2 * Copyright (c) 2009, Microsoft Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, see <http://www.gnu.org/licenses/>.
17 * Haiyang Zhang <haiyangz@microsoft.com>
18 * Hank Janssen <hjanssen@microsoft.com>
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
22 #include <linux/kernel.h>
23 #include <linux/sched.h>
24 #include <linux/wait.h>
26 #include <linux/delay.h>
28 #include <linux/slab.h>
29 #include <linux/netdevice.h>
30 #include <linux/if_ether.h>
31 #include <linux/vmalloc.h>
32 #include <asm/sync_bitops.h>
34 #include "hyperv_net.h"
37 * Switch the data path from the synthetic interface to the VF
40 void netvsc_switch_datapath(struct net_device *ndev, bool vf)
42 struct net_device_context *net_device_ctx = netdev_priv(ndev);
43 struct hv_device *dev = net_device_ctx->device_ctx;
44 struct netvsc_device *nv_dev = net_device_ctx->nvdev;
45 struct nvsp_message *init_pkt = &nv_dev->channel_init_pkt;
47 memset(init_pkt, 0, sizeof(struct nvsp_message));
48 init_pkt->hdr.msg_type = NVSP_MSG4_TYPE_SWITCH_DATA_PATH;
50 init_pkt->msg.v4_msg.active_dp.active_datapath =
53 init_pkt->msg.v4_msg.active_dp.active_datapath =
54 NVSP_DATAPATH_SYNTHETIC;
56 vmbus_sendpacket(dev->channel, init_pkt,
57 sizeof(struct nvsp_message),
58 (unsigned long)init_pkt,
59 VM_PKT_DATA_INBAND, 0);
62 static struct netvsc_device *alloc_net_device(void)
64 struct netvsc_device *net_device;
66 net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
70 net_device->chan_table[0].mrc.buf
71 = vzalloc(NETVSC_RECVSLOT_MAX * sizeof(struct recv_comp_data));
73 init_waitqueue_head(&net_device->wait_drain);
74 net_device->destroy = false;
75 atomic_set(&net_device->open_cnt, 0);
76 net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
77 net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
78 init_completion(&net_device->channel_init_wait);
83 static void free_netvsc_device(struct netvsc_device *nvdev)
87 for (i = 0; i < VRSS_CHANNEL_MAX; i++)
88 vfree(nvdev->chan_table[i].mrc.buf);
94 static inline bool netvsc_channel_idle(const struct netvsc_device *net_device,
97 const struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
99 return atomic_read(&net_device->num_outstanding_recvs) == 0 &&
100 atomic_read(&nvchan->queue_sends) == 0;
103 static struct netvsc_device *get_outbound_net_device(struct hv_device *device)
105 struct netvsc_device *net_device = hv_device_to_netvsc_device(device);
107 if (net_device && net_device->destroy)
113 static void netvsc_destroy_buf(struct hv_device *device)
115 struct nvsp_message *revoke_packet;
116 struct net_device *ndev = hv_get_drvdata(device);
117 struct netvsc_device *net_device = net_device_to_netvsc_device(ndev);
121 * If we got a section count, it means we received a
122 * SendReceiveBufferComplete msg (ie sent
123 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
124 * to send a revoke msg here
126 if (net_device->recv_section_cnt) {
127 /* Send the revoke receive buffer */
128 revoke_packet = &net_device->revoke_packet;
129 memset(revoke_packet, 0, sizeof(struct nvsp_message));
131 revoke_packet->hdr.msg_type =
132 NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
133 revoke_packet->msg.v1_msg.
134 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
136 ret = vmbus_sendpacket(device->channel,
138 sizeof(struct nvsp_message),
139 (unsigned long)revoke_packet,
140 VM_PKT_DATA_INBAND, 0);
142 * If we failed here, we might as well return and
143 * have a leak rather than continue and a bugchk
146 netdev_err(ndev, "unable to send "
147 "revoke receive buffer to netvsp\n");
152 /* Teardown the gpadl on the vsp end */
153 if (net_device->recv_buf_gpadl_handle) {
154 ret = vmbus_teardown_gpadl(device->channel,
155 net_device->recv_buf_gpadl_handle);
157 /* If we failed here, we might as well return and have a leak
158 * rather than continue and a bugchk
162 "unable to teardown receive buffer's gpadl\n");
165 net_device->recv_buf_gpadl_handle = 0;
168 if (net_device->recv_buf) {
169 /* Free up the receive buffer */
170 vfree(net_device->recv_buf);
171 net_device->recv_buf = NULL;
174 if (net_device->recv_section) {
175 net_device->recv_section_cnt = 0;
176 kfree(net_device->recv_section);
177 net_device->recv_section = NULL;
180 /* Deal with the send buffer we may have setup.
