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 static struct netvsc_device *alloc_net_device(struct hv_device *device)
39 struct netvsc_device *net_device;
40 struct net_device *ndev = hv_get_drvdata(device);
43 net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
47 net_device->cb_buffer = kzalloc(NETVSC_PACKET_SIZE, GFP_KERNEL);
48 if (!net_device->cb_buffer) {
53 init_waitqueue_head(&net_device->wait_drain);
54 net_device->start_remove = false;
55 net_device->destroy = false;
56 net_device->dev = device;
57 net_device->ndev = ndev;
58 net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
59 net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
61 for (i = 0; i < num_online_cpus(); i++)
62 spin_lock_init(&net_device->msd[i].lock);
64 hv_set_drvdata(device, net_device);
68 static void free_netvsc_device(struct netvsc_device *nvdev)
70 kfree(nvdev->cb_buffer);
74 static struct netvsc_device *get_outbound_net_device(struct hv_device *device)
76 struct netvsc_device *net_device;
78 net_device = hv_get_drvdata(device);
79 if (net_device && net_device->destroy)
85 static struct netvsc_device *get_inbound_net_device(struct hv_device *device)
87 struct netvsc_device *net_device;
89 net_device = hv_get_drvdata(device);
94 if (net_device->destroy &&
95 atomic_read(&net_device->num_outstanding_sends) == 0)
103 static int netvsc_destroy_buf(struct netvsc_device *net_device)
105 struct nvsp_message *revoke_packet;
107 struct net_device *ndev = net_device->ndev;
110 * If we got a section count, it means we received a
111 * SendReceiveBufferComplete msg (ie sent
112 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
113 * to send a revoke msg here
115 if (net_device->recv_section_cnt) {
116 /* Send the revoke receive buffer */
117 revoke_packet = &net_device->revoke_packet;
118 memset(revoke_packet, 0, sizeof(struct nvsp_message));
120 revoke_packet->hdr.msg_type =
121 NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
122 revoke_packet->msg.v1_msg.
123 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
125 ret = vmbus_sendpacket(net_device->dev->channel,
127 sizeof(struct nvsp_message),
128 (unsigned long)revoke_packet,
129 VM_PKT_DATA_INBAND, 0);
131 * If we failed here, we might as well return and
132 * have a leak rather than continue and a bugchk
135 netdev_err(ndev, "unable to send "
136 "revoke receive buffer to netvsp\n");
141 /* Teardown the gpadl on the vsp end */
142 if (net_device->recv_buf_gpadl_handle) {
143 ret = vmbus_teardown_gpadl(net_device->dev->channel,
144 net_device->recv_buf_gpadl_handle);
146 /* If we failed here, we might as well return and have a leak
147 * rather than continue and a bugchk
151 "unable to teardown receive buffer's gpadl\n");
154 net_device->recv_buf_gpadl_handle = 0;
157 if (net_device->recv_buf) {
158 /* Free up the receive buffer */
159 vfree(net_device->recv_buf);
160 net_device->recv_buf = NULL;
163 if (net_device->recv_section) {
164 net_device->recv_section_cnt = 0;
165 kfree(net_device->recv_section);
166 net_device->recv_section = NULL;
169 /* Deal with the send buffer we may have setup.
