Utilize the drm pagepool code to speed up allocation
performance.

This is similar to the ION pagepool usage, but tries to
utilize generic code instead of a custom implementation.

Cc: Daniel Vetter <dan...@ffwll.ch>
Cc: Christian Koenig <christian.koe...@amd.com>
Cc: Sumit Semwal <sumit.sem...@linaro.org>
Cc: Liam Mark <lm...@codeaurora.org>
Cc: Chris Goldsworthy <cgold...@codeaurora.org>
Cc: Laura Abbott <labb...@kernel.org>
Cc: Brian Starkey <brian.star...@arm.com>
Cc: Hridya Valsaraju <hri...@google.com>
Cc: Suren Baghdasaryan <sur...@google.com>
Cc: Sandeep Patil <sspa...@google.com>
Cc: Daniel Mentz <danielme...@google.com>
Cc: Ørjan Eide <orjan.e...@arm.com>
Cc: Robin Murphy <robin.mur...@arm.com>
Cc: Ezequiel Garcia <ezequ...@collabora.com>
Cc: Simon Ser <cont...@emersion.fr>
Cc: James Jones <jajo...@nvidia.com>
Cc: linux-me...@vger.kernel.org
Cc: dri-de...@lists.freedesktop.org
Signed-off-by: John Stultz <john.stu...@linaro.org>
---
v2:
* Fix build issue caused by selecting PAGE_POOL w/o NET
  as Reported-by: kernel test robot <l...@intel.com>
v3:
* Simplify the page zeroing logic a bit by using kmap_atomic
  instead of vmap as suggested by Daniel Mentz
v5:
* Shift away from networking page pool completely to
  dmabuf page pool implementation
v6:
* Switch again to using the drm_page_pool code shared w/
  ttm_pool
v7:
* Slight rework for drm_page_pool changes
v8:
* Rework to use the rewritten drm_page_pool logic
* Drop explicit buffer zeroing, as the drm_page_pool handles that
---
 drivers/dma-buf/heaps/Kconfig       |  1 +
 drivers/dma-buf/heaps/system_heap.c | 27 ++++++++++++++++++++++++---
 2 files changed, 25 insertions(+), 3 deletions(-)

diff --git a/drivers/dma-buf/heaps/Kconfig b/drivers/dma-buf/heaps/Kconfig
index f7aef8bc7119..7e28934e0def 100644
--- a/drivers/dma-buf/heaps/Kconfig
+++ b/drivers/dma-buf/heaps/Kconfig
@@ -4,6 +4,7 @@ config DMABUF_HEAPS_DEFERRED_FREE
 config DMABUF_HEAPS_SYSTEM
        bool "DMA-BUF System Heap"
        depends on DMABUF_HEAPS
+       select DRM_PAGE_POOL
        help
          Choose this option to enable the system dmabuf heap. The system heap
          is backed by pages from the buddy allocator. If in doubt, say Y.
diff --git a/drivers/dma-buf/heaps/system_heap.c 
b/drivers/dma-buf/heaps/system_heap.c
index 29e49ac17251..006271881d85 100644
--- a/drivers/dma-buf/heaps/system_heap.c
+++ b/drivers/dma-buf/heaps/system_heap.c
@@ -21,6 +21,8 @@
 #include <linux/slab.h>
 #include <linux/vmalloc.h>
 
+#include <drm/page_pool.h>
+
 static struct dma_heap *sys_heap;
 
 struct system_heap_buffer {
@@ -53,6 +55,7 @@ static gfp_t order_flags[] = {HIGH_ORDER_GFP, LOW_ORDER_GFP, 
LOW_ORDER_GFP};
  */
 static const unsigned int orders[] = {8, 4, 0};
 #define NUM_ORDERS ARRAY_SIZE(orders)
+struct drm_page_pool pools[NUM_ORDERS];
 
 static struct sg_table *dup_sg_table(struct sg_table *table)
 {
@@ -281,18 +284,28 @@ static void system_heap_vunmap(struct dma_buf *dmabuf, 
struct dma_buf_map *map)
        dma_buf_map_clear(map);
 }
 
+static unsigned long system_heap_free_pages(struct drm_page_pool *pool, struct 
page *p)
+{
+       __free_pages(p, pool->order);
+       return 1 << pool->order;
+}
+
 static void system_heap_dma_buf_release(struct dma_buf *dmabuf)
 {
        struct system_heap_buffer *buffer = dmabuf->priv;
        struct sg_table *table;
        struct scatterlist *sg;
-       int i;
+       int i, j;
 
        table = &buffer->sg_table;
        for_each_sg(table->sgl, sg, table->nents, i) {
                struct page *page = sg_page(sg);
 
-               __free_pages(page, compound_order(page));
+               for (j = 0; j < NUM_ORDERS; j++) {
+                       if (compound_order(page) == orders[j])
+                               break;
+               }
+               drm_page_pool_add(&pools[j], page);
        }
        sg_free_table(table);
        kfree(buffer);
@@ -323,7 +336,9 @@ static struct page *alloc_largest_available(unsigned long 
size,
                if (max_order < orders[i])
                        continue;
 
-               page = alloc_pages(order_flags[i], orders[i]);
+               page = drm_page_pool_remove(&pools[i]);
+               if (!page)
+                       page = alloc_pages(order_flags[i], orders[i]);
                if (!page)
                        continue;
                return page;
@@ -423,6 +438,12 @@ static const struct dma_heap_ops system_heap_ops = {
 static int system_heap_create(void)
 {
        struct dma_heap_export_info exp_info;
+       int i;
+
+       for (i = 0; i < NUM_ORDERS; i++) {
+               drm_page_pool_init(&pools[i], orders[i],
+                                  system_heap_free_pages);
+       }
 
        exp_info.name = "system";
        exp_info.ops = &system_heap_ops;
-- 
2.25.1

Reply via email to