2 * Copyright (c) 2009, Microsoft Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, see <http://www.gnu.org/licenses/>.
17 * Haiyang Zhang <haiyangz@microsoft.com>
18 * Hank Janssen <hjanssen@microsoft.com>
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
22 #include <linux/kernel.h>
23 #include <linux/sched.h>
24 #include <linux/wait.h>
26 #include <linux/delay.h>
28 #include <linux/slab.h>
29 #include <linux/netdevice.h>
30 #include <linux/if_ether.h>
31 #include <linux/vmalloc.h>
32 #include <asm/sync_bitops.h>
34 #include "hyperv_net.h"
37 * Switch the data path from the synthetic interface to the VF
40 void netvsc_switch_datapath(struct net_device *ndev, bool vf)
42 struct net_device_context *net_device_ctx = netdev_priv(ndev);
43 struct hv_device *dev = net_device_ctx->device_ctx;
44 struct netvsc_device *nv_dev = net_device_ctx->nvdev;
45 struct nvsp_message *init_pkt = &nv_dev->channel_init_pkt;
47 memset(init_pkt, 0, sizeof(struct nvsp_message));
48 init_pkt->hdr.msg_type = NVSP_MSG4_TYPE_SWITCH_DATA_PATH;
50 init_pkt->msg.v4_msg.active_dp.active_datapath =
53 init_pkt->msg.v4_msg.active_dp.active_datapath =
54 NVSP_DATAPATH_SYNTHETIC;
56 vmbus_sendpacket(dev->channel, init_pkt,
57 sizeof(struct nvsp_message),
58 (unsigned long)init_pkt,
59 VM_PKT_DATA_INBAND, 0);
61 net_device_ctx->datapath = vf;
64 static struct netvsc_device *alloc_net_device(void)
66 struct netvsc_device *net_device;
68 net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
72 net_device->chan_table[0].mrc.buf
73 = vzalloc(NETVSC_RECVSLOT_MAX * sizeof(struct recv_comp_data));
75 init_waitqueue_head(&net_device->wait_drain);
76 net_device->destroy = false;
77 atomic_set(&net_device->open_cnt, 0);
78 net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
79 net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
80 init_completion(&net_device->channel_init_wait);
81 init_waitqueue_head(&net_device->subchan_open);
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.buf);
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 netvsc_device *net_device = net_device_to_netvsc_device(ndev);
111 * If we got a section count, it means we received a
112 * SendReceiveBufferComplete msg (ie sent
113 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
114 * to send a revoke msg here
116 if (net_device->recv_section_cnt) {
117 /* Send the revoke receive buffer */
118 revoke_packet = &net_device->revoke_packet;
119 memset(revoke_packet, 0, sizeof(struct nvsp_message));
121 revoke_packet->hdr.msg_type =
122 NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
123 revoke_packet->msg.v1_msg.
124 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
126 ret = vmbus_sendpacket(device->channel,
128 sizeof(struct nvsp_message),
129 (unsigned long)revoke_packet,
130 VM_PKT_DATA_INBAND, 0);
131 /* If the failure is because the channel is rescinded;
132 * ignore the failure since we cannot send on a rescinded
133 * channel. This would allow us to properly cleanup
134 * even when the channel is rescinded.
136 if (device->channel->rescind)
139 * If we failed here, we might as well return and
140 * have a leak rather than continue and a bugchk
143 netdev_err(ndev, "unable to send "
144 "revoke receive buffer to netvsp\n");
149 /* Teardown the gpadl on the vsp end */
150 if (net_device->recv_buf_gpadl_handle) {
151 ret = vmbus_teardown_gpadl(device->channel,
152 net_device->recv_buf_gpadl_handle);
154 /* If we failed here, we might as well return and have a leak
155 * rather than continue and a bugchk
159 "unable to teardown receive buffer's gpadl\n");
162 net_device->recv_buf_gpadl_handle = 0;
165 if (net_device->recv_buf) {
166 /* Free up the receive buffer */
167 vfree(net_device->recv_buf);
168 net_device->recv_buf = NULL;
171 if (net_device->recv_section) {
172 net_device->recv_section_cnt = 0;
173 kfree(net_device->recv_section);
174 net_device->recv_section = NULL;
177 /* Deal with the send buffer we may have setup.
