162306a36Sopenharmony_ci/* SPDX-License-Identifier: GPL-2.0
262306a36Sopenharmony_ci *
362306a36Sopenharmony_ci * page_pool/helpers.h
462306a36Sopenharmony_ci *	Author:	Jesper Dangaard Brouer <netoptimizer@brouer.com>
562306a36Sopenharmony_ci *	Copyright (C) 2016 Red Hat, Inc.
662306a36Sopenharmony_ci */
762306a36Sopenharmony_ci
862306a36Sopenharmony_ci/**
962306a36Sopenharmony_ci * DOC: page_pool allocator
1062306a36Sopenharmony_ci *
1162306a36Sopenharmony_ci * The page_pool allocator is optimized for the XDP mode that
1262306a36Sopenharmony_ci * uses one frame per-page, but it can fallback on the
1362306a36Sopenharmony_ci * regular page allocator APIs.
1462306a36Sopenharmony_ci *
1562306a36Sopenharmony_ci * Basic use involves replacing alloc_pages() calls with the
1662306a36Sopenharmony_ci * page_pool_alloc_pages() call.  Drivers should use
1762306a36Sopenharmony_ci * page_pool_dev_alloc_pages() replacing dev_alloc_pages().
1862306a36Sopenharmony_ci *
1962306a36Sopenharmony_ci * The API keeps track of in-flight pages, in order to let API users know
2062306a36Sopenharmony_ci * when it is safe to free a page_pool object.  Thus, API users
2162306a36Sopenharmony_ci * must call page_pool_put_page() to free the page, or attach
2262306a36Sopenharmony_ci * the page to a page_pool-aware object like skbs marked with
2362306a36Sopenharmony_ci * skb_mark_for_recycle().
2462306a36Sopenharmony_ci *
2562306a36Sopenharmony_ci * API users must call page_pool_put_page() once on a page, as it
2662306a36Sopenharmony_ci * will either recycle the page, or in case of refcnt > 1, it will
2762306a36Sopenharmony_ci * release the DMA mapping and in-flight state accounting.
2862306a36Sopenharmony_ci */
2962306a36Sopenharmony_ci#ifndef _NET_PAGE_POOL_HELPERS_H
3062306a36Sopenharmony_ci#define _NET_PAGE_POOL_HELPERS_H
3162306a36Sopenharmony_ci
3262306a36Sopenharmony_ci#include <net/page_pool/types.h>
3362306a36Sopenharmony_ci
3462306a36Sopenharmony_ci#ifdef CONFIG_PAGE_POOL_STATS
3562306a36Sopenharmony_ciint page_pool_ethtool_stats_get_count(void);
3662306a36Sopenharmony_ciu8 *page_pool_ethtool_stats_get_strings(u8 *data);
3762306a36Sopenharmony_ciu64 *page_pool_ethtool_stats_get(u64 *data, void *stats);
3862306a36Sopenharmony_ci
3962306a36Sopenharmony_ci/*
4062306a36Sopenharmony_ci * Drivers that wish to harvest page pool stats and report them to users
4162306a36Sopenharmony_ci * (perhaps via ethtool, debugfs, or another mechanism) can allocate a
4262306a36Sopenharmony_ci * struct page_pool_stats call page_pool_get_stats to get stats for the specified pool.
4362306a36Sopenharmony_ci */
4462306a36Sopenharmony_cibool page_pool_get_stats(struct page_pool *pool,
4562306a36Sopenharmony_ci			 struct page_pool_stats *stats);
4662306a36Sopenharmony_ci#else
4762306a36Sopenharmony_cistatic inline int page_pool_ethtool_stats_get_count(void)
4862306a36Sopenharmony_ci{
4962306a36Sopenharmony_ci	return 0;
5062306a36Sopenharmony_ci}
5162306a36Sopenharmony_ci
5262306a36Sopenharmony_cistatic inline u8 *page_pool_ethtool_stats_get_strings(u8 *data)
5362306a36Sopenharmony_ci{
5462306a36Sopenharmony_ci	return data;
5562306a36Sopenharmony_ci}
5662306a36Sopenharmony_ci
5762306a36Sopenharmony_cistatic inline u64 *page_pool_ethtool_stats_get(u64 *data, void *stats)
5862306a36Sopenharmony_ci{
5962306a36Sopenharmony_ci	return data;
6062306a36Sopenharmony_ci}
6162306a36Sopenharmony_ci#endif
6262306a36Sopenharmony_ci
6362306a36Sopenharmony_ci/**
6462306a36Sopenharmony_ci * page_pool_dev_alloc_pages() - allocate a page.
