1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (c) 2009, Microsoft Corporation.
6 * Haiyang Zhang <haiyangz@microsoft.com>
7 * Hank Janssen <hjanssen@microsoft.com>
9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11 #include <linux/kernel.h>
12 #include <linux/sched.h>
13 #include <linux/wait.h>
15 #include <linux/delay.h>
17 #include <linux/slab.h>
18 #include <linux/netdevice.h>
19 #include <linux/if_ether.h>
20 #include <linux/vmalloc.h>
21 #include <linux/rtnetlink.h>
22 #include <linux/prefetch.h>
24 #include <asm/sync_bitops.h>
25 #include <asm/mshyperv.h>
27 #include "hyperv_net.h"
28 #include "netvsc_trace.h"
31 * Switch the data path from the synthetic interface to the VF
34 void netvsc_switch_datapath(struct net_device *ndev, bool vf)
36 struct net_device_context *net_device_ctx = netdev_priv(ndev);
37 struct hv_device *dev = net_device_ctx->device_ctx;
38 struct netvsc_device *nv_dev = rtnl_dereference(net_device_ctx->nvdev);
39 struct nvsp_message *init_pkt = &nv_dev->channel_init_pkt;
41 /* Block sending traffic to VF if it's about to be gone */
43 net_device_ctx->data_path_is_vf = vf;
45 memset(init_pkt, 0, sizeof(struct nvsp_message));
46 init_pkt->hdr.msg_type = NVSP_MSG4_TYPE_SWITCH_DATA_PATH;
48 init_pkt->msg.v4_msg.active_dp.active_datapath =
51 init_pkt->msg.v4_msg.active_dp.active_datapath =
52 NVSP_DATAPATH_SYNTHETIC;
54 trace_nvsp_send(ndev, init_pkt);
56 vmbus_sendpacket(dev->channel, init_pkt,
57 sizeof(struct nvsp_message),
58 (unsigned long)init_pkt,
60 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
61 wait_for_completion(&nv_dev->channel_init_wait);
62 net_device_ctx->data_path_is_vf = vf;
65 /* Worker to setup sub channels on initial setup
66 * Initial hotplug event occurs in softirq context
67 * and can't wait for channels.
69 static void netvsc_subchan_work(struct work_struct *w)
71 struct netvsc_device *nvdev =
72 container_of(w, struct netvsc_device, subchan_work);
73 struct rndis_device *rdev;
76 /* Avoid deadlock with device removal already under RTNL */
77 if (!rtnl_trylock()) {
82 rdev = nvdev->extension;
84 ret = rndis_set_subchannel(rdev->ndev, nvdev, NULL);
86 netif_device_attach(rdev->ndev);
88 /* fallback to only primary channel */
89 for (i = 1; i < nvdev->num_chn; i++)
90 netif_napi_del(&nvdev->chan_table[i].napi);
100 static struct netvsc_device *alloc_net_device(void)
102 struct netvsc_device *net_device;
104 net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
108 init_waitqueue_head(&net_device->wait_drain);
109 net_device->destroy = false;
110 net_device->tx_disable = true;
112 net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
113 net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
115 init_completion(&net_device->channel_init_wait);
116 init_waitqueue_head(&net_device->subchan_open);
117 INIT_WORK(&net_device->subchan_work, netvsc_subchan_work);
122 static void free_netvsc_device(struct rcu_head *head)
124 struct netvsc_device *nvdev
125 = container_of(head, struct netvsc_device, rcu);
128 kfree(nvdev->extension);
129 vfree(nvdev->recv_buf);
130 vfree(nvdev->send_buf);
131 kfree(nvdev->send_section_map);
133 for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
134 xdp_rxq_info_unreg(&nvdev->chan_table[i].xdp_rxq);
135 kfree(nvdev->chan_table[i].recv_buf);
136 vfree(nvdev->chan_table[i].mrc.slots);
142 static void free_netvsc_device_rcu(struct netvsc_device *nvdev)
144 call_rcu(&nvdev->rcu, free_netvsc_device);
147 static void netvsc_revoke_recv_buf(struct hv_device *device,
148 struct netvsc_device *net_device,
149 struct net_device *ndev)
151 struct nvsp_message *revoke_packet;
155 * If we got a section count, it means we received a
156 * SendReceiveBufferComplete msg (ie sent
157 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
158 * to send a revoke msg here
160 if (net_device->recv_section_cnt) {
161 /* Send the revoke receive buffer */
162 revoke_packet = &net_device->revoke_packet;
163 memset(revoke_packet, 0, sizeof(struct nvsp_message));
165 revoke_packet->hdr.msg_type =
166 NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
167 revoke_packet->msg.v1_msg.
168 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
170 trace_nvsp_send(ndev, revoke_packet);
172 ret = vmbus_sendpacket(device->channel,
174 sizeof(struct nvsp_message),
175 VMBUS_RQST_ID_NO_RESPONSE,
176 VM_PKT_DATA_INBAND, 0);
177 /* If the failure is because the channel is rescinded;
178 * ignore the failure since we cannot send on a rescinded
179 * channel. This would allow us to properly cleanup
180 * even when the channel is rescinded.
182 if (device->channel->rescind)
185 * If we failed here, we might as well return and
186 * have a leak rather than continue and a bugchk
189 netdev_err(ndev, "unable to send "
190 "revoke receive buffer to netvsp\n");
193 net_device->recv_section_cnt = 0;
197 static void netvsc_revoke_send_buf(struct hv_device *device,
198 struct netvsc_device *net_device,
199 struct net_device *ndev)
201 struct nvsp_message *revoke_packet;
204 /* Deal with the send buffer we may have setup.
205 * If we got a send section size, it means we received a
206 * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
207 * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
208 * to send a revoke msg here
210 if (net_device->send_section_cnt) {
211 /* Send the revoke receive buffer */
212 revoke_packet = &net_device->revoke_packet;
213 memset(revoke_packet, 0, sizeof(struct nvsp_message));
215 revoke_packet->hdr.msg_type =
216 NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
217 revoke_packet->msg.v1_msg.revoke_send_buf.id =
218 NETVSC_SEND_BUFFER_ID;
220 trace_nvsp_send(ndev, revoke_packet);
222 ret = vmbus_sendpacket(device->channel,
224 sizeof(struct nvsp_message),
225 VMBUS_RQST_ID_NO_RESPONSE,
226 VM_PKT_DATA_INBAND, 0);
228 /* If the failure is because the channel is rescinded;
229 * ignore the failure since we cannot send on a rescinded
230 * channel. This would allow us to properly cleanup
231 * even when the channel is rescinded.
