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"
40 * Switch the data path from the synthetic interface to the VF
43 void netvsc_switch_datapath(struct net_device *ndev, bool vf)
45 struct net_device_context *net_device_ctx = netdev_priv(ndev);
46 struct hv_device *dev = net_device_ctx->device_ctx;
47 struct netvsc_device *nv_dev = rtnl_dereference(net_device_ctx->nvdev);
48 struct nvsp_message *init_pkt = &nv_dev->channel_init_pkt;
50 memset(init_pkt, 0, sizeof(struct nvsp_message));
51 init_pkt->hdr.msg_type = NVSP_MSG4_TYPE_SWITCH_DATA_PATH;
53 init_pkt->msg.v4_msg.active_dp.active_datapath =
56 init_pkt->msg.v4_msg.active_dp.active_datapath =
57 NVSP_DATAPATH_SYNTHETIC;
59 vmbus_sendpacket(dev->channel, init_pkt,
60 sizeof(struct nvsp_message),
61 (unsigned long)init_pkt,
62 VM_PKT_DATA_INBAND, 0);
65 static struct netvsc_device *alloc_net_device(void)
67 struct netvsc_device *net_device;
69 net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
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;
79 init_completion(&net_device->channel_init_wait);
80 init_waitqueue_head(&net_device->subchan_open);
81 INIT_WORK(&net_device->subchan_work, rndis_set_subchannel);
86 static void free_netvsc_device(struct rcu_head *head)
88 struct netvsc_device *nvdev
89 = container_of(head, struct netvsc_device, rcu);
92 for (i = 0; i < VRSS_CHANNEL_MAX; i++)
93 vfree(nvdev->chan_table[i].mrc.slots);
98 static void free_netvsc_device_rcu(struct netvsc_device *nvdev)
100 call_rcu(&nvdev->rcu, free_netvsc_device);
103 static void netvsc_destroy_buf(struct hv_device *device)
105 struct nvsp_message *revoke_packet;
106 struct net_device *ndev = hv_get_drvdata(device);
107 struct net_device_context *ndc = netdev_priv(ndev);
108 struct netvsc_device *net_device = rtnl_dereference(ndc->nvdev);
112 * If we got a section count, it means we received a
113 * SendReceiveBufferComplete msg (ie sent
114 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
115 * to send a revoke msg here
117 if (net_device->recv_section_cnt) {
118 /* Send the revoke receive buffer */
119 revoke_packet = &net_device->revoke_packet;
120 memset(revoke_packet, 0, sizeof(struct nvsp_message));
122 revoke_packet->hdr.msg_type =
123 NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
124 revoke_packet->msg.v1_msg.
125 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
127 ret = vmbus_sendpacket(device->channel,
129 sizeof(struct nvsp_message),
130 (unsigned long)revoke_packet,
131 VM_PKT_DATA_INBAND, 0);
132 /* If the failure is because the channel is rescinded;
133 * ignore the failure since we cannot send on a rescinded
134 * channel. This would allow us to properly cleanup
135 * even when the channel is rescinded.
137 if (device->channel->rescind)
140 * If we failed here, we might as well return and
141 * have a leak rather than continue and a bugchk
144 netdev_err(ndev, "unable to send "
145 "revoke receive buffer to netvsp\n");
148 net_device->recv_section_cnt = 0;
151 /* Teardown the gpadl on the vsp end */
152 if (net_device->recv_buf_gpadl_handle) {
153 ret = vmbus_teardown_gpadl(device->channel,
154 net_device->recv_buf_gpadl_handle);
156 /* If we failed here, we might as well return and have a leak
157 * rather than continue and a bugchk
161 "unable to teardown receive buffer's gpadl\n");
164 net_device->recv_buf_gpadl_handle = 0;
167 if (net_device->recv_buf) {
168 /* Free up the receive buffer */
169 vfree(net_device->recv_buf);
170 net_device->recv_buf = NULL;
173 /* Deal with the send buffer we may have setup.
174 * If we got a send section size, it means we received a
175 * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
176 * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
177 * to send a revoke msg here
179 if (net_device->send_section_cnt) {
180 /* Send the revoke receive buffer */
181 revoke_packet = &net_device->revoke_packet;
182 memset(revoke_packet, 0, sizeof(struct nvsp_message));
184 revoke_packet->hdr.msg_type =
185 NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
186 revoke_packet->msg.v1_msg.revoke_send_buf.id =
187 NETVSC_SEND_BUFFER_ID;
189 ret = vmbus_sendpacket(device->channel,
191 sizeof(struct nvsp_message),
192 (unsigned long)revoke_packet,
193 VM_PKT_DATA_INBAND, 0);
195 /* If the failure is because the channel is rescinded;
196 * ignore the failure since we cannot send on a rescinded
197 * channel. This would allow us to properly cleanup
198 * even when the channel is rescinded.