6562306a36Sopenharmony_ci * @pool:	pool from which to allocate
6662306a36Sopenharmony_ci *
6762306a36Sopenharmony_ci * Get a page from the page allocator or page_pool caches.
6862306a36Sopenharmony_ci */
6962306a36Sopenharmony_cistatic inline struct page *page_pool_dev_alloc_pages(struct page_pool *pool)
7062306a36Sopenharmony_ci{
7162306a36Sopenharmony_ci	gfp_t gfp = (GFP_ATOMIC | __GFP_NOWARN);
7262306a36Sopenharmony_ci
7362306a36Sopenharmony_ci	return page_pool_alloc_pages(pool, gfp);
7462306a36Sopenharmony_ci}
7562306a36Sopenharmony_ci
7662306a36Sopenharmony_cistatic inline struct page *page_pool_dev_alloc_frag(struct page_pool *pool,
7762306a36Sopenharmony_ci						    unsigned int *offset,
7862306a36Sopenharmony_ci						    unsigned int size)
7962306a36Sopenharmony_ci{
8062306a36Sopenharmony_ci	gfp_t gfp = (GFP_ATOMIC | __GFP_NOWARN);
8162306a36Sopenharmony_ci
8262306a36Sopenharmony_ci	return page_pool_alloc_frag(pool, offset, size, gfp);
8362306a36Sopenharmony_ci}
8462306a36Sopenharmony_ci
8562306a36Sopenharmony_ci/**
8662306a36Sopenharmony_ci * page_pool_get_dma_dir() - Retrieve the stored DMA direction.
8762306a36Sopenharmony_ci * @pool:	pool from which page was allocated
8862306a36Sopenharmony_ci *
8962306a36Sopenharmony_ci * Get the stored dma direction. A driver might decide to store this locally
9062306a36Sopenharmony_ci * and avoid the extra cache line from page_pool to determine the direction.
9162306a36Sopenharmony_ci */
9262306a36Sopenharmony_cistatic
9362306a36Sopenharmony_ciinline enum dma_data_direction page_pool_get_dma_dir(struct page_pool *pool)
9462306a36Sopenharmony_ci{
9562306a36Sopenharmony_ci	return pool->p.dma_dir;
9662306a36Sopenharmony_ci}
9762306a36Sopenharmony_ci
9862306a36Sopenharmony_ci/* pp_frag_count represents the number of writers who can update the page
9962306a36Sopenharmony_ci * either by updating skb->data or via DMA mappings for the device.
10062306a36Sopenharmony_ci * We can't rely on the page refcnt for that as we don't know who might be
10162306a36Sopenharmony_ci * holding page references and we can't reliably destroy or sync DMA mappings
10262306a36Sopenharmony_ci * of the fragments.
10362306a36Sopenharmony_ci *
10462306a36Sopenharmony_ci * When pp_frag_count reaches 0 we can either recycle the page if the page
10562306a36Sopenharmony_ci * refcnt is 1 or return it back to the memory allocator and destroy any
10662306a36Sopenharmony_ci * mappings we have.
10762306a36Sopenharmony_ci */
10862306a36Sopenharmony_cistatic inline void page_pool_fragment_page(struct page *page, long nr)
10962306a36Sopenharmony_ci{
11062306a36Sopenharmony_ci	atomic_long_set(&page->pp_frag_count, nr);
11162306a36Sopenharmony_ci}
11262306a36Sopenharmony_ci
11362306a36Sopenharmony_cistatic inline long page_pool_defrag_page(struct page *page, long nr)
11462306a36Sopenharmony_ci{
11562306a36Sopenharmony_ci	long ret;
11662306a36Sopenharmony_ci
11762306a36Sopenharmony_ci	/* If nr == pp_frag_count then we have cleared all remaining
11862306a36Sopenharmony_ci	 * references to the page. No need to actually overwrite it, instead
11962306a36Sopenharmony_ci	 * we can leave this to be overwritten by the calling function.
