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
->chan_table
[0].mrc
.buf
71 = vzalloc(NETVSC_RECVSLOT_MAX
* sizeof(struct recv_comp_data
));
73 init_waitqueue_head(&net_device
->wait_drain
);
74 net_device
->destroy
= false;
75 atomic_set(&net_device
->open_cnt
, 0);
76 net_device
->max_pkt
= RNDIS_MAX_PKT_DEFAULT
;
77 net_device
->pkt_align
= RNDIS_PKT_ALIGN_DEFAULT
;
78 init_completion(&net_device
->channel_init_wait
);
83 static void free_netvsc_device(struct netvsc_device
*nvdev
)
87 for (i
= 0; i
< VRSS_CHANNEL_MAX
; i
++)
88 vfree(nvdev
->chan_table
[i
].mrc
.buf
);
94 static inline bool netvsc_channel_idle(const struct netvsc_device
*net_device
,
97 const struct netvsc_channel
*nvchan
= &net_device
->chan_table
[q_idx
];
99 return atomic_read(&net_device
->num_outstanding_recvs
) == 0 &&
100 atomic_read(&nvchan
->queue_sends
) == 0;
103 static struct netvsc_device
*get_outbound_net_device(struct hv_device
*device
)
105 struct netvsc_device
*net_device
= hv_device_to_netvsc_device(device
);
107 if (net_device
&& net_device
->destroy
)
113 static void netvsc_destroy_buf(struct hv_device
*device
)
115 struct nvsp_message
*revoke_packet
;
116 struct net_device
*ndev
= hv_get_drvdata(device
);
117 struct netvsc_device
*net_device
= net_device_to_netvsc_device(ndev
);
121 * If we got a section count, it means we received a
122 * SendReceiveBufferComplete msg (ie sent
123 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
124 * to send a revoke msg here
126 if (net_device
->recv_section_cnt
) {
127 /* Send the revoke receive buffer */
128 revoke_packet
= &net_device
->revoke_packet
;
129 memset(revoke_packet
, 0, sizeof(struct nvsp_message
));
131 revoke_packet
->hdr
.msg_type
=
132 NVSP_MSG1_TYPE_REVOKE_RECV_BUF
;
133 revoke_packet
->msg
.v1_msg
.
134 revoke_recv_buf
.id
= NETVSC_RECEIVE_BUFFER_ID
;
136 ret
= vmbus_sendpacket(device
->channel
,
138 sizeof(struct nvsp_message
),
139 (unsigned long)revoke_packet
,
140 VM_PKT_DATA_INBAND
, 0);
142 * If we failed here, we might as well return and
143 * have a leak rather than continue and a bugchk
146 netdev_err(ndev
, "unable to send "
147 "revoke receive buffer to netvsp\n");
152 /* Teardown the gpadl on the vsp end */
153 if (net_device
->recv_buf_gpadl_handle
) {
154 ret
= vmbus_teardown_gpadl(device
->channel
,
155 net_device
->recv_buf_gpadl_handle
);
157 /* If we failed here, we might as well return and have a leak
158 * rather than continue and a bugchk
162 "unable to teardown receive buffer's gpadl\n");
165 net_device
->recv_buf_gpadl_handle
= 0;
168 if (net_device
->recv_buf
) {
169 /* Free up the receive buffer */
170 vfree(net_device
->recv_buf
);
171 net_device
->recv_buf
= NULL
;
174 if (net_device
->recv_section
) {
175 net_device
->recv_section_cnt
= 0;
176 kfree(net_device
->recv_section
);
177 net_device
->recv_section
= NULL
;
180 /* Deal with the send buffer we may have setup.
181 * If we got a send section size, it means we received a
182 * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
183 * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
184 * to send a revoke msg here
186 if (net_device
->send_section_size
) {
187 /* Send the revoke receive buffer */
188 revoke_packet
= &net_device
->revoke_packet
;
189 memset(revoke_packet
, 0, sizeof(struct nvsp_message
));
191 revoke_packet
->hdr
.msg_type
=
192 NVSP_MSG1_TYPE_REVOKE_SEND_BUF
;
193 revoke_packet
->msg
.v1_msg
.revoke_send_buf
.id
=
194 NETVSC_SEND_BUFFER_ID
;
196 ret
= vmbus_sendpacket(device
->channel
,
198 sizeof(struct nvsp_message
),
199 (unsigned long)revoke_packet
,
200 VM_PKT_DATA_INBAND
, 0);
201 /* If we failed here, we might as well return and
202 * have a leak rather than continue and a bugchk
205 netdev_err(ndev
, "unable to send "
206 "revoke send buffer to netvsp\n");
210 /* Teardown the gpadl on the vsp end */
211 if (net_device
->send_buf_gpadl_handle
) {
212 ret
= vmbus_teardown_gpadl(device
->channel
,
213 net_device
->send_buf_gpadl_handle
);
215 /* If we failed here, we might as well return and have a leak
216 * rather than continue and a bugchk
220 "unable to teardown send buffer's gpadl\n");
223 net_device
->send_buf_gpadl_handle
= 0;
225 if (net_device
->send_buf
) {
226 /* Free up the send buffer */
227 vfree(net_device
->send_buf
);
228 net_device
->send_buf
= NULL
;
230 kfree(net_device
->send_section_map
);
233 static int netvsc_init_buf(struct hv_device
*device
)
236 struct netvsc_device
*net_device
;
237 struct nvsp_message
*init_packet
;
238 struct net_device
*ndev
;
241 net_device
= get_outbound_net_device(device
);
244 ndev
= hv_get_drvdata(device
);
246 node
= cpu_to_node(device
->channel
->target_cpu
);
247 net_device
->recv_buf
= vzalloc_node(net_device
->recv_buf_size
, node
);
248 if (!net_device
->recv_buf
)
249 net_device
->recv_buf
= vzalloc(net_device
->recv_buf_size
);
251 if (!net_device
->recv_buf
) {
252 netdev_err(ndev
, "unable to allocate receive "
253 "buffer of size %d\n", net_device
->recv_buf_size
);
259 * Establish the gpadl handle for this buffer on this
260 * channel. Note: This call uses the vmbus connection rather
261 * than the channel to establish the gpadl handle.
