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, see <http://www.gnu.org/licenses/>.
15 *
16 * Authors:
17 *   Haiyang Zhang <haiyangz@microsoft.com>
18 *   Hank Janssen  <hjanssen@microsoft.com>
19 */
20#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21
22#include <linux/kernel.h>
23#include <linux/sched.h>
24#include <linux/wait.h>
25#include <linux/mm.h>
26#include <linux/delay.h>
27#include <linux/io.h>
28#include <linux/slab.h>
29#include <linux/netdevice.h>
30#include <linux/if_ether.h>
31#include <asm/sync_bitops.h>
32
33#include "hyperv_net.h"
34
35
36static 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	int i;
41
42	net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
43	if (!net_device)
44		return NULL;
45
46	net_device->cb_buffer = kzalloc(NETVSC_PACKET_SIZE, GFP_KERNEL);
47	if (!net_device->cb_buffer) {
48		kfree(net_device);
49		return NULL;
50	}
51
52	init_waitqueue_head(&net_device->wait_drain);
53	net_device->start_remove = false;
54	net_device->destroy = false;
55	net_device->dev = device;
56	net_device->ndev = ndev;
57	net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
58	net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
59
60	for (i = 0; i < num_online_cpus(); i++)
61		spin_lock_init(&net_device->msd[i].lock);
62
63	hv_set_drvdata(device, net_device);
64	return net_device;
65}
66
67static void free_netvsc_device(struct netvsc_device *nvdev)
68{
69	kfree(nvdev->cb_buffer);
70	kfree(nvdev);
71}
72
73static struct netvsc_device *get_outbound_net_device(struct hv_device *device)
74{
75	struct netvsc_device *net_device;
76
77	net_device = hv_get_drvdata(device);
78	if (net_device && net_device->destroy)
79		net_device = NULL;
80
81	return net_device;
82}
83
84static struct netvsc_device *get_inbound_net_device(struct hv_device *device)
85{
86	struct netvsc_device *net_device;
87
88	net_device = hv_get_drvdata(device);
89
90	if (!net_device)
91		goto get_in_err;
92
93	if (net_device->destroy &&
94		atomic_read(&net_device->num_outstanding_sends) == 0)
95		net_device = NULL;
96
97get_in_err:
98	return net_device;
99}
100
101
102static int netvsc_destroy_buf(struct netvsc_device *net_device)
103{
104	struct nvsp_message *revoke_packet;
105	int ret = 0;
106	struct net_device *ndev = net_device->ndev;
107
108	/*
109	 * If we got a section count, it means we received a
110	 * SendReceiveBufferComplete msg (ie sent
111	 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
112	 * to send a revoke msg here
113	 */
114	if (net_device->recv_section_cnt) {
115		/* Send the revoke receive buffer */
116		revoke_packet = &net_device->revoke_packet;
117		memset(revoke_packet, 0, sizeof(struct nvsp_message));
118
119		revoke_packet->hdr.msg_type =
120			NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
121		revoke_packet->msg.v1_msg.
122		revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
123
124		ret = vmbus_sendpacket(net_device->dev->channel,
125				       revoke_packet,
126				       sizeof(struct nvsp_message),
127				       (unsigned long)revoke_packet,
128				       VM_PKT_DATA_INBAND, 0);
129		/*
130		 * If we failed here, we might as well return and
131		 * have a leak rather than continue and a bugchk
132		 */
133		if (ret != 0) {
134			netdev_err(ndev, "unable to send "
135				"revoke receive buffer to netvsp\n");
136			return ret;
137		}
138	}
139
140	/* Teardown the gpadl on the vsp end */
141	if (net_device->recv_buf_gpadl_handle) {
142		ret = vmbus_teardown_gpadl(net_device->dev->channel,
143			   net_device->recv_buf_gpadl_handle);
144
145		/* If we failed here, we might as well return and have a leak
146		 * rather than continue and a bugchk
147		 */
148		if (ret != 0) {
149			netdev_err(ndev,
150				   "unable to teardown receive buffer's gpadl\n");
151			return ret;
152		}
153		net_device->recv_buf_gpadl_handle = 0;
154	}
155
156	if (net_device->recv_buf) {
157		/* Free up the receive buffer */
158		vfree(net_device->recv_buf);
159		net_device->recv_buf = NULL;
160	}
161
162	if (net_device->recv_section) {
163		net_device->recv_section_cnt = 0;
164		kfree(net_device->recv_section);
165		net_device->recv_section = NULL;
166	}
167
168	/* Deal with the send buffer we may have setup.
