ION: Sys_heap: Add cached pool to spead up cached buffer alloc
Add ion cached pool in system heap. This patch add a cached pool
in system heap. It has a great improvement of alloc for cached
buffer.
With memory pressue alloc test 800MB in userspace used iontest.
The result avg is 577ms. Without patch it's avg is about 883ms.
v1: Makes the cached buffer zeroed before going to pool
v2: Add cached param in pool to distinguish wheather need to flush
cache at a fresh alloc.
Rework the shrink function.
Signed-off-by: Chen Feng <puck.chen@hisilicon.com>
Signed-off-by: Xia Qing <saberlily.xia@hisilicon.com>
Reviewed-by: Fu Jun <oliver.fu@hisilicon.com>
Acked-by: Laura Abbott <labbott@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
hifive-unleashed-5.1
Chen Feng
Greg Kroah-Hartman
parent
1f0c9efe56
commit
e7f63771b6
|
|
@ -30,8 +30,9 @@ static void *ion_page_pool_alloc_pages(struct ion_page_pool *pool)
|
|||
|
||||
if (!page)
|
||||
return NULL;
|
||||
ion_pages_sync_for_device(NULL, page, PAGE_SIZE << pool->order,
|
||||
DMA_BIDIRECTIONAL);
|
||||
if (!pool->cached)
|
||||
ion_pages_sync_for_device(NULL, page, PAGE_SIZE << pool->order,
|
||||
DMA_BIDIRECTIONAL);
|
||||
return page;
|
||||
}
|
||||
|
||||
|
|
@ -147,7 +148,8 @@ int ion_page_pool_shrink(struct ion_page_pool *pool, gfp_t gfp_mask,
|
|||
return freed;
|
||||
}
|
||||
|
||||
struct ion_page_pool *ion_page_pool_create(gfp_t gfp_mask, unsigned int order)
|
||||
struct ion_page_pool *ion_page_pool_create(gfp_t gfp_mask, unsigned int order,
|
||||
bool cached)
|
||||
{
|
||||
struct ion_page_pool *pool = kmalloc(sizeof(*pool), GFP_KERNEL);
|
||||
|
||||
|
|
@ -161,6 +163,8 @@ struct ion_page_pool *ion_page_pool_create(gfp_t gfp_mask, unsigned int order)
|
|||
pool->order = order;
|
||||
mutex_init(&pool->mutex);
|
||||
plist_node_init(&pool->list, order);
|
||||
if (cached)
|
||||
pool->cached = true;
|
||||
|
||||
return pool;
|
||||
}
|
||||
|
|
|
|||
|
|
@ -332,6 +332,7 @@ void ion_cma_heap_destroy(struct ion_heap *);
|
|||
* @gfp_mask: gfp_mask to use from alloc
|
||||
* @order: order of pages in the pool
|
||||
* @list: plist node for list of pools
|
||||
* @cached: it's cached pool or not
|
||||
*
|
||||
* Allows you to keep a pool of pre allocated pages to use from your heap.
|
||||
* Keeping a pool of pages that is ready for dma, ie any cached mapping have
|
||||
|
|
@ -341,6 +342,7 @@ void ion_cma_heap_destroy(struct ion_heap *);
|
|||
struct ion_page_pool {
|
||||
int high_count;
|
||||
int low_count;
|
||||
bool cached;
|
||||
struct list_head high_items;
|
||||
struct list_head low_items;
|
||||
struct mutex mutex;
|
||||
|
|
@ -349,7 +351,8 @@ struct ion_page_pool {
|
|||
struct plist_node list;
|
||||
};
|
||||
|
||||
struct ion_page_pool *ion_page_pool_create(gfp_t gfp_mask, unsigned int order);
|
||||
struct ion_page_pool *ion_page_pool_create(gfp_t gfp_mask, unsigned int order,
|
||||
bool cached);
|
||||
void ion_page_pool_destroy(struct ion_page_pool *);
|
||||
struct page *ion_page_pool_alloc(struct ion_page_pool *);
|
||||
void ion_page_pool_free(struct ion_page_pool *, struct page *);
|
||||
|
|
|
|||
|
|
@ -26,16 +26,18 @@
|
|||
#include "ion.h"
|
||||
#include "ion_priv.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 | __GFP_NOWARN);
|
||||
static const unsigned int orders[] = {8, 4, 0};
|
||||
static const int num_orders = ARRAY_SIZE(orders);
|
||||
|
||||
static int order_to_index(unsigned int order)
|
||||
{
|
||||
int i;
|
||||
|
||||
for (i = 0; i < num_orders; i++)
|
||||
for (i = 0; i < NUM_ORDERS; i++)
|
||||
if (order == orders[i])
|
||||
return i;
|
||||
BUG();
|
||||
|
|
@ -49,47 +51,55 @@ static inline unsigned int order_to_size(int order)
|
|||
|
||||
struct ion_system_heap {
|
||||
struct ion_heap heap;
|
||||
struct ion_page_pool *pools[0];
|
||||
struct ion_page_pool *uncached_pools[NUM_ORDERS];
|
||||
struct ion_page_pool *cached_pools[NUM_ORDERS];
|
||||
};
|
||||
|
||||
/**
|
||||
* The page from page-pool are all zeroed before. We need do cache
|
||||
* clean for cached buffer. The uncached buffer are always non-cached
|
||||
* since it's allocated. So no need for non-cached pages.
