2 * Copyright (c) 2009, Microsoft Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, see <http://www.gnu.org/licenses/>.
17 * Haiyang Zhang <haiyangz@microsoft.com>
18 * Hank Janssen <hjanssen@microsoft.com>
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
22 #include <linux/kernel.h>
23 #include <linux/sched.h>
24 #include <linux/wait.h>
26 #include <linux/delay.h>
28 #include <linux/slab.h>
29 #include <linux/netdevice.h>
30 #include <linux/if_ether.h>
31 #include <linux/vmalloc.h>
32 #include <asm/sync_bitops.h>
34 #include "hyperv_net.h"
37 * Switch the data path from the synthetic interface to the VF
40 void netvsc_switch_datapath(struct net_device
*ndev
, bool vf
)
42 struct net_device_context
*net_device_ctx
= netdev_priv(ndev
);
43 struct hv_device
*dev
= net_device_ctx
->device_ctx
;
44 struct netvsc_device
*nv_dev
= net_device_ctx
->nvdev
;
45 struct nvsp_message
*init_pkt
= &nv_dev
->channel_init_pkt
;
47 memset(init_pkt
, 0, sizeof(struct nvsp_message
));
48 init_pkt
->hdr
.msg_type
= NVSP_MSG4_TYPE_SWITCH_DATA_PATH
;
50 init_pkt
->msg
.v4_msg
.active_dp
.active_datapath
=
53 init_pkt
->msg
.v4_msg
.active_dp
.active_datapath
=
54 NVSP_DATAPATH_SYNTHETIC
;
56 vmbus_sendpacket(dev
->channel
, init_pkt
,
57 sizeof(struct nvsp_message
),
58 (unsigned long)init_pkt
,
59 VM_PKT_DATA_INBAND
, 0);
63 static struct netvsc_device
*alloc_net_device(void)
65 struct netvsc_device
*net_device
;
67 net_device
= kzalloc(sizeof(struct netvsc_device
), GFP_KERNEL
);
71 net_device
->cb_buffer
= kzalloc(NETVSC_PACKET_SIZE
, GFP_KERNEL
);
72 if (!net_device
->cb_buffer
) {
77 init_waitqueue_head(&net_device
->wait_drain
);
78 net_device
->destroy
= false;
79 atomic_set(&net_device
->open_cnt
, 0);
80 atomic_set(&net_device
->vf_use_cnt
, 0);
81 net_device
->max_pkt
= RNDIS_MAX_PKT_DEFAULT
;
82 net_device
->pkt_align
= RNDIS_PKT_ALIGN_DEFAULT
;
84 net_device
->vf_netdev
= NULL
;
85 net_device
->vf_inject
= false;
90 static void free_netvsc_device(struct netvsc_device
*nvdev
)
92 kfree(nvdev
->cb_buffer
);
96 static struct netvsc_device
*get_outbound_net_device(struct hv_device
*device
)
98 struct netvsc_device
*net_device
= hv_device_to_netvsc_device(device
);
100 if (net_device
&& net_device
->destroy
)
106 static struct netvsc_device
*get_inbound_net_device(struct hv_device
*device
)
108 struct netvsc_device
*net_device
= hv_device_to_netvsc_device(device
);
113 if (net_device
->destroy
&&
114 atomic_read(&net_device
->num_outstanding_sends
) == 0)
122 static int netvsc_destroy_buf(struct hv_device
*device
)
124 struct nvsp_message
*revoke_packet
;
126 struct net_device
*ndev
= hv_get_drvdata(device
);
127 struct netvsc_device
*net_device
= net_device_to_netvsc_device(ndev
);
130 * If we got a section count, it means we received a
131 * SendReceiveBufferComplete msg (ie sent
132 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
133 * to send a revoke msg here
135 if (net_device
->recv_section_cnt
) {
136 /* Send the revoke receive buffer */
137 revoke_packet
= &net_device
->revoke_packet
;
138 memset(revoke_packet
, 0, sizeof(struct nvsp_message
));
140 revoke_packet
->hdr
.msg_type
=
141 NVSP_MSG1_TYPE_REVOKE_RECV_BUF
;
142 revoke_packet
->msg
.v1_msg
.