200 if (device->channel->rescind)
203 /* If we failed here, we might as well return and
204 * have a leak rather than continue and a bugchk
207 netdev_err(ndev, "unable to send "
208 "revoke send buffer to netvsp\n");
211 net_device->send_section_cnt = 0;
213 /* Teardown the gpadl on the vsp end */
214 if (net_device->send_buf_gpadl_handle) {
215 ret = vmbus_teardown_gpadl(device->channel,
216 net_device->send_buf_gpadl_handle);
218 /* If we failed here, we might as well return and have a leak
219 * rather than continue and a bugchk
223 "unable to teardown send buffer's gpadl\n");
226 net_device->send_buf_gpadl_handle = 0;
228 if (net_device->send_buf) {
229 /* Free up the send buffer */
230 vfree(net_device->send_buf);
231 net_device->send_buf = NULL;
233 kfree(net_device->send_section_map);
236 int netvsc_alloc_recv_comp_ring(struct netvsc_device *net_device, u32 q_idx)
238 struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
239 int node = cpu_to_node(nvchan->channel->target_cpu);
242 size = net_device->recv_completion_cnt * sizeof(struct recv_comp_data);
243 nvchan->mrc.slots = vzalloc_node(size, node);
244 if (!nvchan->mrc.slots)
245 nvchan->mrc.slots = vzalloc(size);
247 return nvchan->mrc.slots ? 0 : -ENOMEM;
250 static int netvsc_init_buf(struct hv_device *device,
251 struct netvsc_device *net_device,
252 const struct netvsc_device_info *device_info)
254 struct nvsp_1_message_send_receive_buffer_complete *resp;
255 struct net_device *ndev = hv_get_drvdata(device);
256 struct nvsp_message *init_packet;
257 unsigned int buf_size;
261 /* Get receive buffer area. */
262 buf_size = device_info->recv_sections * device_info->recv_section_size;
263 buf_size = roundup(buf_size, PAGE_SIZE);
265 /* Legacy hosts only allow smaller receive buffer */
266 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
267 buf_size = min_t(unsigned int, buf_size,
268 NETVSC_RECEIVE_BUFFER_SIZE_LEGACY);
270 net_device->recv_buf = vzalloc(buf_size);
271 if (!net_device->recv_buf) {
273 "unable to allocate receive buffer of size %u\n",
280 * Establish the gpadl handle for this buffer on this
281 * channel. Note: This call uses the vmbus connection rather
282 * than the channel to establish the gpadl handle.
284 ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
286 &net_device->recv_buf_gpadl_handle);
289 "unable to establish receive buffer's gpadl\n");
293 /* Notify the NetVsp of the gpadl handle */
294 init_packet = &net_device->channel_init_pkt;
295 memset(init_packet, 0, sizeof(struct nvsp_message));
296 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
297 init_packet->msg.v1_msg.send_recv_buf.
298 gpadl_handle = net_device->recv_buf_gpadl_handle;
299 init_packet->msg.v1_msg.
300 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
302 /* Send the gpadl notification request */
303 ret = vmbus_sendpacket(device->channel, init_packet,
304 sizeof(struct nvsp_message),
305 (unsigned long)init_packet,
307 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
310 "unable to send receive buffer's gpadl to netvsp\n");
314 wait_for_completion(&net_device->channel_init_wait);
316 /* Check the response */
317 resp = &init_packet->msg.v1_msg.send_recv_buf_complete;
318 if (resp->status != NVSP_STAT_SUCCESS) {
320 "Unable to complete receive buffer initialization with NetVsp - status %d\n",
326 /* Parse the response */
327 netdev_dbg(ndev, "Receive sections: %u sub_allocs: size %u count: %u\n",
328 resp->num_sections, resp->sections[0].sub_alloc_size,
329 resp->sections[0].num_sub_allocs);
331 /* There should only be one section for the entire receive buffer */
332 if (resp->num_sections != 1 || resp->sections[0].offset != 0) {
337 net_device->recv_section_size = resp->sections[0].sub_alloc_size;
338 net_device->recv_section_cnt = resp->sections[0].num_sub_allocs;
340 /* Setup receive completion ring */
341 net_device->recv_completion_cnt
342 = round_up(net_device->recv_section_cnt + 1,
343 PAGE_SIZE / sizeof(u64));
344 ret = netvsc_alloc_recv_comp_ring(net_device, 0);
348 /* Now setup the send buffer. */
349 buf_size = device_info->send_sections * device_info->send_section_size;
350 buf_size = round_up(buf_size, PAGE_SIZE);
352 net_device->send_buf = vzalloc(buf_size);
353 if (!net_device->send_buf) {
354 netdev_err(ndev, "unable to allocate send buffer of size %u\n",
360 /* Establish the gpadl handle for this buffer on this
361 * channel. Note: This call uses the vmbus connection rather
362 * than the channel to establish the gpadl handle.
