[PATCH v2 bpf-next 0/6] xsk: struct diet and cleanups

[PATCH v2 bpf-next 0/6] xsk: struct diet and cleanups

[Maciej Fijalkowski]

Hi all,

this modest work brings back size of xdp_buff_xsk back to two cache
lines which in turn improves performance. Interestingly I was able to
observe on ice with HW rings sized to 512 around 12% better performance
when running xdpsock in l2fwd scenario. First three patches are behind
this. Other setups were not that impressive, I believe results may vary
based on the underlying CPU. Bottom line is that shrinking this struct
takes off a bit of work from CPU's shoulders.

Other three patches are rather cleanups.

Thanks,
Maciej

v1->v2:
* fix build issues on patch 1 (Daniel, CI)
* be smarter about xsk_buff_pool layout in patch 4 (Vadim)

Maciej Fijalkowski (6):
  xsk: get rid of xdp_buff_xsk::xskb_list_node
  xsk: s/free_list_node/list_node
  xsk: get rid of xdp_buff_xsk::orig_addr
  xsk: carry a copy of xdp_zc_max_segs within xsk_buff_pool
  xsk: wrap duplicated code to function
  xsk: use xsk_buff_pool directly for cq functions

 include/net/xdp_sock_drv.h  | 14 +++++-----
 include/net/xsk_buff_pool.h | 23 +++++++++-------
 net/xdp/xsk.c               | 38 +++++++++++++-------------
 net/xdp/xsk_buff_pool.c     | 54 ++++++++++++++++++++-----------------
 net/xdp/xsk_queue.h         |  2 +-
 5 files changed, 69 insertions(+), 62 deletions(-)

-- 
2.34.1

[PATCH v2 bpf-next 1/6] xsk: get rid of xdp_buff_xsk::xskb_list_node

[Maciej Fijalkowski]

Let's bring xdp_buff_xsk back to occupying 2 cachelines by removing
xskb_list_node - for the purpose of gathering the xskb frags
free_list_node can be used, head of the list (xsk_buff_pool::xskb_list)
stays as-is, just reuse the node ptr.

It is safe to do as a single xdp_buff_xsk can never reside in two
pool's lists simultaneously.

Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
---
 include/net/xdp_sock_drv.h  | 14 +++++++-------
 include/net/xsk_buff_pool.h |  1 -
 net/xdp/xsk.c               |  4 ++--
 net/xdp/xsk_buff_pool.c     |  1 -
 4 files changed, 9 insertions(+), 11 deletions(-)

diff --git a/include/net/xdp_sock_drv.h b/include/net/xdp_sock_drv.h
index 0a5dca2b2b3f..360bc1244c6a 100644
--- a/include/net/xdp_sock_drv.h
+++ b/include/net/xdp_sock_drv.h
@@ -126,8 +126,8 @@ static inline void xsk_buff_free(struct xdp_buff *xdp)
 	if (likely(!xdp_buff_has_frags(xdp)))
 		goto out;
 
-	list_for_each_entry_safe(pos, tmp, xskb_list, xskb_list_node) {
-		list_del(&pos->xskb_list_node);
+	list_for_each_entry_safe(pos, tmp, xskb_list, free_list_node) {
+		list_del(&pos->free_list_node);
 		xp_free(pos);
 	}
 
@@ -140,7 +140,7 @@ static inline void xsk_buff_add_frag(struct xdp_buff *xdp)
 {
 	struct xdp_buff_xsk *frag = container_of(xdp, struct xdp_buff_xsk, xdp);
 
-	list_add_tail(&frag->xskb_list_node, &frag->pool->xskb_list);
+	list_add_tail(&frag->free_list_node, &frag->pool->xskb_list);
 }
 
 static inline struct xdp_buff *xsk_buff_get_frag(struct xdp_buff *first)
@@ -150,9 +150,9 @@ static inline struct xdp_buff *xsk_buff_get_frag(struct xdp_buff *first)
 	struct xdp_buff_xsk *frag;
 
 	frag = list_first_entry_or_null(&xskb->pool->xskb_list,
-					struct xdp_buff_xsk, xskb_list_node);
+					struct xdp_buff_xsk, free_list_node);
 	if (frag) {
-		list_del(&frag->xskb_list_node);
+		list_del(&frag->free_list_node);
 		ret = &frag->xdp;
 	}
 
