Merge tag 'arm-soc-defconfig-5.11' of git://git.kernel.org/pub/scm/linux/kernel/git...
[platform/kernel/linux-rpi.git] / drivers / hv / channel.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2009, Microsoft Corporation.
4  *
5  * Authors:
6  *   Haiyang Zhang <haiyangz@microsoft.com>
7  *   Hank Janssen  <hjanssen@microsoft.com>
8  */
9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10
11 #include <linux/kernel.h>
12 #include <linux/sched.h>
13 #include <linux/wait.h>
14 #include <linux/mm.h>
15 #include <linux/slab.h>
16 #include <linux/module.h>
17 #include <linux/hyperv.h>
18 #include <linux/uio.h>
19 #include <linux/interrupt.h>
20 #include <asm/page.h>
21 #include <asm/mshyperv.h>
22
23 #include "hyperv_vmbus.h"
24
25 /*
26  * hv_gpadl_size - Return the real size of a gpadl, the size that Hyper-V uses
27  *
28  * For BUFFER gpadl, Hyper-V uses the exact same size as the guest does.
29  *
30  * For RING gpadl, in each ring, the guest uses one PAGE_SIZE as the header
31  * (because of the alignment requirement), however, the hypervisor only
32  * uses the first HV_HYP_PAGE_SIZE as the header, therefore leaving a
33  * (PAGE_SIZE - HV_HYP_PAGE_SIZE) gap. And since there are two rings in a
34  * ringbuffer, the total size for a RING gpadl that Hyper-V uses is the
35  * total size that the guest uses minus twice of the gap size.
36  */
37 static inline u32 hv_gpadl_size(enum hv_gpadl_type type, u32 size)
38 {
39         switch (type) {
40         case HV_GPADL_BUFFER:
41                 return size;
42         case HV_GPADL_RING:
43                 /* The size of a ringbuffer must be page-aligned */
44                 BUG_ON(size % PAGE_SIZE);
45                 /*
46                  * Two things to notice here:
47                  * 1) We're processing two ring buffers as a unit
48                  * 2) We're skipping any space larger than HV_HYP_PAGE_SIZE in
49                  * the first guest-size page of each of the two ring buffers.
50                  * So we effectively subtract out two guest-size pages, and add
51                  * back two Hyper-V size pages.
52                  */
53                 return size - 2 * (PAGE_SIZE - HV_HYP_PAGE_SIZE);
54         }
55         BUG();
56         return 0;
57 }
58
59 /*
60  * hv_ring_gpadl_send_hvpgoffset - Calculate the send offset (in unit of
61  *                                 HV_HYP_PAGE) in a ring gpadl based on the
62  *                                 offset in the guest
63  *
64  * @offset: the offset (in bytes) where the send ringbuffer starts in the
65  *               virtual address space of the guest
66  */
67 static inline u32 hv_ring_gpadl_send_hvpgoffset(u32 offset)
68 {
69
70         /*
71          * For RING gpadl, in each ring, the guest uses one PAGE_SIZE as the
72          * header (because of the alignment requirement), however, the
73          * hypervisor only uses the first HV_HYP_PAGE_SIZE as the header,
74          * therefore leaving a (PAGE_SIZE - HV_HYP_PAGE_SIZE) gap.
75          *
76          * And to calculate the effective send offset in gpadl, we need to
77          * substract this gap.
78          */
79         return (offset - (PAGE_SIZE - HV_HYP_PAGE_SIZE)) >> HV_HYP_PAGE_SHIFT;
80 }
81
82 /*
83  * hv_gpadl_hvpfn - Return the Hyper-V page PFN of the @i th Hyper-V page in
84  *                  the gpadl
85  *
86  * @type: the type of the gpadl
87  * @kbuffer: the pointer to the gpadl in the guest
88  * @size: the total size (in bytes) of the gpadl
89  * @send_offset: the offset (in bytes) where the send ringbuffer starts in the
90  *               virtual address space of the guest
91  * @i: the index
92  */
93 static inline u64 hv_gpadl_hvpfn(enum hv_gpadl_type type, void *kbuffer,
94                                  u32 size, u32 send_offset, int i)
95 {
96         int send_idx = hv_ring_gpadl_send_hvpgoffset(send_offset);
97         unsigned long delta = 0UL;
98
99         switch (type) {
100         case HV_GPADL_BUFFER:
101                 break;
102         case HV_GPADL_RING:
103                 if (i == 0)
104                         delta = 0;
105                 else if (i <= send_idx)
106                         delta = PAGE_SIZE - HV_HYP_PAGE_SIZE;
107                 else
108                         delta = 2 * (PAGE_SIZE - HV_HYP_PAGE_SIZE);
109                 break;
110         default:
111                 BUG();
112                 break;
113         }
114
115         return virt_to_hvpfn(kbuffer + delta + (HV_HYP_PAGE_SIZE * i));
116 }
117
118 /*
119  * vmbus_setevent- Trigger an event notification on the specified
120  * channel.
