[EARLY,RFC,3/4] dma-buf: pools: Add system/system-contig pools to dmabuf pools

diff --git a/drivers/dma-buf/pools/Kconfig b/drivers/dma-buf/pools/Kconfig
index caa7eb8..787b2a6 100644
--- a/drivers/dma-buf/pools/Kconfig
+++ b/drivers/dma-buf/pools/Kconfig
@@ -8,3 +8,10 @@  menuconfig DMABUF_POOLS
 	  which allow userspace to allocate dma-bufs that can be shared between
 	  drivers.
 	  If you're not using Android its probably safe to say N here.
+
+config DMABUF_POOLS_SYSTEM
+	bool "DMA-BUF System Pool"
+	depends on DMABUF_POOLS
+	help
+	  Choose this option to enable the system dmabuf pool. The system pool
+	  is backed by pages from the buddy allocator. If in doubt, say Y.
diff --git a/drivers/dma-buf/pools/Makefile b/drivers/dma-buf/pools/Makefile
index a51ec25..2ccf2a1 100644
--- a/drivers/dma-buf/pools/Makefile
+++ b/drivers/dma-buf/pools/Makefile
@@ -1,2 +1,3 @@ 
 # SPDX-License-Identifier: GPL-2.0
 obj-$(CONFIG_DMABUF_POOLS)		+= dmabuf-pools.o pool-ioctl.o pool-helpers.o page_pool.o
+obj-$(CONFIG_DMABUF_POOLS_SYSTEM)	+= system_pool.o
diff --git a/drivers/dma-buf/pools/system_pool.c b/drivers/dma-buf/pools/system_pool.c
new file mode 100644
index 0000000..1756990
--- /dev/null
+++ b/drivers/dma-buf/pools/system_pool.c
@@ -0,0 +1,374 @@ 
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * drivers/dma-buf/pools/system_pool.c
+ *
+ * Copyright (C) 2011 Google, Inc.
+ * Copyright (C) 2019 Linaro Ltd.
+ */
+
+#include <asm/page.h>
+#include <linux/dma-mapping.h>
+#include <linux/err.h>
+#include <linux/highmem.h>
+#include <linux/mm.h>
+#include <linux/scatterlist.h>
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+#include "dmabuf-pools.h"
+
+#define NUM_ORDERS ARRAY_SIZE(orders)
+
+static gfp_t high_order_gfp_flags = (GFP_HIGHUSER | __GFP_ZERO | __GFP_NOWARN |
+				     __GFP_NORETRY) & ~__GFP_RECLAIM;
+static gfp_t low_order_gfp_flags  = GFP_HIGHUSER | __GFP_ZERO;
+static const unsigned int orders[] = {8, 4, 0};
+
+static int order_to_index(unsigned int order)
+{
+	int i;
+
+	for (i = 0; i < NUM_ORDERS; i++)
+		if (order == orders[i])
+			return i;
+	WARN_ON(1);
+	return -1;
+}
+
+static inline unsigned int order_to_size(int order)
+{
+	return PAGE_SIZE << order;
+}
+
+struct system_pool {
+	struct dmabuf_pool pool;
+	struct dmabuf_page_pool *page_pools[NUM_ORDERS];
+};
+
+static struct page *alloc_buffer_page(struct system_pool *sys_pool,
+				      struct dmabuf_pool_buffer *buffer,
+				      unsigned long order)
+{
+	struct dmabuf_page_pool *pagepool =
+				sys_pool->page_pools[order_to_index(order)];
+
+	return dmabuf_page_pool_alloc(pagepool);
+}
+
+static void free_buffer_page(struct system_pool *sys_pool,
+			     struct dmabuf_pool_buffer *buffer,
+			     struct page *page)
+{
+	struct dmabuf_page_pool *pagepool;
+	unsigned int order = compound_order(page);
+
+	/* go to system */
+	if (buffer->private_flags & DMABUF_POOL_PRIV_FLAG_SHRINKER_FREE) {
+		__free_pages(page, order);
+		return;
+	}
+
+	pagepool = sys_pool->page_pools[order_to_index(order)];
+
+	dmabuf_page_pool_free(pagepool, page);
+}
+
+static struct page *alloc_largest_available(struct system_pool *sys_pool,
+					    struct dmabuf_pool_buffer *buffer,
+					    unsigned long size,
+					    unsigned int max_order)
+{
+	struct page *page;
+	int i;
+
+	for (i = 0; i < NUM_ORDERS; i++) {