@@ -163,7 +163,7 @@ static inline void xsk_buff_del_tail(struct xdp_buff *tail)
 {
 	struct xdp_buff_xsk *xskb = container_of(tail, struct xdp_buff_xsk, xdp);
 
-	list_del(&xskb->xskb_list_node);
+	list_del(&xskb->free_list_node);
 }
 
 static inline struct xdp_buff *xsk_buff_get_tail(struct xdp_buff *first)
@@ -172,7 +172,7 @@ static inline struct xdp_buff *xsk_buff_get_tail(struct xdp_buff *first)
 	struct xdp_buff_xsk *frag;
 
 	frag = list_last_entry(&xskb->pool->xskb_list, struct xdp_buff_xsk,
-			       xskb_list_node);
+			       free_list_node);
 	return &frag->xdp;
 }
 
diff --git a/include/net/xsk_buff_pool.h b/include/net/xsk_buff_pool.h
index bacb33f1e3e5..aa7f1d0b3a5e 100644
--- a/include/net/xsk_buff_pool.h
+++ b/include/net/xsk_buff_pool.h
@@ -30,7 +30,6 @@ struct xdp_buff_xsk {
 	struct xsk_buff_pool *pool;
 	u64 orig_addr;
 	struct list_head free_list_node;
-	struct list_head xskb_list_node;
 };
 
 #define XSK_CHECK_PRIV_TYPE(t) BUILD_BUG_ON(sizeof(t) > offsetofend(struct xdp_buff_xsk, cb))
diff --git a/net/xdp/xsk.c b/net/xdp/xsk.c
index 1140b2a120ca..9c93064349a8 100644
--- a/net/xdp/xsk.c
+++ b/net/xdp/xsk.c
@@ -171,14 +171,14 @@ static int xsk_rcv_zc(struct xdp_sock *xs, struct xdp_buff *xdp, u32 len)
 		return 0;
 
 	xskb_list = &xskb->pool->xskb_list;
-	list_for_each_entry_safe(pos, tmp, xskb_list, xskb_list_node) {
+	list_for_each_entry_safe(pos, tmp, xskb_list, free_list_node) {
 		if (list_is_singular(xskb_list))
 			contd = 0;
 		len = pos->xdp.data_end - pos->xdp.data;
 		err = __xsk_rcv_zc(xs, pos, len, contd);
 		if (err)
 			goto err;
-		list_del(&pos->xskb_list_node);
+		list_del(&pos->free_list_node);
 	}
 
 	return 0;
diff --git a/net/xdp/xsk_buff_pool.c b/net/xdp/xsk_buff_pool.c
index 521a2938e50a..e5368db7d18e 100644
--- a/net/xdp/xsk_buff_pool.c
+++ b/net/xdp/xsk_buff_pool.c
@@ -102,7 +102,6 @@ struct xsk_buff_pool *xp_create_and_assign_umem(struct xdp_sock *xs,
 		xskb->pool = pool;
 		xskb->xdp.frame_sz = umem->chunk_size - umem->headroom;
 		INIT_LIST_HEAD(&xskb->free_list_node);
-		INIT_LIST_HEAD(&xskb->xskb_list_node);
 		if (pool->unaligned)
 			pool->free_heads[i] = xskb;
 		else
-- 
2.34.1

[PATCH v2 bpf-next 2/6] xsk: s/free_list_node/list_node

[Maciej Fijalkowski]

Now that free_list_node's purpose is two-folded, make it just a
'list_node'.

Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
---
 include/net/xdp_sock_drv.h  | 14 +++++++-------
 include/net/xsk_buff_pool.h |  2 +-
 net/xdp/xsk.c               |  4 ++--
 net/xdp/xsk_buff_pool.c     | 14 +++++++-------
 4 files changed, 17 insertions(+), 17 deletions(-)

diff --git a/include/net/xdp_sock_drv.h b/include/net/xdp_sock_drv.h
index 360bc1244c6a..40085afd9160 100644
--- a/include/net/xdp_sock_drv.h
+++ b/include/net/xdp_sock_drv.h
@@ -126,8 +126,8 @@ static inline void xsk_buff_free(struct xdp_buff *xdp)
 	if (likely(!xdp_buff_has_frags(xdp)))
 		goto out;
 
