[mirror_ubuntu-jammy-kernel.git] / include / net / page_pool.h

/* SPDX-License-Identifier: GPL-2.0
 *
 * page_pool.h
 *	Author:	Jesper Dangaard Brouer <netoptimizer@brouer.com>
 *	Copyright (C) 2016 Red Hat, Inc.
 */

/**
 * DOC: page_pool allocator
 *
 * This page_pool allocator is optimized for the XDP mode that
 * uses one-frame-per-page, but have fallbacks that act like the
 * regular page allocator APIs.
 *
 * Basic use involve replacing alloc_pages() calls with the
 * page_pool_alloc_pages() call.  Drivers should likely use
 * page_pool_dev_alloc_pages() replacing dev_alloc_pages().
 *
 * API keeps track of in-flight pages, in-order to let API user know
 * when it is safe to dealloactor page_pool object.  Thus, API users
 * must make sure to call page_pool_release_page() when a page is
 * "leaving" the page_pool.  Or call page_pool_put_page() where
 * appropiate.  For maintaining correct accounting.
 *
 * API user must only call page_pool_put_page() once on a page, as it
 * will either recycle the page, or in case of elevated refcnt, it
 * will release the DMA mapping and in-flight state accounting.  We
 * hope to lift this requirement in the future.
 */
#ifndef _NET_PAGE_POOL_H
#define _NET_PAGE_POOL_H

#include <linux/mm.h> /* Needed by ptr_ring */
#include <linux/ptr_ring.h>
#include <linux/dma-direction.h>

#define PP_FLAG_DMA_MAP		BIT(0) /* Should page_pool do the DMA
					* map/unmap
					*/
#define PP_FLAG_DMA_SYNC_DEV	BIT(1) /* If set all pages that the driver gets
					* from page_pool will be
					* DMA-synced-for-device according to
					* the length provided by the device
					* driver.
					* Please note DMA-sync-for-CPU is still
					* device driver responsibility
					*/
#define PP_FLAG_PAGE_FRAG	BIT(2) /* for page frag feature */
#define PP_FLAG_ALL		(PP_FLAG_DMA_MAP |\
				 PP_FLAG_DMA_SYNC_DEV |\
				 PP_FLAG_PAGE_FRAG)

/*
 * Fast allocation side cache array/stack
 *
 * The cache size and refill watermark is related to the network
 * use-case.  The NAPI budget is 64 packets.  After a NAPI poll the RX
 * ring is usually refilled and the max consumed elements will be 64,
 * thus a natural max size of objects needed in the cache.
 *
 * Keeping room for more objects, is due to XDP_DROP use-case.  As
 * XDP_DROP allows the opportunity to recycle objects directly into
 * this array, as it shares the same softirq/NAPI protection.  If
 * cache is already full (or partly full) then the XDP_DROP recycles
 * would have to take a slower code path.
 */
#define PP_ALLOC_CACHE_SIZE	128
#define PP_ALLOC_CACHE_REFILL	64
struct pp_alloc_cache {
	u32 count;
	struct page *cache[PP_ALLOC_CACHE_SIZE];
};

struct page_pool_params {
	unsigned int	flags;
	unsigned int	order;
	unsigned int	pool_size;
	int		nid;  /* Numa node id to allocate from pages from */
	struct device	*dev; /* device, for DMA pre-mapping purposes */
	enum dma_data_direction dma_dir; /* DMA mapping direction */
	unsigned int	max_len; /* max DMA sync memory size */
	unsigned int	offset;  /* DMA addr offset */
};

struct page_pool {
	struct page_pool_params p;

	struct delayed_work release_dw;
	void (*disconnect)(void *);
	unsigned long defer_start;
	unsigned long defer_warn;

	u32 pages_state_hold_cnt;
	unsigned int frag_offset;
	struct page *frag_page;
	long frag_users;

