Merge branch 'netvsc-NAPI'
Stephen Hemminger says:
====================
NAPI support for Hyper-V
These patches enable NAPI, GRO and napi_alloc_skb for Hyper-V netvsc
driver.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
diff --git a/drivers/hv/ring_buffer.c b/drivers/hv/ring_buffer.c
index 87799e8..c3f1a9e 100644
--- a/drivers/hv/ring_buffer.c
+++ b/drivers/hv/ring_buffer.c
@@ -32,6 +32,8 @@
#include "hyperv_vmbus.h"
+#define VMBUS_PKT_TRAILER 8
+
/*
* When we write to the ring buffer, check if the host needs to
* be signaled. Here is the details of this protocol:
@@ -336,6 +338,12 @@
return 0;
}
+static inline void
+init_cached_read_index(struct hv_ring_buffer_info *rbi)
+{
+ rbi->cached_read_index = rbi->ring_buffer->read_index;
+}
+
int hv_ringbuffer_read(struct vmbus_channel *channel,
void *buffer, u32 buflen, u32 *buffer_actual_len,
u64 *requestid, bool raw)
@@ -366,7 +374,8 @@
return ret;
}
- init_cached_read_index(channel);
+ init_cached_read_index(inring_info);
+
next_read_location = hv_get_next_read_location(inring_info);
next_read_location = hv_copyfrom_ringbuffer(inring_info, &desc,
sizeof(desc),
@@ -410,3 +419,86 @@
return ret;
}
+
+/*
+ * Determine number of bytes available in ring buffer after
+ * the current iterator (priv_read_index) location.
+ *
+ * This is similar to hv_get_bytes_to_read but with private
+ * read index instead.
+ */
+static u32 hv_pkt_iter_avail(const struct hv_ring_buffer_info *rbi)
+{
+ u32 priv_read_loc = rbi->priv_read_index;
+ u32 write_loc = READ_ONCE(rbi->ring_buffer->write_index);
+
+ if (write_loc >= priv_read_loc)
+ return write_loc - priv_read_loc;
+ else
+ return (rbi->ring_datasize - priv_read_loc) + write_loc;
+}
+
+/*
+ * Get first vmbus packet from ring buffer after read_index
+ *
+ * If ring buffer is empty, returns NULL and no other action needed.
+ */
+struct vmpacket_descriptor *hv_pkt_iter_first(struct vmbus_channel *channel)
+{
+ struct hv_ring_buffer_info *rbi = &channel->inbound;
+
+ /* set state for later hv_signal_on_read() */
+ init_cached_read_index(rbi);
+
+ if (hv_pkt_iter_avail(rbi) < sizeof(struct vmpacket_descriptor))
+ return NULL;
+
+ return hv_get_ring_buffer(rbi) + rbi->priv_read_index;
+}
+EXPORT_SYMBOL_GPL(hv_pkt_iter_first);
+
+/*
+ * Get next vmbus packet from ring buffer.
+ *
+ * Advances the current location (priv_read_index) and checks for more
+ * data. If the end of the ring buffer is reached, then return NULL.
+ */
+struct vmpacket_descriptor *
+__hv_pkt_iter_next(struct vmbus_channel *channel,
+ const struct vmpacket_descriptor *desc)
+{
+ struct hv_ring_buffer_info *rbi = &channel->inbound;
+ u32 packetlen = desc->len8 << 3;
+ u32 dsize = rbi->ring_datasize;
+
+ /* bump offset to next potential packet */
+ rbi->priv_read_index += packetlen + VMBUS_PKT_TRAILER;
+ if (rbi->priv_read_index >= dsize)
+ rbi->priv_read_index -= dsize;
+
+ /* more data? */
+ if (hv_pkt_iter_avail(rbi) < sizeof(struct vmpacket_descriptor))
+ return NULL;
+ else
+ return hv_get_ring_buffer(rbi) + rbi->priv_read_index;
+}
+EXPORT_SYMBOL_GPL(__hv_pkt_iter_next);
+
+/*
+ * Update host ring buffer after iterating over packets.
+ */
+void hv_pkt_iter_close(struct vmbus_channel *channel)
+{
+ struct hv_ring_buffer_info *rbi = &channel->inbound;
+
+ /*
+ * Make sure all reads are done before we update the read index since
+ * the writer may start writing to the read area once the read index
+ * is updated.
