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 net_device->recv_section_size = NETVSC_RECV_SECTION_SIZE;
80 net_device->send_section_size = NETVSC_SEND_SECTION_SIZE;
82 init_completion(&net_device->channel_init_wait);
83 init_waitqueue_head(&net_device->subchan_open);
84 INIT_WORK(&net_device->subchan_work, rndis_set_subchannel);
89 static void free_netvsc_device(struct rcu_head *head)
91 struct netvsc_device *nvdev
92 = container_of(head, struct netvsc_device, rcu);
95 for (i = 0; i < VRSS_CHANNEL_MAX; i++)
96 vfree(nvdev->chan_table[i].mrc.slots);
101 static void free_netvsc_device_rcu(struct netvsc_device *nvdev)
103 call_rcu(&nvdev->rcu, free_netvsc_device);
106 static void netvsc_destroy_buf(struct hv_device *device)
108 struct nvsp_message *revoke_packet;
109 struct net_device *ndev = hv_get_drvdata(device);
110 struct net_device_context *ndc = netdev_priv(ndev);
111 struct netvsc_device *net_device = rtnl_dereference(ndc->nvdev);
115 * If we got a section count, it means we received a
116 * SendReceiveBufferComplete msg (ie sent
117 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
118 * to send a revoke msg here
120 if (net_device->recv_section_cnt) {
121 /* Send the revoke receive buffer */
122 revoke_packet = &net_device->revoke_packet;
123 memset(revoke_packet, 0, sizeof(struct nvsp_message));
125 revoke_packet->hdr.msg_type =
126 NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
127 revoke_packet->msg.v1_msg.
128 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
130 ret = vmbus_sendpacket(device->channel,
132 sizeof(struct nvsp_message),
133 (unsigned long)revoke_packet,
134 VM_PKT_DATA_INBAND, 0);
135 /* If the failure is because the channel is rescinded;
136 * ignore the failure since we cannot send on a rescinded
137 * channel. This would allow us to properly cleanup
138 * even when the channel is rescinded.
140 if (device->channel->rescind)
143 * If we failed here, we might as well return and
144 * have a leak rather than continue and a bugchk
147 netdev_err(ndev, "unable to send "
148 "revoke receive buffer to netvsp\n");
151 net_device->recv_section_cnt = 0;
154 /* Teardown the gpadl on the vsp end */
155 if (net_device->recv_buf_gpadl_handle) {
156 ret = vmbus_teardown_gpadl(device->channel,
157 net_device->recv_buf_gpadl_handle);
159 /* If we failed here, we might as well return and have a leak
160 * rather than continue and a bugchk
164 "unable to teardown receive buffer's gpadl\n");
167 net_device->recv_buf_gpadl_handle = 0;
170 if (net_device->recv_buf) {
171 /* Free up the receive buffer */
172 vfree(net_device->recv_buf);
173 net_device->recv_buf = NULL;
176 /* Deal with the send buffer we may have setup.
177 * If we got a send section size, it means we received a
178 * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
179 * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
180 * to send a revoke msg here
182 if (net_device->send_section_cnt) {
183 /* Send the revoke receive buffer */
184 revoke_packet = &net_device->revoke_packet;
185 memset(revoke_packet, 0, sizeof(struct nvsp_message));
187 revoke_packet->hdr.msg_type =
188 NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
189 revoke_packet->msg.v1_msg.revoke_send_buf.id =
190 NETVSC_SEND_BUFFER_ID;
192 ret = vmbus_sendpacket(device->channel,
194 sizeof(struct nvsp_message),
195 (unsigned long)revoke_packet,
196 VM_PKT_DATA_INBAND, 0);
198 /* If the failure is because the channel is rescinded;
199 * ignore the failure since we cannot send on a rescinded
200 * channel. This would allow us to properly cleanup
201 * even when the channel is rescinded.
