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 net_device
->max_pkt
= RNDIS_MAX_PKT_DEFAULT
;
81 net_device
->pkt_align
= RNDIS_PKT_ALIGN_DEFAULT
;
86 static void free_netvsc_device(struct netvsc_device
*nvdev
)
88 kfree(nvdev
->cb_buffer
);
92 static struct netvsc_device
*get_outbound_net_device(struct hv_device
*device
)
94 struct netvsc_device
*net_device
= hv_device_to_netvsc_device(device
);
96 if (net_device
&& net_device
->destroy
)
102 static struct netvsc_device
*get_inbound_net_device(struct hv_device
*device
)
104 struct netvsc_device
*net_device
= hv_device_to_netvsc_device(device
);
109 if (net_device
->destroy
&&
110 atomic_read(&net_device
->num_outstanding_sends
) == 0)
118 static int netvsc_destroy_buf(struct hv_device
*device
)
120 struct nvsp_message
*revoke_packet
;
122 struct net_device
*ndev
= hv_get_drvdata(device
);
123 struct netvsc_device
*net_device
= net_device_to_netvsc_device(ndev
);
126 * If we got a section count, it means we received a
127 * SendReceiveBufferComplete msg (ie sent
128 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
129 * to send a revoke msg here
131 if (net_device
->recv_section_cnt
) {
132 /* Send the revoke receive buffer */
133 revoke_packet
= &net_device
->revoke_packet
;
134 memset(revoke_packet
, 0, sizeof(struct nvsp_message
));
136 revoke_packet
->hdr
.msg_type
=
137 NVSP_MSG1_TYPE_REVOKE_RECV_BUF
;
138 revoke_packet
->msg
.v1_msg
.
139 revoke_recv_buf
.id
= NETVSC_RECEIVE_BUFFER_ID
;
141 ret
= vmbus_sendpacket(device
->channel
,
143 sizeof(struct nvsp_message
),
144 (unsigned long)revoke_packet
,
145 VM_PKT_DATA_INBAND
, 0);
147 * If we failed here, we might as well return and
148 * have a leak rather than continue and a bugchk
151 netdev_err(ndev
, "unable to send "
152 "revoke receive buffer to netvsp\n");
157 /* Teardown the gpadl on the vsp end */
158 if (net_device
->recv_buf_gpadl_handle
) {
159 ret
= vmbus_teardown_gpadl(device
->channel
,
160 net_device
->recv_buf_gpadl_handle
);
162 /* If we failed here, we might as well return and have a leak
163 * rather than continue and a bugchk
167 "unable to teardown receive buffer's gpadl\n");
170 net_device
->recv_buf_gpadl_handle
= 0;
173 if (net_device
->recv_buf
) {
174 /* Free up the receive buffer */
175 vfree(net_device
->recv_buf
);
176 net_device
->recv_buf
= NULL
;
179 if (net_device
->recv_section
) {
180 net_device
->recv_section_cnt
= 0;
181 kfree(net_device
->recv_section
);
182 net_device
->recv_section
= NULL
;
185 /* Deal with the send buffer we may have setup.
186 * If we got a send section size, it means we received a
187 * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
188 * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
189 * to send a revoke msg here
191 if (net_device
->send_section_size
) {
192 /* Send the revoke receive buffer */
193 revoke_packet
= &net_device
->revoke_packet
;
194 memset(revoke_packet
, 0, sizeof(struct nvsp_message
));
196 revoke_packet
->hdr
.msg_type
=
197 NVSP_MSG1_TYPE_REVOKE_SEND_BUF
;
198 revoke_packet
->msg
.v1_msg
.revoke_send_buf
.id
=
199 NETVSC_SEND_BUFFER_ID
;
201 ret
= vmbus_sendpacket(device
->channel
,
203 sizeof(struct nvsp_message
),
204 (unsigned long)revoke_packet
,
205 VM_PKT_DATA_INBAND
, 0);
206 /* If we failed here, we might as well return and
207 * have a leak rather than continue and a bugchk
210 netdev_err(ndev
, "unable to send "
211 "revoke send buffer to netvsp\n");
215 /* Teardown the gpadl on the vsp end */
216 if (net_device
->send_buf_gpadl_handle
) {
217 ret
= vmbus_teardown_gpadl(device
->channel
,
218 net_device
->send_buf_gpadl_handle
);
220 /* If we failed here, we might as well return and have a leak
221 * rather than continue and a bugchk
225 "unable to teardown send buffer's gpadl\n");
228 net_device
->send_buf_gpadl_handle
= 0;
230 if (net_device
->send_buf
) {
231 /* Free up the send buffer */
232 vfree(net_device
->send_buf
);
233 net_device
->send_buf
= NULL
;
235 kfree(net_device
->send_section_map
);
240 static int netvsc_init_buf(struct hv_device
*device
)
243 struct netvsc_device
*net_device
;
244 struct nvsp_message
*init_packet
;
245 struct net_device
*ndev
;
248 net_device
= get_outbound_net_device(device
);
251 ndev
= hv_get_drvdata(device
);
253 node
= cpu_to_node(device
->channel
->target_cpu
);
254 net_device
->recv_buf
= vzalloc_node(net_device
->recv_buf_size
, node
);
255 if (!net_device
->recv_buf
)
256 net_device
->recv_buf
= vzalloc(net_device
->recv_buf_size
);
258 if (!net_device
->recv_buf
) {
259 netdev_err(ndev
, "unable to allocate receive "
260 "buffer of size %d\n", net_device
->recv_buf_size
);
266 * Establish the gpadl handle for this buffer on this
267 * channel. Note: This call uses the vmbus connection rather
268 * than the channel to establish the gpadl handle.