170 * If we got a send section size, it means we received a
171 * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
172 * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
173 * to send a revoke msg here
175 if (net_device->send_section_size) {
176 /* Send the revoke receive buffer */
177 revoke_packet = &net_device->revoke_packet;
178 memset(revoke_packet, 0, sizeof(struct nvsp_message));
180 revoke_packet->hdr.msg_type =
181 NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
182 revoke_packet->msg.v1_msg.revoke_send_buf.id =
183 NETVSC_SEND_BUFFER_ID;
185 ret = vmbus_sendpacket(net_device->dev->channel,
187 sizeof(struct nvsp_message),
188 (unsigned long)revoke_packet,
189 VM_PKT_DATA_INBAND, 0);
190 /* If we failed here, we might as well return and
191 * have a leak rather than continue and a bugchk
194 netdev_err(ndev, "unable to send "
195 "revoke send buffer to netvsp\n");
199 /* Teardown the gpadl on the vsp end */
200 if (net_device->send_buf_gpadl_handle) {
201 ret = vmbus_teardown_gpadl(net_device->dev->channel,
202 net_device->send_buf_gpadl_handle);
204 /* If we failed here, we might as well return and have a leak
205 * rather than continue and a bugchk
209 "unable to teardown send buffer's gpadl\n");
212 net_device->send_buf_gpadl_handle = 0;
214 if (net_device->send_buf) {
215 /* Free up the send buffer */
216 vfree(net_device->send_buf);
217 net_device->send_buf = NULL;
219 kfree(net_device->send_section_map);
224 static int netvsc_init_buf(struct hv_device *device)
228 struct netvsc_device *net_device;
229 struct nvsp_message *init_packet;
230 struct net_device *ndev;
232 net_device = get_outbound_net_device(device);
235 ndev = net_device->ndev;
237 net_device->recv_buf = vzalloc(net_device->recv_buf_size);
238 if (!net_device->recv_buf) {
239 netdev_err(ndev, "unable to allocate receive "
240 "buffer of size %d\n", net_device->recv_buf_size);
246 * Establish the gpadl handle for this buffer on this
247 * channel. Note: This call uses the vmbus connection rather
248 * than the channel to establish the gpadl handle.
250 ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
251 net_device->recv_buf_size,
252 &net_device->recv_buf_gpadl_handle);
255 "unable to establish receive buffer's gpadl\n");
260 /* Notify the NetVsp of the gpadl handle */
261 init_packet = &net_device->channel_init_pkt;
263 memset(init_packet, 0, sizeof(struct nvsp_message));
265 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
266 init_packet->msg.v1_msg.send_recv_buf.
267 gpadl_handle = net_device->recv_buf_gpadl_handle;
268 init_packet->msg.v1_msg.
269 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
271 /* Send the gpadl notification request */
272 ret = vmbus_sendpacket(device->channel, init_packet,
273 sizeof(struct nvsp_message),
274 (unsigned long)init_packet,
276 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
279 "unable to send receive buffer's gpadl to netvsp\n");
283 t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
287 /* Check the response */
288 if (init_packet->msg.v1_msg.
289 send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
290 netdev_err(ndev, "Unable to complete receive buffer "
291 "initialization with NetVsp - status %d\n",
292 init_packet->msg.v1_msg.
293 send_recv_buf_complete.status);
298 /* Parse the response */
300 net_device->recv_section_cnt = init_packet->msg.
301 v1_msg.send_recv_buf_complete.num_sections;
303 net_device->recv_section = kmemdup(
304 init_packet->msg.v1_msg.send_recv_buf_complete.sections,
305 net_device->recv_section_cnt *
306 sizeof(struct nvsp_1_receive_buffer_section),
308 if (net_device->recv_section == NULL) {
314 * For 1st release, there should only be 1 section that represents the
315 * entire receive buffer
317 if (net_device->recv_section_cnt != 1 ||
318 net_device->recv_section->offset != 0) {
323 /* Now setup the send buffer.
325 net_device->send_buf = vzalloc(net_device->send_buf_size);
326 if (!net_device->send_buf) {
327 netdev_err(ndev, "unable to allocate send "
328 "buffer of size %d\n", net_device->send_buf_size);
333 /* Establish the gpadl handle for this buffer on this
334 * channel. Note: This call uses the vmbus connection rather
335 * than the channel to establish the gpadl handle.
337 ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
338 net_device->send_buf_size,
339 &net_device->send_buf_gpadl_handle);
342 "unable to establish send buffer's gpadl\n");
346 /* Notify the NetVsp of the gpadl handle */
347 init_packet = &net_device->channel_init_pkt;
348 memset(init_packet, 0, sizeof(struct nvsp_message));
349 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
350 init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
351 net_device->send_buf_gpadl_handle;
352 init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
354 /* Send the gpadl notification request */
355 ret = vmbus_sendpacket(device->channel, init_packet,
356 sizeof(struct nvsp_message),
357 (unsigned long)init_packet,
359 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
362 "unable to send send buffer's gpadl to netvsp\n");
366 t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
369 /* Check the response */
370 if (init_packet->msg.v1_msg.
371 send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
372 netdev_err(ndev, "Unable to complete send buffer "
373 "initialization with NetVsp - status %d\n",
374 init_packet->msg.v1_msg.
