[PATCH 2/5] ANDROID: dma-buf: heaps: Add a shrinker controlled page pool

[Olivier Masse <olivier.masse@nxp.com>]

From: John Stultz <john.stultz@linaro.org>

This patch adds a simple shrinker controlled page pool to the
dmabuf heaps subsystem.

This replaces the use of the networking page_pool, over concerns
that the lack of a shrinker for that implementation may cause
additional low-memory kills

TODO: Take another pass at trying to unify this w/ the ttm pool

Thoughts and feedback would be greatly appreciated!

Cc: Sumit Semwal <sumit.semwal@linaro.org>
Cc: Liam Mark <lmark@codeaurora.org>
Cc: Laura Abbott <labbott@kernel.org>
Cc: Brian Starkey <Brian.Starkey@arm.com>
Cc: Hridya Valsaraju <hridya@google.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Sandeep Patil <sspatil@google.com>
Cc: Daniel Mentz <danielmentz@google.com>
Cc: Chris Goldsworthy <cgoldswo@codeaurora.org>
Cc: Ørjan Eide <orjan.eide@arm.com>
Cc: Robin Murphy <robin.murphy@arm.com>
Cc: Ezequiel Garcia <ezequiel@collabora.com>
Cc: Simon Ser <contact@emersion.fr>
Cc: James Jones <jajones@nvidia.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: linux-mm@kvack.org
Cc: linux-media@vger.kernel.org
Cc: dri-devel@lists.freedesktop.org
Signed-off-by: John Stultz <john.stultz@linaro.org>
Signed-off-by: Olivier Masse <olivier.masse@nxp.com>
Bug: 168742043
---
 drivers/dma-buf/heaps/Kconfig     |   3 +
 drivers/dma-buf/heaps/Makefile    |   1 +
 drivers/dma-buf/heaps/page_pool.c | 246 ++++++++++++++++++++++++++++++
 drivers/dma-buf/heaps/page_pool.h |  55 +++++++
 4 files changed, 305 insertions(+)
 create mode 100644 drivers/dma-buf/heaps/page_pool.c
 create mode 100644 drivers/dma-buf/heaps/page_pool.h

diff --git a/drivers/dma-buf/heaps/Kconfig b/drivers/dma-buf/heaps/Kconfig
index 8ee64277a5d2..6a33193a7b3e 100644
--- a/drivers/dma-buf/heaps/Kconfig
+++ b/drivers/dma-buf/heaps/Kconfig
@@ -1,6 +1,9 @@
 config DMABUF_HEAPS_DEFERRED_FREE
 	tristate
 