143 revoke_recv_buf
.id
= NETVSC_RECEIVE_BUFFER_ID
;
145 ret
= vmbus_sendpacket(device
->channel
,
147 sizeof(struct nvsp_message
),
148 (unsigned long)revoke_packet
,
149 VM_PKT_DATA_INBAND
, 0);
151 * If we failed here, we might as well return and
152 * have a leak rather than continue and a bugchk
155 netdev_err(ndev
, "unable to send "
156 "revoke receive buffer to netvsp\n");
161 /* Teardown the gpadl on the vsp end */
162 if (net_device
->recv_buf_gpadl_handle
) {
163 ret
= vmbus_teardown_gpadl(device
->channel
,
164 net_device
->recv_buf_gpadl_handle
);
166 /* If we failed here, we might as well return and have a leak
167 * rather than continue and a bugchk
171 "unable to teardown receive buffer's gpadl\n");
174 net_device
->recv_buf_gpadl_handle
= 0;
177 if (net_device
->recv_buf
) {
178 /* Free up the receive buffer */
179 vfree(net_device
->recv_buf
);
180 net_device
->recv_buf
= NULL
;
183 if (net_device
->recv_section
) {
184 net_device
->recv_section_cnt
= 0;
185 kfree(net_device
->recv_section
);
186 net_device
->recv_section
= NULL
;
189 /* Deal with the send buffer we may have setup.
190 * If we got a send section size, it means we received a
191 * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
192 * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
193 * to send a revoke msg here
195 if (net_device
->send_section_size
) {
196 /* Send the revoke receive buffer */
197 revoke_packet
= &net_device
->revoke_packet
;
198 memset(revoke_packet
, 0, sizeof(struct nvsp_message
));
200 revoke_packet
->hdr
.msg_type
=
201 NVSP_MSG1_TYPE_REVOKE_SEND_BUF
;
202 revoke_packet
->msg
.v1_msg
.revoke_send_buf
.id
=
203 NETVSC_SEND_BUFFER_ID
;
205 ret
= vmbus_sendpacket(device
->channel
,
207 sizeof(struct nvsp_message
),
208 (unsigned long)revoke_packet
,
209 VM_PKT_DATA_INBAND
, 0);
210 /* If we failed here, we might as well return and
211 * have a leak rather than continue and a bugchk
214 netdev_err(ndev
, "unable to send "
215 "revoke send buffer to netvsp\n");
219 /* Teardown the gpadl on the vsp end */
220 if (net_device
->send_buf_gpadl_handle
) {
221 ret
= vmbus_teardown_gpadl(device
->channel
,
222 net_device
->send_buf_gpadl_handle
);
224 /* If we failed here, we might as well return and have a leak
225 * rather than continue and a bugchk
229 "unable to teardown send buffer's gpadl\n");
232 net_device
->send_buf_gpadl_handle
= 0;
234 if (net_device
->send_buf
) {
235 /* Free up the send buffer */
236 vfree(net_device
->send_buf
);
237 net_device
->send_buf
= NULL
;
239 kfree(net_device
->send_section_map
);
244 static int netvsc_init_buf(struct hv_device
*device
)
247 struct netvsc_device
*net_device
;
248 struct nvsp_message
*init_packet
;
249 struct net_device
*ndev
;
252 net_device
= get_outbound_net_device(device
);
255 ndev
= hv_get_drvdata(device
);
257 node
= cpu_to_node(device
->channel
->target_cpu
);
258 net_device
->recv_buf
= vzalloc_node(net_device
->recv_buf_size
, node
);
259 if (!net_device
->recv_buf
)
260 net_device
->recv_buf
= vzalloc(net_device
->recv_buf_size
);
262 if (!net_device
->recv_buf
) {
263 netdev_err(ndev
, "unable to allocate receive "
264 "buffer of size %d\n", net_device
->recv_buf_size
);
270 * Establish the gpadl handle for this buffer on this
271 * channel. Note: This call uses the vmbus connection rather
272 * than the channel to establish the gpadl handle.
274 ret
= vmbus_establish_gpadl(device
->channel
, net_device
->recv_buf
,
275 net_device
->recv_buf_size
,
276 &net_device
->recv_buf_gpadl_handle
);
279 "unable to establish receive buffer's gpadl\n");
284 /* Notify the NetVsp of the gpadl handle */
285 init_packet
= &net_device
->channel_init_pkt
;
287 memset(init_packet
, 0, sizeof(struct nvsp_message
));
289 init_packet
->hdr
.msg_type
= NVSP_MSG1_TYPE_SEND_RECV_BUF
;
290 init_packet
->msg
.v1_msg
.send_recv_buf
.
291 gpadl_handle
= net_device
->recv_buf_gpadl_handle
;
292 init_packet
->msg
.v1_msg
.
293 send_recv_buf
.id
= NETVSC_RECEIVE_BUFFER_ID
;
295 /* Send the gpadl notification request */
296 ret
= vmbus_sendpacket(device
->channel
, init_packet
,
297 sizeof(struct nvsp_message
),
298 (unsigned long)init_packet
,
300 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
);
303 "unable to send receive buffer's gpadl to netvsp\n");
307 wait_for_completion(&net_device
->channel_init_wait
);
309 /* Check the response */
310 if (init_packet
->msg
.v1_msg
.
311 send_recv_buf_complete
.status
!= NVSP_STAT_SUCCESS
) {
312 netdev_err(ndev
, "Unable to complete receive buffer "
313 "initialization with NetVsp - status %d\n",
314 init_packet
->msg
.v1_msg
.