233 if (device->channel->rescind)
236 /* If we failed here, we might as well return and
237 * have a leak rather than continue and a bugchk
240 netdev_err(ndev, "unable to send "
241 "revoke send buffer to netvsp\n");
244 net_device->send_section_cnt = 0;
248 static void netvsc_teardown_recv_gpadl(struct hv_device *device,
249 struct netvsc_device *net_device,
250 struct net_device *ndev)
254 if (net_device->recv_buf_gpadl_handle) {
255 ret = vmbus_teardown_gpadl(device->channel,
256 net_device->recv_buf_gpadl_handle);
258 /* If we failed here, we might as well return and have a leak
259 * rather than continue and a bugchk
263 "unable to teardown receive buffer's gpadl\n");
266 net_device->recv_buf_gpadl_handle = 0;
270 static void netvsc_teardown_send_gpadl(struct hv_device *device,
271 struct netvsc_device *net_device,
272 struct net_device *ndev)
276 if (net_device->send_buf_gpadl_handle) {
277 ret = vmbus_teardown_gpadl(device->channel,
278 net_device->send_buf_gpadl_handle);
280 /* If we failed here, we might as well return and have a leak
281 * rather than continue and a bugchk
285 "unable to teardown send buffer's gpadl\n");
288 net_device->send_buf_gpadl_handle = 0;
292 int netvsc_alloc_recv_comp_ring(struct netvsc_device *net_device, u32 q_idx)
294 struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
295 int node = cpu_to_node(nvchan->channel->target_cpu);
298 size = net_device->recv_completion_cnt * sizeof(struct recv_comp_data);
299 nvchan->mrc.slots = vzalloc_node(size, node);
300 if (!nvchan->mrc.slots)
301 nvchan->mrc.slots = vzalloc(size);
303 return nvchan->mrc.slots ? 0 : -ENOMEM;
306 static int netvsc_init_buf(struct hv_device *device,
307 struct netvsc_device *net_device,
308 const struct netvsc_device_info *device_info)
310 struct nvsp_1_message_send_receive_buffer_complete *resp;
311 struct net_device *ndev = hv_get_drvdata(device);
312 struct nvsp_message *init_packet;
313 unsigned int buf_size;
317 /* Get receive buffer area. */
318 buf_size = device_info->recv_sections * device_info->recv_section_size;
319 buf_size = roundup(buf_size, PAGE_SIZE);
321 /* Legacy hosts only allow smaller receive buffer */
322 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
323 buf_size = min_t(unsigned int, buf_size,
324 NETVSC_RECEIVE_BUFFER_SIZE_LEGACY);
326 net_device->recv_buf = vzalloc(buf_size);
327 if (!net_device->recv_buf) {
329 "unable to allocate receive buffer of size %u\n",
335 net_device->recv_buf_size = buf_size;
338 * Establish the gpadl handle for this buffer on this
339 * channel. Note: This call uses the vmbus connection rather
340 * than the channel to establish the gpadl handle.
342 ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
344 &net_device->recv_buf_gpadl_handle);
347 "unable to establish receive buffer's gpadl\n");
351 /* Notify the NetVsp of the gpadl handle */
352 init_packet = &net_device->channel_init_pkt;
353 memset(init_packet, 0, sizeof(struct nvsp_message));
354 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
355 init_packet->msg.v1_msg.send_recv_buf.
356 gpadl_handle = net_device->recv_buf_gpadl_handle;
357 init_packet->msg.v1_msg.
358 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
360 trace_nvsp_send(ndev, init_packet);
362 /* Send the gpadl notification request */
363 ret = vmbus_sendpacket(device->channel, init_packet,
364 sizeof(struct nvsp_message),
365 (unsigned long)init_packet,
367 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
370 "unable to send receive buffer's gpadl to netvsp\n");
374 wait_for_completion(&net_device->channel_init_wait);
376 /* Check the response */
377 resp = &init_packet->msg.v1_msg.send_recv_buf_complete;
378 if (resp->status != NVSP_STAT_SUCCESS) {
380 "Unable to complete receive buffer initialization with NetVsp - status %d\n",
386 /* Parse the response */
387 netdev_dbg(ndev, "Receive sections: %u sub_allocs: size %u count: %u\n",
388 resp->num_sections, resp->sections[0].sub_alloc_size,
389 resp->sections[0].num_sub_allocs);
391 /* There should only be one section for the entire receive buffer */
392 if (resp->num_sections != 1 || resp->sections[0].offset != 0) {
397 net_device->recv_section_size = resp->sections[0].sub_alloc_size;
398 net_device->recv_section_cnt = resp->sections[0].num_sub_allocs;
400 /* Ensure buffer will not overflow */
401 if (net_device->recv_section_size < NETVSC_MTU_MIN || (u64)net_device->recv_section_size *
402 (u64)net_device->recv_section_cnt > (u64)buf_size) {
403 netdev_err(ndev, "invalid recv_section_size %u\n",
404 net_device->recv_section_size);
409 for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
410 struct netvsc_channel *nvchan = &net_device->chan_table[i];
412 nvchan->recv_buf = kzalloc(net_device->recv_section_size, GFP_KERNEL);
413 if (nvchan->recv_buf == NULL) {
419 /* Setup receive completion ring.
420 * Add 1 to the recv_section_cnt because at least one entry in a
421 * ring buffer has to be empty.
423 net_device->recv_completion_cnt = net_device->recv_section_cnt + 1;
424 ret = netvsc_alloc_recv_comp_ring(net_device, 0);
428 /* Now setup the send buffer. */
429 buf_size = device_info->send_sections * device_info->send_section_size;
430 buf_size = round_up(buf_size, PAGE_SIZE);
432 net_device->send_buf = vzalloc(buf_size);
433 if (!net_device->send_buf) {
434 netdev_err(ndev, "unable to allocate send buffer of size %u\n",
440 /* Establish the gpadl handle for this buffer on this
441 * channel. Note: This call uses the vmbus connection rather
442 * than the channel to establish the gpadl handle.
