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 netvsc_device *nv_dev, bool vf)
42 struct nvsp_message *init_pkt = &nv_dev->channel_init_pkt;
43 struct hv_device *dev = nv_dev->dev;
45 memset(init_pkt, 0, sizeof(struct nvsp_message));
46 init_pkt->hdr.msg_type = NVSP_MSG4_TYPE_SWITCH_DATA_PATH;
48 init_pkt->msg.v4_msg.active_dp.active_datapath =
51 init_pkt->msg.v4_msg.active_dp.active_datapath =
52 NVSP_DATAPATH_SYNTHETIC;
54 vmbus_sendpacket(dev->channel, init_pkt,
55 sizeof(struct nvsp_message),
56 (unsigned long)init_pkt,
57 VM_PKT_DATA_INBAND, 0);
61 static struct netvsc_device *alloc_net_device(struct hv_device *device)
63 struct netvsc_device *net_device;
64 struct net_device *ndev = hv_get_drvdata(device);
66 net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
70 net_device->cb_buffer = kzalloc(NETVSC_PACKET_SIZE, GFP_KERNEL);
71 if (!net_device->cb_buffer) {
76 init_waitqueue_head(&net_device->wait_drain);
77 net_device->start_remove = false;
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->dev = device;
82 net_device->ndev = ndev;
83 net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
84 net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
86 net_device->vf_netdev = NULL;
87 net_device->vf_inject = false;
89 hv_set_drvdata(device, net_device);
93 static void free_netvsc_device(struct netvsc_device *nvdev)
95 kfree(nvdev->cb_buffer);
99 static struct netvsc_device *get_outbound_net_device(struct hv_device *device)
101 struct netvsc_device *net_device;
103 net_device = hv_get_drvdata(device);
104 if (net_device && net_device->destroy)
110 static struct netvsc_device *get_inbound_net_device(struct hv_device *device)
112 struct netvsc_device *net_device;
114 net_device = hv_get_drvdata(device);
119 if (net_device->destroy &&
120 atomic_read(&net_device->num_outstanding_sends) == 0)
128 static int netvsc_destroy_buf(struct netvsc_device *net_device)
130 struct nvsp_message *revoke_packet;
132 struct net_device *ndev = net_device->ndev;
135 * If we got a section count, it means we received a
136 * SendReceiveBufferComplete msg (ie sent
137 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
138 * to send a revoke msg here
140 if (net_device->recv_section_cnt) {
141 /* Send the revoke receive buffer */
142 revoke_packet = &net_device->revoke_packet;
143 memset(revoke_packet, 0, sizeof(struct nvsp_message));
145 revoke_packet->hdr.msg_type =
146 NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
147 revoke_packet->msg.v1_msg.
148 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
150 ret = vmbus_sendpacket(net_device->dev->channel,
152 sizeof(struct nvsp_message),
153 (unsigned long)revoke_packet,
154 VM_PKT_DATA_INBAND, 0);
156 * If we failed here, we might as well return and
157 * have a leak rather than continue and a bugchk
160 netdev_err(ndev, "unable to send "
161 "revoke receive buffer to netvsp\n");
166 /* Teardown the gpadl on the vsp end */
167 if (net_device->recv_buf_gpadl_handle) {
168 ret = vmbus_teardown_gpadl(net_device->dev->channel,
169 net_device->recv_buf_gpadl_handle);
171 /* If we failed here, we might as well return and have a leak
172 * rather than continue and a bugchk
176 "unable to teardown receive buffer's gpadl\n");
179 net_device->recv_buf_gpadl_handle = 0;
182 if (net_device->recv_buf) {
183 /* Free up the receive buffer */
184 vfree(net_device->recv_buf);
185 net_device->recv_buf = NULL;
188 if (net_device->recv_section) {
189 net_device->recv_section_cnt = 0;
190 kfree(net_device->recv_section);
191 net_device->recv_section = NULL;
194 /* Deal with the send buffer we may have setup.
