> 
> 
> Hmm no this is not what was previously suggested. David Hildenbrand proposed
> starting with highest order and iterating until we reach 0.
> And I proposed a new API to allocate some memory between min and max
> orders.
> 
> 
>> @@ -481,6 +513,8 @@ static int init_vqs(struct virtio_balloon *vb)
>> 	names[VIRTIO_BALLOON_VQ_STATS] = NULL;
>> 	callbacks[VIRTIO_BALLOON_VQ_FREE_PAGE] = NULL;
>> 	names[VIRTIO_BALLOON_VQ_FREE_PAGE] = NULL;
>> +	callbacks[VIRTIO_BALLOON_VQ_INFLATE_CONT] = NULL;
>> +	names[VIRTIO_BALLOON_VQ_INFLATE_CONT] = NULL;
>> 
>> 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) {
>> 		names[VIRTIO_BALLOON_VQ_STATS] = "stats";
>> @@ -492,6 +526,11 @@ static int init_vqs(struct virtio_balloon *vb)
>> 		callbacks[VIRTIO_BALLOON_VQ_FREE_PAGE] = NULL;
>> 	}
>> 
>> +	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_CONT_PAGES)) {
>> +		names[VIRTIO_BALLOON_VQ_INFLATE_CONT] = "inflate_cont";
>> +		callbacks[VIRTIO_BALLOON_VQ_INFLATE_CONT] = balloon_ack;
>> +	}
>> +
>> 	err = vb->vdev->config->find_vqs(vb->vdev, VIRTIO_BALLOON_VQ_MAX,
>> 					 vqs, callbacks, names, NULL, NULL);
>> 	if (err)
>> @@ -524,6 +563,10 @@ static int init_vqs(struct virtio_balloon *vb)
>> 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
>> 		vb->free_page_vq = vqs[VIRTIO_BALLOON_VQ_FREE_PAGE];
>> 
>> +	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_CONT_PAGES))
>> +		vb->inflate_cont_vq
>> +			= vqs[VIRTIO_BALLOON_VQ_INFLATE_CONT];
>> +
>> 	return 0;
>> }
>> 
>> @@ -945,6 +988,14 @@ static int virtballoon_probe(struct virtio_device *vdev)
>> 		if (err)
>> 			goto out_del_balloon_wq;
>> 	}
>> +	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_CONT_PAGES)) {
>> +		virtio_cread(vb->vdev, struct virtio_balloon_config,
>> +			     pages_order, &vb->pages_order);
>> +		if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1))
>> +			vb->pages_order
>> +				= le32_to_cpu((__force __le32)vb->pages_order);
>> +	} else
>> +		vb->pages_order = 0;
>> 	virtio_device_ready(vdev);
>> 
>> 	if (towards_target(vb))
>> @@ -1058,6 +1109,7 @@ static unsigned int features[] = {
>> 	VIRTIO_BALLOON_F_DEFLATE_ON_OOM,
>> 	VIRTIO_BALLOON_F_FREE_PAGE_HINT,
>> 	VIRTIO_BALLOON_F_PAGE_POISON,
>> +	VIRTIO_BALLOON_F_CONT_PAGES,
>> };
>> 
>> static struct virtio_driver virtio_balloon_driver = {
>> diff --git a/include/linux/balloon_compaction.h b/include/linux/balloon_compaction.h
>> index 338aa27..8180bbf 100644
>> --- a/include/linux/balloon_compaction.h
>> +++ b/include/linux/balloon_compaction.h
>> @@ -60,15 +60,22 @@ struct balloon_dev_info {
>> 	struct inode *inode;
>> };
>> 
>> -extern struct page *balloon_page_alloc(void);
>> +extern struct page *balloon_pages_alloc(unsigned int order);
>> extern void balloon_page_enqueue(struct balloon_dev_info *b_dev_info,
>> 				 struct page *page);
>> +extern void balloon_pages_enqueue(struct balloon_dev_info *b_dev_info,
>> +				  struct page *page, unsigned int order);
>> extern struct page *balloon_page_dequeue(struct balloon_dev_info *b_dev_info);
>> extern size_t balloon_page_list_enqueue(struct balloon_dev_info *b_dev_info,
>> 				      struct list_head *pages);
>> extern size_t balloon_page_list_dequeue(struct balloon_dev_info *b_dev_info,
>> 				     struct list_head *pages, size_t n_req_pages);
>> 
>> +static inline struct page *balloon_page_alloc(void)
>> +{
>> +	return balloon_pages_alloc(0);
>> +}
>> +
>> static inline void balloon_devinfo_init(struct balloon_dev_info *balloon)
>> {
>> 	balloon->isolated_pages = 0;
