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);
62 static struct netvsc_device *alloc_net_device(void)
64 struct netvsc_device *net_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 net_device->mrc[0].buf = vzalloc(NETVSC_RECVSLOT_MAX *
77 sizeof(struct recv_comp_data));
79 init_waitqueue_head(&net_device->wait_drain);
80 net_device->destroy = false;
81 atomic_set(&net_device->open_cnt, 0);
82 net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
83 net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
84 init_completion(&net_device->channel_init_wait);
89 static void free_netvsc_device(struct netvsc_device *nvdev)
93 for (i = 0; i < VRSS_CHANNEL_MAX; i++)
94 vfree(nvdev->mrc[i].buf);
96 kfree(nvdev->cb_buffer);
100 static struct netvsc_device *get_outbound_net_device(struct hv_device *device)
102 struct netvsc_device *net_device = hv_device_to_netvsc_device(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 = hv_device_to_netvsc_device(device);
117 if (net_device->destroy &&
118 atomic_read(&net_device->num_outstanding_sends) == 0 &&
119 atomic_read(&net_device->num_outstanding_recvs) == 0)
126 static void netvsc_destroy_buf(struct hv_device *device)
128 struct nvsp_message *revoke_packet;
129 struct net_device *ndev = hv_get_drvdata(device);
130 struct netvsc_device *net_device = net_device_to_netvsc_device(ndev);
134 * If we got a section count, it means we received a
135 * SendReceiveBufferComplete msg (ie sent
136 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
137 * to send a revoke msg here
139 if (net_device->recv_section_cnt) {
140 /* Send the revoke receive buffer */
141 revoke_packet = &net_device->revoke_packet;
142 memset(revoke_packet, 0, sizeof(struct nvsp_message));
144 revoke_packet->hdr.msg_type =
145 NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
146 revoke_packet->msg.v1_msg.
147 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
149 ret = vmbus_sendpacket(device->channel,
151 sizeof(struct nvsp_message),
152 (unsigned long)revoke_packet,
153 VM_PKT_DATA_INBAND, 0);
155 * If we failed here, we might as well return and
156 * have a leak rather than continue and a bugchk
159 netdev_err(ndev, "unable to send "
160 "revoke receive buffer to netvsp\n");
165 /* Teardown the gpadl on the vsp end */
166 if (net_device->recv_buf_gpadl_handle) {
167 ret = vmbus_teardown_gpadl(device->channel,
168 net_device->recv_buf_gpadl_handle);
170 /* If we failed here, we might as well return and have a leak
171 * rather than continue and a bugchk
175 "unable to teardown receive buffer's gpadl\n");
178 net_device->recv_buf_gpadl_handle = 0;
181 if (net_device->recv_buf) {
182 /* Free up the receive buffer */
183 vfree(net_device->recv_buf);
184 net_device->recv_buf = NULL;
187 if (net_device->recv_section) {
188 net_device->recv_section_cnt = 0;
189 kfree(net_device->recv_section);
190 net_device->recv_section = NULL;
193 /* Deal with the send buffer we may have setup.
194 * If we got a send section size, it means we received a
195 * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
196 * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
197 * to send a revoke msg here
199 if (net_device->send_section_size) {
200 /* Send the revoke receive buffer */
201 revoke_packet = &net_device->revoke_packet;
202 memset(revoke_packet, 0, sizeof(struct nvsp_message));
204 revoke_packet->hdr.msg_type =
205 NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
206 revoke_packet->msg.v1_msg.revoke_send_buf.id =
207 NETVSC_SEND_BUFFER_ID;
209 ret = vmbus_sendpacket(device->channel,
211 sizeof(struct nvsp_message),
212 (unsigned long)revoke_packet,
213 VM_PKT_DATA_INBAND, 0);
214 /* If we failed here, we might as well return and
215 * have a leak rather than continue and a bugchk
218 netdev_err(ndev, "unable to send "
219 "revoke send buffer to netvsp\n");
223 /* Teardown the gpadl on the vsp end */
224 if (net_device->send_buf_gpadl_handle) {
225 ret = vmbus_teardown_gpadl(device->channel,
226 net_device->send_buf_gpadl_handle);
228 /* If we failed here, we might as well return and have a leak
229 * rather than continue and a bugchk
233 "unable to teardown send buffer's gpadl\n");
236 net_device->send_buf_gpadl_handle = 0;
238 if (net_device->send_buf) {
239 /* Free up the send buffer */
240 vfree(net_device->send_buf);
241 net_device->send_buf = NULL;
243 kfree(net_device->send_section_map);
246 static int netvsc_init_buf(struct hv_device *device)
249 struct netvsc_device *net_device;
250 struct nvsp_message *init_packet;
251 struct net_device *ndev;
254 net_device = get_outbound_net_device(device);
257 ndev = hv_get_drvdata(device);
259 node = cpu_to_node(device->channel->target_cpu);
260 net_device->recv_buf = vzalloc_node(net_device->recv_buf_size, node);
261 if (!net_device->recv_buf)
262 net_device->recv_buf = vzalloc(net_device->recv_buf_size);
264 if (!net_device->recv_buf) {
265 netdev_err(ndev, "unable to allocate receive "
266 "buffer of size %d\n", net_device->recv_buf_size);
272 * Establish the gpadl handle for this buffer on this
273 * channel. Note: This call uses the vmbus connection rather
274 * than the channel to establish the gpadl handle.