203 if (device->channel->rescind)
206 /* If we failed here, we might as well return and
207 * have a leak rather than continue and a bugchk
210 netdev_err(ndev, "unable to send "
211 "revoke send buffer to netvsp\n");
214 net_device->send_section_cnt = 0;
216 /* Teardown the gpadl on the vsp end */
217 if (net_device->send_buf_gpadl_handle) {
218 ret = vmbus_teardown_gpadl(device->channel,
219 net_device->send_buf_gpadl_handle);
221 /* If we failed here, we might as well return and have a leak
222 * rather than continue and a bugchk
226 "unable to teardown send buffer's gpadl\n");
229 net_device->send_buf_gpadl_handle = 0;
231 if (net_device->send_buf) {
232 /* Free up the send buffer */
233 vfree(net_device->send_buf);
234 net_device->send_buf = NULL;
236 kfree(net_device->send_section_map);
239 int netvsc_alloc_recv_comp_ring(struct netvsc_device *net_device, u32 q_idx)
241 struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
242 int node = cpu_to_node(nvchan->channel->target_cpu);
245 size = net_device->recv_completion_cnt * sizeof(struct recv_comp_data);
246 nvchan->mrc.slots = vzalloc_node(size, node);
247 if (!nvchan->mrc.slots)
248 nvchan->mrc.slots = vzalloc(size);
250 return nvchan->mrc.slots ? 0 : -ENOMEM;
253 static int netvsc_init_buf(struct hv_device *device,
254 struct netvsc_device *net_device,
255 const struct netvsc_device_info *device_info)
257 struct nvsp_1_message_send_receive_buffer_complete *resp;
258 struct net_device *ndev = hv_get_drvdata(device);
259 struct nvsp_message *init_packet;
260 unsigned int buf_size;
264 /* Get receive buffer area. */
265 buf_size = device_info->recv_sections * net_device->recv_section_size;
266 buf_size = roundup(buf_size, PAGE_SIZE);
268 net_device->recv_buf = vzalloc(buf_size);
269 if (!net_device->recv_buf) {
271 "unable to allocate receive buffer of size %u\n",
278 * Establish the gpadl handle for this buffer on this
279 * channel. Note: This call uses the vmbus connection rather
280 * than the channel to establish the gpadl handle.
282 ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
284 &net_device->recv_buf_gpadl_handle);
287 "unable to establish receive buffer's gpadl\n");
291 /* Notify the NetVsp of the gpadl handle */
292 init_packet = &net_device->channel_init_pkt;
293 memset(init_packet, 0, sizeof(struct nvsp_message));
294 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
295 init_packet->msg.v1_msg.send_recv_buf.
296 gpadl_handle = net_device->recv_buf_gpadl_handle;
297 init_packet->msg.v1_msg.
298 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
300 /* Send the gpadl notification request */
301 ret = vmbus_sendpacket(device->channel, init_packet,
302 sizeof(struct nvsp_message),
303 (unsigned long)init_packet,
305 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
308 "unable to send receive buffer's gpadl to netvsp\n");
312 wait_for_completion(&net_device->channel_init_wait);
314 /* Check the response */
315 resp = &init_packet->msg.v1_msg.send_recv_buf_complete;
316 if (resp->status != NVSP_STAT_SUCCESS) {
318 "Unable to complete receive buffer initialization with NetVsp - status %d\n",
324 /* Parse the response */
325 netdev_dbg(ndev, "Receive sections: %u sub_allocs: size %u count: %u\n",
326 resp->num_sections, resp->sections[0].sub_alloc_size,
327 resp->sections[0].num_sub_allocs);
329 /* There should only be one section for the entire receive buffer */
330 if (resp->num_sections != 1 || resp->sections[0].offset != 0) {
335 net_device->recv_section_size = resp->sections[0].sub_alloc_size;
336 net_device->recv_section_cnt = resp->sections[0].num_sub_allocs;
338 /* Setup receive completion ring */
339 net_device->recv_completion_cnt
340 = round_up(net_device->recv_section_cnt + 1,
341 PAGE_SIZE / sizeof(u64));
342 ret = netvsc_alloc_recv_comp_ring(net_device, 0);
346 /* Now setup the send buffer. */
347 buf_size = device_info->send_sections * net_device->send_section_size;
348 buf_size = round_up(buf_size, PAGE_SIZE);
350 net_device->send_buf = vzalloc(buf_size);
351 if (!net_device->send_buf) {
352 netdev_err(ndev, "unable to allocate send buffer of size %u\n",
358 /* Establish the gpadl handle for this buffer on this
359 * channel. Note: This call uses the vmbus connection rather
360 * than the channel to establish the gpadl handle.
362 ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
364 &net_device->send_buf_gpadl_handle);
367 "unable to establish send buffer's gpadl\n");
371 /* Notify the NetVsp of the gpadl handle */
372 init_packet = &net_device->channel_init_pkt;
373 memset(init_packet, 0, sizeof(struct nvsp_message));
374 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
375 init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
376 net_device->send_buf_gpadl_handle;
377 init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
379 /* Send the gpadl notification request */
380 ret = vmbus_sendpacket(device->channel, init_packet,
381 sizeof(struct nvsp_message),
382 (unsigned long)init_packet,
384 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
387 "unable to send send buffer's gpadl to netvsp\n");
391 wait_for_completion(&net_device->channel_init_wait);
393 /* Check the response */
394 if (init_packet->msg.v1_msg.
