hyperv: Add support for virtual Receive Side Scaling (vRSS)

This feature allows multiple channels to be used by each virtual NIC.
It is available on Hyper-V host 2012 R2.
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/hyperv/hyperv_net.h

@@ -28,6 +28,96 @@
 #include <linux/hyperv.h>
 #include <linux/rndis.h>
 
+/* RSS related */
+#define OID_GEN_RECEIVE_SCALE_CAPABILITIES 0x00010203  /* query only */
+#define OID_GEN_RECEIVE_SCALE_PARAMETERS 0x00010204  /* query and set */
+
+#define NDIS_OBJECT_TYPE_RSS_CAPABILITIES 0x88
+#define NDIS_OBJECT_TYPE_RSS_PARAMETERS 0x89
+
+#define NDIS_RECEIVE_SCALE_CAPABILITIES_REVISION_2 2
+#define NDIS_RECEIVE_SCALE_PARAMETERS_REVISION_2 2
+
+struct ndis_obj_header {
+	u8 type;
+	u8 rev;
+	u16 size;
+} __packed;
+
+/* ndis_recv_scale_cap/cap_flag */
+#define NDIS_RSS_CAPS_MESSAGE_SIGNALED_INTERRUPTS 0x01000000
+#define NDIS_RSS_CAPS_CLASSIFICATION_AT_ISR       0x02000000
+#define NDIS_RSS_CAPS_CLASSIFICATION_AT_DPC       0x04000000
+#define NDIS_RSS_CAPS_USING_MSI_X                 0x08000000
+#define NDIS_RSS_CAPS_RSS_AVAILABLE_ON_PORTS      0x10000000
+#define NDIS_RSS_CAPS_SUPPORTS_MSI_X              0x20000000
+#define NDIS_RSS_CAPS_HASH_TYPE_TCP_IPV4          0x00000100
+#define NDIS_RSS_CAPS_HASH_TYPE_TCP_IPV6          0x00000200
+#define NDIS_RSS_CAPS_HASH_TYPE_TCP_IPV6_EX       0x00000400
+
+struct ndis_recv_scale_cap { /* NDIS_RECEIVE_SCALE_CAPABILITIES */
+	struct ndis_obj_header hdr;
+	u32 cap_flag;
+	u32 num_int_msg;
+	u32 num_recv_que;
+	u16 num_indirect_tabent;
+} __packed;
+
+
+/* ndis_recv_scale_param flags */
+#define NDIS_RSS_PARAM_FLAG_BASE_CPU_UNCHANGED     0x0001
+#define NDIS_RSS_PARAM_FLAG_HASH_INFO_UNCHANGED    0x0002
+#define NDIS_RSS_PARAM_FLAG_ITABLE_UNCHANGED       0x0004
+#define NDIS_RSS_PARAM_FLAG_HASH_KEY_UNCHANGED     0x0008
+#define NDIS_RSS_PARAM_FLAG_DISABLE_RSS            0x0010
+
+/* Hash info bits */
+#define NDIS_HASH_FUNC_TOEPLITZ 0x00000001
+#define NDIS_HASH_IPV4          0x00000100
+#define NDIS_HASH_TCP_IPV4      0x00000200
+#define NDIS_HASH_IPV6          0x00000400
+#define NDIS_HASH_IPV6_EX       0x00000800
+#define NDIS_HASH_TCP_IPV6      0x00001000
+#define NDIS_HASH_TCP_IPV6_EX   0x00002000
+
+#define NDIS_RSS_INDIRECTION_TABLE_MAX_SIZE_REVISION_2 (128 * 4)
+#define NDIS_RSS_HASH_SECRET_KEY_MAX_SIZE_REVISION_2   40
+
+#define ITAB_NUM 128
+#define HASH_KEYLEN NDIS_RSS_HASH_SECRET_KEY_MAX_SIZE_REVISION_2
+extern u8 netvsc_hash_key[];
+
+struct ndis_recv_scale_param { /* NDIS_RECEIVE_SCALE_PARAMETERS */
+	struct ndis_obj_header hdr;
+
+	/* Qualifies the rest of the information */
+	u16 flag;
+
+	/* The base CPU number to do receive processing. not used */
+	u16 base_cpu_number;
+
+	/* This describes the hash function and type being enabled */
+	u32 hashinfo;
+
+	/* The size of indirection table array */
+	u16 indirect_tabsize;
+
+	/* The offset of the indirection table from the beginning of this
+	 * structure
+	 */
+	u32 indirect_taboffset;
+
+	/* The size of the hash secret key */
+	u16 hashkey_size;
+
+	/* The offset of the secret key from the beginning of this structure */
+	u32 kashkey_offset;
+
+	u32 processor_masks_offset;
+	u32 num_processor_masks;
+	u32 processor_masks_entry_size;
+};
+
 /* Fwd declaration */
 struct hv_netvsc_packet;
 struct ndis_tcp_ip_checksum_info;
@@ -39,6 +129,8 @@ struct xferpage_packet {
 
