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 <linux/rtnetlink.h>
33 #include <linux/prefetch.h>
35 #include <asm/sync_bitops.h>
37 #include "hyperv_net.h"
38 #include "netvsc_trace.h"
41 * Switch the data path from the synthetic interface to the VF
44 void netvsc_switch_datapath(struct net_device *ndev, bool vf)
46 struct net_device_context *net_device_ctx = netdev_priv(ndev);
47 struct hv_device *dev = net_device_ctx->device_ctx;
48 struct netvsc_device *nv_dev = rtnl_dereference(net_device_ctx->nvdev);
49 struct nvsp_message *init_pkt = &nv_dev->channel_init_pkt;
51 memset(init_pkt, 0, sizeof(struct nvsp_message));
52 init_pkt->hdr.msg_type = NVSP_MSG4_TYPE_SWITCH_DATA_PATH;
54 init_pkt->msg.v4_msg.active_dp.active_datapath =
57 init_pkt->msg.v4_msg.active_dp.active_datapath =
58 NVSP_DATAPATH_SYNTHETIC;
60 trace_nvsp_send(ndev, init_pkt);
62 vmbus_sendpacket(dev->channel, init_pkt,
63 sizeof(struct nvsp_message),
64 (unsigned long)init_pkt,
65 VM_PKT_DATA_INBAND, 0);
68 /* Worker to setup sub channels on initial setup
69 * Initial hotplug event occurs in softirq context
70 * and can't wait for channels.
72 static void netvsc_subchan_work(struct work_struct *w)
74 struct netvsc_device *nvdev =
75 container_of(w, struct netvsc_device, subchan_work);
76 struct rndis_device *rdev;
79 /* Avoid deadlock with device removal already under RTNL */
80 if (!rtnl_trylock()) {
85 rdev = nvdev->extension;
87 ret = rndis_set_subchannel(rdev->ndev, nvdev, NULL);
89 netif_device_attach(rdev->ndev);
91 /* fallback to only primary channel */
92 for (i = 1; i < nvdev->num_chn; i++)
93 netif_napi_del(&nvdev->chan_table[i].napi);
103 static struct netvsc_device *alloc_net_device(void)
105 struct netvsc_device *net_device;
107 net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
111 init_waitqueue_head(&net_device->wait_drain);
112 net_device->destroy = false;
113 net_device->tx_disable = false;
115 net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
116 net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
118 init_completion(&net_device->channel_init_wait);
119 init_waitqueue_head(&net_device->subchan_open);
120 INIT_WORK(&net_device->subchan_work, netvsc_subchan_work);
125 static void free_netvsc_device(struct rcu_head *head)
127 struct netvsc_device *nvdev
128 = container_of(head, struct netvsc_device, rcu);
131 kfree(nvdev->extension);
132 vfree(nvdev->recv_buf);
133 vfree(nvdev->send_buf);
134 kfree(nvdev->send_section_map);
136 for (i = 0; i < VRSS_CHANNEL_MAX; i++)
137 vfree(nvdev->chan_table[i].mrc.slots);
142 static void free_netvsc_device_rcu(struct netvsc_device *nvdev)
144 call_rcu(&nvdev->rcu, free_netvsc_device);
147 static void netvsc_revoke_recv_buf(struct hv_device *device,
148 struct netvsc_device *net_device,
149 struct net_device *ndev)
151 struct nvsp_message *revoke_packet;
155 * If we got a section count, it means we received a
156 * SendReceiveBufferComplete msg (ie sent
157 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
158 * to send a revoke msg here
160 if (net_device->recv_section_cnt) {
161 /* Send the revoke receive buffer */
162 revoke_packet = &net_device->revoke_packet;
163 memset(revoke_packet, 0, sizeof(struct nvsp_message));
165 revoke_packet->hdr.msg_type =
166 NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
167 revoke_packet->msg.v1_msg.
168 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
170 trace_nvsp_send(ndev, revoke_packet);
172 ret = vmbus_sendpacket(device->channel,
174 sizeof(struct nvsp_message),
175 (unsigned long)revoke_packet,
176 VM_PKT_DATA_INBAND, 0);
177 /* If the failure is because the channel is rescinded;
178 * ignore the failure since we cannot send on a rescinded
179 * channel. This would allow us to properly cleanup
180 * even when the channel is rescinded.
182 if (device->channel->rescind)
185 * If we failed here, we might as well return and
186 * have a leak rather than continue and a bugchk
189 netdev_err(ndev, "unable to send "
190 "revoke receive buffer to netvsp\n");
193 net_device->recv_section_cnt = 0;
197 static void netvsc_revoke_send_buf(struct hv_device *device,
198 struct netvsc_device *net_device,
199 struct net_device *ndev)
201 struct nvsp_message *revoke_packet;
204 /* Deal with the send buffer we may have setup.
