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>
34 #include <linux/reciprocal_div.h>
36 #include <asm/sync_bitops.h>
38 #include "hyperv_net.h"
39 #include "netvsc_trace.h"
42 * Switch the data path from the synthetic interface to the VF
45 void netvsc_switch_datapath(struct net_device *ndev, bool vf)
47 struct net_device_context *net_device_ctx = netdev_priv(ndev);
48 struct hv_device *dev = net_device_ctx->device_ctx;
49 struct netvsc_device *nv_dev = rtnl_dereference(net_device_ctx->nvdev);
50 struct nvsp_message *init_pkt = &nv_dev->channel_init_pkt;
52 memset(init_pkt, 0, sizeof(struct nvsp_message));
53 init_pkt->hdr.msg_type = NVSP_MSG4_TYPE_SWITCH_DATA_PATH;
55 init_pkt->msg.v4_msg.active_dp.active_datapath =
58 init_pkt->msg.v4_msg.active_dp.active_datapath =
59 NVSP_DATAPATH_SYNTHETIC;
61 trace_nvsp_send(ndev, init_pkt);
63 vmbus_sendpacket(dev->channel, init_pkt,
64 sizeof(struct nvsp_message),
65 (unsigned long)init_pkt,
66 VM_PKT_DATA_INBAND, 0);
69 static struct netvsc_device *alloc_net_device(void)
71 struct netvsc_device *net_device;
73 net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
77 init_waitqueue_head(&net_device->wait_drain);
78 net_device->destroy = false;
80 net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
81 net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
83 init_completion(&net_device->channel_init_wait);
84 init_waitqueue_head(&net_device->subchan_open);
85 INIT_WORK(&net_device->subchan_work, rndis_set_subchannel);
90 static void free_netvsc_device(struct rcu_head *head)
92 struct netvsc_device *nvdev
93 = container_of(head, struct netvsc_device, rcu);
96 kfree(nvdev->extension);
97 vfree(nvdev->recv_buf);
98 vfree(nvdev->send_buf);
99 kfree(nvdev->send_section_map);
101 for (i = 0; i < VRSS_CHANNEL_MAX; i++)
102 vfree(nvdev->chan_table[i].mrc.slots);
107 static void free_netvsc_device_rcu(struct netvsc_device *nvdev)
109 call_rcu(&nvdev->rcu, free_netvsc_device);
112 static void netvsc_revoke_recv_buf(struct hv_device *device,
113 struct netvsc_device *net_device)
115 struct net_device *ndev = hv_get_drvdata(device);
116 struct nvsp_message *revoke_packet;
120 * If we got a section count, it means we received a
121 * SendReceiveBufferComplete msg (ie sent
122 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
123 * to send a revoke msg here
125 if (net_device->recv_section_cnt) {
126 /* Send the revoke receive buffer */
127 revoke_packet = &net_device->revoke_packet;
128 memset(revoke_packet, 0, sizeof(struct nvsp_message));
130 revoke_packet->hdr.msg_type =
131 NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
132 revoke_packet->msg.v1_msg.
133 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
135 trace_nvsp_send(ndev, revoke_packet);
137 ret = vmbus_sendpacket(device->channel,
139 sizeof(struct nvsp_message),
140 (unsigned long)revoke_packet,
141 VM_PKT_DATA_INBAND, 0);
142 /* If the failure is because the channel is rescinded;
143 * ignore the failure since we cannot send on a rescinded
144 * channel. This would allow us to properly cleanup
145 * even when the channel is rescinded.
147 if (device->channel->rescind)
150 * If we failed here, we might as well return and
151 * have a leak rather than continue and a bugchk
154 netdev_err(ndev, "unable to send "
155 "revoke receive buffer to netvsp\n");
158 net_device->recv_section_cnt = 0;
162 static void netvsc_revoke_send_buf(struct hv_device *device,
163 struct netvsc_device *net_device)
165 struct net_device *ndev = hv_get_drvdata(device);
166 struct nvsp_message *revoke_packet;
169 /* Deal with the send buffer we may have setup.
170 * If we got a send section size, it means we received a
171 * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
172 * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
173 * to send a revoke msg here
175 if (net_device->send_section_cnt) {
176 /* Send the revoke receive buffer */
177 revoke_packet = &net_device->revoke_packet;
178 memset(revoke_packet, 0, sizeof(struct nvsp_message));
180 revoke_packet->hdr.msg_type =
181 NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
182 revoke_packet->msg.v1_msg.revoke_send_buf.id =
183 NETVSC_SEND_BUFFER_ID;
185 trace_nvsp_send(ndev, revoke_packet);
187 ret = vmbus_sendpacket(device->channel,
189 sizeof(struct nvsp_message),
190 (unsigned long)revoke_packet,
191 VM_PKT_DATA_INBAND, 0);
193 /* If the failure is because the channel is rescinded;
194 * ignore the failure since we cannot send on a rescinded
195 * channel. This would allow us to properly cleanup
196 * even when the channel is rescinded.