315 send_recv_buf_complete
.status
);
320 /* Parse the response */
322 net_device
->recv_section_cnt
= init_packet
->msg
.
323 v1_msg
.send_recv_buf_complete
.num_sections
;
325 net_device
->recv_section
= kmemdup(
326 init_packet
->msg
.v1_msg
.send_recv_buf_complete
.sections
,
327 net_device
->recv_section_cnt
*
328 sizeof(struct nvsp_1_receive_buffer_section
),
330 if (net_device
->recv_section
== NULL
) {
336 * For 1st release, there should only be 1 section that represents the
337 * entire receive buffer
339 if (net_device
->recv_section_cnt
!= 1 ||
340 net_device
->recv_section
->offset
!= 0) {
345 /* Now setup the send buffer.
347 net_device
->send_buf
= vzalloc_node(net_device
->send_buf_size
, node
);
348 if (!net_device
->send_buf
)
349 net_device
->send_buf
= vzalloc(net_device
->send_buf_size
);
350 if (!net_device
->send_buf
) {
351 netdev_err(ndev
, "unable to allocate send "
352 "buffer of size %d\n", net_device
->send_buf_size
);
357 /* Establish the gpadl handle for this buffer on this
358 * channel. Note: This call uses the vmbus connection rather
359 * than the channel to establish the gpadl handle.
361 ret
= vmbus_establish_gpadl(device
->channel
, net_device
->send_buf
,
362 net_device
->send_buf_size
,
363 &net_device
->send_buf_gpadl_handle
);
366 "unable to establish send buffer's gpadl\n");
370 /* Notify the NetVsp of the gpadl handle */
371 init_packet
= &net_device
->channel_init_pkt
;
372 memset(init_packet
, 0, sizeof(struct nvsp_message
));
373 init_packet
->hdr
.msg_type
= NVSP_MSG1_TYPE_SEND_SEND_BUF
;
374 init_packet
->msg
.v1_msg
.send_send_buf
.gpadl_handle
=
375 net_device
->send_buf_gpadl_handle
;
376 init_packet
->msg
.v1_msg
.send_send_buf
.id
= NETVSC_SEND_BUFFER_ID
;
378 /* Send the gpadl notification request */
379 ret
= vmbus_sendpacket(device
->channel
, init_packet
,
380 sizeof(struct nvsp_message
),
381 (unsigned long)init_packet
,
383 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
);
386 "unable to send send buffer's gpadl to netvsp\n");
390 wait_for_completion(&net_device
->channel_init_wait
);
392 /* Check the response */
393 if (init_packet
->msg
.v1_msg
.
394 send_send_buf_complete
.status
!= NVSP_STAT_SUCCESS
) {
395 netdev_err(ndev
, "Unable to complete send buffer "
396 "initialization with NetVsp - status %d\n",
397 init_packet
->msg
.v1_msg
.
398 send_send_buf_complete
.status
);
403 /* Parse the response */
404 net_device
->send_section_size
= init_packet
->msg
.
405 v1_msg
.send_send_buf_complete
.section_size
;
407 /* Section count is simply the size divided by the section size.
409 net_device
->send_section_cnt
=
410 net_device
->send_buf_size
/net_device
->send_section_size
;
412 dev_info(&device
->device
, "Send section size: %d, Section count:%d\n",
413 net_device
->send_section_size
, net_device
->send_section_cnt
);
415 /* Setup state for managing the send buffer. */
416 net_device
->map_words
= DIV_ROUND_UP(net_device
->send_section_cnt
,
419 net_device
->send_section_map
=
420 kzalloc(net_device
->map_words
* sizeof(ulong
), GFP_KERNEL
);
421 if (net_device
->send_section_map
== NULL
) {
429 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 u32 ver_list
[] = { NVSP_PROTOCOL_VERSION_1
, NVSP_PROTOCOL_VERSION_2
,
497 NVSP_PROTOCOL_VERSION_4
, NVSP_PROTOCOL_VERSION_5
};
498 int i
, num_ver
= 4; /* number of different NVSP versions */
500 net_device
= get_outbound_net_device(device
);
504 init_packet
= &net_device
->channel_init_pkt
;
506 /* Negotiate the latest NVSP protocol supported */
507 for (i
= num_ver
- 1; i
>= 0; i
--)
508 if (negotiate_nvsp_ver(device
, net_device
, init_packet
,
510 net_device
->nvsp_version
= ver_list
[i
];
519 pr_debug("Negotiated NVSP version:%x\n", net_device
->nvsp_version
);
521 /* Send the ndis version */
522 memset(init_packet
, 0, sizeof(struct nvsp_message
));
524 if (net_device
->nvsp_version
<= NVSP_PROTOCOL_VERSION_4
)
525 ndis_version
= 0x00060001;
527 ndis_version
= 0x0006001e;
529 init_packet
->hdr
.msg_type
= NVSP_MSG1_TYPE_SEND_NDIS_VER
;
530 init_packet
->msg
.v1_msg
.