+ */
+ virt_rmb();
+ rbi->ring_buffer->read_index = rbi->priv_read_index;
+
+ hv_signal_on_read(channel);
+}
+EXPORT_SYMBOL_GPL(hv_pkt_iter_close);
diff --git a/drivers/net/hyperv/hyperv_net.h b/drivers/net/hyperv/hyperv_net.h
index d3e73ac..7433b16 100644
--- a/drivers/net/hyperv/hyperv_net.h
+++ b/drivers/net/hyperv/hyperv_net.h
@@ -196,6 +196,7 @@
const struct ndis_tcp_ip_checksum_info *csum_info,
const struct ndis_pkt_8021q_info *vlan);
void netvsc_channel_cb(void *context);
+int netvsc_poll(struct napi_struct *napi, int budget);
int rndis_filter_open(struct netvsc_device *nvdev);
int rndis_filter_close(struct netvsc_device *nvdev);
int rndis_filter_device_add(struct hv_device *dev,
@@ -720,6 +721,7 @@
/* Per channel data */
struct netvsc_channel {
struct vmbus_channel *channel;
+ struct napi_struct napi;
struct multi_send_data msd;
struct multi_recv_comp mrc;
atomic_t queue_sends;
diff --git a/drivers/net/hyperv/netvsc.c b/drivers/net/hyperv/netvsc.c
index d35ebd9..b1328ce 100644
--- a/drivers/net/hyperv/netvsc.c
+++ b/drivers/net/hyperv/netvsc.c
@@ -556,6 +556,7 @@
struct net_device *ndev = hv_get_drvdata(device);
struct net_device_context *net_device_ctx = netdev_priv(ndev);
struct netvsc_device *net_device = net_device_ctx->nvdev;
+ int i;
netvsc_disconnect_vsp(device);
@@ -570,6 +571,9 @@
/* Now, we can close the channel safely */
vmbus_close(device->channel);
+ for (i = 0; i < VRSS_CHANNEL_MAX; i++)
+ napi_disable(&net_device->chan_table[0].napi);
+
/* Release all resources */
free_netvsc_device(net_device);
}
@@ -600,9 +604,9 @@
static void netvsc_send_tx_complete(struct netvsc_device *net_device,
struct vmbus_channel *incoming_channel,
struct hv_device *device,
- struct vmpacket_descriptor *packet)
+ const struct vmpacket_descriptor *desc)
{
- struct sk_buff *skb = (struct sk_buff *)(unsigned long)packet->trans_id;
+ struct sk_buff *skb = (struct sk_buff *)(unsigned long)desc->trans_id;
struct net_device *ndev = hv_get_drvdata(device);
struct net_device_context *net_device_ctx = netdev_priv(ndev);
struct vmbus_channel *channel = device->channel;
@@ -647,14 +651,11 @@
static void netvsc_send_completion(struct netvsc_device *net_device,
struct vmbus_channel *incoming_channel,
struct hv_device *device,
- struct vmpacket_descriptor *packet)
+ const struct vmpacket_descriptor *desc)
{
- struct nvsp_message *nvsp_packet;
+ struct nvsp_message *nvsp_packet = hv_pkt_data(desc);
struct net_device *ndev = hv_get_drvdata(device);
- nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
- (packet->offset8 << 3));
-
switch (nvsp_packet->hdr.msg_type) {
case NVSP_MSG_TYPE_INIT_COMPLETE:
case NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE:
@@ -668,7 +669,7 @@
case NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE:
netvsc_send_tx_complete(net_device, incoming_channel,
- device, packet);
+ device, desc);
break;
default:
@@ -1066,28 +1067,29 @@
return rcd;
}
-static void netvsc_receive(struct net_device *ndev,
+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,
- struct vmtransfer_page_packet_header *vmxferpage_packet,
+ 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;
- struct recv_comp_data *rcd;
- u16 q_idx = channel->offermsg.offer.sub_channel_index;
/* Make sure this is a valid nvsp packet */
if (unlikely(nvsp->hdr.msg_type != NVSP_MSG1_TYPE_SEND_RNDIS_PKT)) {
netif_err(net_device_ctx, rx_err, ndev,
"Unknown nvsp packet type received %u\n",
nvsp->hdr.msg_type);
- return;
+ return 0;
}
if (unlikely(vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID)) {
@@ -1095,7 +1097,7 @@
"Invalid xfer page set id - expecting %x got %x\n",
NETVSC_RECEIVE_BUFFER_ID,
vmxferpage_packet->xfer_pageset_id);
- return;
+ return 0;
}
count = vmxferpage_packet->range_cnt;
@@ -1111,26 +1113,26 @@
channel, data, buflen);
}
- if (!net_device->chan_table[q_idx].mrc.buf) {
+ if (net_device->chan_table[q_idx].mrc.buf) {
+ struct recv_comp_data *rcd;
+
+ 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;
}
-
- rcd = get_recv_comp_slot(net_device, channel, q_idx);
-
- if (!rcd) {
- netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
- q_idx, vmxferpage_packet->d.trans_id);
- return;
- }
-
- rcd->tid = vmxferpage_packet->d.trans_id;
