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 net_device *ndev = hv_get_drvdata(device);
99 struct net_device_context *net_device_ctx = netdev_priv(ndev);
100 struct netvsc_device *net_device = net_device_ctx->nvdev;
102 if (net_device && net_device->destroy)
108 static struct netvsc_device *get_inbound_net_device(struct hv_device *device)
110 struct net_device *ndev = hv_get_drvdata(device);
111 struct net_device_context *net_device_ctx = netdev_priv(ndev);
112 struct netvsc_device *net_device = net_device_ctx->nvdev;
117 if (net_device->destroy &&
118 atomic_read(&net_device->num_outstanding_sends) == 0)
126 static int netvsc_destroy_buf(struct hv_device *device)
128 struct nvsp_message *revoke_packet;
130 struct net_device *ndev = hv_get_drvdata(device);
131 struct net_device_context *net_device_ctx = netdev_priv(ndev);
132 struct netvsc_device *net_device = net_device_ctx->nvdev;
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(device->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(device->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(device->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(device->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 = hv_get_drvdata(device);
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(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,
451 struct net_device *ndev = hv_get_drvdata(device);
455 memset(init_packet, 0, sizeof(struct nvsp_message));
456 init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
457 init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
458 init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
460 /* Send the init request */
461 ret = vmbus_sendpacket(device->channel, init_packet,
462 sizeof(struct nvsp_message),
463 (unsigned long)init_packet,
465 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
470 t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
475 if (init_packet->msg.init_msg.init_complete.status !=
479 if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
482 /* NVSPv2 or later: Send NDIS config */
483 memset(init_packet, 0, sizeof(struct nvsp_message));
484 init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
485 init_packet->msg.v2_msg.send_ndis_config.mtu = ndev->mtu + ETH_HLEN;
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 u32 ver_list[] = { NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
506 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5 };
507 int i, num_ver = 4; /* number of different NVSP versions */
509 net_device = get_outbound_net_device(device);
513 init_packet = &net_device->channel_init_pkt;
515 /* Negotiate the latest NVSP protocol supported */
516 for (i = num_ver - 1; i >= 0; i--)
517 if (negotiate_nvsp_ver(device, net_device, init_packet,
519 net_device->nvsp_version = ver_list[i];
528 pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
530 /* Send the ndis version */
531 memset(init_packet, 0, sizeof(struct nvsp_message));
533 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
534 ndis_version = 0x00060001;
536 ndis_version = 0x0006001e;
538 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
539 init_packet->msg.v1_msg.
540 send_ndis_ver.ndis_major_ver =
541 (ndis_version & 0xFFFF0000) >> 16;
542 init_packet->msg.v1_msg.
543 send_ndis_ver.ndis_minor_ver =
544 ndis_version & 0xFFFF;
546 /* Send the init request */
547 ret = vmbus_sendpacket(device->channel, init_packet,
548 sizeof(struct nvsp_message),
549 (unsigned long)init_packet,
550 VM_PKT_DATA_INBAND, 0);
554 /* Post the big receive buffer to NetVSP */
555 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
556 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
558 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
559 net_device->send_buf_size = NETVSC_SEND_BUFFER_SIZE;
561 ret = netvsc_init_buf(device);
567 static void netvsc_disconnect_vsp(struct hv_device *device)
569 netvsc_destroy_buf(device);
573 * netvsc_device_remove - Callback when the root bus device is removed
575 int netvsc_device_remove(struct hv_device *device)
577 struct net_device *ndev = hv_get_drvdata(device);
578 struct net_device_context *net_device_ctx = netdev_priv(ndev);
579 struct netvsc_device *net_device = net_device_ctx->nvdev;
581 netvsc_disconnect_vsp(device);
583 net_device_ctx->nvdev = NULL;
586 * At this point, no one should be accessing net_device
589 dev_notice(&device->device, "net device safe to remove\n");
591 /* Now, we can close the channel safely */
592 vmbus_close(device->channel);
594 /* Release all resources */
595 vfree(net_device->sub_cb_buf);
596 free_netvsc_device(net_device);
601 #define RING_AVAIL_PERCENT_HIWATER 20
602 #define RING_AVAIL_PERCENT_LOWATER 10
605 * Get the percentage of available bytes to write in the ring.
606 * The return value is in range from 0 to 100.