531 send_ndis_ver
.ndis_major_ver
=
532 (ndis_version
& 0xFFFF0000) >> 16;
533 init_packet
->msg
.v1_msg
.
534 send_ndis_ver
.ndis_minor_ver
=
535 ndis_version
& 0xFFFF;
537 /* Send the init request */
538 ret
= vmbus_sendpacket(device
->channel
, init_packet
,
539 sizeof(struct nvsp_message
),
540 (unsigned long)init_packet
,
541 VM_PKT_DATA_INBAND
, 0);
545 /* Post the big receive buffer to NetVSP */
546 if (net_device
->nvsp_version
<= NVSP_PROTOCOL_VERSION_2
)
547 net_device
->recv_buf_size
= NETVSC_RECEIVE_BUFFER_SIZE_LEGACY
;
549 net_device
->recv_buf_size
= NETVSC_RECEIVE_BUFFER_SIZE
;
550 net_device
->send_buf_size
= NETVSC_SEND_BUFFER_SIZE
;
552 ret
= netvsc_init_buf(device
);
558 static void netvsc_disconnect_vsp(struct hv_device
*device
)
560 netvsc_destroy_buf(device
);
564 * netvsc_device_remove - Callback when the root bus device is removed
566 int netvsc_device_remove(struct hv_device
*device
)
568 struct net_device
*ndev
= hv_get_drvdata(device
);
569 struct net_device_context
*net_device_ctx
= netdev_priv(ndev
);
570 struct netvsc_device
*net_device
= net_device_ctx
->nvdev
;
572 netvsc_disconnect_vsp(device
);
574 net_device_ctx
->nvdev
= NULL
;
577 * At this point, no one should be accessing net_device
580 dev_notice(&device
->device
, "net device safe to remove\n");
582 /* Now, we can close the channel safely */
583 vmbus_close(device
->channel
);
585 /* Release all resources */
586 vfree(net_device
->sub_cb_buf
);
587 free_netvsc_device(net_device
);
592 #define RING_AVAIL_PERCENT_HIWATER 20
593 #define RING_AVAIL_PERCENT_LOWATER 10
596 * Get the percentage of available bytes to write in the ring.
597 * The return value is in range from 0 to 100.
599 static inline u32
hv_ringbuf_avail_percent(
600 struct hv_ring_buffer_info
*ring_info
)
602 u32 avail_read
, avail_write
;
604 hv_get_ringbuffer_availbytes(ring_info
, &avail_read
, &avail_write
);
606 return avail_write
* 100 / ring_info
->ring_datasize
;
609 static inline void netvsc_free_send_slot(struct netvsc_device
*net_device
,
612 sync_change_bit(index
, net_device
->send_section_map
);
615 static void netvsc_send_completion(struct netvsc_device
*net_device
,
616 struct vmbus_channel
*incoming_channel
,
617 struct hv_device
*device
,
618 struct vmpacket_descriptor
*packet
)
620 struct nvsp_message
*nvsp_packet
;
621 struct hv_netvsc_packet
*nvsc_packet
;
622 struct net_device
*ndev
= hv_get_drvdata(device
);
623 struct net_device_context
*net_device_ctx
= netdev_priv(ndev
);
627 nvsp_packet
= (struct nvsp_message
*)((unsigned long)packet
+
628 (packet
->offset8
<< 3));
630 if ((nvsp_packet
->hdr
.msg_type
== NVSP_MSG_TYPE_INIT_COMPLETE
) ||
631 (nvsp_packet
->hdr
.msg_type
==
632 NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE
) ||
633 (nvsp_packet
->hdr
.msg_type
==
634 NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE
) ||
635 (nvsp_packet
->hdr
.msg_type
==
636 NVSP_MSG5_TYPE_SUBCHANNEL
)) {
637 /* Copy the response back */
638 memcpy(&net_device
->channel_init_pkt
, nvsp_packet
,
639 sizeof(struct nvsp_message
));
640 complete(&net_device
->channel_init_wait
);
641 } else if (nvsp_packet
->hdr
.msg_type
==
642 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE
) {
643 int num_outstanding_sends
;
645 struct vmbus_channel
*channel
= device
->channel
;
648 /* Get the send context */
649 skb
= (struct sk_buff
*)(unsigned long)packet
->trans_id
;
651 /* Notify the layer above us */
653 nvsc_packet
= (struct hv_netvsc_packet
*) skb
->cb
;
654 send_index
= nvsc_packet
->send_buf_index
;
655 if (send_index
!= NETVSC_INVALID_INDEX
)
656 netvsc_free_send_slot(net_device
, send_index
);
657 q_idx
= nvsc_packet
->q_idx
;
658 channel
= incoming_channel