276 ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
277 net_device->recv_buf_size,
278 &net_device->recv_buf_gpadl_handle);
281 "unable to establish receive buffer's gpadl\n");
285 /* Notify the NetVsp of the gpadl handle */
286 init_packet = &net_device->channel_init_pkt;
288 memset(init_packet, 0, sizeof(struct nvsp_message));
290 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
291 init_packet->msg.v1_msg.send_recv_buf.
292 gpadl_handle = net_device->recv_buf_gpadl_handle;
293 init_packet->msg.v1_msg.
294 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
296 /* Send the gpadl notification request */
297 ret = vmbus_sendpacket(device->channel, init_packet,
298 sizeof(struct nvsp_message),
299 (unsigned long)init_packet,
301 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
304 "unable to send receive buffer's gpadl to netvsp\n");
308 wait_for_completion(&net_device->channel_init_wait);
310 /* Check the response */
311 if (init_packet->msg.v1_msg.
312 send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
313 netdev_err(ndev, "Unable to complete receive buffer "
314 "initialization with NetVsp - status %d\n",
315 init_packet->msg.v1_msg.
316 send_recv_buf_complete.status);
321 /* Parse the response */
323 net_device->recv_section_cnt = init_packet->msg.
324 v1_msg.send_recv_buf_complete.num_sections;
326 net_device->recv_section = kmemdup(
327 init_packet->msg.v1_msg.send_recv_buf_complete.sections,
328 net_device->recv_section_cnt *
329 sizeof(struct nvsp_1_receive_buffer_section),
331 if (net_device->recv_section == NULL) {
337 * For 1st release, there should only be 1 section that represents the
338 * entire receive buffer
340 if (net_device->recv_section_cnt != 1 ||
341 net_device->recv_section->offset != 0) {
346 /* Now setup the send buffer.
348 net_device->send_buf = vzalloc_node(net_device->send_buf_size, node);
349 if (!net_device->send_buf)
350 net_device->send_buf = vzalloc(net_device->send_buf_size);
351 if (!net_device->send_buf) {
352 netdev_err(ndev, "unable to allocate send "
353 "buffer of size %d\n", net_device->send_buf_size);
358 /* Establish the gpadl handle for this buffer on this
359 * channel. Note: This call uses the vmbus connection rather
360 * than the channel to establish the gpadl handle.
362 ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
363 net_device->send_buf_size,
364 &net_device->send_buf_gpadl_handle);
367 "unable to establish send buffer's gpadl\n");
371 /* Notify the NetVsp of the gpadl handle */
372 init_packet = &net_device->channel_init_pkt;
373 memset(init_packet, 0, sizeof(struct nvsp_message));
374 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
375 init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
376 net_device->send_buf_gpadl_handle;
377 init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
379 /* Send the gpadl notification request */
380 ret = vmbus_sendpacket(device->channel, init_packet,
381 sizeof(struct nvsp_message),
382 (unsigned long)init_packet,
384 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
387 "unable to send send buffer's gpadl to netvsp\n");
391 wait_for_completion(&net_device->channel_init_wait);
393 /* Check the response */
394 if (init_packet->msg.v1_msg.
395 send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
396 netdev_err(ndev, "Unable to complete send buffer "
397 "initialization with NetVsp - status %d\n",
398 init_packet->msg.v1_msg.
399 send_send_buf_complete.status);
404 /* Parse the response */
405 net_device->send_section_size = init_packet->msg.