270 ret
= vmbus_establish_gpadl(device
->channel
, net_device
->recv_buf
,
271 net_device
->recv_buf_size
,
272 &net_device
->recv_buf_gpadl_handle
);
275 "unable to establish receive buffer's gpadl\n");
280 /* Notify the NetVsp of the gpadl handle */
281 init_packet
= &net_device
->channel_init_pkt
;
283 memset(init_packet
, 0, sizeof(struct nvsp_message
));
285 init_packet
->hdr
.msg_type
= NVSP_MSG1_TYPE_SEND_RECV_BUF
;
286 init_packet
->msg
.v1_msg
.send_recv_buf
.
287 gpadl_handle
= net_device
->recv_buf_gpadl_handle
;
288 init_packet
->msg
.v1_msg
.
289 send_recv_buf
.id
= NETVSC_RECEIVE_BUFFER_ID
;
291 /* Send the gpadl notification request */
292 ret
= vmbus_sendpacket(device
->channel
, init_packet
,
293 sizeof(struct nvsp_message
),
294 (unsigned long)init_packet
,
296 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
);
299 "unable to send receive buffer's gpadl to netvsp\n");
303 wait_for_completion(&net_device
->channel_init_wait
);
305 /* Check the response */
306 if (init_packet
->msg
.v1_msg
.
307 send_recv_buf_complete
.status
!= NVSP_STAT_SUCCESS
) {
308 netdev_err(ndev
, "Unable to complete receive buffer "
309 "initialization with NetVsp - status %d\n",
310 init_packet
->msg
.v1_msg
.
311 send_recv_buf_complete
.status
);
316 /* Parse the response */
318 net_device
->recv_section_cnt
= init_packet
->msg
.
319 v1_msg
.send_recv_buf_complete
.num_sections
;
321 net_device
->recv_section
= kmemdup(
322 init_packet
->msg
.v1_msg
.send_recv_buf_complete
.sections
,
323 net_device
->recv_section_cnt
*
324 sizeof(struct nvsp_1_receive_buffer_section
),
326 if (net_device
->recv_section
== NULL
) {
332 * For 1st release, there should only be 1 section that represents the
333 * entire receive buffer
335 if (net_device
->recv_section_cnt
!= 1 ||
336 net_device
->recv_section
->offset
!= 0) {
341 /* Now setup the send buffer.
343 net_device
->send_buf
= vzalloc_node(net_device
->send_buf_size
, node
);
344 if (!net_device
->send_buf
)
345 net_device
->send_buf
= vzalloc(net_device
->send_buf_size
);
346 if (!net_device
->send_buf
) {
347 netdev_err(ndev
, "unable to allocate send "
348 "buffer of size %d\n", net_device
->send_buf_size
);
353 /* Establish the gpadl handle for this buffer on this
354 * channel. Note: This call uses the vmbus connection rather
355 * than the channel to establish the gpadl handle.
357 ret
= vmbus_establish_gpadl(device
->channel
, net_device
->send_buf
,
358 net_device
->send_buf_size
,
359 &net_device
->send_buf_gpadl_handle
);
362 "unable to establish send buffer's gpadl\n");
366 /* Notify the NetVsp of the gpadl handle */
367 init_packet
= &net_device
->channel_init_pkt
;
368 memset(init_packet
, 0, sizeof(struct nvsp_message
));
369 init_packet
->hdr
.msg_type
= NVSP_MSG1_TYPE_SEND_SEND_BUF
;
370 init_packet
->msg
.v1_msg
.send_send_buf
.gpadl_handle
=
371 net_device
->send_buf_gpadl_handle
;
372 init_packet
->msg
.v1_msg
.send_send_buf
.id
= NETVSC_SEND_BUFFER_ID
;
374 /* Send the gpadl notification request */
375 ret
= vmbus_sendpacket(device
->channel
, init_packet
,
376 sizeof(struct nvsp_message
),
377 (unsigned long)init_packet
,
379 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
);
382 "unable to send send buffer's gpadl to netvsp\n");
386 wait_for_completion(&net_device
->channel_init_wait
);
388 /* Check the response */
389 if (init_packet
->msg
.v1_msg
.