263 ret
= vmbus_establish_gpadl(device
->channel
, net_device
->recv_buf
,
264 net_device
->recv_buf_size
,
265 &net_device
->recv_buf_gpadl_handle
);
268 "unable to establish receive buffer's gpadl\n");
272 /* Notify the NetVsp of the gpadl handle */
273 init_packet
= &net_device
->channel_init_pkt
;
275 memset(init_packet
, 0, sizeof(struct nvsp_message
));
277 init_packet
->hdr
.msg_type
= NVSP_MSG1_TYPE_SEND_RECV_BUF
;
278 init_packet
->msg
.v1_msg
.send_recv_buf
.
279 gpadl_handle
= net_device
->recv_buf_gpadl_handle
;
280 init_packet
->msg
.v1_msg
.
281 send_recv_buf
.id
= NETVSC_RECEIVE_BUFFER_ID
;
283 /* Send the gpadl notification request */
284 ret
= vmbus_sendpacket(device
->channel
, init_packet
,
285 sizeof(struct nvsp_message
),
286 (unsigned long)init_packet
,
288 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
);
291 "unable to send receive buffer's gpadl to netvsp\n");
295 wait_for_completion(&net_device
->channel_init_wait
);
297 /* Check the response */
298 if (init_packet
->msg
.v1_msg
.
299 send_recv_buf_complete
.status
!= NVSP_STAT_SUCCESS
) {
300 netdev_err(ndev
, "Unable to complete receive buffer "
301 "initialization with NetVsp - status %d\n",
302 init_packet
->msg
.v1_msg
.
303 send_recv_buf_complete
.status
);
308 /* Parse the response */
310 net_device
->recv_section_cnt
= init_packet
->msg
.
311 v1_msg
.send_recv_buf_complete
.num_sections
;
313 net_device
->recv_section
= kmemdup(
314 init_packet
->msg
.v1_msg
.send_recv_buf_complete
.sections
,
315 net_device
->recv_section_cnt
*
316 sizeof(struct nvsp_1_receive_buffer_section
),
318 if (net_device
->recv_section
== NULL
) {
324 * For 1st release, there should only be 1 section that represents the
325 * entire receive buffer
327 if (net_device
->recv_section_cnt
!= 1 ||
328 net_device
->recv_section
->offset
!= 0) {
333 /* Now setup the send buffer.
335 net_device
->send_buf
= vzalloc_node(net_device
->send_buf_size
, node
);
336 if (!net_device
->send_buf
)
337 net_device
->send_buf
= vzalloc(net_device
->send_buf_size
);
338 if (!net_device
->send_buf
) {
339 netdev_err(ndev
, "unable to allocate send "
340 "buffer of size %d\n", net_device
->send_buf_size
);
345 /* Establish the gpadl handle for this buffer on this
346 * channel. Note: This call uses the vmbus connection rather
347 * than the channel to establish the gpadl handle.
349 ret
= vmbus_establish_gpadl(device
->channel
, net_device
->send_buf
,
350 net_device
->send_buf_size
,
351 &net_device
->send_buf_gpadl_handle
);
354 "unable to establish send buffer's gpadl\n");
358 /* Notify the NetVsp of the gpadl handle */
359 init_packet
= &net_device
->channel_init_pkt
;
360 memset(init_packet
, 0, sizeof(struct nvsp_message
));
361 init_packet
->hdr
.msg_type
= NVSP_MSG1_TYPE_SEND_SEND_BUF
;
362 init_packet
->msg
.v1_msg
.send_send_buf
.gpadl_handle
=
363 net_device
->send_buf_gpadl_handle
;
364 init_packet
->msg
.v1_msg
.send_send_buf
.id
= NETVSC_SEND_BUFFER_ID
;
366 /* Send the gpadl notification request */
367 ret
= vmbus_sendpacket(device
->channel
, init_packet
,
368 sizeof(struct nvsp_message
),
369 (unsigned long)init_packet
,
371 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
);
374 "unable to send send buffer's gpadl to netvsp\n");
378 wait_for_completion(&net_device
->channel_init_wait
);
380 /* Check the response */
381 if (init_packet
->msg
.v1_msg
.
