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 static struct netvsc_device
*alloc_net_device(struct hv_device
*device
)
39 struct netvsc_device
*net_device
;
40 struct net_device
*ndev
= hv_get_drvdata(device
);
43 net_device
= kzalloc(sizeof(struct netvsc_device
), GFP_KERNEL
);
47 net_device
->cb_buffer
= kzalloc(NETVSC_PACKET_SIZE
, GFP_KERNEL
);
48 if (!net_device
->cb_buffer
) {
53 init_waitqueue_head(&net_device
->wait_drain
);
54 net_device
->start_remove
= false;
55 net_device
->destroy
= false;
56 net_device
->dev
= device
;
57 net_device
->ndev
= ndev
;
58 net_device
->max_pkt
= RNDIS_MAX_PKT_DEFAULT
;
59 net_device
->pkt_align
= RNDIS_PKT_ALIGN_DEFAULT
;
61 for (i
= 0; i
< num_online_cpus(); i
++)
62 spin_lock_init(&net_device
->msd
[i
].lock
);
64 hv_set_drvdata(device
, net_device
);
68 static void free_netvsc_device(struct netvsc_device
*nvdev
)
70 kfree(nvdev
->cb_buffer
);
74 static struct netvsc_device
*get_outbound_net_device(struct hv_device
*device
)
76 struct netvsc_device
*net_device
;
78 net_device
= hv_get_drvdata(device
);
79 if (net_device
&& net_device
->destroy
)
85 static struct netvsc_device
*get_inbound_net_device(struct hv_device
*device
)
87 struct netvsc_device
*net_device
;
89 net_device
= hv_get_drvdata(device
);
94 if (net_device
->destroy
&&
95 atomic_read(&net_device
->num_outstanding_sends
) == 0)
103 static int netvsc_destroy_buf(struct netvsc_device
*net_device
)
105 struct nvsp_message
*revoke_packet
;
107 struct net_device
*ndev
= net_device
->ndev
;
110 * If we got a section count, it means we received a
111 * SendReceiveBufferComplete msg (ie sent
112 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
113 * to send a revoke msg here
115 if (net_device
->recv_section_cnt
) {
116 /* Send the revoke receive buffer */
117 revoke_packet
= &net_device
->revoke_packet
;
118 memset(revoke_packet
, 0, sizeof(struct nvsp_message
));
120 revoke_packet
->hdr
.msg_type
=
121 NVSP_MSG1_TYPE_REVOKE_RECV_BUF
;
122 revoke_packet
->msg
.v1_msg
.
123 revoke_recv_buf
.id
= NETVSC_RECEIVE_BUFFER_ID
;
125 ret
= vmbus_sendpacket(net_device
->dev
->channel
,
127 sizeof(struct nvsp_message
),
128 (unsigned long)revoke_packet
,
129 VM_PKT_DATA_INBAND
, 0);
131 * If we failed here, we might as well return and
132 * have a leak rather than continue and a bugchk
135 netdev_err(ndev
, "unable to send "
136 "revoke receive buffer to netvsp\n");
141 /* Teardown the gpadl on the vsp end */
142 if (net_device
->recv_buf_gpadl_handle
) {
143 ret
= vmbus_teardown_gpadl(net_device
->dev
->channel
,
144 net_device
->recv_buf_gpadl_handle
);
146 /* If we failed here, we might as well return and have a leak
147 * rather than continue and a bugchk
151 "unable to teardown receive buffer's gpadl\n");
154 net_device
->recv_buf_gpadl_handle
= 0;
157 if (net_device
->recv_buf
) {
158 /* Free up the receive buffer */
159 vfree(net_device
->recv_buf
);
160 net_device
->recv_buf
= NULL
;
163 if (net_device
->recv_section
) {
164 net_device
->recv_section_cnt
= 0;
165 kfree(net_device
->recv_section
);
166 net_device
->recv_section
= NULL
;
169 /* Deal with the send buffer we may have setup.