169	 * If we got a  send section size, it means we received a
170	 * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
171	 * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
172	 * to send a revoke msg here
173	 */
174	if (net_device->send_section_size) {
175		/* Send the revoke receive buffer */
176		revoke_packet = &net_device->revoke_packet;
177		memset(revoke_packet, 0, sizeof(struct nvsp_message));
178
179		revoke_packet->hdr.msg_type =
180			NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
181		revoke_packet->msg.v1_msg.revoke_send_buf.id =
182			NETVSC_SEND_BUFFER_ID;
183
184		ret = vmbus_sendpacket(net_device->dev->channel,
185				       revoke_packet,
186				       sizeof(struct nvsp_message),
187				       (unsigned long)revoke_packet,
188				       VM_PKT_DATA_INBAND, 0);
189		/* If we failed here, we might as well return and
190		 * have a leak rather than continue and a bugchk
191		 */
192		if (ret != 0) {
193			netdev_err(ndev, "unable to send "
194				   "revoke send buffer to netvsp\n");
195			return ret;
196		}
197	}
198	/* Teardown the gpadl on the vsp end */
199	if (net_device->send_buf_gpadl_handle) {
200		ret = vmbus_teardown_gpadl(net_device->dev->channel,
201					   net_device->send_buf_gpadl_handle);
202
203		/* If we failed here, we might as well return and have a leak
204		 * rather than continue and a bugchk
205		 */
206		if (ret != 0) {
207			netdev_err(ndev,
208				   "unable to teardown send buffer's gpadl\n");
209			return ret;
210		}
211		net_device->send_buf_gpadl_handle = 0;
212	}
213	if (net_device->send_buf) {
214		/* Free up the send buffer */
215		vfree(net_device->send_buf);
216		net_device->send_buf = NULL;
217	}
218	kfree(net_device->send_section_map);
219
220	return ret;
221}
222
223static int netvsc_init_buf(struct hv_device *device)
224{
225	int ret = 0;
226	unsigned long t;
227	struct netvsc_device *net_device;
228	struct nvsp_message *init_packet;
229	struct net_device *ndev;
230
231	net_device = get_outbound_net_device(device);
232	if (!net_device)
233		return -ENODEV;
234	ndev = net_device->ndev;
235
236	net_device->recv_buf = vzalloc(net_device->recv_buf_size);
237	if (!net_device->recv_buf) {
238		netdev_err(ndev, "unable to allocate receive "
239			"buffer of size %d\n", net_device->recv_buf_size);
240		ret = -ENOMEM;
241		goto cleanup;
242	}
243
244	/*
245	 * Establish the gpadl handle for this buffer on this
246	 * channel.  Note: This call uses the vmbus connection rather
247	 * than the channel to establish the gpadl handle.
248	 */
249	ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
250				    net_device->recv_buf_size,
251				    &net_device->recv_buf_gpadl_handle);
252	if (ret != 0) {
253		netdev_err(ndev,
254			"unable to establish receive buffer's gpadl\n");
255		goto cleanup;
256	}
257
258
259	/* Notify the NetVsp of the gpadl handle */
260	init_packet = &net_device->channel_init_pkt;
261
262	memset(init_packet, 0, sizeof(struct nvsp_message));
263
264	init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
265	init_packet->msg.v1_msg.send_recv_buf.
266		gpadl_handle = net_device->recv_buf_gpadl_handle;
267	init_packet->msg.v1_msg.
268		send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
269
270	/* Send the gpadl notification request */
271	ret = vmbus_sendpacket(device->channel, init_packet,
272			       sizeof(struct nvsp_message),
273			       (unsigned long)init_packet,
274			       VM_PKT_DATA_INBAND,
275			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
276	if (ret != 0) {
277		netdev_err(ndev,
278			"unable to send receive buffer's gpadl to netvsp\n");
279		goto cleanup;
280	}
281
282	t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
283	BUG_ON(t == 0);
284
285
286	/* Check the response */
287	if (init_packet->msg.v1_msg.
288	    send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
289		netdev_err(ndev, "Unable to complete receive buffer "
290			   "initialization with NetVsp - status %d\n",
291			   init_packet->msg.v1_msg.
292			   send_recv_buf_complete.status);
293		ret = -EINVAL;
294		goto cleanup;
295	}
296
297	/* Parse the response */
298
299	net_device->recv_section_cnt = init_packet->msg.
300		v1_msg.send_recv_buf_complete.num_sections;
301
302	net_device->recv_section = kmemdup(
303		init_packet->msg.v1_msg.send_recv_buf_complete.sections,
304		net_device->recv_section_cnt *
305		sizeof(struct nvsp_1_receive_buffer_section),
306		GFP_KERNEL);
307	if (net_device->recv_section == NULL) {
308		ret = -EINVAL;
309		goto cleanup;
310	}
311
312	/*
313	 * For 1st release, there should only be 1 section that represents the
314	 * entire receive buffer
315	 */
316	if (net_device->recv_section_cnt != 1 ||
317	    net_device->recv_section->offset != 0) {
318		ret = -EINVAL;
319		goto cleanup;
320	}
321
322	/* Now setup the send buffer.