444 ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
446 &net_device->send_buf_gpadl_handle);
449 "unable to establish send buffer's gpadl\n");
453 /* Notify the NetVsp of the gpadl handle */
454 init_packet = &net_device->channel_init_pkt;
455 memset(init_packet, 0, sizeof(struct nvsp_message));
456 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
457 init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
458 net_device->send_buf_gpadl_handle;
459 init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
461 trace_nvsp_send(ndev, init_packet);
463 /* Send the gpadl notification request */
464 ret = vmbus_sendpacket(device->channel, init_packet,
465 sizeof(struct nvsp_message),
466 (unsigned long)init_packet,
468 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
471 "unable to send send buffer's gpadl to netvsp\n");
475 wait_for_completion(&net_device->channel_init_wait);
477 /* Check the response */
478 if (init_packet->msg.v1_msg.
479 send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
480 netdev_err(ndev, "Unable to complete send buffer "
481 "initialization with NetVsp - status %d\n",
482 init_packet->msg.v1_msg.
483 send_send_buf_complete.status);
488 /* Parse the response */
489 net_device->send_section_size = init_packet->msg.
490 v1_msg.send_send_buf_complete.section_size;
491 if (net_device->send_section_size < NETVSC_MTU_MIN) {
492 netdev_err(ndev, "invalid send_section_size %u\n",
493 net_device->send_section_size);
498 /* Section count is simply the size divided by the section size. */
499 net_device->send_section_cnt = buf_size / net_device->send_section_size;
501 netdev_dbg(ndev, "Send section size: %d, Section count:%d\n",
502 net_device->send_section_size, net_device->send_section_cnt);
504 /* Setup state for managing the send buffer. */
505 map_words = DIV_ROUND_UP(net_device->send_section_cnt, BITS_PER_LONG);
507 net_device->send_section_map = kcalloc(map_words, sizeof(ulong), GFP_KERNEL);
508 if (net_device->send_section_map == NULL) {
516 netvsc_revoke_recv_buf(device, net_device, ndev);
517 netvsc_revoke_send_buf(device, net_device, ndev);
518 netvsc_teardown_recv_gpadl(device, net_device, ndev);
519 netvsc_teardown_send_gpadl(device, net_device, ndev);
525 /* Negotiate NVSP protocol version */
526 static int negotiate_nvsp_ver(struct hv_device *device,
527 struct netvsc_device *net_device,
528 struct nvsp_message *init_packet,
531 struct net_device *ndev = hv_get_drvdata(device);
534 memset(init_packet, 0, sizeof(struct nvsp_message));
535 init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
536 init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
537 init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
538 trace_nvsp_send(ndev, init_packet);
540 /* Send the init request */
541 ret = vmbus_sendpacket(device->channel, init_packet,
542 sizeof(struct nvsp_message),
543 (unsigned long)init_packet,
545 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
550 wait_for_completion(&net_device->channel_init_wait);
552 if (init_packet->msg.init_msg.init_complete.status !=
556 if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
559 /* NVSPv2 or later: Send NDIS config */
560 memset(init_packet, 0, sizeof(struct nvsp_message));
561 init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
562 init_packet->msg.v2_msg.send_ndis_config.mtu = ndev->mtu + ETH_HLEN;
563 init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
565 if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5) {
566 if (hv_is_isolation_supported())
567 netdev_info(ndev, "SR-IOV not advertised by guests on the host supporting isolation\n");
569 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
571 /* Teaming bit is needed to receive link speed updates */
572 init_packet->msg.v2_msg.send_ndis_config.capability.teaming = 1;
575 if (nvsp_ver >= NVSP_PROTOCOL_VERSION_61)
576 init_packet->msg.v2_msg.send_ndis_config.capability.rsc = 1;
578 trace_nvsp_send(ndev, init_packet);
580 ret = vmbus_sendpacket(device->channel, init_packet,
581 sizeof(struct nvsp_message),
582 VMBUS_RQST_ID_NO_RESPONSE,
583 VM_PKT_DATA_INBAND, 0);
588 static int netvsc_connect_vsp(struct hv_device *device,
589 struct netvsc_device *net_device,
590 const struct netvsc_device_info *device_info)
592 struct net_device *ndev = hv_get_drvdata(device);
593 static const u32 ver_list[] = {
594 NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
595 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5,
596 NVSP_PROTOCOL_VERSION_6, NVSP_PROTOCOL_VERSION_61
598 struct nvsp_message *init_packet;
599 int ndis_version, i, ret;
601 init_packet = &net_device->channel_init_pkt;
603 /* Negotiate the latest NVSP protocol supported */
604 for (i = ARRAY_SIZE(ver_list) - 1; i >= 0; i--)
605 if (negotiate_nvsp_ver(device, net_device, init_packet,
607 net_device->nvsp_version = ver_list[i];
616 if (hv_is_isolation_supported() && net_device->nvsp_version < NVSP_PROTOCOL_VERSION_61) {
617 netdev_err(ndev, "Invalid NVSP version 0x%x (expected >= 0x%x) from the host supporting isolation\n",
618 net_device->nvsp_version, NVSP_PROTOCOL_VERSION_61);
623 pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
625 /* Send the ndis version */
626 memset(init_packet, 0, sizeof(struct nvsp_message));
628 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
629 ndis_version = 0x00060001;
631 ndis_version = 0x0006001e;
633 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
634 init_packet->msg.v1_msg.
635 send_ndis_ver.ndis_major_ver =
636 (ndis_version & 0xFFFF0000) >> 16;
637 init_packet->msg.v1_msg.