205 * If we got a send section size, it means we received a
206 * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
207 * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
208 * to send a revoke msg here
210 if (net_device->send_section_cnt) {
211 /* Send the revoke receive buffer */
212 revoke_packet = &net_device->revoke_packet;
213 memset(revoke_packet, 0, sizeof(struct nvsp_message));
215 revoke_packet->hdr.msg_type =
216 NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
217 revoke_packet->msg.v1_msg.revoke_send_buf.id =
218 NETVSC_SEND_BUFFER_ID;
220 trace_nvsp_send(ndev, revoke_packet);
222 ret = vmbus_sendpacket(device->channel,
224 sizeof(struct nvsp_message),
225 (unsigned long)revoke_packet,
226 VM_PKT_DATA_INBAND, 0);
228 /* If the failure is because the channel is rescinded;
229 * ignore the failure since we cannot send on a rescinded
230 * channel. This would allow us to properly cleanup
231 * even when the channel is rescinded.
233 if (device->channel->rescind)
236 /* If we failed here, we might as well return and
237 * have a leak rather than continue and a bugchk
240 netdev_err(ndev, "unable to send "
241 "revoke send buffer to netvsp\n");
244 net_device->send_section_cnt = 0;
248 static void netvsc_teardown_recv_gpadl(struct hv_device *device,
249 struct netvsc_device *net_device,
250 struct net_device *ndev)
254 if (net_device->recv_buf_gpadl_handle) {
255 ret = vmbus_teardown_gpadl(device->channel,
256 net_device->recv_buf_gpadl_handle);
258 /* If we failed here, we might as well return and have a leak
259 * rather than continue and a bugchk
263 "unable to teardown receive buffer's gpadl\n");
266 net_device->recv_buf_gpadl_handle = 0;
270 static void netvsc_teardown_send_gpadl(struct hv_device *device,
271 struct netvsc_device *net_device,
272 struct net_device *ndev)
276 if (net_device->send_buf_gpadl_handle) {
277 ret = vmbus_teardown_gpadl(device->channel,
278 net_device->send_buf_gpadl_handle);
280 /* If we failed here, we might as well return and have a leak
281 * rather than continue and a bugchk
285 "unable to teardown send buffer's gpadl\n");
288 net_device->send_buf_gpadl_handle = 0;
292 int netvsc_alloc_recv_comp_ring(struct netvsc_device *net_device, u32 q_idx)
294 struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
295 int node = cpu_to_node(nvchan->channel->target_cpu);
298 size = net_device->recv_completion_cnt * sizeof(struct recv_comp_data);
299 nvchan->mrc.slots = vzalloc_node(size, node);
300 if (!nvchan->mrc.slots)
301 nvchan->mrc.slots = vzalloc(size);
303 return nvchan->mrc.slots ? 0 : -ENOMEM;
306 static int netvsc_init_buf(struct hv_device *device,
307 struct netvsc_device *net_device,
308 const struct netvsc_device_info *device_info)
310 struct nvsp_1_message_send_receive_buffer_complete *resp;
311 struct net_device *ndev = hv_get_drvdata(device);
312 struct nvsp_message *init_packet;
313 unsigned int buf_size;
317 /* Get receive buffer area. */
318 buf_size = device_info->recv_sections * device_info->recv_section_size;
319 buf_size = roundup(buf_size, PAGE_SIZE);
321 /* Legacy hosts only allow smaller receive buffer */
322 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
323 buf_size = min_t(unsigned int, buf_size,
324 NETVSC_RECEIVE_BUFFER_SIZE_LEGACY);
326 net_device->recv_buf = vzalloc(buf_size);
327 if (!net_device->recv_buf) {
329 "unable to allocate receive buffer of size %u\n",
335 net_device->recv_buf_size = buf_size;
338 * Establish the gpadl handle for this buffer on this
339 * channel. Note: This call uses the vmbus connection rather
340 * than the channel to establish the gpadl handle.
342 ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
344 &net_device->recv_buf_gpadl_handle);
347 "unable to establish receive buffer's gpadl\n");
351 /* Notify the NetVsp of the gpadl handle */
352 init_packet = &net_device->channel_init_pkt;
353 memset(init_packet, 0, sizeof(struct nvsp_message));
354 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
355 init_packet->msg.v1_msg.send_recv_buf.
356 gpadl_handle = net_device->recv_buf_gpadl_handle;
357 init_packet->msg.v1_msg.
358 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
360 trace_nvsp_send(ndev, init_packet);
362 /* Send the gpadl notification request */
363 ret = vmbus_sendpacket(device->channel, init_packet,
364 sizeof(struct nvsp_message),
365 (unsigned long)init_packet,
367 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
370 "unable to send receive buffer's gpadl to netvsp\n");
374 wait_for_completion(&net_device->channel_init_wait);
376 /* Check the response */
377 resp = &init_packet->msg.v1_msg.send_recv_buf_complete;
378 if (resp->status != NVSP_STAT_SUCCESS) {
380 "Unable to complete receive buffer initialization with NetVsp - status %d\n",
386 /* Parse the response */
387 netdev_dbg(ndev, "Receive sections: %u sub_allocs: size %u count: %u\n",
388 resp->num_sections, resp->sections[0].sub_alloc_size,
389 resp->sections[0].num_sub_allocs);
391 /* There should only be one section for the entire receive buffer */
392 if (resp->num_sections != 1 || resp->sections[0].offset != 0) {
397 net_device->recv_section_size = resp->sections[0].sub_alloc_size;
398 net_device->recv_section_cnt = resp->sections[0].num_sub_allocs;
400 /* Setup receive completion ring */
401 net_device->recv_completion_cnt
402 = round_up(net_device->recv_section_cnt + 1,
403 PAGE_SIZE / sizeof(u64));
404 ret = netvsc_alloc_recv_comp_ring(net_device, 0);
408 /* Now setup the send buffer. */
409 buf_size = device_info->send_sections * device_info->send_section_size;
410 buf_size = round_up(buf_size, PAGE_SIZE);
412 net_device->send_buf = vzalloc(buf_size);
413 if (!net_device->send_buf) {
414 netdev_err(ndev, "unable to allocate send buffer of size %u\n",
420 /* Establish the gpadl handle for this buffer on this
421 * channel. Note: This call uses the vmbus connection rather
422 * than the channel to establish the gpadl handle.