198 if (device->channel->rescind)
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");
209 net_device->send_section_cnt = 0;
213 static void netvsc_teardown_recv_gpadl(struct hv_device *device,
214 struct netvsc_device *net_device)
216 struct net_device *ndev = hv_get_drvdata(device);
219 if (net_device->recv_buf_gpadl_handle) {
220 ret = vmbus_teardown_gpadl(device->channel,
221 net_device->recv_buf_gpadl_handle);
223 /* If we failed here, we might as well return and have a leak
224 * rather than continue and a bugchk
228 "unable to teardown receive buffer's gpadl\n");
231 net_device->recv_buf_gpadl_handle = 0;
235 static void netvsc_teardown_send_gpadl(struct hv_device *device,
236 struct netvsc_device *net_device)
238 struct net_device *ndev = hv_get_drvdata(device);
241 if (net_device->send_buf_gpadl_handle) {
242 ret = vmbus_teardown_gpadl(device->channel,
243 net_device->send_buf_gpadl_handle);
245 /* If we failed here, we might as well return and have a leak
246 * rather than continue and a bugchk
250 "unable to teardown send buffer's gpadl\n");
253 net_device->send_buf_gpadl_handle = 0;
257 int netvsc_alloc_recv_comp_ring(struct netvsc_device *net_device, u32 q_idx)
259 struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
260 int node = cpu_to_node(nvchan->channel->target_cpu);
263 size = net_device->recv_completion_cnt * sizeof(struct recv_comp_data);
264 nvchan->mrc.slots = vzalloc_node(size, node);
265 if (!nvchan->mrc.slots)
266 nvchan->mrc.slots = vzalloc(size);
268 return nvchan->mrc.slots ? 0 : -ENOMEM;
271 static int netvsc_init_buf(struct hv_device *device,
272 struct netvsc_device *net_device,
273 const struct netvsc_device_info *device_info)
275 struct nvsp_1_message_send_receive_buffer_complete *resp;
276 struct net_device *ndev = hv_get_drvdata(device);
277 struct nvsp_message *init_packet;
278 unsigned int buf_size;
282 /* Get receive buffer area. */
283 buf_size = device_info->recv_sections * device_info->recv_section_size;
284 buf_size = roundup(buf_size, PAGE_SIZE);
286 /* Legacy hosts only allow smaller receive buffer */
287 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
288 buf_size = min_t(unsigned int, buf_size,
289 NETVSC_RECEIVE_BUFFER_SIZE_LEGACY);
291 net_device->recv_buf = vzalloc(buf_size);
292 if (!net_device->recv_buf) {
294 "unable to allocate receive buffer of size %u\n",
300 net_device->recv_buf_size = buf_size;
303 * Establish the gpadl handle for this buffer on this
304 * channel. Note: This call uses the vmbus connection rather
305 * than the channel to establish the gpadl handle.
307 ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
309 &net_device->recv_buf_gpadl_handle);
312 "unable to establish receive buffer's gpadl\n");
316 /* Notify the NetVsp of the gpadl handle */
317 init_packet = &net_device->channel_init_pkt;
318 memset(init_packet, 0, sizeof(struct nvsp_message));
319 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
320 init_packet->msg.v1_msg.send_recv_buf.
321 gpadl_handle = net_device->recv_buf_gpadl_handle;
322 init_packet->msg.v1_msg.
323 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
325 trace_nvsp_send(ndev, init_packet);
327 /* Send the gpadl notification request */
328 ret = vmbus_sendpacket(device->channel, init_packet,
329 sizeof(struct nvsp_message),
330 (unsigned long)init_packet,
332 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
335 "unable to send receive buffer's gpadl to netvsp\n");
339 wait_for_completion(&net_device->channel_init_wait);
341 /* Check the response */
342 resp = &init_packet->msg.v1_msg.send_recv_buf_complete;
343 if (resp->status != NVSP_STAT_SUCCESS) {
345 "Unable to complete receive buffer initialization with NetVsp - status %d\n",
351 /* Parse the response */
352 netdev_dbg(ndev, "Receive sections: %u sub_allocs: size %u count: %u\n",
353 resp->num_sections, resp->sections[0].sub_alloc_size,
354 resp->sections[0].num_sub_allocs);
356 /* There should only be one section for the entire receive buffer */
357 if (resp->num_sections != 1 || resp->sections[0].offset != 0) {
362 net_device->recv_section_size = resp->sections[0].sub_alloc_size;
363 net_device->recv_section_cnt = resp->sections[0].num_sub_allocs;
365 /* Setup receive completion ring */
366 net_device->recv_completion_cnt
367 = round_up(net_device->recv_section_cnt + 1,
368 PAGE_SIZE / sizeof(u64));
369 ret = netvsc_alloc_recv_comp_ring(net_device, 0);
373 /* Now setup the send buffer. */
374 buf_size = device_info->send_sections * device_info->send_section_size;
375 buf_size = round_up(buf_size, PAGE_SIZE);
377 net_device->send_buf = vzalloc(buf_size);
378 if (!net_device->send_buf) {
379 netdev_err(ndev, "unable to allocate send buffer of size %u\n",
385 /* Establish the gpadl handle for this buffer on this
386 * channel. Note: This call uses the vmbus connection rather
387 * than the channel to establish the gpadl handle.
