hyperv: Remove recv_pkt_list and lock
authorHaiyang Zhang <haiyangz@microsoft.com>
Mon, 21 Apr 2014 21:54:43 +0000 (14:54 -0700)
committerDavid S. Miller <davem@davemloft.net>
Wed, 23 Apr 2014 18:48:39 +0000 (14:48 -0400)
Removed recv_pkt_list and lock, and updated related code, so that
the locking overhead is reduced especially when multiple channels
are in use.

The recv_pkt_list isn't actually necessary because the packets are
processed sequentially in each channel. It has been replaced by a
local variable, and the related lock for this list is also removed.
The is_data_pkt field is not used in receive path, so its assignment
is cleaned up.

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
drivers/net/hyperv/netvsc.c
drivers/net/hyperv/netvsc_drv.c
drivers/net/hyperv/rndis_filter.c

index 57eb3f906d64be9508ae700804cff05d276e347a..a1af0f7711e28a76c74e7f3b59637725770c59a9 100644 (file)
@@ -119,27 +119,14 @@ struct ndis_recv_scale_param { /* NDIS_RECEIVE_SCALE_PARAMETERS */
 };
 
 /* Fwd declaration */
-struct hv_netvsc_packet;
 struct ndis_tcp_ip_checksum_info;
 
-/* Represent the xfer page packet which contains 1 or more netvsc packet */
-struct xferpage_packet {
-       struct list_head list_ent;
-       u32 status;
-
-       /* # of netvsc packets this xfer packet contains */
-       u32 count;
-
-       struct vmbus_channel *channel;
-};
-
 /*
  * Represent netvsc packet which contains 1 RNDIS and 1 ethernet frame
  * within the RNDIS
  */
 struct hv_netvsc_packet {
        /* Bookkeeping stuff */
-       struct list_head list_ent;
        u32 status;
 
        struct hv_device *device;
@@ -149,18 +136,7 @@ struct hv_netvsc_packet {
        u16 q_idx;
        struct vmbus_channel *channel;
 
-       /*
-        * Valid only for receives when we break a xfer page packet
-        * into multiple netvsc packets
-        */
-       struct xferpage_packet *xfer_page_pkt;
-
        union {
-               struct {
-                       u64 recv_completion_tid;
-                       void *recv_completion_ctx;
-                       void (*recv_completion)(void *context);
-               } recv;
                struct {
                        u64 send_completion_tid;
                        void *send_completion_ctx;
@@ -613,9 +589,6 @@ struct nvsp_message {
 
 #define NETVSC_RECEIVE_BUFFER_ID               0xcafe
 
-/* Preallocated receive packets */
-#define NETVSC_RECEIVE_PACKETLIST_COUNT                256
-
 #define NETVSC_PACKET_SIZE                      2048
 
 #define VRSS_SEND_TAB_SIZE 16
@@ -630,12 +603,6 @@ struct netvsc_device {
        wait_queue_head_t wait_drain;
        bool start_remove;
        bool destroy;
-       /*
-        * List of free preallocated hv_netvsc_packet to represent receive
-        * packet
-        */
-       struct list_head recv_pkt_list;
-       spinlock_t recv_pkt_list_lock;
 
        /* Receive buffer allocated by us but manages by NetVSP */
        void *recv_buf;
index e7e77f12bc38872001497378dc4cce7bb92cd0d1..b10334773b328a07121fa4f32bbaa14f2590b40a 100644 (file)
@@ -387,7 +387,6 @@ static void netvsc_disconnect_vsp(struct netvsc_device *net_device)
 int netvsc_device_remove(struct hv_device *device)
 {
        struct netvsc_device *net_device;
-       struct hv_netvsc_packet *netvsc_packet, *pos;
        unsigned long flags;
 
        net_device = hv_get_drvdata(device);
@@ -416,12 +415,6 @@ int netvsc_device_remove(struct hv_device *device)
        vmbus_close(device->channel);
 
        /* Release all resources */
-       list_for_each_entry_safe(netvsc_packet, pos,
-                                &net_device->recv_pkt_list, list_ent) {
-               list_del(&netvsc_packet->list_ent);
-               kfree(netvsc_packet);
-       }
-
        if (net_device->sub_cb_buf)
                vfree(net_device->sub_cb_buf);
 