382 send_send_buf_complete
.status
!= NVSP_STAT_SUCCESS
) {
383 netdev_err(ndev
, "Unable to complete send buffer "
384 "initialization with NetVsp - status %d\n",
385 init_packet
->msg
.v1_msg
.
386 send_send_buf_complete
.status
);
391 /* Parse the response */
392 net_device
->send_section_size
= init_packet
->msg
.
393 v1_msg
.send_send_buf_complete
.section_size
;
395 /* Section count is simply the size divided by the section size.
397 net_device
->send_section_cnt
=
398 net_device
->send_buf_size
/ net_device
->send_section_size
;
400 netdev_dbg(ndev
, "Send section size: %d, Section count:%d\n",
401 net_device
->send_section_size
, net_device
->send_section_cnt
);
403 /* Setup state for managing the send buffer. */
404 net_device
->map_words
= DIV_ROUND_UP(net_device
->send_section_cnt
,
407 net_device
->send_section_map
= kcalloc(net_device
->map_words
,
408 sizeof(ulong
), GFP_KERNEL
);
409 if (net_device
->send_section_map
== NULL
) {
417 netvsc_destroy_buf(device
);
423 /* Negotiate NVSP protocol version */
424 static int negotiate_nvsp_ver(struct hv_device
*device
,
425 struct netvsc_device
*net_device
,
426 struct nvsp_message
*init_packet
,
429 struct net_device
*ndev
= hv_get_drvdata(device
);
432 memset(init_packet
, 0, sizeof(struct nvsp_message
));
433 init_packet
->hdr
.msg_type
= NVSP_MSG_TYPE_INIT
;
434 init_packet
->msg
.init_msg
.init
.min_protocol_ver
= nvsp_ver
;
435 init_packet
->msg
.init_msg
.init
.max_protocol_ver
= nvsp_ver
;
437 /* Send the init request */
438 ret
= vmbus_sendpacket(device
->channel
, init_packet
,
439 sizeof(struct nvsp_message
),
440 (unsigned long)init_packet
,
442 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
);
447 wait_for_completion(&net_device
->channel_init_wait
);
449 if (init_packet
->msg
.init_msg
.init_complete
.status
!=
453 if (nvsp_ver
== NVSP_PROTOCOL_VERSION_1
)
456 /* NVSPv2 or later: Send NDIS config */
457 memset(init_packet
, 0, sizeof(struct nvsp_message
));
458 init_packet
->hdr
.msg_type
= NVSP_MSG2_TYPE_SEND_NDIS_CONFIG
;
459 init_packet
->msg
.v2_msg
.send_ndis_config
.mtu
= ndev
->mtu
+ ETH_HLEN
;
460 init_packet
->msg
.v2_msg
.send_ndis_config
.capability
.ieee8021q
= 1;
462 if (nvsp_ver
>= NVSP_PROTOCOL_VERSION_5
) {
463 init_packet
->msg
.v2_msg
.send_ndis_config
.capability
.sriov
= 1;
465 /* Teaming bit is needed to receive link speed updates */
466 init_packet
->msg
.v2_msg
.send_ndis_config
.capability
.teaming
= 1;
469 ret
= vmbus_sendpacket(device
->channel
, init_packet
,
470 sizeof(struct nvsp_message
),
471 (unsigned long)init_packet
,
472 VM_PKT_DATA_INBAND
, 0);
477 static int netvsc_connect_vsp(struct hv_device
*device
)
480 struct netvsc_device
*net_device
;
481 struct nvsp_message
*init_packet
;
483 const u32 ver_list
[] = {
484 NVSP_PROTOCOL_VERSION_1
, NVSP_PROTOCOL_VERSION_2
,
485 NVSP_PROTOCOL_VERSION_4
, NVSP_PROTOCOL_VERSION_5
};
488 net_device
= get_outbound_net_device(device
);
492 init_packet
= &net_device
->channel_init_pkt
;
494 /* Negotiate the latest NVSP protocol supported */
495 for (i
= ARRAY_SIZE(ver_list
) - 1; i
>= 0; i
--)
496 if (negotiate_nvsp_ver(device
, net_device
, init_packet
,
498 net_device
->nvsp_version
= ver_list
[i
];
507 pr_debug("Negotiated NVSP version:%x\n", net_device
->nvsp_version
);
509 /* Send the ndis version */
510 memset(init_packet
, 0, sizeof(struct nvsp_message
));
512 if (net_device
->nvsp_version
<= NVSP_PROTOCOL_VERSION_4
)
513 ndis_version
= 0x00060001;
515 ndis_version
= 0x0006001e;
517 init_packet
->hdr
.msg_type
= NVSP_MSG1_TYPE_SEND_NDIS_VER
;
518 init_packet
->msg
.v1_msg
.
519 send_ndis_ver
.ndis_major_ver
=
520 (ndis_version
& 0xFFFF0000) >> 16;
521 init_packet
->msg
.v1_msg
.