323	 */
324	net_device->send_buf = vzalloc(net_device->send_buf_size);
325	if (!net_device->send_buf) {
326		netdev_err(ndev, "unable to allocate send "
327			   "buffer of size %d\n", net_device->send_buf_size);
328		ret = -ENOMEM;
329		goto cleanup;
330	}
331
332	/* Establish the gpadl handle for this buffer on this
333	 * channel.  Note: This call uses the vmbus connection rather
334	 * than the channel to establish the gpadl handle.
335	 */
336	ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
337				    net_device->send_buf_size,
338				    &net_device->send_buf_gpadl_handle);
339	if (ret != 0) {
340		netdev_err(ndev,
341			   "unable to establish send buffer's gpadl\n");
342		goto cleanup;
343	}
344
345	/* Notify the NetVsp of the gpadl handle */
346	init_packet = &net_device->channel_init_pkt;
347	memset(init_packet, 0, sizeof(struct nvsp_message));
348	init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
349	init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
350		net_device->send_buf_gpadl_handle;
351	init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
352
353	/* Send the gpadl notification request */
354	ret = vmbus_sendpacket(device->channel, init_packet,
355			       sizeof(struct nvsp_message),
356			       (unsigned long)init_packet,
357			       VM_PKT_DATA_INBAND,
358			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
359	if (ret != 0) {
360		netdev_err(ndev,
361			   "unable to send send buffer's gpadl to netvsp\n");
362		goto cleanup;
363	}
364
365	t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
366	BUG_ON(t == 0);
367
368	/* Check the response */
369	if (init_packet->msg.v1_msg.
370	    send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
371		netdev_err(ndev, "Unable to complete send buffer "
372			   "initialization with NetVsp - status %d\n",
373			   init_packet->msg.v1_msg.
374			   send_send_buf_complete.status);
375		ret = -EINVAL;
376		goto cleanup;
377	}
378
379	/* Parse the response */
380	net_device->send_section_size = init_packet->msg.
381				v1_msg.send_send_buf_complete.section_size;
382
383	/* Section count is simply the size divided by the section size.
384	 */
385	net_device->send_section_cnt =
386		net_device->send_buf_size/net_device->send_section_size;
387
388	dev_info(&device->device, "Send section size: %d, Section count:%d\n",
389		 net_device->send_section_size, net_device->send_section_cnt);
390
391	/* Setup state for managing the send buffer. */
392	net_device->map_words = DIV_ROUND_UP(net_device->send_section_cnt,
393					     BITS_PER_LONG);
394
395	net_device->send_section_map =
396		kzalloc(net_device->map_words * sizeof(ulong), GFP_KERNEL);
397	if (net_device->send_section_map == NULL) {
398		ret = -ENOMEM;
399		goto cleanup;
400	}
401
402	goto exit;
403
404cleanup:
405	netvsc_destroy_buf(net_device);
406
407exit:
408	return ret;
409}
410
411
412/* Negotiate NVSP protocol version */
413static int negotiate_nvsp_ver(struct hv_device *device,
414			      struct netvsc_device *net_device,
415			      struct nvsp_message *init_packet,
416			      u32 nvsp_ver)
417{
418	int ret;
419	unsigned long t;
420
421	memset(init_packet, 0, sizeof(struct nvsp_message));
422	init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
423	init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
424	init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
425
426	/* Send the init request */
427	ret = vmbus_sendpacket(device->channel, init_packet,
428			       sizeof(struct nvsp_message),
429			       (unsigned long)init_packet,
430			       VM_PKT_DATA_INBAND,
431			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
432
433	if (ret != 0)
434		return ret;
435
436	t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
437
438	if (t == 0)
439		return -ETIMEDOUT;
440
441	if (init_packet->msg.init_msg.init_complete.status !=
442	    NVSP_STAT_SUCCESS)
443		return -EINVAL;
444
445	if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
446		return 0;
447
448	/* NVSPv2 only: Send NDIS config */
449	memset(init_packet, 0, sizeof(struct nvsp_message));
450	init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
451	init_packet->msg.v2_msg.send_ndis_config.mtu = net_device->ndev->mtu +
452						       ETH_HLEN;
453	init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
454
455	ret = vmbus_sendpacket(device->channel, init_packet,
456				sizeof(struct nvsp_message),
457				(unsigned long)init_packet,
458				VM_PKT_DATA_INBAND, 0);
459
460	return ret;
461}
462
463static int netvsc_connect_vsp(struct hv_device *device)
464{
465	int ret;
466	struct netvsc_device *net_device;
467	struct nvsp_message *init_packet;
468	int ndis_version;
469	struct net_device *ndev;
470	u32 ver_list[] = { NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
471		NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5 };
472	int i, num_ver = 4; /* number of different NVSP versions */
473
474	net_device = get_outbound_net_device(device);
475	if (!net_device)
476		return -ENODEV;
477	ndev = net_device->ndev;
478
479	init_packet = &net_device->channel_init_pkt;
480
481	/* Negotiate the latest NVSP protocol supported */
482	for (i = num_ver - 1; i >= 0; i--)
483		if (negotiate_nvsp_ver(device, net_device, init_packet,
484				       ver_list[i])  == 0) {
485			net_device->nvsp_version = ver_list[i];
486			break;
487		}
488
489	if (i < 0) {
490		ret = -EPROTO;
491		goto cleanup;
492	}
493
494	pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
495
496	/* Send the ndis version */
497	memset(init_packet, 0, sizeof(struct nvsp_message));
498
499	if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
500		ndis_version = 0x00060001;
501	else
502		ndis_version = 0x0006001e;
503
504	init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
505	init_packet->msg.v1_msg.