 	/* # of netvsc packets this xfer packet contains */
 	u32 count;
+
+	struct vmbus_channel *channel;
 };
 
 /*
@@ -54,6 +146,9 @@ struct hv_netvsc_packet {
 	bool is_data_pkt;
 	u16 vlan_tci;
 
+	u16 q_idx;
+	struct vmbus_channel *channel;
+
 	/*
 	 * Valid only for receives when we break a xfer page packet
 	 * into multiple netvsc packets
@@ -120,6 +215,7 @@ void netvsc_linkstatus_callback(struct hv_device *device_obj,
 int netvsc_recv_callback(struct hv_device *device_obj,
 			struct hv_netvsc_packet *packet,
 			struct ndis_tcp_ip_checksum_info *csum_info);
+void netvsc_channel_cb(void *context);
 int rndis_filter_open(struct hv_device *dev);
 int rndis_filter_close(struct hv_device *dev);
 int rndis_filter_device_add(struct hv_device *dev,
@@ -522,6 +618,8 @@ struct nvsp_message {
 
 #define NETVSC_PACKET_SIZE                      2048
 
+#define VRSS_SEND_TAB_SIZE 16
+
 /* Per netvsc channel-specific */
 struct netvsc_device {
 	struct hv_device *dev;
@@ -555,10 +653,20 @@ struct netvsc_device {
 
 	struct net_device *ndev;
 
+	struct vmbus_channel *chn_table[NR_CPUS];
+	u32 send_table[VRSS_SEND_TAB_SIZE];
+	u32 num_chn;
+	atomic_t queue_sends[NR_CPUS];
+
 	/* Holds rndis device info */
 	void *extension;
-	/* The recive buffer for this device */
+
+	int ring_size;
+
+	/* The primary channel callback buffer */
 	unsigned char cb_buffer[NETVSC_PACKET_SIZE];
+	/* The sub channel callback buffer */
+	unsigned char *sub_cb_buf;
 };
 
 /* NdisInitialize message */

drivers/net/hyperv/netvsc.c

@@ -422,6 +422,9 @@ int netvsc_device_remove(struct hv_device *device)
 		kfree(netvsc_packet);
 	}
 
+	if (net_device->sub_cb_buf)
+		vfree(net_device->sub_cb_buf);
+
 	kfree(net_device);
 	return 0;
 }
@@ -461,7 +464,9 @@ static void netvsc_send_completion(struct netvsc_device *net_device,
 	    (nvsp_packet->hdr.msg_type ==
 	     NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) ||
 	    (nvsp_packet->hdr.msg_type ==
-	     NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE)) {
+	     NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE) ||
+	    (nvsp_packet->hdr.msg_type ==
+	     NVSP_MSG5_TYPE_SUBCHANNEL)) {
 		/* Copy the response back */
 		memcpy(&net_device->channel_init_pkt, nvsp_packet,
 		       sizeof(struct nvsp_message));
@@ -469,28 +474,37 @@ static void netvsc_send_completion(struct netvsc_device *net_device,
 	} else if (nvsp_packet->hdr.msg_type ==
 		   NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) {
 		int num_outstanding_sends;
+		u16 q_idx = 0;
+		struct vmbus_channel *channel = device->channel;
+		int queue_sends;
 
 		/* Get the send context */
 		nvsc_packet = (struct hv_netvsc_packet *)(unsigned long)
 			packet->trans_id;
 
 		/* Notify the layer above us */
-		if (nvsc_packet)
+		if (nvsc_packet) {
+			q_idx = nvsc_packet->q_idx;
+			channel = nvsc_packet->channel;
 			nvsc_packet->completion.send.send_completion(
 				nvsc_packet->completion.send.
 				send_completion_ctx);
+		}
 
 		num_outstanding_sends =
 			atomic_dec_return(&net_device->num_outstanding_sends);
+		queue_sends = atomic_dec_return(&net_device->
+						queue_sends[q_idx]);
 
 		if (net_device->destroy && num_outstanding_sends == 0)
 			wake_up(&net_device->wait_drain);
 