195 * If we got a send section size, it means we received a
196 * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
197 * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
198 * to send a revoke msg here
200 if (net_device->send_section_size) {
201 /* Send the revoke receive buffer */
202 revoke_packet = &net_device->revoke_packet;
203 memset(revoke_packet, 0, sizeof(struct nvsp_message));
205 revoke_packet->hdr.msg_type =
206 NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
207 revoke_packet->msg.v1_msg.revoke_send_buf.id =
208 NETVSC_SEND_BUFFER_ID;
210 ret = vmbus_sendpacket(net_device->dev->channel,
212 sizeof(struct nvsp_message),
213 (unsigned long)revoke_packet,
214 VM_PKT_DATA_INBAND, 0);
215 /* If we failed here, we might as well return and
216 * have a leak rather than continue and a bugchk
219 netdev_err(ndev, "unable to send "
220 "revoke send buffer to netvsp\n");
224 /* Teardown the gpadl on the vsp end */
225 if (net_device->send_buf_gpadl_handle) {
226 ret = vmbus_teardown_gpadl(net_device->dev->channel,
227 net_device->send_buf_gpadl_handle);
229 /* If we failed here, we might as well return and have a leak
230 * rather than continue and a bugchk
234 "unable to teardown send buffer's gpadl\n");
237 net_device->send_buf_gpadl_handle = 0;
239 if (net_device->send_buf) {
240 /* Free up the send buffer */
241 vfree(net_device->send_buf);
242 net_device->send_buf = NULL;
244 kfree(net_device->send_section_map);
249 static int netvsc_init_buf(struct hv_device *device)
253 struct netvsc_device *net_device;
254 struct nvsp_message *init_packet;
255 struct net_device *ndev;
258 net_device = get_outbound_net_device(device);
261 ndev = net_device->ndev;
263 node = cpu_to_node(device->channel->target_cpu);
264 net_device->recv_buf = vzalloc_node(net_device->recv_buf_size, node);
265 if (!net_device->recv_buf)
266 net_device->recv_buf = vzalloc(net_device->recv_buf_size);
268 if (!net_device->recv_buf) {
269 netdev_err(ndev, "unable to allocate receive "
270 "buffer of size %d\n", net_device->recv_buf_size);
276 * Establish the gpadl handle for this buffer on this
277 * channel. Note: This call uses the vmbus connection rather
278 * than the channel to establish the gpadl handle.
280 ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
281 net_device->recv_buf_size,
282 &net_device->recv_buf_gpadl_handle);
285 "unable to establish receive buffer's gpadl\n");
290 /* Notify the NetVsp of the gpadl handle */
291 init_packet = &net_device->channel_init_pkt;
293 memset(init_packet, 0, sizeof(struct nvsp_message));
295 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
296 init_packet->msg.v1_msg.send_recv_buf.
297 gpadl_handle = net_device->recv_buf_gpadl_handle;
298 init_packet->msg.v1_msg.
299 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
301 /* Send the gpadl notification request */
302 ret = vmbus_sendpacket(device->channel, init_packet,
303 sizeof(struct nvsp_message),
304 (unsigned long)init_packet,
306 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
309 "unable to send receive buffer's gpadl to netvsp\n");
313 t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
317 /* Check the response */
318 if (init_packet->msg.v1_msg.
319 send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
320 netdev_err(ndev, "Unable to complete receive buffer "
321 "initialization with NetVsp - status %d\n",
322 init_packet->msg.v1_msg.
323 send_recv_buf_complete.status);
328 /* Parse the response */
330 net_device->recv_section_cnt = init_packet->msg.
331 v1_msg.send_recv_buf_complete.num_sections;
333 net_device->recv_section = kmemdup(
334 init_packet->msg.v1_msg.send_recv_buf_complete.sections,
335 net_device->recv_section_cnt *
336 sizeof(struct nvsp_1_receive_buffer_section),
338 if (net_device->recv_section == NULL) {
344 * For 1st release, there should only be 1 section that represents the
345 * entire receive buffer
347 if (net_device->recv_section_cnt != 1 ||
348 net_device->recv_section->offset != 0) {
353 /* Now setup the send buffer.
