diff mbox series

[v13,05/12] bcache: bch_nvmpg_free_pages() of the buddy allocator

Message ID 20211212170552.2812-6-colyli@suse.de (mailing list archive)
State New, archived
Headers show
Series bcache for 5.17: enable NVDIMM for bcache journal | expand

Commit Message

Coly Li Dec. 12, 2021, 5:05 p.m. UTC 
From: Jianpeng Ma <jianpeng.ma@intel.com>

This patch implements the bch_nvmpg_free_pages() of the buddy allocator.

The difference between this and page-buddy-free:
it need owner_uuid to free owner allocated pages, and must
persistent after free.

Signed-off-by: Jianpeng Ma <jianpeng.ma@intel.com>
Co-developed-by: Qiaowei Ren <qiaowei.ren@intel.com>
Signed-off-by: Qiaowei Ren <qiaowei.ren@intel.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Hannes Reinecke <hare@suse.de>
Cc: Jens Axboe <axboe@kernel.dk>
---
 drivers/md/bcache/nvmpg.c | 164 ++++++++++++++++++++++++++++++++++++--
 drivers/md/bcache/nvmpg.h |   3 +
 2 files changed, 160 insertions(+), 7 deletions(-)

Comments

Jens Axboe Dec. 12, 2021, 8:16 p.m. UTC | #1 
On 12/12/21 10:05 AM, Coly Li wrote:
> diff --git a/drivers/md/bcache/nvmpg.c b/drivers/md/bcache/nvmpg.c
> index a920779eb548..8ce0c4389b42 100644
> --- a/drivers/md/bcache/nvmpg.c
> +++ b/drivers/md/bcache/nvmpg.c
> @@ -248,6 +248,57 @@ static int init_nvmpg_set_header(struct bch_nvmpg_ns *ns)
>  	return rc;
>  }
>  
> +static void __free_space(struct bch_nvmpg_ns *ns, unsigned long nvmpg_offset,
> +			 int order)
> +{
> +	unsigned long add_pages = (1L << order);
> +	pgoff_t pgoff;
> +	struct page *page;
> +	void *va;
> +
> +	if (nvmpg_offset == 0) {
> +		pr_err("free pages on offset 0\n");
> +		return;
> +	}
> +
> +	page = bch_nvmpg_va_to_pg(bch_nvmpg_offset_to_ptr(nvmpg_offset));
> +	WARN_ON((!page) || (page->private != order));
> +	pgoff = page->index;
> +
> +	while (order < BCH_MAX_ORDER - 1) {
> +		struct page *buddy_page;
> +
> +		pgoff_t buddy_pgoff = pgoff ^ (1L << order);
> +		pgoff_t parent_pgoff = pgoff & ~(1L << order);
> +
> +		if ((parent_pgoff + (1L << (order + 1)) > ns->pages_total))
> +			break;
> +
> +		va = bch_nvmpg_pgoff_to_ptr(ns, buddy_pgoff);
> +		buddy_page = bch_nvmpg_va_to_pg(va);
> +		WARN_ON(!buddy_page);
> +
> +		if (PageBuddy(buddy_page) && (buddy_page->private == order)) {
> +			list_del((struct list_head *)&buddy_page->zone_device_data);
> +			__ClearPageBuddy(buddy_page);
> +			pgoff = parent_pgoff;
> +			order++;
> +			continue;
> +		}
> +		break;
> +	}
> +
> +	va = bch_nvmpg_pgoff_to_ptr(ns, pgoff);
> +	page = bch_nvmpg_va_to_pg(va);
> +	WARN_ON(!page);
> +	list_add((struct list_head *)&page->zone_device_data,
> +		 &ns->free_area[order]);
> +	page->index = pgoff;
> +	set_page_private(page, order);
> +	__SetPageBuddy(page);
> +	ns->free += add_pages;
> +}

There are 3 WARN_ON's in here. If you absolutely must use a WARN_ON,
then make them WARN_ON_ONCE(). Ditto in other spots in this patch.