170 * If we got a send section size, it means we received a
171 * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
172 * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
173 * to send a revoke msg here
175 if (net_device
->send_section_size
) {
176 /* Send the revoke receive buffer */
177 revoke_packet
= &net_device
->revoke_packet
;
178 memset(revoke_packet
, 0, sizeof(struct nvsp_message
));
180 revoke_packet
->hdr
.msg_type
=
181 NVSP_MSG1_TYPE_REVOKE_SEND_BUF
;
182 revoke_packet
->msg
.v1_msg
.revoke_send_buf
.id
=
183 NETVSC_SEND_BUFFER_ID
;
185 ret
= vmbus_sendpacket(net_device
->dev
->channel
,
187 sizeof(struct nvsp_message
),
188 (unsigned long)revoke_packet
,
189 VM_PKT_DATA_INBAND
, 0);
190 /* If we failed here, we might as well return and
191 * have a leak rather than continue and a bugchk
194 netdev_err(ndev
, "unable to send "
195 "revoke send buffer to netvsp\n");
199 /* Teardown the gpadl on the vsp end */
200 if (net_device
->send_buf_gpadl_handle
) {
201 ret
= vmbus_teardown_gpadl(net_device
->dev
->channel
,
202 net_device
->send_buf_gpadl_handle
);
204 /* If we failed here, we might as well return and have a leak
205 * rather than continue and a bugchk
209 "unable to teardown send buffer's gpadl\n");
212 net_device
->send_buf_gpadl_handle
= 0;
214 if (net_device
->send_buf
) {
215 /* Free up the send buffer */
216 vfree(net_device
->send_buf
);
217 net_device
->send_buf
= NULL
;
219 kfree(net_device
->send_section_map
);
224 static int netvsc_init_buf(struct hv_device
*device
)
228 struct netvsc_device
*net_device
;
229 struct nvsp_message
*init_packet
;
230 struct net_device
*ndev
;
233 net_device
= get_outbound_net_device(device
);
236 ndev
= net_device
->ndev
;
238 node
= cpu_to_node(device
->channel
->target_cpu
);
239 net_device
->recv_buf
= vzalloc_node(net_device
->recv_buf_size
, node
);
240 if (!net_device
->recv_buf
)
241 net_device
->recv_buf
= vzalloc(net_device
->recv_buf_size
);
243 if (!net_device
->recv_buf
) {
244 netdev_err(ndev
, "unable to allocate receive "
245 "buffer of size %d\n", net_device
->recv_buf_size
);
251 * Establish the gpadl handle for this buffer on this
252 * channel. Note: This call uses the vmbus connection rather
253 * than the channel to establish the gpadl handle.
255 ret
= vmbus_establish_gpadl(device
->channel
, net_device
->recv_buf
,
256 net_device
->recv_buf_size
,
257 &net_device
->recv_buf_gpadl_handle
);
260 "unable to establish receive buffer's gpadl\n");
265 /* Notify the NetVsp of the gpadl handle */
266 init_packet
= &net_device
->channel_init_pkt
;
268 memset(init_packet
, 0, sizeof(struct nvsp_message
));
270 init_packet
->hdr
.msg_type
= NVSP_MSG1_TYPE_SEND_RECV_BUF
;
271 init_packet
->msg
.v1_msg
.send_recv_buf
.
272 gpadl_handle
= net_device
->recv_buf_gpadl_handle
;
273 init_packet
->msg
.v1_msg
.
274 send_recv_buf
.id
= NETVSC_RECEIVE_BUFFER_ID
;
276 /* Send the gpadl notification request */
277 ret
= vmbus_sendpacket(device
->channel
, init_packet
,
278 sizeof(struct nvsp_message
),
279 (unsigned long)init_packet
,
281 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
);
284 "unable to send receive buffer's gpadl to netvsp\n");
288 t
= wait_for_completion_timeout(&net_device
->channel_init_wait
, 5*HZ
);
292 /* Check the response */
293 if (init_packet
->msg
.v1_msg
.
294 send_recv_buf_complete
.status
!= NVSP_STAT_SUCCESS
) {
295 netdev_err(ndev
, "Unable to complete receive buffer "
296 "initialization with NetVsp - status %d\n",
297 init_packet
->msg
.v1_msg
.
298 send_recv_buf_complete
.status
);
303 /* Parse the response */
305 net_device
->recv_section_cnt
= init_packet
->msg
.
306 v1_msg
.send_recv_buf_complete
.num_sections
;
308 net_device
->recv_section
= kmemdup(
309 init_packet
->msg
.v1_msg
.send_recv_buf_complete
.sections
,
310 net_device
->recv_section_cnt
*
311 sizeof(struct nvsp_1_receive_buffer_section
),
313 if (net_device
->recv_section
== NULL
) {
319 * For 1st release, there should only be 1 section that represents the
320 * entire receive buffer
322 if (net_device
->recv_section_cnt
!= 1 ||
323 net_device
->recv_section
->offset
!= 0) {
328 /* Now setup the send buffer.
330 net_device
->send_buf
= vzalloc_node(net_device
->send_buf_size
, node
);
331 if (!net_device
->send_buf
)
332 net_device
->send_buf
= vzalloc(net_device
->send_buf_size
);
333 if (!net_device
->send_buf
) {
334 netdev_err(ndev
, "unable to allocate send "
335 "buffer of size %d\n", net_device
->send_buf_size
);
340 /* Establish the gpadl handle for this buffer on this
341 * channel. Note: This call uses the vmbus connection rather
342 * than the channel to establish the gpadl handle.