178 * If we got a send section size, it means we received a
179 * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
180 * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
181 * to send a revoke msg here
183 if (net_device->send_section_size) {
184 /* Send the revoke receive buffer */
185 revoke_packet = &net_device->revoke_packet;
186 memset(revoke_packet, 0, sizeof(struct nvsp_message));
188 revoke_packet->hdr.msg_type =
189 NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
190 revoke_packet->msg.v1_msg.revoke_send_buf.id =
191 NETVSC_SEND_BUFFER_ID;
193 ret = vmbus_sendpacket(device->channel,
195 sizeof(struct nvsp_message),
196 (unsigned long)revoke_packet,
197 VM_PKT_DATA_INBAND, 0);
199 /* If the failure is because the channel is rescinded;
200 * ignore the failure since we cannot send on a rescinded
201 * channel. This would allow us to properly cleanup
202 * even when the channel is rescinded.
204 if (device->channel->rescind)
207 /* If we failed here, we might as well return and
208 * have a leak rather than continue and a bugchk
211 netdev_err(ndev, "unable to send "
212 "revoke send buffer to netvsp\n");
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 static int netvsc_init_buf(struct hv_device *device,
240 struct netvsc_device *net_device)
243 struct nvsp_message *init_packet;
244 struct net_device *ndev;
248 ndev = hv_get_drvdata(device);
250 node = cpu_to_node(device->channel->target_cpu);
251 net_device->recv_buf = vzalloc_node(net_device->recv_buf_size, node);
252 if (!net_device->recv_buf)
253 net_device->recv_buf = vzalloc(net_device->recv_buf_size);
255 if (!net_device->recv_buf) {
256 netdev_err(ndev, "unable to allocate receive "
257 "buffer of size %d\n", net_device->recv_buf_size);
263 * Establish the gpadl handle for this buffer on this
264 * channel. Note: This call uses the vmbus connection rather
265 * than the channel to establish the gpadl handle.
267 ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
268 net_device->recv_buf_size,
269 &net_device->recv_buf_gpadl_handle);
272 "unable to establish receive buffer's gpadl\n");
276 /* Notify the NetVsp of the gpadl handle */
277 init_packet = &net_device->channel_init_pkt;
278 memset(init_packet, 0, sizeof(struct nvsp_message));
279 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
280 init_packet->msg.v1_msg.send_recv_buf.
281 gpadl_handle = net_device->recv_buf_gpadl_handle;
282 init_packet->msg.v1_msg.
283 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
285 /* Send the gpadl notification request */
286 ret = vmbus_sendpacket(device->channel, init_packet,
287 sizeof(struct nvsp_message),
288 (unsigned long)init_packet,
290 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
293 "unable to send receive buffer's gpadl to netvsp\n");
297 wait_for_completion(&net_device->channel_init_wait);
299 /* Check the response */
300 if (init_packet->msg.v1_msg.
301 send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
302 netdev_err(ndev, "Unable to complete receive buffer "
303 "initialization with NetVsp - status %d\n",
304 init_packet->msg.v1_msg.
305 send_recv_buf_complete.status);
310 /* Parse the response */
312 net_device->recv_section_cnt = init_packet->msg.
313 v1_msg.send_recv_buf_complete.num_sections;
315 net_device->recv_section = kmemdup(
316 init_packet->msg.v1_msg.send_recv_buf_complete.sections,
317 net_device->recv_section_cnt *
318 sizeof(struct nvsp_1_receive_buffer_section),
320 if (net_device->recv_section == NULL) {
326 * For 1st release, there should only be 1 section that represents the
327 * entire receive buffer
329 if (net_device->recv_section_cnt != 1 ||
330 net_device->recv_section->offset != 0) {
335 /* Now setup the send buffer.
337 net_device->send_buf = vzalloc_node(net_device->send_buf_size, node);
338 if (!net_device->send_buf)
339 net_device->send_buf = vzalloc(net_device->send_buf_size);
340 if (!net_device->send_buf) {
341 netdev_err(ndev, "unable to allocate send "
342 "buffer of size %d\n", net_device->send_buf_size);
347 /* Establish the gpadl handle for this buffer on this
348 * channel. Note: This call uses the vmbus connection rather
349 * than the channel to establish the gpadl handle.
351 ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
352 net_device->send_buf_size,
353 &net_device->send_buf_gpadl_handle);
356 "unable to establish send buffer's gpadl\n");
360 /* Notify the NetVsp of the gpadl handle */
361 init_packet = &net_device->channel_init_pkt;
362 memset(init_packet, 0, sizeof(struct nvsp_message));
363 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
364 init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
365 net_device->send_buf_gpadl_handle;
366 init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
368 /* Send the gpadl notification request */
369 ret = vmbus_sendpacket(device->channel, init_packet,
370 sizeof(struct nvsp_message),
371 (unsigned long)init_packet,
373 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
376 "unable to send send buffer's gpadl to netvsp\n");
380 wait_for_completion(&net_device->channel_init_wait);
382 /* Check the response */
383 if (init_packet->msg.v1_msg.