|
||||
*/
|
||||
static struct page *alloc_buffer_page(struct ion_system_heap *heap,
|
||||
struct ion_buffer *buffer,
|
||||
unsigned long order)
|
||||
{
|
||||
bool cached = ion_buffer_cached(buffer);
|
||||
struct ion_page_pool *pool = heap->pools[order_to_index(order)];
|
||||
struct ion_page_pool *pool;
|
||||
struct page *page;
|
||||
|
||||
if (!cached) {
|
||||
page = ion_page_pool_alloc(pool);
|
||||
} else {
|
||||
gfp_t gfp_flags = low_order_gfp_flags;
|
||||
if (!cached)
|
||||
pool = heap->uncached_pools[order_to_index(order)];
|
||||
else
|
||||
pool = heap->cached_pools[order_to_index(order)];
|
||||
|
||||
if (order > 4)
|
||||
gfp_flags = high_order_gfp_flags;
|
||||
page = alloc_pages(gfp_flags | __GFP_COMP, order);
|
||||
if (!page)
|
||||
return NULL;
|
||||
page = ion_page_pool_alloc(pool);
|
||||
|
||||
if (cached)
|
||||
ion_pages_sync_for_device(NULL, page, PAGE_SIZE << order,
|
||||
DMA_BIDIRECTIONAL);
|
||||
}
|
||||
|
||||
DMA_BIDIRECTIONAL);
|
||||
return page;
|
||||
}
|
||||
|
||||
static void free_buffer_page(struct ion_system_heap *heap,
|
||||
struct ion_buffer *buffer, struct page *page)
|
||||
{
|
||||
struct ion_page_pool *pool;
|
||||
unsigned int order = compound_order(page);
|
||||
bool cached = ion_buffer_cached(buffer);
|
||||
|
||||
if (!cached && !(buffer->private_flags & ION_PRIV_FLAG_SHRINKER_FREE)) {
|
||||
struct ion_page_pool *pool = heap->pools[order_to_index(order)];
|
||||
|
||||
ion_page_pool_free(pool, page);
|
||||
} else {
|
||||
/* go to system */
|
||||
if (buffer->private_flags & ION_PRIV_FLAG_SHRINKER_FREE) {
|
||||
__free_pages(page, order);
|
||||
return;
|
||||
}
|
||||
|
||||
if (!cached)
|
||||
pool = heap->uncached_pools[order_to_index(order)];
|
||||
else
|
||||
pool = heap->cached_pools[order_to_index(order)];
|
||||
|
||||
ion_page_pool_free(pool, page);
|
||||
}
|
||||
|
||||
|
||||
|
|
@ -101,7 +111,7 @@ static struct page *alloc_largest_available(struct ion_system_heap *heap,
|
|||
struct page *page;
|
||||
int i;
|
||||
|
||||
for (i = 0; i < num_orders; i++) {
|
||||
for (i = 0; i < NUM_ORDERS; i++) {
|
||||
if (size < order_to_size(orders[i]))
|
||||
continue;
|
||||
if (max_order < orders[i])
|
||||
|
|
@ -181,16 +191,11 @@ static void ion_system_heap_free(struct ion_buffer *buffer)
|
|||
struct ion_system_heap,
|
||||
heap);
|
||||
struct sg_table *table = buffer->sg_table;
|
||||
bool cached = ion_buffer_cached(buffer);
|
||||
struct scatterlist *sg;
|
||||
int i;
|
||||
|
||||
/*
|
||||
* uncached pages come from the page pools, zero them before returning
|
||||
* for security purposes (other allocations are zerod at
|
||||
* alloc time
|
||||
*/
|
||||
if (!cached && !(buffer->private_flags & ION_PRIV_FLAG_SHRINKER_FREE))
|
||||
/* zero the buffer before goto page pool */
|
||||
if (!(buffer->private_flags & ION_PRIV_FLAG_SHRINKER_FREE))
|
||||