;
659 dev_kfree_skb_any(skb
);
662 num_outstanding_sends
=
663 atomic_dec_return(&net_device
->num_outstanding_sends
);
664 queue_sends
= atomic_dec_return(&net_device
->
667 if (net_device
->destroy
&& num_outstanding_sends
== 0)
668 wake_up(&net_device
->wait_drain
);
670 if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev
, q_idx
)) &&
671 !net_device_ctx
->start_remove
&&
672 (hv_ringbuf_avail_percent(&channel
->outbound
) >
673 RING_AVAIL_PERCENT_HIWATER
|| queue_sends
< 1))
674 netif_tx_wake_queue(netdev_get_tx_queue(
677 netdev_err(ndev
, "Unknown send completion packet type- "
678 "%d received!!\n", nvsp_packet
->hdr
.msg_type
);
683 static u32
netvsc_get_next_send_section(struct netvsc_device
*net_device
)
686 u32 max_words
= net_device
->map_words
;
687 unsigned long *map_addr
= (unsigned long *)net_device
->send_section_map
;
688 u32 section_cnt
= net_device
->send_section_cnt
;
689 int ret_val
= NETVSC_INVALID_INDEX
;
693 for (i
= 0; i
< max_words
; i
++) {
696 index
= ffz(map_addr
[i
]);
697 prev_val
= sync_test_and_set_bit(index
, &map_addr
[i
]);
700 if ((index
+ (i
* BITS_PER_LONG
)) >= section_cnt
)
702 ret_val
= (index
+ (i
* BITS_PER_LONG
));
708 static u32
netvsc_copy_to_send_buf(struct netvsc_device
*net_device
,
709 unsigned int section_index
,
711 struct hv_netvsc_packet
*packet
,
712 struct rndis_message
*rndis_msg
,
713 struct hv_page_buffer
**pb
,
716 char *start
= net_device
->send_buf
;
717 char *dest
= start
+ (section_index
* net_device
->send_section_size
)
720 bool is_data_pkt
= (skb
!= NULL
) ? true : false;
721 bool xmit_more
= (skb
!= NULL
) ? skb
->xmit_more
: false;
724 u32 remain
= packet
->total_data_buflen
% net_device
->pkt_align
;
725 u32 page_count
= packet
->cp_partial
? packet
->rmsg_pgcnt
:
726 packet
->page_buf_cnt
;
729 if (is_data_pkt
&& xmit_more
&& remain
&&
730 !packet
->cp_partial
) {
731 padding
= net_device
->pkt_align
- remain
;
732 rndis_msg
->msg_len
+= padding
;
733 packet
->total_data_buflen
+= padding
;
736 for (i
= 0; i
< page_count
; i
++) {
737 char *src
= phys_to_virt((*pb
)[i
].pfn
<< PAGE_SHIFT
);
738 u32 offset
= (*pb
)[i
].offset
;
739 u32 len
= (*pb
)[i
].len
;
741 memcpy(dest
, (src
+ offset
), len
);
747 memset(dest
, 0, padding
);
754 static inline int netvsc_send_pkt(
755 struct hv_device
*device
,
756 struct hv_netvsc_packet
*packet
,
757 struct netvsc_device
*net_device
,
758 struct hv_page_buffer
**pb
,
761 struct nvsp_message nvmsg
;
762 u16 q_idx
= packet
->q_idx
;
763 struct vmbus_channel
*out_channel
= net_device
->chn_table
[q_idx
];
764 struct net_device
*ndev
= hv_get_drvdata(device
);
767 struct hv_page_buffer
*pgbuf
;
768 u32 ring_avail
= hv_ringbuf_avail_percent(&out_channel
->outbound
);
769 bool xmit_more
= (skb
!= NULL
) ? skb
->xmit_more
: false;
771 nvmsg
.hdr
.msg_type
= NVSP_MSG1_TYPE_SEND_RNDIS_PKT
;
774 nvmsg
.msg
.v1_msg
.send_rndis_pkt
.channel_type
= 0;
776 /* 1 is RMC_CONTROL; */
777 nvmsg
.msg
.v1_msg
.send_rndis_pkt
.channel_type
= 1;
780 nvmsg
.msg
.v1_msg
.send_rndis_pkt
.send_buf_section_index
=
781 packet
->send_buf_index
;
782 if (packet
->send_buf_index
== NETVSC_INVALID_INDEX
)
783 nvmsg
.msg
.v1_msg
.send_rndis_pkt
.send_buf_section_size
= 0;
785 nvmsg
.msg
.v1_msg
.send_rndis_pkt
.send_buf_section_size
=
786 packet
->total_data_buflen
;
790 if (out_channel
->rescind
)
794 * It is possible that once we successfully place this packet
795 * on the ringbuffer, we may stop the queue. In that case, we want
796 * to notify the host independent of the xmit_more flag. We don't
797 * need to be precise here; in the worst case we may signal the host
800 if (ring_avail
< (RING_AVAIL_PERCENT_LOWATER
+ 1))
803 if (packet
->page_buf_cnt