389 ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
391 &net_device->send_buf_gpadl_handle);
394 "unable to establish send buffer's gpadl\n");
398 /* Notify the NetVsp of the gpadl handle */
399 init_packet = &net_device->channel_init_pkt;
400 memset(init_packet, 0, sizeof(struct nvsp_message));
401 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
402 init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
403 net_device->send_buf_gpadl_handle;
404 init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
406 trace_nvsp_send(ndev, init_packet);
408 /* Send the gpadl notification request */
409 ret = vmbus_sendpacket(device->channel, init_packet,
410 sizeof(struct nvsp_message),
411 (unsigned long)init_packet,
413 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
416 "unable to send send buffer's gpadl to netvsp\n");
420 wait_for_completion(&net_device->channel_init_wait);
422 /* Check the response */
423 if (init_packet->msg.v1_msg.
424 send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
425 netdev_err(ndev, "Unable to complete send buffer "
426 "initialization with NetVsp - status %d\n",
427 init_packet->msg.v1_msg.
428 send_send_buf_complete.status);
433 /* Parse the response */
434 net_device->send_section_size = init_packet->msg.
435 v1_msg.send_send_buf_complete.section_size;
437 /* Section count is simply the size divided by the section size. */
438 net_device->send_section_cnt = buf_size / net_device->send_section_size;
440 netdev_dbg(ndev, "Send section size: %d, Section count:%d\n",
441 net_device->send_section_size, net_device->send_section_cnt);
443 /* Setup state for managing the send buffer. */
444 map_words = DIV_ROUND_UP(net_device->send_section_cnt, BITS_PER_LONG);
446 net_device->send_section_map = kcalloc(map_words, sizeof(ulong), GFP_KERNEL);
447 if (net_device->send_section_map == NULL) {
455 netvsc_revoke_recv_buf(device, net_device);
456 netvsc_revoke_send_buf(device, net_device);
457 netvsc_teardown_recv_gpadl(device, net_device);
458 netvsc_teardown_send_gpadl(device, net_device);
464 /* Negotiate NVSP protocol version */
465 static int negotiate_nvsp_ver(struct hv_device *device,
466 struct netvsc_device *net_device,
467 struct nvsp_message *init_packet,
470 struct net_device *ndev = hv_get_drvdata(device);
473 memset(init_packet, 0, sizeof(struct nvsp_message));
474 init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
475 init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
476 init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
478 trace_nvsp_send(ndev, init_packet);
480 /* Send the init request */
481 ret = vmbus_sendpacket(device->channel, init_packet,
482 sizeof(struct nvsp_message),
483 (unsigned long)init_packet,
485 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
490 wait_for_completion(&net_device->channel_init_wait);
492 if (init_packet->msg.init_msg.init_complete.status !=
496 if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
499 /* NVSPv2 or later: Send NDIS config */
500 memset(init_packet, 0, sizeof(struct nvsp_message));
501 init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
502 init_packet->msg.v2_msg.send_ndis_config.mtu = ndev->mtu + ETH_HLEN;
503 init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
505 if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5) {
506 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
508 /* Teaming bit is needed to receive link speed updates */
509 init_packet->msg.v2_msg.send_ndis_config.capability.teaming = 1;
512 trace_nvsp_send(ndev, init_packet);
514 ret = vmbus_sendpacket(device->channel, init_packet,
515 sizeof(struct nvsp_message),
516 (unsigned long)init_packet,
517 VM_PKT_DATA_INBAND, 0);
522 static int netvsc_connect_vsp(struct hv_device *device,
523 struct netvsc_device *net_device,
524 const struct netvsc_device_info *device_info)
526 struct net_device *ndev = hv_get_drvdata(device);
527 static const u32 ver_list[] = {
528 NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
529 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5
531 struct nvsp_message *init_packet;
532 int ndis_version, i, ret;
534 init_packet = &net_device->channel_init_pkt;
536 /* Negotiate the latest NVSP protocol supported */
537 for (i = ARRAY_SIZE(ver_list) - 1; i >= 0; i--)
538 if (negotiate_nvsp_ver(device, net_device, init_packet,
540 net_device->nvsp_version = ver_list[i];
549 pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
551 /* Send the ndis version */
552 memset(init_packet, 0, sizeof(struct nvsp_message));
554 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
555 ndis_version = 0x00060001;
557 ndis_version = 0x0006001e;
559 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
560 init_packet->msg.v1_msg.
561 send_ndis_ver.ndis_major_ver =
562 (ndis_version & 0xFFFF0000) >> 16;
563 init_packet->msg.v1_msg.