181 * If we got a send section size, it means we received a
182 * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
183 * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
184 * to send a revoke msg here
186 if (net_device->send_section_size) {
187 /* Send the revoke receive buffer */
188 revoke_packet = &net_device->revoke_packet;
189 memset(revoke_packet, 0, sizeof(struct nvsp_message));
191 revoke_packet->hdr.msg_type =
192 NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
193 revoke_packet->msg.v1_msg.revoke_send_buf.id =
194 NETVSC_SEND_BUFFER_ID;
196 ret = vmbus_sendpacket(device->channel,
198 sizeof(struct nvsp_message),
199 (unsigned long)revoke_packet,
200 VM_PKT_DATA_INBAND, 0);
201 /* If we failed here, we might as well return and
202 * have a leak rather than continue and a bugchk
205 netdev_err(ndev, "unable to send "
206 "revoke send buffer to netvsp\n");
210 /* Teardown the gpadl on the vsp end */
211 if (net_device->send_buf_gpadl_handle) {
212 ret = vmbus_teardown_gpadl(device->channel,
213 net_device->send_buf_gpadl_handle);
215 /* If we failed here, we might as well return and have a leak
216 * rather than continue and a bugchk
220 "unable to teardown send buffer's gpadl\n");
223 net_device->send_buf_gpadl_handle = 0;
225 if (net_device->send_buf) {
226 /* Free up the send buffer */
227 vfree(net_device->send_buf);
228 net_device->send_buf = NULL;
230 kfree(net_device->send_section_map);
233 static int netvsc_init_buf(struct hv_device *device)
236 struct netvsc_device *net_device;
237 struct nvsp_message *init_packet;
238 struct net_device *ndev;
241 net_device = get_outbound_net_device(device);
244 ndev = hv_get_drvdata(device);
246 node = cpu_to_node(device->channel->target_cpu);
247 net_device->recv_buf = vzalloc_node(net_device->recv_buf_size, node);
248 if (!net_device->recv_buf)
249 net_device->recv_buf = vzalloc(net_device->recv_buf_size);
251 if (!net_device->recv_buf) {
252 netdev_err(ndev, "unable to allocate receive "
253 "buffer of size %d\n", net_device->recv_buf_size);
259 * Establish the gpadl handle for this buffer on this
260 * channel. Note: This call uses the vmbus connection rather
261 * than the channel to establish the gpadl handle.
263 ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
264 net_device->recv_buf_size,
265 &net_device->recv_buf_gpadl_handle);
268 "unable to establish receive buffer's gpadl\n");
272 /* Notify the NetVsp of the gpadl handle */
273 init_packet = &net_device->channel_init_pkt;
275 memset(init_packet, 0, sizeof(struct nvsp_message));
277 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
278 init_packet->msg.v1_msg.send_recv_buf.
279 gpadl_handle = net_device->recv_buf_gpadl_handle;
280 init_packet->msg.v1_msg.
281 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
283 /* Send the gpadl notification request */
284 ret = vmbus_sendpacket(device->channel, init_packet,
285 sizeof(struct nvsp_message),
286 (unsigned long)init_packet,
288 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
291 "unable to send receive buffer's gpadl to netvsp\n");
295 wait_for_completion(&net_device->channel_init_wait);
297 /* Check the response */
298 if (init_packet->msg.v1_msg.
299 send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
300 netdev_err(ndev, "Unable to complete receive buffer "
301 "initialization with NetVsp - status %d\n",
302 init_packet->msg.v1_msg.
303 send_recv_buf_complete.status);
308 /* Parse the response */
310 net_device->recv_section_cnt = init_packet->msg.
311 v1_msg.send_recv_buf_complete.num_sections;
313 net_device->recv_section = kmemdup(
314 init_packet->msg.v1_msg.send_recv_buf_complete.sections,
315 net_device->recv_section_cnt *
316 sizeof(struct nvsp_1_receive_buffer_section),
318 if (net_device->recv_section == NULL) {
324 * For 1st release, there should only be 1 section that represents the
325 * entire receive buffer
327 if (net_device->recv_section_cnt != 1 ||
328 net_device->recv_section->offset != 0) {
333 /* Now setup the send buffer.
335 net_device->send_buf = vzalloc_node(net_device->send_buf_size, node);
336 if (!net_device->send_buf)
337 net_device->send_buf = vzalloc(net_device->send_buf_size);
338 if (!net_device->send_buf) {
339 netdev_err(ndev, "unable to allocate send "
340 "buffer of size %d\n", net_device->send_buf_size);
345 /* Establish the gpadl handle for this buffer on this
346 * channel. Note: This call uses the vmbus connection rather
347 * than the channel to establish the gpadl handle.