344 ret
= vmbus_establish_gpadl(device
->channel
, net_device
->send_buf
,
345 net_device
->send_buf_size
,
346 &net_device
->send_buf_gpadl_handle
);
349 "unable to establish send buffer's gpadl\n");
353 /* Notify the NetVsp of the gpadl handle */
354 init_packet
= &net_device
->channel_init_pkt
;
355 memset(init_packet
, 0, sizeof(struct nvsp_message
));
356 init_packet
->hdr
.msg_type
= NVSP_MSG1_TYPE_SEND_SEND_BUF
;
357 init_packet
->msg
.v1_msg
.send_send_buf
.gpadl_handle
=
358 net_device
->send_buf_gpadl_handle
;
359 init_packet
->msg
.v1_msg
.send_send_buf
.id
= NETVSC_SEND_BUFFER_ID
;
361 /* Send the gpadl notification request */
362 ret
= vmbus_sendpacket(device
->channel
, init_packet
,
363 sizeof(struct nvsp_message
),
364 (unsigned long)init_packet
,
366 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
);
369 "unable to send send buffer's gpadl to netvsp\n");
373 t
= wait_for_completion_timeout(&net_device
->channel_init_wait
, 5*HZ
);
376 /* Check the response */
377 if (init_packet
->msg
.v1_msg
.
378 send_send_buf_complete
.status
!= NVSP_STAT_SUCCESS
) {
379 netdev_err(ndev
, "Unable to complete send buffer "
380 "initialization with NetVsp - status %d\n",
381 init_packet
->msg
.v1_msg
.
382 send_send_buf_complete
.status
);
387 /* Parse the response */
388 net_device
->send_section_size
= init_packet
->msg
.
389 v1_msg
.send_send_buf_complete
.section_size
;
391 /* Section count is simply the size divided by the section size.
393 net_device
->send_section_cnt
=
394 net_device
->send_buf_size
/net_device
->send_section_size
;
396 dev_info(&device
->device
, "Send section size: %d, Section count:%d\n",
397 net_device
->send_section_size
, net_device
->send_section_cnt
);
399 /* Setup state for managing the send buffer. */
400 net_device
->map_words
= DIV_ROUND_UP(net_device
->send_section_cnt
,
403 net_device
->send_section_map
=
404 kzalloc(net_device
->map_words
* sizeof(ulong
), GFP_KERNEL
);
405 if (net_device
->send_section_map
== NULL
) {
413 netvsc_destroy_buf(net_device
);
420 /* Negotiate NVSP protocol version */
421 static int negotiate_nvsp_ver(struct hv_device
*device
,
422 struct netvsc_device
*net_device
,
423 struct nvsp_message
*init_packet
,
429 memset(init_packet
, 0, sizeof(struct nvsp_message
));
430 init_packet
->hdr
.msg_type
= NVSP_MSG_TYPE_INIT
;
431 init_packet
->msg
.init_msg
.init
.min_protocol_ver
= nvsp_ver
;
432 init_packet
->msg
.init_msg
.init
.max_protocol_ver
= nvsp_ver
;
434 /* Send the init request */
435 ret
= vmbus_sendpacket(device
->channel
, init_packet
,
436 sizeof(struct nvsp_message
),
437 (unsigned long)init_packet
,
439 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
);
444 t
= wait_for_completion_timeout(&net_device
->channel_init_wait
, 5*HZ
);
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
= net_device
->ndev
->mtu
+
461 init_packet
->msg
.v2_msg
.send_ndis_config
.capability
.ieee8021q
= 1;
463 if (nvsp_ver
>= NVSP_PROTOCOL_VERSION_5
)
464 init_packet
->msg
.v2_msg
.send_ndis_config
.capability
.sriov
= 1;
466 ret
= vmbus_sendpacket(device
->channel
, init_packet
,
467 sizeof(struct nvsp_message
),
468 (unsigned long)init_packet
,
469 VM_PKT_DATA_INBAND
, 0);
474 static int netvsc_connect_vsp(struct hv_device
*device
)
477 struct netvsc_device
*net_device
;
478 struct nvsp_message
*init_packet
;
480 struct net_device
*ndev
;
481 u32 ver_list
[] = { NVSP_PROTOCOL_VERSION_1
, NVSP_PROTOCOL_VERSION_2
,
482 NVSP_PROTOCOL_VERSION_4
, NVSP_PROTOCOL_VERSION_5
};
483 int i
, num_ver
= 4; /* number of different NVSP versions */
485 net_device
= get_outbound_net_device(device
);
488 ndev
= net_device
->ndev
;
490 init_packet
= &net_device
->channel_init_pkt
;
492 /* Negotiate the latest NVSP protocol supported */
493 for (i
= num_ver
- 1; i
>= 0; i
--)
494 if (negotiate_nvsp_ver(device
, net_device
, init_packet
,
496 net_device
->nvsp_version
= ver_list
[i
];
505 pr_debug("Negotiated NVSP version:%x\n", net_device
->nvsp_version
);
507 /* Send the ndis version */
508 memset(init_packet
, 0, sizeof(struct nvsp_message
));
510 if (net_device
->nvsp_version
<= NVSP_PROTOCOL_VERSION_4
)
511 ndis_version
= 0x00060001;
513 ndis_version
= 0x0006001e;
515 init_packet
->hdr
.msg_type
= NVSP_MSG1_TYPE_SEND_NDIS_VER
;
516 init_packet
->msg
.v1_msg
.