-		if (netif_queue_stopped(ndev) && !net_device->start_remove &&
-			(hv_ringbuf_avail_percent(&device->channel->outbound)
-			> RING_AVAIL_PERCENT_HIWATER ||
-			num_outstanding_sends < 1))
-				netif_wake_queue(ndev);
+		if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev, q_idx)) &&
+		    !net_device->start_remove &&
+		    (hv_ringbuf_avail_percent(&channel->outbound) >
+		     RING_AVAIL_PERCENT_HIWATER || queue_sends < 1))
+				netif_tx_wake_queue(netdev_get_tx_queue(
+						    ndev, q_idx));
 	} else {
 		netdev_err(ndev, "Unknown send completion packet type- "
 			   "%d received!!\n", nvsp_packet->hdr.msg_type);
@@ -505,6 +519,7 @@ int netvsc_send(struct hv_device *device,
 	int ret = 0;
 	struct nvsp_message sendMessage;
 	struct net_device *ndev;
+	struct vmbus_channel *out_channel = NULL;
 	u64 req_id;
 
 	net_device = get_outbound_net_device(device);
@@ -531,15 +546,20 @@ int netvsc_send(struct hv_device *device,
 	else
 		req_id = 0;
 
+	out_channel = net_device->chn_table[packet->q_idx];
+	if (out_channel == NULL)
+		out_channel = device->channel;
+	packet->channel = out_channel;
+
 	if (packet->page_buf_cnt) {
-		ret = vmbus_sendpacket_pagebuffer(device->channel,
+		ret = vmbus_sendpacket_pagebuffer(out_channel,
 						  packet->page_buf,
 						  packet->page_buf_cnt,
 						  &sendMessage,
 						  sizeof(struct nvsp_message),
 						  req_id);
 	} else {
-		ret = vmbus_sendpacket(device->channel, &sendMessage,
+		ret = vmbus_sendpacket(out_channel, &sendMessage,
 				sizeof(struct nvsp_message),
 				req_id,
 				VM_PKT_DATA_INBAND,
@@ -548,17 +568,24 @@ int netvsc_send(struct hv_device *device,
 
 	if (ret == 0) {
 		atomic_inc(&net_device->num_outstanding_sends);
-		if (hv_ringbuf_avail_percent(&device->channel->outbound) <
+		atomic_inc(&net_device->queue_sends[packet->q_idx]);
+
+		if (hv_ringbuf_avail_percent(&out_channel->outbound) <
 			RING_AVAIL_PERCENT_LOWATER) {
-			netif_stop_queue(ndev);
+			netif_tx_stop_queue(netdev_get_tx_queue(
+					    ndev, packet->q_idx));
+
 			if (atomic_read(&net_device->
-				num_outstanding_sends) < 1)
-				netif_wake_queue(ndev);
+				queue_sends[packet->q_idx]) < 1)
+				netif_tx_wake_queue(netdev_get_tx_queue(
+						    ndev, packet->q_idx));
 		}
 	} else if (ret == -EAGAIN) {
-		netif_stop_queue(ndev);
-		if (atomic_read(&net_device->num_outstanding_sends) < 1) {
-			netif_wake_queue(ndev);
+		netif_tx_stop_queue(netdev_get_tx_queue(
+				    ndev, packet->q_idx));
+		if (atomic_read(&net_device->queue_sends[packet->q_idx]) < 1) {
+			netif_tx_wake_queue(netdev_get_tx_queue(
+					    ndev, packet->q_idx));
 			ret = -ENOSPC;
 		}
 	} else {
@@ -570,6 +597,7 @@ int netvsc_send(struct hv_device *device,
 }
 
 static void netvsc_send_recv_completion(struct hv_device *device,
+					struct vmbus_channel *channel,
 					struct netvsc_device *net_device,
 					u64 transaction_id, u32 status)
 {
@@ -587,7 +615,7 @@ static void netvsc_send_recv_completion(struct hv_device *device,
 