> +void bch_nvmpg_free_pages(unsigned long nvmpg_offset, int order,
> +			  const char *uuid)
> +{
> +	struct bch_nvmpg_ns *ns;
> +	struct bch_nvmpg_head *head;
> +	struct bch_nvmpg_recs *recs;
> +	int r;
> +
> +	mutex_lock(&global_nvmpg_set->lock);
> +
> +	ns = global_nvmpg_set->ns_tbl[BCH_NVMPG_GET_NS_ID(nvmpg_offset)];
> +	if (!ns) {
> +		pr_err("can't find namespace by given kaddr from namespace\n");
> +		goto unlock;
> +	}
> +
> +	head = find_nvmpg_head(uuid, false);
> +	if (!head) {
> +		pr_err("can't found bch_nvmpg_head by uuid\n");
> +		goto unlock;
> +	}
> +
> +	recs = find_nvmpg_recs(ns, head, false);
> +	if (!recs) {
> +		pr_err("can't find bch_nvmpg_recs by uuid\n");
> +		goto unlock;
> +	}
> +
> +	r = remove_nvmpg_rec(recs, ns->sb->this_ns, nvmpg_offset, order);
> +	if (r < 0) {
> +		pr_err("can't find bch_nvmpg_rec\n");
> +		goto unlock;
> +	}
> +
> +	__free_space(ns, nvmpg_offset, order);
> +
> +unlock:
> +	mutex_unlock(&global_nvmpg_set->lock);
> +}

Again tons of useless error prints. Make them return a proper error
instad of just making things void...

> @@ -686,6 +835,7 @@ struct bch_nvmpg_ns *bch_register_namespace(const char *dev_path)
>  	ns->pages_offset = sb->pages_offset;
>  	ns->pages_total = sb->pages_total;
>  	ns->sb = sb;
> +	/* increase by __free_space() */
>  	ns->free = 0;
>  	ns->bdev = bdev;
>  	ns->set = global_nvmpg_set;