355 net_device->send_buf = vzalloc_node(net_device->send_buf_size, node);
356 if (!net_device->send_buf)
357 net_device->send_buf = vzalloc(net_device->send_buf_size);
358 if (!net_device->send_buf) {
359 netdev_err(ndev, "unable to allocate send "
360 "buffer of size %d\n", net_device->send_buf_size);
365 /* Establish the gpadl handle for this buffer on this
366 * channel. Note: This call uses the vmbus connection rather
367 * than the channel to establish the gpadl handle.
369 ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
370 net_device->send_buf_size,
371 &net_device->send_buf_gpadl_handle);
374 "unable to establish send buffer's gpadl\n");
378 /* Notify the NetVsp of the gpadl handle */
379 init_packet = &net_device->channel_init_pkt;
380 memset(init_packet, 0, sizeof(struct nvsp_message));
381 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
382 init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
383 net_device->send_buf_gpadl_handle;
384 init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
386 /* Send the gpadl notification request */
387 ret = vmbus_sendpacket(device->channel, init_packet,
388 sizeof(struct nvsp_message),
389 (unsigned long)init_packet,
391 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
394 "unable to send send buffer's gpadl to netvsp\n");
398 t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
401 /* Check the response */
402 if (init_packet->msg.v1_msg.
403 send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
404 netdev_err(ndev, "Unable to complete send buffer "
405 "initialization with NetVsp - status %d\n",
406 init_packet->msg.v1_msg.
407 send_send_buf_complete.status);
412 /* Parse the response */
413 net_device->send_section_size = init_packet->msg.
414 v1_msg.send_send_buf_complete.section_size;
416 /* Section count is simply the size divided by the section size.
418 net_device->send_section_cnt =
419 net_device->send_buf_size/net_device->send_section_size;
421 dev_info(&device->device, "Send section size: %d, Section count:%d\n",
422 net_device->send_section_size, net_device->send_section_cnt);
424 /* Setup state for managing the send buffer. */
425 net_device->map_words = DIV_ROUND_UP(net_device->send_section_cnt,
428 net_device->send_section_map =
429 kzalloc(net_device->map_words * sizeof(ulong), GFP_KERNEL);
430 if (net_device->send_section_map == NULL) {
438 netvsc_destroy_buf(net_device);
445 /* Negotiate NVSP protocol version */
446 static int negotiate_nvsp_ver(struct hv_device *device,
447 struct netvsc_device *net_device,
448 struct nvsp_message *init_packet,
454 memset(init_packet, 0, sizeof(struct nvsp_message));
455 init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
456 init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
457 init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
459 /* Send the init request */
460 ret = vmbus_sendpacket(device->channel, init_packet,
461 sizeof(struct nvsp_message),
462 (unsigned long)init_packet,
464 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
469 t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
474 if (init_packet->msg.init_msg.init_complete.status !=
478 if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
481 /* NVSPv2 or later: Send NDIS config */
482 memset(init_packet, 0, sizeof(struct nvsp_message));
483 init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
484 init_packet->msg.v2_msg.send_ndis_config.mtu = net_device->ndev->mtu +
486 init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
488 if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5)
489 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
491 ret = vmbus_sendpacket(device->channel, init_packet,
492 sizeof(struct nvsp_message),
493 (unsigned long)init_packet,
494 VM_PKT_DATA_INBAND, 0);
499 static int netvsc_connect_vsp(struct hv_device *device)
502 struct netvsc_device *net_device;
503 struct nvsp_message *init_packet;
505 struct net_device *ndev;
506 u32 ver_list[] = { NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
507 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5 };
508 int i, num_ver = 4; /* number of different NVSP versions */
510 net_device = get_outbound_net_device(device);
513 ndev = net_device->ndev;
515 init_packet = &net_device->channel_init_pkt;
517 /* Negotiate the latest NVSP protocol supported */
518 for (i = num_ver - 1; i >= 0; i--)
519 if (negotiate_nvsp_ver(device, net_device, init_packet,
521 net_device->nvsp_version = ver_list[i];
530 pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
532 /* Send the ndis version */
533 memset(init_packet, 0, sizeof(struct nvsp_message));
535 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
536 ndis_version = 0x00060001;
538 ndis_version = 0x0006001e;
540 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
541 init_packet->msg.v1_msg.