375 send_send_buf_complete.status);
380 /* Parse the response */
381 net_device->send_section_size = init_packet->msg.
382 v1_msg.send_send_buf_complete.section_size;
384 /* Section count is simply the size divided by the section size.
386 net_device->send_section_cnt =
387 net_device->send_buf_size/net_device->send_section_size;
389 dev_info(&device->device, "Send section size: %d, Section count:%d\n",
390 net_device->send_section_size, net_device->send_section_cnt);
392 /* Setup state for managing the send buffer. */
393 net_device->map_words = DIV_ROUND_UP(net_device->send_section_cnt,
396 net_device->send_section_map =
397 kzalloc(net_device->map_words * sizeof(ulong), GFP_KERNEL);
398 if (net_device->send_section_map == NULL) {
406 netvsc_destroy_buf(net_device);
413 /* Negotiate NVSP protocol version */
414 static int negotiate_nvsp_ver(struct hv_device *device,
415 struct netvsc_device *net_device,
416 struct nvsp_message *init_packet,
422 memset(init_packet, 0, sizeof(struct nvsp_message));
423 init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
424 init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
425 init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
427 /* Send the init request */
428 ret = vmbus_sendpacket(device->channel, init_packet,
429 sizeof(struct nvsp_message),
430 (unsigned long)init_packet,
432 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
437 t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
442 if (init_packet->msg.init_msg.init_complete.status !=
446 if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
449 /* NVSPv2 only: Send NDIS config */
450 memset(init_packet, 0, sizeof(struct nvsp_message));
451 init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
452 init_packet->msg.v2_msg.send_ndis_config.mtu = net_device->ndev->mtu +
454 init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
456 ret = vmbus_sendpacket(device->channel, init_packet,
457 sizeof(struct nvsp_message),
458 (unsigned long)init_packet,
459 VM_PKT_DATA_INBAND, 0);
464 static int netvsc_connect_vsp(struct hv_device *device)
467 struct netvsc_device *net_device;
468 struct nvsp_message *init_packet;
470 struct net_device *ndev;
471 u32 ver_list[] = { NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
472 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5 };
473 int i, num_ver = 4; /* number of different NVSP versions */
475 net_device = get_outbound_net_device(device);
478 ndev = net_device->ndev;
480 init_packet = &net_device->channel_init_pkt;
482 /* Negotiate the latest NVSP protocol supported */
483 for (i = num_ver - 1; i >= 0; i--)
484 if (negotiate_nvsp_ver(device, net_device, init_packet,
486 net_device->nvsp_version = ver_list[i];
495 pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
497 /* Send the ndis version */
498 memset(init_packet, 0, sizeof(struct nvsp_message));
500 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
501 ndis_version = 0x00060001;
503 ndis_version = 0x0006001e;
505 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
506 init_packet->msg.v1_msg.
507 send_ndis_ver.ndis_major_ver =
508 (ndis_version & 0xFFFF0000) >> 16;
509 init_packet->msg.v1_msg.
510 send_ndis_ver.ndis_minor_ver =
511 ndis_version & 0xFFFF;
513 /* Send the init request */
514 ret = vmbus_sendpacket(device->channel, init_packet,
515 sizeof(struct nvsp_message),
516 (unsigned long)init_packet,
517 VM_PKT_DATA_INBAND, 0);
521 /* Post the big receive buffer to NetVSP */
522 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
523 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
525 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
526 net_device->send_buf_size = NETVSC_SEND_BUFFER_SIZE;
528 ret = netvsc_init_buf(device);
534 static void netvsc_disconnect_vsp(struct netvsc_device *net_device)
536 netvsc_destroy_buf(net_device);
540 * netvsc_device_remove - Callback when the root bus device is removed
542 int netvsc_device_remove(struct hv_device *device)
544 struct netvsc_device *net_device;
547 net_device = hv_get_drvdata(device);
549 netvsc_disconnect_vsp(net_device);
552 * Since we have already drained, we don't need to busy wait
553 * as was done in final_release_stor_device()
554 * Note that we cannot set the ext pointer to NULL until
555 * we have drained - to drain the outgoing packets, we need to
556 * allow incoming packets.