121  */
122 void vmbus_setevent(struct vmbus_channel *channel)
123 {
124         struct hv_monitor_page *monitorpage;
125
126         trace_vmbus_setevent(channel);
127
128         /*
129          * For channels marked as in "low latency" mode
130          * bypass the monitor page mechanism.
131          */
132         if (channel->offermsg.monitor_allocated && !channel->low_latency) {
133                 vmbus_send_interrupt(channel->offermsg.child_relid);
134
135                 /* Get the child to parent monitor page */
136                 monitorpage = vmbus_connection.monitor_pages[1];
137
138                 sync_set_bit(channel->monitor_bit,
139                         (unsigned long *)&monitorpage->trigger_group
140                                         [channel->monitor_grp].pending);
141
142         } else {
143                 vmbus_set_event(channel);
144         }
145 }
146 EXPORT_SYMBOL_GPL(vmbus_setevent);
147
148 /* vmbus_free_ring - drop mapping of ring buffer */
149 void vmbus_free_ring(struct vmbus_channel *channel)
150 {
151         hv_ringbuffer_cleanup(&channel->outbound);
152         hv_ringbuffer_cleanup(&channel->inbound);
153
154         if (channel->ringbuffer_page) {
155                 __free_pages(channel->ringbuffer_page,
156                              get_order(channel->ringbuffer_pagecount
157                                        << PAGE_SHIFT));
158                 channel->ringbuffer_page = NULL;
159         }
160 }
161 EXPORT_SYMBOL_GPL(vmbus_free_ring);
162
163 /* vmbus_alloc_ring - allocate and map pages for ring buffer */
164 int vmbus_alloc_ring(struct vmbus_channel *newchannel,
165                      u32 send_size, u32 recv_size)
166 {
167         struct page *page;
168         int order;
169
170         if (send_size % PAGE_SIZE || recv_size % PAGE_SIZE)
171                 return -EINVAL;
172
173         /* Allocate the ring buffer */
174         order = get_order(send_size + recv_size);
175         page = alloc_pages_node(cpu_to_node(newchannel->target_cpu),
176                                 GFP_KERNEL|__GFP_ZERO, order);
177
178         if (!page)
179                 page = alloc_pages(GFP_KERNEL|__GFP_ZERO, order);
180
181         if (!page)
182                 return -ENOMEM;
183
184         newchannel->ringbuffer_page = page;
185         newchannel->ringbuffer_pagecount = (send_size + recv_size) >> PAGE_SHIFT;
186         newchannel->ringbuffer_send_offset = send_size >> PAGE_SHIFT;
187
188         return 0;
189 }
190 EXPORT_SYMBOL_GPL(vmbus_alloc_ring);
191
192 /* Used for Hyper-V Socket: a guest client's connect() to the host */
193 int vmbus_send_tl_connect_request(const guid_t *shv_guest_servie_id,
194                                   const guid_t *shv_host_servie_id)
195 {
196         struct vmbus_channel_tl_connect_request conn_msg;
197         int ret;
198
199         memset(&conn_msg, 0, sizeof(conn_msg));
200         conn_msg.header.msgtype = CHANNELMSG_TL_CONNECT_REQUEST;
201         conn_msg.guest_endpoint_id = *shv_guest_servie_id;
202         conn_msg.host_service_id = *shv_host_servie_id;
203
204         ret = vmbus_post_msg(&conn_msg, sizeof(conn_msg), true);
205
206         trace_vmbus_send_tl_connect_request(&conn_msg, ret);
207
208         return ret;
209 }
210 EXPORT_SYMBOL_GPL(vmbus_send_tl_connect_request);
211
212 /*
213  * Set/change the vCPU (@target_vp) the channel (@child_relid) will interrupt.