- rcd->status = status;
+ return count;
}
static void netvsc_send_table(struct hv_device *hdev,
@@ -1180,17 +1182,15 @@
}
}
-static void netvsc_process_raw_pkt(struct hv_device *device,
- struct vmbus_channel *channel,
- struct netvsc_device *net_device,
- struct net_device *ndev,
- u64 request_id,
- struct vmpacket_descriptor *desc)
+static int netvsc_process_raw_pkt(struct hv_device *device,
+ struct vmbus_channel *channel,
+ struct netvsc_device *net_device,
+ struct net_device *ndev,
+ u64 request_id,
+ const struct vmpacket_descriptor *desc)
{
struct net_device_context *net_device_ctx = netdev_priv(ndev);
- struct nvsp_message *nvmsg
- = (struct nvsp_message *)((unsigned long)desc
- + (desc->offset8 << 3));
+ struct nvsp_message *nvmsg = hv_pkt_data(desc);
switch (desc->type) {
case VM_PKT_COMP:
@@ -1198,10 +1198,8 @@
break;
case VM_PKT_DATA_USING_XFER_PAGES:
- netvsc_receive(ndev, net_device, net_device_ctx,
- device, channel,
- (struct vmtransfer_page_packet_header *)desc,
- nvmsg);
+ return netvsc_receive(ndev, net_device, net_device_ctx,
+ device, channel, desc, nvmsg);
break;
case VM_PKT_DATA_INBAND:
@@ -1213,22 +1211,63 @@
desc->type, request_id);
break;
}
+
+ return 0;
+}
+
+static struct hv_device *netvsc_channel_to_device(struct vmbus_channel *channel)
+{
+ struct vmbus_channel *primary = channel->primary_channel;
+
+ return primary ? primary->device_obj : channel->device_obj;
+}
+
+int netvsc_poll(struct napi_struct *napi, int budget)
+{
+ struct netvsc_channel *nvchan
+ = container_of(napi, struct netvsc_channel, napi);
+ struct vmbus_channel *channel = nvchan->channel;
+ struct hv_device *device = netvsc_channel_to_device(channel);
+ u16 q_idx = channel->offermsg.offer.sub_channel_index;
+ struct net_device *ndev = hv_get_drvdata(device);
+ struct netvsc_device *net_device = net_device_to_netvsc_device(ndev);
+ const struct vmpacket_descriptor *desc;
+ int work_done = 0;
+
+ desc = hv_pkt_iter_first(channel);
+ while (desc) {
+ int count;
+
+ count = netvsc_process_raw_pkt(device, channel, net_device,
+ ndev, desc->trans_id, desc);
+ work_done += count;
+ desc = __hv_pkt_iter_next(channel, desc);
+
+ /* If receive packet budget is exhausted, reschedule */
+ if (work_done >= budget) {
+ work_done = budget;
+ break;
+ }
+ }
+ hv_pkt_iter_close(channel);
+
+ /* If ring is empty and NAPI is not doing polling */
+ if (work_done < budget &&
+ napi_complete_done(napi, work_done) &&
+ hv_end_read(&channel->inbound) != 0)
+ napi_reschedule(napi);
+
+ netvsc_chk_recv_comp(net_device, channel, q_idx);
+ return work_done;
}
void netvsc_channel_cb(void *context)
{
struct vmbus_channel *channel = context;
+ struct hv_device *device = netvsc_channel_to_device(channel);
u16 q_idx = channel->offermsg.offer.sub_channel_index;
- struct hv_device *device;
struct netvsc_device *net_device;
- struct vmpacket_descriptor *desc;
struct net_device *ndev;
- bool need_to_commit = false;
-
- if (channel->primary_channel != NULL)
- device = channel->primary_channel->device_obj;
- else
- device = channel->device_obj;
ndev = hv_get_drvdata(device);
if (unlikely(!ndev))
@@ -1239,21 +1278,9 @@
netvsc_channel_idle(net_device, q_idx))
return;
- /* commit_rd_index() -> hv_signal_on_read() needs this. */
- init_cached_read_index(channel);
-
- while ((desc = get_next_pkt_raw(channel)) != NULL) {
- netvsc_process_raw_pkt(device, channel, net_device,
- ndev, desc->trans_id, desc);
-
- put_pkt_raw(channel, desc);
- need_to_commit = true;
- }
-
- if (need_to_commit)
- commit_rd_index(channel);
-
- netvsc_chk_recv_comp(net_device, channel, q_idx);
+ /* disable interupts from host */
+ hv_begin_read(&channel->inbound);
+ napi_schedule(&net_device->chan_table[q_idx].napi);
}
/*
@@ -1275,6 +1302,11 @@
net_device->ring_size = ring_size;
+ /* Because the device uses NAPI, all the interrupt batching and
+ * control is done via Net softirq, not the channel handling
+ */
+ set_channel_read_mode(device->channel, HV_CALL_ISR);
+
/* Open the channel */
ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
ring_size * PAGE_SIZE, NULL, 0,
@@ -1292,8 +1324,16 @@
* chn_table with the default channel to use it before subchannels are
* opened.