406 v1_msg.send_send_buf_complete.section_size;
408 /* Section count is simply the size divided by the section size.
410 net_device->send_section_cnt =
411 net_device->send_buf_size / net_device->send_section_size;
413 dev_info(&device->device, "Send section size: %d, Section count:%d\n",
414 net_device->send_section_size, net_device->send_section_cnt);
416 /* Setup state for managing the send buffer. */
417 net_device->map_words = DIV_ROUND_UP(net_device->send_section_cnt,
420 net_device->send_section_map = kcalloc(net_device->map_words,
421 sizeof(ulong), GFP_KERNEL);
422 if (net_device->send_section_map == NULL) {
430 netvsc_destroy_buf(device);
436 /* Negotiate NVSP protocol version */
437 static int negotiate_nvsp_ver(struct hv_device *device,
438 struct netvsc_device *net_device,
439 struct nvsp_message *init_packet,
442 struct net_device *ndev = hv_get_drvdata(device);
445 memset(init_packet, 0, sizeof(struct nvsp_message));
446 init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
447 init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
448 init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
450 /* Send the init request */
451 ret = vmbus_sendpacket(device->channel, init_packet,
452 sizeof(struct nvsp_message),
453 (unsigned long)init_packet,
455 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
460 wait_for_completion(&net_device->channel_init_wait);
462 if (init_packet->msg.init_msg.init_complete.status !=
466 if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
469 /* NVSPv2 or later: Send NDIS config */
470 memset(init_packet, 0, sizeof(struct nvsp_message));
471 init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
472 init_packet->msg.v2_msg.send_ndis_config.mtu = ndev->mtu + ETH_HLEN;
473 init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
475 if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5) {
476 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
478 /* Teaming bit is needed to receive link speed updates */
479 init_packet->msg.v2_msg.send_ndis_config.capability.teaming = 1;
482 ret = vmbus_sendpacket(device->channel, init_packet,
483 sizeof(struct nvsp_message),
484 (unsigned long)init_packet,
485 VM_PKT_DATA_INBAND, 0);
490 static int netvsc_connect_vsp(struct hv_device *device)
493 struct netvsc_device *net_device;
494 struct nvsp_message *init_packet;
496 const u32 ver_list[] = {
497 NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
498 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5 };
501 net_device = get_outbound_net_device(device);
505 init_packet = &net_device->channel_init_pkt;
507 /* Negotiate the latest NVSP protocol supported */
508 for (i = ARRAY_SIZE(ver_list) - 1; i >= 0; i--)
509 if (negotiate_nvsp_ver(device, net_device, init_packet,
511 net_device->nvsp_version = ver_list[i];
520 pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
522 /* Send the ndis version */
523 memset(init_packet, 0, sizeof(struct nvsp_message));
525 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
526 ndis_version = 0x00060001;
528 ndis_version = 0x0006001e;
530 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
531 init_packet->msg.v1_msg.
532 send_ndis_ver.ndis_major_ver =
533 (ndis_version & 0xFFFF0000) >> 16;
534 init_packet->msg.v1_msg.
535 send_ndis_ver.ndis_minor_ver =
536 ndis_version & 0xFFFF;
538 /* Send the init request */
539 ret = vmbus_sendpacket(device->channel, init_packet,
540 sizeof(struct nvsp_message),
541 (unsigned long)init_packet,
542 VM_PKT_DATA_INBAND, 0);
546 /* Post the big receive buffer to NetVSP */
547 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
548 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
550 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
551 net_device->send_buf_size = NETVSC_SEND_BUFFER_SIZE;
553 ret = netvsc_init_buf(device);
559 static void netvsc_disconnect_vsp(struct hv_device *device)
561 netvsc_destroy_buf(device);
565 * netvsc_device_remove - Callback when the root bus device is removed
567 void netvsc_device_remove(struct hv_device *device)
569 struct net_device *ndev = hv_get_drvdata(device);
570 struct net_device_context *net_device_ctx = netdev_priv(ndev);
571 struct netvsc_device *net_device = net_device_ctx->nvdev;
573 netvsc_disconnect_vsp(device);
575 net_device_ctx->nvdev = NULL;
578 * At this point, no one should be accessing net_device
581 dev_notice(&device->device, "net device safe to remove\n");
583 /* Now, we can close the channel safely */
584 vmbus_close(device->channel);
586 /* Release all resources */
587 vfree(net_device->sub_cb_buf);
588 free_netvsc_device(net_device);
591 #define RING_AVAIL_PERCENT_HIWATER 20
592 #define RING_AVAIL_PERCENT_LOWATER 10
595 * Get the percentage of available bytes to write in the ring.