542 send_ndis_ver.ndis_major_ver =
543 (ndis_version & 0xFFFF0000) >> 16;
544 init_packet->msg.v1_msg.
545 send_ndis_ver.ndis_minor_ver =
546 ndis_version & 0xFFFF;
548 /* Send the init request */
549 ret = vmbus_sendpacket(device->channel, init_packet,
550 sizeof(struct nvsp_message),
551 (unsigned long)init_packet,
552 VM_PKT_DATA_INBAND, 0);
556 /* Post the big receive buffer to NetVSP */
557 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
558 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
560 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
561 net_device->send_buf_size = NETVSC_SEND_BUFFER_SIZE;
563 ret = netvsc_init_buf(device);
569 static void netvsc_disconnect_vsp(struct netvsc_device *net_device)
571 netvsc_destroy_buf(net_device);
575 * netvsc_device_remove - Callback when the root bus device is removed
577 int netvsc_device_remove(struct hv_device *device)
579 struct netvsc_device *net_device;
582 net_device = hv_get_drvdata(device);
584 netvsc_disconnect_vsp(net_device);
587 * Since we have already drained, we don't need to busy wait
588 * as was done in final_release_stor_device()
589 * Note that we cannot set the ext pointer to NULL until
590 * we have drained - to drain the outgoing packets, we need to
591 * allow incoming packets.
594 spin_lock_irqsave(&device->channel->inbound_lock, flags);
595 hv_set_drvdata(device, NULL);
596 spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
599 * At this point, no one should be accessing net_device
602 dev_notice(&device->device, "net device safe to remove\n");
604 /* Now, we can close the channel safely */
605 vmbus_close(device->channel);
607 /* Release all resources */
608 vfree(net_device->sub_cb_buf);
609 free_netvsc_device(net_device);
614 #define RING_AVAIL_PERCENT_HIWATER 20
615 #define RING_AVAIL_PERCENT_LOWATER 10
618 * Get the percentage of available bytes to write in the ring.
619 * The return value is in range from 0 to 100.
621 static inline u32 hv_ringbuf_avail_percent(
622 struct hv_ring_buffer_info *ring_info)
624 u32 avail_read, avail_write;
626 hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
628 return avail_write * 100 / ring_info->ring_datasize;
631 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
634 sync_change_bit(index, net_device->send_section_map);
637 static void netvsc_send_completion(struct netvsc_device *net_device,
638 struct vmbus_channel *incoming_channel,
639 struct hv_device *device,
640 struct vmpacket_descriptor *packet)
642 struct nvsp_message *nvsp_packet;
643 struct hv_netvsc_packet *nvsc_packet;
644 struct net_device *ndev;
648 ndev = net_device->ndev;
650 nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
651 (packet->offset8 << 3));
653 if ((nvsp_packet->hdr.msg_type == NVSP_MSG_TYPE_INIT_COMPLETE) ||
654 (nvsp_packet->hdr.msg_type ==
655 NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) ||
656 (nvsp_packet->hdr.msg_type ==
657 NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE) ||
658 (nvsp_packet->hdr.msg_type ==
659 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);
664 } else if (nvsp_packet->hdr.msg_type ==
665 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) {
666 int num_outstanding_sends;
668 struct vmbus_channel *channel = device->channel;
671 /* Get the send context */
672 skb = (struct sk_buff *)(unsigned long)packet->trans_id;
674 /* Notify the layer above us */
676 nvsc_packet = (struct hv_netvsc_packet *) skb->cb;
677 send_index = nvsc_packet->send_buf_index;
678 if (send_index != NETVSC_INVALID_INDEX)
679 netvsc_free_send_slot(net_device, send_index);
680 q_idx = nvsc_packet->q_idx;