*/
- for (i = 0; i < VRSS_CHANNEL_MAX; i++)
- net_device->chan_table[i].channel = device->channel;
+ for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
+ struct netvsc_channel *nvchan = &net_device->chan_table[i];
+
+ nvchan->channel = device->channel;
+ netif_napi_add(ndev, &nvchan->napi,
+ netvsc_poll, NAPI_POLL_WEIGHT);
+ }
+
+ /* Enable NAPI handler for init callbacks */
+ napi_enable(&net_device->chan_table[0].napi);
/* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
* populated.
@@ -1313,6 +1353,8 @@
return ret;
close:
+ napi_disable(&net_device->chan_table[0].napi);
+
/* Now, we can close the channel safely */
vmbus_close(device->channel);
diff --git a/drivers/net/hyperv/netvsc_drv.c b/drivers/net/hyperv/netvsc_drv.c
index bc05c89..617dd90 100644
--- a/drivers/net/hyperv/netvsc_drv.c
+++ b/drivers/net/hyperv/netvsc_drv.c
@@ -589,13 +589,14 @@
}
static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
+ struct napi_struct *napi,
const struct ndis_tcp_ip_checksum_info *csum_info,
const struct ndis_pkt_8021q_info *vlan,
void *data, u32 buflen)
{
struct sk_buff *skb;
- skb = netdev_alloc_skb_ip_align(net, buflen);
+ skb = napi_alloc_skb(napi, buflen);
if (!skb)
return skb;
@@ -642,11 +643,11 @@
{
struct net_device_context *net_device_ctx = netdev_priv(net);
struct netvsc_device *net_device = net_device_ctx->nvdev;
+ u16 q_idx = channel->offermsg.offer.sub_channel_index;
+ struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
struct net_device *vf_netdev;
struct sk_buff *skb;
struct netvsc_stats *rx_stats;
- u16 q_idx = channel->offermsg.offer.sub_channel_index;
-
if (net->reg_state != NETREG_REGISTERED)
return NVSP_STAT_FAIL;
@@ -664,7 +665,8 @@
net = vf_netdev;
/* Allocate a skb - TODO direct I/O to pages? */
- skb = netvsc_alloc_recv_skb(net, csum_info, vlan, data, len);
+ skb = netvsc_alloc_recv_skb(net, &nvchan->napi,
+ csum_info, vlan, data, len);
if (unlikely(!skb)) {
++net->stats.rx_dropped;
rcu_read_unlock();
@@ -679,7 +681,7 @@
* on the synthetic device because modifying the VF device
* statistics will not work correctly.
*/
- rx_stats = &net_device->chan_table[q_idx].rx_stats;
+ rx_stats = &nvchan->rx_stats;
u64_stats_update_begin(&rx_stats->syncp);
rx_stats->packets++;
rx_stats->bytes += len;
@@ -690,12 +692,7 @@
++rx_stats->multicast;
u64_stats_update_end(&rx_stats->syncp);
- /*
- * Pass the skb back up. Network stack will deallocate the skb when it
- * is done.
- * TODO - use NAPI?
- */
- netif_receive_skb(skb);
+ napi_gro_receive(&nvchan->napi, skb);
rcu_read_unlock();
return 0;
diff --git a/drivers/net/hyperv/rndis_filter.c b/drivers/net/hyperv/rndis_filter.c
index 19356f5..d7b6311 100644
--- a/drivers/net/hyperv/rndis_filter.c
+++ b/drivers/net/hyperv/rndis_filter.c
@@ -1012,6 +1012,8 @@
if (ret == 0)
nvscdev->chan_table[chn_index].channel = new_sc;
+ napi_enable(&nvscdev->chan_table[chn_index].napi);
+
spin_lock_irqsave(&nvscdev->sc_lock, flags);
nvscdev->num_sc_offered--;
spin_unlock_irqrestore(&nvscdev->sc_lock, flags);
diff --git a/include/linux/hyperv.h b/include/linux/hyperv.h
index 62bbf3c..3616248 100644
--- a/include/linux/hyperv.h
+++ b/include/linux/hyperv.h
@@ -1504,14 +1504,6 @@
return;
}
-static inline void
-init_cached_read_index(struct vmbus_channel *channel)
-{
- struct hv_ring_buffer_info *rbi = &channel->inbound;
-
- rbi->cached_read_index = rbi->ring_buffer->read_index;
-}
-
/*
* Mask off host interrupt callback notifications
*/
@@ -1545,76 +1537,48 @@
/*
* An API to support in-place processing of incoming VMBUS packets.