564 send_ndis_ver.ndis_minor_ver =
565 ndis_version & 0xFFFF;
567 trace_nvsp_send(ndev, init_packet);
569 /* Send the init request */
570 ret = vmbus_sendpacket(device->channel, init_packet,
571 sizeof(struct nvsp_message),
572 (unsigned long)init_packet,
573 VM_PKT_DATA_INBAND, 0);
578 ret = netvsc_init_buf(device, net_device, device_info);
585 * netvsc_device_remove - Callback when the root bus device is removed
587 void netvsc_device_remove(struct hv_device *device)
589 struct net_device *ndev = hv_get_drvdata(device);
590 struct net_device_context *net_device_ctx = netdev_priv(ndev);
591 struct netvsc_device *net_device
592 = rtnl_dereference(net_device_ctx->nvdev);
596 * Revoke receive buffer. If host is pre-Win2016 then tear down
597 * receive buffer GPADL. Do the same for send buffer.
599 netvsc_revoke_recv_buf(device, net_device);
600 if (vmbus_proto_version < VERSION_WIN10)
601 netvsc_teardown_recv_gpadl(device, net_device);
603 netvsc_revoke_send_buf(device, net_device);
604 if (vmbus_proto_version < VERSION_WIN10)
605 netvsc_teardown_send_gpadl(device, net_device);
607 RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
609 /* And disassociate NAPI context from device */
610 for (i = 0; i < net_device->num_chn; i++)
611 netif_napi_del(&net_device->chan_table[i].napi);
614 * At this point, no one should be accessing net_device
617 netdev_dbg(ndev, "net device safe to remove\n");
619 /* Now, we can close the channel safely */
620 vmbus_close(device->channel);
623 * If host is Win2016 or higher then we do the GPADL tear down
624 * here after VMBus is closed.
626 if (vmbus_proto_version >= VERSION_WIN10) {
627 netvsc_teardown_recv_gpadl(device, net_device);
628 netvsc_teardown_send_gpadl(device, net_device);
631 /* Release all resources */
632 free_netvsc_device_rcu(net_device);
635 #define RING_AVAIL_PERCENT_HIWATER 20
636 #define RING_AVAIL_PERCENT_LOWATER 10
639 * Get the percentage of available bytes to write in the ring.
640 * The return value is in range from 0 to 100.
642 static u32 hv_ringbuf_avail_percent(const struct hv_ring_buffer_info *ring_info)
644 u32 avail_write = hv_get_bytes_to_write(ring_info);
646 return reciprocal_divide(avail_write * 100, netvsc_ring_reciprocal);
649 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
652 sync_change_bit(index, net_device->send_section_map);
655 static void netvsc_send_tx_complete(struct netvsc_device *net_device,
656 struct vmbus_channel *incoming_channel,
657 struct hv_device *device,
658 const struct vmpacket_descriptor *desc,
661 struct sk_buff *skb = (struct sk_buff *)(unsigned long)desc->trans_id;
662 struct net_device *ndev = hv_get_drvdata(device);
663 struct net_device_context *ndev_ctx = netdev_priv(ndev);
664 struct vmbus_channel *channel = device->channel;
668 /* Notify the layer above us */
670 const struct hv_netvsc_packet *packet
671 = (struct hv_netvsc_packet *)skb->cb;
672 u32 send_index = packet->send_buf_index;
673 struct netvsc_stats *tx_stats;
675 if (send_index != NETVSC_INVALID_INDEX)
676 netvsc_free_send_slot(net_device, send_index);
677 q_idx = packet->q_idx;
678 channel = incoming_channel;
680 tx_stats = &net_device->chan_table[q_idx].tx_stats;
682 u64_stats_update_begin(&tx_stats->syncp);
683 tx_stats->packets += packet->total_packets;
684 tx_stats->bytes += packet->total_bytes;
685 u64_stats_update_end(&tx_stats->syncp);
687 napi_consume_skb(skb, budget);
691 atomic_dec_return(&net_device->chan_table[q_idx].queue_sends);
693 if (unlikely(net_device->destroy)) {
694 if (queue_sends == 0)
695 wake_up(&net_device->wait_drain);
697 struct netdev_queue *txq = netdev_get_tx_queue(ndev, q_idx);
699 if (netif_tx_queue_stopped(txq) &&
700 (hv_ringbuf_avail_percent(&channel->outbound) > RING_AVAIL_PERCENT_HIWATER ||
702 netif_tx_wake_queue(txq);
703 ndev_ctx->eth_stats.wake_queue++;
708 static void netvsc_send_completion(struct netvsc_device *net_device,
709 struct vmbus_channel *incoming_channel,
710 struct hv_device *device,
711 const struct vmpacket_descriptor *desc,
714 struct nvsp_message *nvsp_packet = hv_pkt_data(desc);
715 struct net_device *ndev = hv_get_drvdata(device);
717 switch (nvsp_packet->hdr.msg_type) {
718 case NVSP_MSG_TYPE_INIT_COMPLETE:
719 case NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE:
720 case NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE:
721 case NVSP_MSG5_TYPE_SUBCHANNEL:
722 /* Copy the response back */
723 memcpy(&net_device->channel_init_pkt, nvsp_packet,
724 sizeof(struct nvsp_message));
725 complete(&net_device->channel_init_wait);
728 case NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE:
