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);
63 static struct netvsc_device *alloc_net_device(void)
65 struct netvsc_device *net_device;
67 net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
71 net_device->cb_buffer = kzalloc(NETVSC_PACKET_SIZE, GFP_KERNEL);
72 if (!net_device->cb_buffer) {
77 init_waitqueue_head(&net_device->wait_drain);
78 net_device->destroy = false;
79 atomic_set(&net_device->open_cnt, 0);
80 atomic_set(&net_device->vf_use_cnt, 0);
81 net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
82 net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
84 net_device->vf_netdev = NULL;
85 net_device->vf_inject = false;
90 static void free_netvsc_device(struct netvsc_device *nvdev)
92 kfree(nvdev->cb_buffer);
96 static struct netvsc_device *get_outbound_net_device(struct hv_device *device)
98 struct netvsc_device *net_device = hv_device_to_netvsc_device(device);
100 if (net_device && net_device->destroy)
106 static struct netvsc_device *get_inbound_net_device(struct hv_device *device)
108 struct netvsc_device *net_device = hv_device_to_netvsc_device(device);
113 if (net_device->destroy &&
114 atomic_read(&net_device->num_outstanding_sends) == 0)
122 static int netvsc_destroy_buf(struct hv_device *device)
124 struct nvsp_message *revoke_packet;
126 struct net_device *ndev = hv_get_drvdata(device);
127 struct netvsc_device *net_device = net_device_to_netvsc_device(ndev);
130 * If we got a section count, it means we received a
131 * SendReceiveBufferComplete msg (ie sent
132 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
133 * to send a revoke msg here
135 if (net_device->recv_section_cnt) {
136 /* Send the revoke receive buffer */
137 revoke_packet = &net_device->revoke_packet;
138 memset(revoke_packet, 0, sizeof(struct nvsp_message));
140 revoke_packet->hdr.msg_type =
141 NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
142 revoke_packet->msg.v1_msg.
143 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
145 ret = vmbus_sendpacket(device->channel,
147 sizeof(struct nvsp_message),
148 (unsigned long)revoke_packet,
149 VM_PKT_DATA_INBAND, 0);
151 * If we failed here, we might as well return and
152 * have a leak rather than continue and a bugchk
155 netdev_err(ndev, "unable to send "
156 "revoke receive buffer to netvsp\n");
161 /* Teardown the gpadl on the vsp end */
162 if (net_device->recv_buf_gpadl_handle) {
163 ret = vmbus_teardown_gpadl(device->channel,
164 net_device->recv_buf_gpadl_handle);
166 /* If we failed here, we might as well return and have a leak
167 * rather than continue and a bugchk
171 "unable to teardown receive buffer's gpadl\n");
174 net_device->recv_buf_gpadl_handle = 0;
177 if (net_device->recv_buf) {
178 /* Free up the receive buffer */
179 vfree(net_device->recv_buf);
180 net_device->recv_buf = NULL;
183 if (net_device->recv_section) {
184 net_device->recv_section_cnt = 0;
185 kfree(net_device->recv_section);
186 net_device->recv_section = NULL;
189 /* Deal with the send buffer we may have setup.
190 * If we got a send section size, it means we received a
191 * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
192 * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
193 * to send a revoke msg here
195 if (net_device->send_section_size) {
196 /* Send the revoke receive buffer */
197 revoke_packet = &net_device->revoke_packet;
198 memset(revoke_packet, 0, sizeof(struct nvsp_message));
200 revoke_packet->hdr.msg_type =
201 NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
202 revoke_packet->msg.v1_msg.revoke_send_buf.id =
203 NETVSC_SEND_BUFFER_ID;
205 ret = vmbus_sendpacket(device->channel,
207 sizeof(struct nvsp_message),
208 (unsigned long)revoke_packet,
209 VM_PKT_DATA_INBAND, 0);
210 /* If we failed here, we might as well return and
211 * have a leak rather than continue and a bugchk
214 netdev_err(ndev, "unable to send "
215 "revoke send buffer to netvsp\n");
219 /* Teardown the gpadl on the vsp end */
220 if (net_device->send_buf_gpadl_handle) {
221 ret = vmbus_teardown_gpadl(device->channel,
222 net_device->send_buf_gpadl_handle);
224 /* If we failed here, we might as well return and have a leak
225 * rather than continue and a bugchk
229 "unable to teardown send buffer's gpadl\n");
232 net_device->send_buf_gpadl_handle = 0;
234 if (net_device->send_buf) {
235 /* Free up the send buffer */
236 vfree(net_device->send_buf);
237 net_device->send_buf = NULL;
239 kfree(net_device->send_section_map);
244 static int netvsc_init_buf(struct hv_device *device)
248 struct netvsc_device *net_device;
249 struct nvsp_message *init_packet;
250 struct net_device *ndev;
253 net_device = get_outbound_net_device(device);
256 ndev = hv_get_drvdata(device);
258 node = cpu_to_node(device->channel->target_cpu);
259 net_device->recv_buf = vzalloc_node(net_device->recv_buf_size, node);
260 if (!net_device->recv_buf)
261 net_device->recv_buf = vzalloc(net_device->recv_buf_size);
263 if (!net_device->recv_buf) {
264 netdev_err(ndev, "unable to allocate receive "
265 "buffer of size %d\n", net_device->recv_buf_size);
271 * Establish the gpadl handle for this buffer on this
272 * channel. Note: This call uses the vmbus connection rather
273 * than the channel to establish the gpadl handle.