	/*
	 * Data structure for allocation side
	 *
	 * Drivers allocation side usually already perform some kind
	 * of resource protection.  Piggyback on this protection, and
	 * require driver to protect allocation side.
	 *
	 * For NIC drivers this means, allocate a page_pool per
	 * RX-queue. As the RX-queue is already protected by
	 * Softirq/BH scheduling and napi_schedule. NAPI schedule
	 * guarantee that a single napi_struct will only be scheduled
	 * on a single CPU (see napi_schedule).
	 */
	struct pp_alloc_cache alloc ____cacheline_aligned_in_smp;

	/* Data structure for storing recycled pages.
	 *
	 * Returning/freeing pages is more complicated synchronization
	 * wise, because free's can happen on remote CPUs, with no
	 * association with allocation resource.
	 *
	 * Use ptr_ring, as it separates consumer and producer
	 * effeciently, it a way that doesn't bounce cache-lines.
	 *
	 * TODO: Implement bulk return pages into this structure.
	 */
	struct ptr_ring ring;

	atomic_t pages_state_release_cnt;

	/* A page_pool is strictly tied to a single RX-queue being
	 * protected by NAPI, due to above pp_alloc_cache. This
	 * refcnt serves purpose is to simplify drivers error handling.
	 */
	refcount_t user_cnt;

	u64 destroy_cnt;
};

struct page *page_pool_alloc_pages(struct page_pool *pool, gfp_t gfp);

static inline struct page *page_pool_dev_alloc_pages(struct page_pool *pool)
{
	gfp_t gfp = (GFP_ATOMIC | __GFP_NOWARN);

	return page_pool_alloc_pages(pool, gfp);
}

struct page *page_pool_alloc_frag(struct page_pool *pool, unsigned int *offset,
				  unsigned int size, gfp_t gfp);

static inline struct page *page_pool_dev_alloc_frag(struct page_pool *pool,
						    unsigned int *offset,
						    unsigned int size)
{
	gfp_t gfp = (GFP_ATOMIC | __GFP_NOWARN);

	return page_pool_alloc_frag(pool, offset, size, gfp);
}

/* get the stored dma direction. A driver might decide to treat this locally and
 * avoid the extra cache line from page_pool to determine the direction
 */
static
inline enum dma_data_direction page_pool_get_dma_dir(struct page_pool *pool)
{
	return pool->p.dma_dir;
}

bool page_pool_return_skb_page(struct page *page);

struct page_pool *page_pool_create(const struct page_pool_params *params);

#ifdef CONFIG_PAGE_POOL
void page_pool_destroy(struct page_pool *pool);
void page_pool_use_xdp_mem(struct page_pool *pool, void (*disconnect)(void *));
void page_pool_release_page(struct page_pool *pool, struct page *page);
void page_pool_put_page_bulk(struct page_pool *pool, void **data,
			     int count);
#else
static inline void page_pool_destroy(struct page_pool *pool)
{
}

static inline void page_pool_use_xdp_mem(struct page_pool *pool,
					 void (*disconnect)(void *))
{
}
static inline void page_pool_release_page(struct page_pool *pool,
					  struct page *page)
{
}

static inline void page_pool_put_page_bulk(struct page_pool *pool, void **data,
					   int count)
{
}
#endif

void page_pool_put_page(struct page_pool *pool, struct page *page,
			unsigned int dma_sync_size, bool allow_direct);

/* Same as above but will try to sync the entire area pool->max_len */
static inline void page_pool_put_full_page(struct page_pool *pool,
					   struct page *page, bool allow_direct)
{
	/* When page_pool isn't compiled-in, net/core/xdp.c doesn't
	 * allow registering MEM_TYPE_PAGE_POOL, but shield linker.
	 */
#ifdef CONFIG_PAGE_POOL
	page_pool_put_page(pool, page, -1, allow_direct);
#endif
}

/* Same as above but the caller must guarantee safe context. e.g NAPI */
static inline void page_pool_recycle_direct(struct page_pool *pool,
					    struct page *page)
{
	page_pool_put_full_page(pool, page, true);
}