349 ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
350 net_device->send_buf_size,
351 &net_device->send_buf_gpadl_handle);
354 "unable to establish send buffer's gpadl\n");
358 /* Notify the NetVsp of the gpadl handle */
359 init_packet = &net_device->channel_init_pkt;
360 memset(init_packet, 0, sizeof(struct nvsp_message));
361 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
362 init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
363 net_device->send_buf_gpadl_handle;
364 init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
366 /* Send the gpadl notification request */
367 ret = vmbus_sendpacket(device->channel, init_packet,
368 sizeof(struct nvsp_message),
369 (unsigned long)init_packet,
371 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
374 "unable to send send buffer's gpadl to netvsp\n");
378 wait_for_completion(&net_device->channel_init_wait);
380 /* Check the response */
381 if (init_packet->msg.v1_msg.
382 send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
383 netdev_err(ndev, "Unable to complete send buffer "
384 "initialization with NetVsp - status %d\n",
385 init_packet->msg.v1_msg.
386 send_send_buf_complete.status);
391 /* Parse the response */
392 net_device->send_section_size = init_packet->msg.
393 v1_msg.send_send_buf_complete.section_size;
395 /* Section count is simply the size divided by the section size.
397 net_device->send_section_cnt =
398 net_device->send_buf_size / net_device->send_section_size;
400 netdev_dbg(ndev, "Send section size: %d, Section count:%d\n",
401 net_device->send_section_size, net_device->send_section_cnt);
403 /* Setup state for managing the send buffer. */
404 net_device->map_words = DIV_ROUND_UP(net_device->send_section_cnt,
407 net_device->send_section_map = kcalloc(net_device->map_words,
408 sizeof(ulong), GFP_KERNEL);
409 if (net_device->send_section_map == NULL) {
417 netvsc_destroy_buf(device);
423 /* Negotiate NVSP protocol version */
424 static int negotiate_nvsp_ver(struct hv_device *device,
425 struct netvsc_device *net_device,
426 struct nvsp_message *init_packet,
429 struct net_device *ndev = hv_get_drvdata(device);
432 memset(init_packet, 0, sizeof(struct nvsp_message));
433 init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
434 init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
435 init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
437 /* Send the init request */
438 ret = vmbus_sendpacket(device->channel, init_packet,
439 sizeof(struct nvsp_message),
440 (unsigned long)init_packet,
442 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
447 wait_for_completion(&net_device->channel_init_wait);
449 if (init_packet->msg.init_msg.init_complete.status !=
453 if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
456 /* NVSPv2 or later: Send NDIS config */
457 memset(init_packet, 0, sizeof(struct nvsp_message));
458 init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
459 init_packet->msg.v2_msg.send_ndis_config.mtu = ndev->mtu + ETH_HLEN;
460 init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
462 if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5) {
463 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
465 /* Teaming bit is needed to receive link speed updates */
466 init_packet->msg.v2_msg.send_ndis_config.capability.teaming = 1;
469 ret = vmbus_sendpacket(device->channel, init_packet,
470 sizeof(struct nvsp_message),
471 (unsigned long)init_packet,
472 VM_PKT_DATA_INBAND, 0);
477 static int netvsc_connect_vsp(struct hv_device *device)
480 struct netvsc_device *net_device;
481 struct nvsp_message *init_packet;
483 const u32 ver_list[] = {
484 NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
485 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5 };
488 net_device = get_outbound_net_device(device);
492 init_packet = &net_device->channel_init_pkt;
494 /* Negotiate the latest NVSP protocol supported */
495 for (i = ARRAY_SIZE(ver_list) - 1; i >= 0; i--)
496 if (negotiate_nvsp_ver(device, net_device, init_packet,
498 net_device->nvsp_version = ver_list[i];
507 pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
509 /* Send the ndis version */
510 memset(init_packet, 0, sizeof(struct nvsp_message));
512 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
513 ndis_version = 0x00060001;
515 ndis_version = 0x0006001e;
517 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
518 init_packet->msg.v1_msg.
519 send_ndis_ver.ndis_major_ver =
520 (ndis_version & 0xFFFF0000) >> 16;
521 init_packet->msg.v1_msg.