214  *
215  * CHANNELMSG_MODIFYCHANNEL messages are aynchronous.  Also, Hyper-V does not
216  * ACK such messages.  IOW we can't know when the host will stop interrupting
217  * the "old" vCPU and start interrupting the "new" vCPU for the given channel.
218  *
219  * The CHANNELMSG_MODIFYCHANNEL message type is supported since VMBus version
220  * VERSION_WIN10_V4_1.
221  */
222 int vmbus_send_modifychannel(u32 child_relid, u32 target_vp)
223 {
224         struct vmbus_channel_modifychannel conn_msg;
225         int ret;
226
227         memset(&conn_msg, 0, sizeof(conn_msg));
228         conn_msg.header.msgtype = CHANNELMSG_MODIFYCHANNEL;
229         conn_msg.child_relid = child_relid;
230         conn_msg.target_vp = target_vp;
231
232         ret = vmbus_post_msg(&conn_msg, sizeof(conn_msg), true);
233
234         trace_vmbus_send_modifychannel(&conn_msg, ret);
235
236         return ret;
237 }
238 EXPORT_SYMBOL_GPL(vmbus_send_modifychannel);
239
240 /*
241  * create_gpadl_header - Creates a gpadl for the specified buffer
242  */
243 static int create_gpadl_header(enum hv_gpadl_type type, void *kbuffer,
244                                u32 size, u32 send_offset,
245                                struct vmbus_channel_msginfo **msginfo)
246 {
247         int i;
248         int pagecount;
249         struct vmbus_channel_gpadl_header *gpadl_header;
250         struct vmbus_channel_gpadl_body *gpadl_body;
251         struct vmbus_channel_msginfo *msgheader;
252         struct vmbus_channel_msginfo *msgbody = NULL;
253         u32 msgsize;
254
255         int pfnsum, pfncount, pfnleft, pfncurr, pfnsize;
256
257         pagecount = hv_gpadl_size(type, size) >> HV_HYP_PAGE_SHIFT;
258
259         /* do we need a gpadl body msg */
260         pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
261                   sizeof(struct vmbus_channel_gpadl_header) -
262                   sizeof(struct gpa_range);
263         pfncount = pfnsize / sizeof(u64);
264
265         if (pagecount > pfncount) {
266                 /* we need a gpadl body */
267                 /* fill in the header */
268                 msgsize = sizeof(struct vmbus_channel_msginfo) +
269                           sizeof(struct vmbus_channel_gpadl_header) +
270                           sizeof(struct gpa_range) + pfncount * sizeof(u64);
271                 msgheader =  kzalloc(msgsize, GFP_KERNEL);
272                 if (!msgheader)
273                         goto nomem;
274
275                 INIT_LIST_HEAD(&msgheader->submsglist);
276                 msgheader->msgsize = msgsize;
277
278                 gpadl_header = (struct vmbus_channel_gpadl_header *)
279                         msgheader->msg;
280                 gpadl_header->rangecount = 1;
281                 gpadl_header->range_buflen = sizeof(struct gpa_range) +
282                                          pagecount * sizeof(u64);
283                 gpadl_header->range[0].byte_offset = 0;
284                 gpadl_header->range[0].byte_count = hv_gpadl_size(type, size);
285                 for (i = 0; i < pfncount; i++)
286                         gpadl_header->range[0].pfn_array[i] = hv_gpadl_hvpfn(
287                                 type, kbuffer, size, send_offset, i);
288                 *msginfo = msgheader;
289
290                 pfnsum = pfncount;
291                 pfnleft = pagecount - pfncount;
292
293                 /* how many pfns can we fit */
294                 pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
295                           sizeof(struct vmbus_channel_gpadl_body);
296                 pfncount = pfnsize / sizeof(u64);
297
298                 /* fill in the body */
299                 while (pfnleft) {
300                         if (pfnleft > pfncount)
301                                 pfncurr = pfncount;
302                         else
303                                 pfncurr = pfnleft;
304
305                         msgsize = sizeof(struct vmbus_channel_msginfo) +
306                                   sizeof(struct vmbus_channel_gpadl_body) +
307                                   pfncurr * sizeof(u64);
308                         msgbody = kzalloc(msgsize, GFP_KERNEL);
309
310                         if (!msgbody) {
311                                 struct vmbus_channel_msginfo *pos = NULL;
312                                 struct vmbus_channel_msginfo *tmp = NULL;
313                                 /*
314                                  * Free up all the allocated messages.
