[PATCH net-next 3/5] netvsc: optimize receive completions

Stephen Hemminger stephen at networkplumber.org
Tue Jul 25 20:04:20 UTC 2017


Optimize how receive completion ring are managed.
   * Allocate only as many slots as needed for all buffers from host
   * Allocate before setting up sub channel for better error detection
   * Don't need to keep copy of initial receive section message
   * Only needt keep the transaction id, status doesn't matter
   * Precompute the watermark for when receive flushing is needed
   * Replace division with conditional test
   * Replace atomic per-device variable with per-channel check.
   * Handle corner case where receive completion send
     fails if ring buffer to host is full.

Signed-off-by: Stephen Hemminger <sthemmin at microsoft.com>
---
 drivers/net/hyperv/hyperv_net.h   |  19 +--
 drivers/net/hyperv/netvsc.c       | 270 +++++++++++++++-----------------------
 drivers/net/hyperv/rndis_filter.c |  20 +--
 3 files changed, 125 insertions(+), 184 deletions(-)

diff --git a/drivers/net/hyperv/hyperv_net.h b/drivers/net/hyperv/hyperv_net.h
index fb62ea632914..b0259b12c5ee 100644
--- a/drivers/net/hyperv/hyperv_net.h
+++ b/drivers/net/hyperv/hyperv_net.h
@@ -186,6 +186,7 @@ struct net_device_context;
 
 struct netvsc_device *netvsc_device_add(struct hv_device *device,
 					const struct netvsc_device_info *info);
+int netvsc_alloc_recv_comp_ring(struct netvsc_device *net_device, u32 q_idx);
 void netvsc_device_remove(struct hv_device *device);
 int netvsc_send(struct net_device_context *ndc,
 		struct hv_netvsc_packet *packet,
@@ -652,18 +653,10 @@ struct multi_send_data {
 	u32 count; /* counter of batched packets */
 };
 
-struct recv_comp_data {
-	u64 tid; /* transaction id */
-	u32 status;
-};
-
-/* Netvsc Receive Slots Max */
-#define NETVSC_RECVSLOT_MAX (NETVSC_RECEIVE_BUFFER_SIZE / ETH_DATA_LEN + 1)
-
 struct multi_recv_comp {
-	void *buf; /* queued receive completions */
-	u32 first; /* first data entry */
-	u32 next; /* next entry for writing */
+	u64 *slots;	/* pending completion transactions */
+	u32 first;	/* first data entry */
+	u32 next;	/* next entry for writing */
 };
 
 struct netvsc_stats {
@@ -750,7 +743,7 @@ struct netvsc_device {
 	u32 recv_buf_size;
 	u32 recv_buf_gpadl_handle;
 	u32 recv_section_cnt;
-	struct nvsp_1_receive_buffer_section *recv_section;
+	u32 recv_completion_cnt;
 
 	/* Send buffer allocated by us */
 	void *send_buf;
@@ -778,8 +771,6 @@ struct netvsc_device {
 	u32 max_pkt; /* max number of pkt in one send, e.g. 8 */
 	u32 pkt_align; /* alignment bytes, e.g. 8 */
 
-	atomic_t num_outstanding_recvs;
-
 	atomic_t open_cnt;
 
 	struct netvsc_channel chan_table[VRSS_CHANNEL_MAX];
diff --git a/drivers/net/hyperv/netvsc.c b/drivers/net/hyperv/netvsc.c
index 94c00acac58a..03c1fec762c7 100644
--- a/drivers/net/hyperv/netvsc.c
+++ b/drivers/net/hyperv/netvsc.c
@@ -72,9 +72,6 @@ static struct netvsc_device *alloc_net_device(void)
 	if (!net_device)
 		return NULL;
 
-	net_device->chan_table[0].mrc.buf
-		= vzalloc(NETVSC_RECVSLOT_MAX * sizeof(struct recv_comp_data));
-
 	init_waitqueue_head(&net_device->wait_drain);
 	net_device->destroy = false;
 	atomic_set(&net_device->open_cnt, 0);
@@ -92,7 +89,7 @@ static void free_netvsc_device(struct rcu_head *head)
 	int i;
 
 	for (i = 0; i < VRSS_CHANNEL_MAX; i++)
-		vfree(nvdev->chan_table[i].mrc.buf);
+		vfree(nvdev->chan_table[i].mrc.slots);
 
 	kfree(nvdev);
 }
@@ -171,12 +168,6 @@ static void netvsc_destroy_buf(struct hv_device *device)
 		net_device->recv_buf = NULL;
 	}
 