 retry_send_cmplt:
 	/* Send the completion */
-	ret = vmbus_sendpacket(device->channel, &recvcompMessage,
+	ret = vmbus_sendpacket(channel, &recvcompMessage,
 			       sizeof(struct nvsp_message), transaction_id,
 			       VM_PKT_COMP, 0);
 	if (ret == 0) {
@@ -618,6 +646,7 @@ static void netvsc_receive_completion(void *context)
 {
 	struct hv_netvsc_packet *packet = context;
 	struct hv_device *device = packet->device;
+	struct vmbus_channel *channel;
 	struct netvsc_device *net_device;
 	u64 transaction_id = 0;
 	bool fsend_receive_comp = false;
@@ -649,6 +678,7 @@ static void netvsc_receive_completion(void *context)
 	 */
 	if (packet->xfer_page_pkt->count == 0) {
 		fsend_receive_comp = true;
+		channel = packet->xfer_page_pkt->channel;
 		transaction_id = packet->completion.recv.recv_completion_tid;
 		status = packet->xfer_page_pkt->status;
 		list_add_tail(&packet->xfer_page_pkt->list_ent,
@@ -662,12 +692,13 @@ static void netvsc_receive_completion(void *context)
 
 	/* Send a receive completion for the xfer page packet */
 	if (fsend_receive_comp)
-		netvsc_send_recv_completion(device, net_device, transaction_id,
-					status);
+		netvsc_send_recv_completion(device, channel, net_device,
+					    transaction_id, status);
 
 }
 
 static void netvsc_receive(struct netvsc_device *net_device,
+			struct vmbus_channel *channel,
 			struct hv_device *device,
 			struct vmpacket_descriptor *packet)
 {
@@ -748,7 +779,7 @@ static void netvsc_receive(struct netvsc_device *net_device,
 		spin_unlock_irqrestore(&net_device->recv_pkt_list_lock,
 				       flags);
 
-		netvsc_send_recv_completion(device, net_device,
+		netvsc_send_recv_completion(device, channel, net_device,
 					    vmxferpage_packet->d.trans_id,
 					    NVSP_STAT_FAIL);
 
@@ -759,6 +790,7 @@ static void netvsc_receive(struct netvsc_device *net_device,
 	xferpage_packet = (struct xferpage_packet *)listHead.next;
 	list_del(&xferpage_packet->list_ent);
 	xferpage_packet->status = NVSP_STAT_SUCCESS;
+	xferpage_packet->channel = channel;
 
 	/* This is how much we can satisfy */
 	xferpage_packet->count = count - 1;
@@ -800,10 +832,45 @@ static void netvsc_receive(struct netvsc_device *net_device,
 
 }
 
-static void netvsc_channel_cb(void *context)
+
+static void netvsc_send_table(struct hv_device *hdev,
+			      struct vmpacket_descriptor *vmpkt)
+{
+	struct netvsc_device *nvscdev;
+	struct net_device *ndev;
+	struct nvsp_message *nvmsg;
+	int i;
+	u32 count, *tab;
+
+	nvscdev = get_outbound_net_device(hdev);
+	if (!nvscdev)
+		return;
+	ndev = nvscdev->ndev;
+
+	nvmsg = (struct nvsp_message *)((unsigned long)vmpkt +
+					(vmpkt->offset8 << 3));
+
+	if (nvmsg->hdr.msg_type != NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE)
+		return;
+
+	count = nvmsg->msg.v5_msg.send_table.count;
+	if (count != VRSS_SEND_TAB_SIZE) {
+		netdev_err(ndev, "Received wrong send-table size:%u\n", count);
+		return;
+	}
+
+	tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
+		      nvmsg->msg.v5_msg.send_table.offset);
+
+	for (i = 0; i < count; i++)
+		nvscdev->send_table[i] = tab[i];
+}
+
+void netvsc_channel_cb(void *context)
 {
 	int ret;
-	struct hv_device *device = context;
+	struct vmbus_channel *channel = (struct vmbus_channel *)context;
+	struct hv_device *device;
 	struct netvsc_device *net_device;
 	u32 bytes_recvd;
 	u64 request_id;
@@ -812,14 +879,19 @@ static void netvsc_channel_cb(void *context)
 	int bufferlen = NETVSC_PACKET_SIZE;
 	struct net_device *ndev;
 
+	if (channel->primary_channel != NULL)
+		device = channel->primary_channel->device_obj;
+	else
+		device = channel->device_obj;
+
 	net_device = get_inbound_net_device(device);
 	if (!net_device)
 		return;
 	ndev = net_device->ndev;
-	buffer = net_device->cb_buffer;
+	buffer = get_per_channel_state(channel);
 
 	do {
-		ret = vmbus_recvpacket_raw(device->channel, buffer, bufferlen,
+		ret = vmbus_recvpacket_raw(channel, buffer, bufferlen,
 					   &bytes_recvd, &request_id);
 		if (ret == 0) {
 			if (bytes_recvd > 0) {
@@ -831,8 +903,12 @@ static void netvsc_channel_cb(void *context)
 					break;
 
 				case VM_PKT_DATA_USING_XFER_PAGES:
-					netvsc_receive(net_device,
-							device, desc);
+					netvsc_receive(net_device, channel,
+						       device, desc);
+					break;
+
+				case VM_PKT_DATA_INBAND:
+					netvsc_send_table(device, desc);
 					break;
 
 				default:
@@ -893,6 +969,8 @@ int netvsc_device_add(struct hv_device *device, void *additional_info)
 		goto cleanup;
 	}
 