424 ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
426 &net_device->send_buf_gpadl_handle);
429 "unable to establish send buffer's gpadl\n");
433 /* Notify the NetVsp of the gpadl handle */
434 init_packet = &net_device->channel_init_pkt;
435 memset(init_packet, 0, sizeof(struct nvsp_message));
436 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
437 init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
438 net_device->send_buf_gpadl_handle;
439 init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
441 trace_nvsp_send(ndev, init_packet);
443 /* Send the gpadl notification request */
444 ret = vmbus_sendpacket(device->channel, init_packet,
445 sizeof(struct nvsp_message),
446 (unsigned long)init_packet,
448 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
451 "unable to send send buffer's gpadl to netvsp\n");
455 wait_for_completion(&net_device->channel_init_wait);
457 /* Check the response */
458 if (init_packet->msg.v1_msg.
459 send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
460 netdev_err(ndev, "Unable to complete send buffer "
461 "initialization with NetVsp - status %d\n",
462 init_packet->msg.v1_msg.
463 send_send_buf_complete.status);
468 /* Parse the response */
469 net_device->send_section_size = init_packet->msg.
470 v1_msg.send_send_buf_complete.section_size;
472 /* Section count is simply the size divided by the section size. */
473 net_device->send_section_cnt = buf_size / net_device->send_section_size;
475 netdev_dbg(ndev, "Send section size: %d, Section count:%d\n",
476 net_device->send_section_size, net_device->send_section_cnt);
478 /* Setup state for managing the send buffer. */
479 map_words = DIV_ROUND_UP(net_device->send_section_cnt, BITS_PER_LONG);
481 net_device->send_section_map = kcalloc(map_words, sizeof(ulong), GFP_KERNEL);
482 if (net_device->send_section_map == NULL) {
490 netvsc_revoke_recv_buf(device, net_device, ndev);
491 netvsc_revoke_send_buf(device, net_device, ndev);
492 netvsc_teardown_recv_gpadl(device, net_device, ndev);
493 netvsc_teardown_send_gpadl(device, net_device, ndev);
499 /* Negotiate NVSP protocol version */
500 static int negotiate_nvsp_ver(struct hv_device *device,
501 struct netvsc_device *net_device,
502 struct nvsp_message *init_packet,
505 struct net_device *ndev = hv_get_drvdata(device);
508 memset(init_packet, 0, sizeof(struct nvsp_message));
509 init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
510 init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
511 init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
512 trace_nvsp_send(ndev, init_packet);
514 /* Send the init request */
515 ret = vmbus_sendpacket(device->channel, init_packet,
516 sizeof(struct nvsp_message),
517 (unsigned long)init_packet,
519 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
524 wait_for_completion(&net_device->channel_init_wait);
526 if (init_packet->msg.init_msg.init_complete.status !=
530 if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
533 /* NVSPv2 or later: Send NDIS config */
534 memset(init_packet, 0, sizeof(struct nvsp_message));
535 init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
536 init_packet->msg.v2_msg.send_ndis_config.mtu = ndev->mtu + ETH_HLEN;
537 init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
539 if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5) {
540 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
542 /* Teaming bit is needed to receive link speed updates */
543 init_packet->msg.v2_msg.send_ndis_config.capability.teaming = 1;
546 if (nvsp_ver >= NVSP_PROTOCOL_VERSION_61)
547 init_packet->msg.v2_msg.send_ndis_config.capability.rsc = 1;
549 trace_nvsp_send(ndev, init_packet);
551 ret = vmbus_sendpacket(device->channel, init_packet,
552 sizeof(struct nvsp_message),
553 (unsigned long)init_packet,
554 VM_PKT_DATA_INBAND, 0);
559 static int netvsc_connect_vsp(struct hv_device *device,
560 struct netvsc_device *net_device,
561 const struct netvsc_device_info *device_info)
563 struct net_device *ndev = hv_get_drvdata(device);
564 static const u32 ver_list[] = {
565 NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
566 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5,
567 NVSP_PROTOCOL_VERSION_6, NVSP_PROTOCOL_VERSION_61
569 struct nvsp_message *init_packet;
570 int ndis_version, i, ret;
572 init_packet = &net_device->channel_init_pkt;
574 /* Negotiate the latest NVSP protocol supported */
575 for (i = ARRAY_SIZE(ver_list) - 1; i >= 0; i--)
576 if (negotiate_nvsp_ver(device, net_device, init_packet,
578 net_device->nvsp_version = ver_list[i];
587 pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
589 /* Send the ndis version */
590 memset(init_packet, 0, sizeof(struct nvsp_message));
592 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
593 ndis_version = 0x00060001;
595 ndis_version = 0x0006001e;
597 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
598 init_packet->msg.v1_msg.