-	if (net_device->recv_section) {
-		net_device->recv_section_cnt = 0;
-		kfree(net_device->recv_section);
-		net_device->recv_section = NULL;
-	}
-
 	/* Deal with the send buffer we may have setup.
 	 * If we got a  send section size, it means we received a
 	 * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
@@ -239,11 +230,26 @@ static void netvsc_destroy_buf(struct hv_device *device)
 	kfree(net_device->send_section_map);
 }
 
+int netvsc_alloc_recv_comp_ring(struct netvsc_device *net_device, u32 q_idx)
+{
+	struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
+	int node = cpu_to_node(nvchan->channel->target_cpu);
+	size_t size;
+
+	size = net_device->recv_completion_cnt * sizeof(u64);
+	nvchan->mrc.slots = vzalloc_node(size, node);
+	if (!nvchan->mrc.slots)
+		nvchan->mrc.slots = vzalloc(size);
+
+	return nvchan->mrc.slots ? 0 : -ENOMEM;
+}
+
 static int netvsc_init_buf(struct hv_device *device,
 			   struct netvsc_device *net_device)
 {
 	int ret = 0;
 	struct nvsp_message *init_packet;
+	struct nvsp_1_message_send_receive_buffer_complete *resp;
 	struct net_device *ndev;
 	size_t map_words;
 	int node;
@@ -300,43 +306,41 @@ static int netvsc_init_buf(struct hv_device *device,
 	wait_for_completion(&net_device->channel_init_wait);
 
 	/* Check the response */
-	if (init_packet->msg.v1_msg.
-	    send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
-		netdev_err(ndev, "Unable to complete receive buffer "
-			   "initialization with NetVsp - status %d\n",
-			   init_packet->msg.v1_msg.
-			   send_recv_buf_complete.status);
+	resp = &init_packet->msg.v1_msg.send_recv_buf_complete;
+	if (resp->status != NVSP_STAT_SUCCESS) {
+		netdev_err(ndev,
+			   "Unable to complete receive buffer initialization with NetVsp - status %d\n",
+			   resp->status);
 		ret = -EINVAL;
 		goto cleanup;
 	}
 
 	/* Parse the response */
+	netdev_dbg(ndev, "Receive sections: %u sub_allocs: size %u count: %u\n",
+		   resp->num_sections, resp->sections[0].sub_alloc_size,
+		   resp->sections[0].num_sub_allocs);
 
-	net_device->recv_section_cnt = init_packet->msg.
-		v1_msg.send_recv_buf_complete.num_sections;
-
-	net_device->recv_section = kmemdup(
-		init_packet->msg.v1_msg.send_recv_buf_complete.sections,
-		net_device->recv_section_cnt *
-		sizeof(struct nvsp_1_receive_buffer_section),
-		GFP_KERNEL);
-	if (net_device->recv_section == NULL) {
-		ret = -EINVAL;
-		goto cleanup;
-	}
+	net_device->recv_section_cnt = resp->num_sections;
 
 	/*
 	 * For 1st release, there should only be 1 section that represents the
 	 * entire receive buffer
 	 */
 	if (net_device->recv_section_cnt != 1 ||
-	    net_device->recv_section->offset != 0) {
+	    resp->sections[0].offset != 0) {
 		ret = -EINVAL;
 		goto cleanup;
 	}
 
-	/* Now setup the send buffer.
-	 */
+	/* Setup receive completion ring */
+	net_device->recv_completion_cnt
+		= round_up(resp->sections[0].num_sub_allocs + 1,
+			   PAGE_SIZE / sizeof(u64));
+	ret = netvsc_alloc_recv_comp_ring(net_device, 0);
+	if (ret)
+		goto cleanup;
+
+	/* Now setup the send buffer. */
 	net_device->send_buf = vzalloc_node(net_device->send_buf_size, node);
 	if (!net_device->send_buf)
 		net_device->send_buf = vzalloc(net_device->send_buf_size);
@@ -950,139 +954,96 @@ int netvsc_send(struct net_device_context *ndev_ctx,
 	return ret;
 }
 