+		if (size < order_to_size(orders[i]))
+			continue;
+		if (max_order < orders[i])
+			continue;
+
+		page = alloc_buffer_page(sys_pool, buffer, orders[i]);
+		if (!page)
+			continue;
+
+		return page;
+	}
+
+	return NULL;
+}
+
+static int system_pool_allocate(struct dmabuf_pool *pool,
+				    struct dmabuf_pool_buffer *buffer,
+				    unsigned long size,
+				    unsigned long flags)
+{
+	struct system_pool *sys_pool = container_of(pool,
+							struct system_pool,
+							pool);
+	struct sg_table *table;
+	struct scatterlist *sg;
+	struct list_head pages;
+	struct page *page, *tmp_page;
+	int i = 0;
+	unsigned long size_remaining = PAGE_ALIGN(size);
+	unsigned int max_order = orders[0];
+
+	if (size / PAGE_SIZE > totalram_pages() / 2)
+		return -ENOMEM;
+
+	INIT_LIST_HEAD(&pages);
+	while (size_remaining > 0) {
+		page = alloc_largest_available(sys_pool, buffer, size_remaining,
+					       max_order);
+		if (!page)
+			goto free_pages;
+		list_add_tail(&page->lru, &pages);
+		size_remaining -= PAGE_SIZE << compound_order(page);
+		max_order = compound_order(page);
+		i++;
+	}
+	table = kmalloc(sizeof(*table), GFP_KERNEL);
+	if (!table)
+		goto free_pages;
+
+	if (sg_alloc_table(table, i, GFP_KERNEL))
+		goto free_table;
+
+	sg = table->sgl;
+	list_for_each_entry_safe(page, tmp_page, &pages, lru) {
+		sg_set_page(sg, page, PAGE_SIZE << compound_order(page), 0);
+		sg = sg_next(sg);
+		list_del(&page->lru);
+	}
+
+	buffer->sg_table = table;
+	return 0;
+
+free_table:
+	kfree(table);
+free_pages:
+	list_for_each_entry_safe(page, tmp_page, &pages, lru)
+		free_buffer_page(sys_pool, buffer, page);
+	return -ENOMEM;
+}
+
+static void system_pool_free(struct dmabuf_pool_buffer *buffer)
+{
+	struct system_pool *sys_pool = container_of(buffer->pool,
+							struct system_pool,
+							pool);
+	struct sg_table *table = buffer->sg_table;
+	struct scatterlist *sg;
+	int i;
+
+	/* zero the buffer before goto page pool */
+	if (!(buffer->private_flags & DMABUF_POOL_PRIV_FLAG_SHRINKER_FREE))
+		dmabuf_pool_buffer_zero(buffer);
+
+	for_each_sg(table->sgl, sg, table->nents, i)
+		free_buffer_page(sys_pool, buffer, sg_page(sg));
+	sg_free_table(table);
+	kfree(table);
+}
+
+static int system_pool_shrink(struct dmabuf_pool *pool, gfp_t gfp_mask,
+				  int nr_to_scan)
+{
+	struct dmabuf_page_pool *page_pool;
+	struct system_pool *sys_pool;
+	int nr_total = 0;
+	int i, nr_freed;
+	int only_scan = 0;
+
+	sys_pool = container_of(pool, struct system_pool, pool);
+
+	if (!nr_to_scan)
+		only_scan = 1;
+
+	for (i = 0; i < NUM_ORDERS; i++) {
+		page_pool = sys_pool->page_pools[i];
+
+		if (only_scan) {
+			nr_total += dmabuf_page_pool_shrink(page_pool,
+							 gfp_mask,
+							 nr_to_scan);
+
+		} else {
+			nr_freed = dmabuf_page_pool_shrink(page_pool,
+							gfp_mask,
+							nr_to_scan);
+			nr_to_scan -= nr_freed;
+			nr_total += nr_freed;
+			if (nr_to_scan <= 0)
+				break;
+		}
+	}
+	return nr_total;
+}
+
+static struct dmabuf_pool_ops system_pool_ops = {
+	.allocate = system_pool_allocate,
+	.free = system_pool_free,
+	.map_kernel = dmabuf_pool_map_kernel,
+	.unmap_kernel = dmabuf_pool_unmap_kernel,
+	.map_user = dmabuf_pool_map_user,
+	.shrink = system_pool_shrink,
+};
+
+static void system_pool_destroy_pools(struct dmabuf_page_pool **page_pools)
+{
+	int i;
+