384 send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
385 netdev_err(ndev, "Unable to complete send buffer "
386 "initialization with NetVsp - status %d\n",
387 init_packet->msg.v1_msg.
388 send_send_buf_complete.status);
393 /* Parse the response */
394 net_device->send_section_size = init_packet->msg.
395 v1_msg.send_send_buf_complete.section_size;
397 /* Section count is simply the size divided by the section size.
399 net_device->send_section_cnt =
400 net_device->send_buf_size / net_device->send_section_size;
402 netdev_dbg(ndev, "Send section size: %d, Section count:%d\n",
403 net_device->send_section_size, net_device->send_section_cnt);
405 /* Setup state for managing the send buffer. */
406 map_words = DIV_ROUND_UP(net_device->send_section_cnt, BITS_PER_LONG);
408 net_device->send_section_map = kcalloc(map_words, sizeof(ulong), GFP_KERNEL);
409 if (net_device->send_section_map == NULL) {
417 netvsc_destroy_buf(device);
423 /* Negotiate NVSP protocol version */
424 static int negotiate_nvsp_ver(struct hv_device *device,
425 struct netvsc_device *net_device,
426 struct nvsp_message *init_packet,
429 struct net_device *ndev = hv_get_drvdata(device);
432 memset(init_packet, 0, sizeof(struct nvsp_message));
433 init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
434 init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
435 init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
437 /* Send the init request */
438 ret = vmbus_sendpacket(device->channel, init_packet,
439 sizeof(struct nvsp_message),
440 (unsigned long)init_packet,
442 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
447 wait_for_completion(&net_device->channel_init_wait);
449 if (init_packet->msg.init_msg.init_complete.status !=
453 if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
456 /* NVSPv2 or later: Send NDIS config */
457 memset(init_packet, 0, sizeof(struct nvsp_message));
458 init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
459 init_packet->msg.v2_msg.send_ndis_config.mtu = ndev->mtu + ETH_HLEN;
460 init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
462 if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5) {
463 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
465 /* Teaming bit is needed to receive link speed updates */
466 init_packet->msg.v2_msg.send_ndis_config.capability.teaming = 1;
469 ret = vmbus_sendpacket(device->channel, init_packet,
470 sizeof(struct nvsp_message),
471 (unsigned long)init_packet,
472 VM_PKT_DATA_INBAND, 0);
477 static int netvsc_connect_vsp(struct hv_device *device,
478 struct netvsc_device *net_device)
480 const u32 ver_list[] = {
481 NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
482 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5
484 struct nvsp_message *init_packet;
485 int ndis_version, i, ret;
487 init_packet = &net_device->channel_init_pkt;
489 /* Negotiate the latest NVSP protocol supported */
490 for (i = ARRAY_SIZE(ver_list) - 1; i >= 0; i--)
491 if (negotiate_nvsp_ver(device, net_device, init_packet,
493 net_device->nvsp_version = ver_list[i];
502 pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
504 /* Send the ndis version */
505 memset(init_packet, 0, sizeof(struct nvsp_message));
507 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
508 ndis_version = 0x00060001;
510 ndis_version = 0x0006001e;
512 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
513 init_packet->msg.v1_msg.
514 send_ndis_ver.ndis_major_ver =
515 (ndis_version & 0xFFFF0000) >> 16;
516 init_packet->msg.v1_msg.
517 send_ndis_ver.ndis_minor_ver =
518 ndis_version & 0xFFFF;
520 /* Send the init request */
521 ret = vmbus_sendpacket(device->channel, init_packet,
522 sizeof(struct nvsp_message),
523 (unsigned long)init_packet,
524 VM_PKT_DATA_INBAND, 0);
528 /* Post the big receive buffer to NetVSP */
529 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
530 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
532 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
533 net_device->send_buf_size = NETVSC_SEND_BUFFER_SIZE;
535 ret = netvsc_init_buf(device, net_device);
541 static void netvsc_disconnect_vsp(struct hv_device *device)
543 netvsc_destroy_buf(device);
547 * netvsc_device_remove - Callback when the root bus device is removed
549 void netvsc_device_remove(struct hv_device *device)
551 struct net_device *ndev = hv_get_drvdata(device);
552 struct net_device_context *net_device_ctx = netdev_priv(ndev);
553 struct netvsc_device *net_device = net_device_ctx->nvdev;
556 netvsc_disconnect_vsp(device);
558 RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
561 * At this point, no one should be accessing net_device
564 netdev_dbg(ndev, "net device safe to remove\n");
566 /* Now, we can close the channel safely */
567 vmbus_close(device->channel);
569 /* And dissassociate NAPI context from device */
570 for (i = 0; i < net_device->num_chn; i++)
571 netif_napi_del(&net_device->chan_table[i].napi);
573 /* Release all resources */
574 free_netvsc_device_rcu(net_device);
577 #define RING_AVAIL_PERCENT_HIWATER 20
578 #define RING_AVAIL_PERCENT_LOWATER 10
581 * Get the percentage of available bytes to write in the ring.
