xref: /openbmc/linux/drivers/net/hyperv/netvsc.c (revision 95e9fd10)
1 /*
2  * Copyright (c) 2009, Microsoft Corporation.
3  *
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.
7  *
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
11  * more details.
12  *
13  * You should have received a copy of the GNU General Public License along with
14  * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
15  * Place - Suite 330, Boston, MA 02111-1307 USA.
16  *
17  * Authors:
18  *   Haiyang Zhang <haiyangz@microsoft.com>
19  *   Hank Janssen  <hjanssen@microsoft.com>
20  */
21 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
22 
23 #include <linux/kernel.h>
24 #include <linux/sched.h>
25 #include <linux/wait.h>
26 #include <linux/mm.h>
27 #include <linux/delay.h>
28 #include <linux/io.h>
29 #include <linux/slab.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_ether.h>
32 
33 #include "hyperv_net.h"
34 
35 
36 static struct netvsc_device *alloc_net_device(struct hv_device *device)
37 {
38 	struct netvsc_device *net_device;
39 	struct net_device *ndev = hv_get_drvdata(device);
40 
41 	net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
42 	if (!net_device)
43 		return NULL;
44 
45 	init_waitqueue_head(&net_device->wait_drain);
46 	net_device->start_remove = false;
47 	net_device->destroy = false;
48 	net_device->dev = device;
49 	net_device->ndev = ndev;
50 
51 	hv_set_drvdata(device, net_device);
52 	return net_device;
53 }
54 
55 static struct netvsc_device *get_outbound_net_device(struct hv_device *device)
56 {
57 	struct netvsc_device *net_device;
58 
59 	net_device = hv_get_drvdata(device);
60 	if (net_device && net_device->destroy)
61 		net_device = NULL;
62 
63 	return net_device;
64 }
65 
66 static struct netvsc_device *get_inbound_net_device(struct hv_device *device)
67 {
68 	struct netvsc_device *net_device;
69 
70 	net_device = hv_get_drvdata(device);
71 
72 	if (!net_device)
73 		goto get_in_err;
74 
75 	if (net_device->destroy &&
76 		atomic_read(&net_device->num_outstanding_sends) == 0)
77 		net_device = NULL;
78 
79 get_in_err:
80 	return net_device;
81 }
82 
83 
84 static int netvsc_destroy_recv_buf(struct netvsc_device *net_device)
85 {
86 	struct nvsp_message *revoke_packet;
87 	int ret = 0;
88 	struct net_device *ndev = net_device->ndev;
89 
90 	/*
91 	 * If we got a section count, it means we received a
92 	 * SendReceiveBufferComplete msg (ie sent
93 	 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
94 	 * to send a revoke msg here
95 	 */
96 	if (net_device->recv_section_cnt) {
97 		/* Send the revoke receive buffer */
98 		revoke_packet = &net_device->revoke_packet;
99 		memset(revoke_packet, 0, sizeof(struct nvsp_message));
100 
101 		revoke_packet->hdr.msg_type =
102 			NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
103 		revoke_packet->msg.v1_msg.