-	list_for_each_entry_safe(pos, tmp, xskb_list, free_list_node) {
-		list_del(&pos->free_list_node);
+	list_for_each_entry_safe(pos, tmp, xskb_list, list_node) {
+		list_del(&pos->list_node);
 		xp_free(pos);
 	}
 
@@ -140,7 +140,7 @@ static inline void xsk_buff_add_frag(struct xdp_buff *xdp)
 {
 	struct xdp_buff_xsk *frag = container_of(xdp, struct xdp_buff_xsk, xdp);
 
-	list_add_tail(&frag->free_list_node, &frag->pool->xskb_list);
+	list_add_tail(&frag->list_node, &frag->pool->xskb_list);
 }
 
 static inline struct xdp_buff *xsk_buff_get_frag(struct xdp_buff *first)
@@ -150,9 +150,9 @@ static inline struct xdp_buff *xsk_buff_get_frag(struct xdp_buff *first)
 	struct xdp_buff_xsk *frag;
 
 	frag = list_first_entry_or_null(&xskb->pool->xskb_list,
-					struct xdp_buff_xsk, free_list_node);
+					struct xdp_buff_xsk, list_node);
 	if (frag) {
-		list_del(&frag->free_list_node);
+		list_del(&frag->list_node);
 		ret = &frag->xdp;
 	}
 
@@ -163,7 +163,7 @@ static inline void xsk_buff_del_tail(struct xdp_buff *tail)
 {
 	struct xdp_buff_xsk *xskb = container_of(tail, struct xdp_buff_xsk, xdp);
 
-	list_del(&xskb->free_list_node);
+	list_del(&xskb->list_node);
 }
 
 static inline struct xdp_buff *xsk_buff_get_tail(struct xdp_buff *first)
@@ -172,7 +172,7 @@ static inline struct xdp_buff *xsk_buff_get_tail(struct xdp_buff *first)
 	struct xdp_buff_xsk *frag;
 
 	frag = list_last_entry(&xskb->pool->xskb_list, struct xdp_buff_xsk,
-			       free_list_node);
+			       list_node);
 	return &frag->xdp;
 }
 
diff --git a/include/net/xsk_buff_pool.h b/include/net/xsk_buff_pool.h
index aa7f1d0b3a5e..af8b6f776f86 100644
--- a/include/net/xsk_buff_pool.h
+++ b/include/net/xsk_buff_pool.h
@@ -29,7 +29,7 @@ struct xdp_buff_xsk {
 	dma_addr_t frame_dma;
 	struct xsk_buff_pool *pool;
 	u64 orig_addr;
-	struct list_head free_list_node;
+	struct list_head list_node;
 };
 
 #define XSK_CHECK_PRIV_TYPE(t) BUILD_BUG_ON(sizeof(t) > offsetofend(struct xdp_buff_xsk, cb))
diff --git a/net/xdp/xsk.c b/net/xdp/xsk.c
index 9c93064349a8..520023405908 100644
--- a/net/xdp/xsk.c
+++ b/net/xdp/xsk.c
@@ -171,14 +171,14 @@ static int xsk_rcv_zc(struct xdp_sock *xs, struct xdp_buff *xdp, u32 len)
 		return 0;
 
 	xskb_list = &xskb->pool->xskb_list;
-	list_for_each_entry_safe(pos, tmp, xskb_list, free_list_node) {
+	list_for_each_entry_safe(pos, tmp, xskb_list, list_node) {
 		if (list_is_singular(xskb_list))
 			contd = 0;
 		len = pos->xdp.data_end - pos->xdp.data;
 		err = __xsk_rcv_zc(xs, pos, len, contd);
 		if (err)
 			goto err;
-		list_del(&pos->free_list_node);
+		list_del(&pos->list_node);
 	}
 