729 netvsc_send_tx_complete(net_device, incoming_channel,
730 device, desc, budget);
735 "Unknown send completion type %d received!!\n",
736 nvsp_packet->hdr.msg_type);
740 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
742 unsigned long *map_addr = net_device->send_section_map;
745 for_each_clear_bit(i, map_addr, net_device->send_section_cnt) {
746 if (sync_test_and_set_bit(i, map_addr) == 0)
750 return NETVSC_INVALID_INDEX;
753 static void netvsc_copy_to_send_buf(struct netvsc_device *net_device,
754 unsigned int section_index,
756 struct hv_netvsc_packet *packet,
757 struct rndis_message *rndis_msg,
758 struct hv_page_buffer *pb,
761 char *start = net_device->send_buf;
762 char *dest = start + (section_index * net_device->send_section_size)
766 u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
767 packet->page_buf_cnt;
771 remain = packet->total_data_buflen & (net_device->pkt_align - 1);
772 if (xmit_more && remain) {
773 padding = net_device->pkt_align - remain;
774 rndis_msg->msg_len += padding;
775 packet->total_data_buflen += padding;
778 for (i = 0; i < page_count; i++) {
779 char *src = phys_to_virt(pb[i].pfn << PAGE_SHIFT);
780 u32 offset = pb[i].offset;
783 memcpy(dest, (src + offset), len);
788 memset(dest, 0, padding);
791 static inline int netvsc_send_pkt(
792 struct hv_device *device,
793 struct hv_netvsc_packet *packet,
794 struct netvsc_device *net_device,
795 struct hv_page_buffer *pb,
798 struct nvsp_message nvmsg;
799 struct nvsp_1_message_send_rndis_packet *rpkt =
800 &nvmsg.msg.v1_msg.send_rndis_pkt;
801 struct netvsc_channel * const nvchan =
802 &net_device->chan_table[packet->q_idx];
803 struct vmbus_channel *out_channel = nvchan->channel;
804 struct net_device *ndev = hv_get_drvdata(device);
805 struct net_device_context *ndev_ctx = netdev_priv(ndev);
806 struct netdev_queue *txq = netdev_get_tx_queue(ndev, packet->q_idx);
809 u32 ring_avail = hv_ringbuf_avail_percent(&out_channel->outbound);
811 nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
813 rpkt->channel_type = 0; /* 0 is RMC_DATA */
815 rpkt->channel_type = 1; /* 1 is RMC_CONTROL */
817 rpkt->send_buf_section_index = packet->send_buf_index;
818 if (packet->send_buf_index == NETVSC_INVALID_INDEX)
819 rpkt->send_buf_section_size = 0;
821 rpkt->send_buf_section_size = packet->total_data_buflen;
825 if (out_channel->rescind)
828 trace_nvsp_send_pkt(ndev, out_channel, rpkt);
830 if (packet->page_buf_cnt) {
831 if (packet->cp_partial)
832 pb += packet->rmsg_pgcnt;
834 ret = vmbus_sendpacket_pagebuffer(out_channel,
835 pb, packet->page_buf_cnt,
836 &nvmsg, sizeof(nvmsg),
839 ret = vmbus_sendpacket(out_channel,
840 &nvmsg, sizeof(nvmsg),
841 req_id, VM_PKT_DATA_INBAND,
842 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
846 atomic_inc_return(&nvchan->queue_sends);
848 if (ring_avail < RING_AVAIL_PERCENT_LOWATER) {
849 netif_tx_stop_queue(txq);
850 ndev_ctx->eth_stats.stop_queue++;
852 } else if (ret == -EAGAIN) {
853 netif_tx_stop_queue(txq);
854 ndev_ctx->eth_stats.stop_queue++;
855 if (atomic_read(&nvchan->queue_sends) < 1) {
856 netif_tx_wake_queue(txq);
857 ndev_ctx->eth_stats.wake_queue++;
862 "Unable to send packet pages %u len %u, ret %d\n",
863 packet->page_buf_cnt, packet->total_data_buflen,
870 /* Move packet out of multi send data (msd), and clear msd */
871 static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
872 struct sk_buff **msd_skb,
873 struct multi_send_data *msdp)
875 *msd_skb = msdp->skb;
876 *msd_send = msdp->pkt;
882 /* RCU already held by caller */
883 int netvsc_send(struct net_device *ndev,
884 struct hv_netvsc_packet *packet,
885 struct rndis_message *rndis_msg,
886 struct hv_page_buffer *pb,
889 struct net_device_context *ndev_ctx = netdev_priv(ndev);
890 struct netvsc_device *net_device
891 = rcu_dereference_bh(ndev_ctx->nvdev);
892 struct hv_device *device = ndev_ctx->device_ctx;
894 struct netvsc_channel *nvchan;
895 u32 pktlen = packet->total_data_buflen, msd_len = 0;
896 unsigned int section_index = NETVSC_INVALID_INDEX;
897 struct multi_send_data *msdp;
898 struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
899 struct sk_buff *msd_skb = NULL;
900 bool try_batch, xmit_more;
902 /* If device is rescinded, return error and packet will get dropped. */
903 if (unlikely(!net_device || net_device->destroy))
906 nvchan = &net_device->chan_table[packet->q_idx];
907 packet->send_buf_index = NETVSC_INVALID_INDEX;
908 packet->cp_partial = false;
910 /* Send control message directly without accessing msd (Multi-Send
911 * Data) field which may be changed during data packet processing.