104 		revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
105 
106 		ret = vmbus_sendpacket(net_device->dev->channel,
107 				       revoke_packet,
108 				       sizeof(struct nvsp_message),
109 				       (unsigned long)revoke_packet,
110 				       VM_PKT_DATA_INBAND, 0);
111 		/*
112 		 * If we failed here, we might as well return and
113 		 * have a leak rather than continue and a bugchk
114 		 */
115 		if (ret != 0) {
116 			netdev_err(ndev, "unable to send "
117 				"revoke receive buffer to netvsp\n");
118 			return ret;
119 		}
120 	}
121 
122 	/* Teardown the gpadl on the vsp end */
123 	if (net_device->recv_buf_gpadl_handle) {
124 		ret = vmbus_teardown_gpadl(net_device->dev->channel,
125 			   net_device->recv_buf_gpadl_handle);
126 
127 		/* If we failed here, we might as well return and have a leak
128 		 * rather than continue and a bugchk
129 		 */
130 		if (ret != 0) {
131 			netdev_err(ndev,
132 				   "unable to teardown receive buffer's gpadl\n");
133 			return ret;
134 		}
135 		net_device->recv_buf_gpadl_handle = 0;
136 	}
137 
138 	if (net_device->recv_buf) {
139 		/* Free up the receive buffer */
140 		free_pages((unsigned long)net_device->recv_buf,
141 			get_order(net_device->recv_buf_size));
142 		net_device->recv_buf = NULL;
143 	}
144 
145 	if (net_device->recv_section) {
146 		net_device->recv_section_cnt = 0;
147 		kfree(net_device->recv_section);
148 		net_device->recv_section = NULL;
149 	}
150 
151 	return ret;
152 }
153 
154 static int netvsc_init_recv_buf(struct hv_device *device)
155 {
156 	int ret = 0;
157 	int t;
158 	struct netvsc_device *net_device;
159 	struct nvsp_message *init_packet;
160 	struct net_device *ndev;
161 
162 	net_device = get_outbound_net_device(device);
163 	if (!net_device)
164 		return -ENODEV;
165 	ndev = net_device->ndev;
166 
167 	net_device->recv_buf =
168 		(void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO,
169 				get_order(net_device->recv_buf_size));
170 	if (!net_device->recv_buf) {
171 		netdev_err(ndev, "unable to allocate receive "
172 			"buffer of size %d\n", net_device->recv_buf_size);
173 		ret = -ENOMEM;
174 		goto cleanup;
175 	}
176 
177 	/*
178 	 * Establish the gpadl handle for this buffer on this
179 	 * channel.  Note: This call uses the vmbus connection rather
180 	 * than the channel to establish the gpadl handle.
181 	 */
182 	ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
183 				    net_device->recv_buf_size,
184 				    &net_device->recv_buf_gpadl_handle);
185 	if (ret != 0) {
186 		netdev_err(ndev,
187 			"unable to establish receive buffer's gpadl\n");
188 		goto cleanup;
189 	}
190 
191 
192 	/* Notify the NetVsp of the gpadl handle */
193 	init_packet = &net_device->channel_init_pkt;
194 
195 	memset(init_packet, 0, sizeof(struct nvsp_message));
196 
197 	init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
198 	init_packet->msg.v1_msg.send_recv_buf.
199 		gpadl_handle = net_device->recv_buf_gpadl_handle;
200 	init_packet->msg.v1_msg.
201 		send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
202 
203 	/* Send the gpadl notification request */
204 	ret = vmbus_sendpacket(device->channel, init_packet,
205 			       sizeof(struct nvsp_message),
206 			       (unsigned long)init_packet,
207 			       VM_PKT_DATA_INBAND,
208 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
209 	if (ret != 0) {
210 		netdev_err(ndev,
211 			"unable to send receive buffer's gpadl to netvsp\n");
212 		goto cleanup;
213 	}
214 
215 	t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
216 	BUG_ON(t == 0);
217 
218 
219 	/* Check the response */
220 	if (init_packet->msg.v1_msg.
221 	    send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
222 		netdev_err(ndev, "Unable to complete receive buffer "
223 			   "initialization with NetVsp - status %d\n",
224 			   init_packet->msg.v1_msg.
225 			   send_recv_buf_complete.status);
226 		ret = -EINVAL;
227 		goto cleanup;
228 	}
229 
230 	/* Parse the response */
231 
232 	net_device->recv_section_cnt = init_packet->msg.