275 ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
276 net_device->recv_buf_size,
277 &net_device->recv_buf_gpadl_handle);
280 "unable to establish receive buffer's gpadl\n");
285 /* Notify the NetVsp of the gpadl handle */
286 init_packet = &net_device->channel_init_pkt;
288 memset(init_packet, 0, sizeof(struct nvsp_message));
290 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
291 init_packet->msg.v1_msg.send_recv_buf.
292 gpadl_handle = net_device->recv_buf_gpadl_handle;
293 init_packet->msg.v1_msg.
294 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
296 /* Send the gpadl notification request */
297 ret = vmbus_sendpacket(device->channel, init_packet,
298 sizeof(struct nvsp_message),
299 (unsigned long)init_packet,
301 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
304 "unable to send receive buffer's gpadl to netvsp\n");
308 t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
312 /* Check the response */
313 if (init_packet->msg.v1_msg.
314 send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
315 netdev_err(ndev, "Unable to complete receive buffer "
316 "initialization with NetVsp - status %d\n",
317 init_packet->msg.v1_msg.
318 send_recv_buf_complete.status);
323 /* Parse the response */
325 net_device->recv_section_cnt = init_packet->msg.
326 v1_msg.send_recv_buf_complete.num_sections;
328 net_device->recv_section = kmemdup(
329 init_packet->msg.v1_msg.send_recv_buf_complete.sections,
330 net_device->recv_section_cnt *
331 sizeof(struct nvsp_1_receive_buffer_section),
333 if (net_device->recv_section == NULL) {
339 * For 1st release, there should only be 1 section that represents the
340 * entire receive buffer
342 if (net_device->recv_section_cnt != 1 ||
343 net_device->recv_section->offset != 0) {
348 /* Now setup the send buffer.
350 net_device->send_buf = vzalloc_node(net_device->send_buf_size, node);
351 if (!net_device->send_buf)
352 net_device->send_buf = vzalloc(net_device->send_buf_size);
353 if (!net_device->send_buf) {
354 netdev_err(ndev, "unable to allocate send "
355 "buffer of size %d\n", net_device->send_buf_size);
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,
365 net_device->send_buf_size,
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 t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
396 /* Check the response */
397 if (init_packet->msg.v1_msg.
398 send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
399 netdev_err(ndev, "Unable to complete send buffer "
400 "initialization with NetVsp - status %d\n",
401 init_packet->msg.v1_msg.
402 send_send_buf_complete.status);
407 /* Parse the response */
408 net_device->send_section_size = init_packet->msg.
409 v1_msg.send_send_buf_complete.section_size;
411 /* Section count is simply the size divided by the section size.