) {
804 pgbuf
= packet
->cp_partial
? (*pb
) +
805 packet
->rmsg_pgcnt
: (*pb
);
806 ret
= vmbus_sendpacket_pagebuffer_ctl(out_channel
,
808 packet
->page_buf_cnt
,
810 sizeof(struct nvsp_message
),
812 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
,
815 ret
= vmbus_sendpacket_ctl(out_channel
, &nvmsg
,
816 sizeof(struct nvsp_message
),
819 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
,
824 atomic_inc(&net_device
->num_outstanding_sends
);
825 atomic_inc(&net_device
->queue_sends
[q_idx
]);
827 if (ring_avail
< RING_AVAIL_PERCENT_LOWATER
) {
828 netif_tx_stop_queue(netdev_get_tx_queue(ndev
, q_idx
));
830 if (atomic_read(&net_device
->
831 queue_sends
[q_idx
]) < 1)
832 netif_tx_wake_queue(netdev_get_tx_queue(
835 } else if (ret
== -EAGAIN
) {
836 netif_tx_stop_queue(netdev_get_tx_queue(
838 if (atomic_read(&net_device
->queue_sends
[q_idx
]) < 1) {
839 netif_tx_wake_queue(netdev_get_tx_queue(
844 netdev_err(ndev
, "Unable to send packet %p ret %d\n",
851 /* Move packet out of multi send data (msd), and clear msd */
852 static inline void move_pkt_msd(struct hv_netvsc_packet
**msd_send
,
853 struct sk_buff
**msd_skb
,
854 struct multi_send_data
*msdp
)
856 *msd_skb
= msdp
->skb
;
857 *msd_send
= msdp
->pkt
;
863 int netvsc_send(struct hv_device
*device
,
864 struct hv_netvsc_packet
*packet
,
865 struct rndis_message
*rndis_msg
,
866 struct hv_page_buffer
**pb
,
869 struct netvsc_device
*net_device
;
870 int ret
= 0, m_ret
= 0;
871 struct vmbus_channel
*out_channel
;
872 u16 q_idx
= packet
->q_idx
;
873 u32 pktlen
= packet
->total_data_buflen
, msd_len
= 0;
874 unsigned int section_index
= NETVSC_INVALID_INDEX
;
875 struct multi_send_data
*msdp
;
876 struct hv_netvsc_packet
*msd_send
= NULL
, *cur_send
= NULL
;
877 struct sk_buff
*msd_skb
= NULL
;
879 bool xmit_more
= (skb
!= NULL
) ? skb
->xmit_more
: false;
881 net_device
= get_outbound_net_device(device
);
885 out_channel
= net_device
->chn_table
[q_idx
];
887 packet
->send_buf_index
= NETVSC_INVALID_INDEX
;
888 packet
->cp_partial
= false;
890 /* Send control message directly without accessing msd (Multi-Send
891 * Data) field which may be changed during data packet processing.
898 msdp
= &net_device
->msd
[q_idx
];
900 /* batch packets in send buffer if possible */
902 msd_len
= msdp
->pkt
->total_data_buflen
;
904 try_batch
= (skb
!= NULL
) && msd_len
> 0 && msdp
->count
<
907 if (try_batch
&& msd_len
+ pktlen
+ net_device
->pkt_align
<
908 net_device
->send_section_size
) {
909 section_index
= msdp
->pkt
->send_buf_index
;
911 } else if (try_batch
&& msd_len
+ packet
->rmsg_size
<
912 net_device
->send_section_size
) {
913 section_index
= msdp
->pkt
->send_buf_index
;
914 packet
->cp_partial
= true;
916 } else if ((skb
!= NULL
) && pktlen
+ net_device
->pkt_align
<
917 net_device
->send_section_size
) {
918 section_index
= netvsc_get_next_send_section(net_device
);
919 if (section_index
!= NETVSC_INVALID_INDEX
) {
920 move_pkt_msd(&msd_send
, &msd_skb
, msdp
);
925 if (section_index
!= NETVSC_INVALID_INDEX
) {
926 netvsc_copy_to_send_buf(net_device
,
927 section_index
, msd_len
,
928 packet
, rndis_msg
, pb
, skb
);
930 packet
->send_buf_index
= section_index
;
932 if (packet
->cp_partial
) {
933 packet
->page_buf_cnt
-= packet
->rmsg_pgcnt
;
934 packet
->total_data_buflen
= msd_len
+ packet
->rmsg_size
;
936 packet
->page_buf_cnt
= 0;
937 packet
->total_data_buflen
+= msd_len
;
941 dev_kfree_skb_any(msdp
->skb
);
943 if (xmit_more
&& !packet
->cp_partial
) {
954 move_pkt_msd(&msd_send
, &msd_skb
, msdp
);
959 m_ret
= netvsc_send_pkt(device
, msd_send
, net_device
,
963 netvsc_free_send_slot(net_device
,
964 msd_send
->send_buf_index
);
965 dev_kfree_skb_any(msd_skb