522 send_ndis_ver.ndis_minor_ver =
523 ndis_version & 0xFFFF;
525 /* Send the init request */
526 ret = vmbus_sendpacket(device->channel, init_packet,
527 sizeof(struct nvsp_message),
528 (unsigned long)init_packet,
529 VM_PKT_DATA_INBAND, 0);
533 /* Post the big receive buffer to NetVSP */
534 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
535 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
537 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
538 net_device->send_buf_size = NETVSC_SEND_BUFFER_SIZE;
540 ret = netvsc_init_buf(device);
546 static void netvsc_disconnect_vsp(struct hv_device *device)
548 netvsc_destroy_buf(device);
552 * netvsc_device_remove - Callback when the root bus device is removed
554 void netvsc_device_remove(struct hv_device *device)
556 struct net_device *ndev = hv_get_drvdata(device);
557 struct net_device_context *net_device_ctx = netdev_priv(ndev);
558 struct netvsc_device *net_device = net_device_ctx->nvdev;
561 netvsc_disconnect_vsp(device);
563 net_device_ctx->nvdev = NULL;
566 * At this point, no one should be accessing net_device
569 netdev_dbg(ndev, "net device safe to remove\n");
571 /* Now, we can close the channel safely */
572 vmbus_close(device->channel);
574 for (i = 0; i < VRSS_CHANNEL_MAX; i++)
575 napi_disable(&net_device->chan_table[0].napi);
577 /* Release all resources */
578 free_netvsc_device(net_device);
581 #define RING_AVAIL_PERCENT_HIWATER 20
582 #define RING_AVAIL_PERCENT_LOWATER 10
585 * Get the percentage of available bytes to write in the ring.
586 * The return value is in range from 0 to 100.
588 static inline u32 hv_ringbuf_avail_percent(
589 struct hv_ring_buffer_info *ring_info)
591 u32 avail_read, avail_write;
593 hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
595 return avail_write * 100 / ring_info->ring_datasize;
598 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
601 sync_change_bit(index, net_device->send_section_map);
604 static void netvsc_send_tx_complete(struct netvsc_device *net_device,
605 struct vmbus_channel *incoming_channel,
606 struct hv_device *device,
607 const struct vmpacket_descriptor *desc)
609 struct sk_buff *skb = (struct sk_buff *)(unsigned long)desc->trans_id;
610 struct net_device *ndev = hv_get_drvdata(device);
611 struct net_device_context *net_device_ctx = netdev_priv(ndev);
612 struct vmbus_channel *channel = device->channel;
616 /* Notify the layer above us */
618 const struct hv_netvsc_packet *packet
619 = (struct hv_netvsc_packet *)skb->cb;
620 u32 send_index = packet->send_buf_index;
621 struct netvsc_stats *tx_stats;
623 if (send_index != NETVSC_INVALID_INDEX)
624 netvsc_free_send_slot(net_device, send_index);
625 q_idx = packet->q_idx;
626 channel = incoming_channel;
628 tx_stats = &net_device->chan_table[q_idx].tx_stats;
630 u64_stats_update_begin(&tx_stats->syncp);
631 tx_stats->packets += packet->total_packets;
632 tx_stats->bytes += packet->total_bytes;
633 u64_stats_update_end(&tx_stats->syncp);
635 dev_consume_skb_any(skb);
639 atomic_dec_return(&net_device->chan_table[q_idx].queue_sends);
641 if (net_device->destroy && queue_sends == 0)
642 wake_up(&net_device->wait_drain);
644 if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev, q_idx)) &&
645 !net_device_ctx->start_remove &&
646 (hv_ringbuf_avail_percent(&channel->outbound) > RING_AVAIL_PERCENT_HIWATER ||
648 netif_tx_wake_queue(netdev_get_tx_queue(ndev, q_idx));
651 static void netvsc_send_completion(struct netvsc_device *net_device,
652 struct vmbus_channel *incoming_channel,
653 struct hv_device *device,
654 const struct vmpacket_descriptor *desc)
656 struct nvsp_message *nvsp_packet = hv_pkt_data(desc);
657 struct net_device *ndev = hv_get_drvdata(device);
659 switch (nvsp_packet->hdr.msg_type) {
660 case NVSP_MSG_TYPE_INIT_COMPLETE:
661 case NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE:
662 case NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE:
663 case NVSP_MSG5_TYPE_SUBCHANNEL:
664 /* Copy the response back */
665 memcpy(&net_device->channel_init_pkt, nvsp_packet,
666 sizeof(struct nvsp_message));
667 complete(&net_device->channel_init_wait);
670 case NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE:
671 netvsc_send_tx_complete(net_device, incoming_channel,
677 "Unknown send completion type %d received!!\n",
678 nvsp_packet->hdr.msg_type);