+config DMABUF_HEAPS_PAGE_POOL
+	tristate
+
 config DMABUF_HEAPS_SYSTEM
 	bool "DMA-BUF System Heap"
 	depends on DMABUF_HEAPS
diff --git a/drivers/dma-buf/heaps/Makefile b/drivers/dma-buf/heaps/Makefile
index 5de95b77e169..e70722ea615e 100644
--- a/drivers/dma-buf/heaps/Makefile
+++ b/drivers/dma-buf/heaps/Makefile
@@ -1,5 +1,6 @@
 # SPDX-License-Identifier: GPL-2.0
 obj-$(CONFIG_DMABUF_HEAPS_DEFERRED_FREE) += deferred-free-helper.o
+obj-$(CONFIG_DMABUF_HEAPS_PAGE_POOL)	+= page_pool.o
 obj-$(CONFIG_DMABUF_HEAPS_SYSTEM)	+= system_heap.o
 obj-$(CONFIG_DMABUF_HEAPS_CMA)		+= cma_heap.o
 obj-$(CONFIG_DMABUF_HEAPS_DSP)          += dsp_heap.o
diff --git a/drivers/dma-buf/heaps/page_pool.c b/drivers/dma-buf/heaps/page_pool.c
new file mode 100644
index 000000000000..3dd4c3862dca
--- /dev/null
+++ b/drivers/dma-buf/heaps/page_pool.c
@@ -0,0 +1,246 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * DMA BUF page pool system
+ *
+ * Copyright (C) 2020 Linaro Ltd.
+ *
+ * Based on the ION page pool code
+ * Copyright (C) 2011 Google, Inc.
+ */
+
+#include <linux/freezer.h>
+#include <linux/list.h>
+#include <linux/sched/signal.h>
+#include <linux/slab.h>
+#include <linux/swap.h>
+#include "page_pool.h"
+
+static LIST_HEAD(pool_list);
+static DEFINE_MUTEX(pool_list_lock);
+
+static struct page *dmabuf_page_pool_alloc_pages(struct dmabuf_page_pool *pool)
+{
+	if (fatal_signal_pending(current))
+		return NULL;
+	return alloc_pages(pool->gfp_mask, pool->order);
+}
+
+static void dmabuf_page_pool_free_pages(struct dmabuf_page_pool *pool,
+					       struct page *page)
+{
+	__free_pages(page, pool->order);
+}
+
+static void dmabuf_page_pool_add(struct dmabuf_page_pool *pool, struct page *page)
+{
+	int index;
+
+	if (PageHighMem(page))
+		index = POOL_HIGHPAGE;
+	else
+		index = POOL_LOWPAGE;
+
+	mutex_lock(&pool->mutex);
+	list_add_tail(&page->lru, &pool->items[index]);
+	pool->count[index]++;
+	mod_node_page_state(page_pgdat(page), NR_KERNEL_MISC_RECLAIMABLE,
+			    1 << pool->order);
+	mutex_unlock(&pool->mutex);
+}
+
+static struct page *dmabuf_page_pool_remove(struct dmabuf_page_pool *pool, int index)
+{
+	struct page *page;
+
+	mutex_lock(&pool->mutex);
+	page = list_first_entry_or_null(&pool->items[index], struct page, lru);
+	if (page) {
+		pool->count[index]--;
+		list_del(&page->lru);
+		mod_node_page_state(page_pgdat(page), NR_KERNEL_MISC_RECLAIMABLE,
+				    -(1 << pool->order));
+	}
+	mutex_unlock(&pool->mutex);
+
+	return page;
+}
+
+static struct page *dmabuf_page_pool_fetch(struct dmabuf_page_pool *pool)
+{
+	struct page *page = NULL;
+
+	page = dmabuf_page_pool_remove(pool, POOL_HIGHPAGE);
+	if (!page)
+		page = dmabuf_page_pool_remove(pool, POOL_LOWPAGE);
+
+	return page;
+}
+
+struct page *dmabuf_page_pool_alloc(struct dmabuf_page_pool *pool)
+{
+	struct page *page = NULL;
+
+	if (WARN_ON(!pool))
+		return NULL;
+
+	page = dmabuf_page_pool_fetch(pool);
+	if (!page)
+		page = dmabuf_page_pool_alloc_pages(pool);
+
+	return page;
+}
+EXPORT_SYMBOL_GPL(dmabuf_page_pool_alloc);
+
+void dmabuf_page_pool_free(struct dmabuf_page_pool *pool, struct page *page)
+{
+	if (WARN_ON(pool->order != compound_order(page)))
+		return;
+
+	dmabuf_page_pool_add(pool, page);
+}
+EXPORT_SYMBOL_GPL(dmabuf_page_pool_free);
+
+static int dmabuf_page_pool_total(struct dmabuf_page_pool *pool, bool high)
+{
+	int count = pool->count[POOL_LOWPAGE];
+
+	if (high)
+		count += pool->count[POOL_HIGHPAGE];
+
+	return count << pool->order;
+}
+
+struct dmabuf_page_pool *dmabuf_page_pool_create(gfp_t gfp_mask, unsigned int order)
+{
+	struct dmabuf_page_pool *pool = kmalloc(sizeof(*pool), GFP_KERNEL);
+	int i;
+
+	if (!pool)
+		return NULL;
+
+	for (i = 0; i < POOL_TYPE_SIZE; i++) {
+		pool->count[i] = 0;
+		INIT_LIST_HEAD(&pool->items[i]);
+	}
+	pool->gfp_mask = gfp_mask | __GFP_COMP;
+	pool->order = order;
+	mutex_init(&pool->mutex);
+
+	mutex_lock(&pool_list_lock);
+	list_add(&pool->list, &pool_list);
+	mutex_unlock(&pool_list_lock);
+
+	return pool;
+}
+EXPORT_SYMBOL_GPL(dmabuf_page_pool_create);
+
+void dmabuf_page_pool_destroy(struct dmabuf_page_pool *pool)
+{
+	struct page *page;
+	int i;
+
+	/* Remove us from the pool list */
+	mutex_lock(&pool_list_lock);
