On Wed, 4 Nov 2020 11:22:55 +0100 Lorenzo Bianconi <lorenzo@xxxxxxxxxx> wrote:
> diff --git a/net/core/page_pool.c b/net/core/page_pool.c
> index ef98372facf6..236c5ed3aa66 100644
> --- a/net/core/page_pool.c
> +++ b/net/core/page_pool.c
[...]
> @@ -408,6 +410,39 @@ void page_pool_put_page(struct page_pool *pool, struct page *page,
> }
> EXPORT_SYMBOL(page_pool_put_page);
>
> +void page_pool_put_page_bulk(struct page_pool *pool, void **data,
> + int count)
> +{
> + int i, len = 0;
> +
> + for (i = 0; i < count; i++) {
> + struct page *page = virt_to_head_page(data[i]);
> +
> + if (likely(page_ref_count(page) == 1 &&
> + pool_page_reusable(pool, page))) {
> + if (pool->p.flags & PP_FLAG_DMA_SYNC_DEV)
> + page_pool_dma_sync_for_device(pool, page, -1);
> +
> + /* bulk pages for ptr_ring cache */
> + data[len++] = page;
> + } else {
> + page_pool_release_page(pool, page);
> + put_page(page);
> + }
> + }
> +
> + /* Grab the producer spinlock for concurrent access to
> + * ptr_ring page_pool cache
> + */
> + page_pool_ring_lock(pool);
> + for (i = 0; i < len; i++) {
> + if (__ptr_ring_produce(&pool->ring, data[i]))
> + page_pool_return_page(pool, data[i]);
> + }
> + page_pool_ring_unlock(pool);
> +}
> +EXPORT_SYMBOL(page_pool_put_page_bulk);
I don't like that you are replicating the core logic from
page_pool_put_page() in this function. This means that we as
maintainers need to keep both of this places up-to-date.
Let me try to re-implement this, while sharing the refcnt logic:
(completely untested, not even compiled)
---
net/core/page_pool.c | 58 +++++++++++++++++++++++++++++++++++++++++++-------
2 files changed, 51 insertions(+), 9 deletions(-)
diff --git a/net/core/page_pool.c b/net/core/page_pool.c
index ef98372facf6..c785e9825a0d 100644
--- a/net/core/page_pool.c
+++ b/net/core/page_pool.c
@@ -362,8 +362,9 @@ static bool pool_page_reusable(struct page_pool *pool, struct page *page)
* If the page refcnt != 1, then the page will be returned to memory
* subsystem.
*/
-void page_pool_put_page(struct page_pool *pool, struct page *page,
- unsigned int dma_sync_size, bool allow_direct)
+static struct page*
+__page_pool_put_page(struct page_pool *pool, struct page *page,
+ unsigned int dma_sync_size, bool allow_direct)
{
/* This allocator is optimized for the XDP mode that uses
* one-frame-per-page, but have fallbacks that act like the
@@ -381,13 +382,10 @@ void page_pool_put_page(struct page_pool *pool, struct page *page,
if (allow_direct && in_serving_softirq())
if (page_pool_recycle_in_cache(page, pool))
- return;
+ return NULL;
- if (!page_pool_recycle_in_ring(pool, page)) {
- /* Cache full, fallback to free pages */
- page_pool_return_page(pool, page);
- }
- return;
+ /* Page found as candidate for recycling */
+ return page;
}
/* Fallback/non-XDP mode: API user have elevated refcnt.
*
@@ -405,9 +403,53 @@ void page_pool_put_page(struct page_pool *pool, struct page *page,
/* Do not replace this with page_pool_return_page() */
page_pool_release_page(pool, page);
put_page(page);
+ return NULL;
+}
+
+void page_pool_put_page(struct page_pool *pool, struct page *page,
+ unsigned int dma_sync_size, bool allow_direct)
+{
+ page = __page_pool_put_page(pool, page, dma_sync_size, allow_direct);
+
+ if (page && !page_pool_recycle_in_ring(pool, page)) {
+ /* Cache full, fallback to free pages */
+ page_pool_return_page(pool, page);
+ }
}
EXPORT_SYMBOL(page_pool_put_page);
+/* Caller must not use data area after call, as this function overwrites it */
+void page_pool_put_page_bulk(struct page_pool *pool, void **data, int count)
+{
+ int i, len = 0, len2 = 0;
+
+ for (i = 0; i < count; i++) {
+ struct page *page = virt_to_head_page(data[i]);
+
+ page = __page_pool_put_page(pool, page, -1 , false);
+
+ /* Approved for recycling for ptr_ring cache */
+ if (page)
+ data[len++] = page;
+ }
+
+ /* Bulk producer into ptr_ring page_pool cache */
+ page_pool_ring_lock(pool);
+ for (i = 0; i < len; i++) {
+ if (__ptr_ring_produce(&pool->ring, data[i]))
+ data[len2++] = data[i];
+ }
+ page_pool_ring_unlock(pool);
+
+ /* Unlikely case of ptr_ring cache full, free pages outside producer
+ * lock, given put_page() with refcnt==1 can be an expensive operation.
+ */
+ for (i = 0; i < len2; i++) {
+ page_pool_return_page(pool, data[i]);
+ }
+}
+EXPORT_SYMBOL(page_pool_put_page_bulk);
+
static void page_pool_empty_ring(struct page_pool *pool)
{
struct page *page;
--
Best regards,
Jesper Dangaard Brouer
MSc.CS, Principal Kernel Engineer at Red Hat
LinkedIn: http://www.linkedin.com/in/brouer
