1 files changed, 155 insertions, 113 deletions

diff --git a/net/xdp/xsk_queue.h b/net/xdp/xsk_queue.h
index 89a01ac4e079..c6fb6b763658 100644
--- a/net/xdp/xsk_queue.h
+++ b/net/xdp/xsk_queue.h
@@ -9,49 +9,56 @@
 #include <linux/types.h>
 #include <linux/if_xdp.h>
 #include <net/xdp_sock.h>
+#include <net/xsk_buff_pool.h>
+
+#include "xsk.h"
 
 struct xdp_ring {
 	u32 producer ____cacheline_aligned_in_smp;
+	/* Hinder the adjacent cache prefetcher to prefetch the consumer
+	 * pointer if the producer pointer is touched and vice versa.
+	 */
+	u32 pad1 ____cacheline_aligned_in_smp;
 	u32 consumer ____cacheline_aligned_in_smp;
+	u32 pad2 ____cacheline_aligned_in_smp;
 	u32 flags;
+	u32 pad3 ____cacheline_aligned_in_smp;
 };
 
 /* Used for the RX and TX queues for packets */
 struct xdp_rxtx_ring {
 	struct xdp_ring ptrs;
-	struct xdp_desc desc[0] ____cacheline_aligned_in_smp;
+	struct xdp_desc desc[] ____cacheline_aligned_in_smp;
 };
 
 /* Used for the fill and completion queues for buffers */
 struct xdp_umem_ring {
 	struct xdp_ring ptrs;
-	u64 desc[0] ____cacheline_aligned_in_smp;
+	u64 desc[] ____cacheline_aligned_in_smp;
 };
 
 struct xsk_queue {
-	u64 chunk_mask;
-	u64 size;
 	u32 ring_mask;
 	u32 nentries;
 	u32 cached_prod;
 	u32 cached_cons;
 	struct xdp_ring *ring;
 	u64 invalid_descs;
+	u64 queue_empty_descs;
 };
 
-/* The structure of the shared state of the rings are the same as the
- * ring buffer in kernel/events/ring_buffer.c. For the Rx and completion
- * ring, the kernel is the producer and user space is the consumer. For
- * the Tx and fill rings, the kernel is the consumer and user space is
- * the producer.
+/* The structure of the shared state of the rings are a simple
+ * circular buffer, as outlined in
+ * Documentation/core-api/circular-buffers.rst. For the Rx and
+ * completion ring, the kernel is the producer and user space is the
+ * consumer. For the Tx and fill rings, the kernel is the consumer and
+ * user space is the producer.
  *
  * producer                         consumer
  *
- * if (LOAD ->consumer) {           LOAD ->producer
- *                    (A)           smp_rmb()       (C)
+ * if (LOAD ->consumer) {  (A)      LOAD.acq ->producer  (C)
  *    STORE $data                   LOAD $data
- *    smp_wmb()       (B)           smp_mb()        (D)
- *    STORE ->producer              STORE ->consumer
+ *    STORE.rel ->producer (B)      STORE.rel ->consumer (D)
  * }
  *
  * (A) pairs with (D), and (B) pairs with (C).
@@ -70,7 +77,8 @@ struct xsk_queue {
  *
  * (A) is a control dependency that separates the load of ->consumer
  * from the stores of $data. In case ->consumer indicates there is no
- * room in the buffer to store $data we do not. So no barrier is needed.
+ * room in the buffer to store $data we do not. The dependency will
+ * order both of the stores after the loads. So no barrier is needed.
  *
  * (D) protects the load of the data to be observed to happen after the
  * store of the consumer pointer. If we did not have this memory
@@ -94,7 +102,7 @@ struct xsk_queue {
  * seen and read by the consumer.
  *
  * The consumer peeks into the ring to see if the producer has written
- * any new entries. If so, the producer can then read these entries
+ * any new entries. If so, the consumer can then read these entries
  * and when it is done reading them release them back to the producer
  * so that the producer can use these slots to fill in new entries.
  *
@@ -103,111 +111,92 @@ struct xsk_queue {
 
 /* Functions that read and validate content from consumer rings. */
 
-static inline bool xskq_cons_crosses_non_contig_pg(struct xdp_umem *umem,
-						   u64 addr,
-						   u64 length)
+static inline void __xskq_cons_read_addr_unchecked(struct xsk_queue *q, u32 cached_cons, u64 *addr)
 {
-	bool cross_pg = (addr & (PAGE_SIZE - 1)) + length > PAGE_SIZE;
-	bool next_pg_contig =
-		(unsigned long)umem->pages[(addr >> PAGE_SHIFT)].addr &
-			XSK_NEXT_PG_CONTIG_MASK;
+	struct xdp_umem_ring *ring = (struct xdp_umem_ring *)q->ring;
+	u32 idx = cached_cons & q->ring_mask;
 