517 send_ndis_ver
.ndis_major_ver
=
518 (ndis_version
& 0xFFFF0000) >> 16;
519 init_packet
->msg
.v1_msg
.
520 send_ndis_ver
.ndis_minor_ver
=
521 ndis_version
& 0xFFFF;
523 /* Send the init request */
524 ret
= vmbus_sendpacket(device
->channel
, init_packet
,
525 sizeof(struct nvsp_message
),
526 (unsigned long)init_packet
,
527 VM_PKT_DATA_INBAND
, 0);
531 /* Post the big receive buffer to NetVSP */
532 if (net_device
->nvsp_version
<= NVSP_PROTOCOL_VERSION_2
)
533 net_device
->recv_buf_size
= NETVSC_RECEIVE_BUFFER_SIZE_LEGACY
;
535 net_device
->recv_buf_size
= NETVSC_RECEIVE_BUFFER_SIZE
;
536 net_device
->send_buf_size
= NETVSC_SEND_BUFFER_SIZE
;
538 ret
= netvsc_init_buf(device
);
544 static void netvsc_disconnect_vsp(struct netvsc_device
*net_device
)
546 netvsc_destroy_buf(net_device
);
550 * netvsc_device_remove - Callback when the root bus device is removed
552 int netvsc_device_remove(struct hv_device
*device
)
554 struct netvsc_device
*net_device
;
557 net_device
= hv_get_drvdata(device
);
559 netvsc_disconnect_vsp(net_device
);
562 * Since we have already drained, we don't need to busy wait
563 * as was done in final_release_stor_device()
564 * Note that we cannot set the ext pointer to NULL until
565 * we have drained - to drain the outgoing packets, we need to
566 * allow incoming packets.
569 spin_lock_irqsave(&device
->channel
->inbound_lock
, flags
);
570 hv_set_drvdata(device
, NULL
);
571 spin_unlock_irqrestore(&device
->channel
->inbound_lock
, flags
);
574 * At this point, no one should be accessing net_device
577 dev_notice(&device
->device
, "net device safe to remove\n");
579 /* Now, we can close the channel safely */
580 vmbus_close(device
->channel
);
582 /* Release all resources */
583 vfree(net_device
->sub_cb_buf
);
584 free_netvsc_device(net_device
);
589 #define RING_AVAIL_PERCENT_HIWATER 20
590 #define RING_AVAIL_PERCENT_LOWATER 10
593 * Get the percentage of available bytes to write in the ring.
594 * The return value is in range from 0 to 100.