608 static inline u32 hv_ringbuf_avail_percent(
609 struct hv_ring_buffer_info *ring_info)
611 u32 avail_read, avail_write;
613 hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
615 return avail_write * 100 / ring_info->ring_datasize;
618 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
621 sync_change_bit(index, net_device->send_section_map);
624 static void netvsc_send_completion(struct netvsc_device *net_device,
625 struct vmbus_channel *incoming_channel,
626 struct hv_device *device,
627 struct vmpacket_descriptor *packet)
629 struct nvsp_message *nvsp_packet;
630 struct hv_netvsc_packet *nvsc_packet;
631 struct net_device *ndev = hv_get_drvdata(device);
632 struct net_device_context *net_device_ctx = netdev_priv(ndev);
636 nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
637 (packet->offset8 << 3));
639 if ((nvsp_packet->hdr.msg_type == NVSP_MSG_TYPE_INIT_COMPLETE) ||
640 (nvsp_packet->hdr.msg_type ==
641 NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) ||
642 (nvsp_packet->hdr.msg_type ==
643 NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE) ||
644 (nvsp_packet->hdr.msg_type ==
645 NVSP_MSG5_TYPE_SUBCHANNEL)) {
646 /* Copy the response back */
647 memcpy(&net_device->channel_init_pkt, nvsp_packet,
648 sizeof(struct nvsp_message));
649 complete(&net_device->channel_init_wait);
650 } else if (nvsp_packet->hdr.msg_type ==
651 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) {
652 int num_outstanding_sends;
654 struct vmbus_channel *channel = device->channel;
657 /* Get the send context */
658 skb = (struct sk_buff *)(unsigned long)packet->trans_id;
660 /* Notify the layer above us */
662 nvsc_packet = (struct hv_netvsc_packet *) skb->cb;
663 send_index = nvsc_packet->send_buf_index;
664 if (send_index != NETVSC_INVALID_INDEX)
665 netvsc_free_send_slot(net_device, send_index);
666 q_idx = nvsc_packet->q_idx;
667 channel = incoming_channel;
668 dev_kfree_skb_any(skb);
671 num_outstanding_sends =
672 atomic_dec_return(&net_device->num_outstanding_sends);
673 queue_sends = atomic_dec_return(&net_device->
676 if (net_device->destroy && num_outstanding_sends == 0)
677 wake_up(&net_device->wait_drain);
679 if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev, q_idx)) &&
680 !net_device_ctx->start_remove &&
681 (hv_ringbuf_avail_percent(&channel->outbound) >
682 RING_AVAIL_PERCENT_HIWATER || queue_sends < 1))
683 netif_tx_wake_queue(netdev_get_tx_queue(
686 netdev_err(ndev, "Unknown send completion packet type- "
687 "%d received!!\n", nvsp_packet->hdr.msg_type);
692 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
695 u32 max_words = net_device->map_words;
696 unsigned long *map_addr = (unsigned long *)net_device->send_section_map;
697 u32 section_cnt = net_device->send_section_cnt;
698 int ret_val = NETVSC_INVALID_INDEX;
702 for (i = 0; i < max_words; i++) {
705 index = ffz(map_addr[i]);
706 prev_val = sync_test_and_set_bit(index, &map_addr[i]);
709 if ((index + (i * BITS_PER_LONG)) >= section_cnt)
711 ret_val = (index + (i * BITS_PER_LONG));
717 static u32 netvsc_copy_to_send_buf(struct netvsc_device *net_device,
718 unsigned int section_index,
720 struct hv_netvsc_packet *packet,
721 struct rndis_message *rndis_msg,
722 struct hv_page_buffer **pb,
725 char *start = net_device->send_buf;
726 char *dest = start + (section_index * net_device->send_section_size)
729 bool is_data_pkt = (skb != NULL) ? true : false;
730 bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
733 u32 remain = packet->total_data_buflen % net_device->pkt_align;
734 u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
735 packet->page_buf_cnt;
738 if (is_data_pkt && xmit_more && remain &&
739 !packet->cp_partial) {
740 padding = net_device->pkt_align - remain;
741 rndis_msg->msg_len += padding;
