page_pool: refurbish version of page_pool code

Need a fast page recycle mechanism for ndo_xdp_xmit API for returning
pages on DMA-TX completion time, which have good cross CPU
performance, given DMA-TX completion time can happen on a remote CPU.

Refurbish my page_pool code, that was presented[1] at MM-summit 2016.
Adapted page_pool code to not depend the page allocator and
integration into struct page.  The DMA mapping feature is kept,
even-though it will not be activated/used in this patchset.

[1] http://people.netfilter.org/hawk/presentations/MM-summit2016/generic_page_pool_mm_summit2016.pdf

V2: Adjustments requested by Tariq
 - Changed page_pool_create return codes, don't return NULL, only
   ERR_PTR, as this simplifies err handling in drivers.

V4: many small improvements and cleanups
- Add DOC comment section, that can be used by kernel-doc
- Improve fallback mode, to work better with refcnt based recycling
  e.g. remove a WARN as pointed out by Tariq
  e.g. quicker fallback if ptr_ring is empty.

V5: Fixed SPDX license as pointed out by Alexei

V6: Adjustments requested by Eric Dumazet
 - Adjust ____cacheline_aligned_in_smp usage/placement
 - Move rcu_head in struct page_pool
 - Free pages quicker on destroy, minimize resources delayed an RCU period
 - Remove code for forward/backward compat ABI interface

V8: Issues found by kbuild test robot
 - Address sparse should be static warnings
 - Only compile+link when a driver use/select page_pool,
   mlx5 selects CONFIG_PAGE_POOL, although its first used in two patches

Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/mellanox/mlx5/core/Kconfig

@@ -30,6 +30,7 @@ config MLX5_CORE_EN
 	bool "Mellanox Technologies ConnectX-4 Ethernet support"
 	depends on NETDEVICES && ETHERNET && INET && PCI && MLX5_CORE
 	depends on IPV6=y || IPV6=n || MLX5_CORE=m
+	select PAGE_POOL
 	default n
 	---help---
 	  Ethernet support in Mellanox Technologies ConnectX-4 NIC.

include/net/page_pool.h

@@ -0,0 +1,129 @@
+/* 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().
+ *
+ * If page_pool handles DMA mapping (use page->private), then API user
+ * is responsible for invoking page_pool_put_page() once.  In-case of
+ * elevated refcnt, the DMA state is released, assuming other users of
+ * the page will eventually call put_page().
+ *
+ * If no DMA mapping is done, then it can act as shim-layer that
+ * fall-through to alloc_page.  As no state is kept on the page, the
+ * regular put_page() call is sufficient.
+ */
+#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 1 /* Should page_pool do the DMA map/unmap */
+#define PP_FLAG_ALL	PP_FLAG_DMA_MAP
+
+/*
+ * 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;
+	void *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 */
+};
+
+struct page_pool {
+	struct rcu_head rcu;
+	struct page_pool_params p;
+
+	/*
+	 * 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;
+};
+
+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_pool *page_pool_create(const struct page_pool_params *params);
+
+void page_pool_destroy(struct page_pool *pool);
+
+/* Never call this directly, use helpers below */
+void __page_pool_put_page(struct page_pool *pool,
+			  struct page *page, bool allow_direct);
+
+static inline void page_pool_put_page(struct page_pool *pool, struct page *page)
+{
+	__page_pool_put_page(pool, page, false);
+}
+/* Very limited use-cases allow recycle direct */
+static inline void page_pool_recycle_direct(struct page_pool *pool,
+					    struct page *page)
+{
+	__page_pool_put_page(pool, page, true);
+}
+
+#endif /* _NET_PAGE_POOL_H */

net/Kconfig

@@ -423,6 +423,9 @@ config MAY_USE_DEVLINK
 	  on MAY_USE_DEVLINK to ensure they do not cause link errors when
 	  devlink is a loadable module and the driver using it is built-in.
 
+config PAGE_POOL
+       bool
+
 endif   # if NET
 
 # Used by archs to tell that they support BPF JIT compiler plus which flavour.

net/core/Makefile

@@ -14,6 +14,7 @@ obj-y		     += dev.o ethtool.o dev_addr_lists.o dst.o netevent.o \
 			fib_notifier.o xdp.o
 
 obj-y += net-sysfs.o
+obj-$(CONFIG_PAGE_POOL) += page_pool.o
 obj-$(CONFIG_PROC_FS) += net-procfs.o
 obj-$(CONFIG_NET_PKTGEN) += pktgen.o
 obj-$(CONFIG_NETPOLL) += netpoll.o