596 static inline u32
hv_ringbuf_avail_percent(
597 struct hv_ring_buffer_info
*ring_info
)
599 u32 avail_read
, avail_write
;
601 hv_get_ringbuffer_availbytes(ring_info
, &avail_read
, &avail_write
);
603 return avail_write
* 100 / ring_info
->ring_datasize
;
606 static inline void netvsc_free_send_slot(struct netvsc_device
*net_device
,
609 sync_change_bit(index
, net_device
->send_section_map
);
612 static void netvsc_send_completion(struct netvsc_device
*net_device
,
613 struct vmbus_channel
*incoming_channel
,
614 struct hv_device
*device
,
615 struct vmpacket_descriptor
*packet
)
617 struct nvsp_message
*nvsp_packet
;
618 struct hv_netvsc_packet
*nvsc_packet
;
619 struct net_device
*ndev
;
622 ndev
= net_device
->ndev
;
624 nvsp_packet
= (struct nvsp_message
*)((unsigned long)packet
+
625 (packet
->offset8
<< 3));
627 if ((nvsp_packet
->hdr
.msg_type
== NVSP_MSG_TYPE_INIT_COMPLETE
) ||
628 (nvsp_packet
->hdr
.msg_type
==
629 NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE
) ||
630 (nvsp_packet
->hdr
.msg_type
==
631 NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE
) ||
632 (nvsp_packet
->hdr
.msg_type
==
633 NVSP_MSG5_TYPE_SUBCHANNEL
)) {
634 /* Copy the response back */
635 memcpy(&net_device
->channel_init_pkt
, nvsp_packet
,
636 sizeof(struct nvsp_message
));
637 complete(&net_device
->channel_init_wait
);
638 } else if (nvsp_packet
->hdr
.msg_type
==
639 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE
) {
640 int num_outstanding_sends
;
642 struct vmbus_channel
*channel
= device
->channel
;
645 /* Get the send context */
646 nvsc_packet
= (struct hv_netvsc_packet
*)(unsigned long)
649 /* Notify the layer above us */
651 send_index
= nvsc_packet
->send_buf_index
;
652 if (send_index
!= NETVSC_INVALID_INDEX
)
653 netvsc_free_send_slot(net_device
, send_index
);
654 q_idx
= nvsc_packet
->q_idx
;
655 channel
= incoming_channel
;
656 netvsc_xmit_completion(nvsc_packet
);
659 num_outstanding_sends
=
660 atomic_dec_return(&net_device
->num_outstanding_sends
);
661 queue_sends
= atomic_dec_return(&net_device
->
664 if (net_device
->destroy
&& num_outstanding_sends
== 0)
665 wake_up(&net_device
->wait_drain
);
667 if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev
, q_idx
)) &&
668 !net_device
->start_remove
&&
669 (hv_ringbuf_avail_percent(&channel
->outbound
) >
670 RING_AVAIL_PERCENT_HIWATER
|| queue_sends
< 1))
671 netif_tx_wake_queue(netdev_get_tx_queue(
674 netdev_err(ndev
, "Unknown send completion packet type- "
675 "%d received!!\n", nvsp_packet
->hdr
.msg_type
);
680 static u32
netvsc_get_next_send_section(struct netvsc_device
*net_device
)
683 u32 max_words
= net_device
->map_words
;
684 unsigned long *map_addr
= (unsigned long *)net_device
->send_section_map
;
685 u32 section_cnt
= net_device
->send_section_cnt
;
686 int ret_val
= NETVSC_INVALID_INDEX
;
690 for (i
= 0; i
< max_words
; i
++) {
693 index
= ffz(map_addr
[i
]);
694 prev_val
= sync_test_and_set_bit(index
, &map_addr
[i
]);
697 if ((index
+ (i
* BITS_PER_LONG
)) >= section_cnt
)
699 ret_val
= (index
+ (i
* BITS_PER_LONG
));
705 static u32
netvsc_copy_to_send_buf(struct netvsc_device
*net_device
,
706 unsigned int section_index
,
708 struct hv_netvsc_packet
*packet
,
709 struct rndis_message
*rndis_msg
)
711 char *start
= net_device
->send_buf
;
712 char *dest
= start
+ (section_index
* net_device
->send_section_size
)
717 u32 remain
= packet
->total_data_buflen
% net_device
->pkt_align
;
718 u32 page_count
= packet
->cp_partial
? packet
->rmsg_pgcnt
:
719 packet
->page_buf_cnt
;
722 if (packet
->is_data_pkt
&& packet
->xmit_more
&& remain
&&
723 !packet
->cp_partial
) {
724 padding
= net_device
->pkt_align
- remain
;
725 rndis_msg
->msg_len
+= padding
;
726 packet
->total_data_buflen
+= padding
;
729 for (i
= 0; i
< page_count
; i
++) {