390 send_send_buf_complete
.status
!= NVSP_STAT_SUCCESS
) {
391 netdev_err(ndev
, "Unable to complete send buffer "
392 "initialization with NetVsp - status %d\n",
393 init_packet
->msg
.v1_msg
.
394 send_send_buf_complete
.status
);
399 /* Parse the response */
400 net_device
->send_section_size
= init_packet
->msg
.
401 v1_msg
.send_send_buf_complete
.section_size
;
403 /* Section count is simply the size divided by the section size.
405 net_device
->send_section_cnt
=
406 net_device
->send_buf_size
/net_device
->send_section_size
;
408 dev_info(&device
->device
, "Send section size: %d, Section count:%d\n",
409 net_device
->send_section_size
, net_device
->send_section_cnt
);
411 /* Setup state for managing the send buffer. */
412 net_device
->map_words
= DIV_ROUND_UP(net_device
->send_section_cnt
,
415 net_device
->send_section_map
=
416 kzalloc(net_device
->map_words
* sizeof(ulong
), GFP_KERNEL
);
417 if (net_device
->send_section_map
== NULL
) {
425 netvsc_destroy_buf(device
);
432 /* Negotiate NVSP protocol version */
433 static int negotiate_nvsp_ver(struct hv_device
*device
,
434 struct netvsc_device
*net_device
,
435 struct nvsp_message
*init_packet
,
438 struct net_device
*ndev
= hv_get_drvdata(device
);
441 memset(init_packet
, 0, sizeof(struct nvsp_message
));
442 init_packet
->hdr
.msg_type
= NVSP_MSG_TYPE_INIT
;
443 init_packet
->msg
.init_msg
.init
.min_protocol_ver
= nvsp_ver
;
444 init_packet
->msg
.init_msg
.init
.max_protocol_ver
= nvsp_ver
;
446 /* Send the init request */
447 ret
= vmbus_sendpacket(device
->channel
, init_packet
,
448 sizeof(struct nvsp_message
),
449 (unsigned long)init_packet
,
451 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
);
456 wait_for_completion(&net_device
->channel_init_wait
);
458 if (init_packet
->msg
.init_msg
.init_complete
.status
!=
462 if (nvsp_ver
== NVSP_PROTOCOL_VERSION_1
)
465 /* NVSPv2 or later: Send NDIS config */
466 memset(init_packet
, 0, sizeof(struct nvsp_message
));
467 init_packet
->hdr
.msg_type
= NVSP_MSG2_TYPE_SEND_NDIS_CONFIG
;
468 init_packet
->msg
.v2_msg
.send_ndis_config
.mtu
= ndev
->mtu
+ ETH_HLEN
;
469 init_packet
->msg
.v2_msg
.send_ndis_config
.capability
.ieee8021q
= 1;
471 if (nvsp_ver
>= NVSP_PROTOCOL_VERSION_5
)
472 init_packet
->msg
.v2_msg
.send_ndis_config
.capability
.sriov
= 1;
474 ret
= vmbus_sendpacket(device
->channel
, init_packet
,
475 sizeof(struct nvsp_message
),
476 (unsigned long)init_packet
,
477 VM_PKT_DATA_INBAND
, 0);
482 static int netvsc_connect_vsp(struct hv_device
*device
)
485 struct netvsc_device
*net_device
;
486 struct nvsp_message
*init_packet
;
488 u32 ver_list
[] = { NVSP_PROTOCOL_VERSION_1
, NVSP_PROTOCOL_VERSION_2
,
489 NVSP_PROTOCOL_VERSION_4
, NVSP_PROTOCOL_VERSION_5
};
490 int i
, num_ver
= 4; /* number of different NVSP versions */
492 net_device
= get_outbound_net_device(device
);
496 init_packet
= &net_device
->channel_init_pkt
;
498 /* Negotiate the latest NVSP protocol supported */
499 for (i
= num_ver
- 1; i
>= 0; i
--)
500 if (negotiate_nvsp_ver(device
, net_device
, init_packet
,
502 net_device
->nvsp_version
= ver_list
[i
];
511 pr_debug("Negotiated NVSP version:%x\n", net_device
->nvsp_version
);
513 /* Send the ndis version */
514 memset(init_packet
, 0, sizeof(struct nvsp_message
));
516 if (net_device
->nvsp_version
<= NVSP_PROTOCOL_VERSION_4
)
517 ndis_version
= 0x00060001;
519 ndis_version
= 0x0006001e;
521 init_packet
->hdr
.msg_type
= NVSP_MSG1_TYPE_SEND_NDIS_VER
;
522 init_packet
->msg
.v1_msg
.
523 send_ndis_ver
.ndis_major_ver
=
524 (ndis_version
& 0xFFFF0000) >> 16;
525 init_packet
->msg
.v1_msg
.