506		send_ndis_ver.ndis_major_ver =
507				(ndis_version & 0xFFFF0000) >> 16;
508	init_packet->msg.v1_msg.
509		send_ndis_ver.ndis_minor_ver =
510				ndis_version & 0xFFFF;
511
512	/* Send the init request */
513	ret = vmbus_sendpacket(device->channel, init_packet,
514				sizeof(struct nvsp_message),
515				(unsigned long)init_packet,
516				VM_PKT_DATA_INBAND, 0);
517	if (ret != 0)
518		goto cleanup;
519
520	/* Post the big receive buffer to NetVSP */
521	if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
522		net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
523	else
524		net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
525	net_device->send_buf_size = NETVSC_SEND_BUFFER_SIZE;
526
527	ret = netvsc_init_buf(device);
528
529cleanup:
530	return ret;
531}
532
533static void netvsc_disconnect_vsp(struct netvsc_device *net_device)
534{
535	netvsc_destroy_buf(net_device);
536}
537
538/*
539 * netvsc_device_remove - Callback when the root bus device is removed
540 */
541int netvsc_device_remove(struct hv_device *device)
542{
543	struct netvsc_device *net_device;
544	unsigned long flags;
545
546	net_device = hv_get_drvdata(device);
547
548	netvsc_disconnect_vsp(net_device);
549
550	/*
551	 * Since we have already drained, we don't need to busy wait
552	 * as was done in final_release_stor_device()
553	 * Note that we cannot set the ext pointer to NULL until
554	 * we have drained - to drain the outgoing packets, we need to
555	 * allow incoming packets.
556	 */
557
558	spin_lock_irqsave(&device->channel->inbound_lock, flags);
559	hv_set_drvdata(device, NULL);
560	spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
561
562	/*
563	 * At this point, no one should be accessing net_device
564	 * except in here
565	 */
566	dev_notice(&device->device, "net device safe to remove\n");
567
568	/* Now, we can close the channel safely */
569	vmbus_close(device->channel);
570
571	/* Release all resources */
572	vfree(net_device->sub_cb_buf);
573	free_netvsc_device(net_device);
574	return 0;
575}
576
577
578#define RING_AVAIL_PERCENT_HIWATER 20
579#define RING_AVAIL_PERCENT_LOWATER 10
580
581/*
582 * Get the percentage of available bytes to write in the ring.
583 * The return value is in range from 0 to 100.
584 */
585static inline u32 hv_ringbuf_avail_percent(
586		struct hv_ring_buffer_info *ring_info)
587{
588	u32 avail_read, avail_write;
589
590	hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
591
592	return avail_write * 100 / ring_info->ring_datasize;
593}
594
595static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
596					 u32 index)
597{
598	sync_change_bit(index, net_device->send_section_map);
599}
600
601static void netvsc_send_completion(struct netvsc_device *net_device,
602				   struct hv_device *device,
603				   struct vmpacket_descriptor *packet)
604{
605	struct nvsp_message *nvsp_packet;
606	struct hv_netvsc_packet *nvsc_packet;
607	struct net_device *ndev;
608	u32 send_index;
609
610	ndev = net_device->ndev;
611
612	nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
613			(packet->offset8 << 3));
614
615	if ((nvsp_packet->hdr.msg_type == NVSP_MSG_TYPE_INIT_COMPLETE) ||
616	    (nvsp_packet->hdr.msg_type ==
617	     NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) ||
618	    (nvsp_packet->hdr.msg_type ==
619	     NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE) ||
620	    (nvsp_packet->hdr.msg_type ==
621	     NVSP_MSG5_TYPE_SUBCHANNEL)) {
622		/* Copy the response back */
623		memcpy(&net_device->channel_init_pkt, nvsp_packet,
624		       sizeof(struct nvsp_message));
625		complete(&net_device->channel_init_wait);
626	} else if (nvsp_packet->hdr.msg_type ==
627		   NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) {
628		int num_outstanding_sends;
629		u16 q_idx = 0;
630		struct vmbus_channel *channel = device->channel;
631		int queue_sends;
632
633		/* Get the send context */
634		nvsc_packet = (struct hv_netvsc_packet *)(unsigned long)
635			packet->trans_id;
636
637		/* Notify the layer above us */
638		if (nvsc_packet) {
639			send_index = nvsc_packet->send_buf_index;
640			if (send_index != NETVSC_INVALID_INDEX)
641				netvsc_free_send_slot(net_device, send_index);
642			q_idx = nvsc_packet->q_idx;
643			channel = nvsc_packet->channel;