681 channel = incoming_channel;
682 dev_kfree_skb_any(skb);
685 num_outstanding_sends =
686 atomic_dec_return(&net_device->num_outstanding_sends);
687 queue_sends = atomic_dec_return(&net_device->
690 if (net_device->destroy && num_outstanding_sends == 0)
691 wake_up(&net_device->wait_drain);
693 if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev, q_idx)) &&
694 !net_device->start_remove &&
695 (hv_ringbuf_avail_percent(&channel->outbound) >
696 RING_AVAIL_PERCENT_HIWATER || queue_sends < 1))
697 netif_tx_wake_queue(netdev_get_tx_queue(
700 netdev_err(ndev, "Unknown send completion packet type- "
701 "%d received!!\n", nvsp_packet->hdr.msg_type);
706 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
709 u32 max_words = net_device->map_words;
710 unsigned long *map_addr = (unsigned long *)net_device->send_section_map;
711 u32 section_cnt = net_device->send_section_cnt;
712 int ret_val = NETVSC_INVALID_INDEX;
716 for (i = 0; i < max_words; i++) {
719 index = ffz(map_addr[i]);
720 prev_val = sync_test_and_set_bit(index, &map_addr[i]);
723 if ((index + (i * BITS_PER_LONG)) >= section_cnt)
725 ret_val = (index + (i * BITS_PER_LONG));
731 static u32 netvsc_copy_to_send_buf(struct netvsc_device *net_device,
732 unsigned int section_index,
734 struct hv_netvsc_packet *packet,
735 struct rndis_message *rndis_msg,
736 struct hv_page_buffer **pb,
739 char *start = net_device->send_buf;
740 char *dest = start + (section_index * net_device->send_section_size)
743 bool is_data_pkt = (skb != NULL) ? true : false;
744 bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
747 u32 remain = packet->total_data_buflen % net_device->pkt_align;
748 u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
749 packet->page_buf_cnt;
752 if (is_data_pkt && xmit_more && remain &&
753 !packet->cp_partial) {
754 padding = net_device->pkt_align - remain;
755 rndis_msg->msg_len += padding;
756 packet->total_data_buflen += padding;
759 for (i = 0; i < page_count; i++) {
760 char *src = phys_to_virt((*pb)[i].pfn << PAGE_SHIFT);
761 u32 offset = (*pb)[i].offset;
762 u32 len = (*pb)[i].len;
764 memcpy(dest, (src + offset), len);
770 memset(dest, 0, padding);
777 static inline int netvsc_send_pkt(
778 struct hv_netvsc_packet *packet,
779 struct netvsc_device *net_device,
780 struct hv_page_buffer **pb,
783 struct nvsp_message nvmsg;
784 u16 q_idx = packet->q_idx;
785 struct vmbus_channel *out_channel = net_device->chn_table[q_idx];
786 struct net_device *ndev = net_device->ndev;
789 struct hv_page_buffer *pgbuf;
790 u32 ring_avail = hv_ringbuf_avail_percent(&out_channel->outbound);
791 bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
793 nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
796 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 0;
798 /* 1 is RMC_CONTROL; */
799 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 1;
802 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
803 packet->send_buf_index;
804 if (packet->send_buf_index == NETVSC_INVALID_INDEX)
805 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
807 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size =
808 packet->total_data_buflen;
812 if (out_channel->rescind)
816 * It is possible that once we successfully place this packet
817 * on the ringbuffer, we may stop the queue. In that case, we want
818 * to notify the host independent of the xmit_more flag. We don't
819 * need to be precise here; in the worst case we may signal the host
822 if (ring_avail < (RING_AVAIL_PERCENT_LOWATER + 1))
825 if (packet->page_buf_cnt) {
826 pgbuf = packet->cp_partial ? (*pb) +
827 packet->rmsg_pgcnt : (*pb);