233 		v1_msg.send_recv_buf_complete.num_sections;
234 
235 	net_device->recv_section = kmemdup(
236 		init_packet->msg.v1_msg.send_recv_buf_complete.sections,
237 		net_device->recv_section_cnt *
238 		sizeof(struct nvsp_1_receive_buffer_section),
239 		GFP_KERNEL);
240 	if (net_device->recv_section == NULL) {
241 		ret = -EINVAL;
242 		goto cleanup;
243 	}
244 
245 	/*
246 	 * For 1st release, there should only be 1 section that represents the
247 	 * entire receive buffer
248 	 */
249 	if (net_device->recv_section_cnt != 1 ||
250 	    net_device->recv_section->offset != 0) {
251 		ret = -EINVAL;
252 		goto cleanup;
253 	}
254 
255 	goto exit;
256 
257 cleanup:
258 	netvsc_destroy_recv_buf(net_device);
259 
260 exit:
261 	return ret;
262 }
263 
264 
265 /* Negotiate NVSP protocol version */
266 static int negotiate_nvsp_ver(struct hv_device *device,
267 			      struct netvsc_device *net_device,
268 			      struct nvsp_message *init_packet,
269 			      u32 nvsp_ver)
270 {
271 	int ret, t;
272 
273 	memset(init_packet, 0, sizeof(struct nvsp_message));
274 	init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
275 	init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
276 	init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
277 
278 	/* Send the init request */
279 	ret = vmbus_sendpacket(device->channel, init_packet,
280 			       sizeof(struct nvsp_message),
281 			       (unsigned long)init_packet,
282 			       VM_PKT_DATA_INBAND,
283 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
284 
285 	if (ret != 0)
286 		return ret;
287 
288 	t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
289 
290 	if (t == 0)
291 		return -ETIMEDOUT;
292 
293 	if (init_packet->msg.init_msg.init_complete.status !=
294 	    NVSP_STAT_SUCCESS)
295 		return -EINVAL;
296 
297 	if (nvsp_ver != NVSP_PROTOCOL_VERSION_2)
298 		return 0;
299 
300 	/* NVSPv2 only: Send NDIS config */
301 	memset(init_packet, 0, sizeof(struct nvsp_message));
302 	init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
303 	init_packet->msg.v2_msg.send_ndis_config.mtu = net_device->ndev->mtu;
304 	init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
305 
306 	ret = vmbus_sendpacket(device->channel, init_packet,
307 				sizeof(struct nvsp_message),
308 				(unsigned long)init_packet,
309 				VM_PKT_DATA_INBAND, 0);
310 
311 	return ret;
312 }
313 
314 static int netvsc_connect_vsp(struct hv_device *device)
315 {
316 	int ret;
317 	struct netvsc_device *net_device;
318 	struct nvsp_message *init_packet;
319 	int ndis_version;
320 	struct net_device *ndev;
321 
322 	net_device = get_outbound_net_device(device);
323 	if (!net_device)
324 		return -ENODEV;
325 	ndev = net_device->ndev;
326 
327 	init_packet = &net_device->channel_init_pkt;
328 
329 	/* Negotiate the latest NVSP protocol supported */
330 	if (negotiate_nvsp_ver(device, net_device, init_packet,
331 			       NVSP_PROTOCOL_VERSION_2) == 0) {
332 		net_device->nvsp_version = NVSP_PROTOCOL_VERSION_2;
333 	} else if (negotiate_nvsp_ver(device, net_device, init_packet,
334 				    NVSP_PROTOCOL_VERSION_1) == 0) {
335 		net_device->nvsp_version = NVSP_PROTOCOL_VERSION_1;
336 	} else {
337 		ret = -EPROTO;
338 		goto cleanup;
339 	}
340 
341 	pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
342 
343 	/* Send the ndis version */
344 	memset(init_packet, 0, sizeof(struct nvsp_message));
345 
346 	ndis_version = 0x00050001;
347 
348 	init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
349 	init_packet->msg.v1_msg.
350 		send_ndis_ver.ndis_major_ver =
351 				(ndis_version & 0xFFFF0000) >> 16;
352 	init_packet->msg.v1_msg.