-	return cross_pg && !next_pg_contig;
+	*addr = ring->desc[idx];
 }
 
-static inline bool xskq_cons_is_valid_unaligned(struct xsk_queue *q,
-						u64 addr,
-						u64 length,
-						struct xdp_umem *umem)
+static inline bool xskq_cons_read_addr_unchecked(struct xsk_queue *q, u64 *addr)
 {
-	u64 base_addr = xsk_umem_extract_addr(addr);
-
-	addr = xsk_umem_add_offset_to_addr(addr);
-	if (base_addr >= q->size || addr >= q->size ||
-	    xskq_cons_crosses_non_contig_pg(umem, addr, length)) {
-		q->invalid_descs++;
-		return false;
+	if (q->cached_cons != q->cached_prod) {
+		__xskq_cons_read_addr_unchecked(q, q->cached_cons, addr);
+		return true;
 	}
 
-	return true;
+	return false;
 }
 
-static inline bool xskq_cons_is_valid_addr(struct xsk_queue *q, u64 addr)
+static inline bool xp_aligned_validate_desc(struct xsk_buff_pool *pool,
+					    struct xdp_desc *desc)
 {
-	if (addr >= q->size) {
-		q->invalid_descs++;
-		return false;
+	u64 chunk, chunk_end;
+
+	chunk = xp_aligned_extract_addr(pool, desc->addr);
+	if (likely(desc->len)) {
+		chunk_end = xp_aligned_extract_addr(pool, desc->addr + desc->len - 1);
+		if (chunk != chunk_end)
+			return false;
 	}
 
+	if (chunk >= pool->addrs_cnt)
+		return false;
+
+	if (desc->options)
+		return false;
 	return true;
 }
 
-static inline bool xskq_cons_read_addr(struct xsk_queue *q, u64 *addr,
-				       struct xdp_umem *umem)
+static inline bool xp_unaligned_validate_desc(struct xsk_buff_pool *pool,
+					      struct xdp_desc *desc)
 {
-	struct xdp_umem_ring *ring = (struct xdp_umem_ring *)q->ring;
+	u64 addr, base_addr;
 
-	while (q->cached_cons != q->cached_prod) {
-		u32 idx = q->cached_cons & q->ring_mask;
+	base_addr = xp_unaligned_extract_addr(desc->addr);
+	addr = xp_unaligned_add_offset_to_addr(desc->addr);
 
-		*addr = ring->desc[idx] & q->chunk_mask;
-
-		if (umem->flags & XDP_UMEM_UNALIGNED_CHUNK_FLAG) {
-			if (xskq_cons_is_valid_unaligned(q, *addr,
-							 umem->chunk_size_nohr,
-							 umem))
-				return true;
-			goto out;
-		}
+	if (desc->len > pool->chunk_size)
+		return false;
 
-		if (xskq_cons_is_valid_addr(q, *addr))
-			return true;
+	if (base_addr >= pool->addrs_cnt || addr >= pool->addrs_cnt ||
+	    xp_desc_crosses_non_contig_pg(pool, addr, desc->len))
+		return false;
 
-out:
-		q->cached_cons++;
-	}
+	if (desc->options)
+		return false;
+	return true;
+}
 
-	return false;
+static inline bool xp_validate_desc(struct xsk_buff_pool *pool,
+				    struct xdp_desc *desc)
+{
+	return pool->unaligned ? xp_unaligned_validate_desc(pool, desc) :
+		xp_aligned_validate_desc(pool, desc);
 }
 
 static inline bool xskq_cons_is_valid_desc(struct xsk_queue *q,
 					   struct xdp_desc *d,
-					   struct xdp_umem *umem)
+					   struct xsk_buff_pool *pool)
 {
-	if (umem->flags & XDP_UMEM_UNALIGNED_CHUNK_FLAG) {
-		if (!xskq_cons_is_valid_unaligned(q, d->addr, d->len, umem))
-			return false;
-
-		if (d->len > umem->chunk_size_nohr || d->options) {
-			q->invalid_descs++;
-			return false;
-		}
-
-		return true;
-	}
-
-	if (!xskq_cons_is_valid_addr(q, d->addr))
-		return false;
-
-	if (((d->addr + d->len) & q->chunk_mask) != (d->addr & q->chunk_mask) ||
-	    d->options) {
+	if (!xp_validate_desc(pool, d)) {
 		q->invalid_descs++;
 		return false;
 	}
-
 	return true;
 }
 
 static inline bool xskq_cons_read_desc(struct xsk_queue *q,
 				       struct xdp_desc *desc,
-				       struct xdp_umem *umem)
+				       struct xsk_buff_pool *pool)
 {
 	while (q->cached_cons != q->cached_prod) {
 		struct xdp_rxtx_ring *ring = (struct xdp_rxtx_ring *)q->ring;
 		u32 idx = q->cached_cons & q->ring_mask;
 