413 net_device->send_section_cnt =
414 net_device->send_buf_size/net_device->send_section_size;
416 dev_info(&device->device, "Send section size: %d, Section count:%d\n",
417 net_device->send_section_size, net_device->send_section_cnt);
419 /* Setup state for managing the send buffer. */
420 net_device->map_words = DIV_ROUND_UP(net_device->send_section_cnt,
423 net_device->send_section_map =
424 kzalloc(net_device->map_words * sizeof(ulong), GFP_KERNEL);
425 if (net_device->send_section_map == NULL) {
433 netvsc_destroy_buf(device);
440 /* Negotiate NVSP protocol version */
441 static int negotiate_nvsp_ver(struct hv_device *device,
442 struct netvsc_device *net_device,
443 struct nvsp_message *init_packet,
446 struct net_device *ndev = hv_get_drvdata(device);
450 memset(init_packet, 0, sizeof(struct nvsp_message));
451 init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
452 init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
453 init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
455 /* Send the init request */
456 ret = vmbus_sendpacket(device->channel, init_packet,
457 sizeof(struct nvsp_message),
458 (unsigned long)init_packet,
460 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
465 t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
470 if (init_packet->msg.init_msg.init_complete.status !=
474 if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
477 /* NVSPv2 or later: Send NDIS config */
478 memset(init_packet, 0, sizeof(struct nvsp_message));
479 init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
480 init_packet->msg.v2_msg.send_ndis_config.mtu = ndev->mtu + ETH_HLEN;
481 init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
483 if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5)
484 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
486 ret = vmbus_sendpacket(device->channel, init_packet,
487 sizeof(struct nvsp_message),
488 (unsigned long)init_packet,
489 VM_PKT_DATA_INBAND, 0);
494 static int netvsc_connect_vsp(struct hv_device *device)
497 struct netvsc_device *net_device;
498 struct nvsp_message *init_packet;
500 u32 ver_list[] = { NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
501 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5 };
502 int i, num_ver = 4; /* number of different NVSP versions */
504 net_device = get_outbound_net_device(device);
508 init_packet = &net_device->channel_init_pkt;
510 /* Negotiate the latest NVSP protocol supported */
511 for (i = num_ver - 1; i >= 0; i--)
512 if (negotiate_nvsp_ver(device, net_device, init_packet,
514 net_device->nvsp_version = ver_list[i];
523 pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
525 /* Send the ndis version */
526 memset(init_packet, 0, sizeof(struct nvsp_message));
528 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
529 ndis_version = 0x00060001;
531 ndis_version = 0x0006001e;
533 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
534 init_packet->msg.v1_msg.
535 send_ndis_ver.ndis_major_ver =
536 (ndis_version & 0xFFFF0000) >> 16;
537 init_packet->msg.v1_msg.
538 send_ndis_ver.ndis_minor_ver =
539 ndis_version & 0xFFFF;
541 /* Send the init request */
542 ret = vmbus_sendpacket(device->channel, init_packet,
543 sizeof(struct nvsp_message),
544 (unsigned long)init_packet,
545 VM_PKT_DATA_INBAND, 0);
549 /* Post the big receive buffer to NetVSP */
550 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
551 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
553 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
554 net_device->send_buf_size = NETVSC_SEND_BUFFER_SIZE;
556 ret = netvsc_init_buf(device);
562 static void netvsc_disconnect_vsp(struct hv_device *device)
564 netvsc_destroy_buf(device);
568 * netvsc_device_remove - Callback when the root bus device is removed
570 int netvsc_device_remove(struct hv_device *device)
572 struct net_device *ndev = hv_get_drvdata(device);
573 struct net_device_context *net_device_ctx = netdev_priv(ndev);
574 struct netvsc_device *net_device = net_device_ctx->nvdev;
576 netvsc_disconnect_vsp(device);
578 net_device_ctx->nvdev = NULL;
581 * At this point, no one should be accessing net_device
584 dev_notice(&device->device, "net device safe to remove\n");
586 /* Now, we can close the channel safely */
587 vmbus_close(device->channel);
589 /* Release all resources */
590 vfree(net_device->sub_cb_buf);
591 free_netvsc_device(net_device);
596 #define RING_AVAIL_PERCENT_HIWATER 20
597 #define RING_AVAIL_PERCENT_LOWATER 10
600 * Get the percentage of available bytes to write in the ring.
601 * The return value is in range from 0 to 100.