+	list_del(&pool->list);
+	mutex_unlock(&pool_list_lock);
+
+	/* Free any remaining pages in the pool */
+	for (i = 0; i < POOL_TYPE_SIZE; i++) {
+		while ((page = dmabuf_page_pool_remove(pool, i)))
+			dmabuf_page_pool_free_pages(pool, page);
+	}
+
+	kfree(pool);
+}
+EXPORT_SYMBOL_GPL(dmabuf_page_pool_destroy);
+
+static int dmabuf_page_pool_do_shrink(struct dmabuf_page_pool *pool, gfp_t gfp_mask,
+				      int nr_to_scan)
+{
+	int freed = 0;
+	bool high;
+
+	if (current_is_kswapd())
+		high = true;
+	else
+		high = !!(gfp_mask & __GFP_HIGHMEM);
+
+	if (nr_to_scan == 0)
+		return dmabuf_page_pool_total(pool, high);
+
+	while (freed < nr_to_scan) {
+		struct page *page;
+
+		/* Try to free low pages first */
+		page = dmabuf_page_pool_remove(pool, POOL_LOWPAGE);
+		if (!page)
+			page = dmabuf_page_pool_remove(pool, POOL_HIGHPAGE);
+
+		if (!page)
+			break;
+
+		dmabuf_page_pool_free_pages(pool, page);
+		freed += (1 << pool->order);
+	}
+
+	return freed;
+}
+
+static int dmabuf_page_pool_shrink(gfp_t gfp_mask, int nr_to_scan)
+{
+	struct dmabuf_page_pool *pool;
+	int nr_total = 0;
+	int nr_freed;
+	bool only_scan = false;
+
+	if (!nr_to_scan)
+		only_scan = true;
+
+	mutex_lock(&pool_list_lock);
+	list_for_each_entry(pool, &pool_list, list) {
+		if (only_scan) {
+			nr_total += dmabuf_page_pool_do_shrink(pool,
+							       gfp_mask,
+							       nr_to_scan);
+		} else {
+			nr_freed = dmabuf_page_pool_do_shrink(pool,
+							      gfp_mask,
+							      nr_to_scan);
+			nr_to_scan -= nr_freed;
+			nr_total += nr_freed;
+			if (nr_to_scan <= 0)
+				break;
+		}
+	}
+	mutex_unlock(&pool_list_lock);
+
+	return nr_total;
+}
+
+static unsigned long dmabuf_page_pool_shrink_count(struct shrinker *shrinker,
+						   struct shrink_control *sc)
+{
+	return dmabuf_page_pool_shrink(sc->gfp_mask, 0);
+}
+
+static unsigned long dmabuf_page_pool_shrink_scan(struct shrinker *shrinker,
+						  struct shrink_control *sc)
+{
+	if (sc->nr_to_scan == 0)
+		return 0;
+	return dmabuf_page_pool_shrink(sc->gfp_mask, sc->nr_to_scan);
+}
+
+struct shrinker pool_shrinker = {
+	.count_objects = dmabuf_page_pool_shrink_count,
+	.scan_objects = dmabuf_page_pool_shrink_scan,
+	.seeks = DEFAULT_SEEKS,
+	.batch = 0,
+};
+
+static int dmabuf_page_pool_init_shrinker(void)
+{
+	return register_shrinker(&pool_shrinker);
+}
+module_init(dmabuf_page_pool_init_shrinker);
+MODULE_LICENSE("GPL v2");
diff --git a/drivers/dma-buf/heaps/page_pool.h b/drivers/dma-buf/heaps/page_pool.h
new file mode 100644
index 000000000000..e3ec9eaacbc2
--- /dev/null
+++ b/drivers/dma-buf/heaps/page_pool.h
@@ -0,0 +1,55 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * DMA BUF PagePool implementation
+ * Based on earlier ION code by Google
+ *
+ * Copyright (C) 2011 Google, Inc.
+ * Copyright (C) 2020 Linaro Ltd.
+ */
+
+#ifndef _DMABUF_PAGE_POOL_H
+#define _DMABUF_PAGE_POOL_H
+
+#include <linux/device.h>
+#include <linux/kref.h>
+#include <linux/mm_types.h>
+#include <linux/mutex.h>
+#include <linux/shrinker.h>
+#include <linux/types.h>
+
+/* page types we track in the pool */
+enum {
+	POOL_LOWPAGE,      /* Clean lowmem pages */
+	POOL_HIGHPAGE,     /* Clean highmem pages */
+
+	POOL_TYPE_SIZE
+};
+
+/**
+ * struct dmabuf_page_pool - pagepool struct
+ * @count[]:		array of number of pages of that type in the pool
+ * @items[]:		array of list of pages of the specific type
+ * @mutex:		lock protecting this struct and especially the count
+ *			item list
+ * @gfp_mask:		gfp_mask to use from alloc
+ * @order:		order of pages in the pool
+ * @list:		list node for list of pools
+ *
+ * Allows you to keep a pool of pre allocated pages to use
+ */
+struct dmabuf_page_pool {
+	int count[POOL_TYPE_SIZE];
+	struct list_head items[POOL_TYPE_SIZE];
+	struct mutex mutex;
+	gfp_t gfp_mask;
+	unsigned int order;
+	struct list_head list;
+};
+
+struct dmabuf_page_pool *dmabuf_page_pool_create(gfp_t gfp_mask,
+						 unsigned int order);
+void dmabuf_page_pool_destroy(struct dmabuf_page_pool *pool);
+struct page *dmabuf_page_pool_alloc(struct dmabuf_page_pool *pool);
+void dmabuf_page_pool_free(struct dmabuf_page_pool *pool, struct page *page);
+
+#endif /* _DMABUF_PAGE_POOL_H */
-- 
2.25.0