596 * The return value is in range from 0 to 100.
598 static inline u32 hv_ringbuf_avail_percent(
599 struct hv_ring_buffer_info *ring_info)
601 u32 avail_read, avail_write;
603 hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
605 return avail_write * 100 / ring_info->ring_datasize;
608 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
611 sync_change_bit(index, net_device->send_section_map);
614 static void netvsc_send_tx_complete(struct netvsc_device *net_device,
615 struct vmbus_channel *incoming_channel,
616 struct hv_device *device,
617 struct vmpacket_descriptor *packet)
619 struct sk_buff *skb = (struct sk_buff *)(unsigned long)packet->trans_id;
620 struct net_device *ndev = hv_get_drvdata(device);
621 struct net_device_context *net_device_ctx = netdev_priv(ndev);
622 struct vmbus_channel *channel = device->channel;
623 int num_outstanding_sends;
627 /* Notify the layer above us */
629 struct hv_netvsc_packet *nvsc_packet
630 = (struct hv_netvsc_packet *)skb->cb;
631 u32 send_index = nvsc_packet->send_buf_index;
633 if (send_index != NETVSC_INVALID_INDEX)
634 netvsc_free_send_slot(net_device, send_index);
635 q_idx = nvsc_packet->q_idx;
636 channel = incoming_channel;
638 dev_consume_skb_any(skb);
641 num_outstanding_sends =
642 atomic_dec_return(&net_device->num_outstanding_sends);
643 queue_sends = atomic_dec_return(&net_device->queue_sends[q_idx]);
645 if (net_device->destroy && num_outstanding_sends == 0)
646 wake_up(&net_device->wait_drain);
648 if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev, q_idx)) &&
649 !net_device_ctx->start_remove &&
650 (hv_ringbuf_avail_percent(&channel->outbound) > RING_AVAIL_PERCENT_HIWATER ||
652 netif_tx_wake_queue(netdev_get_tx_queue(ndev, q_idx));
655 static void netvsc_send_completion(struct netvsc_device *net_device,
656 struct vmbus_channel *incoming_channel,
657 struct hv_device *device,
658 struct vmpacket_descriptor *packet)
660 struct nvsp_message *nvsp_packet;
661 struct net_device *ndev = hv_get_drvdata(device);
663 nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
664 (packet->offset8 << 3));
666 switch (nvsp_packet->hdr.msg_type) {
667 case NVSP_MSG_TYPE_INIT_COMPLETE:
668 case NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE:
669 case NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE:
670 case NVSP_MSG5_TYPE_SUBCHANNEL:
671 /* Copy the response back */
672 memcpy(&net_device->channel_init_pkt, nvsp_packet,
673 sizeof(struct nvsp_message));
674 complete(&net_device->channel_init_wait);
677 case NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE:
678 netvsc_send_tx_complete(net_device, incoming_channel,
684 "Unknown send completion type %d received!!\n",
685 nvsp_packet->hdr.msg_type);
689 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
692 u32 max_words = net_device->map_words;
693 unsigned long *map_addr = (unsigned long *)net_device->send_section_map;
694 u32 section_cnt = net_device->send_section_cnt;
695 int ret_val = NETVSC_INVALID_INDEX;
699 for (i = 0; i < max_words; i++) {
702 index = ffz(map_addr[i]);
703 prev_val = sync_test_and_set_bit(index, &map_addr[i]);
706 if ((index + (i * BITS_PER_LONG)) >= section_cnt)
708 ret_val = (index + (i * BITS_PER_LONG));
714 static u32 netvsc_copy_to_send_buf(struct netvsc_device *net_device,
715 unsigned int section_index,
717 struct hv_netvsc_packet *packet,
718 struct rndis_message *rndis_msg,
719 struct hv_page_buffer **pb,
722 char *start = net_device->send_buf;
723 char *dest = start + (section_index * net_device->send_section_size)
726 bool is_data_pkt = (skb != NULL) ? true : false;
727 bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
730 u32 remain = packet->total_data_buflen % net_device->pkt_align;
731 u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
732 packet->page_buf_cnt;
735 if (is_data_pkt && xmit_more && remain &&