+	net_device->ring_size = ring_size;
+
 	/*
 	 * Coming into this function, struct net_device * is
 	 * registered as the driver private data.
@@ -917,10 +995,12 @@ int netvsc_device_add(struct hv_device *device, void *additional_info)
 	}
 	init_completion(&net_device->channel_init_wait);
 
+	set_per_channel_state(device->channel, net_device->cb_buffer);
+
 	/* Open the channel */
 	ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
 			 ring_size * PAGE_SIZE, NULL, 0,
-			 netvsc_channel_cb, device);
+			 netvsc_channel_cb, device->channel);
 
 	if (ret != 0) {
 		netdev_err(ndev, "unable to open channel: %d\n", ret);
@@ -930,6 +1010,8 @@ int netvsc_device_add(struct hv_device *device, void *additional_info)
 	/* Channel is opened */
 	pr_info("hv_netvsc channel opened successfully\n");
 
+	net_device->chn_table[0] = device->channel;
+
 	/* Connect with the NetVsp */
 	ret = netvsc_connect_vsp(device);
 	if (ret != 0) {

drivers/net/hyperv/netvsc_drv.c

@@ -101,7 +101,7 @@ static int netvsc_open(struct net_device *net)
 		return ret;
 	}
 
-	netif_start_queue(net);
+	netif_tx_start_all_queues(net);
 
 	nvdev = hv_get_drvdata(device_obj);
 	rdev = nvdev->extension;
@@ -149,6 +149,88 @@ static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
 	return ppi;
 }
 
+union sub_key {
+	u64 k;
+	struct {
+		u8 pad[3];
+		u8 kb;
+		u32 ka;
+	};
+};
+
+/* Toeplitz hash function
+ * data: network byte order
+ * return: host byte order
+ */
+static u32 comp_hash(u8 *key, int klen, u8 *data, int dlen)
+{
+	union sub_key subk;
+	int k_next = 4;
+	u8 dt;
+	int i, j;
+	u32 ret = 0;
+
+	subk.k = 0;
+	subk.ka = ntohl(*(u32 *)key);
+
+	for (i = 0; i < dlen; i++) {
+		subk.kb = key[k_next];
+		k_next = (k_next + 1) % klen;
+		dt = data[i];
+		for (j = 0; j < 8; j++) {
+			if (dt & 0x80)
+				ret ^= subk.ka;
+			dt <<= 1;
+			subk.k <<= 1;
+		}
+	}
+
+	return ret;
+}
+
+static bool netvsc_set_hash(u32 *hash, struct sk_buff *skb)
+{
+	struct iphdr *iphdr;
+	int data_len;
+	bool ret = false;
+
+	if (eth_hdr(skb)->h_proto != htons(ETH_P_IP))
+		return false;
+
+	iphdr = ip_hdr(skb);
+
+	if (iphdr->version == 4) {
+		if (iphdr->protocol == IPPROTO_TCP)
+			data_len = 12;
+		else
+			data_len = 8;
+		*hash = comp_hash(netvsc_hash_key, HASH_KEYLEN,
+				  (u8 *)&iphdr->saddr, data_len);
+		ret = true;
+	}
+
+	return ret;
+}
+
+static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
+			void *accel_priv, select_queue_fallback_t fallback)
+{
+	struct net_device_context *net_device_ctx = netdev_priv(ndev);
+	struct hv_device *hdev =  net_device_ctx->device_ctx;
+	struct netvsc_device *nvsc_dev = hv_get_drvdata(hdev);
+	u32 hash;
+	u16 q_idx = 0;
+
+	if (nvsc_dev == NULL || ndev->real_num_tx_queues <= 1)
+		return 0;
+
+	if (netvsc_set_hash(&hash, skb))
+		q_idx = nvsc_dev->send_table[hash % VRSS_SEND_TAB_SIZE] %
+			ndev->real_num_tx_queues;
+
+	return q_idx;
+}
+
 static void netvsc_xmit_completion(void *context)
 {
 	struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
@@ -333,6 +415,8 @@ static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 
 	packet->vlan_tci = skb->vlan_tci;
 