526 send_ndis_ver
.ndis_minor_ver
=
527 ndis_version
& 0xFFFF;
529 /* Send the init request */
530 ret
= vmbus_sendpacket(device
->channel
, init_packet
,
531 sizeof(struct nvsp_message
),
532 (unsigned long)init_packet
,
533 VM_PKT_DATA_INBAND
, 0);
537 /* Post the big receive buffer to NetVSP */
538 if (net_device
->nvsp_version
<= NVSP_PROTOCOL_VERSION_2
)
539 net_device
->recv_buf_size
= NETVSC_RECEIVE_BUFFER_SIZE_LEGACY
;
541 net_device
->recv_buf_size
= NETVSC_RECEIVE_BUFFER_SIZE
;
542 net_device
->send_buf_size
= NETVSC_SEND_BUFFER_SIZE
;
544 ret
= netvsc_init_buf(device
);
550 static void netvsc_disconnect_vsp(struct hv_device
*device
)
552 netvsc_destroy_buf(device
);
556 * netvsc_device_remove - Callback when the root bus device is removed
558 int netvsc_device_remove(struct hv_device
*device
)
560 struct net_device
*ndev
= hv_get_drvdata(device
);
561 struct net_device_context
*net_device_ctx
= netdev_priv(ndev
);
562 struct netvsc_device
*net_device
= net_device_ctx
->nvdev
;
564 netvsc_disconnect_vsp(device
);
566 net_device_ctx
->nvdev
= NULL
;
569 * At this point, no one should be accessing net_device
572 dev_notice(&device
->device
, "net device safe to remove\n");
574 /* Now, we can close the channel safely */
575 vmbus_close(device
->channel
);
577 /* Release all resources */
578 vfree(net_device
->sub_cb_buf
);
579 free_netvsc_device(net_device
);
584 #define RING_AVAIL_PERCENT_HIWATER 20
585 #define RING_AVAIL_PERCENT_LOWATER 10
588 * Get the percentage of available bytes to write in the ring.
589 * The return value is in range from 0 to 100.
591 static inline u32
hv_ringbuf_avail_percent(
592 struct hv_ring_buffer_info
*ring_info
)
594 u32 avail_read
, avail_write
;
596 hv_get_ringbuffer_availbytes(ring_info
, &avail_read
, &avail_write
);
598 return avail_write
* 100 / ring_info
->ring_datasize
;
601 static inline void netvsc_free_send_slot(struct netvsc_device
*net_device
,
604 sync_change_bit(index
, net_device
->send_section_map
);
607 static void netvsc_send_completion(struct netvsc_device
*net_device
,
608 struct vmbus_channel
*incoming_channel
,
609 struct hv_device
*device
,
610 struct vmpacket_descriptor
*packet
)
612 struct nvsp_message
*nvsp_packet
;
613 struct hv_netvsc_packet
*nvsc_packet
;
614 struct net_device
*ndev
= hv_get_drvdata(device
);
615 struct net_device_context
*net_device_ctx
= netdev_priv(ndev
);
619 nvsp_packet
= (struct nvsp_message
*)((unsigned long)packet
+
620 (packet
->offset8
<< 3));
622 if ((nvsp_packet
->hdr
.msg_type
== NVSP_MSG_TYPE_INIT_COMPLETE
) ||
623 (nvsp_packet
->hdr
.msg_type
==
624 NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE
) ||
625 (nvsp_packet
->hdr
.msg_type
==
626 NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE
) ||
627 (nvsp_packet
->hdr
.msg_type
==
628 NVSP_MSG5_TYPE_SUBCHANNEL
)) {
629 /* Copy the response back */
630 memcpy(&net_device
->channel_init_pkt
, nvsp_packet
,
631 sizeof(struct nvsp_message
));
632 complete(&net_device
->channel_init_wait
);
633 } else if (nvsp_packet
->hdr
.msg_type
==
634 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE
) {
635 int num_outstanding_sends
;
637 struct vmbus_channel
*channel
= device
->channel
;
640 /* Get the send context */
641 skb
= (struct sk_buff
*)(unsigned long)packet
->trans_id
;
643 /* Notify the layer above us */
645 nvsc_packet
= (struct hv_netvsc_packet
*) skb
->cb
;
646 send_index
= nvsc_packet
->send_buf_index
;
647 if (send_index
!= NETVSC_INVALID_INDEX
)
648 netvsc_free_send_slot(net_device
, send_index
);
649 q_idx
= nvsc_packet
->q_idx
;
650 channel
= incoming_channel
;
651 dev_kfree_skb_any(skb
);
654 num_outstanding_sends
=
655 atomic_dec_return(&net_device
->num_outstanding_sends
);
656 queue_sends
= atomic_dec_return(&net_device
->
659 if (net_device
->destroy
&& num_outstanding_sends
== 0)
660 wake_up(&net_device