682 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
684 unsigned long *map_addr = net_device->send_section_map;
687 for_each_clear_bit(i, map_addr, net_device->map_words) {
688 if (sync_test_and_set_bit(i, map_addr) == 0)
692 return NETVSC_INVALID_INDEX;
695 static u32 netvsc_copy_to_send_buf(struct netvsc_device *net_device,
696 unsigned int section_index,
698 struct hv_netvsc_packet *packet,
699 struct rndis_message *rndis_msg,
700 struct hv_page_buffer **pb,
703 char *start = net_device->send_buf;
704 char *dest = start + (section_index * net_device->send_section_size)
709 u32 remain = packet->total_data_buflen % net_device->pkt_align;
710 u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
711 packet->page_buf_cnt;
714 if (skb && skb->xmit_more && remain &&
715 !packet->cp_partial) {
716 padding = net_device->pkt_align - remain;
717 rndis_msg->msg_len += padding;
718 packet->total_data_buflen += padding;
721 for (i = 0; i < page_count; i++) {
722 char *src = phys_to_virt((*pb)[i].pfn << PAGE_SHIFT);
723 u32 offset = (*pb)[i].offset;
724 u32 len = (*pb)[i].len;
726 memcpy(dest, (src + offset), len);
732 memset(dest, 0, padding);
739 static inline int netvsc_send_pkt(
740 struct hv_device *device,
741 struct hv_netvsc_packet *packet,
742 struct netvsc_device *net_device,
743 struct hv_page_buffer **pb,
746 struct nvsp_message nvmsg;
747 struct netvsc_channel *nvchan
748 = &net_device->chan_table[packet->q_idx];
749 struct vmbus_channel *out_channel = nvchan->channel;
750 struct net_device *ndev = hv_get_drvdata(device);
751 struct netdev_queue *txq = netdev_get_tx_queue(ndev, packet->q_idx);
754 struct hv_page_buffer *pgbuf;
755 u32 ring_avail = hv_ringbuf_avail_percent(&out_channel->outbound);
757 nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
760 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 0;
762 /* 1 is RMC_CONTROL; */
763 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 1;
766 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
767 packet->send_buf_index;
768 if (packet->send_buf_index == NETVSC_INVALID_INDEX)
769 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
771 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size =
772 packet->total_data_buflen;
776 if (out_channel->rescind)
779 if (packet->page_buf_cnt) {
780 pgbuf = packet->cp_partial ? (*pb) +
781 packet->rmsg_pgcnt : (*pb);
782 ret = vmbus_sendpacket_pagebuffer_ctl(out_channel,
784 packet->page_buf_cnt,
786 sizeof(struct nvsp_message),
788 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
790 ret = vmbus_sendpacket_ctl(out_channel, &nvmsg,
791 sizeof(struct nvsp_message),
794 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
798 atomic_inc_return(&nvchan->queue_sends);
800 if (ring_avail < RING_AVAIL_PERCENT_LOWATER)
801 netif_tx_stop_queue(txq);
802 } else if (ret == -EAGAIN) {
803 netif_tx_stop_queue(txq);
804 if (atomic_read(&nvchan->queue_sends) < 1) {
805 netif_tx_wake_queue(txq);
809 netdev_err(ndev, "Unable to send packet %p ret %d\n",
816 /* Move packet out of multi send data (msd), and clear msd */
817 static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
818 struct sk_buff **msd_skb,
819 struct multi_send_data *msdp)
821 *msd_skb = msdp->skb;
822 *msd_send = msdp->pkt;
828 int netvsc_send(struct hv_device *device,
829 struct hv_netvsc_packet *packet,
830 struct rndis_message *rndis_msg,
831 struct hv_page_buffer **pb,
834 struct netvsc_device *net_device;
836 struct netvsc_channel *nvchan;
837 u32 pktlen = packet->total_data_buflen, msd_len = 0;
838 unsigned int section_index = NETVSC_INVALID_INDEX;
839 struct multi_send_data *msdp;
840 struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
841 struct sk_buff *msd_skb = NULL;
843 bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
845 net_device = get_outbound_net_device(device);
849 /* We may race with netvsc_connect_vsp()/netvsc_init_buf() and get
850 * here before the negotiation with the host is finished and
851 * send_section_map may not be allocated yet.
853 if (!net_device->send_section_map)
856 nvchan = &net_device->chan_table[packet->q_idx];
857 packet->send_buf_index = NETVSC_INVALID_INDEX;
858 packet->cp_partial = false;
860 /* Send control message directly without accessing msd (Multi-Send
861 * Data) field which may be changed during data packet processing.