+	for (i = 0; i < NUM_ORDERS; i++)
+		if (page_pools[i])
+			dmabuf_page_pool_destroy(page_pools[i]);
+}
+
+static int system_pool_create_pools(struct dmabuf_page_pool **page_pools)
+{
+	int i;
+	gfp_t gfp_flags = low_order_gfp_flags;
+
+	for (i = 0; i < NUM_ORDERS; i++) {
+		struct dmabuf_page_pool *pool;
+
+		if (orders[i] > 4)
+			gfp_flags = high_order_gfp_flags;
+
+		pool = dmabuf_page_pool_create(gfp_flags, orders[i]);
+		if (!pool)
+			goto err_create_pool;
+		page_pools[i] = pool;
+	}
+	return 0;
+
+err_create_pool:
+	system_pool_destroy_pools(page_pools);
+	return -ENOMEM;
+}
+
+static struct dmabuf_pool *__system_pool_create(void)
+{
+	struct system_pool *sys_pool;
+
+	sys_pool = kzalloc(sizeof(*sys_pool), GFP_KERNEL);
+	if (!sys_pool)
+		return ERR_PTR(-ENOMEM);
+	sys_pool->pool.ops = &system_pool_ops;
+	sys_pool->pool.flags = DMABUF_POOL_FLAG_DEFER_FREE;
+
+	if (system_pool_create_pools(sys_pool->page_pools))
+		goto free_pool;
+
+	return &sys_pool->pool;
+
+free_pool:
+	kfree(sys_pool);
+	return ERR_PTR(-ENOMEM);
+}
+
+static int system_pool_create(void)
+{
+	struct dmabuf_pool *pool;
+
+	pool = __system_pool_create();
+	if (IS_ERR(pool))
+		return PTR_ERR(pool);
+	pool->name = "system_pool";
+
+	dmabuf_pool_add(pool);
+	return 0;
+}
+device_initcall(system_pool_create);
+
+static int system_contig_pool_allocate(struct dmabuf_pool *pool,
+					   struct dmabuf_pool_buffer *buffer,
+					   unsigned long len,
+					   unsigned long flags)
+{
+	int order = get_order(len);
+	struct page *page;
+	struct sg_table *table;
+	unsigned long i;
+	int ret;
+
+	page = alloc_pages(low_order_gfp_flags | __GFP_NOWARN, order);
+	if (!page)
+		return -ENOMEM;
+
+	split_page(page, order);
+
+	len = PAGE_ALIGN(len);
+	for (i = len >> PAGE_SHIFT; i < (1 << order); i++)
+		__free_page(page + i);
+
+	table = kmalloc(sizeof(*table), GFP_KERNEL);
+	if (!table) {
+		ret = -ENOMEM;
+		goto free_pages;
+	}
+
+	ret = sg_alloc_table(table, 1, GFP_KERNEL);
+	if (ret)
+		goto free_table;
+
+	sg_set_page(table->sgl, page, len, 0);
+
+	buffer->sg_table = table;
+
+	return 0;
+
+free_table:
+	kfree(table);
+free_pages:
+	for (i = 0; i < len >> PAGE_SHIFT; i++)
+		__free_page(page + i);
+
+	return ret;
+}
+
+static void system_contig_pool_free(struct dmabuf_pool_buffer *buffer)
+{
+	struct sg_table *table = buffer->sg_table;
+	struct page *page = sg_page(table->sgl);
+	unsigned long pages = PAGE_ALIGN(buffer->size) >> PAGE_SHIFT;
+	unsigned long i;
+
+	for (i = 0; i < pages; i++)
+		__free_page(page + i);
+	sg_free_table(table);
+	kfree(table);
+}
+
+static struct dmabuf_pool_ops kmalloc_ops = {
+	.allocate = system_contig_pool_allocate,
+	.free = system_contig_pool_free,
+	.map_kernel = dmabuf_pool_map_kernel,
+	.unmap_kernel = dmabuf_pool_unmap_kernel,
+	.map_user = dmabuf_pool_map_user,
+};
+
+static struct dmabuf_pool *__system_contig_pool_create(void)
+{
+	struct dmabuf_pool *pool;
+
+	pool = kzalloc(sizeof(*pool), GFP_KERNEL);
+	if (!pool)
+		return ERR_PTR(-ENOMEM);
+	pool->ops = &kmalloc_ops;
+	pool->name = "system_contig_pool";
+	return pool;
+}
+
+static int system_contig_pool_create(void)
+{
+	struct dmabuf_pool *pool;
+
+	pool = __system_contig_pool_create();
+	if (IS_ERR(pool))
+		return PTR_ERR(pool);
+
+	dmabuf_pool_add(pool);
+	return 0;
+}
+device_initcall(system_contig_pool_create);
+