Does that hunk belong in here?
Coly Li Dec. 28, 2021, 5:29 a.m. UTC | #2 
On 12/13/21 4:16 AM, Jens Axboe wrote:
> On 12/12/21 10:05 AM, Coly Li wrote:
>> diff --git a/drivers/md/bcache/nvmpg.c b/drivers/md/bcache/nvmpg.c
>> index a920779eb548..8ce0c4389b42 100644
>> --- a/drivers/md/bcache/nvmpg.c
>> +++ b/drivers/md/bcache/nvmpg.c
>> @@ -248,6 +248,57 @@ static int init_nvmpg_set_header(struct bch_nvmpg_ns *ns)
>>   	return rc;
>>   }
>>   
>> +static void __free_space(struct bch_nvmpg_ns *ns, unsigned long nvmpg_offset,
>> +			 int order)
>> +{
>> +	unsigned long add_pages = (1L << order);
>> +	pgoff_t pgoff;
>> +	struct page *page;
>> +	void *va;
>> +
>> +	if (nvmpg_offset == 0) {
>> +		pr_err("free pages on offset 0\n");
>> +		return;
>> +	}
>> +
>> +	page = bch_nvmpg_va_to_pg(bch_nvmpg_offset_to_ptr(nvmpg_offset));
>> +	WARN_ON((!page) || (page->private != order));
>> +	pgoff = page->index;
>> +
>> +	while (order < BCH_MAX_ORDER - 1) {
>> +		struct page *buddy_page;
>> +
>> +		pgoff_t buddy_pgoff = pgoff ^ (1L << order);
>> +		pgoff_t parent_pgoff = pgoff & ~(1L << order);
>> +
>> +		if ((parent_pgoff + (1L << (order + 1)) > ns->pages_total))
>> +			break;
>> +
>> +		va = bch_nvmpg_pgoff_to_ptr(ns, buddy_pgoff);
>> +		buddy_page = bch_nvmpg_va_to_pg(va);
>> +		WARN_ON(!buddy_page);
>> +
>> +		if (PageBuddy(buddy_page) && (buddy_page->private == order)) {
>> +			list_del((struct list_head *)&buddy_page->zone_device_data);
>> +			__ClearPageBuddy(buddy_page);
>> +			pgoff = parent_pgoff;
>> +			order++;
>> +			continue;
>> +		}
>> +		break;
>> +	}
>> +
>> +	va = bch_nvmpg_pgoff_to_ptr(ns, pgoff);
>> +	page = bch_nvmpg_va_to_pg(va);
>> +	WARN_ON(!page);
>> +	list_add((struct list_head *)&page->zone_device_data,
>> +		 &ns->free_area[order]);
>> +	page->index = pgoff;
>> +	set_page_private(page, order);
>> +	__SetPageBuddy(page);
>> +	ns->free += add_pages;
>> +}
> There are 3 WARN_ON's in here. If you absolutely must use a WARN_ON,
> then make them WARN_ON_ONCE(). Ditto in other spots in this patch.
>
>> +void bch_nvmpg_free_pages(unsigned long nvmpg_offset, int order,
>> +			  const char *uuid)
>> +{
>> +	struct bch_nvmpg_ns *ns;
>> +	struct bch_nvmpg_head *head;
>> +	struct bch_nvmpg_recs *recs;
>> +	int r;
>> +
>> +	mutex_lock(&global_nvmpg_set->lock);
>> +
>> +	ns = global_nvmpg_set->ns_tbl[BCH_NVMPG_GET_NS_ID(nvmpg_offset)];
>> +	if (!ns) {
>> +		pr_err("can't find namespace by given kaddr from namespace\n");
>> +		goto unlock;
>> +	}
>> +
>> +	head = find_nvmpg_head(uuid, false);
>> +	if (!head) {
>> +		pr_err("can't found bch_nvmpg_head by uuid\n");
>> +		goto unlock;
>> +	}
>> +
>> +	recs = find_nvmpg_recs(ns, head, false);
>> +	if (!recs) {
>> +		pr_err("can't find bch_nvmpg_recs by uuid\n");
>> +		goto unlock;
>> +	}
>> +
>> +	r = remove_nvmpg_rec(recs, ns->sb->this_ns, nvmpg_offset, order);
>> +	if (r < 0) {
>> +		pr_err("can't find bch_nvmpg_rec\n");
>> +		goto unlock;
>> +	}
>> +
>> +	__free_space(ns, nvmpg_offset, order);
>> +
>> +unlock:
>> +	mutex_unlock(&global_nvmpg_set->lock);
>> +}
> Again tons of useless error prints. Make them return a proper error
> instad of just making things void...

Copied. It will be improved as you suggested.


>> @@ -686,6 +835,7 @@ struct bch_nvmpg_ns *bch_register_namespace(const char *dev_path)
>>   	ns->pages_offset = sb->pages_offset;
>>   	ns->pages_total = sb->pages_total;
>>   	ns->sb = sb;
>> +	/* increase by __free_space() */
>>   	ns->free = 0;
>>   	ns->bdev = bdev;
>>   	ns->set = global_nvmpg_set;
> Does that hunk belong in here?
>

Could you please give me more detailed explanation of the above 
question? Do you mention the bch_register_namespace() function, or the 
code block which initializes members of ns?

Thank you.