582 * The return value is in range from 0 to 100.
584 static inline u32 hv_ringbuf_avail_percent(
585 struct hv_ring_buffer_info *ring_info)
587 u32 avail_read, avail_write;
589 hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
591 return avail_write * 100 / ring_info->ring_datasize;
594 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
597 sync_change_bit(index, net_device->send_section_map);
600 static void netvsc_send_tx_complete(struct netvsc_device *net_device,
601 struct vmbus_channel *incoming_channel,
602 struct hv_device *device,
603 const struct vmpacket_descriptor *desc,
606 struct sk_buff *skb = (struct sk_buff *)(unsigned long)desc->trans_id;
607 struct net_device *ndev = hv_get_drvdata(device);
608 struct vmbus_channel *channel = device->channel;
612 /* Notify the layer above us */
614 const struct hv_netvsc_packet *packet
615 = (struct hv_netvsc_packet *)skb->cb;
616 u32 send_index = packet->send_buf_index;
617 struct netvsc_stats *tx_stats;
619 if (send_index != NETVSC_INVALID_INDEX)
620 netvsc_free_send_slot(net_device, send_index);
621 q_idx = packet->q_idx;
622 channel = incoming_channel;
624 tx_stats = &net_device->chan_table[q_idx].tx_stats;
626 u64_stats_update_begin(&tx_stats->syncp);
627 tx_stats->packets += packet->total_packets;
628 tx_stats->bytes += packet->total_bytes;
629 u64_stats_update_end(&tx_stats->syncp);
631 napi_consume_skb(skb, budget);
635 atomic_dec_return(&net_device->chan_table[q_idx].queue_sends);
637 if (net_device->destroy && queue_sends == 0)
638 wake_up(&net_device->wait_drain);
640 if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev, q_idx)) &&
641 (hv_ringbuf_avail_percent(&channel->outbound) > RING_AVAIL_PERCENT_HIWATER ||
643 netif_tx_wake_queue(netdev_get_tx_queue(ndev, q_idx));
646 static void netvsc_send_completion(struct netvsc_device *net_device,
647 struct vmbus_channel *incoming_channel,
648 struct hv_device *device,
649 const struct vmpacket_descriptor *desc,
652 struct nvsp_message *nvsp_packet = hv_pkt_data(desc);
653 struct net_device *ndev = hv_get_drvdata(device);
655 switch (nvsp_packet->hdr.msg_type) {
656 case NVSP_MSG_TYPE_INIT_COMPLETE:
657 case NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE:
658 case NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE:
659 case NVSP_MSG5_TYPE_SUBCHANNEL:
660 /* Copy the response back */
661 memcpy(&net_device->channel_init_pkt, nvsp_packet,
662 sizeof(struct nvsp_message));
663 complete(&net_device->channel_init_wait);
666 case NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE:
667 netvsc_send_tx_complete(net_device, incoming_channel,
668 device, desc, budget);
673 "Unknown send completion type %d received!!\n",
674 nvsp_packet->hdr.msg_type);
678 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
680 unsigned long *map_addr = net_device->send_section_map;
683 for_each_clear_bit(i, map_addr, net_device->send_section_cnt) {
684 if (sync_test_and_set_bit(i, map_addr) == 0)
688 return NETVSC_INVALID_INDEX;
691 static u32 netvsc_copy_to_send_buf(struct netvsc_device *net_device,
692 unsigned int section_index,
694 struct hv_netvsc_packet *packet,
695 struct rndis_message *rndis_msg,
696 struct hv_page_buffer **pb,
699 char *start = net_device->send_buf;
700 char *dest = start + (section_index * net_device->send_section_size)
705 u32 remain = packet->total_data_buflen % net_device->pkt_align;
706 u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