395 send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
396 netdev_err(ndev, "Unable to complete send buffer "
397 "initialization with NetVsp - status %d\n",
398 init_packet->msg.v1_msg.
399 send_send_buf_complete.status);
404 /* Parse the response */
405 net_device->send_section_size = init_packet->msg.
406 v1_msg.send_send_buf_complete.section_size;
408 /* Section count is simply the size divided by the section size. */
409 net_device->send_section_cnt = buf_size / net_device->send_section_size;
411 netdev_dbg(ndev, "Send section size: %d, Section count:%d\n",
412 net_device->send_section_size, net_device->send_section_cnt);
414 /* Setup state for managing the send buffer. */
415 map_words = DIV_ROUND_UP(net_device->send_section_cnt, BITS_PER_LONG);
417 net_device->send_section_map = kcalloc(map_words, sizeof(ulong), GFP_KERNEL);
418 if (net_device->send_section_map == NULL) {
426 netvsc_destroy_buf(device);
432 /* Negotiate NVSP protocol version */
433 static int negotiate_nvsp_ver(struct hv_device *device,
434 struct netvsc_device *net_device,
435 struct nvsp_message *init_packet,
438 struct net_device *ndev = hv_get_drvdata(device);
441 memset(init_packet, 0, sizeof(struct nvsp_message));
442 init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
443 init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
444 init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
446 /* Send the init request */
447 ret = vmbus_sendpacket(device->channel, init_packet,
448 sizeof(struct nvsp_message),
449 (unsigned long)init_packet,
451 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
456 wait_for_completion(&net_device->channel_init_wait);
458 if (init_packet->msg.init_msg.init_complete.status !=
462 if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
465 /* NVSPv2 or later: Send NDIS config */
466 memset(init_packet, 0, sizeof(struct nvsp_message));
467 init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
468 init_packet->msg.v2_msg.send_ndis_config.mtu = ndev->mtu + ETH_HLEN;
469 init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
471 if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5) {
472 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
474 /* Teaming bit is needed to receive link speed updates */
475 init_packet->msg.v2_msg.send_ndis_config.capability.teaming = 1;
478 ret = vmbus_sendpacket(device->channel, init_packet,
479 sizeof(struct nvsp_message),
480 (unsigned long)init_packet,
481 VM_PKT_DATA_INBAND, 0);
486 static int netvsc_connect_vsp(struct hv_device *device,
487 struct netvsc_device *net_device,
488 const struct netvsc_device_info *device_info)
490 const u32 ver_list[] = {
491 NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
492 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5
494 struct nvsp_message *init_packet;
495 int ndis_version, i, ret;
497 init_packet = &net_device->channel_init_pkt;
499 /* Negotiate the latest NVSP protocol supported */
500 for (i = ARRAY_SIZE(ver_list) - 1; i >= 0; i--)
501 if (negotiate_nvsp_ver(device, net_device, init_packet,
503 net_device->nvsp_version = ver_list[i];
512 pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
514 /* Send the ndis version */
515 memset(init_packet, 0, sizeof(struct nvsp_message));
517 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
518 ndis_version = 0x00060001;
520 ndis_version = 0x0006001e;
522 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
523 init_packet->msg.v1_msg.
524 send_ndis_ver.ndis_major_ver =
525 (ndis_version & 0xFFFF0000) >> 16;
526 init_packet->msg.v1_msg.