ion_heap_buffer_zero(buffer);
|
||||
|
||||
for_each_sg(table->sgl, sg, table->nents, i)
|
||||
|
|
@ -202,6 +207,8 @@ static void ion_system_heap_free(struct ion_buffer *buffer)
|
|||
static int ion_system_heap_shrink(struct ion_heap *heap, gfp_t gfp_mask,
|
||||
int nr_to_scan)
|
||||
{
|
||||
struct ion_page_pool *uncached_pool;
|
||||
struct ion_page_pool *cached_pool;
|
||||
struct ion_system_heap *sys_heap;
|
||||
int nr_total = 0;
|
||||
int i, nr_freed;
|
||||
|
|
@ -212,20 +219,35 @@ static int ion_system_heap_shrink(struct ion_heap *heap, gfp_t gfp_mask,
|
|||
if (!nr_to_scan)
|
||||
only_scan = 1;
|
||||
|
||||
for (i = 0; i < num_orders; i++) {
|
||||
struct ion_page_pool *pool = sys_heap->pools[i];
|
||||
for (i = 0; i < NUM_ORDERS; i++) {
|
||||
uncached_pool = sys_heap->uncached_pools[i];
|
||||
cached_pool = sys_heap->cached_pools[i];
|
||||
|
||||
nr_freed = ion_page_pool_shrink(pool, gfp_mask, nr_to_scan);
|
||||
nr_total += nr_freed;
|
||||
if (only_scan) {
|
||||
nr_total += ion_page_pool_shrink(uncached_pool,
|
||||
gfp_mask,
|
||||
nr_to_scan);
|
||||
|
||||
if (!only_scan) {
|
||||
nr_total += ion_page_pool_shrink(cached_pool,
|
||||
gfp_mask,
|
||||
nr_to_scan);
|
||||
} else {
|
||||
nr_freed = ion_page_pool_shrink(uncached_pool,
|
||||
gfp_mask,
|
||||
nr_to_scan);
|
||||
nr_to_scan -= nr_freed;
|
||||
/* shrink completed */
|
||||
nr_total += nr_freed;
|
||||
if (nr_to_scan <= 0)
|
||||
break;
|
||||
nr_freed = ion_page_pool_shrink(cached_pool,
|
||||
gfp_mask,
|
||||
nr_to_scan);
|
||||
nr_to_scan -= nr_freed;
|
||||
nr_total += nr_freed;
|
||||
if (nr_to_scan <= 0)
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
return nr_total;
|
||||
}
|
||||
|
||||
|
|
@ -246,52 +268,89 @@ static int ion_system_heap_debug_show(struct ion_heap *heap, struct seq_file *s,
|
|||
struct ion_system_heap,
|
||||
heap);
|
||||
int i;
|
||||
struct ion_page_pool *pool;
|
||||
|
||||
for (i = 0; i < num_orders; i++) {
|
||||
struct ion_page_pool *pool = sys_heap->pools[i];
|
||||
for (i = 0; i < NUM_ORDERS; i++) {
|
||||
pool = sys_heap->uncached_pools[i];
|
||||
|
||||
seq_printf(s, "%d order %u highmem pages in pool = %lu total\n",
|
||||
seq_printf(s, "%d order %u highmem pages uncached %lu total\n",
|
||||
pool->high_count, pool->order,
|
||||
(PAGE_SIZE << pool->order) * pool->high_count);
|
||||
seq_printf(s, "%d order %u lowmem pages in pool = %lu total\n",
|
||||
seq_printf(s, "%d order %u lowmem pages uncached %lu total\n",
|
||||
pool->low_count, pool->order,
|
||||
(PAGE_SIZE << pool->order) * pool->low_count);
|
||||
}
|
||||
|
||||
for (i = 0; i < NUM_ORDERS; i++) {
|
||||
pool = sys_heap->cached_pools[i];
|
||||
|
||||
seq_printf(s, "%d order %u highmem pages cached %lu total\n",
|
||||
pool->high_count, pool->order,
|
||||
(PAGE_SIZE << pool->order) * pool->high_count);
|
||||
seq_printf(s, "%d order %u lowmem pages cached %lu total\n",