@@ -641,62 +634,6 @@ retry_send_cmplt:
        }
 }
 
-/* Send a receive completion packet to RNDIS device (ie NetVsp) */
-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;
-       unsigned long flags;
-       struct net_device *ndev;
-       u32 status = NVSP_STAT_NONE;
-
-       /*
-        * Even though it seems logical to do a GetOutboundNetDevice() here to
-        * send out receive completion, we are using GetInboundNetDevice()
-        * since we may have disable outbound traffic already.
-        */
-       net_device = get_inbound_net_device(device);
-       if (!net_device)
-               return;
-       ndev = net_device->ndev;
-
-       /* Overloading use of the lock. */
-       spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
-
-       if (packet->status != NVSP_STAT_SUCCESS)
-               packet->xfer_page_pkt->status = NVSP_STAT_FAIL;
-
-       packet->xfer_page_pkt->count--;
-
-       /*
-        * Last one in the line that represent 1 xfer page packet.
-        * Return the xfer page packet itself to the freelist
-        */
-       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,
-                             &net_device->recv_pkt_list);
-
-       }
-
-       /* Put the packet back */
-       list_add_tail(&packet->list_ent, &net_device->recv_pkt_list);
-       spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
-
-       /* Send a receive completion for the xfer page packet */
-       if (fsend_receive_comp)
-               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,
@@ -704,16 +641,13 @@ static void netvsc_receive(struct netvsc_device *net_device,
 {
        struct vmtransfer_page_packet_header *vmxferpage_packet;
        struct nvsp_message *nvsp_packet;
-       struct hv_netvsc_packet *netvsc_packet = NULL;
-       /* struct netvsc_driver *netvscDriver; */
-       struct xferpage_packet *xferpage_packet = NULL;
+       struct hv_netvsc_packet nv_pkt;
+       struct hv_netvsc_packet *netvsc_packet = &nv_pkt;
+       u32 status = NVSP_STAT_SUCCESS;
        int i;
        int count = 0;
-       unsigned long flags;
        struct net_device *ndev;
 
-       LIST_HEAD(listHead);
-
        ndev = net_device->ndev;
 
        /*
@@ -746,78 +680,14 @@ static void netvsc_receive(struct netvsc_device *net_device,
                return;
        }
 
-       /*
-        * Grab free packets (range count + 1) to represent this xfer
-        * page packet. +1 to represent the xfer page packet itself.
-        * We grab it here so that we know exactly how many we can
-        * fulfil
-        */
-       spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
-       while (!list_empty(&net_device->recv_pkt_list)) {
-               list_move_tail(net_device->recv_pkt_list.next, &listHead);
-               if (++count == vmxferpage_packet->range_cnt + 1)
-                       break;
-       }
-       spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
-
-       /*
-        * We need at least 2 netvsc pkts (1 to represent the xfer
-        * page and at least 1 for the range) i.e. we can handled
-        * some of the xfer page packet ranges...
-        */
-       if (count < 2) {
-               netdev_err(ndev, "Got only %d netvsc pkt...needed "
-                       "%d pkts. Dropping this xfer page packet completely!\n",
-                       count, vmxferpage_packet->range_cnt + 1);
-
-               /* Return it to the freelist */
-               spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
-               for (i = count; i != 0; i--) {
-                       list_move_tail(listHead.next,
-                                      &net_device->recv_pkt_list);
-               }
-               spin_unlock_irqrestore(&net_device->recv_pkt_list_lock,
-                                      flags);
-
-               netvsc_send_recv_completion(device, channel, net_device,
-                                           vmxferpage_packet->d.trans_id,
-                                           NVSP_STAT_FAIL);
-
-               return;
-       }
-
-       /* Remove the 1st packet to represent the xfer page packet itself */
-       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;
-
-       if (xferpage_packet->count != vmxferpage_packet->range_cnt) {
-               netdev_err(ndev, "Needed %d netvsc pkts to satisfy "
-                       "this xfer page...got %d\n",
-                       vmxferpage_packet->range_cnt, xferpage_packet->count);
-       }
+       count = vmxferpage_packet->range_cnt;
+       netvsc_packet->device = device;
+       netvsc_packet->channel = channel;
 