828 ret = vmbus_sendpacket_pagebuffer_ctl(out_channel,
830 packet->page_buf_cnt,
832 sizeof(struct nvsp_message),
834 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED,
837 ret = vmbus_sendpacket_ctl(out_channel, &nvmsg,
838 sizeof(struct nvsp_message),
841 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED,
846 atomic_inc(&net_device->num_outstanding_sends);
847 atomic_inc(&net_device->queue_sends[q_idx]);
849 if (ring_avail < RING_AVAIL_PERCENT_LOWATER) {
850 netif_tx_stop_queue(netdev_get_tx_queue(ndev, q_idx));
852 if (atomic_read(&net_device->
853 queue_sends[q_idx]) < 1)
854 netif_tx_wake_queue(netdev_get_tx_queue(
857 } else if (ret == -EAGAIN) {
858 netif_tx_stop_queue(netdev_get_tx_queue(
860 if (atomic_read(&net_device->queue_sends[q_idx]) < 1) {
861 netif_tx_wake_queue(netdev_get_tx_queue(
866 netdev_err(ndev, "Unable to send packet %p ret %d\n",
873 /* Move packet out of multi send data (msd), and clear msd */
874 static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
875 struct sk_buff **msd_skb,
876 struct multi_send_data *msdp)
878 *msd_skb = msdp->skb;
879 *msd_send = msdp->pkt;
885 int netvsc_send(struct hv_device *device,
886 struct hv_netvsc_packet *packet,
887 struct rndis_message *rndis_msg,
888 struct hv_page_buffer **pb,
891 struct netvsc_device *net_device;
892 int ret = 0, m_ret = 0;
893 struct vmbus_channel *out_channel;
894 u16 q_idx = packet->q_idx;
895 u32 pktlen = packet->total_data_buflen, msd_len = 0;
896 unsigned int section_index = NETVSC_INVALID_INDEX;
897 struct multi_send_data *msdp;
898 struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
899 struct sk_buff *msd_skb = NULL;
901 bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
903 net_device = get_outbound_net_device(device);
907 out_channel = net_device->chn_table[q_idx];
909 packet->send_buf_index = NETVSC_INVALID_INDEX;
910 packet->cp_partial = false;
912 /* Send control message directly without accessing msd (Multi-Send
913 * Data) field which may be changed during data packet processing.
920 msdp = &net_device->msd[q_idx];
922 /* batch packets in send buffer if possible */
924 msd_len = msdp->pkt->total_data_buflen;
926 try_batch = (skb != NULL) && msd_len > 0 && msdp->count <
929 if (try_batch && msd_len + pktlen + net_device->pkt_align <
930 net_device->send_section_size) {
931 section_index = msdp->pkt->send_buf_index;
933 } else if (try_batch && msd_len + packet->rmsg_size <
934 net_device->send_section_size) {
935 section_index = msdp->pkt->send_buf_index;
936 packet->cp_partial = true;
938 } else if ((skb != NULL) && pktlen + net_device->pkt_align <
939 net_device->send_section_size) {
940 section_index = netvsc_get_next_send_section(net_device);
941 if (section_index != NETVSC_INVALID_INDEX) {
942 move_pkt_msd(&msd_send, &msd_skb, msdp);
947 if (section_index != NETVSC_INVALID_INDEX) {
948 netvsc_copy_to_send_buf(net_device,
949 section_index, msd_len,
950 packet, rndis_msg, pb, skb);
952 packet->send_buf_index = section_index;
954 if (packet->cp_partial) {
955 packet->page_buf_cnt -= packet->rmsg_pgcnt;
956 packet->total_data_buflen = msd_len + packet->rmsg_size;
958 packet->page_buf_cnt = 0;
959 packet->total_data_buflen += msd_len;
963 dev_kfree_skb_any(msdp->skb);
965 if (xmit_more && !packet->cp_partial) {
976 move_pkt_msd(&msd_send, &msd_skb, msdp);
981 m_ret = netvsc_send_pkt(msd_send, net_device, NULL, msd_skb);
984 netvsc_free_send_slot(net_device,
985 msd_send->send_buf_index);
986 dev_kfree_skb_any(msd_skb);
992 ret = netvsc_send_pkt(cur_send, net_device, pb, skb);
994 if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