522 send_ndis_ver
.ndis_minor_ver
=
523 ndis_version
& 0xFFFF;
525 /* Send the init request */
526 ret
= vmbus_sendpacket(device
->channel
, init_packet
,
527 sizeof(struct nvsp_message
),
528 (unsigned long)init_packet
,
529 VM_PKT_DATA_INBAND
, 0);
533 /* Post the big receive buffer to NetVSP */
534 if (net_device
->nvsp_version
<= NVSP_PROTOCOL_VERSION_2
)
535 net_device
->recv_buf_size
= NETVSC_RECEIVE_BUFFER_SIZE_LEGACY
;
537 net_device
->recv_buf_size
= NETVSC_RECEIVE_BUFFER_SIZE
;
538 net_device
->send_buf_size
= NETVSC_SEND_BUFFER_SIZE
;
540 ret
= netvsc_init_buf(device
);
546 static void netvsc_disconnect_vsp(struct hv_device
*device
)
548 netvsc_destroy_buf(device
);
552 * netvsc_device_remove - Callback when the root bus device is removed
554 void netvsc_device_remove(struct hv_device
*device
)
556 struct net_device
*ndev
= hv_get_drvdata(device
);
557 struct net_device_context
*net_device_ctx
= netdev_priv(ndev
);
558 struct netvsc_device
*net_device
= net_device_ctx
->nvdev
;
560 netvsc_disconnect_vsp(device
);
562 net_device_ctx
->nvdev
= NULL
;
565 * At this point, no one should be accessing net_device
568 netdev_dbg(ndev
, "net device safe to remove\n");
570 /* Now, we can close the channel safely */
571 vmbus_close(device
->channel
);
573 /* Release all resources */
574 free_netvsc_device(net_device
);
577 #define RING_AVAIL_PERCENT_HIWATER 20
578 #define RING_AVAIL_PERCENT_LOWATER 10
581 * Get the percentage of available bytes to write in the ring.
582 * The return value is in range from 0 to 100.
584 static inline u32
hv_ringbuf_avail_percent(
585 struct hv_ring_buffer_info
*ring_info
)
587 u32 avail_read
, avail_write
;
589 hv_get_ringbuffer_availbytes(ring_info
, &avail_read
, &avail_write
);
591 return avail_write
* 100 / ring_info
->ring_datasize
;
594 static inline void netvsc_free_send_slot(struct netvsc_device
*net_device
,
597 sync_change_bit(index
, net_device
->send_section_map
);
600 static void netvsc_send_tx_complete(struct netvsc_device
*net_device
,
601 struct vmbus_channel
*incoming_channel
,
602 struct hv_device
*device
,
603 struct vmpacket_descriptor
*packet
)
605 struct sk_buff
*skb
= (struct sk_buff
*)(unsigned long)packet
->trans_id
;
606 struct net_device
*ndev
= hv_get_drvdata(device
);
607 struct net_device_context
*net_device_ctx
= netdev_priv(ndev
);
608 struct vmbus_channel
*channel
= device
->channel
;
612 /* Notify the layer above us */
614 const struct hv_netvsc_packet
*packet
615 = (struct hv_netvsc_packet
*)skb
->cb
;
616 u32 send_index
= packet
->send_buf_index
;
617 struct netvsc_stats
*tx_stats
;
619 if (send_index
!= NETVSC_INVALID_INDEX
)
620 netvsc_free_send_slot(net_device
, send_index
);
621 q_idx
= packet
->q_idx
;
622 channel
= incoming_channel
;
624 tx_stats
= &net_device
->chan_table
[q_idx
].tx_stats
;
626 u64_stats_update_begin(&tx_stats
->syncp
);
627 tx_stats
->packets
+= packet
->total_packets
;
628 tx_stats
->bytes
+= packet
->total_bytes
;
629 u64_stats_update_end(&tx_stats
->syncp
);
631 dev_consume_skb_any(skb
);
635 atomic_dec_return(&net_device
->chan_table
[q_idx
].queue_sends
);
637 if (net_device
->destroy
&& queue_sends
== 0)
638 wake_up(&net_device
->wait_drain
);
640 if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev
, q_idx
)) &&
641 !net_device_ctx
->start_remove
&&
642 (hv_ringbuf_avail_percent(&channel
->outbound
) > RING_AVAIL_PERCENT_HIWATER
||
644 netif_tx_wake_queue(netdev_get_tx_queue(ndev
, q_idx
));
647 static void netvsc_send_completion(struct netvsc_device
*net_device
,
648 struct vmbus_channel
*incoming_channel
,
649 struct hv_device
*device
,
650 struct vmpacket_descriptor
*packet
)
652 struct nvsp_message
*nvsp_packet
;
653 struct net_device
*ndev
= hv_get_drvdata(device
);
655 nvsp_packet
= (struct nvsp_message
*)((unsigned long)packet
+
656 (packet
->offset8
<< 3));
658 switch (nvsp_packet
->hdr
.msg_type
) {
659 case NVSP_MSG_TYPE_INIT_COMPLETE
:
660 case NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE
:
661 case NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE
:
662 case NVSP_MSG5_TYPE_SUBCHANNEL
:
663 /* Copy the response back */
664 memcpy(&net_device
->channel_init_pkt
, nvsp_packet
,
665 sizeof(struct nvsp_message
));
666 complete(&net_device
->channel_init_wait
);
669 case NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE
:
670 netvsc_send_tx_complete(net_device
, incoming_channel
,
676 "Unknown send completion type %d received!!\n",
677 nvsp_packet
->hdr
.msg_type
);
681 static u32
netvsc_get_next_send_section(struct netvsc_device
*net_device
)
683 unsigned long *map_addr
= net_device
->send_section_map
;
686 for_each_clear_bit(i
, map_addr
, net_device
->map_words
) {
687 if (sync_test_and_set_bit(i
, map_addr
) == 0)
691 return NETVSC_INVALID_INDEX
;
694 static u32
netvsc_copy_to_send_buf(struct netvsc_device
*net_device
,
695 unsigned int section_index
,
697 struct hv_netvsc_packet
*packet
,
698 struct rndis_message
*rndis_msg
,
699 struct hv_page_buffer
**pb
,
702 char *start
= net_device
->send_buf
;
703 char *dest
= start
+ (section_index
* net_device
->send_section_size
)
708 u32 remain
= packet
->total_data_buflen
% net_device
->pkt_align
;
709 u32 page_count
= packet
->cp_partial
? packet
->rmsg_pgcnt
:
710 packet
->page_buf_cnt
;
713 if (skb
&& skb
->xmit_more
&& remain
&&
714 !packet
->cp_partial
) {
715 padding
= net_device
->pkt_align
- remain
;
716 rndis_msg
->msg_len
+= padding
;
717 packet
->total_data_buflen
+= padding
;
720 for (i
= 0; i
< page_count
; i
++) {
721 char *src
= phys_to_virt((*pb
)[i
].pfn
<< PAGE_SHIFT
);
722 u32 offset
= (*pb
)[i
].offset
;
723 u32 len
= (*pb
)[i
].len
;
725 memcpy(dest
, (src
+ offset
), len
);
731 memset(dest
, 0, padding
);
738 static inline int netvsc_send_pkt(
739 struct hv_device
*device
,
740 struct hv_netvsc_packet
*packet
,
741 struct netvsc_device
*net_device
,
742 struct hv_page_buffer
**pb
,
745 struct nvsp_message nvmsg
;
746 struct netvsc_channel
*nvchan
747 = &net_device
->chan_table
[packet
->q_idx
];
748 struct vmbus_channel
*out_channel
= nvchan
->channel
;
749 struct net_device
*ndev
= hv_get_drvdata(device
);
750 struct netdev_queue
*txq
= netdev_get_tx_queue(ndev
, packet
->q_idx
);
753 struct hv_page_buffer
*pgbuf
;
754 u32 ring_avail
= hv_ringbuf_avail_percent(&out_channel
->outbound
);
756 nvmsg
.hdr
.msg_type
= NVSP_MSG1_TYPE_SEND_RNDIS_PKT
;
759 nvmsg
.msg
.v1_msg
.send_rndis_pkt
.channel_type
= 0;
761 /* 1 is RMC_CONTROL; */
762 nvmsg
.msg
.v1_msg
.send_rndis_pkt
.channel_type
= 1;
765 nvmsg
.msg
.v1_msg
.send_rndis_pkt
.send_buf_section_index
=
766 packet
->send_buf_index
;
767 if (packet
->send_buf_index
== NETVSC_INVALID_INDEX
)
768 nvmsg
.msg
.v1_msg
.send_rndis_pkt
.send_buf_section_size
= 0;
770 nvmsg
.msg
.v1_msg
.send_rndis_pkt
.send_buf_section_size
=
771 packet
->total_data_buflen
;
775 if (out_channel
->rescind
)
778 if (packet
->page_buf_cnt
) {
779 pgbuf
= packet
->cp_partial
? (*pb
) +
780 packet
->rmsg_pgcnt
: (*pb
);
781 ret
= vmbus_sendpacket_pagebuffer_ctl(out_channel
,
783 packet
->page_buf_cnt
,
785 sizeof(struct nvsp_message
),
787 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
);
789 ret
= vmbus_sendpacket_ctl(out_channel
, &nvmsg
,
790 sizeof(struct nvsp_message
),
793 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
);
797 atomic_inc_return(&nvchan
->queue_sends
);
799 if (ring_avail
< RING_AVAIL_PERCENT_LOWATER
)
800 netif_tx_stop_queue(txq
);
801 } else if (ret
== -EAGAIN
) {
802 netif_tx_stop_queue(txq
);
803 if (atomic_read(&nvchan
->queue_sends
) < 1) {
804 netif_tx_wake_queue(txq
);
808 netdev_err(ndev
, "Unable to send packet %p ret %d\n",
815 /* Move packet out of multi send data (msd), and clear msd */
816 static inline void move_pkt_msd(struct hv_netvsc_packet
**msd_send
,
817 struct sk_buff
**msd_skb
,
818 struct multi_send_data
*msdp
)
820 *msd_skb
= msdp
->skb
;
821 *msd_send
= msdp
->pkt
;
827 int netvsc_send(struct hv_device
*device
,
828 struct hv_netvsc_packet
*packet
,
829 struct rndis_message
*rndis_msg
,
830 struct hv_page_buffer
**pb
,
833 struct netvsc_device
*net_device
;
835 struct netvsc_channel
*nvchan
;
836 u32 pktlen
= packet
->total_data_buflen
, msd_len
= 0;
837 unsigned int section_index
= NETVSC_INVALID_INDEX
;
838 struct multi_send_data
*msdp
;
839 struct hv_netvsc_packet
*msd_send
= NULL
, *cur_send
= NULL
;
840 struct sk_buff
*msd_skb
= NULL
;
842 bool xmit_more
= (skb
!= NULL
) ? skb
->xmit_more
: false;
844 net_device
= get_outbound_net_device(device
);
848 /* We may race with netvsc_connect_vsp()/netvsc_init_buf() and get
849 * here before the negotiation with the host is finished and
850 * send_section_map may not be allocated yet.