#define PAGE_POOL_DMA_USE_PP_FRAG_COUNT	\
		(sizeof(dma_addr_t) > sizeof(unsigned long))

static inline dma_addr_t page_pool_get_dma_addr(struct page *page)
{
	dma_addr_t ret = page->dma_addr;

	if (PAGE_POOL_DMA_USE_PP_FRAG_COUNT)
		ret |= (dma_addr_t)page->dma_addr_upper << 16 << 16;

	return ret;
}

static inline void page_pool_set_dma_addr(struct page *page, dma_addr_t addr)
{
	page->dma_addr = addr;
	if (PAGE_POOL_DMA_USE_PP_FRAG_COUNT)
		page->dma_addr_upper = upper_32_bits(addr);
}

static inline void page_pool_set_frag_count(struct page *page, long nr)
{
	atomic_long_set(&page->pp_frag_count, nr);
}

static inline long page_pool_atomic_sub_frag_count_return(struct page *page,
							  long nr)
{
	long ret;

	/* As suggested by Alexander, atomic_long_read() may cover up the
	 * reference count errors, so avoid calling atomic_long_read() in
	 * the cases of freeing or draining the page_frags, where we would
	 * not expect it to match or that are slowpath anyway.
	 */
	if (__builtin_constant_p(nr) &&
	    atomic_long_read(&page->pp_frag_count) == nr)
		return 0;

	ret = atomic_long_sub_return(nr, &page->pp_frag_count);
	WARN_ON(ret < 0);
	return ret;
}

static inline bool is_page_pool_compiled_in(void)
{
#ifdef CONFIG_PAGE_POOL
	return true;
#else
	return false;
#endif
}

static inline bool page_pool_put(struct page_pool *pool)
{
	return refcount_dec_and_test(&pool->user_cnt);
}

/* Caller must provide appropriate safe context, e.g. NAPI. */
void page_pool_update_nid(struct page_pool *pool, int new_nid);
static inline void page_pool_nid_changed(struct page_pool *pool, int new_nid)
{
	if (unlikely(pool->p.nid != new_nid))
		page_pool_update_nid(pool, new_nid);
}

static inline void page_pool_ring_lock(struct page_pool *pool)
	__acquires(&pool->ring.producer_lock)
{
	if (in_serving_softirq())
		spin_lock(&pool->ring.producer_lock);
	else
		spin_lock_bh(&pool->ring.producer_lock);
}

static inline void page_pool_ring_unlock(struct page_pool *pool)
	__releases(&pool->ring.producer_lock)
{
	if (in_serving_softirq())
		spin_unlock(&pool->ring.producer_lock);
	else
		spin_unlock_bh(&pool->ring.producer_lock);
}