736 !packet->cp_partial) {
737 padding = net_device->pkt_align - remain;
738 rndis_msg->msg_len += padding;
739 packet->total_data_buflen += padding;
742 for (i = 0; i < page_count; i++) {
743 char *src = phys_to_virt((*pb)[i].pfn << PAGE_SHIFT);
744 u32 offset = (*pb)[i].offset;
745 u32 len = (*pb)[i].len;
747 memcpy(dest, (src + offset), len);
753 memset(dest, 0, padding);
760 static inline int netvsc_send_pkt(
761 struct hv_device *device,
762 struct hv_netvsc_packet *packet,
763 struct netvsc_device *net_device,
764 struct hv_page_buffer **pb,
767 struct nvsp_message nvmsg;
768 u16 q_idx = packet->q_idx;
769 struct vmbus_channel *out_channel = net_device->chn_table[q_idx];
770 struct net_device *ndev = hv_get_drvdata(device);
773 struct hv_page_buffer *pgbuf;
774 u32 ring_avail = hv_ringbuf_avail_percent(&out_channel->outbound);
775 bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
777 nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
780 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 0;
782 /* 1 is RMC_CONTROL; */
783 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 1;
786 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
787 packet->send_buf_index;
788 if (packet->send_buf_index == NETVSC_INVALID_INDEX)
789 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
791 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size =
792 packet->total_data_buflen;
796 if (out_channel->rescind)
800 * It is possible that once we successfully place this packet
801 * on the ringbuffer, we may stop the queue. In that case, we want
802 * to notify the host independent of the xmit_more flag. We don't
803 * need to be precise here; in the worst case we may signal the host
806 if (ring_avail < (RING_AVAIL_PERCENT_LOWATER + 1))
809 if (packet->page_buf_cnt) {
810 pgbuf = packet->cp_partial ? (*pb) +
811 packet->rmsg_pgcnt : (*pb);
812 ret = vmbus_sendpacket_pagebuffer_ctl(out_channel,
814 packet->page_buf_cnt,
816 sizeof(struct nvsp_message),
818 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED,
821 ret = vmbus_sendpacket_ctl(out_channel, &nvmsg,
822 sizeof(struct nvsp_message),
825 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED,
830 atomic_inc(&net_device->num_outstanding_sends);
831 atomic_inc(&net_device->queue_sends[q_idx]);
833 if (ring_avail < RING_AVAIL_PERCENT_LOWATER) {
834 netif_tx_stop_queue(netdev_get_tx_queue(ndev, q_idx));
836 if (atomic_read(&net_device->
837 queue_sends[q_idx]) < 1)
838 netif_tx_wake_queue(netdev_get_tx_queue(
841 } else if (ret == -EAGAIN) {
842 netif_tx_stop_queue(netdev_get_tx_queue(
844 if (atomic_read(&net_device->queue_sends[q_idx]) < 1) {
845 netif_tx_wake_queue(netdev_get_tx_queue(
850 netdev_err(ndev, "Unable to send packet %p ret %d\n",
857 /* Move packet out of multi send data (msd), and clear msd */
858 static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
859 struct sk_buff **msd_skb,
860 struct multi_send_data *msdp)
862 *msd_skb = msdp->skb;
863 *msd_send = msdp->pkt;
869 int netvsc_send(struct hv_device *device,
870 struct hv_netvsc_packet *packet,
871 struct rndis_message *rndis_msg,
872 struct hv_page_buffer **pb,
875 struct netvsc_device *net_device;
877 struct vmbus_channel *out_channel;
878 u16 q_idx = packet->q_idx;
879 u32 pktlen = packet->total_data_buflen, msd_len = 0;
880 unsigned int section_index = NETVSC_INVALID_INDEX;
881 struct multi_send_data *msdp;
882 struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
883 struct sk_buff *msd_skb = NULL;
885 bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
887 net_device = get_outbound_net_device(device);
891 out_channel = net_device->chn_table[q_idx];
893 packet->send_buf_index = NETVSC_INVALID_INDEX;
894 packet->cp_partial = false;
896 /* Send control message directly without accessing msd (Multi-Send
897 * Data) field which may be changed during data packet processing.
