[PATCH net-next v2 08/18] page_pool: make page_pool_put_page_bulk() actually handle array of pages

Alexander Lobakin posted 18 patches 1 month, 1 week ago
[PATCH net-next v2 08/18] page_pool: make page_pool_put_page_bulk() actually handle array of pages
Posted by Alexander Lobakin 1 month, 1 week ago
Currently, page_pool_put_page_bulk() indeed takes an array of pointers
to the data, not pages, despite the name. As one side effect, when
you're freeing frags from &skb_shared_info, xdp_return_frame_bulk()
converts page pointers to virtual addresses and then
page_pool_put_page_bulk() converts them back.
Make page_pool_put_page_bulk() actually handle array of pages. Pass
frags directly and use virt_to_page() when freeing xdpf->data, so that
the PP core will then get the compound head and take care of the rest.

Signed-off-by: Alexander Lobakin <aleksander.lobakin@intel.com>
---
 include/net/page_pool/types.h | 8 ++++----
 include/net/xdp.h             | 2 +-
 net/core/page_pool.c          | 6 +++---
 net/core/xdp.c                | 4 ++--
 4 files changed, 10 insertions(+), 10 deletions(-)

diff --git a/include/net/page_pool/types.h b/include/net/page_pool/types.h
index c022c410abe3..6c1be99a5959 100644
--- a/include/net/page_pool/types.h
+++ b/include/net/page_pool/types.h
@@ -259,8 +259,8 @@ void page_pool_disable_direct_recycling(struct page_pool *pool);
 void page_pool_destroy(struct page_pool *pool);
 void page_pool_use_xdp_mem(struct page_pool *pool, void (*disconnect)(void *),
 			   const struct xdp_mem_info *mem);
-void page_pool_put_page_bulk(struct page_pool *pool, void **data,
-			     int count);
+void page_pool_put_page_bulk(struct page_pool *pool, struct page **data,
+			     u32 count);
 #else
 static inline void page_pool_destroy(struct page_pool *pool)
 {
@@ -272,8 +272,8 @@ static inline void page_pool_use_xdp_mem(struct page_pool *pool,
 {
 }
 
-static inline void page_pool_put_page_bulk(struct page_pool *pool, void **data,
-					   int count)
+static inline void page_pool_put_page_bulk(struct page_pool *pool,
+					   struct page **data, u32 count)
 {
 }
 #endif
diff --git a/include/net/xdp.h b/include/net/xdp.h
index 3e748bb916d3..4416cd4b5086 100644
--- a/include/net/xdp.h
+++ b/include/net/xdp.h
@@ -194,7 +194,7 @@ xdp_frame_is_frag_pfmemalloc(const struct xdp_frame *frame)
 struct xdp_frame_bulk {
 	int count;
 	void *xa;
-	void *q[XDP_BULK_QUEUE_SIZE];
+	struct page *q[XDP_BULK_QUEUE_SIZE];
 };
 
 static __always_inline void xdp_frame_bulk_init(struct xdp_frame_bulk *bq)
diff --git a/net/core/page_pool.c b/net/core/page_pool.c
index a813d30d2135..ad219206ee8d 100644
--- a/net/core/page_pool.c
+++ b/net/core/page_pool.c
@@ -854,8 +854,8 @@ EXPORT_SYMBOL(page_pool_put_unrefed_page);
  * Please note the caller must not use data area after running
  * page_pool_put_page_bulk(), as this function overwrites it.
  */
-void page_pool_put_page_bulk(struct page_pool *pool, void **data,
-			     int count)
+void page_pool_put_page_bulk(struct page_pool *pool, struct page **data,
+			     u32 count)
 {
 	int i, bulk_len = 0;
 	bool allow_direct;
@@ -864,7 +864,7 @@ void page_pool_put_page_bulk(struct page_pool *pool, void **data,
 	allow_direct = page_pool_napi_local(pool);
 
 	for (i = 0; i < count; i++) {
-		netmem_ref netmem = page_to_netmem(virt_to_head_page(data[i]));
+		netmem_ref netmem = page_to_netmem(compound_head(data[i]));
 