364 ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
366 &net_device->send_buf_gpadl_handle);
369 "unable to establish send buffer's gpadl\n");
373 /* Notify the NetVsp of the gpadl handle */
374 init_packet = &net_device->channel_init_pkt;
375 memset(init_packet, 0, sizeof(struct nvsp_message));
376 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
377 init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
378 net_device->send_buf_gpadl_handle;
379 init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
381 /* Send the gpadl notification request */
382 ret = vmbus_sendpacket(device->channel, init_packet,
383 sizeof(struct nvsp_message),
384 (unsigned long)init_packet,
386 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
389 "unable to send send buffer's gpadl to netvsp\n");
393 wait_for_completion(&net_device->channel_init_wait);
395 /* Check the response */
396 if (init_packet->msg.v1_msg.
397 send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
398 netdev_err(ndev, "Unable to complete send buffer "
399 "initialization with NetVsp - status %d\n",
400 init_packet->msg.v1_msg.
401 send_send_buf_complete.status);
406 /* Parse the response */
407 net_device->send_section_size = init_packet->msg.
408 v1_msg.send_send_buf_complete.section_size;
410 /* Section count is simply the size divided by the section size. */
411 net_device->send_section_cnt = buf_size / net_device->send_section_size;
413 netdev_dbg(ndev, "Send section size: %d, Section count:%d\n",
414 net_device->send_section_size, net_device->send_section_cnt);
416 /* Setup state for managing the send buffer. */
417 map_words = DIV_ROUND_UP(net_device->send_section_cnt, BITS_PER_LONG);
419 net_device->send_section_map = kcalloc(map_words, sizeof(ulong), GFP_KERNEL);
420 if (net_device->send_section_map == NULL) {
428 netvsc_destroy_buf(device);
434 /* Negotiate NVSP protocol version */
435 static int negotiate_nvsp_ver(struct hv_device *device,
436 struct netvsc_device *net_device,
437 struct nvsp_message *init_packet,
440 struct net_device *ndev = hv_get_drvdata(device);
443 memset(init_packet, 0, sizeof(struct nvsp_message));
444 init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
445 init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
446 init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
448 /* Send the init request */
449 ret = vmbus_sendpacket(device->channel, init_packet,
450 sizeof(struct nvsp_message),
451 (unsigned long)init_packet,
453 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
458 wait_for_completion(&net_device->channel_init_wait);
460 if (init_packet->msg.init_msg.init_complete.status !=
464 if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
467 /* NVSPv2 or later: Send NDIS config */
468 memset(init_packet, 0, sizeof(struct nvsp_message));
469 init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
470 init_packet->msg.v2_msg.send_ndis_config.mtu = ndev->mtu + ETH_HLEN;
471 init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
473 if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5) {
474 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
476 /* Teaming bit is needed to receive link speed updates */
477 init_packet->msg.v2_msg.send_ndis_config.capability.teaming = 1;
480 ret = vmbus_sendpacket(device->channel, init_packet,
481 sizeof(struct nvsp_message),
482 (unsigned long)init_packet,
483 VM_PKT_DATA_INBAND, 0);
488 static int netvsc_connect_vsp(struct hv_device *device,
489 struct netvsc_device *net_device,
490 const struct netvsc_device_info *device_info)
492 const u32 ver_list[] = {
493 NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
494 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5
496 struct nvsp_message *init_packet;
497 int ndis_version, i, ret;
499 init_packet = &net_device->channel_init_pkt;
501 /* Negotiate the latest NVSP protocol supported */
502 for (i = ARRAY_SIZE(ver_list) - 1; i >= 0; i--)
503 if (negotiate_nvsp_ver(device, net_device, init_packet,
505 net_device->nvsp_version = ver_list[i];
514 pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
516 /* Send the ndis version */
517 memset(init_packet, 0, sizeof(struct nvsp_message));
519 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
520 ndis_version = 0x00060001;
522 ndis_version = 0x0006001e;
524 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
525 init_packet->msg.v1_msg.