707 packet->page_buf_cnt;
710 if (skb->xmit_more && remain && !packet->cp_partial) {
711 padding = net_device->pkt_align - remain;
712 rndis_msg->msg_len += padding;
713 packet->total_data_buflen += padding;
716 for (i = 0; i < page_count; i++) {
717 char *src = phys_to_virt((*pb)[i].pfn << PAGE_SHIFT);
718 u32 offset = (*pb)[i].offset;
719 u32 len = (*pb)[i].len;
721 memcpy(dest, (src + offset), len);
727 memset(dest, 0, padding);
734 static inline int netvsc_send_pkt(
735 struct hv_device *device,
736 struct hv_netvsc_packet *packet,
737 struct netvsc_device *net_device,
738 struct hv_page_buffer **pb,
741 struct nvsp_message nvmsg;
742 struct netvsc_channel *nvchan
743 = &net_device->chan_table[packet->q_idx];
744 struct vmbus_channel *out_channel = nvchan->channel;
745 struct net_device *ndev = hv_get_drvdata(device);
746 struct netdev_queue *txq = netdev_get_tx_queue(ndev, packet->q_idx);
749 struct hv_page_buffer *pgbuf;
750 u32 ring_avail = hv_ringbuf_avail_percent(&out_channel->outbound);
752 nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
755 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 0;
757 /* 1 is RMC_CONTROL; */
758 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 1;
761 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
762 packet->send_buf_index;
763 if (packet->send_buf_index == NETVSC_INVALID_INDEX)
764 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
766 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size =
767 packet->total_data_buflen;
771 if (out_channel->rescind)
774 if (packet->page_buf_cnt) {
775 pgbuf = packet->cp_partial ? (*pb) +
776 packet->rmsg_pgcnt : (*pb);
777 ret = vmbus_sendpacket_pagebuffer_ctl(out_channel,
779 packet->page_buf_cnt,
781 sizeof(struct nvsp_message),
783 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
785 ret = vmbus_sendpacket_ctl(out_channel, &nvmsg,
786 sizeof(struct nvsp_message),
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);
804 netdev_err(ndev, "Unable to send packet %p ret %d\n",
811 /* Move packet out of multi send data (msd), and clear msd */
812 static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
813 struct sk_buff **msd_skb,
814 struct multi_send_data *msdp)
816 *msd_skb = msdp->skb;
817 *msd_send = msdp->pkt;
823 int netvsc_send(struct hv_device *device,
824 struct hv_netvsc_packet *packet,
825 struct rndis_message *rndis_msg,
826 struct hv_page_buffer **pb,
829 struct netvsc_device *net_device = hv_device_to_netvsc_device(device);
831 struct netvsc_channel *nvchan;
832 u32 pktlen = packet->total_data_buflen, msd_len = 0;
833 unsigned int section_index = NETVSC_INVALID_INDEX;
834 struct multi_send_data *msdp;
835 struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
836 struct sk_buff *msd_skb = NULL;
838 bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
840 /* If device is rescinded, return error and packet will get dropped. */
841 if (unlikely(net_device->destroy))
844 /* We may race with netvsc_connect_vsp()/netvsc_init_buf() and get
845 * here before the negotiation with the host is finished and
846 * send_section_map may not be allocated yet.
848 if (unlikely(!net_device->send_section_map))
851 nvchan = &net_device->chan_table[packet->q_idx];
852 packet->send_buf_index = NETVSC_INVALID_INDEX;
853 packet->cp_partial = false;
855 /* Send control message directly without accessing msd (Multi-Send
856 * Data) field which may be changed during data packet processing.