 	return 0;
diff --git a/net/xdp/xsk_buff_pool.c b/net/xdp/xsk_buff_pool.c
index e5368db7d18e..973557d5e4f7 100644
--- a/net/xdp/xsk_buff_pool.c
+++ b/net/xdp/xsk_buff_pool.c
@@ -101,7 +101,7 @@ struct xsk_buff_pool *xp_create_and_assign_umem(struct xdp_sock *xs,
 		xskb = &pool->heads[i];
 		xskb->pool = pool;
 		xskb->xdp.frame_sz = umem->chunk_size - umem->headroom;
-		INIT_LIST_HEAD(&xskb->free_list_node);
+		INIT_LIST_HEAD(&xskb->list_node);
 		if (pool->unaligned)
 			pool->free_heads[i] = xskb;
 		else
@@ -549,8 +549,8 @@ struct xdp_buff *xp_alloc(struct xsk_buff_pool *pool)
 	} else {
 		pool->free_list_cnt--;
 		xskb = list_first_entry(&pool->free_list, struct xdp_buff_xsk,
-					free_list_node);
-		list_del_init(&xskb->free_list_node);
+					list_node);
+		list_del_init(&xskb->list_node);
 	}
 
 	xskb->xdp.data = xskb->xdp.data_hard_start + XDP_PACKET_HEADROOM;
@@ -616,8 +616,8 @@ static u32 xp_alloc_reused(struct xsk_buff_pool *pool, struct xdp_buff **xdp, u3
 
 	i = nb_entries;
 	while (i--) {
-		xskb = list_first_entry(&pool->free_list, struct xdp_buff_xsk, free_list_node);
-		list_del_init(&xskb->free_list_node);
+		xskb = list_first_entry(&pool->free_list, struct xdp_buff_xsk, list_node);
+		list_del_init(&xskb->list_node);
 
 		*xdp = &xskb->xdp;
 		xdp++;
@@ -687,11 +687,11 @@ EXPORT_SYMBOL(xp_can_alloc);
 
 void xp_free(struct xdp_buff_xsk *xskb)
 {
-	if (!list_empty(&xskb->free_list_node))
+	if (!list_empty(&xskb->list_node))
 		return;
 
 	xskb->pool->free_list_cnt++;
-	list_add(&xskb->free_list_node, &xskb->pool->free_list);
+	list_add(&xskb->list_node, &xskb->pool->free_list);
 }
 EXPORT_SYMBOL(xp_free);
 
-- 
2.34.1

[PATCH v2 bpf-next 3/6] xsk: get rid of xdp_buff_xsk::orig_addr

[Maciej Fijalkowski]

Continue the process of dieting xdp_buff_xsk by removing orig_addr
member. It can be calculated from xdp->data_hard_start where it was
previously used, so it is not anything that has to be carried around in
struct used widely in hot path.

This has been used for initializing xdp_buff_xsk::frame_dma during pool
setup and as a shortcut in xp_get_handle() to retrieve address provided
to xsk Rx queue.

Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
---
 include/net/xsk_buff_pool.h | 19 +++++++++++--------
 net/xdp/xsk.c               |  2 +-
 net/xdp/xsk_buff_pool.c     |  4 +++-
 3 files changed, 15 insertions(+), 10 deletions(-)

diff --git a/include/net/xsk_buff_pool.h b/include/net/xsk_buff_pool.h
index af8b6f776f86..468a23b1b4c5 100644
--- a/include/net/xsk_buff_pool.h
+++ b/include/net/xsk_buff_pool.h
@@ -28,7 +28,6 @@ struct xdp_buff_xsk {
 	dma_addr_t dma;
 	dma_addr_t frame_dma;
 	struct xsk_buff_pool *pool;
-	u64 orig_addr;
 	struct list_head list_node;
 };
 
@@ -119,7 +118,6 @@ void xp_free(struct xdp_buff_xsk *xskb);
 static inline void xp_init_xskb_addr(struct xdp_buff_xsk *xskb, struct xsk_buff_pool *pool,
 				     u64 addr)
 {
-	xskb->orig_addr = addr;
 	xskb->xdp.data_hard_start = pool->addrs + addr + pool->headroom;
 }
 
@@ -221,14 +219,19 @@ static inline void xp_release(struct xdp_buff_xsk *xskb)
 		xskb->pool->free_heads[xskb->pool->free_heads_cnt++] = xskb;
 }
 