353 		send_ndis_ver.ndis_minor_ver =
354 				ndis_version & 0xFFFF;
355 
356 	/* Send the init request */
357 	ret = vmbus_sendpacket(device->channel, init_packet,
358 				sizeof(struct nvsp_message),
359 				(unsigned long)init_packet,
360 				VM_PKT_DATA_INBAND, 0);
361 	if (ret != 0)
362 		goto cleanup;
363 
364 	/* Post the big receive buffer to NetVSP */
365 	ret = netvsc_init_recv_buf(device);
366 
367 cleanup:
368 	return ret;
369 }
370 
371 static void netvsc_disconnect_vsp(struct netvsc_device *net_device)
372 {
373 	netvsc_destroy_recv_buf(net_device);
374 }
375 
376 /*
377  * netvsc_device_remove - Callback when the root bus device is removed
378  */
379 int netvsc_device_remove(struct hv_device *device)
380 {
381 	struct netvsc_device *net_device;
382 	struct hv_netvsc_packet *netvsc_packet, *pos;
383 	unsigned long flags;
384 
385 	net_device = hv_get_drvdata(device);
386 
387 	netvsc_disconnect_vsp(net_device);
388 
389 	/*
390 	 * Since we have already drained, we don't need to busy wait
391 	 * as was done in final_release_stor_device()
392 	 * Note that we cannot set the ext pointer to NULL until
393 	 * we have drained - to drain the outgoing packets, we need to
394 	 * allow incoming packets.
395 	 */
396 
397 	spin_lock_irqsave(&device->channel->inbound_lock, flags);
398 	hv_set_drvdata(device, NULL);
399 	spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
400 
401 	/*
402 	 * At this point, no one should be accessing net_device
403 	 * except in here
404 	 */
405 	dev_notice(&device->device, "net device safe to remove\n");
406 
407 	/* Now, we can close the channel safely */
408 	vmbus_close(device->channel);
409 
410 	/* Release all resources */
411 	list_for_each_entry_safe(netvsc_packet, pos,
412 				 &net_device->recv_pkt_list, list_ent) {
413 		list_del(&netvsc_packet->list_ent);
414 		kfree(netvsc_packet);
415 	}
416 
417 	kfree(net_device);
418 	return 0;
419 }
420 
421 
422 #define RING_AVAIL_PERCENT_HIWATER 20
423 #define RING_AVAIL_PERCENT_LOWATER 10
424 
425 /*
426  * Get the percentage of available bytes to write in the ring.
427  * The return value is in range from 0 to 100.
428  */
429 static inline u32 hv_ringbuf_avail_percent(
430 		struct hv_ring_buffer_info *ring_info)
431 {
432 	u32 avail_read, avail_write;
433 
434 	hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
435 
436 	return avail_write * 100 / ring_info->ring_datasize;
437 }
438 
439 static void netvsc_send_completion(struct hv_device *device,
440 				   struct vmpacket_descriptor *packet)
441 {
442 	struct netvsc_device *net_device;
443 	struct nvsp_message *nvsp_packet;
444 	struct hv_netvsc_packet *nvsc_packet;
445 	struct net_device *ndev;
446 
447 	net_device = get_inbound_net_device(device);
448 	if (!net_device)
449 		return;
450 	ndev = net_device->ndev;
451 
452 	nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
453 			(packet->offset8 << 3));
454 
455 	if ((nvsp_packet->hdr.msg_type == NVSP_MSG_TYPE_INIT_COMPLETE) ||
456 	    (nvsp_packet->hdr.msg_type ==
457 	     NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) ||
458 	    (nvsp_packet->hdr.msg_type ==
459 	     NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE)) {
460 		/* Copy the response back */
461 		memcpy(&net_device->channel_init_pkt, nvsp_packet,
462 		       sizeof(struct nvsp_message));
463 		complete(&net_device->channel_init_wait);
464 	} else if (nvsp_packet->hdr.msg_type ==
465 		   NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) {
466 		int num_outstanding_sends;
467 
468 		/* Get the send context */
469 		nvsc_packet = (struct hv_netvsc_packet *)(unsigned long)
470 			packet->trans_id;
471 
472 		/* Notify the layer above us */
473 		nvsc_packet->completion.send.send_completion(
474 			nvsc_packet->completion.send.send_completion_ctx);
475 
476 		num_outstanding_sends =
477 			atomic_dec_return(&net_device->num_outstanding_sends);
478 
479 		if (net_device->destroy && num_outstanding_sends == 0)
480 			wake_up(&net_device->wait_drain);
481 
482 		if (netif_queue_stopped(ndev) && !net_device->start_remove &&
483 			(hv_ringbuf_avail_percent(&device->channel->outbound)
484 			> RING_AVAIL_PERCENT_HIWATER ||
485 			num_outstanding_sends < 1))