638 send_ndis_ver.ndis_minor_ver =
639 ndis_version & 0xFFFF;
641 trace_nvsp_send(ndev, init_packet);
643 /* Send the init request */
644 ret = vmbus_sendpacket(device->channel, init_packet,
645 sizeof(struct nvsp_message),
646 VMBUS_RQST_ID_NO_RESPONSE,
647 VM_PKT_DATA_INBAND, 0);
652 ret = netvsc_init_buf(device, net_device, device_info);
659 * netvsc_device_remove - Callback when the root bus device is removed
661 void netvsc_device_remove(struct hv_device *device)
663 struct net_device *ndev = hv_get_drvdata(device);
664 struct net_device_context *net_device_ctx = netdev_priv(ndev);
665 struct netvsc_device *net_device
666 = rtnl_dereference(net_device_ctx->nvdev);
670 * Revoke receive buffer. If host is pre-Win2016 then tear down
671 * receive buffer GPADL. Do the same for send buffer.
673 netvsc_revoke_recv_buf(device, net_device, ndev);
674 if (vmbus_proto_version < VERSION_WIN10)
675 netvsc_teardown_recv_gpadl(device, net_device, ndev);
677 netvsc_revoke_send_buf(device, net_device, ndev);
678 if (vmbus_proto_version < VERSION_WIN10)
679 netvsc_teardown_send_gpadl(device, net_device, ndev);
681 RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
683 /* Disable NAPI and disassociate its context from the device. */
684 for (i = 0; i < net_device->num_chn; i++) {
685 /* See also vmbus_reset_channel_cb(). */
686 napi_disable(&net_device->chan_table[i].napi);
687 netif_napi_del(&net_device->chan_table[i].napi);
691 * At this point, no one should be accessing net_device
694 netdev_dbg(ndev, "net device safe to remove\n");
696 /* Now, we can close the channel safely */
697 vmbus_close(device->channel);
700 * If host is Win2016 or higher then we do the GPADL tear down
701 * here after VMBus is closed.
703 if (vmbus_proto_version >= VERSION_WIN10) {
704 netvsc_teardown_recv_gpadl(device, net_device, ndev);
705 netvsc_teardown_send_gpadl(device, net_device, ndev);
708 /* Release all resources */
709 free_netvsc_device_rcu(net_device);
712 #define RING_AVAIL_PERCENT_HIWATER 20
713 #define RING_AVAIL_PERCENT_LOWATER 10
715 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
718 sync_change_bit(index, net_device->send_section_map);
721 static void netvsc_send_tx_complete(struct net_device *ndev,
722 struct netvsc_device *net_device,
723 struct vmbus_channel *channel,
724 const struct vmpacket_descriptor *desc,
727 struct net_device_context *ndev_ctx = netdev_priv(ndev);
733 cmd_rqst = vmbus_request_addr(&channel->requestor, (u64)desc->trans_id);
734 if (cmd_rqst == VMBUS_RQST_ERROR) {
735 netdev_err(ndev, "Incorrect transaction id\n");
739 skb = (struct sk_buff *)(unsigned long)cmd_rqst;
741 /* Notify the layer above us */
743 const struct hv_netvsc_packet *packet
744 = (struct hv_netvsc_packet *)skb->cb;
745 u32 send_index = packet->send_buf_index;
746 struct netvsc_stats *tx_stats;
748 if (send_index != NETVSC_INVALID_INDEX)
749 netvsc_free_send_slot(net_device, send_index);
750 q_idx = packet->q_idx;
752 tx_stats = &net_device->chan_table[q_idx].tx_stats;
754 u64_stats_update_begin(&tx_stats->syncp);
755 tx_stats->packets += packet->total_packets;
756 tx_stats->bytes += packet->total_bytes;
757 u64_stats_update_end(&tx_stats->syncp);
759 napi_consume_skb(skb, budget);
763 atomic_dec_return(&net_device->chan_table[q_idx].queue_sends);
765 if (unlikely(net_device->destroy)) {
766 if (queue_sends == 0)
767 wake_up(&net_device->wait_drain);
769 struct netdev_queue *txq = netdev_get_tx_queue(ndev, q_idx);
771 if (netif_tx_queue_stopped(txq) && !net_device->tx_disable &&
772 (hv_get_avail_to_write_percent(&channel->outbound) >
773 RING_AVAIL_PERCENT_HIWATER || queue_sends < 1)) {
774 netif_tx_wake_queue(txq);
775 ndev_ctx->eth_stats.wake_queue++;
780 static void netvsc_send_completion(struct net_device *ndev,
781 struct netvsc_device *net_device,
782 struct vmbus_channel *incoming_channel,
783 const struct vmpacket_descriptor *desc,
786 const struct nvsp_message *nvsp_packet;
787 u32 msglen = hv_pkt_datalen(desc);
788 struct nvsp_message *pkt_rqst;
791 /* First check if this is a VMBUS completion without data payload */
793 cmd_rqst = vmbus_request_addr(&incoming_channel->requestor,
794 (u64)desc->trans_id);
795 if (cmd_rqst == VMBUS_RQST_ERROR) {
796 netdev_err(ndev, "Invalid transaction id\n");
800 pkt_rqst = (struct nvsp_message *)(uintptr_t)cmd_rqst;
801 switch (pkt_rqst->hdr.msg_type) {
802 case NVSP_MSG4_TYPE_SWITCH_DATA_PATH:
803 complete(&net_device->channel_init_wait);
807 netdev_err(ndev, "Unexpected VMBUS completion!!\n");
812 /* Ensure packet is big enough to read header fields */
813 if (msglen < sizeof(struct nvsp_message_header)) {
814 netdev_err(ndev, "nvsp_message length too small: %u\n", msglen);
818 nvsp_packet = hv_pkt_data(desc);
819 switch (nvsp_packet->hdr.msg_type) {
820 case NVSP_MSG_TYPE_INIT_COMPLETE:
821 if (msglen < sizeof(struct nvsp_message_header) +
822 sizeof(struct nvsp_message_init_complete)) {
823 netdev_err(ndev, "nvsp_msg length too small: %u\n",
829 case NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE:
830 if (msglen < sizeof(struct nvsp_message_header) +
831 sizeof(struct nvsp_1_message_send_receive_buffer_complete)) {
832 netdev_err(ndev, "nvsp_msg1 length too small: %u\n",
838 case NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE:
839 if (msglen < sizeof(struct nvsp_message_header) +
840 sizeof(struct nvsp_1_message_send_send_buffer_complete)) {
841 netdev_err(ndev, "nvsp_msg1 length too small: %u\n",
847 case NVSP_MSG5_TYPE_SUBCHANNEL:
848 if (msglen < sizeof(struct nvsp_message_header) +
849 sizeof(struct nvsp_5_subchannel_complete)) {
850 netdev_err(ndev, "nvsp_msg5 length too small: %u\n",
854 /* Copy the response back */
855 memcpy(&net_device->channel_init_pkt, nvsp_packet,
856 sizeof(struct nvsp_message));