);
971 ret
= netvsc_send_pkt(device
, cur_send
, net_device
, pb
, skb
);
973 if (ret
!= 0 && section_index
!= NETVSC_INVALID_INDEX
)
974 netvsc_free_send_slot(net_device
, section_index
);
979 static void netvsc_send_recv_completion(struct hv_device
*device
,
980 struct vmbus_channel
*channel
,
981 struct netvsc_device
*net_device
,
982 u64 transaction_id
, u32 status
)
984 struct nvsp_message recvcompMessage
;
987 struct net_device
*ndev
= hv_get_drvdata(device
);
989 recvcompMessage
.hdr
.msg_type
=
990 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE
;
992 recvcompMessage
.msg
.v1_msg
.send_rndis_pkt_complete
.status
= status
;
995 /* Send the completion */
996 ret
= vmbus_sendpacket(channel
, &recvcompMessage
,
997 sizeof(struct nvsp_message
), transaction_id
,
1002 } else if (ret
== -EAGAIN
) {
1003 /* no more room...wait a bit and attempt to retry 3 times */
1005 netdev_err(ndev
, "unable to send receive completion pkt"
1006 " (tid %llx)...retrying %d\n", transaction_id
, retries
);
1010 goto retry_send_cmplt
;
1012 netdev_err(ndev
, "unable to send receive "
1013 "completion pkt (tid %llx)...give up retrying\n",
1017 netdev_err(ndev
, "unable to send receive "
1018 "completion pkt - %llx\n", transaction_id
);
1022 static void netvsc_receive(struct netvsc_device
*net_device
,
1023 struct vmbus_channel
*channel
,
1024 struct hv_device
*device
,
1025 struct vmpacket_descriptor
*packet
)
1027 struct vmtransfer_page_packet_header
*vmxferpage_packet
;
1028 struct nvsp_message
*nvsp_packet
;
1029 struct hv_netvsc_packet nv_pkt
;
1030 struct hv_netvsc_packet
*netvsc_packet
= &nv_pkt
;
1031 u32 status
= NVSP_STAT_SUCCESS
;
1034 struct net_device
*ndev
= hv_get_drvdata(device
);
1038 * All inbound packets other than send completion should be xfer page
1041 if (packet
->type
!= VM_PKT_DATA_USING_XFER_PAGES
) {
1042 netdev_err(ndev
, "Unknown packet type received - %d\n",
1047 nvsp_packet
= (struct nvsp_message
*)((unsigned long)packet
+
1048 (packet
->offset8
<< 3));
1050 /* Make sure this is a valid nvsp packet */
1051 if (nvsp_packet
->hdr
.msg_type
!=
1052 NVSP_MSG1_TYPE_SEND_RNDIS_PKT
) {
1053 netdev_err(ndev
, "Unknown nvsp packet type received-"
1054 " %d\n", nvsp_packet
->hdr
.msg_type
);
1058 vmxferpage_packet
= (struct vmtransfer_page_packet_header
*)packet
;
1060 if (vmxferpage_packet
->xfer_pageset_id
!= NETVSC_RECEIVE_BUFFER_ID
) {
1061 netdev_err(ndev
, "Invalid xfer page set id - "
1062 "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID
,
1063 vmxferpage_packet
->xfer_pageset_id
);
1067 count
= vmxferpage_packet
->range_cnt
;
1069 /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1070 for (i
= 0; i
< count
; i
++) {
1071 /* Initialize the netvsc packet */
1072 data
= (void *)((unsigned long)net_device
->
1073 recv_buf
+ vmxferpage_packet
->ranges
[i
].byte_offset
);
1074 netvsc_packet
->total_data_buflen
=
1075 vmxferpage_packet
->ranges
[i
].byte_count
;
1077 /* Pass it to the upper layer */
1078 status
= rndis_filter_receive(device
, netvsc_packet
, &data
,
1083 netvsc_send_recv_completion(device
, channel
, net_device
,
1084 vmxferpage_packet
->d
.trans_id
, status
);
1088 static void netvsc_send_table(struct hv_device
*hdev
,
1089 struct nvsp_message
*nvmsg
)
1091 struct netvsc_device
*nvscdev
;
1092 struct net_device
*ndev
= hv_get_drvdata(hdev
);
1096 nvscdev
= get_outbound_net_device(hdev
);
1100 count
= nvmsg
->msg
.v5_msg
.send_table
.count
;
1101 if (count
!= VRSS_SEND_TAB_SIZE
) {
1102 netdev_err(ndev
, "Received wrong send-table size:%u\n", count
);
1106 tab
= (u32
*)((unsigned long)&nvmsg
->msg
.v5_msg
.send_table
+
1107 nvmsg
->msg
.v5_msg
.send_table
.offset
);
1109 for (i
= 0; i
< count