>> diff --git a/include/uapi/linux/virtio_balloon.h b/include/uapi/linux/virtio_balloon.h
>> index a1966cd7..e21b0aa 100644
>> --- a/include/uapi/linux/virtio_balloon.h
>> +++ b/include/uapi/linux/virtio_balloon.h
>> @@ -36,6 +36,7 @@
>> #define VIRTIO_BALLOON_F_DEFLATE_ON_OOM	2 /* Deflate balloon on OOM */
>> #define VIRTIO_BALLOON_F_FREE_PAGE_HINT	3 /* VQ to report free pages */
>> #define VIRTIO_BALLOON_F_PAGE_POISON	4 /* Guest is using page poisoning */
>> +#define VIRTIO_BALLOON_F_CONT_PAGES	5 /* VQ to report continuous pages */
>> 
>> /* Size of a PFN in the balloon interface. */
>> #define VIRTIO_BALLOON_PFN_SHIFT 12
>> @@ -51,6 +52,8 @@ struct virtio_balloon_config {
>> 	__u32 free_page_report_cmd_id;
>> 	/* Stores PAGE_POISON if page poisoning is in use */
>> 	__u32 poison_val;
>> +	/* Pages order if VIRTIO_BALLOON_F_CONT_PAGES is set */
>> +	__u32 pages_order;
>> };
>> 
>> #define VIRTIO_BALLOON_S_SWAP_IN  0   /* Amount of memory swapped in */
>> diff --git a/mm/balloon_compaction.c b/mm/balloon_compaction.c
>> index 26de020..87df4b2 100644
>> --- a/mm/balloon_compaction.c
>> +++ b/mm/balloon_compaction.c
>> @@ -112,8 +112,8 @@ size_t balloon_page_list_dequeue(struct balloon_dev_info *b_dev_info,
>> EXPORT_SYMBOL_GPL(balloon_page_list_dequeue);
>> 
>> /*
>> - * balloon_page_alloc - allocates a new page for insertion into the balloon
>> - *			page list.
>> + * balloon_pages_alloc - allocates a new page for insertion into the balloon
>> + *			 page list.
>>  *
>>  * Driver must call this function to properly allocate a new balloon page.
>>  * Driver must call balloon_page_enqueue before definitively removing the page
>> @@ -121,14 +121,21 @@ EXPORT_SYMBOL_GPL(balloon_page_list_dequeue);
>>  *
>>  * Return: struct page for the allocated page or NULL on allocation failure.
>>  */
>> -struct page *balloon_page_alloc(void)
>> +struct page *balloon_pages_alloc(unsigned int order)
>> {
>> -	struct page *page = alloc_page(balloon_mapping_gfp_mask() |
>> -				       __GFP_NOMEMALLOC | __GFP_NORETRY |
>> -				       __GFP_NOWARN);
>> +	gfp_t gfp_mask;
>> +
>> +	if (order > 1)
>> +		gfp_mask = __GFP_RETRY_MAYFAIL;
>> +	else
>> +		gfp_mask = __GFP_NORETRY;
>> +
>> +	struct page *page = alloc_pages(balloon_mapping_gfp_mask() |
>> +					gfp_mask | __GFP_NOMEMALLOC |
>> +					__GFP_NOWARN, order);
>> 	return page;
>> }
>> -EXPORT_SYMBOL_GPL(balloon_page_alloc);
>> +EXPORT_SYMBOL_GPL(balloon_pages_alloc);
>> 
>> /*
>>  * balloon_page_enqueue - inserts a new page into the balloon page list.
>> @@ -155,6 +162,25 @@ void balloon_page_enqueue(struct balloon_dev_info *b_dev_info,
>> EXPORT_SYMBOL_GPL(balloon_page_enqueue);
>> 
>> /*
>> + * balloon_pages_enqueue - inserts continuous pages into the balloon page list.
>> + */
>> +void balloon_pages_enqueue(struct balloon_dev_info *b_dev_info,
>> +			   struct page *page, unsigned int order)
>> +{
>> +	unsigned long flags;
>> +	unsigned long pfn = page_to_pfn(page);
>> +	unsigned long last_pfn = pfn + (1 << order) - 1;
>> +
>> +	spin_lock_irqsave(&b_dev_info->pages_lock, flags);
>> +	for (; pfn <= last_pfn; pfn++) {
>> +		page = pfn_to_page(pfn);
>> +		balloon_page_enqueue_one(b_dev_info, page);
>> +	}
>> +	spin_unlock_irqrestore(&b_dev_info->pages_lock, flags);
>> +}
>> +EXPORT_SYMBOL_GPL(balloon_pages_enqueue);
>> +
>> +/*
>>  * balloon_page_dequeue - removes a page from balloon's page list and returns
>>  *			  its address to allow the driver to release the page.
>>  * @b_dev_info: balloon device decriptor where we will grab a page from.
>> -- 
>> 2.7.4