857 complete(&net_device->channel_init_wait);
860 case NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE:
861 netvsc_send_tx_complete(ndev, net_device, incoming_channel,
867 "Unknown send completion type %d received!!\n",
868 nvsp_packet->hdr.msg_type);
872 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
874 unsigned long *map_addr = net_device->send_section_map;
877 for_each_clear_bit(i, map_addr, net_device->send_section_cnt) {
878 if (sync_test_and_set_bit(i, map_addr) == 0)
882 return NETVSC_INVALID_INDEX;
885 static void netvsc_copy_to_send_buf(struct netvsc_device *net_device,
886 unsigned int section_index,
888 struct hv_netvsc_packet *packet,
889 struct rndis_message *rndis_msg,
890 struct hv_page_buffer *pb,
893 char *start = net_device->send_buf;
894 char *dest = start + (section_index * net_device->send_section_size)
898 u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
899 packet->page_buf_cnt;
903 remain = packet->total_data_buflen & (net_device->pkt_align - 1);
904 if (xmit_more && remain) {
905 padding = net_device->pkt_align - remain;
906 rndis_msg->msg_len += padding;
907 packet->total_data_buflen += padding;
910 for (i = 0; i < page_count; i++) {
911 char *src = phys_to_virt(pb[i].pfn << HV_HYP_PAGE_SHIFT);
912 u32 offset = pb[i].offset;
915 memcpy(dest, (src + offset), len);
920 memset(dest, 0, padding);
923 static inline int netvsc_send_pkt(
924 struct hv_device *device,
925 struct hv_netvsc_packet *packet,
926 struct netvsc_device *net_device,
927 struct hv_page_buffer *pb,
930 struct nvsp_message nvmsg;
931 struct nvsp_1_message_send_rndis_packet *rpkt =
932 &nvmsg.msg.v1_msg.send_rndis_pkt;
933 struct netvsc_channel * const nvchan =
934 &net_device->chan_table[packet->q_idx];
935 struct vmbus_channel *out_channel = nvchan->channel;
936 struct net_device *ndev = hv_get_drvdata(device);
937 struct net_device_context *ndev_ctx = netdev_priv(ndev);
938 struct netdev_queue *txq = netdev_get_tx_queue(ndev, packet->q_idx);
941 u32 ring_avail = hv_get_avail_to_write_percent(&out_channel->outbound);
943 memset(&nvmsg, 0, sizeof(struct nvsp_message));
944 nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
946 rpkt->channel_type = 0; /* 0 is RMC_DATA */
948 rpkt->channel_type = 1; /* 1 is RMC_CONTROL */
950 rpkt->send_buf_section_index = packet->send_buf_index;
951 if (packet->send_buf_index == NETVSC_INVALID_INDEX)
952 rpkt->send_buf_section_size = 0;
954 rpkt->send_buf_section_size = packet->total_data_buflen;
958 if (out_channel->rescind)
961 trace_nvsp_send_pkt(ndev, out_channel, rpkt);
963 if (packet->page_buf_cnt) {
964 if (packet->cp_partial)
965 pb += packet->rmsg_pgcnt;
967 ret = vmbus_sendpacket_pagebuffer(out_channel,
968 pb, packet->page_buf_cnt,
969 &nvmsg, sizeof(nvmsg),
972 ret = vmbus_sendpacket(out_channel,
973 &nvmsg, sizeof(nvmsg),
974 req_id, VM_PKT_DATA_INBAND,
975 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
979 atomic_inc_return(&nvchan->queue_sends);
981 if (ring_avail < RING_AVAIL_PERCENT_LOWATER) {
982 netif_tx_stop_queue(txq);
983 ndev_ctx->eth_stats.stop_queue++;
985 } else if (ret == -EAGAIN) {
986 netif_tx_stop_queue(txq);
987 ndev_ctx->eth_stats.stop_queue++;
990 "Unable to send packet pages %u len %u, ret %d\n",
991 packet->page_buf_cnt, packet->total_data_buflen,
995 if (netif_tx_queue_stopped(txq) &&
996 atomic_read(&nvchan->queue_sends) < 1 &&
997 !net_device->tx_disable) {
998 netif_tx_wake_queue(txq);
999 ndev_ctx->eth_stats.wake_queue++;
1007 /* Move packet out of multi send data (msd), and clear msd */
1008 static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
1009 struct sk_buff **msd_skb,
1010 struct multi_send_data *msdp)
1012 *msd_skb = msdp->skb;
1013 *msd_send = msdp->pkt;
1019 /* RCU already held by caller */
1020 int netvsc_send(struct net_device *ndev,
1021 struct hv_netvsc_packet *packet,
1022 struct rndis_message *rndis_msg,
1023 struct hv_page_buffer *pb,
1024 struct sk_buff *skb,
1027 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1028 struct netvsc_device *net_device
1029 = rcu_dereference_bh(ndev_ctx->nvdev);
1030 struct hv_device *device = ndev_ctx->device_ctx;
1032 struct netvsc_channel *nvchan;
1033 u32 pktlen = packet->total_data_buflen, msd_len = 0;
1034 unsigned int section_index = NETVSC_INVALID_INDEX;
1035 struct multi_send_data *msdp;
1036 struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
1037 struct sk_buff *msd_skb = NULL;
1038 bool try_batch, xmit_more;
1040 /* If device is rescinded, return error and packet will get dropped. */
1041 if (unlikely(!net_device || net_device->destroy))
1044 nvchan = &net_device->chan_table[packet->q_idx];
1045 packet->send_buf_index = NETVSC_INVALID_INDEX;
1046 packet->cp_partial = false;
1048 /* Send a control message or XDP packet directly without accessing
1049 * msd (Multi-Send Data) field which may be changed during data packet
1053 return netvsc_send_pkt(device, packet, net_device, pb, skb);
1055 /* batch packets in send buffer if possible */
1056 msdp = &nvchan->msd;
1058 msd_len = msdp->pkt->total_data_buflen;
1060 try_batch = msd_len > 0 && msdp->count < net_device->max_pkt;
1061 if (try_batch && msd_len + pktlen + net_device->pkt_align <
1062 net_device->send_section_size) {
1063 section_index = msdp->pkt->send_buf_index;
1065 } else if (try_batch && msd_len + packet->rmsg_size <
1066 net_device->send_section_size) {
1067 section_index = msdp->pkt->send_buf_index;
1068 packet->cp_partial = true;
1070 } else if (pktlen + net_device->pkt_align <
1071 net_device->send_section_size) {
1072 section_index = netvsc_get_next_send_section(net_device);
1073 if (unlikely(section_index == NETVSC_INVALID_INDEX)) {
1074 ++ndev_ctx->eth_stats.tx_send_full;
1076 move_pkt_msd(&msd_send, &msd_skb, msdp);