->wait_drain
);
662 if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev
, q_idx
)) &&
663 !net_device_ctx
->start_remove
&&
664 (hv_ringbuf_avail_percent(&channel
->outbound
) >
665 RING_AVAIL_PERCENT_HIWATER
|| queue_sends
< 1))
666 netif_tx_wake_queue(netdev_get_tx_queue(
669 netdev_err(ndev
, "Unknown send completion packet type- "
670 "%d received!!\n", nvsp_packet
->hdr
.msg_type
);
675 static u32
netvsc_get_next_send_section(struct netvsc_device
*net_device
)
678 u32 max_words
= net_device
->map_words
;
679 unsigned long *map_addr
= (unsigned long *)net_device
->send_section_map
;
680 u32 section_cnt
= net_device
->send_section_cnt
;
681 int ret_val
= NETVSC_INVALID_INDEX
;
685 for (i
= 0; i
< max_words
; i
++) {
688 index
= ffz(map_addr
[i
]);
689 prev_val
= sync_test_and_set_bit(index
, &map_addr
[i
]);
692 if ((index
+ (i
* BITS_PER_LONG
)) >= section_cnt
)
694 ret_val
= (index
+ (i
* BITS_PER_LONG
));
700 static u32
netvsc_copy_to_send_buf(struct netvsc_device
*net_device
,
701 unsigned int section_index
,
703 struct hv_netvsc_packet
*packet
,
704 struct rndis_message
*rndis_msg
,
705 struct hv_page_buffer
**pb
,
708 char *start
= net_device
->send_buf
;
709 char *dest
= start
+ (section_index
* net_device
->send_section_size
)
712 bool is_data_pkt
= (skb
!= NULL
) ? true : false;
713 bool xmit_more
= (skb
!= NULL
) ? skb
->xmit_more
: false;
716 u32 remain
= packet
->total_data_buflen
% net_device
->pkt_align
;
717 u32 page_count
= packet
->cp_partial
? packet
->rmsg_pgcnt
:
718 packet
->page_buf_cnt
;
721 if (is_data_pkt
&& xmit_more
&& remain
&&
722 !packet
->cp_partial
) {
723 padding
= net_device
->pkt_align
- remain
;
724 rndis_msg
->msg_len
+= padding
;
725 packet
->total_data_buflen
+= padding
;
728 for (i
= 0; i
< page_count
; i
++) {
729 char *src
= phys_to_virt((*pb
)[i
].pfn
<< PAGE_SHIFT
);
730 u32 offset
= (*pb
)[i
].offset
;
731 u32 len
= (*pb
)[i
].len
;
733 memcpy(dest
, (src
+ offset
), len
);
739 memset(dest
, 0, padding
);
746 static inline int netvsc_send_pkt(
747 struct hv_device
*device
,
748 struct hv_netvsc_packet
*packet
,
749 struct netvsc_device
*net_device
,
750 struct hv_page_buffer
**pb
,
753 struct nvsp_message nvmsg
;
754 u16 q_idx
= packet
->q_idx
;
755 struct vmbus_channel
*out_channel
= net_device
->chn_table
[q_idx
];
756 struct net_device
*ndev
= hv_get_drvdata(device
);
759 struct hv_page_buffer
*pgbuf
;
760 u32 ring_avail
= hv_ringbuf_avail_percent(&out_channel
->outbound
);
761 bool xmit_more
= (skb
!= NULL
) ? skb
->xmit_more
: false;
763 nvmsg
.hdr
.msg_type
= NVSP_MSG1_TYPE_SEND_RNDIS_PKT
;
766 nvmsg
.msg
.v1_msg
.send_rndis_pkt
.channel_type
= 0;
768 /* 1 is RMC_CONTROL; */
769 nvmsg
.msg
.v1_msg
.send_rndis_pkt
.channel_type
= 1;
772 nvmsg
.msg
.v1_msg
.send_rndis_pkt
.send_buf_section_index
=
773 packet
->send_buf_index
;
774 if (packet
->send_buf_index
== NETVSC_INVALID_INDEX
)
775 nvmsg
.msg
.v1_msg
.send_rndis_pkt
.send_buf_section_size
= 0;
777 nvmsg
.msg
.v1_msg
.send_rndis_pkt
.send_buf_section_size
=
778 packet
->total_data_buflen
;
782 if (out_channel
->rescind
)
786 * It is possible that once we successfully place this packet
787 * on the ringbuffer, we may stop the queue. In that case, we want
788 * to notify the host independent of the xmit_more flag. We don't
789 * need to be precise here; in the worst case we may signal the host
792 if (ring_avail
< (RING_AVAIL_PERCENT_LOWATER
+ 1))
795 if (packet
->page_buf_cnt
) {
796 pgbuf
= packet
->cp_partial
? (*pb
) +
797 packet
->rmsg_pgcnt
: (*pb
);
798 ret
= vmbus_sendpacket_pagebuffer_ctl(out_channel
,
800 packet
->page_buf_cnt
,
802 sizeof(struct nvsp_message
),
804 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
,
807 ret
= vmbus_sendpacket_ctl(out_channel
, &nvmsg
,
808 sizeof(struct nvsp_message
),
811 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