527 send_ndis_ver.ndis_minor_ver =
528 ndis_version & 0xFFFF;
530 /* Send the init request */
531 ret = vmbus_sendpacket(device->channel, init_packet,
532 sizeof(struct nvsp_message),
533 (unsigned long)init_packet,
534 VM_PKT_DATA_INBAND, 0);
539 ret = netvsc_init_buf(device, net_device, device_info);
545 static void netvsc_disconnect_vsp(struct hv_device *device)
547 netvsc_destroy_buf(device);
551 * netvsc_device_remove - Callback when the root bus device is removed
553 void netvsc_device_remove(struct hv_device *device)
555 struct net_device *ndev = hv_get_drvdata(device);
556 struct net_device_context *net_device_ctx = netdev_priv(ndev);
557 struct netvsc_device *net_device
558 = rtnl_dereference(net_device_ctx->nvdev);
561 cancel_work_sync(&net_device->subchan_work);
563 netvsc_disconnect_vsp(device);
565 RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
568 * At this point, no one should be accessing net_device
571 netdev_dbg(ndev, "net device safe to remove\n");
573 /* Now, we can close the channel safely */
574 vmbus_close(device->channel);
576 /* And dissassociate NAPI context from device */
577 for (i = 0; i < net_device->num_chn; i++)
578 netif_napi_del(&net_device->chan_table[i].napi);
580 /* Release all resources */
581 free_netvsc_device_rcu(net_device);
584 #define RING_AVAIL_PERCENT_HIWATER 20
585 #define RING_AVAIL_PERCENT_LOWATER 10
588 * Get the percentage of available bytes to write in the ring.
589 * The return value is in range from 0 to 100.
591 static inline u32 hv_ringbuf_avail_percent(
592 struct hv_ring_buffer_info *ring_info)
594 u32 avail_read, avail_write;
596 hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
598 return avail_write * 100 / ring_info->ring_datasize;
601 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
604 sync_change_bit(index, net_device->send_section_map);
607 static void netvsc_send_tx_complete(struct netvsc_device *net_device,
608 struct vmbus_channel *incoming_channel,
609 struct hv_device *device,
610 const struct vmpacket_descriptor *desc,
613 struct sk_buff *skb = (struct sk_buff *)(unsigned long)desc->trans_id;
614 struct net_device *ndev = hv_get_drvdata(device);
615 struct vmbus_channel *channel = device->channel;
619 /* Notify the layer above us */
621 const struct hv_netvsc_packet *packet
622 = (struct hv_netvsc_packet *)skb->cb;
623 u32 send_index = packet->send_buf_index;
624 struct netvsc_stats *tx_stats;
626 if (send_index != NETVSC_INVALID_INDEX)
627 netvsc_free_send_slot(net_device, send_index);
628 q_idx = packet->q_idx;
629 channel = incoming_channel;
631 tx_stats = &net_device->chan_table[q_idx].tx_stats;
633 u64_stats_update_begin(&tx_stats->syncp);
634 tx_stats->packets += packet->total_packets;
635 tx_stats->bytes += packet->total_bytes;
636 u64_stats_update_end(&tx_stats->syncp);
638 napi_consume_skb(skb, budget);
642 atomic_dec_return(&net_device->chan_table[q_idx].queue_sends);
644 if (net_device->destroy && queue_sends == 0)
645 wake_up(&net_device->wait_drain);
647 if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev, q_idx)) &&
648 (hv_ringbuf_avail_percent(&channel->outbound) > RING_AVAIL_PERCENT_HIWATER ||
650 netif_tx_wake_queue(netdev_get_tx_queue(ndev, q_idx));
653 static void netvsc_send_completion(struct netvsc_device *net_device,
654 struct vmbus_channel *incoming_channel,
655 struct hv_device *device,
656 const struct vmpacket_descriptor *desc,
659 struct nvsp_message *nvsp_packet = hv_pkt_data(desc);
660 struct net_device *ndev = hv_get_drvdata(device);
662 switch (nvsp_packet->hdr.msg_type) {
663 case NVSP_MSG_TYPE_INIT_COMPLETE:
664 case NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE:
665 case NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE:
666 case NVSP_MSG5_TYPE_SUBCHANNEL:
667 /* Copy the response back */
668 memcpy(&net_device->channel_init_pkt, nvsp_packet,
669 sizeof(struct nvsp_message));
670 complete(&net_device->channel_init_wait);
673 case NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE:
674 netvsc_send_tx_complete(net_device, incoming_channel,
675 device, desc, budget);
680 "Unknown send completion type %d received!!\n",
681 nvsp_packet->hdr.msg_type);
685 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
687 unsigned long *map_addr = net_device->send_section_map;