914 return netvsc_send_pkt(device, packet, net_device, pb, skb);
916 /* batch packets in send buffer if possible */
919 msd_len = msdp->pkt->total_data_buflen;
921 try_batch = msd_len > 0 && msdp->count < net_device->max_pkt;
922 if (try_batch && msd_len + pktlen + net_device->pkt_align <
923 net_device->send_section_size) {
924 section_index = msdp->pkt->send_buf_index;
926 } else if (try_batch && msd_len + packet->rmsg_size <
927 net_device->send_section_size) {
928 section_index = msdp->pkt->send_buf_index;
929 packet->cp_partial = true;
931 } else if (pktlen + net_device->pkt_align <
932 net_device->send_section_size) {
933 section_index = netvsc_get_next_send_section(net_device);
934 if (unlikely(section_index == NETVSC_INVALID_INDEX)) {
935 ++ndev_ctx->eth_stats.tx_send_full;
937 move_pkt_msd(&msd_send, &msd_skb, msdp);
942 /* Keep aggregating only if stack says more data is coming
943 * and not doing mixed modes send and not flow blocked
945 xmit_more = skb->xmit_more &&
946 !packet->cp_partial &&
947 !netif_xmit_stopped(netdev_get_tx_queue(ndev, packet->q_idx));
949 if (section_index != NETVSC_INVALID_INDEX) {
950 netvsc_copy_to_send_buf(net_device,
951 section_index, msd_len,
952 packet, rndis_msg, pb, xmit_more);
954 packet->send_buf_index = section_index;
956 if (packet->cp_partial) {
957 packet->page_buf_cnt -= packet->rmsg_pgcnt;
958 packet->total_data_buflen = msd_len + packet->rmsg_size;
960 packet->page_buf_cnt = 0;
961 packet->total_data_buflen += msd_len;
965 packet->total_packets += msdp->pkt->total_packets;
966 packet->total_bytes += msdp->pkt->total_bytes;
970 dev_consume_skb_any(msdp->skb);
983 move_pkt_msd(&msd_send, &msd_skb, msdp);
988 int m_ret = netvsc_send_pkt(device, msd_send, net_device,
992 netvsc_free_send_slot(net_device,
993 msd_send->send_buf_index);
994 dev_kfree_skb_any(msd_skb);
999 ret = netvsc_send_pkt(device, cur_send, net_device, pb, skb);
1001 if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
1002 netvsc_free_send_slot(net_device, section_index);
1007 /* Send pending recv completions */
1008 static int send_recv_completions(struct net_device *ndev,
1009 struct netvsc_device *nvdev,
1010 struct netvsc_channel *nvchan)
1012 struct multi_recv_comp *mrc = &nvchan->mrc;
1013 struct recv_comp_msg {
1014 struct nvsp_message_header hdr;
1017 struct recv_comp_msg msg = {
1018 .hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE,
1022 while (mrc->first != mrc->next) {
1023 const struct recv_comp_data *rcd
1024 = mrc->slots + mrc->first;
1026 msg.status = rcd->status;
1027 ret = vmbus_sendpacket(nvchan->channel, &msg, sizeof(msg),
1028 rcd->tid, VM_PKT_COMP, 0);
1029 if (unlikely(ret)) {
1030 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1032 ++ndev_ctx->eth_stats.rx_comp_busy;
1036 if (++mrc->first == nvdev->recv_completion_cnt)
1040 /* receive completion ring has been emptied */
1041 if (unlikely(nvdev->destroy))
1042 wake_up(&nvdev->wait_drain);
1047 /* Count how many receive completions are outstanding */
1048 static void recv_comp_slot_avail(const struct netvsc_device *nvdev,
1049 const struct multi_recv_comp *mrc,
1050 u32 *filled, u32 *avail)
1052 u32 count = nvdev->recv_completion_cnt;
1054 if (mrc->next >= mrc->first)
1055 *filled = mrc->next - mrc->first;
1057 *filled = (count - mrc->first) + mrc->next;
1059 *avail = count - *filled - 1;
1062 /* Add receive complete to ring to send to host. */
1063 static void enq_receive_complete(struct net_device *ndev,
1064 struct netvsc_device *nvdev, u16 q_idx,
1065 u64 tid, u32 status)
1067 struct netvsc_channel *nvchan = &nvdev->chan_table[q_idx];
1068 struct multi_recv_comp *mrc = &nvchan->mrc;
1069 struct recv_comp_data *rcd;
1072 recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1074 if (unlikely(filled > NAPI_POLL_WEIGHT)) {
1075 send_recv_completions(ndev, nvdev, nvchan);
1076 recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1079 if (unlikely(!avail)) {
1080 netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
1085 rcd = mrc->slots + mrc->next;
1087 rcd->status = status;
1089 if (++mrc->next == nvdev->recv_completion_cnt)
1093 static int netvsc_receive(struct net_device *ndev,