Coly Li
Yu Kuai Feb. 21, 2022, 12:36 p.m. UTC | #3 
在 2021/12/13 1:05, Coly Li 写道:
> From: Jianpeng Ma <jianpeng.ma@intel.com>
> 
> This patch implements the bch_nvmpg_free_pages() of the buddy allocator.
> 
> The difference between this and page-buddy-free:
> it need owner_uuid to free owner allocated pages, and must
> persistent after free.
> 
> Signed-off-by: Jianpeng Ma <jianpeng.ma@intel.com>
> Co-developed-by: Qiaowei Ren <qiaowei.ren@intel.com>
> Signed-off-by: Qiaowei Ren <qiaowei.ren@intel.com>
> Cc: Christoph Hellwig <hch@lst.de>
> Cc: Dan Williams <dan.j.williams@intel.com>
> Cc: Hannes Reinecke <hare@suse.de>
> Cc: Jens Axboe <axboe@kernel.dk>
> ---
>   drivers/md/bcache/nvmpg.c | 164 ++++++++++++++++++++++++++++++++++++--
>   drivers/md/bcache/nvmpg.h |   3 +
>   2 files changed, 160 insertions(+), 7 deletions(-)
> 
> diff --git a/drivers/md/bcache/nvmpg.c b/drivers/md/bcache/nvmpg.c
> index a920779eb548..8ce0c4389b42 100644
> --- a/drivers/md/bcache/nvmpg.c
> +++ b/drivers/md/bcache/nvmpg.c
> @@ -248,6 +248,57 @@ static int init_nvmpg_set_header(struct bch_nvmpg_ns *ns)
>   	return rc;
>   }
>   
> +static void __free_space(struct bch_nvmpg_ns *ns, unsigned long nvmpg_offset,
> +			 int order)
> +{
> +	unsigned long add_pages = (1L << order);
> +	pgoff_t pgoff;
> +	struct page *page;
> +	void *va;
> +
> +	if (nvmpg_offset == 0) {
> +		pr_err("free pages on offset 0\n");
> +		return;
> +	}
> +
> +	page = bch_nvmpg_va_to_pg(bch_nvmpg_offset_to_ptr(nvmpg_offset));
> +	WARN_ON((!page) || (page->private != order));
> +	pgoff = page->index;
> +
> +	while (order < BCH_MAX_ORDER - 1) {
> +		struct page *buddy_page;
> +
> +		pgoff_t buddy_pgoff = pgoff ^ (1L << order);
> +		pgoff_t parent_pgoff = pgoff & ~(1L << order);
> +
> +		if ((parent_pgoff + (1L << (order + 1)) > ns->pages_total))
> +			break;
> +
> +		va = bch_nvmpg_pgoff_to_ptr(ns, buddy_pgoff);
> +		buddy_page = bch_nvmpg_va_to_pg(va);
> +		WARN_ON(!buddy_page);
> +
> +		if (PageBuddy(buddy_page) && (buddy_page->private == order)) {
> +			list_del((struct list_head *)&buddy_page->zone_device_data);
> +			__ClearPageBuddy(buddy_page);
> +			pgoff = parent_pgoff;
> +			order++;
> +			continue;
> +		}
> +		break;
> +	}
> +
> +	va = bch_nvmpg_pgoff_to_ptr(ns, pgoff);
> +	page = bch_nvmpg_va_to_pg(va);
> +	WARN_ON(!page);
> +	list_add((struct list_head *)&page->zone_device_data,
> +		 &ns->free_area[order]);
> +	page->index = pgoff;
> +	set_page_private(page, order);
> +	__SetPageBuddy(page);
> +	ns->free += add_pages;
> +}
> +
>   static void bch_nvmpg_init_free_space(struct bch_nvmpg_ns *ns)
>   {
>   	unsigned int start, end, pages;
> @@ -261,21 +312,19 @@ static void bch_nvmpg_init_free_space(struct bch_nvmpg_ns *ns)
>   		pages = end - start;
>   
>   		while (pages) {
> -			void *addr;
> -
>   			for (i = BCH_MAX_ORDER - 1; i >= 0; i--) {
>   				if ((pgoff_start % (1L << i) == 0) &&
>   				    (pages >= (1L << i)))
>   					break;
>   			}
>   
> -			addr = bch_nvmpg_pgoff_to_ptr(ns, pgoff_start);
> -			page = bch_nvmpg_va_to_pg(addr);
> +			page = bch_nvmpg_va_to_pg(
> +					bch_nvmpg_pgoff_to_ptr(ns, pgoff_start));
>   			set_page_private(page, i);
>   			page->index = pgoff_start;
> -			__SetPageBuddy(page);
> -			list_add((struct list_head *)&page->zone_device_data,
> -				 &ns->free_area[i]);
> +
> +			/* In order to update ns->free */
> +			__free_space(ns, pgoff_start, i);