904 msdp = &net_device->msd[q_idx];
906 /* batch packets in send buffer if possible */
908 msd_len = msdp->pkt->total_data_buflen;
910 try_batch = (skb != NULL) && msd_len > 0 && msdp->count <
913 if (try_batch && msd_len + pktlen + net_device->pkt_align <
914 net_device->send_section_size) {
915 section_index = msdp->pkt->send_buf_index;
917 } else if (try_batch && msd_len + packet->rmsg_size <
918 net_device->send_section_size) {
919 section_index = msdp->pkt->send_buf_index;
920 packet->cp_partial = true;
922 } else if ((skb != NULL) && pktlen + net_device->pkt_align <
923 net_device->send_section_size) {
924 section_index = netvsc_get_next_send_section(net_device);
925 if (section_index != NETVSC_INVALID_INDEX) {
926 move_pkt_msd(&msd_send, &msd_skb, msdp);
931 if (section_index != NETVSC_INVALID_INDEX) {
932 netvsc_copy_to_send_buf(net_device,
933 section_index, msd_len,
934 packet, rndis_msg, pb, skb);
936 packet->send_buf_index = section_index;
938 if (packet->cp_partial) {
939 packet->page_buf_cnt -= packet->rmsg_pgcnt;
940 packet->total_data_buflen = msd_len + packet->rmsg_size;
942 packet->page_buf_cnt = 0;
943 packet->total_data_buflen += msd_len;
947 dev_consume_skb_any(msdp->skb);
949 if (xmit_more && !packet->cp_partial) {
960 move_pkt_msd(&msd_send, &msd_skb, msdp);
965 int m_ret = netvsc_send_pkt(device, msd_send, net_device,
969 netvsc_free_send_slot(net_device,
970 msd_send->send_buf_index);
971 dev_kfree_skb_any(msd_skb);
977 ret = netvsc_send_pkt(device, cur_send, net_device, pb, skb);
979 if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
980 netvsc_free_send_slot(net_device, section_index);
985 static int netvsc_send_recv_completion(struct vmbus_channel *channel,
986 u64 transaction_id, u32 status)
988 struct nvsp_message recvcompMessage;
991 recvcompMessage.hdr.msg_type =
992 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
994 recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status = status;
996 /* Send the completion */
997 ret = vmbus_sendpacket(channel, &recvcompMessage,
998 sizeof(struct nvsp_message_header) + sizeof(u32),
999 transaction_id, VM_PKT_COMP, 0);
1004 static inline void count_recv_comp_slot(struct netvsc_device *nvdev, u16 q_idx,
1005 u32 *filled, u32 *avail)
1007 u32 first = nvdev->mrc[q_idx].first;
1008 u32 next = nvdev->mrc[q_idx].next;
1010 *filled = (first > next) ? NETVSC_RECVSLOT_MAX - first + next :
1013 *avail = NETVSC_RECVSLOT_MAX - *filled - 1;
1016 /* Read the first filled slot, no change to index */
1017 static inline struct recv_comp_data *read_recv_comp_slot(struct netvsc_device
1022 if (!nvdev->mrc[q_idx].buf)
1025 count_recv_comp_slot(nvdev, q_idx, &filled, &avail);
1029 return nvdev->mrc[q_idx].buf + nvdev->mrc[q_idx].first *
1030 sizeof(struct recv_comp_data);
1033 /* Put the first filled slot back to available pool */
1034 static inline void put_recv_comp_slot(struct netvsc_device *nvdev, u16 q_idx)
1038 nvdev->mrc[q_idx].first = (nvdev->mrc[q_idx].first + 1) %
1039 NETVSC_RECVSLOT_MAX;
1041 num_recv = atomic_dec_return(&nvdev->num_outstanding_recvs);
1043 if (nvdev->destroy && num_recv == 0)
1044 wake_up(&nvdev->wait_drain);
1047 /* Check and send pending recv completions */
1048 static void netvsc_chk_recv_comp(struct netvsc_device *nvdev,
1049 struct vmbus_channel *channel, u16 q_idx)
1051 struct recv_comp_data *rcd;
1055 rcd = read_recv_comp_slot(nvdev, q_idx);
1059 ret = netvsc_send_recv_completion(channel, rcd->tid,
1064 put_recv_comp_slot(nvdev, q_idx);
1068 #define NETVSC_RCD_WATERMARK 80
1070 /* Get next available slot */
1071 static inline struct recv_comp_data *get_recv_comp_slot(
1072 struct netvsc_device *nvdev, struct vmbus_channel *channel, u16 q_idx)
1074 u32 filled, avail, next;
1075 struct recv_comp_data *rcd;
1077 if (!nvdev->recv_section)
1080 if (!nvdev->mrc[q_idx].buf)