559 spin_lock_irqsave(&device->channel->inbound_lock, flags);
560 hv_set_drvdata(device, NULL);
561 spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
564 * At this point, no one should be accessing net_device
567 dev_notice(&device->device, "net device safe to remove\n");
569 /* Now, we can close the channel safely */
570 vmbus_close(device->channel);
572 /* Release all resources */
573 vfree(net_device->sub_cb_buf);
574 free_netvsc_device(net_device);
579 #define RING_AVAIL_PERCENT_HIWATER 20
580 #define RING_AVAIL_PERCENT_LOWATER 10
583 * Get the percentage of available bytes to write in the ring.
584 * The return value is in range from 0 to 100.
586 static inline u32 hv_ringbuf_avail_percent(
587 struct hv_ring_buffer_info *ring_info)
589 u32 avail_read, avail_write;
591 hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
593 return avail_write * 100 / ring_info->ring_datasize;
596 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
599 sync_change_bit(index, net_device->send_section_map);
602 static void netvsc_send_completion(struct netvsc_device *net_device,
603 struct hv_device *device,
604 struct vmpacket_descriptor *packet)
606 struct nvsp_message *nvsp_packet;
607 struct hv_netvsc_packet *nvsc_packet;
608 struct net_device *ndev;
611 ndev = net_device->ndev;
613 nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
614 (packet->offset8 << 3));
616 if ((nvsp_packet->hdr.msg_type == NVSP_MSG_TYPE_INIT_COMPLETE) ||
617 (nvsp_packet->hdr.msg_type ==
618 NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) ||
619 (nvsp_packet->hdr.msg_type ==
620 NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE) ||
621 (nvsp_packet->hdr.msg_type ==
622 NVSP_MSG5_TYPE_SUBCHANNEL)) {
623 /* Copy the response back */
624 memcpy(&net_device->channel_init_pkt, nvsp_packet,
625 sizeof(struct nvsp_message));
626 complete(&net_device->channel_init_wait);
627 } else if (nvsp_packet->hdr.msg_type ==
628 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) {
629 int num_outstanding_sends;
631 struct vmbus_channel *channel = device->channel;
634 /* Get the send context */
635 nvsc_packet = (struct hv_netvsc_packet *)(unsigned long)
638 /* Notify the layer above us */
640 send_index = nvsc_packet->send_buf_index;
641 if (send_index != NETVSC_INVALID_INDEX)
642 netvsc_free_send_slot(net_device, send_index);
643 q_idx = nvsc_packet->q_idx;
644 channel = nvsc_packet->channel;
645 nvsc_packet->send_completion(nvsc_packet->
646 send_completion_ctx);
649 num_outstanding_sends =
650 atomic_dec_return(&net_device->num_outstanding_sends);
651 queue_sends = atomic_dec_return(&net_device->
654 if (net_device->destroy && num_outstanding_sends == 0)
655 wake_up(&net_device->wait_drain);
657 if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev, q_idx)) &&
658 !net_device->start_remove &&
659 (hv_ringbuf_avail_percent(&channel->outbound) >
660 RING_AVAIL_PERCENT_HIWATER || queue_sends < 1))
661 netif_tx_wake_queue(netdev_get_tx_queue(
664 netdev_err(ndev, "Unknown send completion packet type- "
665 "%d received!!\n", nvsp_packet->hdr.msg_type);
670 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
673 u32 max_words = net_device->map_words;
674 unsigned long *map_addr = (unsigned long *)net_device->send_section_map;
675 u32 section_cnt = net_device->send_section_cnt;
676 int ret_val = NETVSC_INVALID_INDEX;
680 for (i = 0; i < max_words; i++) {
683 index = ffz(map_addr[i]);
684 prev_val = sync_test_and_set_bit(index, &map_addr[i]);
687 if ((index + (i * BITS_PER_LONG)) >= section_cnt)
689 ret_val = (index + (i * BITS_PER_LONG));
695 static u32 netvsc_copy_to_send_buf(struct netvsc_device *net_device,
696 unsigned int section_index,
698 struct hv_netvsc_packet *packet)
700 char *start = net_device->send_buf;
701 char *dest = start + (section_index * net_device->send_section_size)
706 u32 remain = packet->total_data_buflen % net_device->pkt_align;