Hi,

While testing this patchset, we found that this is probably wrong,
pgoff_start represent page number here, however, __free_space expect
this to be page offset. Maybe this should be:

__free_space(ns, pgoff_start << PAGE_SHIFT, i);

Thanks,
Kuai
Ma, Jianpeng Feb. 22, 2022, 5:03 a.m. UTC | #4 
> -----Original Message-----
> From: yukuai (C) <yukuai3@huawei.com>
> Sent: Monday, February 21, 2022 8:36 PM
> To: Li, Coly <colyli@suse.de>; axboe@kernel.dk
> Cc: linux-bcache@vger.kernel.org; linux-block@vger.kernel.org; Ma, Jianpeng
> <jianpeng.ma@intel.com>; Ren, Qiaowei <qiaowei.ren@intel.com>; Christoph
> Hellwig <hch@lst.de>; Williams, Dan J <dan.j.williams@intel.com>; Hannes
> Reinecke <hare@suse.de>
> Subject: Re: [PATCH v13 05/12] bcache: bch_nvmpg_free_pages() of the buddy
> allocator
> 
> 在 2021/12/13 1:05, Coly Li 写道:
> > From: Jianpeng Ma <jianpeng.ma@intel.com>
> >
> > This patch implements the bch_nvmpg_free_pages() of the buddy allocator.
> >
> > The difference between this and page-buddy-free:
> > it need owner_uuid to free owner allocated pages, and must persistent
> > after free.
> >
> > Signed-off-by: Jianpeng Ma <jianpeng.ma@intel.com>
> > Co-developed-by: Qiaowei Ren <qiaowei.ren@intel.com>
> > Signed-off-by: Qiaowei Ren <qiaowei.ren@intel.com>
> > Cc: Christoph Hellwig <hch@lst.de>
> > Cc: Dan Williams <dan.j.williams@intel.com>
> > Cc: Hannes Reinecke <hare@suse.de>
> > Cc: Jens Axboe <axboe@kernel.dk>
> > ---
> >   drivers/md/bcache/nvmpg.c | 164
> ++++++++++++++++++++++++++++++++++++--
> >   drivers/md/bcache/nvmpg.h |   3 +
> >   2 files changed, 160 insertions(+), 7 deletions(-)
> >
> > diff --git a/drivers/md/bcache/nvmpg.c b/drivers/md/bcache/nvmpg.c
> > index a920779eb548..8ce0c4389b42 100644
> > --- a/drivers/md/bcache/nvmpg.c
> > +++ b/drivers/md/bcache/nvmpg.c
> > @@ -248,6 +248,57 @@ static int init_nvmpg_set_header(struct
> bch_nvmpg_ns *ns)
> >   	return rc;
> >   }
> >
> > +static void __free_space(struct bch_nvmpg_ns *ns, unsigned long
> nvmpg_offset,
> > +			 int order)
> > +{
> > +	unsigned long add_pages = (1L << order);
> > +	pgoff_t pgoff;
> > +	struct page *page;
> > +	void *va;
> > +
> > +	if (nvmpg_offset == 0) {
> > +		pr_err("free pages on offset 0\n");
> > +		return;
> > +	}
> > +
> > +	page = bch_nvmpg_va_to_pg(bch_nvmpg_offset_to_ptr(nvmpg_offset));