#endif /* _NET_PAGE_POOL_H */
Commit	Line	Data
ff7d6b27 JDB	1	/* SPDX-License-Identifier: GPL-2.0
	2	*
	3	* page_pool.h
	4	* Author: Jesper Dangaard Brouer <netoptimizer@brouer.com>
	5	* Copyright (C) 2016 Red Hat, Inc.
	6	*/
	7
	8	/**
	9	* DOC: page_pool allocator
	10	*
	11	* This page_pool allocator is optimized for the XDP mode that
	12	* uses one-frame-per-page, but have fallbacks that act like the
	13	* regular page allocator APIs.
	14	*
	15	* Basic use involve replacing alloc_pages() calls with the
	16	* page_pool_alloc_pages() call. Drivers should likely use
	17	* page_pool_dev_alloc_pages() replacing dev_alloc_pages().
	18	*
99c07c43 JDB	19	* API keeps track of in-flight pages, in-order to let API user know
	20	* when it is safe to dealloactor page_pool object. Thus, API users
	21	* must make sure to call page_pool_release_page() when a page is
	22	* "leaving" the page_pool. Or call page_pool_put_page() where
	23	* appropiate. For maintaining correct accounting.
ff7d6b27	24	*
99c07c43 JDB	25	* API user must only call page_pool_put_page() once on a page, as it
	26	* will either recycle the page, or in case of elevated refcnt, it
	27	* will release the DMA mapping and in-flight state accounting. We
	28	* hope to lift this requirement in the future.
ff7d6b27 JDB	29	*/
	30	#ifndef _NET_PAGE_POOL_H
	31	#define _NET_PAGE_POOL_H
	32
	33	#include <linux/mm.h> /* Needed by ptr_ring */
	34	#include <linux/ptr_ring.h>
	35	#include <linux/dma-direction.h>
	36
e68bc756 LB	37	#define PP_FLAG_DMA_MAP BIT(0) /* Should page_pool do the DMA
	38	* map/unmap
	39	*/
	40	#define PP_FLAG_DMA_SYNC_DEV BIT(1) /* If set all pages that the driver gets
	41	* from page_pool will be
	42	* DMA-synced-for-device according to
	43	* the length provided by the device
	44	* driver.
	45	* Please note DMA-sync-for-CPU is still
	46	* device driver responsibility
	47	*/
0e9d2a0a YL	48	#define PP_FLAG_PAGE_FRAG BIT(2) /* for page frag feature */
	49	#define PP_FLAG_ALL (PP_FLAG_DMA_MAP \|\
	50	PP_FLAG_DMA_SYNC_DEV \|\
	51	PP_FLAG_PAGE_FRAG)
ff7d6b27 JDB	52
	53	/*
	54	* Fast allocation side cache array/stack
	55	*
	56	* The cache size and refill watermark is related to the network
	57	* use-case. The NAPI budget is 64 packets. After a NAPI poll the RX
	58	* ring is usually refilled and the max consumed elements will be 64,
	59	* thus a natural max size of objects needed in the cache.
	60	*
	61	* Keeping room for more objects, is due to XDP_DROP use-case. As
	62	* XDP_DROP allows the opportunity to recycle objects directly into
	63	* this array, as it shares the same softirq/NAPI protection. If
	64	* cache is already full (or partly full) then the XDP_DROP recycles
	65	* would have to take a slower code path.
	66	*/
	67	#define PP_ALLOC_CACHE_SIZE 128
	68	#define PP_ALLOC_CACHE_REFILL 64
	69	struct pp_alloc_cache {
	70	u32 count;
be5dba25	71	struct page *cache[PP_ALLOC_CACHE_SIZE];
ff7d6b27 JDB	72	};
	73
	74	struct page_pool_params {
	75	unsigned int flags;
	76	unsigned int order;
	77	unsigned int pool_size;
	78	int nid; /* Numa node id to allocate from pages from */
	79	struct device dev; / device, for DMA pre-mapping purposes */
	80	enum dma_data_direction dma_dir; /* DMA mapping direction */
e68bc756 LB	81	unsigned int max_len; /* max DMA sync memory size */
e68bc756 LB	82	unsigned int offset; /* DMA addr offset */
ff7d6b27 JDB	83	};
	84
	85	struct page_pool {
ff7d6b27 JDB	86	struct page_pool_params p;
ff7d6b27 JDB	87
c3f812ce JL	88	struct delayed_work release_dw;
	89	void (disconnect)(void );
	90	unsigned long defer_start;
	91	unsigned long defer_warn;
	92
	93	u32 pages_state_hold_cnt;
53e0961d YL	94	unsigned int frag_offset;