+	packet->q_idx = skb_get_queue_mapping(skb);
+
 	packet->is_data_pkt = true;
 	packet->total_data_buflen = skb->len;
 
@@ -554,6 +638,10 @@ int netvsc_recv_callback(struct hv_device *device_obj,
 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 				       packet->vlan_tci);
 
+	skb_record_rx_queue(skb, packet->xfer_page_pkt->channel->
+			    offermsg.offer.sub_channel_index %
+			    net->real_num_rx_queues);
+
 	net->stats.rx_packets++;
 	net->stats.rx_bytes += packet->total_data_buflen;
 
@@ -602,7 +690,7 @@ static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 	hv_set_drvdata(hdev, ndev);
 	device_info.ring_size = ring_size;
 	rndis_filter_device_add(hdev, &device_info);
-	netif_wake_queue(ndev);
+	netif_tx_wake_all_queues(ndev);
 
 	return 0;
 }
@@ -648,6 +736,7 @@ static const struct net_device_ops device_ops = {
 	.ndo_change_mtu =		netvsc_change_mtu,
 	.ndo_validate_addr =		eth_validate_addr,
 	.ndo_set_mac_address =		netvsc_set_mac_addr,
+	.ndo_select_queue =		netvsc_select_queue,
 };
 
 /*
@@ -694,9 +783,11 @@ static int netvsc_probe(struct hv_device *dev,
 	struct net_device *net = NULL;
 	struct net_device_context *net_device_ctx;
 	struct netvsc_device_info device_info;
+	struct netvsc_device *nvdev;
 	int ret;
 
-	net = alloc_etherdev(sizeof(struct net_device_context));
+	net = alloc_etherdev_mq(sizeof(struct net_device_context),
+				num_online_cpus());
 	if (!net)
 		return -ENOMEM;
 
@@ -729,6 +820,12 @@ static int netvsc_probe(struct hv_device *dev,
 	}
 	memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
 
+	nvdev = hv_get_drvdata(dev);
+	netif_set_real_num_tx_queues(net, nvdev->num_chn);
+	netif_set_real_num_rx_queues(net, nvdev->num_chn);
+	dev_info(&dev->device, "real num tx,rx queues:%u, %u\n",
+		 net->real_num_tx_queues, net->real_num_rx_queues);
+
 	ret = register_netdev(net);
 	if (ret != 0) {
 		pr_err("Unable to register netdev.\n");

drivers/net/hyperv/rndis_filter.c

@@ -31,7 +31,7 @@
 #include "hyperv_net.h"
 
 
-#define RNDIS_EXT_LEN 100
+#define RNDIS_EXT_LEN PAGE_SIZE
 struct rndis_request {
 	struct list_head list_ent;
 	struct completion  wait_event;
@@ -94,6 +94,8 @@ static struct rndis_request *get_rndis_request(struct rndis_device *dev,
 	rndis_msg->ndis_msg_type = msg_type;
 	rndis_msg->msg_len = msg_len;
 
+	request->pkt.q_idx = 0;
+
 	/*
 	 * Set the request id. This field is always after the rndis header for
 	 * request/response packet types so we just used the SetRequest as a
@@ -509,6 +511,19 @@ static int rndis_filter_query_device(struct rndis_device *dev, u32 oid,
 	query->info_buflen = 0;
 	query->dev_vc_handle = 0;
 
+	if (oid == OID_GEN_RECEIVE_SCALE_CAPABILITIES) {
+		struct ndis_recv_scale_cap *cap;
+
+		request->request_msg.msg_len +=
+			sizeof(struct ndis_recv_scale_cap);
+		query->info_buflen = sizeof(struct ndis_recv_scale_cap);
+		cap = (struct ndis_recv_scale_cap *)((unsigned long)query +
+						     query->info_buf_offset);
+		cap->hdr.type = NDIS_OBJECT_TYPE_RSS_CAPABILITIES;
+		cap->hdr.rev = NDIS_RECEIVE_SCALE_CAPABILITIES_REVISION_2;
+		cap->hdr.size = sizeof(struct ndis_recv_scale_cap);
+	}
+
 	ret = rndis_filter_send_request(dev, request);
 	if (ret != 0)
 		goto cleanup;
@@ -695,6 +710,89 @@ int rndis_filter_set_offload_params(struct hv_device *hdev,
 	return ret;
 }
 