707 u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
708 packet->page_buf_cnt;
711 if (packet->is_data_pkt && packet->xmit_more && remain &&
712 !packet->cp_partial) {
713 padding = net_device->pkt_align - remain;
714 packet->rndis_msg->msg_len += padding;
715 packet->total_data_buflen += padding;
718 for (i = 0; i < page_count; i++) {
719 char *src = phys_to_virt(packet->page_buf[i].pfn << PAGE_SHIFT);
720 u32 offset = packet->page_buf[i].offset;
721 u32 len = packet->page_buf[i].len;
723 memcpy(dest, (src + offset), len);
729 memset(dest, 0, padding);
736 static inline int netvsc_send_pkt(
737 struct hv_netvsc_packet *packet,
738 struct netvsc_device *net_device)
740 struct nvsp_message nvmsg;
741 struct vmbus_channel *out_channel = packet->channel;
742 u16 q_idx = packet->q_idx;
743 struct net_device *ndev = net_device->ndev;
746 struct hv_page_buffer *pgbuf;
748 nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
749 if (packet->is_data_pkt) {
751 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 0;
753 /* 1 is RMC_CONTROL; */
754 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 1;
757 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
758 packet->send_buf_index;
759 if (packet->send_buf_index == NETVSC_INVALID_INDEX)
760 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
762 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size =
763 packet->total_data_buflen;
765 if (packet->send_completion)
766 req_id = (ulong)packet;
770 if (out_channel->rescind)
773 if (packet->page_buf_cnt) {
774 pgbuf = packet->cp_partial ? packet->page_buf +
775 packet->rmsg_pgcnt : packet->page_buf;
776 ret = vmbus_sendpacket_pagebuffer(out_channel,
778 packet->page_buf_cnt,
780 sizeof(struct nvsp_message),
783 ret = vmbus_sendpacket(
785 sizeof(struct nvsp_message),
788 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
792 atomic_inc(&net_device->num_outstanding_sends);
793 atomic_inc(&net_device->queue_sends[q_idx]);
795 if (hv_ringbuf_avail_percent(&out_channel->outbound) <
796 RING_AVAIL_PERCENT_LOWATER) {
797 netif_tx_stop_queue(netdev_get_tx_queue(
800 if (atomic_read(&net_device->
801 queue_sends[q_idx]) < 1)
802 netif_tx_wake_queue(netdev_get_tx_queue(
805 } else if (ret == -EAGAIN) {
806 netif_tx_stop_queue(netdev_get_tx_queue(
808 if (atomic_read(&net_device->queue_sends[q_idx]) < 1) {
809 netif_tx_wake_queue(netdev_get_tx_queue(
814 netdev_err(ndev, "Unable to send packet %p ret %d\n",
821 int netvsc_send(struct hv_device *device,
822 struct hv_netvsc_packet *packet)
824 struct netvsc_device *net_device;
825 int ret = 0, m_ret = 0;
826 struct vmbus_channel *out_channel;
827 u16 q_idx = packet->q_idx;
828 u32 pktlen = packet->total_data_buflen, msd_len = 0;
829 unsigned int section_index = NETVSC_INVALID_INDEX;
831 struct multi_send_data *msdp;
832 struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
835 net_device = get_outbound_net_device(device);
839 out_channel = net_device->chn_table[q_idx];
841 out_channel = device->channel;
845 packet->channel = out_channel;
846 packet->send_buf_index = NETVSC_INVALID_INDEX;
847 packet->cp_partial = false;
849 msdp = &net_device->msd[q_idx];
851 /* batch packets in send buffer if possible */
852 spin_lock_irqsave(&msdp->lock, flag);
854 msd_len = msdp->pkt->total_data_buflen;
856 try_batch = packet->is_data_pkt && msd_len > 0 && msdp->count <
859 if (try_batch && msd_len + pktlen + net_device->pkt_align <
860 net_device->send_section_size) {
861 section_index = msdp->pkt->send_buf_index;
863 } else if (try_batch && msd_len + packet->rmsg_size <
864 net_device->send_section_size) {
865 section_index = msdp->pkt->send_buf_index;
866 packet->cp_partial = true;
868 } else if (packet->is_data_pkt && pktlen + net_device->pkt_align <
869 net_device->send_section_size) {
870 section_index = netvsc_get_next_send_section(net_device);