852 if (!net_device
->send_section_map
)
855 nvchan
= &net_device
->chan_table
[packet
->q_idx
];
856 packet
->send_buf_index
= NETVSC_INVALID_INDEX
;
857 packet
->cp_partial
= false;
859 /* Send control message directly without accessing msd (Multi-Send
860 * Data) field which may be changed during data packet processing.
867 /* batch packets in send buffer if possible */
870 msd_len
= msdp
->pkt
->total_data_buflen
;
872 try_batch
= (skb
!= NULL
) && msd_len
> 0 && msdp
->count
<
875 if (try_batch
&& msd_len
+ pktlen
+ net_device
->pkt_align
<
876 net_device
->send_section_size
) {
877 section_index
= msdp
->pkt
->send_buf_index
;
879 } else if (try_batch
&& msd_len
+ packet
->rmsg_size
<
880 net_device
->send_section_size
) {
881 section_index
= msdp
->pkt
->send_buf_index
;
882 packet
->cp_partial
= true;
884 } else if ((skb
!= NULL
) && pktlen
+ net_device
->pkt_align
<
885 net_device
->send_section_size
) {
886 section_index
= netvsc_get_next_send_section(net_device
);
887 if (section_index
!= NETVSC_INVALID_INDEX
) {
888 move_pkt_msd(&msd_send
, &msd_skb
, msdp
);
893 if (section_index
!= NETVSC_INVALID_INDEX
) {
894 netvsc_copy_to_send_buf(net_device
,
895 section_index
, msd_len
,
896 packet
, rndis_msg
, pb
, skb
);
898 packet
->send_buf_index
= section_index
;
900 if (packet
->cp_partial
) {
901 packet
->page_buf_cnt
-= packet
->rmsg_pgcnt
;
902 packet
->total_data_buflen
= msd_len
+ packet
->rmsg_size
;
904 packet
->page_buf_cnt
= 0;
905 packet
->total_data_buflen
+= msd_len
;
909 packet
->total_packets
+= msdp
->pkt
->total_packets
;
910 packet
->total_bytes
+= msdp
->pkt
->total_bytes
;
914 dev_consume_skb_any(msdp
->skb
);
916 if (xmit_more
&& !packet
->cp_partial
) {
927 move_pkt_msd(&msd_send
, &msd_skb
, msdp
);
932 int m_ret
= netvsc_send_pkt(device
, msd_send
, net_device
,
936 netvsc_free_send_slot(net_device
,
937 msd_send
->send_buf_index
);
938 dev_kfree_skb_any(msd_skb
);
944 ret
= netvsc_send_pkt(device
, cur_send
, net_device
, pb
, skb
);
946 if (ret
!= 0 && section_index
!= NETVSC_INVALID_INDEX
)
947 netvsc_free_send_slot(net_device
, section_index
);
952 static int netvsc_send_recv_completion(struct vmbus_channel
*channel
,
953 u64 transaction_id
, u32 status
)
955 struct nvsp_message recvcompMessage
;
958 recvcompMessage
.hdr
.msg_type
=
959 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE
;
961 recvcompMessage
.msg
.v1_msg
.send_rndis_pkt_complete
.status
= status
;
963 /* Send the completion */
964 ret
= vmbus_sendpacket(channel
, &recvcompMessage
,
965 sizeof(struct nvsp_message_header
) + sizeof(u32
),
966 transaction_id
, VM_PKT_COMP
, 0);
971 static inline void count_recv_comp_slot(struct netvsc_device
*nvdev
, u16 q_idx
,
972 u32
*filled
, u32
*avail
)
974 struct multi_recv_comp
*mrc
= &nvdev
->chan_table
[q_idx
].mrc
;
975 u32 first
= mrc
->first
;
976 u32 next
= mrc
->next
;
978 *filled
= (first
> next
) ? NETVSC_RECVSLOT_MAX
- first
+ next
:
981 *avail
= NETVSC_RECVSLOT_MAX
- *filled
- 1;
984 /* Read the first filled slot, no change to index */
985 static inline struct recv_comp_data
*read_recv_comp_slot(struct netvsc_device
988 struct multi_recv_comp
*mrc
= &nvdev
->chan_table
[q_idx
].mrc
;
991 if (unlikely(!mrc
->buf
))
994 count_recv_comp_slot(nvdev
, q_idx
, &filled
, &avail
);
998 return mrc
->buf
+ mrc
->first
* sizeof(struct recv_comp_data
);
1001 /* Put the first filled slot back to available pool */
1002 static inline void put_recv_comp_slot(struct netvsc_device
*nvdev
, u16 q_idx
)
1004 struct multi_recv_comp
*mrc
= &nvdev
->chan_table
[q_idx
].mrc
;
1007 mrc
->first
= (mrc
->first
+ 1) % NETVSC_RECVSLOT_MAX
;
1009 num_recv
= atomic_dec_return(&nvdev
->num_outstanding_recvs
);
1011 if (nvdev
->destroy
&& num_recv
== 0)
1012 wake_up(&nvdev
->wait_drain
);
1015 /* Check and send pending recv completions */