1081 /* Keep aggregating only if stack says more data is coming
1082 * and not doing mixed modes send and not flow blocked
1084 xmit_more = netdev_xmit_more() &&
1085 !packet->cp_partial &&
1086 !netif_xmit_stopped(netdev_get_tx_queue(ndev, packet->q_idx));
1088 if (section_index != NETVSC_INVALID_INDEX) {
1089 netvsc_copy_to_send_buf(net_device,
1090 section_index, msd_len,
1091 packet, rndis_msg, pb, xmit_more);
1093 packet->send_buf_index = section_index;
1095 if (packet->cp_partial) {
1096 packet->page_buf_cnt -= packet->rmsg_pgcnt;
1097 packet->total_data_buflen = msd_len + packet->rmsg_size;
1099 packet->page_buf_cnt = 0;
1100 packet->total_data_buflen += msd_len;
1104 packet->total_packets += msdp->pkt->total_packets;
1105 packet->total_bytes += msdp->pkt->total_bytes;
1109 dev_consume_skb_any(msdp->skb);
1122 move_pkt_msd(&msd_send, &msd_skb, msdp);
1127 int m_ret = netvsc_send_pkt(device, msd_send, net_device,
1131 netvsc_free_send_slot(net_device,
1132 msd_send->send_buf_index);
1133 dev_kfree_skb_any(msd_skb);
1138 ret = netvsc_send_pkt(device, cur_send, net_device, pb, skb);
1140 if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
1141 netvsc_free_send_slot(net_device, section_index);
1146 /* Send pending recv completions */
1147 static int send_recv_completions(struct net_device *ndev,
1148 struct netvsc_device *nvdev,
1149 struct netvsc_channel *nvchan)
1151 struct multi_recv_comp *mrc = &nvchan->mrc;
1152 struct recv_comp_msg {
1153 struct nvsp_message_header hdr;
1156 struct recv_comp_msg msg = {
1157 .hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE,
1161 while (mrc->first != mrc->next) {
1162 const struct recv_comp_data *rcd
1163 = mrc->slots + mrc->first;
1165 msg.status = rcd->status;
1166 ret = vmbus_sendpacket(nvchan->channel, &msg, sizeof(msg),
1167 rcd->tid, VM_PKT_COMP, 0);
1168 if (unlikely(ret)) {
1169 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1171 ++ndev_ctx->eth_stats.rx_comp_busy;
1175 if (++mrc->first == nvdev->recv_completion_cnt)
1179 /* receive completion ring has been emptied */
1180 if (unlikely(nvdev->destroy))
1181 wake_up(&nvdev->wait_drain);
1186 /* Count how many receive completions are outstanding */
1187 static void recv_comp_slot_avail(const struct netvsc_device *nvdev,
1188 const struct multi_recv_comp *mrc,
1189 u32 *filled, u32 *avail)
1191 u32 count = nvdev->recv_completion_cnt;
1193 if (mrc->next >= mrc->first)
1194 *filled = mrc->next - mrc->first;
1196 *filled = (count - mrc->first) + mrc->next;
1198 *avail = count - *filled - 1;
1201 /* Add receive complete to ring to send to host. */
1202 static void enq_receive_complete(struct net_device *ndev,
1203 struct netvsc_device *nvdev, u16 q_idx,
1204 u64 tid, u32 status)
1206 struct netvsc_channel *nvchan = &nvdev->chan_table[q_idx];
1207 struct multi_recv_comp *mrc = &nvchan->mrc;
1208 struct recv_comp_data *rcd;
1211 recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1213 if (unlikely(filled > NAPI_POLL_WEIGHT)) {
1214 send_recv_completions(ndev, nvdev, nvchan);
1215 recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1218 if (unlikely(!avail)) {
1219 netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
1224 rcd = mrc->slots + mrc->next;
1226 rcd->status = status;
1228 if (++mrc->next == nvdev->recv_completion_cnt)
1232 static int netvsc_receive(struct net_device *ndev,
1233 struct netvsc_device *net_device,
1234 struct netvsc_channel *nvchan,
1235 const struct vmpacket_descriptor *desc)
1237 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1238 struct vmbus_channel *channel = nvchan->channel;
1239 const struct vmtransfer_page_packet_header *vmxferpage_packet
1240 = container_of(desc, const struct vmtransfer_page_packet_header, d);
1241 const struct nvsp_message *nvsp = hv_pkt_data(desc);
1242 u32 msglen = hv_pkt_datalen(desc);
1243 u16 q_idx = channel->offermsg.offer.sub_channel_index;
1244 char *recv_buf = net_device->recv_buf;
1245 u32 status = NVSP_STAT_SUCCESS;
1249 /* Ensure packet is big enough to read header fields */
1250 if (msglen < sizeof(struct nvsp_message_header)) {
1251 netif_err(net_device_ctx, rx_err, ndev,
1252 "invalid nvsp header, length too small: %u\n",
1257 /* Make sure this is a valid nvsp packet */
1258 if (unlikely(nvsp->hdr.msg_type != NVSP_MSG1_TYPE_SEND_RNDIS_PKT)) {
1259 netif_err(net_device_ctx, rx_err, ndev,
1260 "Unknown nvsp packet type received %u\n",
1261 nvsp->hdr.msg_type);
1265 /* Validate xfer page pkt header */
1266 if ((desc->offset8 << 3) < sizeof(struct vmtransfer_page_packet_header)) {
1267 netif_err(net_device_ctx, rx_err, ndev,
1268 "Invalid xfer page pkt, offset too small: %u\n",
1269 desc->offset8 << 3);
1273 if (unlikely(vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID)) {
1274 netif_err(net_device_ctx, rx_err, ndev,
1275 "Invalid xfer page set id - expecting %x got %x\n",
1276 NETVSC_RECEIVE_BUFFER_ID,
1277 vmxferpage_packet->xfer_pageset_id);
1281 count = vmxferpage_packet->range_cnt;
1283 /* Check count for a valid value */
1284 if (NETVSC_XFER_HEADER_SIZE(count) > desc->offset8 << 3) {
1285 netif_err(net_device_ctx, rx_err, ndev,
1286 "Range count is not valid: %d\n",
1291 /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1292 for (i = 0; i < count; i++) {
1293 u32 offset = vmxferpage_packet->ranges[i].byte_offset;
1294 u32 buflen = vmxferpage_packet->ranges[i].byte_count;
1298 if (unlikely(offset > net_device->recv_buf_size ||
1299 buflen > net_device->recv_buf_size - offset)) {
1300 nvchan->rsc.cnt = 0;
1301 status = NVSP_STAT_FAIL;
1302 netif_err(net_device_ctx, rx_err, ndev,
1303 "Packet offset:%u + len:%u too big\n",
1309 /* We're going to copy (sections of) the packet into nvchan->recv_buf;
1310 * make sure that nvchan->recv_buf is large enough to hold the packet.