742 packet->total_data_buflen += padding;
745 for (i = 0; i < page_count; i++) {
746 char *src = phys_to_virt((*pb)[i].pfn << PAGE_SHIFT);
747 u32 offset = (*pb)[i].offset;
748 u32 len = (*pb)[i].len;
750 memcpy(dest, (src + offset), len);
756 memset(dest, 0, padding);
763 static inline int netvsc_send_pkt(
764 struct hv_device *device,
765 struct hv_netvsc_packet *packet,
766 struct netvsc_device *net_device,
767 struct hv_page_buffer **pb,
770 struct nvsp_message nvmsg;
771 u16 q_idx = packet->q_idx;
772 struct vmbus_channel *out_channel = net_device->chn_table[q_idx];
773 struct net_device *ndev = hv_get_drvdata(device);
776 struct hv_page_buffer *pgbuf;
777 u32 ring_avail = hv_ringbuf_avail_percent(&out_channel->outbound);
778 bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
780 nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
783 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 0;
785 /* 1 is RMC_CONTROL; */
786 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 1;
789 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
790 packet->send_buf_index;
791 if (packet->send_buf_index == NETVSC_INVALID_INDEX)
792 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
794 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size =
795 packet->total_data_buflen;
799 if (out_channel->rescind)
803 * It is possible that once we successfully place this packet
804 * on the ringbuffer, we may stop the queue. In that case, we want
805 * to notify the host independent of the xmit_more flag. We don't
806 * need to be precise here; in the worst case we may signal the host
809 if (ring_avail < (RING_AVAIL_PERCENT_LOWATER + 1))
812 if (packet->page_buf_cnt) {
813 pgbuf = packet->cp_partial ? (*pb) +
814 packet->rmsg_pgcnt : (*pb);
815 ret = vmbus_sendpacket_pagebuffer_ctl(out_channel,
817 packet->page_buf_cnt,
819 sizeof(struct nvsp_message),
821 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED,
824 ret = vmbus_sendpacket_ctl(out_channel, &nvmsg,
825 sizeof(struct nvsp_message),
828 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED,
833 atomic_inc(&net_device->num_outstanding_sends);
834 atomic_inc(&net_device->queue_sends[q_idx]);
836 if (ring_avail < RING_AVAIL_PERCENT_LOWATER) {
837 netif_tx_stop_queue(netdev_get_tx_queue(ndev, q_idx));
839 if (atomic_read(&net_device->
840 queue_sends[q_idx]) < 1)
841 netif_tx_wake_queue(netdev_get_tx_queue(
844 } else if (ret == -EAGAIN) {
845 netif_tx_stop_queue(netdev_get_tx_queue(
847 if (atomic_read(&net_device->queue_sends[q_idx]) < 1) {
848 netif_tx_wake_queue(netdev_get_tx_queue(
853 netdev_err(ndev, "Unable to send packet %p ret %d\n",
860 /* Move packet out of multi send data (msd), and clear msd */
861 static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
862 struct sk_buff **msd_skb,
863 struct multi_send_data *msdp)
865 *msd_skb = msdp->skb;
866 *msd_send = msdp->pkt;
872 int netvsc_send(struct hv_device *device,
873 struct hv_netvsc_packet *packet,
874 struct rndis_message *rndis_msg,
875 struct hv_page_buffer **pb,
878 struct netvsc_device *net_device;
879 int ret = 0, m_ret = 0;
880 struct vmbus_channel *out_channel;
881 u16 q_idx = packet->q_idx;
882 u32 pktlen = packet->total_data_buflen, msd_len = 0;
883 unsigned int section_index = NETVSC_INVALID_INDEX;
884 struct multi_send_data *msdp;
885 struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
886 struct sk_buff *msd_skb = NULL;
888 bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
890 net_device = get_outbound_net_device(device);
894 out_channel = net_device->chn_table[q_idx];
896 packet->send_buf_index = NETVSC_INVALID_INDEX;
897 packet->cp_partial = false;
899 /* Send control message directly without accessing msd (Multi-Send
900 * Data) field which may be changed during data packet processing.
