[07/20] io_uring: add ZC pool API

Commit Message

David Wei Nov. 7, 2023, 9:40 p.m. UTC

This patch adds an API to get/put bufs from a ZC pool added in the
previous patch.

Recall that there is an rbuf refill ring in an ifq that is shared w/
userspace, which puts bufs it is done with back into it. A new tier is
added to the ZC pool that drains entries from the refill ring to put
into the cache. So when the cache is empty, it is refilled from the
refill ring first, then the freelist.

ZC bufs are refcounted. Userspace is given an off + len into the entire
ZC pool region, not individual pages from ZC bufs. A net device may pack
multiple packets into the same page it gets from a ZC buf, so it is
possible for the same ZC buf to be handed out to userspace multiple
times.

This means it is possible to drain the entire refill ring, and have no
usable free bufs. Suggestions for dealing w/ this are very welcome!

Only up to POOL_REFILL_COUNT entries are refilled from the refill ring.
Given the above, we may want to limit the amount of work being done
since refilling happens inside the NAPI softirq context.

Co-developed-by: Pavel Begunkov <[e-mail]>
Signed-off-by: Pavel Begunkov <[e-mail]>
Signed-off-by: David Wei <[e-mail]>
---
 include/linux/io_uring.h | 19 ++++++++
 io_uring/zc_rx.c         | 95 ++++++++++++++++++++++++++++++++++++++++
 io_uring/zc_rx.h         | 11 +++++
 3 files changed, 125 insertions(+)

Patch

diff --git a/include/linux/io_uring.h b/include/linux/io_uring.h
index abfb73e257a4..624515a8bdd5 100644
--- a/include/linux/io_uring.h
+++ b/include/linux/io_uring.h
@@ -70,6 +70,18 @@  static inline void io_uring_cmd_complete_in_task(struct io_uring_cmd *ioucmd,
 	__io_uring_cmd_do_in_task(ioucmd, task_work_cb, 0);
 }
 
+struct io_zc_rx_ifq;
+struct io_zc_rx_buf *io_zc_rx_get_buf(struct io_zc_rx_ifq *ifq);
+void io_zc_rx_put_buf(struct io_zc_rx_ifq *ifq, struct io_zc_rx_buf *buf);
+
+static inline dma_addr_t io_zc_rx_buf_dma(struct io_zc_rx_buf *buf)
+{
+	return buf->dma;
+}
+static inline struct page *io_zc_rx_buf_page(struct io_zc_rx_buf *buf)
+{
+	return buf->page;
+}
 static inline void io_uring_files_cancel(void)
 {
 	if (current->io_uring) {
@@ -106,6 +118,13 @@  static inline void io_uring_cmd_do_in_task_lazy(struct io_uring_cmd *ioucmd,
 			void (*task_work_cb)(struct io_uring_cmd *, unsigned))
 {
 }
+static inline struct io_zc_rx_buf *io_zc_rx_get_buf(struct io_zc_rx_ifq *ifq)
+{
+	return NULL;
+}
+static inline void io_zc_rx_put_buf(struct io_zc_rx_ifq *ifq, struct io_zc_rx_buf *buf)
+{
+}
 static inline struct sock *io_uring_get_socket(struct file *file)
 {
 	return NULL;
diff --git a/io_uring/zc_rx.c b/io_uring/zc_rx.c
index 0f5fa9ab5cec..840a21549d89 100644
--- a/io_uring/zc_rx.c
+++ b/io_uring/zc_rx.c
@@ -16,6 +16,9 @@ 
 #include "rsrc.h"
 
 #define POOL_CACHE_SIZE	128
+#define POOL_REFILL_COUNT	64
+#define IO_ZC_RX_UREF		0x10000
+#define IO_ZC_RX_KREF_MASK	(IO_ZC_RX_UREF - 1)
 
 struct io_zc_rx_pool {
 	struct io_zc_rx_ifq  	*ifq;
@@ -269,6 +272,8 @@  int io_register_zc_rx_ifq(struct io_ring_ctx *ctx,
 
 	ifq->rq_entries = reg.rq_entries;
 	ifq->cq_entries = reg.cq_entries;
+	ifq->cached_rq_head = 0;
+	ifq->cached_cq_tail = 0;
 	ifq->if_rxq_id = reg.if_rxq_id;
 	ctx->ifq = ifq;
 