526 send_ndis_ver.ndis_major_ver =
527 (ndis_version & 0xFFFF0000) >> 16;
528 init_packet->msg.v1_msg.
529 send_ndis_ver.ndis_minor_ver =
530 ndis_version & 0xFFFF;
532 /* Send the init request */
533 ret = vmbus_sendpacket(device->channel, init_packet,
534 sizeof(struct nvsp_message),
535 (unsigned long)init_packet,
536 VM_PKT_DATA_INBAND, 0);
541 ret = netvsc_init_buf(device, net_device, device_info);
547 static void netvsc_disconnect_vsp(struct hv_device *device)
549 netvsc_destroy_buf(device);
553 * netvsc_device_remove - Callback when the root bus device is removed
555 void netvsc_device_remove(struct hv_device *device)
557 struct net_device *ndev = hv_get_drvdata(device);
558 struct net_device_context *net_device_ctx = netdev_priv(ndev);
559 struct netvsc_device *net_device
560 = rtnl_dereference(net_device_ctx->nvdev);
563 cancel_work_sync(&net_device->subchan_work);
565 netvsc_disconnect_vsp(device);
567 RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
570 * At this point, no one should be accessing net_device
573 netdev_dbg(ndev, "net device safe to remove\n");
575 /* Now, we can close the channel safely */
576 vmbus_close(device->channel);
578 /* And dissassociate NAPI context from device */
579 for (i = 0; i < net_device->num_chn; i++)
580 netif_napi_del(&net_device->chan_table[i].napi);
582 /* Release all resources */
583 free_netvsc_device_rcu(net_device);
586 #define RING_AVAIL_PERCENT_HIWATER 20
587 #define RING_AVAIL_PERCENT_LOWATER 10
590 * Get the percentage of available bytes to write in the ring.
591 * The return value is in range from 0 to 100.
593 static inline u32 hv_ringbuf_avail_percent(
594 struct hv_ring_buffer_info *ring_info)
596 u32 avail_read, avail_write;
598 hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
600 return avail_write * 100 / ring_info->ring_datasize;
603 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
606 sync_change_bit(index, net_device->send_section_map);
609 static void netvsc_send_tx_complete(struct netvsc_device *net_device,
610 struct vmbus_channel *incoming_channel,
611 struct hv_device *device,
612 const struct vmpacket_descriptor *desc,
615 struct sk_buff *skb = (struct sk_buff *)(unsigned long)desc->trans_id;
616 struct net_device *ndev = hv_get_drvdata(device);
617 struct vmbus_channel *channel = device->channel;
621 /* Notify the layer above us */
623 const struct hv_netvsc_packet *packet
624 = (struct hv_netvsc_packet *)skb->cb;
625 u32 send_index = packet->send_buf_index;
626 struct netvsc_stats *tx_stats;
628 if (send_index != NETVSC_INVALID_INDEX)
629 netvsc_free_send_slot(net_device, send_index);
630 q_idx = packet->q_idx;
631 channel = incoming_channel;
633 tx_stats = &net_device->chan_table[q_idx].tx_stats;
635 u64_stats_update_begin(&tx_stats->syncp);
636 tx_stats->packets += packet->total_packets;
637 tx_stats->bytes += packet->total_bytes;
638 u64_stats_update_end(&tx_stats->syncp);
640 napi_consume_skb(skb, budget);
644 atomic_dec_return(&net_device->chan_table[q_idx].queue_sends);
646 if (net_device->destroy && queue_sends == 0)
647 wake_up(&net_device->wait_drain);
649 if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev, q_idx)) &&
650 (hv_ringbuf_avail_percent(&channel->outbound) > RING_AVAIL_PERCENT_HIWATER ||
652 netif_tx_wake_queue(netdev_get_tx_queue(ndev, q_idx));
655 static void netvsc_send_completion(struct netvsc_device *net_device,
656 struct vmbus_channel *incoming_channel,
657 struct hv_device *device,
658 const struct vmpacket_descriptor *desc,
661 struct nvsp_message *nvsp_packet = hv_pkt_data(desc);
662 struct net_device *ndev = hv_get_drvdata(device);
664 switch (nvsp_packet->hdr.msg_type) {
665 case NVSP_MSG_TYPE_INIT_COMPLETE:
666 case NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE:
667 case NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE:
668 case NVSP_MSG5_TYPE_SUBCHANNEL:
669 /* Copy the response back */
670 memcpy(&net_device->channel_init_pkt, nvsp_packet,
671 sizeof(struct nvsp_message));
672 complete(&net_device->channel_init_wait);
675 case NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE:
676 netvsc_send_tx_complete(net_device, incoming_channel,
677 device, desc, budget);