12062306a36Sopenharmony_ci	 *
12162306a36Sopenharmony_ci	 * The main advantage to doing this is that an atomic_read is
12262306a36Sopenharmony_ci	 * generally a much cheaper operation than an atomic update,
12362306a36Sopenharmony_ci	 * especially when dealing with a page that may be partitioned
12462306a36Sopenharmony_ci	 * into only 2 or 3 pieces.
12562306a36Sopenharmony_ci	 */
12662306a36Sopenharmony_ci	if (atomic_long_read(&page->pp_frag_count) == nr)
12762306a36Sopenharmony_ci		return 0;
12862306a36Sopenharmony_ci
12962306a36Sopenharmony_ci	ret = atomic_long_sub_return(nr, &page->pp_frag_count);
13062306a36Sopenharmony_ci	WARN_ON(ret < 0);
13162306a36Sopenharmony_ci	return ret;
13262306a36Sopenharmony_ci}
13362306a36Sopenharmony_ci
13462306a36Sopenharmony_cistatic inline bool page_pool_is_last_frag(struct page_pool *pool,
13562306a36Sopenharmony_ci					  struct page *page)
13662306a36Sopenharmony_ci{
13762306a36Sopenharmony_ci	/* If fragments aren't enabled or count is 0 we were the last user */
13862306a36Sopenharmony_ci	return !(pool->p.flags & PP_FLAG_PAGE_FRAG) ||
13962306a36Sopenharmony_ci	       (page_pool_defrag_page(page, 1) == 0);
14062306a36Sopenharmony_ci}
14162306a36Sopenharmony_ci
14262306a36Sopenharmony_ci/**
14362306a36Sopenharmony_ci * page_pool_put_page() - release a reference to a page pool page
14462306a36Sopenharmony_ci * @pool:	pool from which page was allocated
14562306a36Sopenharmony_ci * @page:	page to release a reference on
14662306a36Sopenharmony_ci * @dma_sync_size: how much of the page may have been touched by the device
14762306a36Sopenharmony_ci * @allow_direct: released by the consumer, allow lockless caching
14862306a36Sopenharmony_ci *
14962306a36Sopenharmony_ci * The outcome of this depends on the page refcnt. If the driver bumps
15062306a36Sopenharmony_ci * the refcnt > 1 this will unmap the page. If the page refcnt is 1
15162306a36Sopenharmony_ci * the allocator owns the page and will try to recycle it in one of the pool
15262306a36Sopenharmony_ci * caches. If PP_FLAG_DMA_SYNC_DEV is set, the page will be synced for_device
15362306a36Sopenharmony_ci * using dma_sync_single_range_for_device().
15462306a36Sopenharmony_ci */
15562306a36Sopenharmony_cistatic inline void page_pool_put_page(struct page_pool *pool,
15662306a36Sopenharmony_ci				      struct page *page,
15762306a36Sopenharmony_ci				      unsigned int dma_sync_size,
15862306a36Sopenharmony_ci				      bool allow_direct)
15962306a36Sopenharmony_ci{
16062306a36Sopenharmony_ci	/* When page_pool isn't compiled-in, net/core/xdp.c doesn't
16162306a36Sopenharmony_ci	 * allow registering MEM_TYPE_PAGE_POOL, but shield linker.
16262306a36Sopenharmony_ci	 */
16362306a36Sopenharmony_ci#ifdef CONFIG_PAGE_POOL
16462306a36Sopenharmony_ci	if (!page_pool_is_last_frag(pool, page))
16562306a36Sopenharmony_ci		return;
16662306a36Sopenharmony_ci
16762306a36Sopenharmony_ci	page_pool_put_defragged_page(pool, page, dma_sync_size, allow_direct);
16862306a36Sopenharmony_ci#endif
16962306a36Sopenharmony_ci}
17062306a36Sopenharmony_ci
17162306a36Sopenharmony_ci/**
17262306a36Sopenharmony_ci * page_pool_put_full_page() - release a reference on a page pool page
17362306a36Sopenharmony_ci * @pool:	pool from which page was allocated
17462306a36Sopenharmony_ci * @page:	page to release a reference on
17562306a36Sopenharmony_ci * @allow_direct: released by the consumer, allow lockless caching
17662306a36Sopenharmony_ci *
17762306a36Sopenharmony_ci * Similar to page_pool_put_page(), but will DMA sync the entire memory area
17862306a36Sopenharmony_ci * as configured in &page_pool_params.max_len.