1312 if (unlikely(buflen > net_device->recv_section_size)) {
1313 nvchan->rsc.cnt = 0;
1314 status = NVSP_STAT_FAIL;
1315 netif_err(net_device_ctx, rx_err, ndev,
1316 "Packet too big: buflen=%u recv_section_size=%u\n",
1317 buflen, net_device->recv_section_size);
1322 data = recv_buf + offset;
1324 nvchan->rsc.is_last = (i == count - 1);
1326 trace_rndis_recv(ndev, q_idx, data);
1328 /* Pass it to the upper layer */
1329 ret = rndis_filter_receive(ndev, net_device,
1330 nvchan, data, buflen);
1332 if (unlikely(ret != NVSP_STAT_SUCCESS)) {
1333 /* Drop incomplete packet */
1334 nvchan->rsc.cnt = 0;
1335 status = NVSP_STAT_FAIL;
1339 enq_receive_complete(ndev, net_device, q_idx,
1340 vmxferpage_packet->d.trans_id, status);
1345 static void netvsc_send_table(struct net_device *ndev,
1346 struct netvsc_device *nvscdev,
1347 const struct nvsp_message *nvmsg,
1350 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1351 u32 count, offset, *tab;
1354 /* Ensure packet is big enough to read send_table fields */
1355 if (msglen < sizeof(struct nvsp_message_header) +
1356 sizeof(struct nvsp_5_send_indirect_table)) {
1357 netdev_err(ndev, "nvsp_v5_msg length too small: %u\n", msglen);
1361 count = nvmsg->msg.v5_msg.send_table.count;
1362 offset = nvmsg->msg.v5_msg.send_table.offset;
1364 if (count != VRSS_SEND_TAB_SIZE) {
1365 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1369 /* If negotiated version <= NVSP_PROTOCOL_VERSION_6, the offset may be
1370 * wrong due to a host bug. So fix the offset here.
1372 if (nvscdev->nvsp_version <= NVSP_PROTOCOL_VERSION_6 &&
1373 msglen >= sizeof(struct nvsp_message_header) +
1374 sizeof(union nvsp_6_message_uber) + count * sizeof(u32))
1375 offset = sizeof(struct nvsp_message_header) +
1376 sizeof(union nvsp_6_message_uber);
1378 /* Boundary check for all versions */
1379 if (msglen < count * sizeof(u32) || offset > msglen - count * sizeof(u32)) {
1380 netdev_err(ndev, "Received send-table offset too big:%u\n",
1385 tab = (void *)nvmsg + offset;
1387 for (i = 0; i < count; i++)
1388 net_device_ctx->tx_table[i] = tab[i];
1391 static void netvsc_send_vf(struct net_device *ndev,
1392 const struct nvsp_message *nvmsg,
1395 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1397 /* Ensure packet is big enough to read its fields */
1398 if (msglen < sizeof(struct nvsp_message_header) +
1399 sizeof(struct nvsp_4_send_vf_association)) {
1400 netdev_err(ndev, "nvsp_v4_msg length too small: %u\n", msglen);
1404 net_device_ctx->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1405 net_device_ctx->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1406 netdev_info(ndev, "VF slot %u %s\n",
1407 net_device_ctx->vf_serial,
1408 net_device_ctx->vf_alloc ? "added" : "removed");
1411 static void netvsc_receive_inband(struct net_device *ndev,
1412 struct netvsc_device *nvscdev,
1413 const struct vmpacket_descriptor *desc)
1415 const struct nvsp_message *nvmsg = hv_pkt_data(desc);
1416 u32 msglen = hv_pkt_datalen(desc);
1418 /* Ensure packet is big enough to read header fields */
1419 if (msglen < sizeof(struct nvsp_message_header)) {
1420 netdev_err(ndev, "inband nvsp_message length too small: %u\n", msglen);
1424 switch (nvmsg->hdr.msg_type) {
1425 case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1426 netvsc_send_table(ndev, nvscdev, nvmsg, msglen);
1429 case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1430 if (hv_is_isolation_supported())
1431 netdev_err(ndev, "Ignore VF_ASSOCIATION msg from the host supporting isolation\n");
1433 netvsc_send_vf(ndev, nvmsg, msglen);
1438 static int netvsc_process_raw_pkt(struct hv_device *device,
1439 struct netvsc_channel *nvchan,
1440 struct netvsc_device *net_device,
1441 struct net_device *ndev,
1442 const struct vmpacket_descriptor *desc,
1445 struct vmbus_channel *channel = nvchan->channel;
1446 const struct nvsp_message *nvmsg = hv_pkt_data(desc);
1448 trace_nvsp_recv(ndev, channel, nvmsg);
1450 switch (desc->type) {
1452 netvsc_send_completion(ndev, net_device, channel, desc, budget);
1455 case VM_PKT_DATA_USING_XFER_PAGES:
1456 return netvsc_receive(ndev, net_device, nvchan, desc);
1459 case VM_PKT_DATA_INBAND:
1460 netvsc_receive_inband(ndev, net_device, desc);
1464 netdev_err(ndev, "unhandled packet type %d, tid %llx\n",
1465 desc->type, desc->trans_id);
1472 static struct hv_device *netvsc_channel_to_device(struct vmbus_channel *channel)
1474 struct vmbus_channel *primary = channel->primary_channel;
1476 return primary ? primary->device_obj : channel->device_obj;
1479 /* Network processing softirq
1480 * Process data in incoming ring buffer from host
1481 * Stops when ring is empty or budget is met or exceeded.