863 /* batch packets in send buffer if possible */
866 msd_len = msdp->pkt->total_data_buflen;
868 try_batch = msd_len > 0 && msdp->count < net_device->max_pkt;
869 if (try_batch && msd_len + pktlen + net_device->pkt_align <
870 net_device->send_section_size) {
871 section_index = msdp->pkt->send_buf_index;
873 } else if (try_batch && msd_len + packet->rmsg_size <
874 net_device->send_section_size) {
875 section_index = msdp->pkt->send_buf_index;
876 packet->cp_partial = true;
878 } else if (pktlen + net_device->pkt_align <
879 net_device->send_section_size) {
880 section_index = netvsc_get_next_send_section(net_device);
881 if (section_index != NETVSC_INVALID_INDEX) {
882 move_pkt_msd(&msd_send, &msd_skb, msdp);
887 if (section_index != NETVSC_INVALID_INDEX) {
888 netvsc_copy_to_send_buf(net_device,
889 section_index, msd_len,
890 packet, rndis_msg, pb, skb);
892 packet->send_buf_index = section_index;
894 if (packet->cp_partial) {
895 packet->page_buf_cnt -= packet->rmsg_pgcnt;
896 packet->total_data_buflen = msd_len + packet->rmsg_size;
898 packet->page_buf_cnt = 0;
899 packet->total_data_buflen += msd_len;
903 packet->total_packets += msdp->pkt->total_packets;
904 packet->total_bytes += msdp->pkt->total_bytes;
908 dev_consume_skb_any(msdp->skb);
910 if (xmit_more && !packet->cp_partial) {
921 move_pkt_msd(&msd_send, &msd_skb, msdp);
926 int m_ret = netvsc_send_pkt(device, msd_send, net_device,
930 netvsc_free_send_slot(net_device,
931 msd_send->send_buf_index);
932 dev_kfree_skb_any(msd_skb);
938 ret = netvsc_send_pkt(device, cur_send, net_device, pb, skb);
940 if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
941 netvsc_free_send_slot(net_device, section_index);
946 static int netvsc_send_recv_completion(struct vmbus_channel *channel,
947 u64 transaction_id, u32 status)
949 struct nvsp_message recvcompMessage;
952 recvcompMessage.hdr.msg_type =
953 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
955 recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status = status;
957 /* Send the completion */
958 ret = vmbus_sendpacket(channel, &recvcompMessage,
959 sizeof(struct nvsp_message_header) + sizeof(u32),
960 transaction_id, VM_PKT_COMP, 0);
965 static inline void count_recv_comp_slot(struct netvsc_device *nvdev, u16 q_idx,
966 u32 *filled, u32 *avail)
968 struct multi_recv_comp *mrc = &nvdev->chan_table[q_idx].mrc;
969 u32 first = mrc->first;
970 u32 next = mrc->next;
972 *filled = (first > next) ? NETVSC_RECVSLOT_MAX - first + next :
975 *avail = NETVSC_RECVSLOT_MAX - *filled - 1;
978 /* Read the first filled slot, no change to index */
979 static inline struct recv_comp_data *read_recv_comp_slot(struct netvsc_device
982 struct multi_recv_comp *mrc = &nvdev->chan_table[q_idx].mrc;
985 if (unlikely(!mrc->buf))
988 count_recv_comp_slot(nvdev, q_idx, &filled, &avail);
992 return mrc->buf + mrc->first * sizeof(struct recv_comp_data);
995 /* Put the first filled slot back to available pool */
996 static inline void put_recv_comp_slot(struct netvsc_device *nvdev, u16 q_idx)
998 struct multi_recv_comp *mrc = &nvdev->chan_table[q_idx].mrc;
1001 mrc->first = (mrc->first + 1) % NETVSC_RECVSLOT_MAX;
1003 num_recv = atomic_dec_return(&nvdev->num_outstanding_recvs);
1005 if (nvdev->destroy && num_recv == 0)
1006 wake_up(&nvdev->wait_drain);
1009 /* Check and send pending recv completions */
1010 static void netvsc_chk_recv_comp(struct netvsc_device *nvdev,
1011 struct vmbus_channel *channel, u16 q_idx)
1013 struct recv_comp_data *rcd;
1017 rcd = read_recv_comp_slot(nvdev, q_idx);
1021 ret = netvsc_send_recv_completion(channel, rcd->tid,
1026 put_recv_comp_slot(nvdev, q_idx);
1030 #define NETVSC_RCD_WATERMARK 80
1032 /* Get next available slot */
1033 static inline struct recv_comp_data *get_recv_comp_slot(
1034 struct netvsc_device *nvdev, struct vmbus_channel *channel, u16 q_idx)
1036 struct multi_recv_comp *mrc = &nvdev->chan_table[q_idx].mrc;
1037 u32 filled, avail, next;
1038 struct recv_comp_data *rcd;
1040 if (unlikely(!nvdev->recv_section))
1043 if (unlikely(!mrc->buf))
1046 if (atomic_read(&nvdev->num_outstanding_recvs) >
1047 nvdev->recv_section->num_sub_allocs * NETVSC_RCD_WATERMARK / 100)
1048 netvsc_chk_recv_comp(nvdev, channel, q_idx);
1050 count_recv_comp_slot(nvdev, q_idx, &filled, &avail);
1055 rcd = mrc->buf + next * sizeof(struct recv_comp_data);