599 send_ndis_ver.ndis_major_ver =
600 (ndis_version & 0xFFFF0000) >> 16;
601 init_packet->msg.v1_msg.
602 send_ndis_ver.ndis_minor_ver =
603 ndis_version & 0xFFFF;
605 trace_nvsp_send(ndev, init_packet);
607 /* Send the init request */
608 ret = vmbus_sendpacket(device->channel, init_packet,
609 sizeof(struct nvsp_message),
610 (unsigned long)init_packet,
611 VM_PKT_DATA_INBAND, 0);
616 ret = netvsc_init_buf(device, net_device, device_info);
623 * netvsc_device_remove - Callback when the root bus device is removed
625 void netvsc_device_remove(struct hv_device *device)
627 struct net_device *ndev = hv_get_drvdata(device);
628 struct net_device_context *net_device_ctx = netdev_priv(ndev);
629 struct netvsc_device *net_device
630 = rtnl_dereference(net_device_ctx->nvdev);
634 * Revoke receive buffer. If host is pre-Win2016 then tear down
635 * receive buffer GPADL. Do the same for send buffer.
637 netvsc_revoke_recv_buf(device, net_device, ndev);
638 if (vmbus_proto_version < VERSION_WIN10)
639 netvsc_teardown_recv_gpadl(device, net_device, ndev);
641 netvsc_revoke_send_buf(device, net_device, ndev);
642 if (vmbus_proto_version < VERSION_WIN10)
643 netvsc_teardown_send_gpadl(device, net_device, ndev);
645 RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
647 /* And disassociate NAPI context from device */
648 for (i = 0; i < net_device->num_chn; i++)
649 netif_napi_del(&net_device->chan_table[i].napi);
652 * At this point, no one should be accessing net_device
655 netdev_dbg(ndev, "net device safe to remove\n");
657 /* Now, we can close the channel safely */
658 vmbus_close(device->channel);
661 * If host is Win2016 or higher then we do the GPADL tear down
662 * here after VMBus is closed.
664 if (vmbus_proto_version >= VERSION_WIN10) {
665 netvsc_teardown_recv_gpadl(device, net_device, ndev);
666 netvsc_teardown_send_gpadl(device, net_device, ndev);
669 /* Release all resources */
670 free_netvsc_device_rcu(net_device);
673 #define RING_AVAIL_PERCENT_HIWATER 20
674 #define RING_AVAIL_PERCENT_LOWATER 10
676 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
679 sync_change_bit(index, net_device->send_section_map);
682 static void netvsc_send_tx_complete(struct net_device *ndev,
683 struct netvsc_device *net_device,
684 struct vmbus_channel *channel,
685 const struct vmpacket_descriptor *desc,
688 struct sk_buff *skb = (struct sk_buff *)(unsigned long)desc->trans_id;
689 struct net_device_context *ndev_ctx = netdev_priv(ndev);
693 /* Notify the layer above us */
695 const struct hv_netvsc_packet *packet
696 = (struct hv_netvsc_packet *)skb->cb;
697 u32 send_index = packet->send_buf_index;
698 struct netvsc_stats *tx_stats;
700 if (send_index != NETVSC_INVALID_INDEX)
701 netvsc_free_send_slot(net_device, send_index);
702 q_idx = packet->q_idx;
704 tx_stats = &net_device->chan_table[q_idx].tx_stats;
706 u64_stats_update_begin(&tx_stats->syncp);
707 tx_stats->packets += packet->total_packets;
708 tx_stats->bytes += packet->total_bytes;
709 u64_stats_update_end(&tx_stats->syncp);
711 napi_consume_skb(skb, budget);
715 atomic_dec_return(&net_device->chan_table[q_idx].queue_sends);
717 if (unlikely(net_device->destroy)) {
718 if (queue_sends == 0)
719 wake_up(&net_device->wait_drain);
721 struct netdev_queue *txq = netdev_get_tx_queue(ndev, q_idx);
723 if (netif_tx_queue_stopped(txq) && !net_device->tx_disable &&
724 (hv_get_avail_to_write_percent(&channel->outbound) >
725 RING_AVAIL_PERCENT_HIWATER || queue_sends < 1)) {
726 netif_tx_wake_queue(txq);
727 ndev_ctx->eth_stats.wake_queue++;
732 static void netvsc_send_completion(struct net_device *ndev,
733 struct netvsc_device *net_device,
734 struct vmbus_channel *incoming_channel,
735 const struct vmpacket_descriptor *desc,
738 const struct nvsp_message *nvsp_packet = hv_pkt_data(desc);
740 switch (nvsp_packet->hdr.msg_type) {
741 case NVSP_MSG_TYPE_INIT_COMPLETE:
742 case NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE:
743 case NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE:
744 case NVSP_MSG5_TYPE_SUBCHANNEL:
745 /* Copy the response back */
746 memcpy(&net_device->channel_init_pkt, nvsp_packet,
747 sizeof(struct nvsp_message));
748 complete(&net_device->channel_init_wait);
751 case NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE:
752 netvsc_send_tx_complete(ndev, net_device, incoming_channel,
758 "Unknown send completion type %d received!!\n",
759 nvsp_packet->hdr.msg_type);