603 static inline u32 hv_ringbuf_avail_percent(
604 struct hv_ring_buffer_info *ring_info)
606 u32 avail_read, avail_write;
608 hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
610 return avail_write * 100 / ring_info->ring_datasize;
613 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
616 sync_change_bit(index, net_device->send_section_map);
619 static void netvsc_send_completion(struct netvsc_device *net_device,
620 struct vmbus_channel *incoming_channel,
621 struct hv_device *device,
622 struct vmpacket_descriptor *packet)
624 struct nvsp_message *nvsp_packet;
625 struct hv_netvsc_packet *nvsc_packet;
626 struct net_device *ndev = hv_get_drvdata(device);
627 struct net_device_context *net_device_ctx = netdev_priv(ndev);
631 nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
632 (packet->offset8 << 3));
634 if ((nvsp_packet->hdr.msg_type == NVSP_MSG_TYPE_INIT_COMPLETE) ||
635 (nvsp_packet->hdr.msg_type ==
636 NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) ||
637 (nvsp_packet->hdr.msg_type ==
638 NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE) ||
639 (nvsp_packet->hdr.msg_type ==
640 NVSP_MSG5_TYPE_SUBCHANNEL)) {
641 /* Copy the response back */
642 memcpy(&net_device->channel_init_pkt, nvsp_packet,
643 sizeof(struct nvsp_message));
644 complete(&net_device->channel_init_wait);
645 } else if (nvsp_packet->hdr.msg_type ==
646 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) {
647 int num_outstanding_sends;
649 struct vmbus_channel *channel = device->channel;
652 /* Get the send context */
653 skb = (struct sk_buff *)(unsigned long)packet->trans_id;
655 /* Notify the layer above us */
657 nvsc_packet = (struct hv_netvsc_packet *) skb->cb;
658 send_index = nvsc_packet->send_buf_index;
659 if (send_index != NETVSC_INVALID_INDEX)
660 netvsc_free_send_slot(net_device, send_index);
661 q_idx = nvsc_packet->q_idx;
662 channel = incoming_channel;
663 dev_kfree_skb_any(skb);
666 num_outstanding_sends =
667 atomic_dec_return(&net_device->num_outstanding_sends);
668 queue_sends = atomic_dec_return(&net_device->
671 if (net_device->destroy && num_outstanding_sends == 0)
672 wake_up(&net_device->wait_drain);
674 if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev, q_idx)) &&
675 !net_device_ctx->start_remove &&
676 (hv_ringbuf_avail_percent(&channel->outbound) >
677 RING_AVAIL_PERCENT_HIWATER || queue_sends < 1))
678 netif_tx_wake_queue(netdev_get_tx_queue(
681 netdev_err(ndev, "Unknown send completion packet type- "
682 "%d received!!\n", nvsp_packet->hdr.msg_type);
687 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
690 u32 max_words = net_device->map_words;
691 unsigned long *map_addr = (unsigned long *)net_device->send_section_map;
692 u32 section_cnt = net_device->send_section_cnt;
693 int ret_val = NETVSC_INVALID_INDEX;
697 for (i = 0; i < max_words; i++) {
700 index = ffz(map_addr[i]);
701 prev_val = sync_test_and_set_bit(index, &map_addr[i]);
704 if ((index + (i * BITS_PER_LONG)) >= section_cnt)
706 ret_val = (index + (i * BITS_PER_LONG));
712 static u32 netvsc_copy_to_send_buf(struct netvsc_device *net_device,
713 unsigned int section_index,
715 struct hv_netvsc_packet *packet,
716 struct rndis_message *rndis_msg,
717 struct hv_page_buffer **pb,
720 char *start = net_device->send_buf;
721 char *dest = start + (section_index * net_device->send_section_size)
724 bool is_data_pkt = (skb != NULL) ? true : false;
725 bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
728 u32 remain = packet->total_data_buflen % net_device->pkt_align;
729 u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
730 packet->page_buf_cnt;
733 if (is_data_pkt && xmit_more && remain &&
734 !packet->cp_partial) {
735 padding = net_device->pkt_align - remain;
736 rndis_msg->msg_len += padding;