1056 mrc->next = (next + 1) % NETVSC_RECVSLOT_MAX;
1058 atomic_inc(&nvdev->num_outstanding_recvs);
1063 static int netvsc_receive(struct net_device *ndev,
1064 struct netvsc_device *net_device,
1065 struct net_device_context *net_device_ctx,
1066 struct hv_device *device,
1067 struct vmbus_channel *channel,
1068 const struct vmpacket_descriptor *desc,
1069 struct nvsp_message *nvsp)
1071 const struct vmtransfer_page_packet_header *vmxferpage_packet
1072 = container_of(desc, const struct vmtransfer_page_packet_header, d);
1073 u16 q_idx = channel->offermsg.offer.sub_channel_index;
1074 char *recv_buf = net_device->recv_buf;
1075 u32 status = NVSP_STAT_SUCCESS;
1080 /* Make sure this is a valid nvsp packet */
1081 if (unlikely(nvsp->hdr.msg_type != NVSP_MSG1_TYPE_SEND_RNDIS_PKT)) {
1082 netif_err(net_device_ctx, rx_err, ndev,
1083 "Unknown nvsp packet type received %u\n",
1084 nvsp->hdr.msg_type);
1088 if (unlikely(vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID)) {
1089 netif_err(net_device_ctx, rx_err, ndev,
1090 "Invalid xfer page set id - expecting %x got %x\n",
1091 NETVSC_RECEIVE_BUFFER_ID,
1092 vmxferpage_packet->xfer_pageset_id);
1096 count = vmxferpage_packet->range_cnt;
1098 /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1099 for (i = 0; i < count; i++) {
1100 void *data = recv_buf
1101 + vmxferpage_packet->ranges[i].byte_offset;
1102 u32 buflen = vmxferpage_packet->ranges[i].byte_count;
1104 /* Pass it to the upper layer */
1105 status = rndis_filter_receive(ndev, net_device, device,
1106 channel, data, buflen);
1109 if (net_device->chan_table[q_idx].mrc.buf) {
1110 struct recv_comp_data *rcd;
1112 rcd = get_recv_comp_slot(net_device, channel, q_idx);
1114 rcd->tid = vmxferpage_packet->d.trans_id;
1115 rcd->status = status;
1117 netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
1118 q_idx, vmxferpage_packet->d.trans_id);
1121 ret = netvsc_send_recv_completion(channel,
1122 vmxferpage_packet->d.trans_id,
1125 netdev_err(ndev, "Recv_comp q:%hd, tid:%llx, err:%d\n",
1126 q_idx, vmxferpage_packet->d.trans_id, ret);
1131 static void netvsc_send_table(struct hv_device *hdev,
1132 struct nvsp_message *nvmsg)
1134 struct net_device *ndev = hv_get_drvdata(hdev);
1135 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1139 count = nvmsg->msg.v5_msg.send_table.count;
1140 if (count != VRSS_SEND_TAB_SIZE) {
1141 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1145 tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
1146 nvmsg->msg.v5_msg.send_table.offset);
1148 for (i = 0; i < count; i++)
1149 net_device_ctx->tx_send_table[i] = tab[i];
1152 static void netvsc_send_vf(struct net_device_context *net_device_ctx,
1153 struct nvsp_message *nvmsg)
1155 net_device_ctx->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1156 net_device_ctx->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1159 static inline void netvsc_receive_inband(struct hv_device *hdev,
1160 struct net_device_context *net_device_ctx,
1161 struct nvsp_message *nvmsg)
1163 switch (nvmsg->hdr.msg_type) {
1164 case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1165 netvsc_send_table(hdev, nvmsg);
1168 case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1169 netvsc_send_vf(net_device_ctx, nvmsg);
1174 static int netvsc_process_raw_pkt(struct hv_device *device,
1175 struct vmbus_channel *channel,
1176 struct netvsc_device *net_device,
1177 struct net_device *ndev,
1178 const struct vmpacket_descriptor *desc,
1181 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1182 struct nvsp_message *nvmsg = hv_pkt_data(desc);
1184 switch (desc->type) {
1186 netvsc_send_completion(net_device, channel, device,
1190 case VM_PKT_DATA_USING_XFER_PAGES:
1191 return netvsc_receive(ndev, net_device, net_device_ctx,
1192 device, channel, desc, nvmsg);
1195 case VM_PKT_DATA_INBAND:
1196 netvsc_receive_inband(device, net_device_ctx, nvmsg);
1200 netdev_err(ndev, "unhandled packet type %d, tid %llx\n",
1201 desc->type, desc->trans_id);
1208 static struct hv_device *netvsc_channel_to_device(struct vmbus_channel *channel)
1210 struct vmbus_channel *primary = channel->primary_channel;
1212 return primary ? primary->device_obj : channel->device_obj;
1215 /* Network processing softirq
1216 * Process data in incoming ring buffer from host
1217 * Stops when ring is empty or budget is met or exceeded.