> > +	WARN_ON((!page) || (page->private != order));
> > +	pgoff = page->index;
> > +
> > +	while (order < BCH_MAX_ORDER - 1) {
> > +		struct page *buddy_page;
> > +
> > +		pgoff_t buddy_pgoff = pgoff ^ (1L << order);
> > +		pgoff_t parent_pgoff = pgoff & ~(1L << order);
> > +
> > +		if ((parent_pgoff + (1L << (order + 1)) > ns->pages_total))
> > +			break;
> > +
> > +		va = bch_nvmpg_pgoff_to_ptr(ns, buddy_pgoff);
> > +		buddy_page = bch_nvmpg_va_to_pg(va);
> > +		WARN_ON(!buddy_page);
> > +
> > +		if (PageBuddy(buddy_page) && (buddy_page->private == order)) {
> > +			list_del((struct list_head *)&buddy_page->zone_device_data);
> > +			__ClearPageBuddy(buddy_page);
> > +			pgoff = parent_pgoff;
> > +			order++;
> > +			continue;
> > +		}
> > +		break;
> > +	}
> > +
> > +	va = bch_nvmpg_pgoff_to_ptr(ns, pgoff);
> > +	page = bch_nvmpg_va_to_pg(va);
> > +	WARN_ON(!page);
> > +	list_add((struct list_head *)&page->zone_device_data,
> > +		 &ns->free_area[order]);
> > +	page->index = pgoff;
> > +	set_page_private(page, order);
> > +	__SetPageBuddy(page);
> > +	ns->free += add_pages;
> > +}
> > +
> >   static void bch_nvmpg_init_free_space(struct bch_nvmpg_ns *ns)
> >   {
> >   	unsigned int start, end, pages;
> > @@ -261,21 +312,19 @@ static void bch_nvmpg_init_free_space(struct
> bch_nvmpg_ns *ns)
> >   		pages = end - start;
> >
> >   		while (pages) {
> > -			void *addr;
> > -
> >   			for (i = BCH_MAX_ORDER - 1; i >= 0; i--) {
> >   				if ((pgoff_start % (1L << i) == 0) &&
> >   				    (pages >= (1L << i)))
> >   					break;
> >   			}
> >
> > -			addr = bch_nvmpg_pgoff_to_ptr(ns, pgoff_start);
> > -			page = bch_nvmpg_va_to_pg(addr);
> > +			page = bch_nvmpg_va_to_pg(
> > +					bch_nvmpg_pgoff_to_ptr(ns, pgoff_start));
> >   			set_page_private(page, i);
> >   			page->index = pgoff_start;
> > -			__SetPageBuddy(page);
> > -			list_add((struct list_head *)&page->zone_device_data,
> > -				 &ns->free_area[i]);
> > +
> > +			/* In order to update ns->free */
> > +			__free_space(ns, pgoff_start, i);
> 
> Hi,
> 
> While testing this patchset, we found that this is probably wrong, pgoff_start
> represent page number here, however, __free_space expect this to be page
> offset. Maybe this should be:
> 
> __free_space(ns, pgoff_start << PAGE_SHIFT, i);
> 
Yes，you are right. I'll fix in the next version.