690 for_each_clear_bit(i, map_addr, net_device->send_section_cnt) {
691 if (sync_test_and_set_bit(i, map_addr) == 0)
695 return NETVSC_INVALID_INDEX;
698 static u32 netvsc_copy_to_send_buf(struct netvsc_device *net_device,
699 unsigned int section_index,
701 struct hv_netvsc_packet *packet,
702 struct rndis_message *rndis_msg,
703 struct hv_page_buffer *pb,
706 char *start = net_device->send_buf;
707 char *dest = start + (section_index * net_device->send_section_size)
712 u32 remain = packet->total_data_buflen % net_device->pkt_align;
713 u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
714 packet->page_buf_cnt;
717 if (skb->xmit_more && remain && !packet->cp_partial) {
718 padding = net_device->pkt_align - remain;
719 rndis_msg->msg_len += padding;
720 packet->total_data_buflen += padding;
723 for (i = 0; i < page_count; i++) {
724 char *src = phys_to_virt(pb[i].pfn << PAGE_SHIFT);
725 u32 offset = pb[i].offset;
728 memcpy(dest, (src + offset), len);
734 memset(dest, 0, padding);
741 static inline int netvsc_send_pkt(
742 struct hv_device *device,
743 struct hv_netvsc_packet *packet,
744 struct netvsc_device *net_device,
745 struct hv_page_buffer *pb,
748 struct nvsp_message nvmsg;
749 struct nvsp_1_message_send_rndis_packet * const rpkt =
750 &nvmsg.msg.v1_msg.send_rndis_pkt;
751 struct netvsc_channel * const nvchan =
752 &net_device->chan_table[packet->q_idx];
753 struct vmbus_channel *out_channel = nvchan->channel;
754 struct net_device *ndev = hv_get_drvdata(device);
755 struct netdev_queue *txq = netdev_get_tx_queue(ndev, packet->q_idx);
758 u32 ring_avail = hv_ringbuf_avail_percent(&out_channel->outbound);
760 nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
762 rpkt->channel_type = 0; /* 0 is RMC_DATA */
764 rpkt->channel_type = 1; /* 1 is RMC_CONTROL */
766 rpkt->send_buf_section_index = packet->send_buf_index;
767 if (packet->send_buf_index == NETVSC_INVALID_INDEX)
768 rpkt->send_buf_section_size = 0;
770 rpkt->send_buf_section_size = packet->total_data_buflen;
774 if (out_channel->rescind)
777 if (packet->page_buf_cnt) {
778 if (packet->cp_partial)
779 pb += packet->rmsg_pgcnt;
781 ret = vmbus_sendpacket_pagebuffer(out_channel,
782 pb, packet->page_buf_cnt,
783 &nvmsg, sizeof(nvmsg),
786 ret = vmbus_sendpacket(out_channel,
787 &nvmsg, sizeof(nvmsg),
788 req_id, VM_PKT_DATA_INBAND,
789 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
793 atomic_inc_return(&nvchan->queue_sends);
795 if (ring_avail < RING_AVAIL_PERCENT_LOWATER)
796 netif_tx_stop_queue(txq);
797 } else if (ret == -EAGAIN) {
798 netif_tx_stop_queue(txq);
799 if (atomic_read(&nvchan->queue_sends) < 1) {
800 netif_tx_wake_queue(txq);
805 "Unable to send packet pages %u len %u, ret %d\n",
806 packet->page_buf_cnt, packet->total_data_buflen,
813 /* Move packet out of multi send data (msd), and clear msd */
814 static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
815 struct sk_buff **msd_skb,
816 struct multi_send_data *msdp)
818 *msd_skb = msdp->skb;
819 *msd_send = msdp->pkt;
825 /* RCU already held by caller */
826 int netvsc_send(struct net_device_context *ndev_ctx,
827 struct hv_netvsc_packet *packet,
828 struct rndis_message *rndis_msg,
829 struct hv_page_buffer *pb,
832 struct netvsc_device *net_device
833 = rcu_dereference_bh(ndev_ctx->nvdev);
834 struct hv_device *device = ndev_ctx->device_ctx;
836 struct netvsc_channel *nvchan;
837 u32 pktlen = packet->total_data_buflen, msd_len = 0;
838 unsigned int section_index = NETVSC_INVALID_INDEX;
839 struct multi_send_data *msdp;
840 struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
841 struct sk_buff *msd_skb = NULL;
843 bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
845 /* If device is rescinded, return error and packet will get dropped. */
846 if (unlikely(!net_device || net_device->destroy))
849 /* We may race with netvsc_connect_vsp()/netvsc_init_buf() and get
850 * here before the negotiation with the host is finished and
851 * send_section_map may not be allocated yet.
853 if (unlikely(!net_device->send_section_map))
856 nvchan = &net_device->chan_table[packet->q_idx];
857 packet->send_buf_index = NETVSC_INVALID_INDEX;
858 packet->cp_partial = false;
860 /* Send control message directly without accessing msd (Multi-Send
861 * Data) field which may be changed during data packet processing.