,
816 atomic_inc(&net_device
->num_outstanding_sends
);
817 atomic_inc(&net_device
->queue_sends
[q_idx
]);
819 if (ring_avail
< RING_AVAIL_PERCENT_LOWATER
) {
820 netif_tx_stop_queue(netdev_get_tx_queue(ndev
, q_idx
));
822 if (atomic_read(&net_device
->
823 queue_sends
[q_idx
]) < 1)
824 netif_tx_wake_queue(netdev_get_tx_queue(
827 } else if (ret
== -EAGAIN
) {
828 netif_tx_stop_queue(netdev_get_tx_queue(
830 if (atomic_read(&net_device
->queue_sends
[q_idx
]) < 1) {
831 netif_tx_wake_queue(netdev_get_tx_queue(
836 netdev_err(ndev
, "Unable to send packet %p ret %d\n",
843 /* Move packet out of multi send data (msd), and clear msd */
844 static inline void move_pkt_msd(struct hv_netvsc_packet
**msd_send
,
845 struct sk_buff
**msd_skb
,
846 struct multi_send_data
*msdp
)
848 *msd_skb
= msdp
->skb
;
849 *msd_send
= msdp
->pkt
;
855 int netvsc_send(struct hv_device
*device
,
856 struct hv_netvsc_packet
*packet
,
857 struct rndis_message
*rndis_msg
,
858 struct hv_page_buffer
**pb
,
861 struct netvsc_device
*net_device
;
862 int ret
= 0, m_ret
= 0;
863 struct vmbus_channel
*out_channel
;
864 u16 q_idx
= packet
->q_idx
;
865 u32 pktlen
= packet
->total_data_buflen
, msd_len
= 0;
866 unsigned int section_index
= NETVSC_INVALID_INDEX
;
867 struct multi_send_data
*msdp
;
868 struct hv_netvsc_packet
*msd_send
= NULL
, *cur_send
= NULL
;
869 struct sk_buff
*msd_skb
= NULL
;
871 bool xmit_more
= (skb
!= NULL
) ? skb
->xmit_more
: false;
873 net_device
= get_outbound_net_device(device
);
877 out_channel
= net_device
->chn_table
[q_idx
];
879 packet
->send_buf_index
= NETVSC_INVALID_INDEX
;
880 packet
->cp_partial
= false;
882 /* Send control message directly without accessing msd (Multi-Send
883 * Data) field which may be changed during data packet processing.
890 msdp
= &net_device
->msd
[q_idx
];
892 /* batch packets in send buffer if possible */
894 msd_len
= msdp
->pkt
->total_data_buflen
;
896 try_batch
= (skb
!= NULL
) && msd_len
> 0 && msdp
->count
<
899 if (try_batch
&& msd_len
+ pktlen
+ net_device
->pkt_align
<
900 net_device
->send_section_size
) {
901 section_index
= msdp
->pkt
->send_buf_index
;
903 } else if (try_batch
&& msd_len
+ packet
->rmsg_size
<
904 net_device
->send_section_size
) {
905 section_index
= msdp
->pkt
->send_buf_index
;
906 packet
->cp_partial
= true;
908 } else if ((skb
!= NULL
) && pktlen
+ net_device
->pkt_align
<
909 net_device
->send_section_size
) {
910 section_index
= netvsc_get_next_send_section(net_device
);
911 if (section_index
!= NETVSC_INVALID_INDEX
) {
912 move_pkt_msd(&msd_send
, &msd_skb
, msdp
);
917 if (section_index
!= NETVSC_INVALID_INDEX
) {
918 netvsc_copy_to_send_buf(net_device
,
919 section_index
, msd_len
,
920 packet
, rndis_msg
, pb
, skb
);
922 packet
->send_buf_index
= section_index
;
924 if (packet
->cp_partial
) {
925 packet
->page_buf_cnt
-= packet
->rmsg_pgcnt
;
926 packet
->total_data_buflen
= msd_len
+ packet
->rmsg_size
;
928 packet
->page_buf_cnt
= 0;
929 packet
->total_data_buflen
+= msd_len
;
933 dev_kfree_skb_any(msdp
->skb
);
935 if (xmit_more
&& !packet
->cp_partial
) {
946 move_pkt_msd(&msd_send
, &msd_skb
, msdp
);
951 m_ret
= netvsc_send_pkt(device
, msd_send
, net_device
,
955 netvsc_free_send_slot(net_device
,
956 msd_send
->send_buf_index
);
957 dev_kfree_skb_any(msd_skb
);
963 ret
= netvsc_send_pkt(device
, cur_send
, net_device
, pb
, skb
);
965 if (ret
!= 0 && section_index
!= NETVSC_INVALID_INDEX
)
966 netvsc_free_send_slot(net_device
, section_index
);
971 static void netvsc_send_recv_completion(struct hv_device
*device
,
972 struct vmbus_channel
*channel
,
973 struct netvsc_device
*net_device
,
974 u64 transaction_id
, u32 status
)
976 struct nvsp_message recvcompMessage
;
979 struct net_device
*ndev
= hv_get_drvdata(device
);
981 recvcompMessage
.hdr
.msg_type
=
982 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE
;
984 recvcompMessage
.msg
.v1_msg
.send_rndis_pkt_complete
.status
= status
;
987 /* Send the completion */
988 ret
= vmbus_sendpacket(channel
, &recvcompMessage
,
989 sizeof(struct nvsp_message
), transaction_id
,
994 } else if (ret
== -EAGAIN
) {
995 /* no more room...wait a bit and attempt to retry 3 times */
997 netdev_err(ndev
, "unable to send receive completion pkt"
998 " (tid %llx)...retrying %d\n", transaction_id
, retries
);
1002 goto retry_send_cmplt
;
1004 netdev_err(ndev
, "unable to send receive "
1005 "completion pkt (tid %llx)...give up retrying\n",
1009 netdev_err(ndev
, "unable to send receive "
1010 "completion pkt - %llx\n", transaction_id
);
1014 static void netvsc_receive(struct netvsc_device
*net_device
,
1015 struct vmbus_channel
*channel
,
1016 struct hv_device
*device
,
1017 struct vmpacket_descriptor
*packet
)
1019 struct vmtransfer_page_packet_header
*vmxferpage_packet
;
1020 struct nvsp_message
*nvsp_packet
;
1021 struct hv_netvsc_packet nv_pkt
;
1022 struct hv_netvsc_packet
*netvsc_packet
= &nv_pkt
;
1023 u32 status
= NVSP_STAT_SUCCESS
;
1026 struct net_device
*ndev
= hv_get_drvdata(device
);
1030 * All inbound packets other than send completion should be xfer page
1033 if (packet
->type
!= VM_PKT_DATA_USING_XFER_PAGES
) {
1034 netdev_err(ndev
, "Unknown packet type received - %d\n",
1039 nvsp_packet
= (struct nvsp_message
*)((unsigned long)packet
+
1040 (packet
->offset8
<< 3));
1042 /* Make sure this is a valid nvsp packet */
1043 if (nvsp_packet
->hdr
.msg_type
!=
1044 NVSP_MSG1_TYPE_SEND_RNDIS_PKT
) {
1045 netdev_err(ndev
, "Unknown nvsp packet type received-"
1046 " %d\n", nvsp_packet
->hdr
.msg_type
);
1050 vmxferpage_packet
= (struct vmtransfer_page_packet_header
*)packet
;
1052 if (vmxferpage_packet
->xfer_pageset_id
!= NETVSC_RECEIVE_BUFFER_ID
) {
1053 netdev_err(ndev
, "Invalid xfer page set id - "
1054 "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID
,
1055 vmxferpage_packet
->xfer_pageset_id
);
1059 count
= vmxferpage_packet
->range_cnt
;
1061 /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1062 for (i
= 0; i
< count
; i
++) {
1063 /* Initialize the netvsc packet */
1064 data
= (void *)((unsigned long)net_device
->
1065 recv_buf
+ vmxferpage_packet
->ranges
[i
].byte_offset
);
1066 netvsc_packet
->total_data_buflen
=
1067 vmxferpage_packet
->ranges
[i
].byte_count
;
1069 /* Pass it to the upper layer */
1070 status
= rndis_filter_receive(device
, netvsc_packet
, &data
,
1075 netvsc_send_recv_completion(device
, channel
, net_device
,
1076 vmxferpage_packet
->d
.trans_id
, status
);
1080 static void netvsc_send_table(struct hv_device
*hdev
,
1081 struct nvsp_message
*nvmsg
)
1083 struct netvsc_device
*nvscdev
;
1084 struct net_device
*ndev
= hv_get_drvdata(hdev
);
1088 nvscdev
= get_outbound_net_device(hdev
);
1092 count
= nvmsg
->msg
.v5_msg
.send_table
.count
;
1093 if (count
!= VRSS_SEND_TAB_SIZE
) {
1094 netdev_err(ndev
, "Received wrong send-table size:%u\n", count
);
1098 tab
= (u32
*)((unsigned long)&nvmsg
->msg
.v5_msg
.send_table
+
1099 nvmsg
->msg
.v5_msg
.send_table
.offset
);
1101 for (i
= 0; i
< count
; i
++)
1102 nvscdev
->send_table
[i
] = tab
[i
];
1105 static void netvsc_send_vf(struct net_device_context
*net_device_ctx
,
1106 struct nvsp_message
*nvmsg
)
1108 net_device_ctx
->vf_alloc
= nvmsg
->msg
.v4_msg
.vf_assoc
.allocated
;
1109 net_device_ctx
->vf_serial
= nvmsg
->msg
.v4_msg
.vf_assoc
.serial
;
1112 static inline void netvsc_receive_inband(struct hv_device
*hdev
,
1113 struct net_device_context
*net_device_ctx
,
1114 struct nvsp_message
*nvmsg
)
1116 switch (nvmsg
->hdr
.msg_type
) {
1117 case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE
:
1118 netvsc_send_table(hdev
, nvmsg
);
1121 case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION
:
1122 netvsc_send_vf(net_device_ctx