Thanks!
Jianpeng.
> Thanks,
> Kuai
diff mbox series

Patch

diff --git a/drivers/md/bcache/nvmpg.c b/drivers/md/bcache/nvmpg.c
index a920779eb548..8ce0c4389b42 100644
--- a/drivers/md/bcache/nvmpg.c
+++ b/drivers/md/bcache/nvmpg.c
@@ -248,6 +248,57 @@  static int init_nvmpg_set_header(struct bch_nvmpg_ns *ns)
 	return rc;
 }
 
+static void __free_space(struct bch_nvmpg_ns *ns, unsigned long nvmpg_offset,
+			 int order)
+{
+	unsigned long add_pages = (1L << order);
+	pgoff_t pgoff;
+	struct page *page;
+	void *va;
+
+	if (nvmpg_offset == 0) {
+		pr_err("free pages on offset 0\n");
+		return;
+	}
+
+	page = bch_nvmpg_va_to_pg(bch_nvmpg_offset_to_ptr(nvmpg_offset));
+	WARN_ON((!page) || (page->private != order));
+	pgoff = page->index;
+
+	while (order < BCH_MAX_ORDER - 1) {
+		struct page *buddy_page;
+
+		pgoff_t buddy_pgoff = pgoff ^ (1L << order);
+		pgoff_t parent_pgoff = pgoff & ~(1L << order);
+
+		if ((parent_pgoff + (1L << (order + 1)) > ns->pages_total))
+			break;
+
+		va = bch_nvmpg_pgoff_to_ptr(ns, buddy_pgoff);
+		buddy_page = bch_nvmpg_va_to_pg(va);
+		WARN_ON(!buddy_page);
+
+		if (PageBuddy(buddy_page) && (buddy_page->private == order)) {
+			list_del((struct list_head *)&buddy_page->zone_device_data);
+			__ClearPageBuddy(buddy_page);
+			pgoff = parent_pgoff;
+			order++;
+			continue;
+		}
+		break;
+	}
+
+	va = bch_nvmpg_pgoff_to_ptr(ns, pgoff);
+	page = bch_nvmpg_va_to_pg(va);
+	WARN_ON(!page);
+	list_add((struct list_head *)&page->zone_device_data,
+		 &ns->free_area[order]);
+	page->index = pgoff;
+	set_page_private(page, order);
+	__SetPageBuddy(page);
+	ns->free += add_pages;
+}
+
 static void bch_nvmpg_init_free_space(struct bch_nvmpg_ns *ns)
 {
 	unsigned int start, end, pages;
@@ -261,21 +312,19 @@  static void bch_nvmpg_init_free_space(struct bch_nvmpg_ns *ns)
 		pages = end - start;
 
 		while (pages) {
-			void *addr;
-
 			for (i = BCH_MAX_ORDER - 1; i >= 0; i--) {
 				if ((pgoff_start % (1L << i) == 0) &&
 				    (pages >= (1L << i)))
 					break;
 			}
 
-			addr = bch_nvmpg_pgoff_to_ptr(ns, pgoff_start);
-			page = bch_nvmpg_va_to_pg(addr);
+			page = bch_nvmpg_va_to_pg(
+					bch_nvmpg_pgoff_to_ptr(ns, pgoff_start));
 			set_page_private(page, i);
 			page->index = pgoff_start;
-			__SetPageBuddy(page);
-			list_add((struct list_head *)&page->zone_device_data,
-				 &ns->free_area[i]);
+
+			/* In order to update ns->free */
+			__free_space(ns, pgoff_start, i);
 
 			pgoff_start += 1L << i;
 			pages -= 1L << i;
@@ -490,6 +539,106 @@  unsigned long bch_nvmpg_alloc_pages(int order, const char *uuid)
 	return nvmpg_offset;
 }
 