-static inline u64 xp_get_handle(struct xdp_buff_xsk *xskb)
+static inline u64 xp_get_handle(struct xdp_buff_xsk *xskb,
+				struct xsk_buff_pool *pool)
 {
-	u64 offset = xskb->xdp.data - xskb->xdp.data_hard_start;
+	u64 orig_addr = xskb->xdp.data - pool->addrs;
+	u64 offset;
 
-	offset += xskb->pool->headroom;
-	if (!xskb->pool->unaligned)
-		return xskb->orig_addr + offset;
-	return xskb->orig_addr + (offset << XSK_UNALIGNED_BUF_OFFSET_SHIFT);
+	if (!pool->unaligned)
+		return orig_addr;
+
+	offset = xskb->xdp.data - xskb->xdp.data_hard_start;
+	orig_addr -= offset;
+	offset += pool->headroom;
+	return orig_addr + (offset << XSK_UNALIGNED_BUF_OFFSET_SHIFT);
 }
 
 static inline bool xp_tx_metadata_enabled(const struct xsk_buff_pool *pool)
diff --git a/net/xdp/xsk.c b/net/xdp/xsk.c
index 520023405908..6c31c1de1619 100644
--- a/net/xdp/xsk.c
+++ b/net/xdp/xsk.c
@@ -141,7 +141,7 @@ static int __xsk_rcv_zc(struct xdp_sock *xs, struct xdp_buff_xsk *xskb, u32 len,
 	u64 addr;
 	int err;
 
-	addr = xp_get_handle(xskb);
+	addr = xp_get_handle(xskb, xskb->pool);
 	err = xskq_prod_reserve_desc(xs->rx, addr, len, flags);
 	if (err) {
 		xs->rx_queue_full++;
diff --git a/net/xdp/xsk_buff_pool.c b/net/xdp/xsk_buff_pool.c
index 973557d5e4f7..7ecd4ccd2473 100644
--- a/net/xdp/xsk_buff_pool.c
+++ b/net/xdp/xsk_buff_pool.c
@@ -416,8 +416,10 @@ static int xp_init_dma_info(struct xsk_buff_pool *pool, struct xsk_dma_map *dma_
 
 		for (i = 0; i < pool->heads_cnt; i++) {
 			struct xdp_buff_xsk *xskb = &pool->heads[i];
+			u64 orig_addr;
 
-			xp_init_xskb_dma(xskb, pool, dma_map->dma_pages, xskb->orig_addr);
+			orig_addr = xskb->xdp.data_hard_start - pool->addrs - pool->headroom;
+			xp_init_xskb_dma(xskb, pool, dma_map->dma_pages, orig_addr);
 		}
 	}
 
-- 
2.34.1

[PATCH v2 bpf-next 4/6] xsk: carry a copy of xdp_zc_max_segs within xsk_buff_pool

[Maciej Fijalkowski]

This so we avoid dereferencing struct net_device within hot path.

Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
---
 include/net/xsk_buff_pool.h | 1 +
 net/xdp/xsk_buff_pool.c     | 1 +
 net/xdp/xsk_queue.h         | 2 +-
 3 files changed, 3 insertions(+), 1 deletion(-)

diff --git a/include/net/xsk_buff_pool.h b/include/net/xsk_buff_pool.h
index 468a23b1b4c5..bb03cee716b3 100644
--- a/include/net/xsk_buff_pool.h
+++ b/include/net/xsk_buff_pool.h
@@ -76,6 +76,7 @@ struct xsk_buff_pool {
 	u32 chunk_size;
 	u32 chunk_shift;
 	u32 frame_len;
+	u32 xdp_zc_max_segs;
 	u8 tx_metadata_len; /* inherited from umem */
 	u8 cached_need_wakeup;
 	bool uses_need_wakeup;
diff --git a/net/xdp/xsk_buff_pool.c b/net/xdp/xsk_buff_pool.c
index 7ecd4ccd2473..e946ba4a5ccf 100644
--- a/net/xdp/xsk_buff_pool.c
+++ b/net/xdp/xsk_buff_pool.c
@@ -229,6 +229,7 @@ int xp_assign_dev(struct xsk_buff_pool *pool,
 		goto err_unreg_xsk;
 	}
 	pool->umem->zc = true;
+	pool->xdp_zc_max_segs = netdev->xdp_zc_max_segs;
 	return 0;
 