@@ -371,4 +376,94 @@  int io_register_zc_rx_sock(struct io_ring_ctx *ctx,
 	ifq->nr_sockets++;
 	return 0;
 }
+
+static bool io_zc_rx_put_buf_uref(struct io_zc_rx_buf *buf)
+{
+	if (atomic_read(&buf->refcount) < IO_ZC_RX_UREF)
+		return false;
+
+	return atomic_sub_and_test(IO_ZC_RX_UREF, &buf->refcount);
+}
+
+static void io_zc_rx_refill_cache(struct io_zc_rx_ifq *ifq, int count)
+{
+	unsigned int entries = io_zc_rx_rqring_entries(ifq);
+	unsigned int mask = ifq->rq_entries - 1;
+	struct io_zc_rx_pool *pool = ifq->pool;
+	struct io_uring_rbuf_rqe *rqe;
+	struct io_zc_rx_buf *buf;
+	int i, filled;
+
+	if (!entries)
+		return;
+
+	for (i = 0, filled = 0; i < entries && filled < count; i++) {
+		unsigned int rq_idx = ifq->cached_rq_head++ & mask;
+		u32 pgid;
+
+		rqe = &ifq->rqes[rq_idx];
+		pgid = rqe->off / PAGE_SIZE;
+		buf = &pool->bufs[pgid];
+		if (!io_zc_rx_put_buf_uref(buf))
+			continue;
+		pool->cache[filled++] = pgid;
+	}
+
+	smp_store_release(&ifq->ring->rq.head, ifq->cached_rq_head);
+	pool->cache_count += filled;
+}
+
+struct io_zc_rx_buf *io_zc_rx_get_buf(struct io_zc_rx_ifq *ifq)
+{
+	struct io_zc_rx_pool *pool = ifq->pool;
+	struct io_zc_rx_buf *buf;
+	int count;
+	u32 pgid;
+
+	lockdep_assert_no_hardirq();
+
+	if (likely(pool->cache_count))
+		goto out;
+
+	io_zc_rx_refill_cache(ifq, POOL_REFILL_COUNT);
+	if (pool->cache_count)
+		goto out;
+
+	spin_lock_bh(&pool->freelist_lock);
+	count = min_t(u32, pool->free_count, POOL_CACHE_SIZE);
+	pool->free_count -= count;
+	pool->cache_count += count;
+	memcpy(pool->cache, &pool->freelist[pool->free_count],
+	       count * sizeof(u32));
+	spin_unlock_bh(&pool->freelist_lock);
+
+	if (!pool->cache_count)
+		return NULL;
+out:
+	pgid = pool->cache[--pool->cache_count];
+	buf = &pool->bufs[pgid];
+	atomic_set(&buf->refcount, 1);
+	return buf;
+}
+EXPORT_SYMBOL(io_zc_rx_get_buf);
+
+static void io_zc_rx_recycle_buf(struct io_zc_rx_pool *pool,
+				 struct io_zc_rx_buf *buf)
+{
+	spin_lock_bh(&pool->freelist_lock);
+	pool->freelist[pool->free_count++] = buf - pool->bufs;
+	spin_unlock_bh(&pool->freelist_lock);
+}
+
+void io_zc_rx_put_buf(struct io_zc_rx_ifq *ifq, struct io_zc_rx_buf *buf)
+{
+	struct io_zc_rx_pool *pool = ifq->pool;
+
+	if (!atomic_dec_and_test(&buf->refcount))
+		return;
+
+	io_zc_rx_recycle_buf(pool, buf);
+}
+EXPORT_SYMBOL(io_zc_rx_put_buf);
+
 #endif
diff --git a/io_uring/zc_rx.h b/io_uring/zc_rx.h
index ab25f8dbb433..a3df820e52e7 100644
--- a/io_uring/zc_rx.h
+++ b/io_uring/zc_rx.h
@@ -16,6 +16,8 @@  struct io_zc_rx_ifq {
 	struct io_uring_rbuf_rqe *rqes;
 	struct io_uring_rbuf_cqe *cqes;
 	u32			rq_entries, cq_entries;
+	u32			cached_rq_head;
+	u32			cached_cq_tail;
 	void			*pool;
 
 	unsigned		nr_sockets;
@@ -26,6 +28,15 @@  struct io_zc_rx_ifq {
 };
 
 #if defined(CONFIG_NET)
+static inline u32 io_zc_rx_rqring_entries(struct io_zc_rx_ifq *ifq)
+{
+	struct io_rbuf_ring *ring = ifq->ring;
+	u32 entries;
+
+	entries = smp_load_acquire(&ring->rq.tail) - ifq->cached_rq_head;
+	return min(entries, ifq->rq_entries);
+}
+
 int io_register_zc_rx_ifq(struct io_ring_ctx *ctx,
 			  struct io_uring_zc_rx_ifq_reg __user *arg);
 int io_unregister_zc_rx_ifq(struct io_ring_ctx *ctx);