 		/* It is not the last user for the page frag case */
 		if (!page_pool_is_last_ref(netmem))
diff --git a/net/core/xdp.c b/net/core/xdp.c
index bd2aa340baad..779e646f347b 100644
--- a/net/core/xdp.c
+++ b/net/core/xdp.c
@@ -556,12 +556,12 @@ void xdp_return_frame_bulk(struct xdp_frame *xdpf,
 		for (i = 0; i < sinfo->nr_frags; i++) {
 			skb_frag_t *frag = &sinfo->frags[i];
 
-			bq->q[bq->count++] = skb_frag_address(frag);
+			bq->q[bq->count++] = skb_frag_page(frag);
 			if (bq->count == XDP_BULK_QUEUE_SIZE)
 				xdp_flush_frame_bulk(bq);
 		}
 	}
-	bq->q[bq->count++] = xdpf->data;
+	bq->q[bq->count++] = virt_to_page(xdpf->data);
 }
 EXPORT_SYMBOL_GPL(xdp_return_frame_bulk);
 
-- 
2.46.2
Re: [PATCH net-next v2 08/18] page_pool: make page_pool_put_page_bulk() actually handle array of pages
Posted by Maciej Fijalkowski 1 month, 1 week ago
On Tue, Oct 15, 2024 at 04:53:40PM +0200, Alexander Lobakin wrote:
> Currently, page_pool_put_page_bulk() indeed takes an array of pointers
> to the data, not pages, despite the name. As one side effect, when
> you're freeing frags from &skb_shared_info, xdp_return_frame_bulk()
> converts page pointers to virtual addresses and then
> page_pool_put_page_bulk() converts them back.
> Make page_pool_put_page_bulk() actually handle array of pages. Pass
> frags directly and use virt_to_page() when freeing xdpf->data, so that
> the PP core will then get the compound head and take care of the rest.

OTOH this one makes sense to me as a part of this patchset, plus i like
that you are getting rid of virt to page dances.

Reviewed-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>

Any small improvements observed when doing micro benchmarks on your side?