1219 int netvsc_poll(struct napi_struct *napi, int budget)
1221 struct netvsc_channel *nvchan
1222 = container_of(napi, struct netvsc_channel, napi);
1223 struct vmbus_channel *channel = nvchan->channel;
1224 struct hv_device *device = netvsc_channel_to_device(channel);
1225 u16 q_idx = channel->offermsg.offer.sub_channel_index;
1226 struct net_device *ndev = hv_get_drvdata(device);
1227 struct netvsc_device *net_device = net_device_to_netvsc_device(ndev);
1230 /* If starting a new interval */
1232 nvchan->desc = hv_pkt_iter_first(channel);
1234 while (nvchan->desc && work_done < budget) {
1235 work_done += netvsc_process_raw_pkt(device, channel, net_device,
1236 ndev, nvchan->desc, budget);
1237 nvchan->desc = hv_pkt_iter_next(channel, nvchan->desc);
1240 /* If receive ring was exhausted
1241 * and not doing busy poll
1242 * then re-enable host interrupts
1243 * and reschedule if ring is not empty.
1245 if (work_done < budget &&
1246 napi_complete_done(napi, work_done) &&
1247 hv_end_read(&channel->inbound) != 0)
1248 napi_reschedule(napi);
1250 netvsc_chk_recv_comp(net_device, channel, q_idx);
1252 /* Driver may overshoot since multiple packets per descriptor */
1253 return min(work_done, budget);
1256 /* Call back when data is available in host ring buffer.
1257 * Processing is deferred until network softirq (NAPI)
1259 void netvsc_channel_cb(void *context)
1261 struct netvsc_channel *nvchan = context;
1263 if (napi_schedule_prep(&nvchan->napi)) {
1264 /* disable interupts from host */
1265 hv_begin_read(&nvchan->channel->inbound);
1267 __napi_schedule(&nvchan->napi);
1272 * netvsc_device_add - Callback when the device belonging to this
1275 int netvsc_device_add(struct hv_device *device,
1276 const struct netvsc_device_info *device_info)
1279 int ring_size = device_info->ring_size;
1280 struct netvsc_device *net_device;
1281 struct net_device *ndev = hv_get_drvdata(device);
1282 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1284 net_device = alloc_net_device();
1288 net_device->ring_size = ring_size;
1290 /* Because the device uses NAPI, all the interrupt batching and
1291 * control is done via Net softirq, not the channel handling
1293 set_channel_read_mode(device->channel, HV_CALL_ISR);
1295 /* If we're reopening the device we may have multiple queues, fill the
1296 * chn_table with the default channel to use it before subchannels are
1298 * Initialize the channel state before we open;
1299 * we can be interrupted as soon as we open the channel.
1302 for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
1303 struct netvsc_channel *nvchan = &net_device->chan_table[i];
1305 nvchan->channel = device->channel;
1306 u64_stats_init(&nvchan->tx_stats.syncp);
1307 u64_stats_init(&nvchan->rx_stats.syncp);
1310 /* Enable NAPI handler before init callbacks */
1311 netif_napi_add(ndev, &net_device->chan_table[0].napi,
1312 netvsc_poll, NAPI_POLL_WEIGHT);
1314 /* Open the channel */
1315 ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
1316 ring_size * PAGE_SIZE, NULL, 0,
1318 net_device->chan_table);
1321 netif_napi_del(&net_device->chan_table[0].napi);
1322 netdev_err(ndev, "unable to open channel: %d\n", ret);
1326 /* Channel is opened */
1327 netdev_dbg(ndev, "hv_netvsc channel opened successfully\n");
1329 napi_enable(&net_device->chan_table[0].napi);
1331 /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1334 rcu_assign_pointer(net_device_ctx->nvdev, net_device);
1336 /* Connect with the NetVsp */
1337 ret = netvsc_connect_vsp(device, net_device);
1340 "unable to connect to NetVSP - %d\n", ret);
1347 netif_napi_del(&net_device->chan_table[0].napi);
1349 /* Now, we can close the channel safely */
1350 vmbus_close(device->channel);
1353 free_netvsc_device(&net_device->rcu);