1094 struct netvsc_device *net_device,
1095 struct net_device_context *net_device_ctx,
1096 struct hv_device *device,
1097 struct vmbus_channel *channel,
1098 const struct vmpacket_descriptor *desc,
1099 struct nvsp_message *nvsp)
1101 const struct vmtransfer_page_packet_header *vmxferpage_packet
1102 = container_of(desc, const struct vmtransfer_page_packet_header, d);
1103 u16 q_idx = channel->offermsg.offer.sub_channel_index;
1104 char *recv_buf = net_device->recv_buf;
1105 u32 status = NVSP_STAT_SUCCESS;
1109 /* Make sure this is a valid nvsp packet */
1110 if (unlikely(nvsp->hdr.msg_type != NVSP_MSG1_TYPE_SEND_RNDIS_PKT)) {
1111 netif_err(net_device_ctx, rx_err, ndev,
1112 "Unknown nvsp packet type received %u\n",
1113 nvsp->hdr.msg_type);
1117 if (unlikely(vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID)) {
1118 netif_err(net_device_ctx, rx_err, ndev,
1119 "Invalid xfer page set id - expecting %x got %x\n",
1120 NETVSC_RECEIVE_BUFFER_ID,
1121 vmxferpage_packet->xfer_pageset_id);
1125 count = vmxferpage_packet->range_cnt;
1127 /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1128 for (i = 0; i < count; i++) {
1129 u32 offset = vmxferpage_packet->ranges[i].byte_offset;
1130 u32 buflen = vmxferpage_packet->ranges[i].byte_count;
1134 if (unlikely(offset + buflen > net_device->recv_buf_size)) {
1135 status = NVSP_STAT_FAIL;
1136 netif_err(net_device_ctx, rx_err, ndev,
1137 "Packet offset:%u + len:%u too big\n",
1143 data = recv_buf + offset;
1145 trace_rndis_recv(ndev, q_idx, data);
1147 /* Pass it to the upper layer */
1148 ret = rndis_filter_receive(ndev, net_device,
1149 channel, data, buflen);
1151 if (unlikely(ret != NVSP_STAT_SUCCESS))
1152 status = NVSP_STAT_FAIL;
1155 enq_receive_complete(ndev, net_device, q_idx,
1156 vmxferpage_packet->d.trans_id, status);
1161 static void netvsc_send_table(struct hv_device *hdev,
1162 struct nvsp_message *nvmsg)
1164 struct net_device *ndev = hv_get_drvdata(hdev);
1165 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1169 count = nvmsg->msg.v5_msg.send_table.count;
1170 if (count != VRSS_SEND_TAB_SIZE) {
1171 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1175 tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
1176 nvmsg->msg.v5_msg.send_table.offset);
1178 for (i = 0; i < count; i++)
1179 net_device_ctx->tx_table[i] = tab[i];
1182 static void netvsc_send_vf(struct net_device_context *net_device_ctx,
1183 struct nvsp_message *nvmsg)
1185 net_device_ctx->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1186 net_device_ctx->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1189 static inline void netvsc_receive_inband(struct hv_device *hdev,
1190 struct net_device_context *net_device_ctx,
1191 struct nvsp_message *nvmsg)
1193 switch (nvmsg->hdr.msg_type) {
1194 case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1195 netvsc_send_table(hdev, nvmsg);
1198 case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1199 netvsc_send_vf(net_device_ctx, nvmsg);
1204 static int netvsc_process_raw_pkt(struct hv_device *device,
1205 struct vmbus_channel *channel,
1206 struct netvsc_device *net_device,
1207 struct net_device *ndev,
1208 const struct vmpacket_descriptor *desc,
1211 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1212 struct nvsp_message *nvmsg = hv_pkt_data(desc);
1214 trace_nvsp_recv(ndev, channel, nvmsg);
1216 switch (desc->type) {
1218 netvsc_send_completion(net_device, channel, device,
1222 case VM_PKT_DATA_USING_XFER_PAGES:
1223 return netvsc_receive(ndev, net_device, net_device_ctx,
1224 device, channel, desc, nvmsg);
1227 case VM_PKT_DATA_INBAND:
1228 netvsc_receive_inband(device, net_device_ctx, nvmsg);
1232 netdev_err(ndev, "unhandled packet type %d, tid %llx\n",
1233 desc->type, desc->trans_id);
1240 static struct hv_device *netvsc_channel_to_device(struct vmbus_channel *channel)
1242 struct vmbus_channel *primary = channel->primary_channel;
1244 return primary ? primary->device_obj : channel->device_obj;
1247 /* Network processing softirq
1248 * Process data in incoming ring buffer from host
1249 * Stops when ring is empty or budget is met or exceeded.