 err_unreg_xsk:
diff --git a/net/xdp/xsk_queue.h b/net/xdp/xsk_queue.h
index 406b20dfee8d..46d87e961ad6 100644
--- a/net/xdp/xsk_queue.h
+++ b/net/xdp/xsk_queue.h
@@ -260,7 +260,7 @@ u32 xskq_cons_read_desc_batch(struct xsk_queue *q, struct xsk_buff_pool *pool,
 			nr_frags = 0;
 		} else {
 			nr_frags++;
-			if (nr_frags == pool->netdev->xdp_zc_max_segs) {
+			if (nr_frags == pool->xdp_zc_max_segs) {
 				nr_frags = 0;
 				break;
 			}
-- 
2.34.1

[PATCH v2 bpf-next 5/6] xsk: wrap duplicated code to function

[Maciej Fijalkowski]

Both allocation paths have exactly the same code responsible for getting
and initializing xskb. Pull it out to common function.

Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
---
 net/xdp/xsk_buff_pool.c | 34 ++++++++++++++++++----------------
 1 file changed, 18 insertions(+), 16 deletions(-)

diff --git a/net/xdp/xsk_buff_pool.c b/net/xdp/xsk_buff_pool.c
index e946ba4a5ccf..ae71da7d2cd6 100644
--- a/net/xdp/xsk_buff_pool.c
+++ b/net/xdp/xsk_buff_pool.c
@@ -503,6 +503,22 @@ static bool xp_check_aligned(struct xsk_buff_pool *pool, u64 *addr)
 	return *addr < pool->addrs_cnt;
 }
 
+static struct xdp_buff_xsk *xp_get_xskb(struct xsk_buff_pool *pool, u64 addr)
+{
+	struct xdp_buff_xsk *xskb;
+
+	if (pool->unaligned) {
+		xskb = pool->free_heads[--pool->free_heads_cnt];
+		xp_init_xskb_addr(xskb, pool, addr);
+		if (pool->dma_pages)
+			xp_init_xskb_dma(xskb, pool, pool->dma_pages, addr);
+	} else {
+		xskb = &pool->heads[xp_aligned_extract_idx(pool, addr)];
+	}
+
+	return xskb;
+}
+
 static struct xdp_buff_xsk *__xp_alloc(struct xsk_buff_pool *pool)
 {
 	struct xdp_buff_xsk *xskb;
@@ -528,14 +544,7 @@ static struct xdp_buff_xsk *__xp_alloc(struct xsk_buff_pool *pool)
 		break;
 	}
 
-	if (pool->unaligned) {
-		xskb = pool->free_heads[--pool->free_heads_cnt];
-		xp_init_xskb_addr(xskb, pool, addr);
-		if (pool->dma_pages)
-			xp_init_xskb_dma(xskb, pool, pool->dma_pages, addr);
-	} else {
-		xskb = &pool->heads[xp_aligned_extract_idx(pool, addr)];
-	}
+	xskb = xp_get_xskb(pool, addr);
 
 	xskq_cons_release(pool->fq);
 	return xskb;
@@ -593,14 +602,7 @@ static u32 xp_alloc_new_from_fq(struct xsk_buff_pool *pool, struct xdp_buff **xd
 			continue;
 		}
 
-		if (pool->unaligned) {
-			xskb = pool->free_heads[--pool->free_heads_cnt];
-			xp_init_xskb_addr(xskb, pool, addr);
-			if (pool->dma_pages)
-				xp_init_xskb_dma(xskb, pool, pool->dma_pages, addr);
-		} else {
-			xskb = &pool->heads[xp_aligned_extract_idx(pool, addr)];
-		}
+		xskb = xp_get_xskb(pool, addr);
 
 		*xdp = &xskb->xdp;
 		xdp++;
-- 
2.34.1

[PATCH v2 bpf-next 6/6] xsk: use xsk_buff_pool directly for cq functions

[Maciej Fijalkowski]

Currently xsk_cq_{reserve_addr,submit,cancel}_locked() take xdp_sock as
an input argument but it is only used for pulling out xsk_buff_pool
pointer from it.

Change mentioned functions to take pool pointer as an input argument to
avoid unnecessary dereferences.

Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
---
 net/xdp/xsk.c | 32 ++++++++++++++++----------------
 1 file changed, 16 insertions(+), 16 deletions(-)

diff --git a/net/xdp/xsk.c b/net/xdp/xsk.c
index 6c31c1de1619..7d7e37f53708 100644
--- a/net/xdp/xsk.c
+++ b/net/xdp/xsk.c
@@ -527,34 +527,34 @@ static int xsk_wakeup(struct xdp_sock *xs, u8 flags)
 	return dev->netdev_ops->ndo_xsk_wakeup(dev, xs->queue_id, flags);
 }
 
-static int xsk_cq_reserve_addr_locked(struct xdp_sock *xs, u64 addr)
+static int xsk_cq_reserve_addr_locked(struct xsk_buff_pool *pool, u64 addr)
 {
 	unsigned long flags;
 	int ret;
 
-	spin_lock_irqsave(&xs->pool->cq_lock, flags);
-	ret = xskq_prod_reserve_addr(xs->pool->cq, addr);
-	spin_unlock_irqrestore(&xs->pool->cq_lock, flags);
+	spin_lock_irqsave(&pool->cq_lock, flags);
+	ret = xskq_prod_reserve_addr(pool->cq, addr);
+	spin_unlock_irqrestore(&pool->cq_lock, flags);
 
 	return ret;
 }
 
-static void xsk_cq_submit_locked(struct xdp_sock *xs, u32 n)
+static void xsk_cq_submit_locked(struct xsk_buff_pool *pool, u32 n)
 {
 	unsigned long flags;
 
-	spin_lock_irqsave(&xs->pool->cq_lock, flags);
-	xskq_prod_submit_n(xs->pool->cq, n);
-	spin_unlock_irqrestore(&xs->pool->cq_lock, flags);
+	spin_lock_irqsave(&pool->cq_lock, flags);
+	xskq_prod_submit_n(pool->cq, n);
+	spin_unlock_irqrestore(&pool->cq_lock, flags);
 }
 
-static void xsk_cq_cancel_locked(struct xdp_sock *xs, u32 n)
+static void xsk_cq_cancel_locked(struct xsk_buff_pool *pool, u32 n)
 {
 	unsigned long flags;
 
-	spin_lock_irqsave(&xs->pool->cq_lock, flags);
-	xskq_prod_cancel_n(xs->pool->cq, n);
-	spin_unlock_irqrestore(&xs->pool->cq_lock, flags);
+	spin_lock_irqsave(&pool->cq_lock, flags);
+	xskq_prod_cancel_n(pool->cq, n);
+	spin_unlock_irqrestore(&pool->cq_lock, flags);
 }
 
 static u32 xsk_get_num_desc(struct sk_buff *skb)
@@ -571,7 +571,7 @@ static void xsk_destruct_skb(struct sk_buff *skb)
 		*compl->tx_timestamp = ktime_get_tai_fast_ns();
 	}
 
-	xsk_cq_submit_locked(xdp_sk(skb->sk), xsk_get_num_desc(skb));
+	xsk_cq_submit_locked(xdp_sk(skb->sk)->pool, xsk_get_num_desc(skb));
 	sock_wfree(skb);
 }
 
@@ -587,7 +587,7 @@ static void xsk_consume_skb(struct sk_buff *skb)
 	struct xdp_sock *xs = xdp_sk(skb->sk);
 
 	skb->destructor = sock_wfree;
-	xsk_cq_cancel_locked(xs, xsk_get_num_desc(skb));
+	xsk_cq_cancel_locked(xs->pool, xsk_get_num_desc(skb));
 	/* Free skb without triggering the perf drop trace */
 	consume_skb(skb);
 	xs->skb = NULL;
@@ -765,7 +765,7 @@ static struct sk_buff *xsk_build_skb(struct xdp_sock *xs,
 		xskq_cons_release(xs->tx);
 	} else {
 		/* Let application retry */
-		xsk_cq_cancel_locked(xs, 1);
+		xsk_cq_cancel_locked(xs->pool, 1);
 	}
 
 	return ERR_PTR(err);
@@ -802,7 +802,7 @@ static int __xsk_generic_xmit(struct sock *sk)
 		 * if there is space in it. This avoids having to implement
 		 * any buffering in the Tx path.
 		 */
-		if (xsk_cq_reserve_addr_locked(xs, desc.addr))
+		if (xsk_cq_reserve_addr_locked(xs->pool, desc.addr))
 			goto out;
 
 		skb = xsk_build_skb(xs, &desc);
-- 
2.34.1

Re: [PATCH v2 bpf-next 0/6] xsk: struct diet and cleanups

[Magnus Karlsson]