682 "Unknown send completion type %d received!!\n",
683 nvsp_packet->hdr.msg_type);
687 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
689 unsigned long *map_addr = net_device->send_section_map;
692 for_each_clear_bit(i, map_addr, net_device->send_section_cnt) {
693 if (sync_test_and_set_bit(i, map_addr) == 0)
697 return NETVSC_INVALID_INDEX;
700 static u32 netvsc_copy_to_send_buf(struct netvsc_device *net_device,
701 unsigned int section_index,
703 struct hv_netvsc_packet *packet,
704 struct rndis_message *rndis_msg,
705 struct hv_page_buffer *pb,
708 char *start = net_device->send_buf;
709 char *dest = start + (section_index * net_device->send_section_size)
714 u32 remain = packet->total_data_buflen % net_device->pkt_align;
715 u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
716 packet->page_buf_cnt;
719 if (skb->xmit_more && remain && !packet->cp_partial) {
720 padding = net_device->pkt_align - remain;
721 rndis_msg->msg_len += padding;
722 packet->total_data_buflen += padding;
725 for (i = 0; i < page_count; i++) {
726 char *src = phys_to_virt(pb[i].pfn << PAGE_SHIFT);
727 u32 offset = pb[i].offset;
730 memcpy(dest, (src + offset), len);
736 memset(dest, 0, padding);
743 static inline int netvsc_send_pkt(
744 struct hv_device *device,
745 struct hv_netvsc_packet *packet,
746 struct netvsc_device *net_device,
747 struct hv_page_buffer *pb,
750 struct nvsp_message nvmsg;
751 struct nvsp_1_message_send_rndis_packet * const rpkt =
752 &nvmsg.msg.v1_msg.send_rndis_pkt;
753 struct netvsc_channel * const nvchan =
754 &net_device->chan_table[packet->q_idx];
755 struct vmbus_channel *out_channel = nvchan->channel;
756 struct net_device *ndev = hv_get_drvdata(device);
757 struct netdev_queue *txq = netdev_get_tx_queue(ndev, packet->q_idx);
760 u32 ring_avail = hv_ringbuf_avail_percent(&out_channel->outbound);
762 nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
764 rpkt->channel_type = 0; /* 0 is RMC_DATA */
766 rpkt->channel_type = 1; /* 1 is RMC_CONTROL */
768 rpkt->send_buf_section_index = packet->send_buf_index;
769 if (packet->send_buf_index == NETVSC_INVALID_INDEX)
770 rpkt->send_buf_section_size = 0;
772 rpkt->send_buf_section_size = packet->total_data_buflen;
776 if (out_channel->rescind)
779 if (packet->page_buf_cnt) {
780 if (packet->cp_partial)
781 pb += packet->rmsg_pgcnt;
783 ret = vmbus_sendpacket_pagebuffer(out_channel,
784 pb, packet->page_buf_cnt,
785 &nvmsg, sizeof(nvmsg),
788 ret = vmbus_sendpacket(out_channel,
789 &nvmsg, sizeof(nvmsg),
790 req_id, VM_PKT_DATA_INBAND,
791 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
795 atomic_inc_return(&nvchan->queue_sends);
797 if (ring_avail < RING_AVAIL_PERCENT_LOWATER)
798 netif_tx_stop_queue(txq);
799 } else if (ret == -EAGAIN) {
800 netif_tx_stop_queue(txq);
801 if (atomic_read(&nvchan->queue_sends) < 1) {
802 netif_tx_wake_queue(txq);
807 "Unable to send packet pages %u len %u, ret %d\n",
808 packet->page_buf_cnt, packet->total_data_buflen,
815 /* Move packet out of multi send data (msd), and clear msd */
816 static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
817 struct sk_buff **msd_skb,
818 struct multi_send_data *msdp)
820 *msd_skb = msdp->skb;
821 *msd_send = msdp->pkt;
827 /* RCU already held by caller */
828 int netvsc_send(struct net_device_context *ndev_ctx,
829 struct hv_netvsc_packet *packet,
830 struct rndis_message *rndis_msg,
831 struct hv_page_buffer *pb,
834 struct netvsc_device *net_device
835 = rcu_dereference_bh(ndev_ctx->nvdev);
836 struct hv_device *device = ndev_ctx->device_ctx;
838 struct netvsc_channel *nvchan;
839 u32 pktlen = packet->total_data_buflen, msd_len = 0;
840 unsigned int section_index = NETVSC_INVALID_INDEX;
841 struct multi_send_data *msdp;
842 struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
843 struct sk_buff *msd_skb = NULL;
845 bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
847 /* If device is rescinded, return error and packet will get dropped. */
848 if (unlikely(!net_device || net_device->destroy))
851 /* We may race with netvsc_connect_vsp()/netvsc_init_buf() and get
852 * here before the negotiation with the host is finished and
853 * send_section_map may not be allocated yet.