|
||||
pool->low_count, pool->order,
|
||||
(PAGE_SIZE << pool->order) * pool->low_count);
|
||||
}
|
||||
return 0;
|
||||
}
|
||||
|
||||
static void ion_system_heap_destroy_pools(struct ion_page_pool **pools)
|
||||
{
|
||||
int i;
|
||||
|
||||
for (i = 0; i < NUM_ORDERS; i++)
|
||||
if (pools[i])
|
||||
ion_page_pool_destroy(pools[i]);
|
||||
}
|
||||
|
||||
static int ion_system_heap_create_pools(struct ion_page_pool **pools,
|
||||
bool cached)
|
||||
{
|
||||
int i;
|
||||
gfp_t gfp_flags = low_order_gfp_flags;
|
||||
|
||||
for (i = 0; i < NUM_ORDERS; i++) {
|
||||
struct ion_page_pool *pool;
|
||||
|
||||
if (orders[i] > 4)
|
||||
gfp_flags = high_order_gfp_flags;
|
||||
|
||||
pool = ion_page_pool_create(gfp_flags, orders[i], cached);
|
||||
if (!pool)
|
||||
goto err_create_pool;
|
||||
pools[i] = pool;
|
||||
}
|
||||
return 0;
|
||||
|
||||
err_create_pool:
|
||||
ion_system_heap_destroy_pools(pools);
|
||||
return -ENOMEM;
|
||||
}
|
||||
|
||||
struct ion_heap *ion_system_heap_create(struct ion_platform_heap *unused)
|
||||
{
|
||||
struct ion_system_heap *heap;
|
||||
int i;
|
||||
|
||||
heap = kzalloc(sizeof(struct ion_system_heap) +
|
||||
sizeof(struct ion_page_pool *) * num_orders,
|
||||
GFP_KERNEL);
|
||||
heap = kzalloc(sizeof(*heap), GFP_KERNEL);
|
||||
if (!heap)
|
||||
return ERR_PTR(-ENOMEM);
|
||||
heap->heap.ops = &system_heap_ops;
|
||||
heap->heap.type = ION_HEAP_TYPE_SYSTEM;
|
||||
heap->heap.flags = ION_HEAP_FLAG_DEFER_FREE;
|
||||
|
||||
for (i = 0; i < num_orders; i++) {
|
||||
struct ion_page_pool *pool;
|
||||
gfp_t gfp_flags = low_order_gfp_flags;
|
||||
if (ion_system_heap_create_pools(heap->uncached_pools, false))
|
||||
goto free_heap;
|
||||
|
||||
if (orders[i] > 4)
|
||||
gfp_flags = high_order_gfp_flags;
|
||||
pool = ion_page_pool_create(gfp_flags, orders[i]);
|
||||
if (!pool)
|
||||
goto destroy_pools;
|
||||
heap->pools[i] = pool;
|
||||
}
|
||||
if (ion_system_heap_create_pools(heap->cached_pools, true))
|
||||
goto destroy_uncached_pools;
|
||||
|
||||
heap->heap.debug_show = ion_system_heap_debug_show;
|
||||
return &heap->heap;
|
||||
|
||||
destroy_pools:
|
||||
while (i--)
|
||||
ion_page_pool_destroy(heap->pools[i]);
|
||||
destroy_uncached_pools:
|
||||
ion_system_heap_destroy_pools(heap->uncached_pools);
|
||||
|
||||
free_heap:
|
||||
kfree(heap);
|
||||
return ERR_PTR(-ENOMEM);
|
||||
}
|
||||
|
|
@ -303,8 +362,10 @@ void ion_system_heap_destroy(struct ion_heap *heap)
|
|||
heap);
|
||||
int i;
|
||||
|
||||
for (i = 0; i < num_orders; i++)
|
||||
ion_page_pool_destroy(sys_heap->pools[i]);
|
||||
for (i = 0; i < NUM_ORDERS; i++) {
|
||||
ion_page_pool_destroy(sys_heap->uncached_pools[i]);
|
||||
ion_page_pool_destroy(sys_heap->cached_pools[i]);
|
||||
}
|
||||
kfree(sys_heap);
|
||||
}
|
||||
|
||||
|
|
|
|||
Loading…
Reference in New Issue