868 /* batch packets in send buffer if possible */
871 msd_len = msdp->pkt->total_data_buflen;
873 try_batch = (skb != NULL) && msd_len > 0 && msdp->count <
876 if (try_batch && msd_len + pktlen + net_device->pkt_align <
877 net_device->send_section_size) {
878 section_index = msdp->pkt->send_buf_index;
880 } else if (try_batch && msd_len + packet->rmsg_size <
881 net_device->send_section_size) {
882 section_index = msdp->pkt->send_buf_index;
883 packet->cp_partial = true;
885 } else if ((skb != NULL) && pktlen + net_device->pkt_align <
886 net_device->send_section_size) {
887 section_index = netvsc_get_next_send_section(net_device);
888 if (section_index != NETVSC_INVALID_INDEX) {
889 move_pkt_msd(&msd_send, &msd_skb, msdp);
894 if (section_index != NETVSC_INVALID_INDEX) {
895 netvsc_copy_to_send_buf(net_device,
896 section_index, msd_len,
897 packet, rndis_msg, pb, skb);
899 packet->send_buf_index = section_index;
901 if (packet->cp_partial) {
902 packet->page_buf_cnt -= packet->rmsg_pgcnt;
903 packet->total_data_buflen = msd_len + packet->rmsg_size;
905 packet->page_buf_cnt = 0;
906 packet->total_data_buflen += msd_len;
910 packet->total_packets += msdp->pkt->total_packets;
911 packet->total_bytes += msdp->pkt->total_bytes;
915 dev_consume_skb_any(msdp->skb);
917 if (xmit_more && !packet->cp_partial) {
928 move_pkt_msd(&msd_send, &msd_skb, msdp);
933 int m_ret = netvsc_send_pkt(device, msd_send, net_device,
937 netvsc_free_send_slot(net_device,
938 msd_send->send_buf_index);
939 dev_kfree_skb_any(msd_skb);
945 ret = netvsc_send_pkt(device, cur_send, net_device, pb, skb);
947 if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
948 netvsc_free_send_slot(net_device, section_index);
953 static int netvsc_send_recv_completion(struct vmbus_channel *channel,
954 u64 transaction_id, u32 status)
956 struct nvsp_message recvcompMessage;
959 recvcompMessage.hdr.msg_type =
960 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
962 recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status = status;
964 /* Send the completion */
965 ret = vmbus_sendpacket(channel, &recvcompMessage,
966 sizeof(struct nvsp_message_header) + sizeof(u32),
967 transaction_id, VM_PKT_COMP, 0);
972 static inline void count_recv_comp_slot(struct netvsc_device *nvdev, u16 q_idx,
973 u32 *filled, u32 *avail)
975 struct multi_recv_comp *mrc = &nvdev->chan_table[q_idx].mrc;
976 u32 first = mrc->first;
977 u32 next = mrc->next;
979 *filled = (first > next) ? NETVSC_RECVSLOT_MAX - first + next :
982 *avail = NETVSC_RECVSLOT_MAX - *filled - 1;
985 /* Read the first filled slot, no change to index */
986 static inline struct recv_comp_data *read_recv_comp_slot(struct netvsc_device
989 struct multi_recv_comp *mrc = &nvdev->chan_table[q_idx].mrc;
992 if (unlikely(!mrc->buf))
995 count_recv_comp_slot(nvdev, q_idx, &filled, &avail);
999 return mrc->buf + mrc->first * sizeof(struct recv_comp_data);
1002 /* Put the first filled slot back to available pool */
1003 static inline void put_recv_comp_slot(struct netvsc_device *nvdev, u16 q_idx)
1005 struct multi_recv_comp *mrc = &nvdev->chan_table[q_idx].mrc;
1008 mrc->first = (mrc->first + 1) % NETVSC_RECVSLOT_MAX;
1010 num_recv = atomic_dec_return(&nvdev->num_outstanding_recvs);
1012 if (nvdev->destroy && num_recv == 0)
1013 wake_up(&nvdev->wait_drain);
1016 /* Check and send pending recv completions */
1017 static void netvsc_chk_recv_comp(struct netvsc_device *nvdev,
1018 struct vmbus_channel *channel, u16 q_idx)
1020 struct recv_comp_data *rcd;
1024 rcd = read_recv_comp_slot(nvdev, q_idx);
1028 ret = netvsc_send_recv_completion(channel, rcd->tid,
1033 put_recv_comp_slot(nvdev, q_idx);
1037 #define NETVSC_RCD_WATERMARK 80
1039 /* Get next available slot */
1040 static inline struct recv_comp_data *get_recv_comp_slot(
1041 struct netvsc_device *nvdev, struct vmbus_channel *channel, u16 q_idx)
1043 struct multi_recv_comp *mrc = &nvdev->chan_table[q_idx].mrc;
1044 u32 filled, avail, next;
1045 struct recv_comp_data *rcd;
1047 if (unlikely(!nvdev->recv_section))