On Mon, 7 Oct 2024 at 14:27, Maciej Fijalkowski
<maciej.fijalkowski@intel.com> wrote:
>
> Hi all,
>
> this modest work brings back size of xdp_buff_xsk back to two cache
> lines which in turn improves performance. Interestingly I was able to
> observe on ice with HW rings sized to 512 around 12% better performance
> when running xdpsock in l2fwd scenario. First three patches are behind
> this. Other setups were not that impressive, I believe results may vary
> based on the underlying CPU. Bottom line is that shrinking this struct
> takes off a bit of work from CPU's shoulders.
>
> Other three patches are rather cleanups.
>
> Thanks,
> Maciej

Thanks for these improvements Maciej.

For the series:
Acked-by: Magnus Karlsson <magnus.karlsson@intel.com>

> v1->v2:
> * fix build issues on patch 1 (Daniel, CI)
> * be smarter about xsk_buff_pool layout in patch 4 (Vadim)
>
> Maciej Fijalkowski (6):
>   xsk: get rid of xdp_buff_xsk::xskb_list_node
>   xsk: s/free_list_node/list_node
>   xsk: get rid of xdp_buff_xsk::orig_addr
>   xsk: carry a copy of xdp_zc_max_segs within xsk_buff_pool
>   xsk: wrap duplicated code to function
>   xsk: use xsk_buff_pool directly for cq functions
>
>  include/net/xdp_sock_drv.h  | 14 +++++-----
>  include/net/xsk_buff_pool.h | 23 +++++++++-------
>  net/xdp/xsk.c               | 38 +++++++++++++-------------
>  net/xdp/xsk_buff_pool.c     | 54 ++++++++++++++++++++-----------------
>  net/xdp/xsk_queue.h         |  2 +-
>  5 files changed, 69 insertions(+), 62 deletions(-)
>
> --
> 2.34.1
>
>

Re: [PATCH v2 bpf-next 0/6] xsk: struct diet and cleanups

[patchwork-bot+netdevbpf]

Hello:

This series was applied to bpf/bpf-next.git (net)
by Daniel Borkmann <daniel@iogearbox.net>:

On Mon,  7 Oct 2024 14:24:52 +0200 you wrote:
> Hi all,
> 
> this modest work brings back size of xdp_buff_xsk back to two cache
> lines which in turn improves performance. Interestingly I was able to
> observe on ice with HW rings sized to 512 around 12% better performance
> when running xdpsock in l2fwd scenario. First three patches are behind
> this. Other setups were not that impressive, I believe results may vary
> based on the underlying CPU. Bottom line is that shrinking this struct
> takes off a bit of work from CPU's shoulders.
> 
> [...]

Here is the summary with links:
  - [v2,bpf-next,1/6] xsk: get rid of xdp_buff_xsk::xskb_list_node
    https://git.kernel.org/bpf/bpf-next/c/b692bf9a7543
  - [v2,bpf-next,2/6] xsk: s/free_list_node/list_node
    https://git.kernel.org/bpf/bpf-next/c/30ec2c1baaea
  - [v2,bpf-next,3/6] xsk: get rid of xdp_buff_xsk::orig_addr
    https://git.kernel.org/bpf/bpf-next/c/bea14124bacb
  - [v2,bpf-next,4/6] xsk: carry a copy of xdp_zc_max_segs within xsk_buff_pool
    https://git.kernel.org/bpf/bpf-next/c/6e126872191d
  - [v2,bpf-next,5/6] xsk: wrap duplicated code to function
    https://git.kernel.org/bpf/bpf-next/c/1d10b2bed2d4
  - [v2,bpf-next,6/6] xsk: use xsk_buff_pool directly for cq functions
    https://git.kernel.org/bpf/bpf-next/c/e6c4047f5122

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html