907 msdp = &net_device->msd[q_idx];
909 /* batch packets in send buffer if possible */
911 msd_len = msdp->pkt->total_data_buflen;
913 try_batch = (skb != NULL) && msd_len > 0 && msdp->count <
916 if (try_batch && msd_len + pktlen + net_device->pkt_align <
917 net_device->send_section_size) {
918 section_index = msdp->pkt->send_buf_index;
920 } else if (try_batch && msd_len + packet->rmsg_size <
921 net_device->send_section_size) {
922 section_index = msdp->pkt->send_buf_index;
923 packet->cp_partial = true;
925 } else if ((skb != NULL) && pktlen + net_device->pkt_align <
926 net_device->send_section_size) {
927 section_index = netvsc_get_next_send_section(net_device);
928 if (section_index != NETVSC_INVALID_INDEX) {
929 move_pkt_msd(&msd_send, &msd_skb, msdp);
934 if (section_index != NETVSC_INVALID_INDEX) {
935 netvsc_copy_to_send_buf(net_device,
936 section_index, msd_len,
937 packet, rndis_msg, pb, skb);
939 packet->send_buf_index = section_index;
941 if (packet->cp_partial) {
942 packet->page_buf_cnt -= packet->rmsg_pgcnt;
943 packet->total_data_buflen = msd_len + packet->rmsg_size;
945 packet->page_buf_cnt = 0;
946 packet->total_data_buflen += msd_len;
950 dev_kfree_skb_any(msdp->skb);
952 if (xmit_more && !packet->cp_partial) {
963 move_pkt_msd(&msd_send, &msd_skb, msdp);
968 m_ret = netvsc_send_pkt(device, msd_send, net_device,
972 netvsc_free_send_slot(net_device,
973 msd_send->send_buf_index);
974 dev_kfree_skb_any(msd_skb);
980 ret = netvsc_send_pkt(device, cur_send, net_device, pb, skb);
982 if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
983 netvsc_free_send_slot(net_device, section_index);
988 static void netvsc_send_recv_completion(struct hv_device *device,
989 struct vmbus_channel *channel,
990 struct netvsc_device *net_device,
991 u64 transaction_id, u32 status)
993 struct nvsp_message recvcompMessage;
996 struct net_device *ndev = hv_get_drvdata(device);
998 recvcompMessage.hdr.msg_type =
999 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
1001 recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status = status;
1004 /* Send the completion */
1005 ret = vmbus_sendpacket(channel, &recvcompMessage,
1006 sizeof(struct nvsp_message), transaction_id,
1011 } else if (ret == -EAGAIN) {
1012 /* no more room...wait a bit and attempt to retry 3 times */
1014 netdev_err(ndev, "unable to send receive completion pkt"
1015 " (tid %llx)...retrying %d\n", transaction_id, retries);
1019 goto retry_send_cmplt;
1021 netdev_err(ndev, "unable to send receive "
1022 "completion pkt (tid %llx)...give up retrying\n",
1026 netdev_err(ndev, "unable to send receive "
1027 "completion pkt - %llx\n", transaction_id);
1031 static void netvsc_receive(struct netvsc_device *net_device,
1032 struct vmbus_channel *channel,
1033 struct hv_device *device,
1034 struct vmpacket_descriptor *packet)
1036 struct vmtransfer_page_packet_header *vmxferpage_packet;
1037 struct nvsp_message *nvsp_packet;
1038 struct hv_netvsc_packet nv_pkt;
1039 struct hv_netvsc_packet *netvsc_packet = &nv_pkt;
1040 u32 status = NVSP_STAT_SUCCESS;
1043 struct net_device *ndev = hv_get_drvdata(device);
1047 * All inbound packets other than send completion should be xfer page
1050 if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
1051 netdev_err(ndev, "Unknown packet type received - %d\n",
1056 nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
1057 (packet->offset8 << 3));
1059 /* Make sure this is a valid nvsp packet */
1060 if (nvsp_packet->hdr.msg_type !=
1061 NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
1062 netdev_err(ndev, "Unknown nvsp packet type received-"
1063 " %d\n", nvsp_packet->hdr.msg_type);
1067 vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
1069 if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
1070 netdev_err(ndev, "Invalid xfer page set id - "
1071 "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
1072 vmxferpage_packet->xfer_pageset_id);
1076 count = vmxferpage_packet->range_cnt;
1078 /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1079 for (i = 0; i < count; i++) {
1080 /* Initialize the netvsc packet */
1081 data = (void *)((unsigned long)net_device->
1082 recv_buf + vmxferpage_packet->ranges[i].byte_offset);
1083 netvsc_packet->total_data_buflen =
1084 vmxferpage_packet->ranges[i].byte_count;
1086 /* Pass it to the upper layer */