486 				netif_wake_queue(ndev);
487 	} else {
488 		netdev_err(ndev, "Unknown send completion packet type- "
489 			   "%d received!!\n", nvsp_packet->hdr.msg_type);
490 	}
491 
492 }
493 
494 int netvsc_send(struct hv_device *device,
495 			struct hv_netvsc_packet *packet)
496 {
497 	struct netvsc_device *net_device;
498 	int ret = 0;
499 	struct nvsp_message sendMessage;
500 	struct net_device *ndev;
501 
502 	net_device = get_outbound_net_device(device);
503 	if (!net_device)
504 		return -ENODEV;
505 	ndev = net_device->ndev;
506 
507 	sendMessage.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
508 	if (packet->is_data_pkt) {
509 		/* 0 is RMC_DATA; */
510 		sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 0;
511 	} else {
512 		/* 1 is RMC_CONTROL; */
513 		sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 1;
514 	}
515 
516 	/* Not using send buffer section */
517 	sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
518 		0xFFFFFFFF;
519 	sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
520 
521 	if (packet->page_buf_cnt) {
522 		ret = vmbus_sendpacket_pagebuffer(device->channel,
523 						  packet->page_buf,
524 						  packet->page_buf_cnt,
525 						  &sendMessage,
526 						  sizeof(struct nvsp_message),
527 						  (unsigned long)packet);
528 	} else {
529 		ret = vmbus_sendpacket(device->channel, &sendMessage,
530 				sizeof(struct nvsp_message),
531 				(unsigned long)packet,
532 				VM_PKT_DATA_INBAND,
533 				VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
534 
535 	}
536 
537 	if (ret == 0) {
538 		atomic_inc(&net_device->num_outstanding_sends);
539 		if (hv_ringbuf_avail_percent(&device->channel->outbound) <
540 			RING_AVAIL_PERCENT_LOWATER) {
541 			netif_stop_queue(ndev);
542 			if (atomic_read(&net_device->
543 				num_outstanding_sends) < 1)
544 				netif_wake_queue(ndev);
545 		}
546 	} else if (ret == -EAGAIN) {
547 		netif_stop_queue(ndev);
548 		if (atomic_read(&net_device->num_outstanding_sends) < 1) {
549 			netif_wake_queue(ndev);
550 			ret = -ENOSPC;
551 		}
552 	} else {
553 		netdev_err(ndev, "Unable to send packet %p ret %d\n",
554 			   packet, ret);
555 	}
556 
557 	return ret;
558 }
559 
560 static void netvsc_send_recv_completion(struct hv_device *device,
561 					u64 transaction_id)
562 {
563 	struct nvsp_message recvcompMessage;
564 	int retries = 0;
565 	int ret;
566 	struct net_device *ndev;
567 	struct netvsc_device *net_device = hv_get_drvdata(device);
568 
569 	ndev = net_device->ndev;
570 
571 	recvcompMessage.hdr.msg_type =
572 				NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
573 
574 	/* FIXME: Pass in the status */
575 	recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status =
576 		NVSP_STAT_SUCCESS;
577 
578 retry_send_cmplt:
579 	/* Send the completion */
580 	ret = vmbus_sendpacket(device->channel, &recvcompMessage,
581 			       sizeof(struct nvsp_message), transaction_id,
582 			       VM_PKT_COMP, 0);
583 	if (ret == 0) {
584 		/* success */
585 		/* no-op */
586 	} else if (ret == -EAGAIN) {
587 		/* no more room...wait a bit and attempt to retry 3 times */
588 		retries++;
589 		netdev_err(ndev, "unable to send receive completion pkt"
590 			" (tid %llx)...retrying %d\n", transaction_id, retries);
591 
592 		if (retries < 4) {
593 			udelay(100);
594 			goto retry_send_cmplt;
595 		} else {
596 			netdev_err(ndev, "unable to send receive "
597 				"completion pkt (tid %llx)...give up retrying\n",
598 				transaction_id);
599 		}
600 	} else {
601 		netdev_err(ndev, "unable to send receive "
602 			"completion pkt - %llx\n", transaction_id);
603 	}
604 }
605 
606 /* Send a receive completion packet to RNDIS device (ie NetVsp) */
607 static void netvsc_receive_completion(void *context)
608 {
609 	struct hv_netvsc_packet *packet = context;
610 	struct hv_device *device = packet->device;
611 	struct netvsc_device *net_device;
612 	u64 transaction_id = 0;
613 	bool fsend_receive_comp = false;
614 	unsigned long flags;
615 	struct net_device *ndev;
616 
617 	/*
618 	 * Even though it seems logical to do a GetOutboundNetDevice() here to
619 	 * send out receive completion, we are using GetInboundNetDevice()
620 	 * since we may have disable outbound traffic already.