1083 if (atomic_read(&nvdev->num_outstanding_recvs) >
1084 nvdev->recv_section->num_sub_allocs * NETVSC_RCD_WATERMARK / 100)
1085 netvsc_chk_recv_comp(nvdev, channel, q_idx);
1087 count_recv_comp_slot(nvdev, q_idx, &filled, &avail);
1091 next = nvdev->mrc[q_idx].next;
1092 rcd = nvdev->mrc[q_idx].buf + next * sizeof(struct recv_comp_data);
1093 nvdev->mrc[q_idx].next = (next + 1) % NETVSC_RECVSLOT_MAX;
1095 atomic_inc(&nvdev->num_outstanding_recvs);
1100 static void netvsc_receive(struct netvsc_device *net_device,
1101 struct vmbus_channel *channel,
1102 struct hv_device *device,
1103 struct vmpacket_descriptor *packet)
1105 struct vmtransfer_page_packet_header *vmxferpage_packet;
1106 struct nvsp_message *nvsp_packet;
1107 struct hv_netvsc_packet nv_pkt;
1108 struct hv_netvsc_packet *netvsc_packet = &nv_pkt;
1109 u32 status = NVSP_STAT_SUCCESS;
1112 struct net_device *ndev = hv_get_drvdata(device);
1115 struct recv_comp_data *rcd;
1116 u16 q_idx = channel->offermsg.offer.sub_channel_index;
1119 * All inbound packets other than send completion should be xfer page
1122 if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
1123 netdev_err(ndev, "Unknown packet type received - %d\n",
1128 nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
1129 (packet->offset8 << 3));
1131 /* Make sure this is a valid nvsp packet */
1132 if (nvsp_packet->hdr.msg_type !=
1133 NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
1134 netdev_err(ndev, "Unknown nvsp packet type received-"
1135 " %d\n", nvsp_packet->hdr.msg_type);
1139 vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
1141 if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
1142 netdev_err(ndev, "Invalid xfer page set id - "
1143 "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
1144 vmxferpage_packet->xfer_pageset_id);
1148 count = vmxferpage_packet->range_cnt;
1150 /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1151 for (i = 0; i < count; i++) {
1152 /* Initialize the netvsc packet */
1153 data = (void *)((unsigned long)net_device->
1154 recv_buf + vmxferpage_packet->ranges[i].byte_offset);
1155 netvsc_packet->total_data_buflen =
1156 vmxferpage_packet->ranges[i].byte_count;
1158 /* Pass it to the upper layer */
1159 status = rndis_filter_receive(device, netvsc_packet, &data,
1163 if (!net_device->mrc[q_idx].buf) {
1164 ret = netvsc_send_recv_completion(channel,
1165 vmxferpage_packet->d.trans_id,
1168 netdev_err(ndev, "Recv_comp q:%hd, tid:%llx, err:%d\n",
1169 q_idx, vmxferpage_packet->d.trans_id, ret);
1173 rcd = get_recv_comp_slot(net_device, channel, q_idx);
1176 netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
1177 q_idx, vmxferpage_packet->d.trans_id);
1181 rcd->tid = vmxferpage_packet->d.trans_id;
1182 rcd->status = status;
1185 static void netvsc_send_table(struct hv_device *hdev,
1186 struct nvsp_message *nvmsg)
1188 struct netvsc_device *nvscdev;
1189 struct net_device *ndev = hv_get_drvdata(hdev);
1193 nvscdev = get_outbound_net_device(hdev);
1197 count = nvmsg->msg.v5_msg.send_table.count;
1198 if (count != VRSS_SEND_TAB_SIZE) {
1199 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1203 tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
1204 nvmsg->msg.v5_msg.send_table.offset);
1206 for (i = 0; i < count; i++)
1207 nvscdev->send_table[i] = tab[i];
1210 static void netvsc_send_vf(struct net_device_context *net_device_ctx,
1211 struct nvsp_message *nvmsg)
1213 net_device_ctx->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1214 net_device_ctx->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1217 static inline void netvsc_receive_inband(struct hv_device *hdev,
1218 struct net_device_context *net_device_ctx,
1219 struct nvsp_message *nvmsg)
1221 switch (nvmsg->hdr.msg_type) {
1222 case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1223 netvsc_send_table(hdev, nvmsg);