855 if (unlikely(!net_device->send_section_map))
858 nvchan = &net_device->chan_table[packet->q_idx];
859 packet->send_buf_index = NETVSC_INVALID_INDEX;
860 packet->cp_partial = false;
862 /* Send control message directly without accessing msd (Multi-Send
863 * Data) field which may be changed during data packet processing.
870 /* batch packets in send buffer if possible */
873 msd_len = msdp->pkt->total_data_buflen;
875 try_batch = msd_len > 0 && msdp->count < net_device->max_pkt;
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 (pktlen + net_device->pkt_align <
886 net_device->send_section_size) {
887 section_index = netvsc_get_next_send_section(net_device);
888 if (unlikely(section_index == NETVSC_INVALID_INDEX)) {
889 ++ndev_ctx->eth_stats.tx_send_full;
891 move_pkt_msd(&msd_send, &msd_skb, msdp);
896 if (section_index != NETVSC_INVALID_INDEX) {
897 netvsc_copy_to_send_buf(net_device,
898 section_index, msd_len,
899 packet, rndis_msg, pb, skb);
901 packet->send_buf_index = section_index;
903 if (packet->cp_partial) {
904 packet->page_buf_cnt -= packet->rmsg_pgcnt;
905 packet->total_data_buflen = msd_len + packet->rmsg_size;
907 packet->page_buf_cnt = 0;
908 packet->total_data_buflen += msd_len;
912 packet->total_packets += msdp->pkt->total_packets;
913 packet->total_bytes += msdp->pkt->total_bytes;
917 dev_consume_skb_any(msdp->skb);
919 if (xmit_more && !packet->cp_partial) {
930 move_pkt_msd(&msd_send, &msd_skb, msdp);
935 int m_ret = netvsc_send_pkt(device, msd_send, net_device,
939 netvsc_free_send_slot(net_device,
940 msd_send->send_buf_index);
941 dev_kfree_skb_any(msd_skb);
947 ret = netvsc_send_pkt(device, cur_send, net_device, pb, skb);
949 if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
950 netvsc_free_send_slot(net_device, section_index);
955 /* Send pending recv completions */
956 static int send_recv_completions(struct net_device *ndev,
957 struct netvsc_device *nvdev,
958 struct netvsc_channel *nvchan)
960 struct multi_recv_comp *mrc = &nvchan->mrc;
961 struct recv_comp_msg {
962 struct nvsp_message_header hdr;
965 struct recv_comp_msg msg = {
966 .hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE,
970 while (mrc->first != mrc->next) {
971 const struct recv_comp_data *rcd
972 = mrc->slots + mrc->first;
974 msg.status = rcd->status;
975 ret = vmbus_sendpacket(nvchan->channel, &msg, sizeof(msg),
976 rcd->tid, VM_PKT_COMP, 0);
978 struct net_device_context *ndev_ctx = netdev_priv(ndev);
980 ++ndev_ctx->eth_stats.rx_comp_busy;
984 if (++mrc->first == nvdev->recv_completion_cnt)
988 /* receive completion ring has been emptied */
989 if (unlikely(nvdev->destroy))
990 wake_up(&nvdev->wait_drain);
995 /* Count how many receive completions are outstanding */
996 static void recv_comp_slot_avail(const struct netvsc_device *nvdev,
997 const struct multi_recv_comp *mrc,
998 u32 *filled, u32 *avail)
1000 u32 count = nvdev->recv_completion_cnt;
1002 if (mrc->next >= mrc->first)
1003 *filled = mrc->next - mrc->first;
1005 *filled = (count - mrc->first) + mrc->next;
1007 *avail = count - *filled - 1;
1010 /* Add receive complete to ring to send to host. */
1011 static void enq_receive_complete(struct net_device *ndev,
1012 struct netvsc_device *nvdev, u16 q_idx,
1013 u64 tid, u32 status)
1015 struct netvsc_channel *nvchan = &nvdev->chan_table[q_idx];
1016 struct multi_recv_comp *mrc = &nvchan->mrc;
1017 struct recv_comp_data *rcd;
1020 recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1022 if (unlikely(filled > NAPI_POLL_WEIGHT)) {
1023 send_recv_completions(ndev, nvdev, nvchan);
1024 recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1027 if (unlikely(!avail)) {
1028 netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
1033 rcd = mrc->slots + mrc->next;
1035 rcd->status = status;
1037 if (++mrc->next == nvdev->recv_completion_cnt)