621 	 */
622 	net_device = get_inbound_net_device(device);
623 	if (!net_device)
624 		return;
625 	ndev = net_device->ndev;
626 
627 	/* Overloading use of the lock. */
628 	spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
629 
630 	packet->xfer_page_pkt->count--;
631 
632 	/*
633 	 * Last one in the line that represent 1 xfer page packet.
634 	 * Return the xfer page packet itself to the freelist
635 	 */
636 	if (packet->xfer_page_pkt->count == 0) {
637 		fsend_receive_comp = true;
638 		transaction_id = packet->completion.recv.recv_completion_tid;
639 		list_add_tail(&packet->xfer_page_pkt->list_ent,
640 			      &net_device->recv_pkt_list);
641 
642 	}
643 
644 	/* Put the packet back */
645 	list_add_tail(&packet->list_ent, &net_device->recv_pkt_list);
646 	spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
647 
648 	/* Send a receive completion for the xfer page packet */
649 	if (fsend_receive_comp)
650 		netvsc_send_recv_completion(device, transaction_id);
651 
652 }
653 
654 static void netvsc_receive(struct hv_device *device,
655 			    struct vmpacket_descriptor *packet)
656 {
657 	struct netvsc_device *net_device;
658 	struct vmtransfer_page_packet_header *vmxferpage_packet;
659 	struct nvsp_message *nvsp_packet;
660 	struct hv_netvsc_packet *netvsc_packet = NULL;
661 	/* struct netvsc_driver *netvscDriver; */
662 	struct xferpage_packet *xferpage_packet = NULL;
663 	int i;
664 	int count = 0;
665 	unsigned long flags;
666 	struct net_device *ndev;
667 
668 	LIST_HEAD(listHead);
669 
670 	net_device = get_inbound_net_device(device);
671 	if (!net_device)
672 		return;
673 	ndev = net_device->ndev;
674 
675 	/*
676 	 * All inbound packets other than send completion should be xfer page
677 	 * packet
678 	 */
679 	if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
680 		netdev_err(ndev, "Unknown packet type received - %d\n",
681 			   packet->type);
682 		return;
683 	}
684 
685 	nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
686 			(packet->offset8 << 3));
687 
688 	/* Make sure this is a valid nvsp packet */
689 	if (nvsp_packet->hdr.msg_type !=
690 	    NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
691 		netdev_err(ndev, "Unknown nvsp packet type received-"
692 			" %d\n", nvsp_packet->hdr.msg_type);
693 		return;
694 	}
695 
696 	vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
697 
698 	if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
699 		netdev_err(ndev, "Invalid xfer page set id - "
700 			   "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
701 			   vmxferpage_packet->xfer_pageset_id);
702 		return;
703 	}
704 
705 	/*
706 	 * Grab free packets (range count + 1) to represent this xfer
707 	 * page packet. +1 to represent the xfer page packet itself.
708 	 * We grab it here so that we know exactly how many we can
709 	 * fulfil
710 	 */
711 	spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
712 	while (!list_empty(&net_device->recv_pkt_list)) {
713 		list_move_tail(net_device->recv_pkt_list.next, &listHead);
714 		if (++count == vmxferpage_packet->range_cnt + 1)
715 			break;
716 	}
717 	spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
718 
719 	/*
720 	 * We need at least 2 netvsc pkts (1 to represent the xfer
721 	 * page and at least 1 for the range) i.e. we can handled
722 	 * some of the xfer page packet ranges...