1226 case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1227 netvsc_send_vf(net_device_ctx, nvmsg);
1232 static void netvsc_process_raw_pkt(struct hv_device *device,
1233 struct vmbus_channel *channel,
1234 struct netvsc_device *net_device,
1235 struct net_device *ndev,
1237 struct vmpacket_descriptor *desc)
1239 struct nvsp_message *nvmsg;
1240 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1242 nvmsg = (struct nvsp_message *)((unsigned long)
1243 desc + (desc->offset8 << 3));
1245 switch (desc->type) {
1247 netvsc_send_completion(net_device, channel, device, desc);
1250 case VM_PKT_DATA_USING_XFER_PAGES:
1251 netvsc_receive(net_device, channel, device, desc);
1254 case VM_PKT_DATA_INBAND:
1255 netvsc_receive_inband(device, net_device_ctx, nvmsg);
1259 netdev_err(ndev, "unhandled packet type %d, tid %llx\n",
1260 desc->type, request_id);
1265 void netvsc_channel_cb(void *context)
1268 struct vmbus_channel *channel = (struct vmbus_channel *)context;
1269 u16 q_idx = channel->offermsg.offer.sub_channel_index;
1270 struct hv_device *device;
1271 struct netvsc_device *net_device;
1274 struct vmpacket_descriptor *desc;
1275 unsigned char *buffer;
1276 int bufferlen = NETVSC_PACKET_SIZE;
1277 struct net_device *ndev;
1278 bool need_to_commit = false;
1280 if (channel->primary_channel != NULL)
1281 device = channel->primary_channel->device_obj;
1283 device = channel->device_obj;
1285 net_device = get_inbound_net_device(device);
1288 ndev = hv_get_drvdata(device);
1289 buffer = get_per_channel_state(channel);
1292 desc = get_next_pkt_raw(channel);
1294 netvsc_process_raw_pkt(device,
1301 put_pkt_raw(channel, desc);
1302 need_to_commit = true;
1305 if (need_to_commit) {
1306 need_to_commit = false;
1307 commit_rd_index(channel);
1310 ret = vmbus_recvpacket_raw(channel, buffer, bufferlen,
1311 &bytes_recvd, &request_id);
1313 if (bytes_recvd > 0) {
1314 desc = (struct vmpacket_descriptor *)buffer;
1315 netvsc_process_raw_pkt(device,
1323 * We are done for this pass.
1328 } else if (ret == -ENOBUFS) {
1329 if (bufferlen > NETVSC_PACKET_SIZE)
1331 /* Handle large packet */
1332 buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
1333 if (buffer == NULL) {
1334 /* Try again next time around */
1336 "unable to allocate buffer of size "
1337 "(%d)!!\n", bytes_recvd);
1341 bufferlen = bytes_recvd;
1345 if (bufferlen > NETVSC_PACKET_SIZE)
1348 netvsc_chk_recv_comp(net_device, channel, q_idx);
1352 * netvsc_device_add - Callback when the device belonging to this
1355 int netvsc_device_add(struct hv_device *device, void *additional_info)
1359 ((struct netvsc_device_info *)additional_info)->ring_size;
1360 struct netvsc_device *net_device;
1361 struct net_device *ndev = hv_get_drvdata(device);
1362 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1364 net_device = alloc_net_device();
1368 net_device->ring_size = ring_size;
1370 set_per_channel_state(device->channel, net_device->cb_buffer);
1372 /* Open the channel */
1373 ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
1374 ring_size * PAGE_SIZE, NULL, 0,
1375 netvsc_channel_cb, device->channel);
1378 netdev_err(ndev, "unable to open channel: %d\n", ret);
1382 /* Channel is opened */
1383 pr_info("hv_netvsc channel opened successfully\n");
1385 /* If we're reopening the device we may have multiple queues, fill the
1386 * chn_table with the default channel to use it before subchannels are
1389 for (i = 0; i < VRSS_CHANNEL_MAX; i++)
1390 net_device->chn_table[i] = device->channel;
1392 /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1397 net_device_ctx->nvdev = net_device;
1399 /* Connect with the NetVsp */
1400 ret = netvsc_connect_vsp(device);
1403 "unable to connect to NetVSP - %d\n", ret);
1410 /* Now, we can close the channel safely */
1411 vmbus_close(device->channel);
1414 free_netvsc_device(net_device);