-static int netvsc_send_recv_completion(struct vmbus_channel *channel,
-				       u64 transaction_id, u32 status)
-{
-	struct nvsp_message recvcompMessage;
-	int ret;
-
-	recvcompMessage.hdr.msg_type =
-				NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
-
-	recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status = status;
-
-	/* Send the completion */
-	ret = vmbus_sendpacket(channel, &recvcompMessage,
-			       sizeof(struct nvsp_message_header) + sizeof(u32),
-			       transaction_id, VM_PKT_COMP, 0);
-
-	return ret;
-}
-
-static inline void count_recv_comp_slot(struct netvsc_device *nvdev, u16 q_idx,
-					u32 *filled, u32 *avail)
+/* Send pending recv completions */
+static int send_recv_completions(struct netvsc_device *nvdev,
+				 struct vmbus_channel *channel, u16 q_idx)
 {
 	struct multi_recv_comp *mrc = &nvdev->chan_table[q_idx].mrc;
-	u32 first = mrc->first;
-	u32 next = mrc->next;
+	struct recv_comp_msg {
+		struct nvsp_message_header hdr;
+		u32 status;
+	}  __packed;
+	struct recv_comp_msg msg = {
+		.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE,
+		.status = NVSP_STAT_SUCCESS,
+	};
+	int ret;
 
-	*filled = (first > next) ? NETVSC_RECVSLOT_MAX - first + next :
-		  next - first;
+	while (mrc->first != mrc->next) {
+		u64 tid = mrc->slots[mrc->first];
 
-	*avail = NETVSC_RECVSLOT_MAX - *filled - 1;
-}
+		ret = vmbus_sendpacket(channel, &msg, sizeof(msg),
+				       tid, VM_PKT_COMP, 0);
+		if (unlikely(ret))
+			return ret;
 
-/* Read the first filled slot, no change to index */
-static inline struct recv_comp_data *read_recv_comp_slot(struct netvsc_device
-							 *nvdev, u16 q_idx)
-{
-	struct multi_recv_comp *mrc = &nvdev->chan_table[q_idx].mrc;
-	u32 filled, avail;
-
-	if (unlikely(!mrc->buf))
-		return NULL;
+		if (++mrc->first == nvdev->recv_completion_cnt)
+			mrc->first = 0;
+	}
 
-	count_recv_comp_slot(nvdev, q_idx, &filled, &avail);
-	if (!filled)
-		return NULL;
+	/* receive completion ring has been emptied */
+	if (unlikely(nvdev->destroy))
+		wake_up(&nvdev->wait_drain);
 
-	return mrc->buf + mrc->first * sizeof(struct recv_comp_data);
+	return 0;
 }
 
-/* Put the first filled slot back to available pool */
-static inline void put_recv_comp_slot(struct netvsc_device *nvdev, u16 q_idx)
+/* Count how many receive completions are outstanding */
+static void recv_comp_slot_avail(const struct netvsc_device *nvdev,
+				 const struct multi_recv_comp *mrc,
+				 u32 *filled, u32 *avail)
 {
-	struct multi_recv_comp *mrc = &nvdev->chan_table[q_idx].mrc;
-	int num_recv;
+	u32 count = nvdev->recv_completion_cnt;
 
-	mrc->first = (mrc->first + 1) % NETVSC_RECVSLOT_MAX;
-
-	num_recv = atomic_dec_return(&nvdev->num_outstanding_recvs);
+	if (mrc->next >= mrc->first)
+		*filled = mrc->next - mrc->first;
+	else
+		*filled = (count - mrc->first) + mrc->next;
 
-	if (nvdev->destroy && num_recv == 0)
-		wake_up(&nvdev->wait_drain);
+	*avail = count - *filled - 1;
 }
 