644			nvsc_packet->send_completion(nvsc_packet->
645						     send_completion_ctx);
646		}
647
648		num_outstanding_sends =
649			atomic_dec_return(&net_device->num_outstanding_sends);
650		queue_sends = atomic_dec_return(&net_device->
651						queue_sends[q_idx]);
652
653		if (net_device->destroy && num_outstanding_sends == 0)
654			wake_up(&net_device->wait_drain);
655
656		if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev, q_idx)) &&
657		    !net_device->start_remove &&
658		    (hv_ringbuf_avail_percent(&channel->outbound) >
659		     RING_AVAIL_PERCENT_HIWATER || queue_sends < 1))
660				netif_tx_wake_queue(netdev_get_tx_queue(
661						    ndev, q_idx));
662	} else {
663		netdev_err(ndev, "Unknown send completion packet type- "
664			   "%d received!!\n", nvsp_packet->hdr.msg_type);
665	}
666
667}
668
669static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
670{
671	unsigned long index;
672	u32 max_words = net_device->map_words;
673	unsigned long *map_addr = (unsigned long *)net_device->send_section_map;
674	u32 section_cnt = net_device->send_section_cnt;
675	int ret_val = NETVSC_INVALID_INDEX;
676	int i;
677	int prev_val;
678
679	for (i = 0; i < max_words; i++) {
680		if (!~(map_addr[i]))
681			continue;
682		index = ffz(map_addr[i]);
683		prev_val = sync_test_and_set_bit(index, &map_addr[i]);
684		if (prev_val)
685			continue;
686		if ((index + (i * BITS_PER_LONG)) >= section_cnt)
687			break;
688		ret_val = (index + (i * BITS_PER_LONG));
689		break;
690	}
691	return ret_val;
692}
693
694static u32 netvsc_copy_to_send_buf(struct netvsc_device *net_device,
695				   unsigned int section_index,
696				   u32 pend_size,
697				   struct hv_netvsc_packet *packet)
698{
699	char *start = net_device->send_buf;
700	char *dest = start + (section_index * net_device->send_section_size)
701		     + pend_size;
702	int i;
703	u32 msg_size = 0;
704	u32 padding = 0;
705	u32 remain = packet->total_data_buflen % net_device->pkt_align;
706	u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
707		packet->page_buf_cnt;
708
709	/* Add padding */
710	if (packet->is_data_pkt && packet->xmit_more && remain &&
711	    !packet->cp_partial) {
712		padding = net_device->pkt_align - remain;
713		packet->rndis_msg->msg_len += padding;
714		packet->total_data_buflen += padding;
715	}
716
717	for (i = 0; i < page_count; i++) {
718		char *src = phys_to_virt(packet->page_buf[i].pfn << PAGE_SHIFT);
719		u32 offset = packet->page_buf[i].offset;
720		u32 len = packet->page_buf[i].len;
721
722		memcpy(dest, (src + offset), len);
723		msg_size += len;
724		dest += len;
725	}
726
727	if (padding) {
728		memset(dest, 0, padding);
729		msg_size += padding;
730	}
731
732	return msg_size;
733}
734
735static inline int netvsc_send_pkt(
736	struct hv_netvsc_packet *packet,
737	struct netvsc_device *net_device)
738{
739	struct nvsp_message nvmsg;
740	struct vmbus_channel *out_channel = packet->channel;
741	u16 q_idx = packet->q_idx;
742	struct net_device *ndev = net_device->ndev;
743	u64 req_id;
744	int ret;
745	struct hv_page_buffer *pgbuf;
746
747	nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
748	if (packet->is_data_pkt) {
749		/* 0 is RMC_DATA; */
750		nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 0;
751	} else {
752		/* 1 is RMC_CONTROL; */
753		nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 1;
754	}
755
756	nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
757		packet->send_buf_index;
758	if (packet->send_buf_index == NETVSC_INVALID_INDEX)
759		nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
760	else
761		nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size =
762			packet->total_data_buflen;
763
764	if (packet->send_completion)
765		req_id = (ulong)packet;
766	else
767		req_id = 0;
768
769	if (out_channel->rescind)
770		return -ENODEV;
771
772	if (packet->page_buf_cnt) {
773		pgbuf = packet->cp_partial ? packet->page_buf +
774			packet->rmsg_pgcnt : packet->page_buf;
775		ret = vmbus_sendpacket_pagebuffer(out_channel,
776						  pgbuf,
777						  packet->page_buf_cnt,
778						  &nvmsg,
779						  sizeof(struct nvsp_message),
780						  req_id);
781	} else {
782		ret = vmbus_sendpacket(
783				out_channel, &nvmsg,
784				sizeof(struct nvsp_message),
785				req_id,