*/
-#define VMBUS_PKT_TRAILER 8
+/* Get data payload associated with descriptor */
+static inline void *hv_pkt_data(const struct vmpacket_descriptor *desc)
+{
+ return (void *)((unsigned long)desc + (desc->offset8 << 3));
+}
+
+/* Get data size associated with descriptor */
+static inline u32 hv_pkt_datalen(const struct vmpacket_descriptor *desc)
+{
+ return (desc->len8 << 3) - (desc->offset8 << 3);
+}
+
+
+struct vmpacket_descriptor *
+hv_pkt_iter_first(struct vmbus_channel *channel);
+
+struct vmpacket_descriptor *
+__hv_pkt_iter_next(struct vmbus_channel *channel,
+ const struct vmpacket_descriptor *pkt);
+
+void hv_pkt_iter_close(struct vmbus_channel *channel);
+
+/*
+ * Get next packet descriptor from iterator
+ * If at end of list, return NULL and update host.
+ */
static inline struct vmpacket_descriptor *
-get_next_pkt_raw(struct vmbus_channel *channel)
+hv_pkt_iter_next(struct vmbus_channel *channel,
+ const struct vmpacket_descriptor *pkt)
{
- struct hv_ring_buffer_info *ring_info = &channel->inbound;
- u32 priv_read_loc = ring_info->priv_read_index;
- void *ring_buffer = hv_get_ring_buffer(ring_info);
- u32 dsize = ring_info->ring_datasize;
- /*
- * delta is the difference between what is available to read and
- * what was already consumed in place. We commit read index after
- * the whole batch is processed.
- */
- u32 delta = priv_read_loc >= ring_info->ring_buffer->read_index ?
- priv_read_loc - ring_info->ring_buffer->read_index :
- (dsize - ring_info->ring_buffer->read_index) + priv_read_loc;
- u32 bytes_avail_toread = (hv_get_bytes_to_read(ring_info) - delta);
+ struct vmpacket_descriptor *nxt;
- if (bytes_avail_toread < sizeof(struct vmpacket_descriptor))
- return NULL;
+ nxt = __hv_pkt_iter_next(channel, pkt);
+ if (!nxt)
+ hv_pkt_iter_close(channel);
- return ring_buffer + priv_read_loc;
+ return nxt;
}
-/*
- * A helper function to step through packets "in-place"
- * This API is to be called after each successful call
- * get_next_pkt_raw().
- */
-static inline void put_pkt_raw(struct vmbus_channel *channel,
- struct vmpacket_descriptor *desc)
-{
- struct hv_ring_buffer_info *ring_info = &channel->inbound;
- u32 packetlen = desc->len8 << 3;
- u32 dsize = ring_info->ring_datasize;
-
- /*
- * Include the packet trailer.
- */
- ring_info->priv_read_index += packetlen + VMBUS_PKT_TRAILER;
- ring_info->priv_read_index %= dsize;
-}
-
-/*
- * This call commits the read index and potentially signals the host.
- * Here is the pattern for using the "in-place" consumption APIs:
- *
- * init_cached_read_index();
- *
- * while (get_next_pkt_raw() {
- * process the packet "in-place";
- * put_pkt_raw();
- * }
- * if (packets processed in place)
- * commit_rd_index();
- */
-static inline void commit_rd_index(struct vmbus_channel *channel)
-{
- struct hv_ring_buffer_info *ring_info = &channel->inbound;
- /*
- * Make sure all reads are done before we update the read index since
- * the writer may start writing to the read area once the read index
- * is updated.
- */
- virt_rmb();
- ring_info->ring_buffer->read_index = ring_info->priv_read_index;
-
- hv_signal_on_read(channel);
-}
-
+#define foreach_vmbus_pkt(pkt, channel) \
+ for (pkt = hv_pkt_iter_first(channel); pkt; \
+ pkt = hv_pkt_iter_next(channel, pkt))
#endif /* _HYPERV_H */