1016 static void netvsc_chk_recv_comp(struct netvsc_device
*nvdev
,
1017 struct vmbus_channel
*channel
, u16 q_idx
)
1019 struct recv_comp_data
*rcd
;
1023 rcd
= read_recv_comp_slot(nvdev
, q_idx
);
1027 ret
= netvsc_send_recv_completion(channel
, rcd
->tid
,
1032 put_recv_comp_slot(nvdev
, q_idx
);
1036 #define NETVSC_RCD_WATERMARK 80
1038 /* Get next available slot */
1039 static inline struct recv_comp_data
*get_recv_comp_slot(
1040 struct netvsc_device
*nvdev
, struct vmbus_channel
*channel
, u16 q_idx
)
1042 struct multi_recv_comp
*mrc
= &nvdev
->chan_table
[q_idx
].mrc
;
1043 u32 filled
, avail
, next
;
1044 struct recv_comp_data
*rcd
;
1046 if (unlikely(!nvdev
->recv_section
))
1049 if (unlikely(!mrc
->buf
))
1052 if (atomic_read(&nvdev
->num_outstanding_recvs
) >
1053 nvdev
->recv_section
->num_sub_allocs
* NETVSC_RCD_WATERMARK
/ 100)
1054 netvsc_chk_recv_comp(nvdev
, channel
, q_idx
);
1056 count_recv_comp_slot(nvdev
, q_idx
, &filled
, &avail
);
1061 rcd
= mrc
->buf
+ next
* sizeof(struct recv_comp_data
);
1062 mrc
->next
= (next
+ 1) % NETVSC_RECVSLOT_MAX
;
1064 atomic_inc(&nvdev
->num_outstanding_recvs
);
1069 static void netvsc_receive(struct net_device
*ndev
,
1070 struct netvsc_device
*net_device
,
1071 struct net_device_context
*net_device_ctx
,
1072 struct hv_device
*device
,
1073 struct vmbus_channel
*channel
,
1074 struct vmtransfer_page_packet_header
*vmxferpage_packet
,
1075 struct nvsp_message
*nvsp
)
1077 char *recv_buf
= net_device
->recv_buf
;
1078 u32 status
= NVSP_STAT_SUCCESS
;
1082 struct recv_comp_data
*rcd
;
1083 u16 q_idx
= channel
->offermsg
.offer
.sub_channel_index
;
1085 /* Make sure this is a valid nvsp packet */
1086 if (unlikely(nvsp
->hdr
.msg_type
!= NVSP_MSG1_TYPE_SEND_RNDIS_PKT
)) {
1087 netif_err(net_device_ctx
, rx_err
, ndev
,
1088 "Unknown nvsp packet type received %u\n",
1089 nvsp
->hdr
.msg_type
);
1093 if (unlikely(vmxferpage_packet
->xfer_pageset_id
!= NETVSC_RECEIVE_BUFFER_ID
)) {
1094 netif_err(net_device_ctx
, rx_err
, ndev
,
1095 "Invalid xfer page set id - expecting %x got %x\n",
1096 NETVSC_RECEIVE_BUFFER_ID
,
1097 vmxferpage_packet
->xfer_pageset_id
);
1101 count
= vmxferpage_packet
->range_cnt
;
1103 /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1104 for (i
= 0; i
< count
; i
++) {
1105 void *data
= recv_buf
1106 + vmxferpage_packet
->ranges
[i
].byte_offset
;
1107 u32 buflen
= vmxferpage_packet
->ranges
[i
].byte_count
;
1109 /* Pass it to the upper layer */
1110 status
= rndis_filter_receive(ndev
, net_device
, device
,
1111 channel
, data
, buflen
);
1114 if (!net_device
->chan_table
[q_idx
].mrc
.buf
) {
1115 ret
= netvsc_send_recv_completion(channel
,
1116 vmxferpage_packet
->d
.trans_id
,
1119 netdev_err(ndev
, "Recv_comp q:%hd, tid:%llx, err:%d\n",
1120 q_idx
, vmxferpage_packet
->d
.trans_id
, ret
);
1124 rcd
= get_recv_comp_slot(net_device
, channel
, q_idx
);
1127 netdev_err(ndev
, "Recv_comp full buf q:%hd, tid:%llx\n",
1128 q_idx
, vmxferpage_packet
->d
.trans_id
);
1132 rcd
->tid
= vmxferpage_packet
->d
.trans_id
;
1133 rcd
->status
= status
;
1136 static void netvsc_send_table(struct hv_device
*hdev
,
1137 struct nvsp_message
*nvmsg
)
1139 struct netvsc_device
*nvscdev
;
1140 struct net_device
*ndev
= hv_get_drvdata(hdev
);
1144 nvscdev
= get_outbound_net_device(hdev
);
1148 count
= nvmsg
->msg
.v5_msg
.send_table
.count
;
1149 if (count
!= VRSS_SEND_TAB_SIZE
) {
1150 netdev_err(ndev
, "Received wrong send-table size:%u\n", count
);
1154 tab
= (u32
*)((unsigned long)&nvmsg
->msg
.v5_msg
.send_table
+
1155 nvmsg
->msg
.v5_msg
.send_table
.offset
);
1157 for (i
= 0; i
< count
; i
++)
1158 nvscdev
->send_table
[i
] = tab
[i
];
1161 static void netvsc_send_vf(struct net_device_context
*net_device_ctx
,
1162 struct nvsp_message
*nvmsg
)
1164 net_device_ctx
->vf_alloc
= nvmsg
->msg
.v4_msg
.vf_assoc