1087 status = rndis_filter_receive(device, netvsc_packet, &data,
1092 netvsc_send_recv_completion(device, channel, net_device,
1093 vmxferpage_packet->d.trans_id, status);
1097 static void netvsc_send_table(struct hv_device *hdev,
1098 struct nvsp_message *nvmsg)
1100 struct netvsc_device *nvscdev;
1101 struct net_device *ndev = hv_get_drvdata(hdev);
1105 nvscdev = get_outbound_net_device(hdev);
1109 count = nvmsg->msg.v5_msg.send_table.count;
1110 if (count != VRSS_SEND_TAB_SIZE) {
1111 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1115 tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
1116 nvmsg->msg.v5_msg.send_table.offset);
1118 for (i = 0; i < count; i++)
1119 nvscdev->send_table[i] = tab[i];
1122 static void netvsc_send_vf(struct netvsc_device *nvdev,
1123 struct nvsp_message *nvmsg)
1125 nvdev->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1126 nvdev->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1129 static inline void netvsc_receive_inband(struct hv_device *hdev,
1130 struct netvsc_device *nvdev,
1131 struct nvsp_message *nvmsg)
1133 switch (nvmsg->hdr.msg_type) {
1134 case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1135 netvsc_send_table(hdev, nvmsg);
1138 case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1139 netvsc_send_vf(nvdev, nvmsg);
1144 void netvsc_channel_cb(void *context)
1147 struct vmbus_channel *channel = (struct vmbus_channel *)context;
1148 struct hv_device *device;
1149 struct netvsc_device *net_device;
1152 struct vmpacket_descriptor *desc;
1153 unsigned char *buffer;
1154 int bufferlen = NETVSC_PACKET_SIZE;
1155 struct net_device *ndev;
1156 struct nvsp_message *nvmsg;
1158 if (channel->primary_channel != NULL)
1159 device = channel->primary_channel->device_obj;
1161 device = channel->device_obj;
1163 net_device = get_inbound_net_device(device);
1166 ndev = hv_get_drvdata(device);
1167 buffer = get_per_channel_state(channel);
1170 ret = vmbus_recvpacket_raw(channel, buffer, bufferlen,
1171 &bytes_recvd, &request_id);
1173 if (bytes_recvd > 0) {
1174 desc = (struct vmpacket_descriptor *)buffer;
1175 nvmsg = (struct nvsp_message *)((unsigned long)
1176 desc + (desc->offset8 << 3));
1177 switch (desc->type) {
1179 netvsc_send_completion(net_device,
1184 case VM_PKT_DATA_USING_XFER_PAGES:
1185 netvsc_receive(net_device, channel,
1189 case VM_PKT_DATA_INBAND:
1190 netvsc_receive_inband(device,
1197 "unhandled packet type %d, "
1198 "tid %llx len %d\n",
1199 desc->type, request_id,
1206 * We are done for this pass.
1211 } else if (ret == -ENOBUFS) {
1212 if (bufferlen > NETVSC_PACKET_SIZE)
1214 /* Handle large packet */
1215 buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
1216 if (buffer == NULL) {
1217 /* Try again next time around */
1219 "unable to allocate buffer of size "
1220 "(%d)!!\n", bytes_recvd);
1224 bufferlen = bytes_recvd;
1228 if (bufferlen > NETVSC_PACKET_SIZE)
1234 * netvsc_device_add - Callback when the device belonging to this
1237 int netvsc_device_add(struct hv_device *device, void *additional_info)
1241 ((struct netvsc_device_info *)additional_info)->ring_size;
1242 struct netvsc_device *net_device;
1243 struct net_device *ndev = hv_get_drvdata(device);
1244 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1246 net_device = alloc_net_device();
1250 net_device->ring_size = ring_size;
1252 /* Initialize the NetVSC channel extension */
1253 init_completion(&net_device->channel_init_wait);
1255 set_per_channel_state(device->channel, net_device->cb_buffer);
1257 /* Open the channel */
1258 ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
1259 ring_size * PAGE_SIZE, NULL, 0,
1260 netvsc_channel_cb, device->channel);
1263 netdev_err(ndev, "unable to open channel: %d\n", ret);
1267 /* Channel is opened */
1268 pr_info("hv_netvsc channel opened successfully\n");
1270 /* If we're reopening the device we may have multiple queues, fill the
1271 * chn_table with the default channel to use it before subchannels are
1274 for (i = 0; i < VRSS_CHANNEL_MAX; i++)
1275 net_device->chn_table[i] = device->channel;
1277 /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1282 net_device_ctx->nvdev = net_device;
1284 /* Connect with the NetVsp */
1285 ret = netvsc_connect_vsp(device);
1288 "unable to connect to NetVSP - %d\n", ret);
1295 /* Now, we can close the channel safely */
1296 vmbus_close(device->channel);
1299 free_netvsc_device(net_device);