723 	 */
724 	if (count < 2) {
725 		netdev_err(ndev, "Got only %d netvsc pkt...needed "
726 			"%d pkts. Dropping this xfer page packet completely!\n",
727 			count, vmxferpage_packet->range_cnt + 1);
728 
729 		/* Return it to the freelist */
730 		spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
731 		for (i = count; i != 0; i--) {
732 			list_move_tail(listHead.next,
733 				       &net_device->recv_pkt_list);
734 		}
735 		spin_unlock_irqrestore(&net_device->recv_pkt_list_lock,
736 				       flags);
737 
738 		netvsc_send_recv_completion(device,
739 					    vmxferpage_packet->d.trans_id);
740 
741 		return;
742 	}
743 
744 	/* Remove the 1st packet to represent the xfer page packet itself */
745 	xferpage_packet = (struct xferpage_packet *)listHead.next;
746 	list_del(&xferpage_packet->list_ent);
747 
748 	/* This is how much we can satisfy */
749 	xferpage_packet->count = count - 1;
750 
751 	if (xferpage_packet->count != vmxferpage_packet->range_cnt) {
752 		netdev_err(ndev, "Needed %d netvsc pkts to satisfy "
753 			"this xfer page...got %d\n",
754 			vmxferpage_packet->range_cnt, xferpage_packet->count);
755 	}
756 
757 	/* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
758 	for (i = 0; i < (count - 1); i++) {
759 		netvsc_packet = (struct hv_netvsc_packet *)listHead.next;
760 		list_del(&netvsc_packet->list_ent);
761 
762 		/* Initialize the netvsc packet */
763 		netvsc_packet->xfer_page_pkt = xferpage_packet;
764 		netvsc_packet->completion.recv.recv_completion =
765 					netvsc_receive_completion;
766 		netvsc_packet->completion.recv.recv_completion_ctx =
767 					netvsc_packet;
768 		netvsc_packet->device = device;
769 		/* Save this so that we can send it back */
770 		netvsc_packet->completion.recv.recv_completion_tid =
771 					vmxferpage_packet->d.trans_id;
772 
773 		netvsc_packet->data = (void *)((unsigned long)net_device->
774 			recv_buf + vmxferpage_packet->ranges[i].byte_offset);
775 		netvsc_packet->total_data_buflen =
776 					vmxferpage_packet->ranges[i].byte_count;
777 
778 		/* Pass it to the upper layer */
779 		rndis_filter_receive(device, netvsc_packet);
780 
781 		netvsc_receive_completion(netvsc_packet->
782 				completion.recv.recv_completion_ctx);
783 	}
784 
785 }
786 
787 static void netvsc_channel_cb(void *context)
788 {
789 	int ret;
790 	struct hv_device *device = context;
791 	struct netvsc_device *net_device;
792 	u32 bytes_recvd;
793 	u64 request_id;
794 	unsigned char *packet;
795 	struct vmpacket_descriptor *desc;
796 	unsigned char *buffer;
797 	int bufferlen = NETVSC_PACKET_SIZE;
798 	struct net_device *ndev;
799 
800 	packet = kzalloc(NETVSC_PACKET_SIZE * sizeof(unsigned char),
801 			 GFP_ATOMIC);
802 	if (!packet)
803 		return;
804 	buffer = packet;
805 
806 	net_device = get_inbound_net_device(device);
807 	if (!net_device)
808 		goto out;
809 	ndev = net_device->ndev;
810 
811 	do {
812 		ret = vmbus_recvpacket_raw(device->channel, buffer, bufferlen,
813 					   &bytes_recvd, &request_id);
814 		if (ret == 0) {
815 			if (bytes_recvd > 0) {
816 				desc = (struct vmpacket_descriptor *)buffer;
817 				switch (desc->type) {
818 				case VM_PKT_COMP:
819 					netvsc_send_completion(device, desc);
820 					break;
821 
822 				case VM_PKT_DATA_USING_XFER_PAGES:
823 					netvsc_receive(device, desc);
824 					break;
825 
826 				default:
827 					netdev_err(ndev,
828 						   "unhandled packet type %d, "