730 char *src
= phys_to_virt(packet
->page_buf
[i
].pfn
<< PAGE_SHIFT
);
731 u32 offset
= packet
->page_buf
[i
].offset
;
732 u32 len
= packet
->page_buf
[i
].len
;
734 memcpy(dest
, (src
+ offset
), len
);
740 memset(dest
, 0, padding
);
747 static inline int netvsc_send_pkt(
748 struct hv_netvsc_packet
*packet
,
749 struct netvsc_device
*net_device
)
751 struct nvsp_message nvmsg
;
752 u16 q_idx
= packet
->q_idx
;
753 struct vmbus_channel
*out_channel
= net_device
->chn_table
[q_idx
];
754 struct net_device
*ndev
= net_device
->ndev
;
757 struct hv_page_buffer
*pgbuf
;
758 u32 ring_avail
= hv_ringbuf_avail_percent(&out_channel
->outbound
);
760 nvmsg
.hdr
.msg_type
= NVSP_MSG1_TYPE_SEND_RNDIS_PKT
;
761 if (packet
->is_data_pkt
) {
763 nvmsg
.msg
.v1_msg
.send_rndis_pkt
.channel_type
= 0;
765 /* 1 is RMC_CONTROL; */
766 nvmsg
.msg
.v1_msg
.send_rndis_pkt
.channel_type
= 1;
769 nvmsg
.msg
.v1_msg
.send_rndis_pkt
.send_buf_section_index
=
770 packet
->send_buf_index
;
771 if (packet
->send_buf_index
== NETVSC_INVALID_INDEX
)
772 nvmsg
.msg
.v1_msg
.send_rndis_pkt
.send_buf_section_size
= 0;
774 nvmsg
.msg
.v1_msg
.send_rndis_pkt
.send_buf_section_size
=
775 packet
->total_data_buflen
;
777 if (packet
->completion_func
)
778 req_id
= (ulong
)packet
;
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))
793 packet
->xmit_more
= false;
795 if (packet
->page_buf_cnt
) {
796 pgbuf
= packet
->cp_partial
? packet
->page_buf
+
797 packet
->rmsg_pgcnt
: packet
->page_buf
;
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 int netvsc_send(struct hv_device
*device
,
844 struct hv_netvsc_packet
*packet
,
845 struct rndis_message
*rndis_msg
)
847 struct netvsc_device
*net_device
;
848 int ret
= 0, m_ret
= 0;
849 struct vmbus_channel
*out_channel
;
850 u16 q_idx
= packet
->q_idx
;
851 u32 pktlen
= packet
->total_data_buflen
, msd_len
= 0;
852 unsigned int section_index
= NETVSC_INVALID_INDEX
;
854 struct multi_send_data
*msdp
;
855 struct hv_netvsc_packet
*msd_send
= NULL
, *cur_send
= NULL
;
858 net_device
= get_outbound_net_device(device
);
862 out_channel
= net_device
->chn_table
[q_idx
];
864 packet
->send_buf_index
= NETVSC_INVALID_INDEX
;
865 packet
->cp_partial
= false;
867 msdp
= &net_device
->msd
[q_idx
];
869 /* batch packets in send buffer if possible */
870 spin_lock_irqsave(&msdp
->lock
, flag
);
872 msd_len
= msdp
->pkt
->total_data_buflen
;
874 try_batch
= packet
->is_data_pkt
&& msd_len
> 0 && msdp
->count
<
877 if (try_batch
&& msd_len
+ pktlen
+ net_device
->pkt_align
<
878 net_device
->send_section_size
) {
879 section_index
= msdp
->pkt
->send_buf_index
;
881 } else if (try_batch
&& msd_len
+ packet
->rmsg_size
<
882 net_device
->send_section_size
) {
883 section_index
= msdp
->pkt
->send_buf_index
;
884 packet
->cp_partial
= true;
886 } else if (packet
->is_data_pkt
&& pktlen
+ net_device
->pkt_align
<
887 net_device
->send_section_size
) {
888 section_index
= netvsc_get_next_send_section(net_device
);
889 if (section_index
!= NETVSC_INVALID_INDEX
) {
890 msd_send
= msdp
->pkt
;
897 if (section_index
!= NETVSC_INVALID_INDEX
) {
898 netvsc_copy_to_send_buf(net_device
,
899 section_index
, msd_len
,
902 packet
->send_buf_index
= section_index
;
904 if (packet
->cp_partial
) {
905 packet
->page_buf_cnt
-= packet
->rmsg_pgcnt
;
906 packet
->total_data_buflen
= msd_len
+ packet
->rmsg_size
;
908 packet
->page_buf_cnt
= 0;
909 packet
->total_data_buflen
+= msd_len
;
913 netvsc_xmit_completion(msdp
->pkt
);
915 if (packet
->xmit_more
&& !packet
->cp_partial
) {
924 msd_send
= msdp
->pkt
;
930 spin_unlock_irqrestore(&msdp
->lock
, flag
);
933 m_ret
= netvsc_send_pkt(msd_send
, net_device
);
936 netvsc_free_send_slot(net_device
,
937 msd_send
->send_buf_index
);
938 netvsc_xmit_completion(msd_send
);
943 ret
= netvsc_send_pkt(cur_send
, net_device