763 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
765 unsigned long *map_addr = net_device->send_section_map;
768 for_each_clear_bit(i, map_addr, net_device->send_section_cnt) {
769 if (sync_test_and_set_bit(i, map_addr) == 0)
773 return NETVSC_INVALID_INDEX;
776 static void netvsc_copy_to_send_buf(struct netvsc_device *net_device,
777 unsigned int section_index,
779 struct hv_netvsc_packet *packet,
780 struct rndis_message *rndis_msg,
781 struct hv_page_buffer *pb,
784 char *start = net_device->send_buf;
785 char *dest = start + (section_index * net_device->send_section_size)
789 u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
790 packet->page_buf_cnt;
794 remain = packet->total_data_buflen & (net_device->pkt_align - 1);
795 if (xmit_more && remain) {
796 padding = net_device->pkt_align - remain;
797 rndis_msg->msg_len += padding;
798 packet->total_data_buflen += padding;
801 for (i = 0; i < page_count; i++) {
802 char *src = phys_to_virt(pb[i].pfn << PAGE_SHIFT);
803 u32 offset = pb[i].offset;
806 memcpy(dest, (src + offset), len);
811 memset(dest, 0, padding);
814 static inline int netvsc_send_pkt(
815 struct hv_device *device,
816 struct hv_netvsc_packet *packet,
817 struct netvsc_device *net_device,
818 struct hv_page_buffer *pb,
821 struct nvsp_message nvmsg;
822 struct nvsp_1_message_send_rndis_packet *rpkt =
823 &nvmsg.msg.v1_msg.send_rndis_pkt;
824 struct netvsc_channel * const nvchan =
825 &net_device->chan_table[packet->q_idx];
826 struct vmbus_channel *out_channel = nvchan->channel;
827 struct net_device *ndev = hv_get_drvdata(device);
828 struct net_device_context *ndev_ctx = netdev_priv(ndev);
829 struct netdev_queue *txq = netdev_get_tx_queue(ndev, packet->q_idx);
832 u32 ring_avail = hv_get_avail_to_write_percent(&out_channel->outbound);
834 nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
836 rpkt->channel_type = 0; /* 0 is RMC_DATA */
838 rpkt->channel_type = 1; /* 1 is RMC_CONTROL */
840 rpkt->send_buf_section_index = packet->send_buf_index;
841 if (packet->send_buf_index == NETVSC_INVALID_INDEX)
842 rpkt->send_buf_section_size = 0;
844 rpkt->send_buf_section_size = packet->total_data_buflen;
848 if (out_channel->rescind)
851 trace_nvsp_send_pkt(ndev, out_channel, rpkt);
853 if (packet->page_buf_cnt) {
854 if (packet->cp_partial)
855 pb += packet->rmsg_pgcnt;
857 ret = vmbus_sendpacket_pagebuffer(out_channel,
858 pb, packet->page_buf_cnt,
859 &nvmsg, sizeof(nvmsg),
862 ret = vmbus_sendpacket(out_channel,
863 &nvmsg, sizeof(nvmsg),
864 req_id, VM_PKT_DATA_INBAND,
865 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
869 atomic_inc_return(&nvchan->queue_sends);
871 if (ring_avail < RING_AVAIL_PERCENT_LOWATER) {
872 netif_tx_stop_queue(txq);
873 ndev_ctx->eth_stats.stop_queue++;
875 } else if (ret == -EAGAIN) {
876 netif_tx_stop_queue(txq);
877 ndev_ctx->eth_stats.stop_queue++;
878 if (atomic_read(&nvchan->queue_sends) < 1 &&
879 !net_device->tx_disable) {
880 netif_tx_wake_queue(txq);
881 ndev_ctx->eth_stats.wake_queue++;
886 "Unable to send packet pages %u len %u, ret %d\n",
887 packet->page_buf_cnt, packet->total_data_buflen,
894 /* Move packet out of multi send data (msd), and clear msd */
895 static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
896 struct sk_buff **msd_skb,
897 struct multi_send_data *msdp)
899 *msd_skb = msdp->skb;
900 *msd_send = msdp->pkt;
906 /* RCU already held by caller */
907 int netvsc_send(struct net_device *ndev,
908 struct hv_netvsc_packet *packet,
909 struct rndis_message *rndis_msg,
910 struct hv_page_buffer *pb,
913 struct net_device_context *ndev_ctx = netdev_priv(ndev);
914 struct netvsc_device *net_device
915 = rcu_dereference_bh(ndev_ctx->nvdev);
916 struct hv_device *device = ndev_ctx->device_ctx;
918 struct netvsc_channel *nvchan;
919 u32 pktlen = packet->total_data_buflen, msd_len = 0;
920 unsigned int section_index = NETVSC_INVALID_INDEX;
921 struct multi_send_data *msdp;
922 struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
923 struct sk_buff *msd_skb = NULL;
924 bool try_batch, xmit_more;
926 /* If device is rescinded, return error and packet will get dropped. */
927 if (unlikely(!net_device || net_device->destroy))
930 nvchan = &net_device->chan_table[packet->q_idx];
931 packet->send_buf_index = NETVSC_INVALID_INDEX;
932 packet->cp_partial = false;
934 /* Send control message directly without accessing msd (Multi-Send
935 * Data) field which may be changed during data packet processing.