737 packet->total_data_buflen += padding;
740 for (i = 0; i < page_count; i++) {
741 char *src = phys_to_virt((*pb)[i].pfn << PAGE_SHIFT);
742 u32 offset = (*pb)[i].offset;
743 u32 len = (*pb)[i].len;
745 memcpy(dest, (src + offset), len);
751 memset(dest, 0, padding);
758 static inline int netvsc_send_pkt(
759 struct hv_device *device,
760 struct hv_netvsc_packet *packet,
761 struct netvsc_device *net_device,
762 struct hv_page_buffer **pb,
765 struct nvsp_message nvmsg;
766 u16 q_idx = packet->q_idx;
767 struct vmbus_channel *out_channel = net_device->chn_table[q_idx];
768 struct net_device *ndev = hv_get_drvdata(device);
771 struct hv_page_buffer *pgbuf;
772 u32 ring_avail = hv_ringbuf_avail_percent(&out_channel->outbound);
773 bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
775 nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
778 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 0;
780 /* 1 is RMC_CONTROL; */
781 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 1;
784 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
785 packet->send_buf_index;
786 if (packet->send_buf_index == NETVSC_INVALID_INDEX)
787 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
789 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size =
790 packet->total_data_buflen;
794 if (out_channel->rescind)
798 * It is possible that once we successfully place this packet
799 * on the ringbuffer, we may stop the queue. In that case, we want
800 * to notify the host independent of the xmit_more flag. We don't
801 * need to be precise here; in the worst case we may signal the host
804 if (ring_avail < (RING_AVAIL_PERCENT_LOWATER + 1))
807 if (packet->page_buf_cnt) {
808 pgbuf = packet->cp_partial ? (*pb) +
809 packet->rmsg_pgcnt : (*pb);
810 ret = vmbus_sendpacket_pagebuffer_ctl(out_channel,
812 packet->page_buf_cnt,
814 sizeof(struct nvsp_message),
816 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED,
819 ret = vmbus_sendpacket_ctl(out_channel, &nvmsg,
820 sizeof(struct nvsp_message),
823 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED,
828 atomic_inc(&net_device->num_outstanding_sends);
829 atomic_inc(&net_device->queue_sends[q_idx]);
831 if (ring_avail < RING_AVAIL_PERCENT_LOWATER) {
832 netif_tx_stop_queue(netdev_get_tx_queue(ndev, q_idx));
834 if (atomic_read(&net_device->
835 queue_sends[q_idx]) < 1)
836 netif_tx_wake_queue(netdev_get_tx_queue(
839 } else if (ret == -EAGAIN) {
840 netif_tx_stop_queue(netdev_get_tx_queue(
842 if (atomic_read(&net_device->queue_sends[q_idx]) < 1) {
843 netif_tx_wake_queue(netdev_get_tx_queue(
848 netdev_err(ndev, "Unable to send packet %p ret %d\n",
855 /* Move packet out of multi send data (msd), and clear msd */
856 static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
857 struct sk_buff **msd_skb,
858 struct multi_send_data *msdp)
860 *msd_skb = msdp->skb;
861 *msd_send = msdp->pkt;
867 int netvsc_send(struct hv_device *device,
868 struct hv_netvsc_packet *packet,
869 struct rndis_message *rndis_msg,
870 struct hv_page_buffer **pb,
873 struct netvsc_device *net_device;
874 int ret = 0, m_ret = 0;
875 struct vmbus_channel *out_channel;
876 u16 q_idx = packet->q_idx;
877 u32 pktlen = packet->total_data_buflen, msd_len = 0;
878 unsigned int section_index = NETVSC_INVALID_INDEX;
879 struct multi_send_data *msdp;
880 struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
881 struct sk_buff *msd_skb = NULL;
883 bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
885 net_device = get_outbound_net_device(device);
889 out_channel = net_device->chn_table[q_idx];
891 packet->send_buf_index = NETVSC_INVALID_INDEX;
892 packet->cp_partial = false;
894 /* Send control message directly without accessing msd (Multi-Send
895 * Data) field which may be changed during data packet processing.