868 /* batch packets in send buffer if possible */
871 msd_len = msdp->pkt->total_data_buflen;
873 try_batch = msd_len > 0 && msdp->count < net_device->max_pkt;
874 if (try_batch && msd_len + pktlen + net_device->pkt_align <
875 net_device->send_section_size) {
876 section_index = msdp->pkt->send_buf_index;
878 } else if (try_batch && msd_len + packet->rmsg_size <
879 net_device->send_section_size) {
880 section_index = msdp->pkt->send_buf_index;
881 packet->cp_partial = true;
883 } else if (pktlen + net_device->pkt_align <
884 net_device->send_section_size) {
885 section_index = netvsc_get_next_send_section(net_device);
886 if (unlikely(section_index == NETVSC_INVALID_INDEX)) {
887 ++ndev_ctx->eth_stats.tx_send_full;
889 move_pkt_msd(&msd_send, &msd_skb, msdp);
894 if (section_index != NETVSC_INVALID_INDEX) {
895 netvsc_copy_to_send_buf(net_device,
896 section_index, msd_len,
897 packet, rndis_msg, pb, skb);
899 packet->send_buf_index = section_index;
901 if (packet->cp_partial) {
902 packet->page_buf_cnt -= packet->rmsg_pgcnt;
903 packet->total_data_buflen = msd_len + packet->rmsg_size;
905 packet->page_buf_cnt = 0;
906 packet->total_data_buflen += msd_len;
910 packet->total_packets += msdp->pkt->total_packets;
911 packet->total_bytes += msdp->pkt->total_bytes;
915 dev_consume_skb_any(msdp->skb);
917 if (xmit_more && !packet->cp_partial) {
928 move_pkt_msd(&msd_send, &msd_skb, msdp);
933 int m_ret = netvsc_send_pkt(device, msd_send, net_device,
937 netvsc_free_send_slot(net_device,
938 msd_send->send_buf_index);
939 dev_kfree_skb_any(msd_skb);
945 ret = netvsc_send_pkt(device, cur_send, net_device, pb, skb);
947 if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
948 netvsc_free_send_slot(net_device, section_index);
953 /* Send pending recv completions */
954 static int send_recv_completions(struct net_device *ndev,
955 struct netvsc_device *nvdev,
956 struct netvsc_channel *nvchan)
958 struct multi_recv_comp *mrc = &nvchan->mrc;
959 struct recv_comp_msg {
960 struct nvsp_message_header hdr;
963 struct recv_comp_msg msg = {
964 .hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE,
968 while (mrc->first != mrc->next) {
969 const struct recv_comp_data *rcd
970 = mrc->slots + mrc->first;
972 msg.status = rcd->status;
973 ret = vmbus_sendpacket(nvchan->channel, &msg, sizeof(msg),
974 rcd->tid, VM_PKT_COMP, 0);
976 struct net_device_context *ndev_ctx = netdev_priv(ndev);
978 ++ndev_ctx->eth_stats.rx_comp_busy;
982 if (++mrc->first == nvdev->recv_completion_cnt)
986 /* receive completion ring has been emptied */
987 if (unlikely(nvdev->destroy))
988 wake_up(&nvdev->wait_drain);
993 /* Count how many receive completions are outstanding */
994 static void recv_comp_slot_avail(const struct netvsc_device *nvdev,
995 const struct multi_recv_comp *mrc,
996 u32 *filled, u32 *avail)
998 u32 count = nvdev->recv_completion_cnt;
1000 if (mrc->next >= mrc->first)
1001 *filled = mrc->next - mrc->first;
1003 *filled = (count - mrc->first) + mrc->next;
1005 *avail = count - *filled - 1;
1008 /* Add receive complete to ring to send to host. */
1009 static void enq_receive_complete(struct net_device *ndev,
1010 struct netvsc_device *nvdev, u16 q_idx,
1011 u64 tid, u32 status)
1013 struct netvsc_channel *nvchan = &nvdev->chan_table[q_idx];
1014 struct multi_recv_comp *mrc = &nvchan->mrc;
1015 struct recv_comp_data *rcd;
1018 recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1020 if (unlikely(filled > NAPI_POLL_WEIGHT)) {
1021 send_recv_completions(ndev, nvdev, nvchan);
1022 recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1025 if (unlikely(!avail)) {
1026 netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
1031 rcd = mrc->slots + mrc->next;
1033 rcd->status = status;
1035 if (++mrc->next == nvdev->recv_completion_cnt)