);
945 if (ret
!= 0 && section_index
!= NETVSC_INVALID_INDEX
)
946 netvsc_free_send_slot(net_device
, section_index
);
951 static void netvsc_send_recv_completion(struct hv_device
*device
,
952 struct vmbus_channel
*channel
,
953 struct netvsc_device
*net_device
,
954 u64 transaction_id
, u32 status
)
956 struct nvsp_message recvcompMessage
;
959 struct net_device
*ndev
;
961 ndev
= net_device
->ndev
;
963 recvcompMessage
.hdr
.msg_type
=
964 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE
;
966 recvcompMessage
.msg
.v1_msg
.send_rndis_pkt_complete
.status
= status
;
969 /* Send the completion */
970 ret
= vmbus_sendpacket(channel
, &recvcompMessage
,
971 sizeof(struct nvsp_message
), transaction_id
,
976 } else if (ret
== -EAGAIN
) {
977 /* no more room...wait a bit and attempt to retry 3 times */
979 netdev_err(ndev
, "unable to send receive completion pkt"
980 " (tid %llx)...retrying %d\n", transaction_id
, retries
);
984 goto retry_send_cmplt
;
986 netdev_err(ndev
, "unable to send receive "
987 "completion pkt (tid %llx)...give up retrying\n",
991 netdev_err(ndev
, "unable to send receive "
992 "completion pkt - %llx\n", transaction_id
);
996 static void netvsc_receive(struct netvsc_device
*net_device
,
997 struct vmbus_channel
*channel
,
998 struct hv_device
*device
,
999 struct vmpacket_descriptor
*packet
)
1001 struct vmtransfer_page_packet_header
*vmxferpage_packet
;
1002 struct nvsp_message
*nvsp_packet
;
1003 struct hv_netvsc_packet nv_pkt
;
1004 struct hv_netvsc_packet
*netvsc_packet
= &nv_pkt
;
1005 u32 status
= NVSP_STAT_SUCCESS
;
1008 struct net_device
*ndev
;
1011 ndev
= net_device
->ndev
;
1014 * All inbound packets other than send completion should be xfer page
1017 if (packet
->type
!= VM_PKT_DATA_USING_XFER_PAGES
) {
1018 netdev_err(ndev
, "Unknown packet type received - %d\n",
1023 nvsp_packet
= (struct nvsp_message
*)((unsigned long)packet
+
1024 (packet
->offset8
<< 3));
1026 /* Make sure this is a valid nvsp packet */
1027 if (nvsp_packet
->hdr
.msg_type
!=
1028 NVSP_MSG1_TYPE_SEND_RNDIS_PKT
) {
1029 netdev_err(ndev
, "Unknown nvsp packet type received-"
1030 " %d\n", nvsp_packet
->hdr
.msg_type
);
1034 vmxferpage_packet
= (struct vmtransfer_page_packet_header
*)packet
;
1036 if (vmxferpage_packet
->xfer_pageset_id
!= NETVSC_RECEIVE_BUFFER_ID
) {
1037 netdev_err(ndev
, "Invalid xfer page set id - "
1038 "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID
,
1039 vmxferpage_packet
->xfer_pageset_id
);
1043 count
= vmxferpage_packet
->range_cnt
;
1045 /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1046 for (i
= 0; i
< count
; i
++) {
1047 /* Initialize the netvsc packet */
1048 netvsc_packet
->status
= NVSP_STAT_SUCCESS
;
1049 data
= (void *)((unsigned long)net_device
->
1050 recv_buf
+ vmxferpage_packet
->ranges
[i
].byte_offset
);
1051 netvsc_packet
->total_data_buflen
=
1052 vmxferpage_packet
->ranges
[i
].byte_count
;
1054 /* Pass it to the upper layer */
1055 rndis_filter_receive(device
, netvsc_packet
, &data
, channel
);
1057 if (netvsc_packet
->status
!= NVSP_STAT_SUCCESS
)
1058 status
= NVSP_STAT_FAIL
;
1061 netvsc_send_recv_completion(device
, channel
, net_device
,
1062 vmxferpage_packet
->d
.trans_id
, status
);
1066 static void netvsc_send_table(struct hv_device
*hdev
,
1067 struct nvsp_message
*nvmsg
)
1069 struct netvsc_device
*nvscdev
;
1070 struct net_device
*ndev
;
1074 nvscdev
= get_outbound_net_device(hdev
);
1077 ndev
= nvscdev
->ndev
;
1079 count
= nvmsg
->msg
.v5_msg
.send_table
.count
;
1080 if (count
!= VRSS_SEND_TAB_SIZE
) {
1081 netdev_err(ndev
, "Received wrong send-table size:%u\n", count
);
1085 tab
= (u32
*)((unsigned long)&nvmsg
->msg
.v5_msg
.send_table
+
1086 nvmsg
->msg
.v5_msg
.send_table
.offset
);
1088 for (i
= 0; i
< count
; i
++)
1089 nvscdev
->send_table
[i
] = tab
[i