938 return netvsc_send_pkt(device, packet, net_device, pb, skb);
940 /* batch packets in send buffer if possible */
943 msd_len = msdp->pkt->total_data_buflen;
945 try_batch = msd_len > 0 && msdp->count < net_device->max_pkt;
946 if (try_batch && msd_len + pktlen + net_device->pkt_align <
947 net_device->send_section_size) {
948 section_index = msdp->pkt->send_buf_index;
950 } else if (try_batch && msd_len + packet->rmsg_size <
951 net_device->send_section_size) {
952 section_index = msdp->pkt->send_buf_index;
953 packet->cp_partial = true;
955 } else if (pktlen + net_device->pkt_align <
956 net_device->send_section_size) {
957 section_index = netvsc_get_next_send_section(net_device);
958 if (unlikely(section_index == NETVSC_INVALID_INDEX)) {
959 ++ndev_ctx->eth_stats.tx_send_full;
961 move_pkt_msd(&msd_send, &msd_skb, msdp);
966 /* Keep aggregating only if stack says more data is coming
967 * and not doing mixed modes send and not flow blocked
969 xmit_more = skb->xmit_more &&
970 !packet->cp_partial &&
971 !netif_xmit_stopped(netdev_get_tx_queue(ndev, packet->q_idx));
973 if (section_index != NETVSC_INVALID_INDEX) {
974 netvsc_copy_to_send_buf(net_device,
975 section_index, msd_len,
976 packet, rndis_msg, pb, xmit_more);
978 packet->send_buf_index = section_index;
980 if (packet->cp_partial) {
981 packet->page_buf_cnt -= packet->rmsg_pgcnt;
982 packet->total_data_buflen = msd_len + packet->rmsg_size;
984 packet->page_buf_cnt = 0;
985 packet->total_data_buflen += msd_len;
989 packet->total_packets += msdp->pkt->total_packets;
990 packet->total_bytes += msdp->pkt->total_bytes;
994 dev_consume_skb_any(msdp->skb);
1007 move_pkt_msd(&msd_send, &msd_skb, msdp);
1012 int m_ret = netvsc_send_pkt(device, msd_send, net_device,
1016 netvsc_free_send_slot(net_device,
1017 msd_send->send_buf_index);
1018 dev_kfree_skb_any(msd_skb);
1023 ret = netvsc_send_pkt(device, cur_send, net_device, pb, skb);
1025 if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
1026 netvsc_free_send_slot(net_device, section_index);
1031 /* Send pending recv completions */
1032 static int send_recv_completions(struct net_device *ndev,
1033 struct netvsc_device *nvdev,
1034 struct netvsc_channel *nvchan)
1036 struct multi_recv_comp *mrc = &nvchan->mrc;
1037 struct recv_comp_msg {
1038 struct nvsp_message_header hdr;
1041 struct recv_comp_msg msg = {
1042 .hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE,
1046 while (mrc->first != mrc->next) {
1047 const struct recv_comp_data *rcd
1048 = mrc->slots + mrc->first;
1050 msg.status = rcd->status;
1051 ret = vmbus_sendpacket(nvchan->channel, &msg, sizeof(msg),
1052 rcd->tid, VM_PKT_COMP, 0);
1053 if (unlikely(ret)) {
1054 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1056 ++ndev_ctx->eth_stats.rx_comp_busy;
1060 if (++mrc->first == nvdev->recv_completion_cnt)
1064 /* receive completion ring has been emptied */
1065 if (unlikely(nvdev->destroy))
1066 wake_up(&nvdev->wait_drain);
1071 /* Count how many receive completions are outstanding */
1072 static void recv_comp_slot_avail(const struct netvsc_device *nvdev,
1073 const struct multi_recv_comp *mrc,
1074 u32 *filled, u32 *avail)
1076 u32 count = nvdev->recv_completion_cnt;
1078 if (mrc->next >= mrc->first)
1079 *filled = mrc->next - mrc->first;
1081 *filled = (count - mrc->first) + mrc->next;
1083 *avail = count - *filled - 1;
1086 /* Add receive complete to ring to send to host. */
1087 static void enq_receive_complete(struct net_device *ndev,
1088 struct netvsc_device *nvdev, u16 q_idx,
1089 u64 tid, u32 status)
1091 struct netvsc_channel *nvchan = &nvdev->chan_table[q_idx];
1092 struct multi_recv_comp *mrc = &nvchan->mrc;
1093 struct recv_comp_data *rcd;
1096 recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1098 if (unlikely(filled > NAPI_POLL_WEIGHT)) {
1099 send_recv_completions(ndev, nvdev, nvchan);
1100 recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1103 if (unlikely(!avail)) {
1104 netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
1109 rcd = mrc->slots + mrc->next;
1111 rcd->status = status;
1113 if (++mrc->next == nvdev->recv_completion_cnt)
1117 static int netvsc_receive(struct net_device *ndev,
1118 struct netvsc_device *net_device,