	95	struct page *frag_page;
	96	long frag_users;
99c07c43	97
ff7d6b27 JDB	98	/*
	99	* Data structure for allocation side
	100	*
	101	* Drivers allocation side usually already perform some kind
	102	* of resource protection. Piggyback on this protection, and
	103	* require driver to protect allocation side.
	104	*
	105	* For NIC drivers this means, allocate a page_pool per
	106	* RX-queue. As the RX-queue is already protected by
	107	* Softirq/BH scheduling and napi_schedule. NAPI schedule
	108	* guarantee that a single napi_struct will only be scheduled
	109	* on a single CPU (see napi_schedule).
	110	*/
	111	struct pp_alloc_cache alloc ____cacheline_aligned_in_smp;
	112
	113	/* Data structure for storing recycled pages.
	114	*
	115	* Returning/freeing pages is more complicated synchronization
	116	* wise, because free's can happen on remote CPUs, with no
	117	* association with allocation resource.
	118	*
	119	* Use ptr_ring, as it separates consumer and producer
	120	* effeciently, it a way that doesn't bounce cache-lines.
	121	*
	122	* TODO: Implement bulk return pages into this structure.
	123	*/
	124	struct ptr_ring ring;
99c07c43 JDB	125
99c07c43 JDB	126	atomic_t pages_state_release_cnt;
1da4bbef IK	127
	128	/* A page_pool is strictly tied to a single RX-queue being
	129	* protected by NAPI, due to above pp_alloc_cache. This
	130	* refcnt serves purpose is to simplify drivers error handling.
	131	*/
	132	refcount_t user_cnt;
7c9e6942 JDB	133
7c9e6942 JDB	134	u64 destroy_cnt;
ff7d6b27 JDB	135	};
	136
	137	struct page page_pool_alloc_pages(struct page_pool pool, gfp_t gfp);
	138
	139	static inline struct page page_pool_dev_alloc_pages(struct page_pool pool)
	140	{
	141	gfp_t gfp = (GFP_ATOMIC \| __GFP_NOWARN);
	142
	143	return page_pool_alloc_pages(pool, gfp);
	144	}
	145
53e0961d YL	146	struct page page_pool_alloc_frag(struct page_pool pool, unsigned int *offset,
	147	unsigned int size, gfp_t gfp);
	148
	149	static inline struct page page_pool_dev_alloc_frag(struct page_pool pool,
	150	unsigned int *offset,
	151	unsigned int size)
	152	{
	153	gfp_t gfp = (GFP_ATOMIC \| __GFP_NOWARN);
	154
	155	return page_pool_alloc_frag(pool, offset, size, gfp);
	156	}
	157
bb005f2a IA	158	/* get the stored dma direction. A driver might decide to treat this locally and
	159	* avoid the extra cache line from page_pool to determine the direction
	160	*/
	161	static
	162	inline enum dma_data_direction page_pool_get_dma_dir(struct page_pool *pool)
	163	{
	164	return pool->p.dma_dir;
	165	}
	166
6a5bcd84 IA	167	bool page_pool_return_skb_page(struct page *page);
6a5bcd84 IA	168
ff7d6b27 JDB	169	struct page_pool page_pool_create(const struct page_pool_params params);
ff7d6b27 JDB	170
e54cfd7e	171	#ifdef CONFIG_PAGE_POOL
c3f812ce JL	172	void page_pool_destroy(struct page_pool *pool);
c3f812ce JL	173	void page_pool_use_xdp_mem(struct page_pool pool, void (disconnect)(void *));
458de8a9	174	void page_pool_release_page(struct page_pool pool, struct page page);
78862447 LB	175	void page_pool_put_page_bulk(struct page_pool pool, void *data,
78862447 LB	176	int count);
c3f812ce	177	#else
1da4bbef IK	178	static inline void page_pool_destroy(struct page_pool *pool)
1da4bbef IK	179	{
c3f812ce	180	}
1da4bbef	181
c3f812ce JL	182	static inline void page_pool_use_xdp_mem(struct page_pool *pool,
	183	void (disconnect)(void ))
	184	{
1da4bbef	185	}
458de8a9 IA	186	static inline void page_pool_release_page(struct page_pool *pool,
	187	struct page *page)
	188	{
	189	}
78862447 LB	190
	191	static inline void page_pool_put_page_bulk(struct page_pool pool, void *data,
	192	int count)
	193	{
	194	}
c3f812ce	195	#endif