786				VM_PKT_DATA_INBAND,
787				VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
788	}
789
790	if (ret == 0) {
791		atomic_inc(&net_device->num_outstanding_sends);
792		atomic_inc(&net_device->queue_sends[q_idx]);
793
794		if (hv_ringbuf_avail_percent(&out_channel->outbound) <
795			RING_AVAIL_PERCENT_LOWATER) {
796			netif_tx_stop_queue(netdev_get_tx_queue(
797					    ndev, q_idx));
798
799			if (atomic_read(&net_device->
800				queue_sends[q_idx]) < 1)
801				netif_tx_wake_queue(netdev_get_tx_queue(
802						    ndev, q_idx));
803		}
804	} else if (ret == -EAGAIN) {
805		netif_tx_stop_queue(netdev_get_tx_queue(
806				    ndev, q_idx));
807		if (atomic_read(&net_device->queue_sends[q_idx]) < 1) {
808			netif_tx_wake_queue(netdev_get_tx_queue(
809					    ndev, q_idx));
810			ret = -ENOSPC;
811		}
812	} else {
813		netdev_err(ndev, "Unable to send packet %p ret %d\n",
814			   packet, ret);
815	}
816
817	return ret;
818}
819
820int netvsc_send(struct hv_device *device,
821		struct hv_netvsc_packet *packet)
822{
823	struct netvsc_device *net_device;
824	int ret = 0, m_ret = 0;
825	struct vmbus_channel *out_channel;
826	u16 q_idx = packet->q_idx;
827	u32 pktlen = packet->total_data_buflen, msd_len = 0;
828	unsigned int section_index = NETVSC_INVALID_INDEX;
829	unsigned long flag;
830	struct multi_send_data *msdp;
831	struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
832	bool try_batch;
833
834	net_device = get_outbound_net_device(device);
835	if (!net_device)
836		return -ENODEV;
837
838	out_channel = net_device->chn_table[q_idx];
839	if (!out_channel) {
840		out_channel = device->channel;
841		q_idx = 0;
842		packet->q_idx = 0;
843	}
844	packet->channel = out_channel;
845	packet->send_buf_index = NETVSC_INVALID_INDEX;
846	packet->cp_partial = false;
847
848	msdp = &net_device->msd[q_idx];
849
850	/* batch packets in send buffer if possible */
851	spin_lock_irqsave(&msdp->lock, flag);
852	if (msdp->pkt)
853		msd_len = msdp->pkt->total_data_buflen;
854
855	try_batch = packet->is_data_pkt && msd_len > 0 && msdp->count <
856		    net_device->max_pkt;
857
858	if (try_batch && msd_len + pktlen + net_device->pkt_align <
859	    net_device->send_section_size) {
860		section_index = msdp->pkt->send_buf_index;
861
862	} else if (try_batch && msd_len + packet->rmsg_size <
863		   net_device->send_section_size) {
864		section_index = msdp->pkt->send_buf_index;
865		packet->cp_partial = true;
866
867	} else if (packet->is_data_pkt && pktlen + net_device->pkt_align <
868		   net_device->send_section_size) {
869		section_index = netvsc_get_next_send_section(net_device);
870		if (section_index != NETVSC_INVALID_INDEX) {
871				msd_send = msdp->pkt;
872				msdp->pkt = NULL;
873				msdp->count = 0;
874				msd_len = 0;
875		}
876	}
877
878	if (section_index != NETVSC_INVALID_INDEX) {
879		netvsc_copy_to_send_buf(net_device,
880					section_index, msd_len,
881					packet);
882
883		packet->send_buf_index = section_index;
884
885		if (packet->cp_partial) {
886			packet->page_buf_cnt -= packet->rmsg_pgcnt;
887			packet->total_data_buflen = msd_len + packet->rmsg_size;
888		} else {
889			packet->page_buf_cnt = 0;
890			packet->total_data_buflen += msd_len;
891		}
892
893		if (msdp->pkt)
894			netvsc_xmit_completion(msdp->pkt);
895
896		if (packet->xmit_more && !packet->cp_partial) {
897			msdp->pkt = packet;
898			msdp->count++;
899		} else {
900			cur_send = packet;
901			msdp->pkt = NULL;
902			msdp->count = 0;
903		}
904	} else {
905		msd_send = msdp->pkt;
906		msdp->pkt = NULL;
907		msdp->count = 0;
908		cur_send = packet;
909	}
910
911	spin_unlock_irqrestore(&msdp->lock, flag);
912
913	if (msd_send) {
914		m_ret = netvsc_send_pkt(msd_send, net_device);
915
916		if (m_ret != 0) {
917			netvsc_free_send_slot(net_device,
918					      msd_send->send_buf_index);
919			netvsc_xmit_completion(msd_send);
920		}
921	}
922
923	if (cur_send)
924		ret = netvsc_send_pkt(cur_send, net_device);
925
926	if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
927		netvsc_free_send_slot(net_device, section_index);
928
929	return ret;
930}
931
932static void netvsc_send_recv_completion(struct hv_device *device,
933					struct vmbus_channel *channel,
934					struct netvsc_device *net_device,
935					u64 transaction_id, u32 status)