+u8 netvsc_hash_key[HASH_KEYLEN] = {
+	0x6d, 0x5a, 0x56, 0xda, 0x25, 0x5b, 0x0e, 0xc2,
+	0x41, 0x67, 0x25, 0x3d, 0x43, 0xa3, 0x8f, 0xb0,
+	0xd0, 0xca, 0x2b, 0xcb, 0xae, 0x7b, 0x30, 0xb4,
+	0x77, 0xcb, 0x2d, 0xa3, 0x80, 0x30, 0xf2, 0x0c,
+	0x6a, 0x42, 0xb7, 0x3b, 0xbe, 0xac, 0x01, 0xfa
+};
+
+int rndis_filter_set_rss_param(struct rndis_device *rdev, int num_queue)
+{
+	struct net_device *ndev = rdev->net_dev->ndev;
+	struct rndis_request *request;
+	struct rndis_set_request *set;
+	struct rndis_set_complete *set_complete;
+	u32 extlen = sizeof(struct ndis_recv_scale_param) +
+		     4*ITAB_NUM + HASH_KEYLEN;
+	struct ndis_recv_scale_param *rssp;
+	u32 *itab;
+	u8 *keyp;
+	int i, t, ret;
+
+	request = get_rndis_request(
+			rdev, RNDIS_MSG_SET,
+			RNDIS_MESSAGE_SIZE(struct rndis_set_request) + extlen);
+	if (!request)
+		return -ENOMEM;
+
+	set = &request->request_msg.msg.set_req;
+	set->oid = OID_GEN_RECEIVE_SCALE_PARAMETERS;
+	set->info_buflen = extlen;
+	set->info_buf_offset = sizeof(struct rndis_set_request);
+	set->dev_vc_handle = 0;
+
+	rssp = (struct ndis_recv_scale_param *)(set + 1);
+	rssp->hdr.type = NDIS_OBJECT_TYPE_RSS_PARAMETERS;
+	rssp->hdr.rev = NDIS_RECEIVE_SCALE_PARAMETERS_REVISION_2;
+	rssp->hdr.size = sizeof(struct ndis_recv_scale_param);
+	rssp->flag = 0;
+	rssp->hashinfo = NDIS_HASH_FUNC_TOEPLITZ | NDIS_HASH_IPV4 |
+			 NDIS_HASH_TCP_IPV4;
+	rssp->indirect_tabsize = 4*ITAB_NUM;
+	rssp->indirect_taboffset = sizeof(struct ndis_recv_scale_param);
+	rssp->hashkey_size = HASH_KEYLEN;
+	rssp->kashkey_offset = rssp->indirect_taboffset +
+			       rssp->indirect_tabsize;
+
+	/* Set indirection table entries */
+	itab = (u32 *)(rssp + 1);
+	for (i = 0; i < ITAB_NUM; i++)
+		itab[i] = i % num_queue;
+
+	/* Set hask key values */
+	keyp = (u8 *)((unsigned long)rssp + rssp->kashkey_offset);
+	for (i = 0; i < HASH_KEYLEN; i++)
+		keyp[i] = netvsc_hash_key[i];
+
+
+	ret = rndis_filter_send_request(rdev, request);
+	if (ret != 0)
+		goto cleanup;
+
+	t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
+	if (t == 0) {
+		netdev_err(ndev, "timeout before we got a set response...\n");
+		/* can't put_rndis_request, since we may still receive a
+		 * send-completion.
+		 */
+		return -ETIMEDOUT;
+	} else {
+		set_complete = &request->response_msg.msg.set_complete;
+		if (set_complete->status != RNDIS_STATUS_SUCCESS) {
+			netdev_err(ndev, "Fail to set RSS parameters:0x%x\n",
+				   set_complete->status);
+			ret = -EINVAL;
+		}
+	}
+
+cleanup:
+	put_rndis_request(rdev, request);
+	return ret;
+}
+
+
 static int rndis_filter_query_device_link_status(struct rndis_device *dev)
 {
 	u32 size = sizeof(u32);
@@ -886,6 +984,28 @@ static int rndis_filter_close_device(struct rndis_device *dev)
 	return ret;
 }
 