1251 int netvsc_poll(struct napi_struct *napi, int budget)
1253 struct netvsc_channel *nvchan
1254 = container_of(napi, struct netvsc_channel, napi);
1255 struct netvsc_device *net_device = nvchan->net_device;
1256 struct vmbus_channel *channel = nvchan->channel;
1257 struct hv_device *device = netvsc_channel_to_device(channel);
1258 struct net_device *ndev = hv_get_drvdata(device);
1261 /* If starting a new interval */
1263 nvchan->desc = hv_pkt_iter_first(channel);
1265 while (nvchan->desc && work_done < budget) {
1266 work_done += netvsc_process_raw_pkt(device, channel, net_device,
1267 ndev, nvchan->desc, budget);
1268 nvchan->desc = hv_pkt_iter_next(channel, nvchan->desc);
1271 /* If send of pending receive completions suceeded
1272 * and did not exhaust NAPI budget this time
1273 * and not doing busy poll
1274 * then re-enable host interrupts
1275 * and reschedule if ring is not empty.
1277 if (send_recv_completions(ndev, net_device, nvchan) == 0 &&
1278 work_done < budget &&
1279 napi_complete_done(napi, work_done) &&
1280 hv_end_read(&channel->inbound) &&
1281 napi_schedule_prep(napi)) {
1282 hv_begin_read(&channel->inbound);
1283 __napi_schedule(napi);
1286 /* Driver may overshoot since multiple packets per descriptor */
1287 return min(work_done, budget);
1290 /* Call back when data is available in host ring buffer.
1291 * Processing is deferred until network softirq (NAPI)
1293 void netvsc_channel_cb(void *context)
1295 struct netvsc_channel *nvchan = context;
1296 struct vmbus_channel *channel = nvchan->channel;
1297 struct hv_ring_buffer_info *rbi = &channel->inbound;
1299 /* preload first vmpacket descriptor */
1300 prefetch(hv_get_ring_buffer(rbi) + rbi->priv_read_index);
1302 if (napi_schedule_prep(&nvchan->napi)) {
1303 /* disable interupts from host */
1306 __napi_schedule_irqoff(&nvchan->napi);
1311 * netvsc_device_add - Callback when the device belonging to this
1314 struct netvsc_device *netvsc_device_add(struct hv_device *device,
1315 const struct netvsc_device_info *device_info)
1318 struct netvsc_device *net_device;
1319 struct net_device *ndev = hv_get_drvdata(device);
1320 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1322 net_device = alloc_net_device();
1324 return ERR_PTR(-ENOMEM);
1326 for (i = 0; i < VRSS_SEND_TAB_SIZE; i++)
1327 net_device_ctx->tx_table[i] = 0;
1329 /* Because the device uses NAPI, all the interrupt batching and
1330 * control is done via Net softirq, not the channel handling
1332 set_channel_read_mode(device->channel, HV_CALL_ISR);
1334 /* If we're reopening the device we may have multiple queues, fill the
1335 * chn_table with the default channel to use it before subchannels are
1337 * Initialize the channel state before we open;
1338 * we can be interrupted as soon as we open the channel.
1341 for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
1342 struct netvsc_channel *nvchan = &net_device->chan_table[i];
1344 nvchan->channel = device->channel;
1345 nvchan->net_device = net_device;
1346 u64_stats_init(&nvchan->tx_stats.syncp);
1347 u64_stats_init(&nvchan->rx_stats.syncp);
1350 /* Enable NAPI handler before init callbacks */
1351 netif_napi_add(ndev, &net_device->chan_table[0].napi,
1352 netvsc_poll, NAPI_POLL_WEIGHT);
1354 /* Open the channel */
1355 ret = vmbus_open(device->channel, netvsc_ring_bytes,
1356 netvsc_ring_bytes, NULL, 0,
1357 netvsc_channel_cb, net_device->chan_table);
1360 netdev_err(ndev, "unable to open channel: %d\n", ret);
1364 /* Channel is opened */
1365 netdev_dbg(ndev, "hv_netvsc channel opened successfully\n");
1367 napi_enable(&net_device->chan_table[0].napi);
1369 /* Connect with the NetVsp */
1370 ret = netvsc_connect_vsp(device, net_device, device_info);
1373 "unable to connect to NetVSP - %d\n", ret);
1377 /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1380 rcu_assign_pointer(net_device_ctx->nvdev, net_device);
1385 RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
1386 napi_disable(&net_device->chan_table[0].napi);
1388 /* Now, we can close the channel safely */
1389 vmbus_close(device->channel);
1392 netif_napi_del(&net_device->chan_table[0].napi);
1393 free_netvsc_device(&net_device->rcu);
1395 return ERR_PTR(ret);