936{
937	struct nvsp_message recvcompMessage;
938	int retries = 0;
939	int ret;
940	struct net_device *ndev;
941
942	ndev = net_device->ndev;
943
944	recvcompMessage.hdr.msg_type =
945				NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
946
947	recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status = status;
948
949retry_send_cmplt:
950	/* Send the completion */
951	ret = vmbus_sendpacket(channel, &recvcompMessage,
952			       sizeof(struct nvsp_message), transaction_id,
953			       VM_PKT_COMP, 0);
954	if (ret == 0) {
955		/* success */
956		/* no-op */
957	} else if (ret == -EAGAIN) {
958		/* no more room...wait a bit and attempt to retry 3 times */
959		retries++;
960		netdev_err(ndev, "unable to send receive completion pkt"
961			" (tid %llx)...retrying %d\n", transaction_id, retries);
962
963		if (retries < 4) {
964			udelay(100);
965			goto retry_send_cmplt;
966		} else {
967			netdev_err(ndev, "unable to send receive "
968				"completion pkt (tid %llx)...give up retrying\n",
969				transaction_id);
970		}
971	} else {
972		netdev_err(ndev, "unable to send receive "
973			"completion pkt - %llx\n", transaction_id);
974	}
975}
976
977static void netvsc_receive(struct netvsc_device *net_device,
978			struct vmbus_channel *channel,
979			struct hv_device *device,
980			struct vmpacket_descriptor *packet)
981{
982	struct vmtransfer_page_packet_header *vmxferpage_packet;
983	struct nvsp_message *nvsp_packet;
984	struct hv_netvsc_packet nv_pkt;
985	struct hv_netvsc_packet *netvsc_packet = &nv_pkt;
986	u32 status = NVSP_STAT_SUCCESS;
987	int i;
988	int count = 0;
989	struct net_device *ndev;
990
991	ndev = net_device->ndev;
992
993	/*
994	 * All inbound packets other than send completion should be xfer page
995	 * packet
996	 */
997	if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
998		netdev_err(ndev, "Unknown packet type received - %d\n",
999			   packet->type);
1000		return;
1001	}
1002
1003	nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
1004			(packet->offset8 << 3));
1005
1006	/* Make sure this is a valid nvsp packet */
1007	if (nvsp_packet->hdr.msg_type !=
1008	    NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
1009		netdev_err(ndev, "Unknown nvsp packet type received-"
1010			" %d\n", nvsp_packet->hdr.msg_type);
1011		return;
1012	}
1013
1014	vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
1015
1016	if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
1017		netdev_err(ndev, "Invalid xfer page set id - "
1018			   "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
1019			   vmxferpage_packet->xfer_pageset_id);
1020		return;
1021	}
1022
1023	count = vmxferpage_packet->range_cnt;
1024	netvsc_packet->channel = channel;
1025
1026	/* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1027	for (i = 0; i < count; i++) {
1028		/* Initialize the netvsc packet */
1029		netvsc_packet->status = NVSP_STAT_SUCCESS;
1030		netvsc_packet->data = (void *)((unsigned long)net_device->
1031			recv_buf + vmxferpage_packet->ranges[i].byte_offset);
1032		netvsc_packet->total_data_buflen =
1033					vmxferpage_packet->ranges[i].byte_count;
1034
1035		/* Pass it to the upper layer */
1036		rndis_filter_receive(device, netvsc_packet);
1037
1038		if (netvsc_packet->status != NVSP_STAT_SUCCESS)
1039			status = NVSP_STAT_FAIL;
1040	}
1041
1042	netvsc_send_recv_completion(device, channel, net_device,
1043				    vmxferpage_packet->d.trans_id, status);
1044}
1045
1046
1047static void netvsc_send_table(struct hv_device *hdev,
1048			      struct vmpacket_descriptor *vmpkt)
1049{
1050	struct netvsc_device *nvscdev;
1051	struct net_device *ndev;
1052	struct nvsp_message *nvmsg;
1053	int i;
1054	u32 count, *tab;
1055
1056	nvscdev = get_outbound_net_device(hdev);
1057	if (!nvscdev)
1058		return;
1059	ndev = nvscdev->ndev;
1060
1061	nvmsg = (struct nvsp_message *)((unsigned long)vmpkt +
1062					(vmpkt->offset8 << 3));
1063
1064	if (nvmsg->hdr.msg_type != NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE)
1065		return;
1066
1067	count = nvmsg->msg.v5_msg.send_table.count;
1068	if (count != VRSS_SEND_TAB_SIZE) {
1069		netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1070		return;
1071	}
1072
1073	tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