1041 static int netvsc_receive(struct net_device *ndev,
1042 struct netvsc_device *net_device,
1043 struct net_device_context *net_device_ctx,
1044 struct hv_device *device,
1045 struct vmbus_channel *channel,
1046 const struct vmpacket_descriptor *desc,
1047 struct nvsp_message *nvsp)
1049 const struct vmtransfer_page_packet_header *vmxferpage_packet
1050 = container_of(desc, const struct vmtransfer_page_packet_header, d);
1051 u16 q_idx = channel->offermsg.offer.sub_channel_index;
1052 char *recv_buf = net_device->recv_buf;
1053 u32 status = NVSP_STAT_SUCCESS;
1057 /* Make sure this is a valid nvsp packet */
1058 if (unlikely(nvsp->hdr.msg_type != NVSP_MSG1_TYPE_SEND_RNDIS_PKT)) {
1059 netif_err(net_device_ctx, rx_err, ndev,
1060 "Unknown nvsp packet type received %u\n",
1061 nvsp->hdr.msg_type);
1065 if (unlikely(vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID)) {
1066 netif_err(net_device_ctx, rx_err, ndev,
1067 "Invalid xfer page set id - expecting %x got %x\n",
1068 NETVSC_RECEIVE_BUFFER_ID,
1069 vmxferpage_packet->xfer_pageset_id);
1073 count = vmxferpage_packet->range_cnt;
1075 /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1076 for (i = 0; i < count; i++) {
1077 void *data = recv_buf
1078 + vmxferpage_packet->ranges[i].byte_offset;
1079 u32 buflen = vmxferpage_packet->ranges[i].byte_count;
1081 /* Pass it to the upper layer */
1082 status = rndis_filter_receive(ndev, net_device, device,
1083 channel, data, buflen);
1086 enq_receive_complete(ndev, net_device, q_idx,
1087 vmxferpage_packet->d.trans_id, status);
1092 static void netvsc_send_table(struct hv_device *hdev,
1093 struct nvsp_message *nvmsg)
1095 struct net_device *ndev = hv_get_drvdata(hdev);
1096 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1100 count = nvmsg->msg.v5_msg.send_table.count;
1101 if (count != VRSS_SEND_TAB_SIZE) {
1102 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1106 tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
1107 nvmsg->msg.v5_msg.send_table.offset);
1109 for (i = 0; i < count; i++)
1110 net_device_ctx->tx_send_table[i] = tab[i];
1113 static void netvsc_send_vf(struct net_device_context *net_device_ctx,
1114 struct nvsp_message *nvmsg)
1116 net_device_ctx->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1117 net_device_ctx->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1120 static inline void netvsc_receive_inband(struct hv_device *hdev,
1121 struct net_device_context *net_device_ctx,
1122 struct nvsp_message *nvmsg)
1124 switch (nvmsg->hdr.msg_type) {
1125 case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1126 netvsc_send_table(hdev, nvmsg);
1129 case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1130 netvsc_send_vf(net_device_ctx, nvmsg);
1135 static int netvsc_process_raw_pkt(struct hv_device *device,
1136 struct vmbus_channel *channel,
1137 struct netvsc_device *net_device,
1138 struct net_device *ndev,
1139 const struct vmpacket_descriptor *desc,
1142 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1143 struct nvsp_message *nvmsg = hv_pkt_data(desc);
1145 switch (desc->type) {
1147 netvsc_send_completion(net_device, channel, device,
1151 case VM_PKT_DATA_USING_XFER_PAGES:
1152 return netvsc_receive(ndev, net_device, net_device_ctx,
1153 device, channel, desc, nvmsg);
1156 case VM_PKT_DATA_INBAND:
1157 netvsc_receive_inband(device, net_device_ctx, nvmsg);
1161 netdev_err(ndev, "unhandled packet type %d, tid %llx\n",
1162 desc->type, desc->trans_id);
1169 static struct hv_device *netvsc_channel_to_device(struct vmbus_channel *channel)
1171 struct vmbus_channel *primary = channel->primary_channel;
1173 return primary ? primary->device_obj : channel->device_obj;
1176 /* Network processing softirq
1177 * Process data in incoming ring buffer from host
1178 * Stops when ring is empty or budget is met or exceeded.