        /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
-       for (i = 0; i < (count - 1); i++) {
-               netvsc_packet = (struct hv_netvsc_packet *)listHead.next;
-               list_del(&netvsc_packet->list_ent);
-
+       for (i = 0; i < count; i++) {
                /* Initialize the netvsc packet */
                netvsc_packet->status = NVSP_STAT_SUCCESS;
-               netvsc_packet->xfer_page_pkt = xferpage_packet;
-               netvsc_packet->completion.recv.recv_completion =
-                                       netvsc_receive_completion;
-               netvsc_packet->completion.recv.recv_completion_ctx =
-                                       netvsc_packet;
-               netvsc_packet->device = device;
-               /* Save this so that we can send it back */
-               netvsc_packet->completion.recv.recv_completion_tid =
-                                       vmxferpage_packet->d.trans_id;
-
                netvsc_packet->data = (void *)((unsigned long)net_device->
                        recv_buf + vmxferpage_packet->ranges[i].byte_offset);
                netvsc_packet->total_data_buflen =
@@ -826,10 +696,12 @@ static void netvsc_receive(struct netvsc_device *net_device,
                /* Pass it to the upper layer */
                rndis_filter_receive(device, netvsc_packet);
 
-               netvsc_receive_completion(netvsc_packet->
-                               completion.recv.recv_completion_ctx);
+               if (netvsc_packet->status != NVSP_STAT_SUCCESS)
+                       status = NVSP_STAT_FAIL;
        }
 
+       netvsc_send_recv_completion(device, channel, net_device,
+                                   vmxferpage_packet->d.trans_id, status);
 }
 
 
@@ -956,11 +828,9 @@ void netvsc_channel_cb(void *context)
 int netvsc_device_add(struct hv_device *device, void *additional_info)
 {
        int ret = 0;
-       int i;
        int ring_size =
        ((struct netvsc_device_info *)additional_info)->ring_size;
        struct netvsc_device *net_device;
-       struct hv_netvsc_packet *packet, *pos;
        struct net_device *ndev;
 
        net_device = alloc_net_device(device);
@@ -981,18 +851,6 @@ int netvsc_device_add(struct hv_device *device, void *additional_info)
        ndev = net_device->ndev;
 
        /* Initialize the NetVSC channel extension */
-       spin_lock_init(&net_device->recv_pkt_list_lock);
-
-       INIT_LIST_HEAD(&net_device->recv_pkt_list);
-
-       for (i = 0; i < NETVSC_RECEIVE_PACKETLIST_COUNT; i++) {
-               packet = kzalloc(sizeof(struct hv_netvsc_packet), GFP_KERNEL);
-               if (!packet)
-                       break;
-
-               list_add_tail(&packet->list_ent,
-                             &net_device->recv_pkt_list);
-       }
        init_completion(&net_device->channel_init_wait);
 
        set_per_channel_state(device->channel, net_device->cb_buffer);
@@ -1028,16 +886,8 @@ close:
 
 cleanup:
 
-       if (net_device) {
-               list_for_each_entry_safe(packet, pos,
-                                        &net_device->recv_pkt_list,
-                                        list_ent) {
-                       list_del(&packet->list_ent);
-                       kfree(packet);
-               }
-
+       if (net_device)
                kfree(net_device);
-       }
 
        return ret;
 }
index 093cf3fc46b8683390892c48fe7dc5a2382b4f98..8f6d53a2ed95ee14e1e9e705779b9de3e145f020 100644 (file)
@@ -638,7 +638,7 @@ 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->
+       skb_record_rx_queue(skb, packet->channel->
                            offermsg.offer.sub_channel_index %
                            net->real_num_rx_queues);
 
index d92cfbe4341036bbeffdd8db834d220d5b514b61..48f5a0fbd6748d072f123f068d346aff4e42f06a 100644 (file)
@@ -401,8 +401,6 @@ static void rndis_filter_receive_data(struct rndis_device *dev,
        pkt->total_data_buflen = rndis_pkt->data_len;
        pkt->data = (void *)((unsigned long)pkt->data + data_offset);
 
-       pkt->is_data_pkt = true;
-
        vlan = rndis_get_ppi(rndis_pkt, IEEE_8021Q_INFO);
        if (vlan) {
                pkt->vlan_tci = VLAN_TAG_PRESENT | vlan->vlanid |