 		*desc = ring->desc[idx];
-		if (xskq_cons_is_valid_desc(q, desc, umem))
+		if (xskq_cons_is_valid_desc(q, desc, pool))
 			return true;
 
 		q->cached_cons++;
@@ -216,19 +205,48 @@ static inline bool xskq_cons_read_desc(struct xsk_queue *q,
 	return false;
 }
 
+static inline void xskq_cons_release_n(struct xsk_queue *q, u32 cnt)
+{
+	q->cached_cons += cnt;
+}
+
+static inline u32 xskq_cons_read_desc_batch(struct xsk_queue *q, struct xsk_buff_pool *pool,
+					    u32 max)
+{
+	u32 cached_cons = q->cached_cons, nb_entries = 0;
+	struct xdp_desc *descs = pool->tx_descs;
+
+	while (cached_cons != q->cached_prod && nb_entries < max) {
+		struct xdp_rxtx_ring *ring = (struct xdp_rxtx_ring *)q->ring;
+		u32 idx = cached_cons & q->ring_mask;
+
+		descs[nb_entries] = ring->desc[idx];
+		if (unlikely(!xskq_cons_is_valid_desc(q, &descs[nb_entries], pool))) {
+			/* Skip the entry */
+			cached_cons++;
+			continue;
+		}
+
+		nb_entries++;
+		cached_cons++;
+	}
+
+	/* Release valid plus any invalid entries */
+	xskq_cons_release_n(q, cached_cons - q->cached_cons);
+	return nb_entries;
+}
+
 /* Functions for consumers */
 
 static inline void __xskq_cons_release(struct xsk_queue *q)
 {
-	smp_mb(); /* D, matches A */
-	WRITE_ONCE(q->ring->consumer, q->cached_cons);
+	smp_store_release(&q->ring->consumer, q->cached_cons); /* D, matchees A */
 }
 
 static inline void __xskq_cons_peek(struct xsk_queue *q)
 {
 	/* Refresh the local pointer */
-	q->cached_prod = READ_ONCE(q->ring->producer);
-	smp_rmb(); /* C, matches B */
+	q->cached_prod = smp_load_acquire(&q->ring->producer);  /* C, matches B */
 }
 
 static inline void xskq_cons_get_entries(struct xsk_queue *q)
@@ -237,67 +255,79 @@ static inline void xskq_cons_get_entries(struct xsk_queue *q)
 	__xskq_cons_peek(q);
 }
 
-static inline bool xskq_cons_has_entries(struct xsk_queue *q, u32 cnt)
+static inline u32 xskq_cons_nb_entries(struct xsk_queue *q, u32 max)
 {
 	u32 entries = q->cached_prod - q->cached_cons;
 
-	if (entries >= cnt)
-		return true;
+	if (entries >= max)
+		return max;
 
 	__xskq_cons_peek(q);
 	entries = q->cached_prod - q->cached_cons;
 
-	return entries >= cnt;
+	return entries >= max ? max : entries;
+}
+
+static inline bool xskq_cons_has_entries(struct xsk_queue *q, u32 cnt)
+{
+	return xskq_cons_nb_entries(q, cnt) >= cnt;
 }
 
-static inline bool xskq_cons_peek_addr(struct xsk_queue *q, u64 *addr,
-				       struct xdp_umem *umem)
+static inline bool xskq_cons_peek_addr_unchecked(struct xsk_queue *q, u64 *addr)
 {
 	if (q->cached_prod == q->cached_cons)
 		xskq_cons_get_entries(q);
-	return xskq_cons_read_addr(q, addr, umem);
+	return xskq_cons_read_addr_unchecked(q, addr);
 }
 
 static inline bool xskq_cons_peek_desc(struct xsk_queue *q,
 				       struct xdp_desc *desc,
-				       struct xdp_umem *umem)
+				       struct xsk_buff_pool *pool)
 {
 	if (q->cached_prod == q->cached_cons)
 		xskq_cons_get_entries(q);
-	return xskq_cons_read_desc(q, desc, umem);
+	return xskq_cons_read_desc(q, desc, pool);
 }
 