1180 int netvsc_poll(struct napi_struct *napi, int budget)
1182 struct netvsc_channel *nvchan
1183 = container_of(napi, struct netvsc_channel, napi);
1184 struct netvsc_device *net_device = nvchan->net_device;
1185 struct vmbus_channel *channel = nvchan->channel;
1186 struct hv_device *device = netvsc_channel_to_device(channel);
1187 struct net_device *ndev = hv_get_drvdata(device);
1190 /* If starting a new interval */
1192 nvchan->desc = hv_pkt_iter_first(channel);
1194 while (nvchan->desc && work_done < budget) {
1195 work_done += netvsc_process_raw_pkt(device, channel, net_device,
1196 ndev, nvchan->desc, budget);
1197 nvchan->desc = hv_pkt_iter_next(channel, nvchan->desc);
1200 /* If send of pending receive completions suceeded
1201 * and did not exhaust NAPI budget this time
1202 * and not doing busy poll
1203 * then re-enable host interrupts
1204 * and reschedule if ring is not empty.
1206 if (send_recv_completions(ndev, net_device, nvchan) == 0 &&
1207 work_done < budget &&
1208 napi_complete_done(napi, work_done) &&
1209 hv_end_read(&channel->inbound)) {
1210 hv_begin_read(&channel->inbound);
1211 napi_reschedule(napi);
1214 /* Driver may overshoot since multiple packets per descriptor */
1215 return min(work_done, budget);
1218 /* Call back when data is available in host ring buffer.
1219 * Processing is deferred until network softirq (NAPI)
1221 void netvsc_channel_cb(void *context)
1223 struct netvsc_channel *nvchan = context;
1224 struct vmbus_channel *channel = nvchan->channel;
1225 struct hv_ring_buffer_info *rbi = &channel->inbound;
1227 /* preload first vmpacket descriptor */
1228 prefetch(hv_get_ring_buffer(rbi) + rbi->priv_read_index);
1230 if (napi_schedule_prep(&nvchan->napi)) {
1231 /* disable interupts from host */
1234 __napi_schedule(&nvchan->napi);
1239 * netvsc_device_add - Callback when the device belonging to this
1242 struct netvsc_device *netvsc_device_add(struct hv_device *device,
1243 const struct netvsc_device_info *device_info)
1246 int ring_size = device_info->ring_size;
1247 struct netvsc_device *net_device;
1248 struct net_device *ndev = hv_get_drvdata(device);
1249 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1251 net_device = alloc_net_device();
1253 return ERR_PTR(-ENOMEM);
1255 net_device->ring_size = ring_size;
1257 /* Because the device uses NAPI, all the interrupt batching and
1258 * control is done via Net softirq, not the channel handling
1260 set_channel_read_mode(device->channel, HV_CALL_ISR);
1262 /* If we're reopening the device we may have multiple queues, fill the
1263 * chn_table with the default channel to use it before subchannels are
1265 * Initialize the channel state before we open;
1266 * we can be interrupted as soon as we open the channel.
1269 for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
1270 struct netvsc_channel *nvchan = &net_device->chan_table[i];
1272 nvchan->channel = device->channel;
1273 nvchan->net_device = net_device;
1274 u64_stats_init(&nvchan->tx_stats.syncp);
1275 u64_stats_init(&nvchan->rx_stats.syncp);
1278 /* Enable NAPI handler before init callbacks */
1279 netif_napi_add(ndev, &net_device->chan_table[0].napi,
1280 netvsc_poll, NAPI_POLL_WEIGHT);
1282 /* Open the channel */
1283 ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
1284 ring_size * PAGE_SIZE, NULL, 0,
1286 net_device->chan_table);
1289 netif_napi_del(&net_device->chan_table[0].napi);
1290 netdev_err(ndev, "unable to open channel: %d\n", ret);
1294 /* Channel is opened */
1295 netdev_dbg(ndev, "hv_netvsc channel opened successfully\n");
1297 napi_enable(&net_device->chan_table[0].napi);
1299 /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1302 rcu_assign_pointer(net_device_ctx->nvdev, net_device);
1304 /* Connect with the NetVsp */
1305 ret = netvsc_connect_vsp(device, net_device, device_info);
1308 "unable to connect to NetVSP - %d\n", ret);
1315 RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
1316 napi_disable(&net_device->chan_table[0].napi);
1318 /* Now, we can close the channel safely */
1319 vmbus_close(device->channel);
1322 free_netvsc_device(&net_device->rcu);
1324 return ERR_PTR(ret);