871 if (section_index != NETVSC_INVALID_INDEX) {
872 msd_send = msdp->pkt;
879 if (section_index != NETVSC_INVALID_INDEX) {
880 netvsc_copy_to_send_buf(net_device,
881 section_index, msd_len,
884 packet->send_buf_index = section_index;
886 if (packet->cp_partial) {
887 packet->page_buf_cnt -= packet->rmsg_pgcnt;
888 packet->total_data_buflen = msd_len + packet->rmsg_size;
890 packet->page_buf_cnt = 0;
891 packet->total_data_buflen += msd_len;
895 netvsc_xmit_completion(msdp->pkt);
897 if (packet->xmit_more && !packet->cp_partial) {
906 msd_send = msdp->pkt;
912 spin_unlock_irqrestore(&msdp->lock, flag);
915 m_ret = netvsc_send_pkt(msd_send, net_device);
918 netvsc_free_send_slot(net_device,
919 msd_send->send_buf_index);
920 netvsc_xmit_completion(msd_send);
925 ret = netvsc_send_pkt(cur_send, net_device);
927 if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
928 netvsc_free_send_slot(net_device, section_index);
933 static void netvsc_send_recv_completion(struct hv_device *device,
934 struct vmbus_channel *channel,
935 struct netvsc_device *net_device,
936 u64 transaction_id, u32 status)
938 struct nvsp_message recvcompMessage;
941 struct net_device *ndev;
943 ndev = net_device->ndev;
945 recvcompMessage.hdr.msg_type =
946 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
948 recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status = status;
951 /* Send the completion */
952 ret = vmbus_sendpacket(channel, &recvcompMessage,
953 sizeof(struct nvsp_message), transaction_id,
958 } else if (ret == -EAGAIN) {
959 /* no more room...wait a bit and attempt to retry 3 times */
961 netdev_err(ndev, "unable to send receive completion pkt"
962 " (tid %llx)...retrying %d\n", transaction_id, retries);
966 goto retry_send_cmplt;
968 netdev_err(ndev, "unable to send receive "
969 "completion pkt (tid %llx)...give up retrying\n",
973 netdev_err(ndev, "unable to send receive "
974 "completion pkt - %llx\n", transaction_id);
978 static void netvsc_receive(struct netvsc_device *net_device,
979 struct vmbus_channel *channel,
980 struct hv_device *device,
981 struct vmpacket_descriptor *packet)
983 struct vmtransfer_page_packet_header *vmxferpage_packet;
984 struct nvsp_message *nvsp_packet;
985 struct hv_netvsc_packet nv_pkt;
986 struct hv_netvsc_packet *netvsc_packet = &nv_pkt;
987 u32 status = NVSP_STAT_SUCCESS;
990 struct net_device *ndev;
992 ndev = net_device->ndev;
995 * All inbound packets other than send completion should be xfer page
998 if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
999 netdev_err(ndev, "Unknown packet type received - %d\n",
1004 nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
1005 (packet->offset8 << 3));
1007 /* Make sure this is a valid nvsp packet */
1008 if (nvsp_packet->hdr.msg_type !=
1009 NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
1010 netdev_err(ndev, "Unknown nvsp packet type received-"
1011 " %d\n", nvsp_packet->hdr.msg_type);
1015 vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
1017 if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
1018 netdev_err(ndev, "Invalid xfer page set id - "
1019 "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
1020 vmxferpage_packet->xfer_pageset_id);
1024 count = vmxferpage_packet->range_cnt;
1025 netvsc_packet->channel = channel;
1027 /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1028 for (i = 0; i < count; i++) {
1029 /* Initialize the netvsc packet */
1030 netvsc_packet->status = NVSP_STAT_SUCCESS;
1031 netvsc_packet->data = (void *)((unsigned long)net_device->
1032 recv_buf + vmxferpage_packet->ranges[i].byte_offset);
1033 netvsc_packet->total_data_buflen =
1034 vmxferpage_packet->ranges[i].byte_count;
1036 /* Pass it to the upper layer */
1037 rndis_filter_receive(device, netvsc_packet);
1039 if (netvsc_packet->status != NVSP_STAT_SUCCESS)
1040 status = NVSP_STAT_FAIL;