, nvmsg
);
1127 static void netvsc_process_raw_pkt(struct hv_device
*device
,
1128 struct vmbus_channel
*channel
,
1129 struct netvsc_device
*net_device
,
1130 struct net_device
*ndev
,
1132 struct vmpacket_descriptor
*desc
)
1134 struct nvsp_message
*nvmsg
;
1135 struct net_device_context
*net_device_ctx
= netdev_priv(ndev
);
1137 nvmsg
= (struct nvsp_message
*)((unsigned long)
1138 desc
+ (desc
->offset8
<< 3));
1140 switch (desc
->type
) {
1142 netvsc_send_completion(net_device
, channel
, device
, desc
);
1145 case VM_PKT_DATA_USING_XFER_PAGES
:
1146 netvsc_receive(net_device
, channel
, device
, desc
);
1149 case VM_PKT_DATA_INBAND
:
1150 netvsc_receive_inband(device
, net_device_ctx
, nvmsg
);
1154 netdev_err(ndev
, "unhandled packet type %d, tid %llx\n",
1155 desc
->type
, request_id
);
1161 void netvsc_channel_cb(void *context
)
1164 struct vmbus_channel
*channel
= (struct vmbus_channel
*)context
;
1165 struct hv_device
*device
;
1166 struct netvsc_device
*net_device
;
1169 struct vmpacket_descriptor
*desc
;
1170 unsigned char *buffer
;
1171 int bufferlen
= NETVSC_PACKET_SIZE
;
1172 struct net_device
*ndev
;
1173 bool need_to_commit
= false;
1175 if (channel
->primary_channel
!= NULL
)
1176 device
= channel
->primary_channel
->device_obj
;
1178 device
= channel
->device_obj
;
1180 net_device
= get_inbound_net_device(device
);
1183 ndev
= hv_get_drvdata(device
);
1184 buffer
= get_per_channel_state(channel
);
1187 desc
= get_next_pkt_raw(channel
);
1189 netvsc_process_raw_pkt(device
,
1196 put_pkt_raw(channel
, desc
);
1197 need_to_commit
= true;
1200 if (need_to_commit
) {
1201 need_to_commit
= false;
1202 commit_rd_index(channel
);
1205 ret
= vmbus_recvpacket_raw(channel
, buffer
, bufferlen
,
1206 &bytes_recvd
, &request_id
);
1208 if (bytes_recvd
> 0) {
1209 desc
= (struct vmpacket_descriptor
*)buffer
;
1210 netvsc_process_raw_pkt(device
,
1220 * We are done for this pass.
1225 } else if (ret
== -ENOBUFS
) {
1226 if (bufferlen
> NETVSC_PACKET_SIZE
)
1228 /* Handle large packet */
1229 buffer
= kmalloc(bytes_recvd
, GFP_ATOMIC
);
1230 if (buffer
== NULL
) {
1231 /* Try again next time around */
1233 "unable to allocate buffer of size "
1234 "(%d)!!\n", bytes_recvd
);
1238 bufferlen
= bytes_recvd
;
1242 if (bufferlen
> NETVSC_PACKET_SIZE
)
1248 * netvsc_device_add - Callback when the device belonging to this
1251 int netvsc_device_add(struct hv_device
*device
, void *additional_info
)
1255 ((struct netvsc_device_info
*)additional_info
)->ring_size
;
1256 struct netvsc_device
*net_device
;
1257 struct net_device
*ndev
= hv_get_drvdata(device
);
1258 struct net_device_context
*net_device_ctx
= netdev_priv(ndev
);
1260 net_device
= alloc_net_device();
1264 net_device
->ring_size
= ring_size
;
1266 /* Initialize the NetVSC channel extension */
1267 init_completion(&net_device
->channel_init_wait
);
1269 set_per_channel_state(device
->channel
, net_device
->cb_buffer
);
1271 /* Open the channel */
1272 ret
= vmbus_open(device
->channel
, ring_size
* PAGE_SIZE
,
1273 ring_size
* PAGE_SIZE
, NULL
, 0,
1274 netvsc_channel_cb
, device
->channel
);
1277 netdev_err(ndev
, "unable to open channel: %d\n", ret
);
1281 /* Channel is opened */
1282 pr_info("hv_netvsc channel opened successfully\n");
1284 /* If we're reopening the device we may have multiple queues, fill the
1285 * chn_table with the default channel to use it before subchannels are
1288 for (i
= 0; i
< VRSS_CHANNEL_MAX
; i
++)
1289 net_device
->chn_table
[i
] = device
->channel
;
1291 /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1296 net_device_ctx
->nvdev
= net_device
;
1298 /* Connect with the NetVsp */
1299 ret
= netvsc_connect_vsp(device
);
1302 "unable to connect to NetVSP - %d\n", ret
);
1309 /* Now, we can close the channel safely */
1310 vmbus_close(device
->channel
);
1313 free_netvsc_device(net_device
);