+static inline void *nvm_end_addr(struct bch_nvmpg_ns *ns)
+{
+	return ns->base_addr + (ns->pages_total << PAGE_SHIFT);
+}
+
+static inline bool in_nvmpg_ns_range(struct bch_nvmpg_ns *ns,
+				     void *start_addr, void *end_addr)
+{
+	return (start_addr >= ns->base_addr) && (end_addr < nvm_end_addr(ns));
+}
+
+static int remove_nvmpg_rec(struct bch_nvmpg_recs *recs, int ns_id,
+			    unsigned long nvmpg_offset, int order)
+{
+	struct bch_nvmpg_head *head;
+	struct bch_nvmpg_recs *prev_recs, *sys_recs;
+	struct bch_nvmpg_ns *ns;
+	unsigned long pgoff;
+	int i;
+
+	ns = global_nvmpg_set->ns_tbl[0];
+	pgoff = bch_nvmpg_offset_to_pgoff(nvmpg_offset);
+
+	head = bch_nvmpg_offset_to_ptr(recs->head_offset);
+	prev_recs = recs;
+	sys_recs = bch_nvmpg_offset_to_ptr(BCH_NVMPG_SYSRECS_OFFSET);
+	while (recs) {
+		for (i = 0; i < recs->size; i++) {
+			struct bch_nvmpg_rec *rec = &(recs->recs[i]);
+
+			if ((rec->pgoff == pgoff) && (rec->ns_id == ns_id)) {
+				WARN_ON(rec->order != order);
+				rec->_v = 0;
+				recs->used--;
+
+				if (recs->used == 0) {
+					int recs_pos = recs - sys_recs;
+
+					if (recs == prev_recs)
+						head->recs_offset[ns_id] =
+							recs->next_offset;
+					else
+						prev_recs->next_offset =
+							recs->next_offset;
+
+					recs->next_offset = 0;
+					recs->head_offset = 0;
+
+					bitmap_clear(ns->recs_bitmap, recs_pos, 1);
+				}
+				goto out;
+			}
+		}
+		prev_recs = recs;
+		recs = bch_nvmpg_offset_to_ptr(recs->next_offset);
+	}
+out:
+	return (recs ? 0 : -ENOENT);
+}
+
+void bch_nvmpg_free_pages(unsigned long nvmpg_offset, int order,
+			  const char *uuid)
+{
+	struct bch_nvmpg_ns *ns;
+	struct bch_nvmpg_head *head;
+	struct bch_nvmpg_recs *recs;
+	int r;
+
+	mutex_lock(&global_nvmpg_set->lock);
+
+	ns = global_nvmpg_set->ns_tbl[BCH_NVMPG_GET_NS_ID(nvmpg_offset)];
+	if (!ns) {
+		pr_err("can't find namespace by given kaddr from namespace\n");
+		goto unlock;
+	}
+
+	head = find_nvmpg_head(uuid, false);
+	if (!head) {
+		pr_err("can't found bch_nvmpg_head by uuid\n");
+		goto unlock;
+	}
+
+	recs = find_nvmpg_recs(ns, head, false);
+	if (!recs) {
+		pr_err("can't find bch_nvmpg_recs by uuid\n");
+		goto unlock;
+	}
+
+	r = remove_nvmpg_rec(recs, ns->sb->this_ns, nvmpg_offset, order);
+	if (r < 0) {
+		pr_err("can't find bch_nvmpg_rec\n");
+		goto unlock;
+	}
+
+	__free_space(ns, nvmpg_offset, order);
+
+unlock:
+	mutex_unlock(&global_nvmpg_set->lock);
+}
+
 static int attach_nvmpg_set(struct bch_nvmpg_ns *ns)
 {
 	struct bch_nvmpg_sb *sb = ns->sb;
@@ -686,6 +835,7 @@  struct bch_nvmpg_ns *bch_register_namespace(const char *dev_path)
 	ns->pages_offset = sb->pages_offset;
 	ns->pages_total = sb->pages_total;
 	ns->sb = sb;
+	/* increase by __free_space() */
 	ns->free = 0;
 	ns->bdev = bdev;
 	ns->set = global_nvmpg_set;
diff --git a/drivers/md/bcache/nvmpg.h b/drivers/md/bcache/nvmpg.h
index d03f3241b45a..e089936e7f13 100644
--- a/drivers/md/bcache/nvmpg.h
+++ b/drivers/md/bcache/nvmpg.h
@@ -93,6 +93,7 @@  struct bch_nvmpg_ns *bch_register_namespace(const char *dev_path);
 int bch_nvmpg_init(void);
 void bch_nvmpg_exit(void);
 unsigned long bch_nvmpg_alloc_pages(int order, const char *uuid);
+void bch_nvmpg_free_pages(unsigned long nvmpg_offset, int order, const char *uuid);
 
 #else
 
@@ -113,6 +114,8 @@  static inline unsigned long bch_nvmpg_alloc_pages(int order, const char *uuid)
 	return 0;
 }
 
+static inline void bch_nvmpg_free_pages(void *addr, int order, const char *uuid) { }
+
 #endif /* CONFIG_BCACHE_NVM_PAGES */
 
 #endif /* _BCACHE_NVM_PAGES_H */