315                                  */
316                                 list_for_each_entry_safe(pos, tmp,
317                                         &msgheader->submsglist,
318                                         msglistentry) {
319
320                                         list_del(&pos->msglistentry);
321                                         kfree(pos);
322                                 }
323
324                                 goto nomem;
325                         }
326
327                         msgbody->msgsize = msgsize;
328                         gpadl_body =
329                                 (struct vmbus_channel_gpadl_body *)msgbody->msg;
330
331                         /*
332                          * Gpadl is u32 and we are using a pointer which could
333                          * be 64-bit
334                          * This is governed by the guest/host protocol and
335                          * so the hypervisor guarantees that this is ok.
336                          */
337                         for (i = 0; i < pfncurr; i++)
338                                 gpadl_body->pfn[i] = hv_gpadl_hvpfn(type,
339                                         kbuffer, size, send_offset, pfnsum + i);
340
341                         /* add to msg header */
342                         list_add_tail(&msgbody->msglistentry,
343                                       &msgheader->submsglist);
344                         pfnsum += pfncurr;
345                         pfnleft -= pfncurr;
346                 }
347         } else {
348                 /* everything fits in a header */
349                 msgsize = sizeof(struct vmbus_channel_msginfo) +
350                           sizeof(struct vmbus_channel_gpadl_header) +
351                           sizeof(struct gpa_range) + pagecount * sizeof(u64);
352                 msgheader = kzalloc(msgsize, GFP_KERNEL);
353                 if (msgheader == NULL)
354                         goto nomem;
355
356                 INIT_LIST_HEAD(&msgheader->submsglist);
357                 msgheader->msgsize = msgsize;
358
359                 gpadl_header = (struct vmbus_channel_gpadl_header *)
360                         msgheader->msg;
361                 gpadl_header->rangecount = 1;
362                 gpadl_header->range_buflen = sizeof(struct gpa_range) +
363                                          pagecount * sizeof(u64);
364                 gpadl_header->range[0].byte_offset = 0;
365                 gpadl_header->range[0].byte_count = hv_gpadl_size(type, size);
366                 for (i = 0; i < pagecount; i++)
367                         gpadl_header->range[0].pfn_array[i] = hv_gpadl_hvpfn(
368                                 type, kbuffer, size, send_offset, i);
369
370                 *msginfo = msgheader;
371         }
372
373         return 0;
374 nomem:
375         kfree(msgheader);
376         kfree(msgbody);
377         return -ENOMEM;
378 }
379
380 /*
381  * __vmbus_establish_gpadl - Establish a GPADL for a buffer or ringbuffer
382  *
383  * @channel: a channel
384  * @type: the type of the corresponding GPADL, only meaningful for the guest.
385  * @kbuffer: from kmalloc or vmalloc
386  * @size: page-size multiple
387  * @send_offset: the offset (in bytes) where the send ring buffer starts,
388  *               should be 0 for BUFFER type gpadl
389  * @gpadl_handle: some funky thing
390  */
391 static int __vmbus_establish_gpadl(struct vmbus_channel *channel,
392                                    enum hv_gpadl_type type, void *kbuffer,
393                                    u32 size, u32 send_offset,
394                                    u32 *gpadl_handle)
395 {
396         struct vmbus_channel_gpadl_header *gpadlmsg;
397         struct vmbus_channel_gpadl_body *gpadl_body;
398         struct vmbus_channel_msginfo *msginfo = NULL;
399         struct vmbus_channel_msginfo *submsginfo, *tmp;
400         struct list_head *curr;
401         u32 next_gpadl_handle;
402         unsigned long flags;
403         int ret = 0;
404
405         next_gpadl_handle =
406                 (atomic_inc_return(&vmbus_connection.next_gpadl_handle) - 1);
407
408         ret = create_gpadl_header(type, kbuffer, size, send_offset, &msginfo);