1119 struct netvsc_channel *nvchan,
1120 const struct vmpacket_descriptor *desc,
1121 const struct nvsp_message *nvsp)
1123 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1124 struct vmbus_channel *channel = nvchan->channel;
1125 const struct vmtransfer_page_packet_header *vmxferpage_packet
1126 = container_of(desc, const struct vmtransfer_page_packet_header, d);
1127 u16 q_idx = channel->offermsg.offer.sub_channel_index;
1128 char *recv_buf = net_device->recv_buf;
1129 u32 status = NVSP_STAT_SUCCESS;
1133 /* Make sure this is a valid nvsp packet */
1134 if (unlikely(nvsp->hdr.msg_type != NVSP_MSG1_TYPE_SEND_RNDIS_PKT)) {
1135 netif_err(net_device_ctx, rx_err, ndev,
1136 "Unknown nvsp packet type received %u\n",
1137 nvsp->hdr.msg_type);
1141 if (unlikely(vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID)) {
1142 netif_err(net_device_ctx, rx_err, ndev,
1143 "Invalid xfer page set id - expecting %x got %x\n",
1144 NETVSC_RECEIVE_BUFFER_ID,
1145 vmxferpage_packet->xfer_pageset_id);
1149 count = vmxferpage_packet->range_cnt;
1151 /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1152 for (i = 0; i < count; i++) {
1153 u32 offset = vmxferpage_packet->ranges[i].byte_offset;
1154 u32 buflen = vmxferpage_packet->ranges[i].byte_count;
1158 if (unlikely(offset + buflen > net_device->recv_buf_size)) {
1159 nvchan->rsc.cnt = 0;
1160 status = NVSP_STAT_FAIL;
1161 netif_err(net_device_ctx, rx_err, ndev,
1162 "Packet offset:%u + len:%u too big\n",
1168 data = recv_buf + offset;
1170 nvchan->rsc.is_last = (i == count - 1);
1172 trace_rndis_recv(ndev, q_idx, data);
1174 /* Pass it to the upper layer */
1175 ret = rndis_filter_receive(ndev, net_device,
1176 nvchan, data, buflen);
1178 if (unlikely(ret != NVSP_STAT_SUCCESS))
1179 status = NVSP_STAT_FAIL;
1182 enq_receive_complete(ndev, net_device, q_idx,
1183 vmxferpage_packet->d.trans_id, status);
1188 static void netvsc_send_table(struct net_device *ndev,
1189 const struct nvsp_message *nvmsg)
1191 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1195 count = nvmsg->msg.v5_msg.send_table.count;
1196 if (count != VRSS_SEND_TAB_SIZE) {
1197 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1201 tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
1202 nvmsg->msg.v5_msg.send_table.offset);
1204 for (i = 0; i < count; i++)
1205 net_device_ctx->tx_table[i] = tab[i];
1208 static void netvsc_send_vf(struct net_device *ndev,
1209 const struct nvsp_message *nvmsg)
1211 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1213 net_device_ctx->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1214 net_device_ctx->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1215 netdev_info(ndev, "VF slot %u %s\n",
1216 net_device_ctx->vf_serial,
1217 net_device_ctx->vf_alloc ? "added" : "removed");
1220 static void netvsc_receive_inband(struct net_device *ndev,
1221 const struct nvsp_message *nvmsg)
1223 switch (nvmsg->hdr.msg_type) {
1224 case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1225 netvsc_send_table(ndev, nvmsg);
1228 case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1229 netvsc_send_vf(ndev, nvmsg);
1234 static int netvsc_process_raw_pkt(struct hv_device *device,
1235 struct netvsc_channel *nvchan,
1236 struct netvsc_device *net_device,
1237 struct net_device *ndev,
1238 const struct vmpacket_descriptor *desc,
1241 struct vmbus_channel *channel = nvchan->channel;
1242 const struct nvsp_message *nvmsg = hv_pkt_data(desc);
1244 trace_nvsp_recv(ndev, channel, nvmsg);
1246 switch (desc->type) {
1248 netvsc_send_completion(ndev, net_device, channel,
1252 case VM_PKT_DATA_USING_XFER_PAGES:
1253 return netvsc_receive(ndev, net_device, nvchan,
1257 case VM_PKT_DATA_INBAND:
1258 netvsc_receive_inband(ndev, nvmsg);
1262 netdev_err(ndev, "unhandled packet type %d, tid %llx\n",
1263 desc->type, desc->trans_id);
1270 static struct hv_device *netvsc_channel_to_device(struct vmbus_channel *channel)
1272 struct vmbus_channel *primary = channel->primary_channel;
1274 return primary ? primary->device_obj : channel->device_obj;
1277 /* Network processing softirq
1278 * Process data in incoming ring buffer from host
1279 * Stops when ring is empty or budget is met or exceeded.