1074		      nvmsg->msg.v5_msg.send_table.offset);
1075
1076	for (i = 0; i < count; i++)
1077		nvscdev->send_table[i] = tab[i];
1078}
1079
1080void netvsc_channel_cb(void *context)
1081{
1082	int ret;
1083	struct vmbus_channel *channel = (struct vmbus_channel *)context;
1084	struct hv_device *device;
1085	struct netvsc_device *net_device;
1086	u32 bytes_recvd;
1087	u64 request_id;
1088	struct vmpacket_descriptor *desc;
1089	unsigned char *buffer;
1090	int bufferlen = NETVSC_PACKET_SIZE;
1091	struct net_device *ndev;
1092
1093	if (channel->primary_channel != NULL)
1094		device = channel->primary_channel->device_obj;
1095	else
1096		device = channel->device_obj;
1097
1098	net_device = get_inbound_net_device(device);
1099	if (!net_device)
1100		return;
1101	ndev = net_device->ndev;
1102	buffer = get_per_channel_state(channel);
1103
1104	do {
1105		ret = vmbus_recvpacket_raw(channel, buffer, bufferlen,
1106					   &bytes_recvd, &request_id);
1107		if (ret == 0) {
1108			if (bytes_recvd > 0) {
1109				desc = (struct vmpacket_descriptor *)buffer;
1110				switch (desc->type) {
1111				case VM_PKT_COMP:
1112					netvsc_send_completion(net_device,
1113								device, desc);
1114					break;
1115
1116				case VM_PKT_DATA_USING_XFER_PAGES:
1117					netvsc_receive(net_device, channel,
1118						       device, desc);
1119					break;
1120
1121				case VM_PKT_DATA_INBAND:
1122					netvsc_send_table(device, desc);
1123					break;
1124
1125				default:
1126					netdev_err(ndev,
1127						   "unhandled packet type %d, "
1128						   "tid %llx len %d\n",
1129						   desc->type, request_id,
1130						   bytes_recvd);
1131					break;
1132				}
1133
1134			} else {
1135				/*
1136				 * We are done for this pass.
1137				 */
1138				break;
1139			}
1140
1141		} else if (ret == -ENOBUFS) {
1142			if (bufferlen > NETVSC_PACKET_SIZE)
1143				kfree(buffer);
1144			/* Handle large packet */
1145			buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
1146			if (buffer == NULL) {
1147				/* Try again next time around */
1148				netdev_err(ndev,
1149					   "unable to allocate buffer of size "
1150					   "(%d)!!\n", bytes_recvd);
1151				break;
1152			}
1153
1154			bufferlen = bytes_recvd;
1155		}
1156	} while (1);
1157
1158	if (bufferlen > NETVSC_PACKET_SIZE)
1159		kfree(buffer);
1160	return;
1161}
1162
1163/*
1164 * netvsc_device_add - Callback when the device belonging to this
1165 * driver is added
1166 */
1167int netvsc_device_add(struct hv_device *device, void *additional_info)
1168{
1169	int ret = 0;
1170	int ring_size =
1171	((struct netvsc_device_info *)additional_info)->ring_size;
1172	struct netvsc_device *net_device;
1173	struct net_device *ndev;
1174
1175	net_device = alloc_net_device(device);
1176	if (!net_device)
1177		return -ENOMEM;
1178
1179	net_device->ring_size = ring_size;
1180
1181	/*
1182	 * Coming into this function, struct net_device * is
1183	 * registered as the driver private data.
1184	 * In alloc_net_device(), we register struct netvsc_device *
1185	 * as the driver private data and stash away struct net_device *
1186	 * in struct netvsc_device *.
1187	 */
1188	ndev = net_device->ndev;
1189
1190	/* Add netvsc_device context to netvsc_device */
1191	net_device->nd_ctx = netdev_priv(ndev);
1192
1193	/* Initialize the NetVSC channel extension */
1194	init_completion(&net_device->channel_init_wait);
1195
1196	set_per_channel_state(device->channel, net_device->cb_buffer);
1197
1198	/* Open the channel */
1199	ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
1200			 ring_size * PAGE_SIZE, NULL, 0,
1201			 netvsc_channel_cb, device->channel);
1202
1203	if (ret != 0) {
1204		netdev_err(ndev, "unable to open channel: %d\n", ret);
1205		goto cleanup;
1206	}
1207
1208	/* Channel is opened */
1209	pr_info("hv_netvsc channel opened successfully\n");
1210
1211	net_device->chn_table[0] = device->channel;
1212
1213	/* Connect with the NetVsp */
1214	ret = netvsc_connect_vsp(device);
1215	if (ret != 0) {
1216		netdev_err(ndev,
1217			"unable to connect to NetVSP - %d\n", ret);
1218		goto close;
1219	}
1220
1221	return ret;
1222
1223close:
1224	/* Now, we can close the channel safely */
1225	vmbus_close(device->channel);
1226
1227cleanup:
1228	free_netvsc_device(net_device);
1229
1230	return ret;
1231}
1232