1483 int netvsc_poll(struct napi_struct *napi, int budget)
1485 struct netvsc_channel *nvchan
1486 = container_of(napi, struct netvsc_channel, napi);
1487 struct netvsc_device *net_device = nvchan->net_device;
1488 struct vmbus_channel *channel = nvchan->channel;
1489 struct hv_device *device = netvsc_channel_to_device(channel);
1490 struct net_device *ndev = hv_get_drvdata(device);
1494 /* If starting a new interval */
1496 nvchan->desc = hv_pkt_iter_first(channel);
1498 while (nvchan->desc && work_done < budget) {
1499 work_done += netvsc_process_raw_pkt(device, nvchan, net_device,
1500 ndev, nvchan->desc, budget);
1501 nvchan->desc = hv_pkt_iter_next(channel, nvchan->desc);
1504 /* Send any pending receive completions */
1505 ret = send_recv_completions(ndev, net_device, nvchan);
1507 /* If it did not exhaust NAPI budget this time
1508 * and not doing busy poll
1509 * then re-enable host interrupts
1510 * and reschedule if ring is not empty
1511 * or sending receive completion failed.
1513 if (work_done < budget &&
1514 napi_complete_done(napi, work_done) &&
1515 (ret || hv_end_read(&channel->inbound)) &&
1516 napi_schedule_prep(napi)) {
1517 hv_begin_read(&channel->inbound);
1518 __napi_schedule(napi);
1521 /* Driver may overshoot since multiple packets per descriptor */
1522 return min(work_done, budget);
1525 /* Call back when data is available in host ring buffer.
1526 * Processing is deferred until network softirq (NAPI)
1528 void netvsc_channel_cb(void *context)
1530 struct netvsc_channel *nvchan = context;
1531 struct vmbus_channel *channel = nvchan->channel;
1532 struct hv_ring_buffer_info *rbi = &channel->inbound;
1534 /* preload first vmpacket descriptor */
1535 prefetch(hv_get_ring_buffer(rbi) + rbi->priv_read_index);
1537 if (napi_schedule_prep(&nvchan->napi)) {
1538 /* disable interrupts from host */
1541 __napi_schedule_irqoff(&nvchan->napi);
1546 * netvsc_device_add - Callback when the device belonging to this
1549 struct netvsc_device *netvsc_device_add(struct hv_device *device,
1550 const struct netvsc_device_info *device_info)
1553 struct netvsc_device *net_device;
1554 struct net_device *ndev = hv_get_drvdata(device);
1555 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1557 net_device = alloc_net_device();
1559 return ERR_PTR(-ENOMEM);
1561 for (i = 0; i < VRSS_SEND_TAB_SIZE; i++)
1562 net_device_ctx->tx_table[i] = 0;
1564 /* Because the device uses NAPI, all the interrupt batching and
1565 * control is done via Net softirq, not the channel handling
1567 set_channel_read_mode(device->channel, HV_CALL_ISR);
1569 /* If we're reopening the device we may have multiple queues, fill the
1570 * chn_table with the default channel to use it before subchannels are
1572 * Initialize the channel state before we open;
1573 * we can be interrupted as soon as we open the channel.
1576 for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
1577 struct netvsc_channel *nvchan = &net_device->chan_table[i];
1579 nvchan->channel = device->channel;
1580 nvchan->net_device = net_device;
1581 u64_stats_init(&nvchan->tx_stats.syncp);
1582 u64_stats_init(&nvchan->rx_stats.syncp);
1584 ret = xdp_rxq_info_reg(&nvchan->xdp_rxq, ndev, i, 0);
1587 netdev_err(ndev, "xdp_rxq_info_reg fail: %d\n", ret);
1591 ret = xdp_rxq_info_reg_mem_model(&nvchan->xdp_rxq,
1592 MEM_TYPE_PAGE_SHARED, NULL);
1595 netdev_err(ndev, "xdp reg_mem_model fail: %d\n", ret);
1600 /* Enable NAPI handler before init callbacks */
1601 netif_napi_add(ndev, &net_device->chan_table[0].napi,
1602 netvsc_poll, NAPI_POLL_WEIGHT);
1604 /* Open the channel */
1605 device->channel->rqstor_size = netvsc_rqstor_size(netvsc_ring_bytes);
1606 ret = vmbus_open(device->channel, netvsc_ring_bytes,
1607 netvsc_ring_bytes, NULL, 0,
1608 netvsc_channel_cb, net_device->chan_table);
1611 netdev_err(ndev, "unable to open channel: %d\n", ret);
1615 /* Channel is opened */
1616 netdev_dbg(ndev, "hv_netvsc channel opened successfully\n");
1618 napi_enable(&net_device->chan_table[0].napi);
1620 /* Connect with the NetVsp */
1621 ret = netvsc_connect_vsp(device, net_device, device_info);
1624 "unable to connect to NetVSP - %d\n", ret);
1628 /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1631 rcu_assign_pointer(net_device_ctx->nvdev, net_device);
1636 RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
1637 napi_disable(&net_device->chan_table[0].napi);
1639 /* Now, we can close the channel safely */
1640 vmbus_close(device->channel);
1643 netif_napi_del(&net_device->chan_table[0].napi);
1646 free_netvsc_device(&net_device->rcu);
1648 return ERR_PTR(ret);