1281 int netvsc_poll(struct napi_struct *napi, int budget)
1283 struct netvsc_channel *nvchan
1284 = container_of(napi, struct netvsc_channel, napi);
1285 struct netvsc_device *net_device = nvchan->net_device;
1286 struct vmbus_channel *channel = nvchan->channel;
1287 struct hv_device *device = netvsc_channel_to_device(channel);
1288 struct net_device *ndev = hv_get_drvdata(device);
1292 /* If starting a new interval */
1294 nvchan->desc = hv_pkt_iter_first(channel);
1296 while (nvchan->desc && work_done < budget) {
1297 work_done += netvsc_process_raw_pkt(device, nvchan, net_device,
1298 ndev, nvchan->desc, budget);
1299 nvchan->desc = hv_pkt_iter_next(channel, nvchan->desc);
1302 /* Send any pending receive completions */
1303 ret = send_recv_completions(ndev, net_device, nvchan);
1305 /* If it did not exhaust NAPI budget this time
1306 * and not doing busy poll
1307 * then re-enable host interrupts
1308 * and reschedule if ring is not empty
1309 * or sending receive completion failed.
1311 if (work_done < budget &&
1312 napi_complete_done(napi, work_done) &&
1313 (ret || hv_end_read(&channel->inbound)) &&
1314 napi_schedule_prep(napi)) {
1315 hv_begin_read(&channel->inbound);
1316 __napi_schedule(napi);
1319 /* Driver may overshoot since multiple packets per descriptor */
1320 return min(work_done, budget);
1323 /* Call back when data is available in host ring buffer.
1324 * Processing is deferred until network softirq (NAPI)
1326 void netvsc_channel_cb(void *context)
1328 struct netvsc_channel *nvchan = context;
1329 struct vmbus_channel *channel = nvchan->channel;
1330 struct hv_ring_buffer_info *rbi = &channel->inbound;
1332 /* preload first vmpacket descriptor */
1333 prefetch(hv_get_ring_buffer(rbi) + rbi->priv_read_index);
1335 if (napi_schedule_prep(&nvchan->napi)) {
1336 /* disable interrupts from host */
1339 __napi_schedule_irqoff(&nvchan->napi);
1344 * netvsc_device_add - Callback when the device belonging to this
1347 struct netvsc_device *netvsc_device_add(struct hv_device *device,
1348 const struct netvsc_device_info *device_info)
1351 struct netvsc_device *net_device;
1352 struct net_device *ndev = hv_get_drvdata(device);
1353 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1355 net_device = alloc_net_device();
1357 return ERR_PTR(-ENOMEM);
1359 for (i = 0; i < VRSS_SEND_TAB_SIZE; i++)
1360 net_device_ctx->tx_table[i] = 0;
1362 /* Because the device uses NAPI, all the interrupt batching and
1363 * control is done via Net softirq, not the channel handling
1365 set_channel_read_mode(device->channel, HV_CALL_ISR);
1367 /* If we're reopening the device we may have multiple queues, fill the
1368 * chn_table with the default channel to use it before subchannels are
1370 * Initialize the channel state before we open;
1371 * we can be interrupted as soon as we open the channel.
1374 for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
1375 struct netvsc_channel *nvchan = &net_device->chan_table[i];
1377 nvchan->channel = device->channel;
1378 nvchan->net_device = net_device;
1379 u64_stats_init(&nvchan->tx_stats.syncp);
1380 u64_stats_init(&nvchan->rx_stats.syncp);
1383 /* Enable NAPI handler before init callbacks */
1384 netif_napi_add(ndev, &net_device->chan_table[0].napi,
1385 netvsc_poll, NAPI_POLL_WEIGHT);
1387 /* Open the channel */
1388 ret = vmbus_open(device->channel, netvsc_ring_bytes,
1389 netvsc_ring_bytes, NULL, 0,
1390 netvsc_channel_cb, net_device->chan_table);
1393 netdev_err(ndev, "unable to open channel: %d\n", ret);
1397 /* Channel is opened */
1398 netdev_dbg(ndev, "hv_netvsc channel opened successfully\n");
1400 napi_enable(&net_device->chan_table[0].napi);
1402 /* Connect with the NetVsp */
1403 ret = netvsc_connect_vsp(device, net_device, device_info);
1406 "unable to connect to NetVSP - %d\n", ret);
1410 /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1413 rcu_assign_pointer(net_device_ctx->nvdev, net_device);
1418 RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
1419 napi_disable(&net_device->chan_table[0].napi);
1421 /* Now, we can close the channel safely */
1422 vmbus_close(device->channel);
1425 netif_napi_del(&net_device->chan_table[0].napi);
1426 free_netvsc_device(&net_device->rcu);
1428 return ERR_PTR(ret);