-/* Check and send pending recv completions */
-static void netvsc_chk_recv_comp(struct netvsc_device *nvdev,
-				 struct vmbus_channel *channel, u16 q_idx)
+/* Add receive complete to ring to send to host. */
+static void enq_receive_complete(struct net_device *ndev,
+				 struct netvsc_device *nvdev, u16 q_idx,
+				 u64 tid)
 {
-	struct recv_comp_data *rcd;
-	int ret;
-
-	while (true) {
-		rcd = read_recv_comp_slot(nvdev, q_idx);
-		if (!rcd)
-			break;
+	struct netvsc_channel *nvchan = &nvdev->chan_table[q_idx];
+	struct multi_recv_comp *mrc = &nvchan->mrc;
+	u32 filled, avail;
 
-		ret = netvsc_send_recv_completion(channel, rcd->tid,
-						  rcd->status);
-		if (ret)
-			break;
+	recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
 
-		put_recv_comp_slot(nvdev, q_idx);
+	if (unlikely(filled > NAPI_POLL_WEIGHT)) {
+		send_recv_completions(nvdev, nvchan->channel, q_idx);
+		recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
 	}
-}
-
-#define NETVSC_RCD_WATERMARK 80
-
-/* Get next available slot */
-static inline struct recv_comp_data *get_recv_comp_slot(
-	struct netvsc_device *nvdev, struct vmbus_channel *channel, u16 q_idx)
-{
-	struct multi_recv_comp *mrc = &nvdev->chan_table[q_idx].mrc;
-	u32 filled, avail, next;
-	struct recv_comp_data *rcd;
-
-	if (unlikely(!nvdev->recv_section))
-		return NULL;
-
-	if (unlikely(!mrc->buf))
-		return NULL;
-
-	if (atomic_read(&nvdev->num_outstanding_recvs) >
-	    nvdev->recv_section->num_sub_allocs * NETVSC_RCD_WATERMARK / 100)
-		netvsc_chk_recv_comp(nvdev, channel, q_idx);
-
-	count_recv_comp_slot(nvdev, q_idx, &filled, &avail);
-	if (!avail)
-		return NULL;
-
-	next = mrc->next;
-	rcd = mrc->buf + next * sizeof(struct recv_comp_data);
-	mrc->next = (next + 1) % NETVSC_RECVSLOT_MAX;
 
-	atomic_inc(&nvdev->num_outstanding_recvs);
+	if (unlikely(!avail)) {
+		netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
+			   q_idx, tid);
+		return;
+	}
 
-	return rcd;
+	mrc->slots[mrc->next] = tid;
+	if (++mrc->next == nvdev->recv_completion_cnt)
+		mrc->next = 0;
 }
 
 static int netvsc_receive(struct net_device *ndev,
-		   struct netvsc_device *net_device,
-		   struct net_device_context *net_device_ctx,
-		   struct hv_device *device,
-		   struct vmbus_channel *channel,
-		   const struct vmpacket_descriptor *desc,
-		   struct nvsp_message *nvsp)
+			  struct netvsc_device *net_device,
+			  struct net_device_context *net_device_ctx,
+			  struct hv_device *device,
+			  struct vmbus_channel *channel,
+			  const struct vmpacket_descriptor *desc,
+			  struct nvsp_message *nvsp)
 {
 	const struct vmtransfer_page_packet_header *vmxferpage_packet
 		= container_of(desc, const struct vmtransfer_page_packet_header, d);
 	u16 q_idx = channel->offermsg.offer.sub_channel_index;
 	char *recv_buf = net_device->recv_buf;
-	u32 status = NVSP_STAT_SUCCESS;
 	int i;
 	int count = 0;
-	int ret;
 
 	/* Make sure this is a valid nvsp packet */
 	if (unlikely(nvsp->hdr.msg_type != NVSP_MSG1_TYPE_SEND_RNDIS_PKT)) {
@@ -1109,29 +1070,13 @@ static int netvsc_receive(struct net_device *ndev,
 		u32 buflen = vmxferpage_packet->ranges[i].byte_count;
 
 		/* Pass it to the upper layer */
-		status = rndis_filter_receive(ndev, net_device, device,
-					      channel, data, buflen);
+		rndis_filter_receive(ndev, net_device, device,
+				     channel, data, buflen);
 	}
 