1039 static int netvsc_receive(struct net_device *ndev,
1040 struct netvsc_device *net_device,
1041 struct net_device_context *net_device_ctx,
1042 struct hv_device *device,
1043 struct vmbus_channel *channel,
1044 const struct vmpacket_descriptor *desc,
1045 struct nvsp_message *nvsp)
1047 const struct vmtransfer_page_packet_header *vmxferpage_packet
1048 = container_of(desc, const struct vmtransfer_page_packet_header, d);
1049 u16 q_idx = channel->offermsg.offer.sub_channel_index;
1050 char *recv_buf = net_device->recv_buf;
1051 u32 status = NVSP_STAT_SUCCESS;
1055 /* Make sure this is a valid nvsp packet */
1056 if (unlikely(nvsp->hdr.msg_type != NVSP_MSG1_TYPE_SEND_RNDIS_PKT)) {
1057 netif_err(net_device_ctx, rx_err, ndev,
1058 "Unknown nvsp packet type received %u\n",
1059 nvsp->hdr.msg_type);
1063 if (unlikely(vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID)) {
1064 netif_err(net_device_ctx, rx_err, ndev,
1065 "Invalid xfer page set id - expecting %x got %x\n",
1066 NETVSC_RECEIVE_BUFFER_ID,
1067 vmxferpage_packet->xfer_pageset_id);
1071 count = vmxferpage_packet->range_cnt;
1073 /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1074 for (i = 0; i < count; i++) {
1075 void *data = recv_buf
1076 + vmxferpage_packet->ranges[i].byte_offset;
1077 u32 buflen = vmxferpage_packet->ranges[i].byte_count;
1079 /* Pass it to the upper layer */
1080 status = rndis_filter_receive(ndev, net_device, device,
1081 channel, data, buflen);
1084 enq_receive_complete(ndev, net_device, q_idx,
1085 vmxferpage_packet->d.trans_id, status);
1090 static void netvsc_send_table(struct hv_device *hdev,
1091 struct nvsp_message *nvmsg)
1093 struct net_device *ndev = hv_get_drvdata(hdev);
1094 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1098 count = nvmsg->msg.v5_msg.send_table.count;
1099 if (count != VRSS_SEND_TAB_SIZE) {
1100 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1104 tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
1105 nvmsg->msg.v5_msg.send_table.offset);
1107 for (i = 0; i < count; i++)
1108 net_device_ctx->tx_send_table[i] = tab[i];
1111 static void netvsc_send_vf(struct net_device_context *net_device_ctx,
1112 struct nvsp_message *nvmsg)
1114 net_device_ctx->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1115 net_device_ctx->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1118 static inline void netvsc_receive_inband(struct hv_device *hdev,
1119 struct net_device_context *net_device_ctx,
1120 struct nvsp_message *nvmsg)
1122 switch (nvmsg->hdr.msg_type) {
1123 case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1124 netvsc_send_table(hdev, nvmsg);
1127 case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1128 netvsc_send_vf(net_device_ctx, nvmsg);
1133 static int netvsc_process_raw_pkt(struct hv_device *device,
1134 struct vmbus_channel *channel,
1135 struct netvsc_device *net_device,
1136 struct net_device *ndev,
1137 const struct vmpacket_descriptor *desc,
1140 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1141 struct nvsp_message *nvmsg = hv_pkt_data(desc);
1143 switch (desc->type) {
1145 netvsc_send_completion(net_device, channel, device,
1149 case VM_PKT_DATA_USING_XFER_PAGES:
1150 return netvsc_receive(ndev, net_device, net_device_ctx,
1151 device, channel, desc, nvmsg);
1154 case VM_PKT_DATA_INBAND:
1155 netvsc_receive_inband(device, net_device_ctx, nvmsg);
1159 netdev_err(ndev, "unhandled packet type %d, tid %llx\n",
1160 desc->type, desc->trans_id);
1167 static struct hv_device *netvsc_channel_to_device(struct vmbus_channel *channel)
1169 struct vmbus_channel *primary = channel->primary_channel;
1171 return primary ? primary->device_obj : channel->device_obj;
1174 /* Network processing softirq
1175 * Process data in incoming ring buffer from host
1176 * Stops when ring is empty or budget is met or exceeded.