409         if (ret)
410                 return ret;
411
412         init_completion(&msginfo->waitevent);
413         msginfo->waiting_channel = channel;
414
415         gpadlmsg = (struct vmbus_channel_gpadl_header *)msginfo->msg;
416         gpadlmsg->header.msgtype = CHANNELMSG_GPADL_HEADER;
417         gpadlmsg->child_relid = channel->offermsg.child_relid;
418         gpadlmsg->gpadl = next_gpadl_handle;
419
420
421         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
422         list_add_tail(&msginfo->msglistentry,
423                       &vmbus_connection.chn_msg_list);
424
425         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
426
427         if (channel->rescind) {
428                 ret = -ENODEV;
429                 goto cleanup;
430         }
431
432         ret = vmbus_post_msg(gpadlmsg, msginfo->msgsize -
433                              sizeof(*msginfo), true);
434
435         trace_vmbus_establish_gpadl_header(gpadlmsg, ret);
436
437         if (ret != 0)
438                 goto cleanup;
439
440         list_for_each(curr, &msginfo->submsglist) {
441                 submsginfo = (struct vmbus_channel_msginfo *)curr;
442                 gpadl_body =
443                         (struct vmbus_channel_gpadl_body *)submsginfo->msg;
444
445                 gpadl_body->header.msgtype =
446                         CHANNELMSG_GPADL_BODY;
447                 gpadl_body->gpadl = next_gpadl_handle;
448
449                 ret = vmbus_post_msg(gpadl_body,
450                                      submsginfo->msgsize - sizeof(*submsginfo),
451                                      true);
452
453                 trace_vmbus_establish_gpadl_body(gpadl_body, ret);
454
455                 if (ret != 0)
456                         goto cleanup;
457
458         }
459         wait_for_completion(&msginfo->waitevent);
460
461         if (msginfo->response.gpadl_created.creation_status != 0) {
462                 pr_err("Failed to establish GPADL: err = 0x%x\n",
463                        msginfo->response.gpadl_created.creation_status);
464
465                 ret = -EDQUOT;
466                 goto cleanup;
467         }
468
469         if (channel->rescind) {
470                 ret = -ENODEV;
471                 goto cleanup;
472         }
473
474         /* At this point, we received the gpadl created msg */
475         *gpadl_handle = gpadlmsg->gpadl;
476
477 cleanup:
478         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
479         list_del(&msginfo->msglistentry);
480         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
481         list_for_each_entry_safe(submsginfo, tmp, &msginfo->submsglist,
482                                  msglistentry) {
483                 kfree(submsginfo);
484         }
485
486         kfree(msginfo);
487         return ret;
488 }
489
490 /*
491  * vmbus_establish_gpadl - Establish a GPADL for the specified buffer
492  *
493  * @channel: a channel
494  * @kbuffer: from kmalloc or vmalloc
495  * @size: page-size multiple
496  * @gpadl_handle: some funky thing
497  */
498 int vmbus_establish_gpadl(struct vmbus_channel *channel, void *kbuffer,
499                           u32 size, u32 *gpadl_handle)
500 {
501         return __vmbus_establish_gpadl(channel, HV_GPADL_BUFFER, kbuffer, size,
502                                        0U, gpadl_handle);
503 }
504 EXPORT_SYMBOL_GPL(vmbus_establish_gpadl);
505
506 /**
507  * request_arr_init - Allocates memory for the requestor array. Each slot
508  * keeps track of the next available slot in the array. Initially, each
509  * slot points to the next one (as in a Linked List). The last slot
510  * does not point to anything, so its value is U64_MAX by default.
511  * @size The size of the array
512  */
513 static u64 *request_arr_init(u32 size)
514 {
515         int i;
516         u64 *req_arr;
517
518         req_arr = kcalloc(size, sizeof(u64), GFP_KERNEL);
519         if (!req_arr)
520                 return NULL;
521
522         for (i = 0; i < size - 1; i++)
523                 req_arr[i] = i + 1;
524
525         /* Last slot (no more available slots) */
526         req_arr[i] = U64_MAX;
527
528         return req_arr;
529 }
530
531 /*
532  * vmbus_alloc_requestor - Initializes @rqstor's fields.