995 netvsc_free_send_slot(net_device, section_index);
1000 static void netvsc_send_recv_completion(struct hv_device *device,
1001 struct vmbus_channel *channel,
1002 struct netvsc_device *net_device,
1003 u64 transaction_id, u32 status)
1005 struct nvsp_message recvcompMessage;
1008 struct net_device *ndev;
1010 ndev = net_device->ndev;
1012 recvcompMessage.hdr.msg_type =
1013 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
1015 recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status = status;
1018 /* Send the completion */
1019 ret = vmbus_sendpacket(channel, &recvcompMessage,
1020 sizeof(struct nvsp_message), transaction_id,
1025 } else if (ret == -EAGAIN) {
1026 /* no more room...wait a bit and attempt to retry 3 times */
1028 netdev_err(ndev, "unable to send receive completion pkt"
1029 " (tid %llx)...retrying %d\n", transaction_id, retries);
1033 goto retry_send_cmplt;
1035 netdev_err(ndev, "unable to send receive "
1036 "completion pkt (tid %llx)...give up retrying\n",
1040 netdev_err(ndev, "unable to send receive "
1041 "completion pkt - %llx\n", transaction_id);
1045 static void netvsc_receive(struct netvsc_device *net_device,
1046 struct vmbus_channel *channel,
1047 struct hv_device *device,
1048 struct vmpacket_descriptor *packet)
1050 struct vmtransfer_page_packet_header *vmxferpage_packet;
1051 struct nvsp_message *nvsp_packet;
1052 struct hv_netvsc_packet nv_pkt;
1053 struct hv_netvsc_packet *netvsc_packet = &nv_pkt;
1054 u32 status = NVSP_STAT_SUCCESS;
1057 struct net_device *ndev;
1060 ndev = net_device->ndev;
1063 * All inbound packets other than send completion should be xfer page
1066 if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
1067 netdev_err(ndev, "Unknown packet type received - %d\n",
1072 nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
1073 (packet->offset8 << 3));
1075 /* Make sure this is a valid nvsp packet */
1076 if (nvsp_packet->hdr.msg_type !=
1077 NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
1078 netdev_err(ndev, "Unknown nvsp packet type received-"
1079 " %d\n", nvsp_packet->hdr.msg_type);
1083 vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
1085 if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
1086 netdev_err(ndev, "Invalid xfer page set id - "
1087 "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
1088 vmxferpage_packet->xfer_pageset_id);
1092 count = vmxferpage_packet->range_cnt;
1094 /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1095 for (i = 0; i < count; i++) {
1096 /* Initialize the netvsc packet */
1097 data = (void *)((unsigned long)net_device->
1098 recv_buf + vmxferpage_packet->ranges[i].byte_offset);
1099 netvsc_packet->total_data_buflen =
1100 vmxferpage_packet->ranges[i].byte_count;
1102 /* Pass it to the upper layer */
1103 status = rndis_filter_receive(device, netvsc_packet, &data,
1108 netvsc_send_recv_completion(device, channel, net_device,
1109 vmxferpage_packet->d.trans_id, status);
1113 static void netvsc_send_table(struct hv_device *hdev,
1114 struct nvsp_message *nvmsg)
1116 struct netvsc_device *nvscdev;
1117 struct net_device *ndev;
1121 nvscdev = get_outbound_net_device(hdev);
1124 ndev = nvscdev->ndev;
1126 count = nvmsg->msg.v5_msg.send_table.count;
1127 if (count != VRSS_SEND_TAB_SIZE) {
1128 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1132 tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
1133 nvmsg->msg.v5_msg.send_table.offset);
1135 for (i = 0; i < count; i++)
1136 nvscdev->send_table[i] = tab[i];
1139 static void netvsc_send_vf(struct netvsc_device *nvdev,
1140 struct nvsp_message *nvmsg)