902 msdp = &net_device->msd[q_idx];
904 /* batch packets in send buffer if possible */
906 msd_len = msdp->pkt->total_data_buflen;
908 try_batch = (skb != NULL) && msd_len > 0 && msdp->count <
911 if (try_batch && msd_len + pktlen + net_device->pkt_align <
912 net_device->send_section_size) {
913 section_index = msdp->pkt->send_buf_index;
915 } else if (try_batch && msd_len + packet->rmsg_size <
916 net_device->send_section_size) {
917 section_index = msdp->pkt->send_buf_index;
918 packet->cp_partial = true;
920 } else if ((skb != NULL) && pktlen + net_device->pkt_align <
921 net_device->send_section_size) {
922 section_index = netvsc_get_next_send_section(net_device);
923 if (section_index != NETVSC_INVALID_INDEX) {
924 move_pkt_msd(&msd_send, &msd_skb, msdp);
929 if (section_index != NETVSC_INVALID_INDEX) {
930 netvsc_copy_to_send_buf(net_device,
931 section_index, msd_len,
932 packet, rndis_msg, pb, skb);
934 packet->send_buf_index = section_index;
936 if (packet->cp_partial) {
937 packet->page_buf_cnt -= packet->rmsg_pgcnt;
938 packet->total_data_buflen = msd_len + packet->rmsg_size;
940 packet->page_buf_cnt = 0;
941 packet->total_data_buflen += msd_len;
945 dev_kfree_skb_any(msdp->skb);
947 if (xmit_more && !packet->cp_partial) {
958 move_pkt_msd(&msd_send, &msd_skb, msdp);
963 m_ret = netvsc_send_pkt(device, msd_send, net_device,
967 netvsc_free_send_slot(net_device,
968 msd_send->send_buf_index);
969 dev_kfree_skb_any(msd_skb);
975 ret = netvsc_send_pkt(device, cur_send, net_device, pb, skb);
977 if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
978 netvsc_free_send_slot(net_device, section_index);
983 static void netvsc_send_recv_completion(struct hv_device *device,
984 struct vmbus_channel *channel,
985 struct netvsc_device *net_device,
986 u64 transaction_id, u32 status)
988 struct nvsp_message recvcompMessage;
991 struct net_device *ndev = hv_get_drvdata(device);
993 recvcompMessage.hdr.msg_type =
994 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
996 recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status = status;
999 /* Send the completion */
1000 ret = vmbus_sendpacket(channel, &recvcompMessage,
1001 sizeof(struct nvsp_message), transaction_id,
1006 } else if (ret == -EAGAIN) {
1007 /* no more room...wait a bit and attempt to retry 3 times */
1009 netdev_err(ndev, "unable to send receive completion pkt"
1010 " (tid %llx)...retrying %d\n", transaction_id, retries);
1014 goto retry_send_cmplt;
1016 netdev_err(ndev, "unable to send receive "
1017 "completion pkt (tid %llx)...give up retrying\n",
1021 netdev_err(ndev, "unable to send receive "
1022 "completion pkt - %llx\n", transaction_id);
1026 static void netvsc_receive(struct netvsc_device *net_device,
1027 struct vmbus_channel *channel,
1028 struct hv_device *device,
1029 struct vmpacket_descriptor *packet)
1031 struct vmtransfer_page_packet_header *vmxferpage_packet;
1032 struct nvsp_message *nvsp_packet;
1033 struct hv_netvsc_packet nv_pkt;
1034 struct hv_netvsc_packet *netvsc_packet = &nv_pkt;
1035 u32 status = NVSP_STAT_SUCCESS;
1038 struct net_device *ndev = hv_get_drvdata(device);
1042 * All inbound packets other than send completion should be xfer page
1045 if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
1046 netdev_err(ndev, "Unknown packet type received - %d\n",
1051 nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
1052 (packet->offset8 << 3));
1054 /* Make sure this is a valid nvsp packet */
1055 if (nvsp_packet->hdr.msg_type !=
1056 NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
1057 netdev_err(ndev, "Unknown nvsp packet type received-"
1058 " %d\n", nvsp_packet->hdr.msg_type);
1062 vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
1064 if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
1065 netdev_err(ndev, "Invalid xfer page set id - "
1066 "expecting %x got %x\n", 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 /* Initialize the netvsc packet */
1076 data = (void *)((unsigned long)net_device->
1077 recv_buf + vmxferpage_packet->ranges[i].byte_offset);
1078 netvsc_packet->total_data_buflen =
1079 vmxferpage_packet->ranges[i].byte_count;
1081 /* Pass it to the upper layer */
1082 status = rndis_filter_receive(device, netvsc_packet, &data,
1087 netvsc_send_recv_completion(device, channel, net_device,