> 
> Signed-off-by: Alexander Lobakin <aleksander.lobakin@intel.com>
> ---
>  include/net/page_pool/types.h | 8 ++++----
>  include/net/xdp.h             | 2 +-
>  net/core/page_pool.c          | 6 +++---
>  net/core/xdp.c                | 4 ++--
>  4 files changed, 10 insertions(+), 10 deletions(-)
> 
> diff --git a/include/net/page_pool/types.h b/include/net/page_pool/types.h
> index c022c410abe3..6c1be99a5959 100644
> --- a/include/net/page_pool/types.h
> +++ b/include/net/page_pool/types.h
> @@ -259,8 +259,8 @@ void page_pool_disable_direct_recycling(struct page_pool *pool);
>  void page_pool_destroy(struct page_pool *pool);
>  void page_pool_use_xdp_mem(struct page_pool *pool, void (*disconnect)(void *),
>  			   const struct xdp_mem_info *mem);
> -void page_pool_put_page_bulk(struct page_pool *pool, void **data,
> -			     int count);
> +void page_pool_put_page_bulk(struct page_pool *pool, struct page **data,
> +			     u32 count);
>  #else
>  static inline void page_pool_destroy(struct page_pool *pool)
>  {
> @@ -272,8 +272,8 @@ static inline void page_pool_use_xdp_mem(struct page_pool *pool,
>  {
>  }
>  
> -static inline void page_pool_put_page_bulk(struct page_pool *pool, void **data,
> -					   int count)
> +static inline void page_pool_put_page_bulk(struct page_pool *pool,
> +					   struct page **data, u32 count)
>  {
>  }
>  #endif
> diff --git a/include/net/xdp.h b/include/net/xdp.h
> index 3e748bb916d3..4416cd4b5086 100644
> --- a/include/net/xdp.h
> +++ b/include/net/xdp.h
> @@ -194,7 +194,7 @@ xdp_frame_is_frag_pfmemalloc(const struct xdp_frame *frame)
>  struct xdp_frame_bulk {
>  	int count;
>  	void *xa;
> -	void *q[XDP_BULK_QUEUE_SIZE];
> +	struct page *q[XDP_BULK_QUEUE_SIZE];
>  };
>  
>  static __always_inline void xdp_frame_bulk_init(struct xdp_frame_bulk *bq)
> diff --git a/net/core/page_pool.c b/net/core/page_pool.c
> index a813d30d2135..ad219206ee8d 100644
> --- a/net/core/page_pool.c
> +++ b/net/core/page_pool.c
> @@ -854,8 +854,8 @@ EXPORT_SYMBOL(page_pool_put_unrefed_page);
>   * Please note the caller must not use data area after running
>   * page_pool_put_page_bulk(), as this function overwrites it.
>   */
> -void page_pool_put_page_bulk(struct page_pool *pool, void **data,
> -			     int count)
> +void page_pool_put_page_bulk(struct page_pool *pool, struct page **data,
> +			     u32 count)
>  {
>  	int i, bulk_len = 0;
>  	bool allow_direct;
> @@ -864,7 +864,7 @@ void page_pool_put_page_bulk(struct page_pool *pool, void **data,
>  	allow_direct = page_pool_napi_local(pool);
>  
>  	for (i = 0; i < count; i++) {
> -		netmem_ref netmem = page_to_netmem(virt_to_head_page(data[i]));
> +		netmem_ref netmem = page_to_netmem(compound_head(data[i]));
>  
>  		/* It is not the last user for the page frag case */
>  		if (!page_pool_is_last_ref(netmem))
> diff --git a/net/core/xdp.c b/net/core/xdp.c
> index bd2aa340baad..779e646f347b 100644
> --- a/net/core/xdp.c
> +++ b/net/core/xdp.c
> @@ -556,12 +556,12 @@ void xdp_return_frame_bulk(struct xdp_frame *xdpf,
>  		for (i = 0; i < sinfo->nr_frags; i++) {
>  			skb_frag_t *frag = &sinfo->frags[i];
>  
> -			bq->q[bq->count++] = skb_frag_address(frag);
> +			bq->q[bq->count++] = skb_frag_page(frag);
>  			if (bq->count == XDP_BULK_QUEUE_SIZE)
>  				xdp_flush_frame_bulk(bq);
>  		}
>  	}
> -	bq->q[bq->count++] = xdpf->data;
> +	bq->q[bq->count++] = virt_to_page(xdpf->data);
>  }
>  EXPORT_SYMBOL_GPL(xdp_return_frame_bulk);
>  
> -- 
> 2.46.2
>
Re: [PATCH net-next v2 08/18] page_pool: make page_pool_put_page_bulk() actually handle array of pages
Posted by Alexander Lobakin 1 month ago
From: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Date: Thu, 17 Oct 2024 13:33:40 +0200

> On Tue, Oct 15, 2024 at 04:53:40PM +0200, Alexander Lobakin wrote:
>> Currently, page_pool_put_page_bulk() indeed takes an array of pointers
>> to the data, not pages, despite the name. As one side effect, when
>> you're freeing frags from &skb_shared_info, xdp_return_frame_bulk()
>> converts page pointers to virtual addresses and then
>> page_pool_put_page_bulk() converts them back.
>> Make page_pool_put_page_bulk() actually handle array of pages. Pass
>> frags directly and use virt_to_page() when freeing xdpf->data, so that
>> the PP core will then get the compound head and take care of the rest.
> 
> OTOH this one makes sense to me as a part of this patchset, plus i like
> that you are getting rid of virt to page dances.
> 
> Reviewed-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
> 
> Any small improvements observed when doing micro benchmarks on your side?

I didn't test the perf on this one without patch 9, so dunno =\

BTW this one is needed for #9, while #9 is needed for idpf as idpf can
mix up to 3 page_pools within one frame (1 for the header, 1 for a large
frag, 1 for a small frag).

Thanks,
Olek