.allocated
;
1165 net_device_ctx
->vf_serial
= nvmsg
->msg
.v4_msg
.vf_assoc
.serial
;
1168 static inline void netvsc_receive_inband(struct hv_device
*hdev
,
1169 struct net_device_context
*net_device_ctx
,
1170 struct nvsp_message
*nvmsg
)
1172 switch (nvmsg
->hdr
.msg_type
) {
1173 case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE
:
1174 netvsc_send_table(hdev
, nvmsg
);
1177 case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION
:
1178 netvsc_send_vf(net_device_ctx
, nvmsg
);
1183 static void netvsc_process_raw_pkt(struct hv_device
*device
,
1184 struct vmbus_channel
*channel
,
1185 struct netvsc_device
*net_device
,
1186 struct net_device
*ndev
,
1188 struct vmpacket_descriptor
*desc
)
1190 struct net_device_context
*net_device_ctx
= netdev_priv(ndev
);
1191 struct nvsp_message
*nvmsg
1192 = (struct nvsp_message
*)((unsigned long)desc
1193 + (desc
->offset8
<< 3));
1195 switch (desc
->type
) {
1197 netvsc_send_completion(net_device
, channel
, device
, desc
);
1200 case VM_PKT_DATA_USING_XFER_PAGES
:
1201 netvsc_receive(ndev
, net_device
, net_device_ctx
,
1203 (struct vmtransfer_page_packet_header
*)desc
,
1207 case VM_PKT_DATA_INBAND
:
1208 netvsc_receive_inband(device
, net_device_ctx
, nvmsg
);
1212 netdev_err(ndev
, "unhandled packet type %d, tid %llx\n",
1213 desc
->type
, request_id
);
1218 void netvsc_channel_cb(void *context
)
1220 struct vmbus_channel
*channel
= context
;
1221 u16 q_idx
= channel
->offermsg
.offer
.sub_channel_index
;
1222 struct hv_device
*device
;
1223 struct netvsc_device
*net_device
;
1224 struct vmpacket_descriptor
*desc
;
1225 struct net_device
*ndev
;
1226 bool need_to_commit
= false;
1228 if (channel
->primary_channel
!= NULL
)
1229 device
= channel
->primary_channel
->device_obj
;
1231 device
= channel
->device_obj
;
1233 ndev
= hv_get_drvdata(device
);
1234 if (unlikely(!ndev
))
1237 net_device
= net_device_to_netvsc_device(ndev
);
1238 if (unlikely(net_device
->destroy
) &&
1239 netvsc_channel_idle(net_device
, q_idx
))
1242 /* commit_rd_index() -> hv_signal_on_read() needs this. */
1243 init_cached_read_index(channel
);
1245 while ((desc
= get_next_pkt_raw(channel
)) != NULL
) {
1246 netvsc_process_raw_pkt(device
, channel
, net_device
,
1247 ndev
, desc
->trans_id
, desc
);
1249 put_pkt_raw(channel
, desc
);
1250 need_to_commit
= true;
1254 commit_rd_index(channel
);
1256 netvsc_chk_recv_comp(net_device
, channel
, q_idx
);
1260 * netvsc_device_add - Callback when the device belonging to this
1263 int netvsc_device_add(struct hv_device
*device
,
1264 const struct netvsc_device_info
*device_info
)
1267 int ring_size
= device_info
->ring_size
;
1268 struct netvsc_device
*net_device
;
1269 struct net_device
*ndev
= hv_get_drvdata(device
);
1270 struct net_device_context
*net_device_ctx
= netdev_priv(ndev
);
1272 net_device
= alloc_net_device();
1276 net_device
->ring_size
= ring_size
;
1278 /* Open the channel */
1279 ret
= vmbus_open(device
->channel
, ring_size
* PAGE_SIZE
,
1280 ring_size
* PAGE_SIZE
, NULL
, 0,
1281 netvsc_channel_cb
, device
->channel
);
1284 netdev_err(ndev
, "unable to open channel: %d\n", ret
);
1288 /* Channel is opened */
1289 netdev_dbg(ndev
, "hv_netvsc channel opened successfully\n");
1291 /* If we're reopening the device we may have multiple queues, fill the
1292 * chn_table with the default channel to use it before subchannels are
1295 for (i
= 0; i
< VRSS_CHANNEL_MAX
; i
++)
1296 net_device
->chan_table
[i
].channel
= device
->channel
;
1298 /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1303 net_device_ctx
->nvdev
= net_device
;
1305 /* Connect with the NetVsp */
1306 ret
= netvsc_connect_vsp(device
);
1309 "unable to connect to NetVSP - %d\n", ret
);
1316 /* Now, we can close the channel safely */
1317 vmbus_close(device
->channel
);
1320 free_netvsc_device(net_device
);