1da4bbef	196
458de8a9 IA	197	void page_pool_put_page(struct page_pool pool, struct page page,
458de8a9 IA	198	unsigned int dma_sync_size, bool allow_direct);
ff7d6b27	199
458de8a9 IA	200	/* Same as above but will try to sync the entire area pool->max_len */
	201	static inline void page_pool_put_full_page(struct page_pool *pool,
	202	struct page *page, bool allow_direct)
ff7d6b27	203	{
57d0a1c1 JDB	204	/* When page_pool isn't compiled-in, net/core/xdp.c doesn't
	205	* allow registering MEM_TYPE_PAGE_POOL, but shield linker.
	206	*/
	207	#ifdef CONFIG_PAGE_POOL
458de8a9	208	page_pool_put_page(pool, page, -1, allow_direct);
57d0a1c1	209	#endif
ff7d6b27	210	}
458de8a9 IA	211
458de8a9 IA	212	/* Same as above but the caller must guarantee safe context. e.g NAPI */
ff7d6b27 JDB	213	static inline void page_pool_recycle_direct(struct page_pool *pool,
	214	struct page *page)
	215	{
458de8a9	216	page_pool_put_full_page(pool, page, true);
6bf071bf JDB	217	}
6bf071bf JDB	218
0e9d2a0a YL	219	#define PAGE_POOL_DMA_USE_PP_FRAG_COUNT \
	220	(sizeof(dma_addr_t) > sizeof(unsigned long))
	221
0afdeeed IA	222	static inline dma_addr_t page_pool_get_dma_addr(struct page *page)
0afdeeed IA	223	{
0e9d2a0a YL	224	dma_addr_t ret = page->dma_addr;
	225
	226	if (PAGE_POOL_DMA_USE_PP_FRAG_COUNT)
	227	ret \|= (dma_addr_t)page->dma_addr_upper << 16 << 16;
	228
9ddb3c14 MWO	229	return ret;
	230	}
	231
	232	static inline void page_pool_set_dma_addr(struct page *page, dma_addr_t addr)
	233	{
0e9d2a0a YL	234	page->dma_addr = addr;
	235	if (PAGE_POOL_DMA_USE_PP_FRAG_COUNT)
	236	page->dma_addr_upper = upper_32_bits(addr);
	237	}
	238
	239	static inline void page_pool_set_frag_count(struct page *page, long nr)
	240	{
	241	atomic_long_set(&page->pp_frag_count, nr);
	242	}
	243
	244	static inline long page_pool_atomic_sub_frag_count_return(struct page *page,
	245	long nr)
	246	{
	247	long ret;
	248
	249	/* As suggested by Alexander, atomic_long_read() may cover up the
	250	* reference count errors, so avoid calling atomic_long_read() in
	251	* the cases of freeing or draining the page_frags, where we would
	252	* not expect it to match or that are slowpath anyway.
	253	*/
	254	if (__builtin_constant_p(nr) &&
	255	atomic_long_read(&page->pp_frag_count) == nr)
	256	return 0;
	257
	258	ret = atomic_long_sub_return(nr, &page->pp_frag_count);
	259	WARN_ON(ret < 0);
	260	return ret;
0afdeeed IA	261	}
0afdeeed IA	262
57d0a1c1 JDB	263	static inline bool is_page_pool_compiled_in(void)
	264	{
	265	#ifdef CONFIG_PAGE_POOL
	266	return true;
	267	#else
	268	return false;
	269	#endif
	270	}
	271
1da4bbef IK	272	static inline bool page_pool_put(struct page_pool *pool)
	273	{
	274	return refcount_dec_and_test(&pool->user_cnt);
	275	}
	276
bc836748 SM	277	/* Caller must provide appropriate safe context, e.g. NAPI. */
	278	void page_pool_update_nid(struct page_pool *pool, int new_nid);
	279	static inline void page_pool_nid_changed(struct page_pool *pool, int new_nid)
	280	{
	281	if (unlikely(pool->p.nid != new_nid))
	282	page_pool_update_nid(pool, new_nid);
	283	}
78862447 LB	284
	285	static inline void page_pool_ring_lock(struct page_pool *pool)
	286	__acquires(&pool->ring.producer_lock)
	287	{
	288	if (in_serving_softirq())
	289	spin_lock(&pool->ring.producer_lock);
	290	else
	291	spin_lock_bh(&pool->ring.producer_lock);
	292	}
	293
	294	static inline void page_pool_ring_unlock(struct page_pool *pool)
	295	__releases(&pool->ring.producer_lock)
	296	{
	297	if (in_serving_softirq())
	298	spin_unlock(&pool->ring.producer_lock);
	299	else
	300	spin_unlock_bh(&pool->ring.producer_lock);
	301	}
	302
ff7d6b27	303	#endif /* _NET_PAGE_POOL_H */