+static void netvsc_sc_open(struct vmbus_channel *new_sc)
+{
+	struct netvsc_device *nvscdev;
+	u16 chn_index = new_sc->offermsg.offer.sub_channel_index;
+	int ret;
+
+	nvscdev = hv_get_drvdata(new_sc->primary_channel->device_obj);
+
+	if (chn_index >= nvscdev->num_chn)
+		return;
+
+	set_per_channel_state(new_sc, nvscdev->sub_cb_buf + (chn_index - 1) *
+			      NETVSC_PACKET_SIZE);
+
+	ret = vmbus_open(new_sc, nvscdev->ring_size * PAGE_SIZE,
+			 nvscdev->ring_size * PAGE_SIZE, NULL, 0,
+			 netvsc_channel_cb, new_sc);
+
+	if (ret == 0)
+		nvscdev->chn_table[chn_index] = new_sc;
+}
+
 int rndis_filter_device_add(struct hv_device *dev,
 				  void *additional_info)
 {
@@ -894,6 +1014,10 @@ int rndis_filter_device_add(struct hv_device *dev,
 	struct rndis_device *rndis_device;
 	struct netvsc_device_info *device_info = additional_info;
 	struct ndis_offload_params offloads;
+	struct nvsp_message *init_packet;
+	int t;
+	struct ndis_recv_scale_cap rsscap;
+	u32 rsscap_size = sizeof(struct ndis_recv_scale_cap);
 
 	rndis_device = get_rndis_device();
 	if (!rndis_device)
@@ -913,6 +1037,7 @@ int rndis_filter_device_add(struct hv_device *dev,
 
 	/* Initialize the rndis device */
 	net_device = hv_get_drvdata(dev);
+	net_device->num_chn = 1;
 
 	net_device->extension = rndis_device;
 	rndis_device->net_dev = net_device;
@@ -952,7 +1077,6 @@ int rndis_filter_device_add(struct hv_device *dev,
 	if (ret)
 		goto err_dev_remv;
 
-
 	rndis_filter_query_device_link_status(rndis_device);
 
 	device_info->link_state = rndis_device->link_state;
@@ -961,7 +1085,66 @@ int rndis_filter_device_add(struct hv_device *dev,
 		 rndis_device->hw_mac_adr,
 		 device_info->link_state ? "down" : "up");
 
-	return ret;
+	if (net_device->nvsp_version < NVSP_PROTOCOL_VERSION_5)
+		return 0;
+
+	/* vRSS setup */
+	memset(&rsscap, 0, rsscap_size);
+	ret = rndis_filter_query_device(rndis_device,
+					OID_GEN_RECEIVE_SCALE_CAPABILITIES,
+					&rsscap, &rsscap_size);
+	if (ret || rsscap.num_recv_que < 2)
+		goto out;
+
+	net_device->num_chn = (num_online_cpus() < rsscap.num_recv_que) ?
+			       num_online_cpus() : rsscap.num_recv_que;
+	if (net_device->num_chn == 1)
+		goto out;
+
+	net_device->sub_cb_buf = vzalloc((net_device->num_chn - 1) *
+					 NETVSC_PACKET_SIZE);
+	if (!net_device->sub_cb_buf) {
+		net_device->num_chn = 1;
+		dev_info(&dev->device, "No memory for subchannels.\n");
+		goto out;
+	}
+
+	vmbus_set_sc_create_callback(dev->channel, netvsc_sc_open);
+
+	init_packet = &net_device->channel_init_pkt;
+	memset(init_packet, 0, sizeof(struct nvsp_message));
+	init_packet->hdr.msg_type = NVSP_MSG5_TYPE_SUBCHANNEL;
+	init_packet->msg.v5_msg.subchn_req.op = NVSP_SUBCHANNEL_ALLOCATE;
+	init_packet->msg.v5_msg.subchn_req.num_subchannels =
+						net_device->num_chn - 1;
+	ret = vmbus_sendpacket(dev->channel, init_packet,
+			       sizeof(struct nvsp_message),
+			       (unsigned long)init_packet,
+			       VM_PKT_DATA_INBAND,
+			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
+	if (ret)
+		goto out;
+	t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
+	if (t == 0) {
+		ret = -ETIMEDOUT;
+		goto out;
+	}
+	if (init_packet->msg.v5_msg.subchn_comp.status !=
+	    NVSP_STAT_SUCCESS) {
+		ret = -ENODEV;
+		goto out;
+	}
+	net_device->num_chn = 1 +
+		init_packet->msg.v5_msg.subchn_comp.num_subchannels;
+
+	vmbus_are_subchannels_present(dev->channel);
+
+	ret = rndis_filter_set_rss_param(rndis_device, net_device->num_chn);
+
+out:
+	if (ret)
+		net_device->num_chn = 1;
+	return 0; /* return 0 because primary channel can be used alone */
 
 err_dev_remv:
 	rndis_filter_device_remove(dev);