829 						   "tid %llx len %d\n",
830 						   desc->type, request_id,
831 						   bytes_recvd);
832 					break;
833 				}
834 
835 				/* reset */
836 				if (bufferlen > NETVSC_PACKET_SIZE) {
837 					kfree(buffer);
838 					buffer = packet;
839 					bufferlen = NETVSC_PACKET_SIZE;
840 				}
841 			} else {
842 				/* reset */
843 				if (bufferlen > NETVSC_PACKET_SIZE) {
844 					kfree(buffer);
845 					buffer = packet;
846 					bufferlen = NETVSC_PACKET_SIZE;
847 				}
848 
849 				break;
850 			}
851 		} else if (ret == -ENOBUFS) {
852 			/* Handle large packet */
853 			buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
854 			if (buffer == NULL) {
855 				/* Try again next time around */
856 				netdev_err(ndev,
857 					   "unable to allocate buffer of size "
858 					   "(%d)!!\n", bytes_recvd);
859 				break;
860 			}
861 
862 			bufferlen = bytes_recvd;
863 		}
864 	} while (1);
865 
866 out:
867 	kfree(buffer);
868 	return;
869 }
870 
871 /*
872  * netvsc_device_add - Callback when the device belonging to this
873  * driver is added
874  */
875 int netvsc_device_add(struct hv_device *device, void *additional_info)
876 {
877 	int ret = 0;
878 	int i;
879 	int ring_size =
880 	((struct netvsc_device_info *)additional_info)->ring_size;
881 	struct netvsc_device *net_device;
882 	struct hv_netvsc_packet *packet, *pos;
883 	struct net_device *ndev;
884 
885 	net_device = alloc_net_device(device);
886 	if (!net_device) {
887 		ret = -ENOMEM;
888 		goto cleanup;
889 	}
890 
891 	/*
892 	 * Coming into this function, struct net_device * is
893 	 * registered as the driver private data.
894 	 * In alloc_net_device(), we register struct netvsc_device *
895 	 * as the driver private data and stash away struct net_device *
896 	 * in struct netvsc_device *.
897 	 */
898 	ndev = net_device->ndev;
899 
900 	/* Initialize the NetVSC channel extension */
901 	net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
902 	spin_lock_init(&net_device->recv_pkt_list_lock);
903 
904 	INIT_LIST_HEAD(&net_device->recv_pkt_list);
905 
906 	for (i = 0; i < NETVSC_RECEIVE_PACKETLIST_COUNT; i++) {
907 		packet = kzalloc(sizeof(struct hv_netvsc_packet) +
908 				 (NETVSC_RECEIVE_SG_COUNT *
909 				  sizeof(struct hv_page_buffer)), GFP_KERNEL);
910 		if (!packet)
911 			break;
912 
913 		list_add_tail(&packet->list_ent,
914 			      &net_device->recv_pkt_list);
915 	}
916 	init_completion(&net_device->channel_init_wait);
917 
918 	/* Open the channel */
919 	ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
920 			 ring_size * PAGE_SIZE, NULL, 0,
921 			 netvsc_channel_cb, device);
922 
923 	if (ret != 0) {
924 		netdev_err(ndev, "unable to open channel: %d\n", ret);
925 		goto cleanup;
926 	}
927 
928 	/* Channel is opened */
929 	pr_info("hv_netvsc channel opened successfully\n");
930 
931 	/* Connect with the NetVsp */
932 	ret = netvsc_connect_vsp(device);
933 	if (ret != 0) {
934 		netdev_err(ndev,
935 			"unable to connect to NetVSP - %d\n", ret);
936 		goto close;
937 	}
938 
939 	return ret;
940 
941 close:
942 	/* Now, we can close the channel safely */
943 	vmbus_close(device->channel);
944 
945 cleanup:
946 
947 	if (net_device) {
948 		list_for_each_entry_safe(packet, pos,
949 					 &net_device->recv_pkt_list,
950 					 list_ent) {
951 			list_del(&packet->list_ent);
952 			kfree(packet);
953 		}
954 
955 		kfree(net_device);
956 	}
957 
958 	return ret;
959 }
960