; i
++)
1110 nvscdev
->send_table
[i
] = tab
[i
];
1113 static void netvsc_send_vf(struct netvsc_device
*nvdev
,
1114 struct nvsp_message
*nvmsg
)
1116 nvdev
->vf_alloc
= nvmsg
->msg
.v4_msg
.vf_assoc
.allocated
;
1117 nvdev
->vf_serial
= nvmsg
->msg
.v4_msg
.vf_assoc
.serial
;
1120 static inline void netvsc_receive_inband(struct hv_device
*hdev
,
1121 struct netvsc_device
*nvdev
,
1122 struct nvsp_message
*nvmsg
)
1124 switch (nvmsg
->hdr
.msg_type
) {
1125 case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE
:
1126 netvsc_send_table(hdev
, nvmsg
);
1129 case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION
:
1130 netvsc_send_vf(nvdev
, nvmsg
);
1135 static void netvsc_process_raw_pkt(struct hv_device
*device
,
1136 struct vmbus_channel
*channel
,
1137 struct netvsc_device
*net_device
,
1138 struct net_device
*ndev
,
1140 struct vmpacket_descriptor
*desc
)
1142 struct nvsp_message
*nvmsg
;
1144 nvmsg
= (struct nvsp_message
*)((unsigned long)
1145 desc
+ (desc
->offset8
<< 3));
1147 switch (desc
->type
) {
1149 netvsc_send_completion(net_device
, channel
, device
, desc
);
1152 case VM_PKT_DATA_USING_XFER_PAGES
:
1153 netvsc_receive(net_device
, channel
, device
, desc
);
1156 case VM_PKT_DATA_INBAND
:
1157 netvsc_receive_inband(device
, net_device
, nvmsg
);
1161 netdev_err(ndev
, "unhandled packet type %d, tid %llx\n",
1162 desc
->type
, request_id
);
1168 void netvsc_channel_cb(void *context
)
1171 struct vmbus_channel
*channel
= (struct vmbus_channel
*)context
;
1172 struct hv_device
*device
;
1173 struct netvsc_device
*net_device
;
1176 struct vmpacket_descriptor
*desc
;
1177 unsigned char *buffer
;
1178 int bufferlen
= NETVSC_PACKET_SIZE
;
1179 struct net_device
*ndev
;
1180 bool need_to_commit
= false;
1182 if (channel
->primary_channel
!= NULL
)
1183 device
= channel
->primary_channel
->device_obj
;
1185 device
= channel
->device_obj
;
1187 net_device
= get_inbound_net_device(device
);
1190 ndev
= hv_get_drvdata(device
);
1191 buffer
= get_per_channel_state(channel
);
1194 desc
= get_next_pkt_raw(channel
);
1196 netvsc_process_raw_pkt(device
,
1203 put_pkt_raw(channel
, desc
);
1204 need_to_commit
= true;
1207 if (need_to_commit
) {
1208 need_to_commit
= false;
1209 commit_rd_index(channel
);
1212 ret
= vmbus_recvpacket_raw(channel
, buffer
, bufferlen
,
1213 &bytes_recvd
, &request_id
);
1215 if (bytes_recvd
> 0) {
1216 desc
= (struct vmpacket_descriptor
*)buffer
;
1217 netvsc_process_raw_pkt(device
,
1227 * We are done for this pass.
1232 } else if (ret
== -ENOBUFS
) {
1233 if (bufferlen
> NETVSC_PACKET_SIZE
)
1235 /* Handle large packet */
1236 buffer
= kmalloc(bytes_recvd
, GFP_ATOMIC
);
1237 if (buffer
== NULL
) {
1238 /* Try again next time around */
1240 "unable to allocate buffer of size "
1241 "(%d)!!\n", bytes_recvd
);
1245 bufferlen
= bytes_recvd
;
1249 if (bufferlen
> NETVSC_PACKET_SIZE
)
1255 * netvsc_device_add - Callback when the device belonging to this
1258 int netvsc_device_add(struct hv_device
*device
, void *additional_info
)
1262 ((struct netvsc_device_info
*)additional_info
)->ring_size
;
1263 struct netvsc_device
*net_device
;
1264 struct net_device
*ndev
= hv_get_drvdata(device
);
1265 struct net_device_context
*net_device_ctx
= netdev_priv(ndev
);
1267 net_device
= alloc_net_device();
1271 net_device
->ring_size
= ring_size
;
1273 /* Initialize the NetVSC channel extension */
1274 init_completion(&net_device
->channel_init_wait
);
1276 set_per_channel_state(device
->channel
, net_device
->cb_buffer
);
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 pr_info("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
->chn_table
[i
] = 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
);