1050 if (unlikely(!mrc->buf))
1053 if (atomic_read(&nvdev->num_outstanding_recvs) >
1054 nvdev->recv_section->num_sub_allocs * NETVSC_RCD_WATERMARK / 100)
1055 netvsc_chk_recv_comp(nvdev, channel, q_idx);
1057 count_recv_comp_slot(nvdev, q_idx, &filled, &avail);
1062 rcd = mrc->buf + next * sizeof(struct recv_comp_data);
1063 mrc->next = (next + 1) % NETVSC_RECVSLOT_MAX;
1065 atomic_inc(&nvdev->num_outstanding_recvs);
1070 static int netvsc_receive(struct net_device *ndev,
1071 struct netvsc_device *net_device,
1072 struct net_device_context *net_device_ctx,
1073 struct hv_device *device,
1074 struct vmbus_channel *channel,
1075 const struct vmpacket_descriptor *desc,
1076 struct nvsp_message *nvsp)
1078 const struct vmtransfer_page_packet_header *vmxferpage_packet
1079 = container_of(desc, const struct vmtransfer_page_packet_header, d);
1080 u16 q_idx = channel->offermsg.offer.sub_channel_index;
1081 char *recv_buf = net_device->recv_buf;
1082 u32 status = NVSP_STAT_SUCCESS;
1087 /* Make sure this is a valid nvsp packet */
1088 if (unlikely(nvsp->hdr.msg_type != NVSP_MSG1_TYPE_SEND_RNDIS_PKT)) {
1089 netif_err(net_device_ctx, rx_err, ndev,
1090 "Unknown nvsp packet type received %u\n",
1091 nvsp->hdr.msg_type);
1095 if (unlikely(vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID)) {
1096 netif_err(net_device_ctx, rx_err, ndev,
1097 "Invalid xfer page set id - expecting %x got %x\n",
1098 NETVSC_RECEIVE_BUFFER_ID,
1099 vmxferpage_packet->xfer_pageset_id);
1103 count = vmxferpage_packet->range_cnt;
1105 /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1106 for (i = 0; i < count; i++) {
1107 void *data = recv_buf
1108 + vmxferpage_packet->ranges[i].byte_offset;
1109 u32 buflen = vmxferpage_packet->ranges[i].byte_count;
1111 /* Pass it to the upper layer */
1112 status = rndis_filter_receive(ndev, net_device, device,
1113 channel, data, buflen);
1116 if (net_device->chan_table[q_idx].mrc.buf) {
1117 struct recv_comp_data *rcd;
1119 rcd = get_recv_comp_slot(net_device, channel, q_idx);
1121 rcd->tid = vmxferpage_packet->d.trans_id;
1122 rcd->status = status;
1124 netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
1125 q_idx, vmxferpage_packet->d.trans_id);
1128 ret = netvsc_send_recv_completion(channel,
1129 vmxferpage_packet->d.trans_id,
1132 netdev_err(ndev, "Recv_comp q:%hd, tid:%llx, err:%d\n",
1133 q_idx, vmxferpage_packet->d.trans_id, ret);
1138 static void netvsc_send_table(struct hv_device *hdev,
1139 struct nvsp_message *nvmsg)
1141 struct netvsc_device *nvscdev;
1142 struct net_device *ndev = hv_get_drvdata(hdev);
1146 nvscdev = get_outbound_net_device(hdev);
1150 count = nvmsg->msg.v5_msg.send_table.count;
1151 if (count != VRSS_SEND_TAB_SIZE) {
1152 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1156 tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
1157 nvmsg->msg.v5_msg.send_table.offset);
1159 for (i = 0; i < count; i++)
1160 nvscdev->send_table[i] = tab[i];
1163 static void netvsc_send_vf(struct net_device_context *net_device_ctx,
1164 struct nvsp_message *nvmsg)
1166 net_device_ctx->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1167 net_device_ctx->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1170 static inline void netvsc_receive_inband(struct hv_device *hdev,
1171 struct net_device_context *net_device_ctx,
1172 struct nvsp_message *nvmsg)
1174 switch (nvmsg->hdr.msg_type) {
1175 case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1176 netvsc_send_table(hdev, nvmsg);
1179 case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1180 netvsc_send_vf(net_device_ctx, nvmsg);
1185 static int netvsc_process_raw_pkt(struct hv_device *device,
1186 struct vmbus_channel *channel,
1187 struct netvsc_device *net_device,
1188 struct net_device *ndev,
1190 const struct vmpacket_descriptor *desc)
1192 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1193 struct nvsp_message *nvmsg = hv_pkt_data(desc);
1195 switch (desc->type) {
1197 netvsc_send_completion(net_device, channel, device, desc);
1200 case VM_PKT_DATA_USING_XFER_PAGES:
1201 return netvsc_receive(ndev, net_device, net_device_ctx,
1202 device, channel, desc, nvmsg);
1205 case VM_PKT_DATA_INBAND:
1206 netvsc_receive_inband(device, net_device_ctx, nvmsg);
1210 netdev_err(ndev, "unhandled packet type %d, tid %llx\n",
1211 desc->type, request_id);
1218 static struct hv_device *netvsc_channel_to_device(struct vmbus_channel *channel)
1220 struct vmbus_channel *primary = channel->primary_channel;
1222 return primary ? primary->device_obj : channel->device_obj;
1225 int netvsc_poll(struct napi_struct *napi, int budget)
1227 struct netvsc_channel *nvchan
1228 = container_of(napi, struct netvsc_channel, napi);
1229 struct vmbus_channel *channel = nvchan->channel;
1230 struct hv_device *device = netvsc_channel_to_device(channel);
1231 u16 q_idx = channel->offermsg.offer.sub_channel_index;
1232 struct net_device *ndev = hv_get_drvdata(device);
1233 struct netvsc_device *net_device = net_device_to_netvsc_device(ndev);
1234 const struct vmpacket_descriptor *desc;
1237 desc = hv_pkt_iter_first(channel);
1241 count = netvsc_process_raw_pkt(device, channel, net_device,
1242 ndev, desc->trans_id, desc);
1244 desc = __hv_pkt_iter_next(channel, desc);
1246 /* If receive packet budget is exhausted, reschedule */
1247 if (work_done >= budget) {
1252 hv_pkt_iter_close(channel);
1254 /* If ring is empty and NAPI is not doing polling */
1255 if (work_done < budget &&
1256 napi_complete_done(napi, work_done) &&
1257 hv_end_read(&channel->inbound) != 0)
1258 napi_reschedule(napi);
1260 netvsc_chk_recv_comp(net_device, channel, q_idx);
1264 void netvsc_channel_cb(void *context)
1266 struct vmbus_channel *channel = context;
1267 struct hv_device *device = netvsc_channel_to_device(channel);
1268 u16 q_idx = channel->offermsg.offer.sub_channel_index;
1269 struct netvsc_device *net_device;
1270 struct net_device *ndev;
1272 ndev = hv_get_drvdata(device);
1273 if (unlikely(!ndev))
1276 net_device = net_device_to_netvsc_device(ndev);
1277 if (unlikely(net_device->destroy) &&
1278 netvsc_channel_idle(net_device, q_idx))
1281 /* disable interupts from host */
1282 hv_begin_read(&channel->inbound);
1283 napi_schedule(&net_device->chan_table[q_idx].napi);
1287 * netvsc_device_add - Callback when the device belonging to this
1290 int netvsc_device_add(struct hv_device *device,
1291 const struct netvsc_device_info *device_info)
1294 int ring_size = device_info->ring_size;
1295 struct netvsc_device *net_device;
1296 struct net_device *ndev = hv_get_drvdata(device);
1297 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1299 net_device = alloc_net_device();
1303 net_device->ring_size = ring_size;
1305 /* Because the device uses NAPI, all the interrupt batching and
1306 * control is done via Net softirq, not the channel handling
1308 set_channel_read_mode(device->channel, HV_CALL_ISR);
1310 /* Open the channel */
1311 ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
1312 ring_size * PAGE_SIZE, NULL, 0,
1313 netvsc_channel_cb, device->channel);
1316 netdev_err(ndev, "unable to open channel: %d\n", ret);
1320 /* Channel is opened */
1321 netdev_dbg(ndev, "hv_netvsc channel opened successfully\n");
1323 /* If we're reopening the device we may have multiple queues, fill the
1324 * chn_table with the default channel to use it before subchannels are
1327 for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
1328 struct netvsc_channel *nvchan = &net_device->chan_table[i];
1330 nvchan->channel = device->channel;
1331 netif_napi_add(ndev, &nvchan->napi,
1332 netvsc_poll, NAPI_POLL_WEIGHT);
1335 /* Enable NAPI handler for init callbacks */
1336 napi_enable(&net_device->chan_table[0].napi);
1338 /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1343 net_device_ctx->nvdev = net_device;
1345 /* Connect with the NetVsp */
1346 ret = netvsc_connect_vsp(device);
1349 "unable to connect to NetVSP - %d\n", ret);
1356 napi_disable(&net_device->chan_table[0].napi);
1358 /* Now, we can close the channel safely */
1359 vmbus_close(device->channel);
1362 free_netvsc_device(net_device);