-	if (net_device->chan_table[q_idx].mrc.buf) {
-		struct recv_comp_data *rcd;
+	enq_receive_complete(ndev, net_device, q_idx,
+			     vmxferpage_packet->d.trans_id);
 
-		rcd = get_recv_comp_slot(net_device, channel, q_idx);
-		if (rcd) {
-			rcd->tid = vmxferpage_packet->d.trans_id;
-			rcd->status = status;
-		} else {
-			netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
-				   q_idx, vmxferpage_packet->d.trans_id);
-		}
-	} else {
-		ret = netvsc_send_recv_completion(channel,
-						  vmxferpage_packet->d.trans_id,
-						  status);
-		if (ret)
-			netdev_err(ndev, "Recv_comp q:%hd, tid:%llx, err:%d\n",
-				   q_idx, vmxferpage_packet->d.trans_id, ret);
-	}
 	return count;
 }
 
@@ -1244,17 +1189,18 @@ int netvsc_poll(struct napi_struct *napi, int budget)
 		nvchan->desc = hv_pkt_iter_next(channel, nvchan->desc);
 	}
 
-	/* If receive ring was exhausted
-	 * and not doing busy poll
-	 * then re-enable host interrupts
-	 *  and reschedule if ring is not empty.
+	/* If send of  pending receive completions suceeded
+	 *   and did not exhaust NAPI budget
+	 *   and not doing busy poll
+	 * then reschedule if more data has arrived from host
 	 */
-	if (work_done < budget &&
+	if (send_recv_completions(net_device, channel, q_idx) == 0 &&
+	    work_done < budget &&
 	    napi_complete_done(napi, work_done) &&
-	    hv_end_read(&channel->inbound) != 0)
+	    hv_end_read(&channel->inbound)) {
+		hv_begin_read(&channel->inbound);
 		napi_reschedule(napi);
-
-	netvsc_chk_recv_comp(net_device, channel, q_idx);
+	}
 
 	/* Driver may overshoot since multiple packets per descriptor */
 	return min(work_done, budget);
diff --git a/drivers/net/hyperv/rndis_filter.c b/drivers/net/hyperv/rndis_filter.c
index bf21ea92c743..db9bac31dd09 100644
--- a/drivers/net/hyperv/rndis_filter.c
+++ b/drivers/net/hyperv/rndis_filter.c
@@ -928,12 +928,12 @@ static bool netvsc_device_idle(const struct netvsc_device *nvdev)
 {
 	int i;
 
-	if (atomic_read(&nvdev->num_outstanding_recvs) > 0)
-		return false;
-
 	for (i = 0; i < nvdev->num_chn; i++) {
 		const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
 
+		if (nvchan->mrc.first != nvchan->mrc.next)
+			return false;
+
 		if (atomic_read(&nvchan->queue_sends) > 0)
 			return false;
 	}
@@ -1031,11 +1031,6 @@ static void netvsc_sc_open(struct vmbus_channel *new_sc)
 		return;
 
 	nvchan = nvscdev->chan_table + chn_index;
-	nvchan->mrc.buf
-		= vzalloc(NETVSC_RECVSLOT_MAX * sizeof(struct recv_comp_data));
-
-	if (!nvchan->mrc.buf)
-		return;
 
 	/* Because the device uses NAPI, all the interrupt batching and
 	 * control is done via Net softirq, not the channel handling
@@ -1225,6 +1220,15 @@ struct netvsc_device *rndis_filter_device_add(struct hv_device *dev,
 	if (num_rss_qs == 0)
 		return net_device;
 
+	for (i = 1; i < net_device->num_chn; i++) {
+		ret = netvsc_alloc_recv_comp_ring(net_device, i);
+		if (ret) {
+			while (--i != 0)
+				vfree(net_device->chan_table[i].mrc.slots);
+			goto out;
+		}
+	}
+
 	refcount_set(&net_device->sc_offered, num_rss_qs);
 	vmbus_set_sc_create_callback(dev->channel, netvsc_sc_open);
 
-- 
2.11.0



More information about the devel mailing list