1142 nvdev->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1143 nvdev->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1146 static inline void netvsc_receive_inband(struct hv_device *hdev,
1147 struct netvsc_device *nvdev,
1148 struct nvsp_message *nvmsg)
1150 switch (nvmsg->hdr.msg_type) {
1151 case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1152 netvsc_send_table(hdev, nvmsg);
1155 case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1156 netvsc_send_vf(nvdev, nvmsg);
1161 void netvsc_channel_cb(void *context)
1164 struct vmbus_channel *channel = (struct vmbus_channel *)context;
1165 struct hv_device *device;
1166 struct netvsc_device *net_device;
1169 struct vmpacket_descriptor *desc;
1170 unsigned char *buffer;
1171 int bufferlen = NETVSC_PACKET_SIZE;
1172 struct net_device *ndev;
1173 struct nvsp_message *nvmsg;
1175 if (channel->primary_channel != NULL)
1176 device = channel->primary_channel->device_obj;
1178 device = channel->device_obj;
1180 net_device = get_inbound_net_device(device);
1183 ndev = net_device->ndev;
1184 buffer = get_per_channel_state(channel);
1187 ret = vmbus_recvpacket_raw(channel, buffer, bufferlen,
1188 &bytes_recvd, &request_id);
1190 if (bytes_recvd > 0) {
1191 desc = (struct vmpacket_descriptor *)buffer;
1192 nvmsg = (struct nvsp_message *)((unsigned long)
1193 desc + (desc->offset8 << 3));
1194 switch (desc->type) {
1196 netvsc_send_completion(net_device,
1201 case VM_PKT_DATA_USING_XFER_PAGES:
1202 netvsc_receive(net_device, channel,
1206 case VM_PKT_DATA_INBAND:
1207 netvsc_receive_inband(device,
1214 "unhandled packet type %d, "
1215 "tid %llx len %d\n",
1216 desc->type, request_id,
1223 * We are done for this pass.
1228 } else if (ret == -ENOBUFS) {
1229 if (bufferlen > NETVSC_PACKET_SIZE)
1231 /* Handle large packet */
1232 buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
1233 if (buffer == NULL) {
1234 /* Try again next time around */
1236 "unable to allocate buffer of size "
1237 "(%d)!!\n", bytes_recvd);
1241 bufferlen = bytes_recvd;
1245 if (bufferlen > NETVSC_PACKET_SIZE)
1251 * netvsc_device_add - Callback when the device belonging to this
1254 int netvsc_device_add(struct hv_device *device, void *additional_info)
1258 ((struct netvsc_device_info *)additional_info)->ring_size;
1259 struct netvsc_device *net_device;
1260 struct net_device *ndev;
1262 net_device = alloc_net_device(device);
1266 net_device->ring_size = ring_size;
1269 * Coming into this function, struct net_device * is
1270 * registered as the driver private data.
1271 * In alloc_net_device(), we register struct netvsc_device *
1272 * as the driver private data and stash away struct net_device *
1273 * in struct netvsc_device *.
1275 ndev = net_device->ndev;
1277 /* Add netvsc_device context to netvsc_device */
1278 net_device->nd_ctx = netdev_priv(ndev);
1280 /* Initialize the NetVSC channel extension */
1281 init_completion(&net_device->channel_init_wait);
1283 set_per_channel_state(device->channel, net_device->cb_buffer);
1285 /* Open the channel */
1286 ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
1287 ring_size * PAGE_SIZE, NULL, 0,
1288 netvsc_channel_cb, device->channel);
1291 netdev_err(ndev, "unable to open channel: %d\n", ret);
1295 /* Channel is opened */
1296 pr_info("hv_netvsc channel opened successfully\n");
1298 net_device->chn_table[0] = device->channel;
1300 /* Connect with the NetVsp */
1301 ret = netvsc_connect_vsp(device);
1304 "unable to connect to NetVSP - %d\n", ret);
1311 /* Now, we can close the channel safely */
1312 vmbus_close(device->channel);
1315 free_netvsc_device(net_device);