];
1092 static void netvsc_send_vf(struct netvsc_device
*nvdev
,
1093 struct nvsp_message
*nvmsg
)
1095 nvdev
->vf_alloc
= nvmsg
->msg
.v4_msg
.vf_assoc
.allocated
;
1096 nvdev
->vf_serial
= nvmsg
->msg
.v4_msg
.vf_assoc
.serial
;
1099 static inline void netvsc_receive_inband(struct hv_device
*hdev
,
1100 struct netvsc_device
*nvdev
,
1101 struct nvsp_message
*nvmsg
)
1103 switch (nvmsg
->hdr
.msg_type
) {
1104 case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE
:
1105 netvsc_send_table(hdev
, nvmsg
);
1108 case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION
:
1109 netvsc_send_vf(nvdev
, nvmsg
);
1114 void netvsc_channel_cb(void *context
)
1117 struct vmbus_channel
*channel
= (struct vmbus_channel
*)context
;
1118 struct hv_device
*device
;
1119 struct netvsc_device
*net_device
;
1122 struct vmpacket_descriptor
*desc
;
1123 unsigned char *buffer
;
1124 int bufferlen
= NETVSC_PACKET_SIZE
;
1125 struct net_device
*ndev
;
1126 struct nvsp_message
*nvmsg
;
1128 if (channel
->primary_channel
!= NULL
)
1129 device
= channel
->primary_channel
->device_obj
;
1131 device
= channel
->device_obj
;
1133 net_device
= get_inbound_net_device(device
);
1136 ndev
= net_device
->ndev
;
1137 buffer
= get_per_channel_state(channel
);
1140 ret
= vmbus_recvpacket_raw(channel
, buffer
, bufferlen
,
1141 &bytes_recvd
, &request_id
);
1143 if (bytes_recvd
> 0) {
1144 desc
= (struct vmpacket_descriptor
*)buffer
;
1145 nvmsg
= (struct nvsp_message
*)((unsigned long)
1146 desc
+ (desc
->offset8
<< 3));
1147 switch (desc
->type
) {
1149 netvsc_send_completion(net_device
,
1154 case VM_PKT_DATA_USING_XFER_PAGES
:
1155 netvsc_receive(net_device
, channel
,
1159 case VM_PKT_DATA_INBAND
:
1160 netvsc_receive_inband(device
,
1167 "unhandled packet type %d, "
1168 "tid %llx len %d\n",
1169 desc
->type
, request_id
,
1176 * We are done for this pass.
1181 } else if (ret
== -ENOBUFS
) {
1182 if (bufferlen
> NETVSC_PACKET_SIZE
)
1184 /* Handle large packet */
1185 buffer
= kmalloc(bytes_recvd
, GFP_ATOMIC
);
1186 if (buffer
== NULL
) {
1187 /* Try again next time around */
1189 "unable to allocate buffer of size "
1190 "(%d)!!\n", bytes_recvd
);
1194 bufferlen
= bytes_recvd
;
1198 if (bufferlen
> NETVSC_PACKET_SIZE
)
1204 * netvsc_device_add - Callback when the device belonging to this
1207 int netvsc_device_add(struct hv_device
*device
, void *additional_info
)
1211 ((struct netvsc_device_info
*)additional_info
)->ring_size
;
1212 struct netvsc_device
*net_device
;
1213 struct net_device
*ndev
;
1215 net_device
= alloc_net_device(device
);
1219 net_device
->ring_size
= ring_size
;
1222 * Coming into this function, struct net_device * is
1223 * registered as the driver private data.
1224 * In alloc_net_device(), we register struct netvsc_device *
1225 * as the driver private data and stash away struct net_device *
1226 * in struct netvsc_device *.
1228 ndev
= net_device
->ndev
;
1230 /* Add netvsc_device context to netvsc_device */
1231 net_device
->nd_ctx
= netdev_priv(ndev
);
1233 /* Initialize the NetVSC channel extension */
1234 init_completion(&net_device
->channel_init_wait
);
1236 set_per_channel_state(device
->channel
, net_device
->cb_buffer
);
1238 /* Open the channel */
1239 ret
= vmbus_open(device
->channel
, ring_size
* PAGE_SIZE
,
1240 ring_size
* PAGE_SIZE
, NULL
, 0,
1241 netvsc_channel_cb
, device
->channel
);
1244 netdev_err(ndev
, "unable to open channel: %d\n", ret
);
1248 /* Channel is opened */
1249 pr_info("hv_netvsc channel opened successfully\n");
1251 net_device
->chn_table
[0] = device
->channel
;
1253 /* Connect with the NetVsp */
1254 ret
= netvsc_connect_vsp(device
);
1257 "unable to connect to NetVSP - %d\n", ret
);
1264 /* Now, we can close the channel safely */
1265 vmbus_close(device
->channel
);
1268 free_netvsc_device(net_device
);