+/* To improve performance in the xskq_cons_release functions, only update local state here.
+ * Reflect this to global state when we get new entries from the ring in
+ * xskq_cons_get_entries() and whenever Rx or Tx processing are completed in the NAPI loop.
+ */
 static inline void xskq_cons_release(struct xsk_queue *q)
 {
-	/* To improve performance, only update local state here.
-	 * Reflect this to global state when we get new entries
-	 * from the ring in xskq_cons_get_entries() and whenever
-	 * Rx or Tx processing are completed in the NAPI loop.
-	 */
 	q->cached_cons++;
 }
 
-static inline bool xskq_cons_is_full(struct xsk_queue *q)
+static inline u32 xskq_cons_present_entries(struct xsk_queue *q)
 {
 	/* No barriers needed since data is not accessed */
-	return READ_ONCE(q->ring->producer) - READ_ONCE(q->ring->consumer) ==
-		q->nentries;
+	return READ_ONCE(q->ring->producer) - READ_ONCE(q->ring->consumer);
 }
 
 /* Functions for producers */
 
-static inline bool xskq_prod_is_full(struct xsk_queue *q)
+static inline u32 xskq_prod_nb_free(struct xsk_queue *q, u32 max)
 {
 	u32 free_entries = q->nentries - (q->cached_prod - q->cached_cons);
 
-	if (free_entries)
-		return false;
+	if (free_entries >= max)
+		return max;
 
 	/* Refresh the local tail pointer */
 	q->cached_cons = READ_ONCE(q->ring->consumer);
 	free_entries = q->nentries - (q->cached_prod - q->cached_cons);
 
-	return !free_entries;
+	return free_entries >= max ? max : free_entries;
+}
+
+static inline bool xskq_prod_is_full(struct xsk_queue *q)
+{
+	return xskq_prod_nb_free(q, 1) ? false : true;
+}
+
+static inline void xskq_prod_cancel(struct xsk_queue *q)
+{
+	q->cached_prod--;
 }
 
 static inline int xskq_prod_reserve(struct xsk_queue *q)
@@ -322,6 +352,19 @@ static inline int xskq_prod_reserve_addr(struct xsk_queue *q, u64 addr)
 	return 0;
 }
 
+static inline void xskq_prod_write_addr_batch(struct xsk_queue *q, struct xdp_desc *descs,
+					      u32 nb_entries)
+{
+	struct xdp_umem_ring *ring = (struct xdp_umem_ring *)q->ring;
+	u32 i, cached_prod;
+
+	/* A, matches D */
+	cached_prod = q->cached_prod;
+	for (i = 0; i < nb_entries; i++)
+		ring->desc[cached_prod++ & q->ring_mask] = descs[i].addr;
+	q->cached_prod = cached_prod;
+}
+
 static inline int xskq_prod_reserve_desc(struct xsk_queue *q,
 					 u64 addr, u32 len)
 {
@@ -329,7 +372,7 @@ static inline int xskq_prod_reserve_desc(struct xsk_queue *q,
 	u32 idx;
 
 	if (xskq_prod_is_full(q))
-		return -ENOSPC;
+		return -ENOBUFS;
 
 	/* A, matches D */
 	idx = q->cached_prod++ & q->ring_mask;
@@ -341,9 +384,7 @@ static inline int xskq_prod_reserve_desc(struct xsk_queue *q,
 
 static inline void __xskq_prod_submit(struct xsk_queue *q, u32 idx)
 {
-	smp_wmb(); /* B, matches C */
-
-	WRITE_ONCE(q->ring->producer, idx);
+	smp_store_release(&q->ring->producer, idx); /* B, matches C */
 }
 
 static inline void xskq_prod_submit(struct xsk_queue *q)
@@ -379,11 +420,12 @@ static inline u64 xskq_nb_invalid_descs(struct xsk_queue *q)
 	return q ? q->invalid_descs : 0;
 }
 
-void xskq_set_umem(struct xsk_queue *q, u64 size, u64 chunk_mask);
+static inline u64 xskq_nb_queue_empty_descs(struct xsk_queue *q)
+{
+	return q ? q->queue_empty_descs : 0;
+}
+
 struct xsk_queue *xskq_create(u32 nentries, bool umem_queue);
 void xskq_destroy(struct xsk_queue *q_ops);
 
-/* Executed by the core when the entire UMEM gets freed */
-void xsk_reuseq_destroy(struct xdp_umem *umem);
-
 #endif /* _LINUX_XSK_QUEUE_H */