1088 vmxferpage_packet->d.trans_id, status);
1092 static void netvsc_send_table(struct hv_device *hdev,
1093 struct nvsp_message *nvmsg)
1095 struct netvsc_device *nvscdev;
1096 struct net_device *ndev = hv_get_drvdata(hdev);
1100 nvscdev = get_outbound_net_device(hdev);
1104 count = nvmsg->msg.v5_msg.send_table.count;
1105 if (count != VRSS_SEND_TAB_SIZE) {
1106 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1110 tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
1111 nvmsg->msg.v5_msg.send_table.offset);
1113 for (i = 0; i < count; i++)
1114 nvscdev->send_table[i] = tab[i];
1117 static void netvsc_send_vf(struct netvsc_device *nvdev,
1118 struct nvsp_message *nvmsg)
1120 nvdev->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1121 nvdev->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1124 static inline void netvsc_receive_inband(struct hv_device *hdev,
1125 struct netvsc_device *nvdev,
1126 struct nvsp_message *nvmsg)
1128 switch (nvmsg->hdr.msg_type) {
1129 case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1130 netvsc_send_table(hdev, nvmsg);
1133 case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1134 netvsc_send_vf(nvdev, nvmsg);
1139 void netvsc_channel_cb(void *context)
1142 struct vmbus_channel *channel = (struct vmbus_channel *)context;
1143 struct hv_device *device;
1144 struct netvsc_device *net_device;
1147 struct vmpacket_descriptor *desc;
1148 unsigned char *buffer;
1149 int bufferlen = NETVSC_PACKET_SIZE;
1150 struct net_device *ndev;
1151 struct nvsp_message *nvmsg;
1153 if (channel->primary_channel != NULL)
1154 device = channel->primary_channel->device_obj;
1156 device = channel->device_obj;
1158 net_device = get_inbound_net_device(device);
1161 ndev = hv_get_drvdata(device);
1162 buffer = get_per_channel_state(channel);
1165 ret = vmbus_recvpacket_raw(channel, buffer, bufferlen,
1166 &bytes_recvd, &request_id);
1168 if (bytes_recvd > 0) {
1169 desc = (struct vmpacket_descriptor *)buffer;
1170 nvmsg = (struct nvsp_message *)((unsigned long)
1171 desc + (desc->offset8 << 3));
1172 switch (desc->type) {
1174 netvsc_send_completion(net_device,
1179 case VM_PKT_DATA_USING_XFER_PAGES:
1180 netvsc_receive(net_device, channel,
1184 case VM_PKT_DATA_INBAND:
1185 netvsc_receive_inband(device,
1192 "unhandled packet type %d, "
1193 "tid %llx len %d\n",
1194 desc->type, request_id,
1201 * We are done for this pass.
1206 } else if (ret == -ENOBUFS) {
1207 if (bufferlen > NETVSC_PACKET_SIZE)
1209 /* Handle large packet */
1210 buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
1211 if (buffer == NULL) {
1212 /* Try again next time around */
1214 "unable to allocate buffer of size "
1215 "(%d)!!\n", bytes_recvd);
1219 bufferlen = bytes_recvd;
1223 if (bufferlen > NETVSC_PACKET_SIZE)
1229 * netvsc_device_add - Callback when the device belonging to this
1232 int netvsc_device_add(struct hv_device *device, void *additional_info)
1236 ((struct netvsc_device_info *)additional_info)->ring_size;
1237 struct netvsc_device *net_device;
1238 struct net_device *ndev = hv_get_drvdata(device);
1239 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1241 net_device = alloc_net_device();
1245 net_device->ring_size = ring_size;
1247 /* Initialize the NetVSC channel extension */
1248 init_completion(&net_device->channel_init_wait);
1250 set_per_channel_state(device->channel, net_device->cb_buffer);
1252 /* Open the channel */
1253 ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
1254 ring_size * PAGE_SIZE, NULL, 0,
1255 netvsc_channel_cb, device->channel);
1258 netdev_err(ndev, "unable to open channel: %d\n", ret);
1262 /* Channel is opened */
1263 pr_info("hv_netvsc channel opened successfully\n");
1265 /* If we're reopening the device we may have multiple queues, fill the
1266 * chn_table with the default channel to use it before subchannels are
1269 for (i = 0; i < VRSS_CHANNEL_MAX; i++)
1270 net_device->chn_table[i] = device->channel;
1272 /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1277 net_device_ctx->nvdev = net_device;
1279 /* Connect with the NetVsp */
1280 ret = netvsc_connect_vsp(device);
1283 "unable to connect to NetVSP - %d\n", ret);
1290 /* Now, we can close the channel safely */
1291 vmbus_close(device->channel);
1294 free_netvsc_device(net_device);