1178 int netvsc_poll(struct napi_struct *napi, int budget)
1180 struct netvsc_channel *nvchan
1181 = container_of(napi, struct netvsc_channel, napi);
1182 struct netvsc_device *net_device = nvchan->net_device;
1183 struct vmbus_channel *channel = nvchan->channel;
1184 struct hv_device *device = netvsc_channel_to_device(channel);
1185 struct net_device *ndev = hv_get_drvdata(device);
1188 /* If starting a new interval */
1190 nvchan->desc = hv_pkt_iter_first(channel);
1192 while (nvchan->desc && work_done < budget) {
1193 work_done += netvsc_process_raw_pkt(device, channel, net_device,
1194 ndev, nvchan->desc, budget);
1195 nvchan->desc = hv_pkt_iter_next(channel, nvchan->desc);
1198 /* If send of pending receive completions suceeded
1199 * and did not exhaust NAPI budget this time
1200 * and not doing busy poll
1201 * then re-enable host interrupts
1202 * and reschedule if ring is not empty.
1204 if (send_recv_completions(ndev, net_device, nvchan) == 0 &&
1205 work_done < budget &&
1206 napi_complete_done(napi, work_done) &&
1207 hv_end_read(&channel->inbound)) {
1208 hv_begin_read(&channel->inbound);
1209 napi_reschedule(napi);
1212 /* Driver may overshoot since multiple packets per descriptor */
1213 return min(work_done, budget);
1216 /* Call back when data is available in host ring buffer.
1217 * Processing is deferred until network softirq (NAPI)
1219 void netvsc_channel_cb(void *context)
1221 struct netvsc_channel *nvchan = context;
1222 struct vmbus_channel *channel = nvchan->channel;
1223 struct hv_ring_buffer_info *rbi = &channel->inbound;
1225 /* preload first vmpacket descriptor */
1226 prefetch(hv_get_ring_buffer(rbi) + rbi->priv_read_index);
1228 if (napi_schedule_prep(&nvchan->napi)) {
1229 /* disable interupts from host */
1232 __napi_schedule(&nvchan->napi);
1237 * netvsc_device_add - Callback when the device belonging to this
1240 struct netvsc_device *netvsc_device_add(struct hv_device *device,
1241 const struct netvsc_device_info *device_info)
1244 int ring_size = device_info->ring_size;
1245 struct netvsc_device *net_device;
1246 struct net_device *ndev = hv_get_drvdata(device);
1247 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1249 net_device = alloc_net_device();
1251 return ERR_PTR(-ENOMEM);
1253 net_device->ring_size = ring_size;
1255 /* Because the device uses NAPI, all the interrupt batching and
1256 * control is done via Net softirq, not the channel handling
1258 set_channel_read_mode(device->channel, HV_CALL_ISR);
1260 /* If we're reopening the device we may have multiple queues, fill the
1261 * chn_table with the default channel to use it before subchannels are
1263 * Initialize the channel state before we open;
1264 * we can be interrupted as soon as we open the channel.
1267 for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
1268 struct netvsc_channel *nvchan = &net_device->chan_table[i];
1270 nvchan->channel = device->channel;
1271 nvchan->net_device = net_device;
1272 u64_stats_init(&nvchan->tx_stats.syncp);
1273 u64_stats_init(&nvchan->rx_stats.syncp);
1276 /* Enable NAPI handler before init callbacks */
1277 netif_napi_add(ndev, &net_device->chan_table[0].napi,
1278 netvsc_poll, NAPI_POLL_WEIGHT);
1280 /* Open the channel */
1281 ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
1282 ring_size * PAGE_SIZE, NULL, 0,
1284 net_device->chan_table);
1287 netif_napi_del(&net_device->chan_table[0].napi);
1288 netdev_err(ndev, "unable to open channel: %d\n", ret);
1292 /* Channel is opened */
1293 netdev_dbg(ndev, "hv_netvsc channel opened successfully\n");
1295 napi_enable(&net_device->chan_table[0].napi);
1297 /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1300 rcu_assign_pointer(net_device_ctx->nvdev, net_device);
1302 /* Connect with the NetVsp */
1303 ret = netvsc_connect_vsp(device, net_device, device_info);
1306 "unable to connect to NetVSP - %d\n", ret);
1313 RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
1314 napi_disable(&net_device->chan_table[0].napi);
1316 /* Now, we can close the channel safely */
1317 vmbus_close(device->channel);
1320 free_netvsc_device(&net_device->rcu);
1322 return ERR_PTR(ret);