17962306a36Sopenharmony_ci */
18062306a36Sopenharmony_cistatic inline void page_pool_put_full_page(struct page_pool *pool,
18162306a36Sopenharmony_ci					   struct page *page, bool allow_direct)
18262306a36Sopenharmony_ci{
18362306a36Sopenharmony_ci	page_pool_put_page(pool, page, -1, allow_direct);
18462306a36Sopenharmony_ci}
18562306a36Sopenharmony_ci
18662306a36Sopenharmony_ci/**
18762306a36Sopenharmony_ci * page_pool_recycle_direct() - release a reference on a page pool page
18862306a36Sopenharmony_ci * @pool:	pool from which page was allocated
18962306a36Sopenharmony_ci * @page:	page to release a reference on
19062306a36Sopenharmony_ci *
19162306a36Sopenharmony_ci * Similar to page_pool_put_full_page() but caller must guarantee safe context
19262306a36Sopenharmony_ci * (e.g NAPI), since it will recycle the page directly into the pool fast cache.
19362306a36Sopenharmony_ci */
19462306a36Sopenharmony_cistatic inline void page_pool_recycle_direct(struct page_pool *pool,
19562306a36Sopenharmony_ci					    struct page *page)
19662306a36Sopenharmony_ci{
19762306a36Sopenharmony_ci	page_pool_put_full_page(pool, page, true);
19862306a36Sopenharmony_ci}
19962306a36Sopenharmony_ci
20062306a36Sopenharmony_ci#define PAGE_POOL_DMA_USE_PP_FRAG_COUNT	\
20162306a36Sopenharmony_ci		(sizeof(dma_addr_t) > sizeof(unsigned long))
20262306a36Sopenharmony_ci
20362306a36Sopenharmony_ci/**
20462306a36Sopenharmony_ci * page_pool_get_dma_addr() - Retrieve the stored DMA address.
20562306a36Sopenharmony_ci * @page:	page allocated from a page pool
20662306a36Sopenharmony_ci *
20762306a36Sopenharmony_ci * Fetch the DMA address of the page. The page pool to which the page belongs
20862306a36Sopenharmony_ci * must had been created with PP_FLAG_DMA_MAP.
20962306a36Sopenharmony_ci */
21062306a36Sopenharmony_cistatic inline dma_addr_t page_pool_get_dma_addr(struct page *page)
21162306a36Sopenharmony_ci{
21262306a36Sopenharmony_ci	dma_addr_t ret = page->dma_addr;
21362306a36Sopenharmony_ci
21462306a36Sopenharmony_ci	if (PAGE_POOL_DMA_USE_PP_FRAG_COUNT)
21562306a36Sopenharmony_ci		ret |= (dma_addr_t)page->dma_addr_upper << 16 << 16;
21662306a36Sopenharmony_ci
21762306a36Sopenharmony_ci	return ret;
21862306a36Sopenharmony_ci}
21962306a36Sopenharmony_ci
22062306a36Sopenharmony_cistatic inline void page_pool_set_dma_addr(struct page *page, dma_addr_t addr)
22162306a36Sopenharmony_ci{
22262306a36Sopenharmony_ci	page->dma_addr = addr;
22362306a36Sopenharmony_ci	if (PAGE_POOL_DMA_USE_PP_FRAG_COUNT)
22462306a36Sopenharmony_ci		page->dma_addr_upper = upper_32_bits(addr);
22562306a36Sopenharmony_ci}
22662306a36Sopenharmony_ci
22762306a36Sopenharmony_cistatic inline bool page_pool_put(struct page_pool *pool)
22862306a36Sopenharmony_ci{
22962306a36Sopenharmony_ci	return refcount_dec_and_test(&pool->user_cnt);
23062306a36Sopenharmony_ci}
23162306a36Sopenharmony_ci
23262306a36Sopenharmony_cistatic inline void page_pool_nid_changed(struct page_pool *pool, int new_nid)
23362306a36Sopenharmony_ci{
23462306a36Sopenharmony_ci	if (unlikely(pool->p.nid != new_nid))
23562306a36Sopenharmony_ci		page_pool_update_nid(pool, new_nid);
23662306a36Sopenharmony_ci}
23762306a36Sopenharmony_ci
23862306a36Sopenharmony_ci#endif /* _NET_PAGE_POOL_HELPERS_H */
239