1043 netvsc_send_recv_completion(device, channel, net_device,
1044 vmxferpage_packet->d.trans_id, status);
1048 static void netvsc_send_table(struct hv_device *hdev,
1049 struct vmpacket_descriptor *vmpkt)
1051 struct netvsc_device *nvscdev;
1052 struct net_device *ndev;
1053 struct nvsp_message *nvmsg;
1057 nvscdev = get_outbound_net_device(hdev);
1060 ndev = nvscdev->ndev;
1062 nvmsg = (struct nvsp_message *)((unsigned long)vmpkt +
1063 (vmpkt->offset8 << 3));
1065 if (nvmsg->hdr.msg_type != NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE)
1068 count = nvmsg->msg.v5_msg.send_table.count;
1069 if (count != VRSS_SEND_TAB_SIZE) {
1070 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1074 tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
1075 nvmsg->msg.v5_msg.send_table.offset);
1077 for (i = 0; i < count; i++)
1078 nvscdev->send_table[i] = tab[i];
1081 void netvsc_channel_cb(void *context)
1084 struct vmbus_channel *channel = (struct vmbus_channel *)context;
1085 struct hv_device *device;
1086 struct netvsc_device *net_device;
1089 struct vmpacket_descriptor *desc;
1090 unsigned char *buffer;
1091 int bufferlen = NETVSC_PACKET_SIZE;
1092 struct net_device *ndev;
1094 if (channel->primary_channel != NULL)
1095 device = channel->primary_channel->device_obj;
1097 device = channel->device_obj;
1099 net_device = get_inbound_net_device(device);
1102 ndev = net_device->ndev;
1103 buffer = get_per_channel_state(channel);
1106 ret = vmbus_recvpacket_raw(channel, buffer, bufferlen,
1107 &bytes_recvd, &request_id);
1109 if (bytes_recvd > 0) {
1110 desc = (struct vmpacket_descriptor *)buffer;
1111 switch (desc->type) {
1113 netvsc_send_completion(net_device,
1117 case VM_PKT_DATA_USING_XFER_PAGES:
1118 netvsc_receive(net_device, channel,
1122 case VM_PKT_DATA_INBAND:
1123 netvsc_send_table(device, desc);
1128 "unhandled packet type %d, "
1129 "tid %llx len %d\n",
1130 desc->type, request_id,
1137 * We are done for this pass.
1142 } else if (ret == -ENOBUFS) {
1143 if (bufferlen > NETVSC_PACKET_SIZE)
1145 /* Handle large packet */
1146 buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
1147 if (buffer == NULL) {
1148 /* Try again next time around */
1150 "unable to allocate buffer of size "
1151 "(%d)!!\n", bytes_recvd);
1155 bufferlen = bytes_recvd;
1159 if (bufferlen > NETVSC_PACKET_SIZE)
1165 * netvsc_device_add - Callback when the device belonging to this
1168 int netvsc_device_add(struct hv_device *device, void *additional_info)
1172 ((struct netvsc_device_info *)additional_info)->ring_size;
1173 struct netvsc_device *net_device;
1174 struct net_device *ndev;
1176 net_device = alloc_net_device(device);
1180 net_device->ring_size = ring_size;
1183 * Coming into this function, struct net_device * is
1184 * registered as the driver private data.
1185 * In alloc_net_device(), we register struct netvsc_device *
1186 * as the driver private data and stash away struct net_device *
1187 * in struct netvsc_device *.
1189 ndev = net_device->ndev;
1191 /* Add netvsc_device context to netvsc_device */
1192 net_device->nd_ctx = netdev_priv(ndev);
1194 /* Initialize the NetVSC channel extension */
1195 init_completion(&net_device->channel_init_wait);
1197 set_per_channel_state(device->channel, net_device->cb_buffer);
1199 /* Open the channel */
1200 ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
1201 ring_size * PAGE_SIZE, NULL, 0,
1202 netvsc_channel_cb, device->channel);
1205 netdev_err(ndev, "unable to open channel: %d\n", ret);
1209 /* Channel is opened */
1210 pr_info("hv_netvsc channel opened successfully\n");
1212 net_device->chn_table[0] = device->channel;
1214 /* Connect with the NetVsp */
1215 ret = netvsc_connect_vsp(device);
1218 "unable to connect to NetVSP - %d\n", ret);
1225 /* Now, we can close the channel safely */
1226 vmbus_close(device->channel);
1229 free_netvsc_device(net_device);