533  * Index 0 is the first free slot
534  * @size: Size of the requestor array
535  */
536 static int vmbus_alloc_requestor(struct vmbus_requestor *rqstor, u32 size)
537 {
538         u64 *rqst_arr;
539         unsigned long *bitmap;
540
541         rqst_arr = request_arr_init(size);
542         if (!rqst_arr)
543                 return -ENOMEM;
544
545         bitmap = bitmap_zalloc(size, GFP_KERNEL);
546         if (!bitmap) {
547                 kfree(rqst_arr);
548                 return -ENOMEM;
549         }
550
551         rqstor->req_arr = rqst_arr;
552         rqstor->req_bitmap = bitmap;
553         rqstor->size = size;
554         rqstor->next_request_id = 0;
555         spin_lock_init(&rqstor->req_lock);
556
557         return 0;
558 }
559
560 /*
561  * vmbus_free_requestor - Frees memory allocated for @rqstor
562  * @rqstor: Pointer to the requestor struct
563  */
564 static void vmbus_free_requestor(struct vmbus_requestor *rqstor)
565 {
566         kfree(rqstor->req_arr);
567         bitmap_free(rqstor->req_bitmap);
568 }
569
570 static int __vmbus_open(struct vmbus_channel *newchannel,
571                        void *userdata, u32 userdatalen,
572                        void (*onchannelcallback)(void *context), void *context)
573 {
574         struct vmbus_channel_open_channel *open_msg;
575         struct vmbus_channel_msginfo *open_info = NULL;
576         struct page *page = newchannel->ringbuffer_page;
577         u32 send_pages, recv_pages;
578         unsigned long flags;
579         int err;
580
581         if (userdatalen > MAX_USER_DEFINED_BYTES)
582                 return -EINVAL;
583
584         send_pages = newchannel->ringbuffer_send_offset;
585         recv_pages = newchannel->ringbuffer_pagecount - send_pages;
586
587         if (newchannel->state != CHANNEL_OPEN_STATE)
588                 return -EINVAL;
589
590         /* Create and init requestor */
591         if (newchannel->rqstor_size) {
592                 if (vmbus_alloc_requestor(&newchannel->requestor, newchannel->rqstor_size))
593                         return -ENOMEM;
594         }
595
596         newchannel->state = CHANNEL_OPENING_STATE;
597         newchannel->onchannel_callback = onchannelcallback;
598         newchannel->channel_callback_context = context;
599
600         err = hv_ringbuffer_init(&newchannel->outbound, page, send_pages);
601         if (err)
602                 goto error_clean_ring;
603
604         err = hv_ringbuffer_init(&newchannel->inbound,
605                                  &page[send_pages], recv_pages);
606         if (err)
607                 goto error_clean_ring;
608
609         /* Establish the gpadl for the ring buffer */
610         newchannel->ringbuffer_gpadlhandle = 0;
611
612         err = __vmbus_establish_gpadl(newchannel, HV_GPADL_RING,
613                                       page_address(newchannel->ringbuffer_page),
614                                       (send_pages + recv_pages) << PAGE_SHIFT,
615                                       newchannel->ringbuffer_send_offset << PAGE_SHIFT,
616                                       &newchannel->ringbuffer_gpadlhandle);
617         if (err)
618                 goto error_clean_ring;
619
620         /* Create and init the channel open message */
621         open_info = kmalloc(sizeof(*open_info) +
622                            sizeof(struct vmbus_channel_open_channel),
623                            GFP_KERNEL);
624         if (!open_info) {
625                 err = -ENOMEM;
626                 goto error_free_gpadl;
627         }
628
629         init_completion(&open_info->waitevent);
630         open_info->waiting_channel = newchannel;
631
632         open_msg = (struct vmbus_channel_open_channel *)open_info->msg;
633         open_msg->header.msgtype = CHANNELMSG_OPENCHANNEL;
634         open_msg->openid = newchannel->offermsg.child_relid;
635         open_msg->child_relid = newchannel->offermsg.child_relid;
636         open_msg->ringbuffer_gpadlhandle = newchannel->ringbuffer_gpadlhandle;
637         /*
638          * The unit of ->downstream_ringbuffer_pageoffset is HV_HYP_PAGE and
639          * the unit of ->ringbuffer_send_offset (i.e. send_pages) is PAGE, so
640          * here we calculate it into HV_HYP_PAGE.
641          */
642         open_msg->downstream_ringbuffer_pageoffset =
643                 hv_ring_gpadl_send_hvpgoffset(send_pages << PAGE_SHIFT);
644         open_msg->target_vp = hv_cpu_number_to_vp_number(newchannel->target_cpu);
645
646         if (userdatalen)
647                 memcpy(open_msg->userdata, userdata, userdatalen);
648
649         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
650         list_add_tail(&open_info->msglistentry,
651                       &vmbus_connection.chn_msg_list);
652         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
653
654         if (newchannel->rescind) {
655                 err = -ENODEV;
656                 goto error_free_info;
657         }
658
659         err = vmbus_post_msg(open_msg,
660                              sizeof(struct vmbus_channel_open_channel), true);
661
662         trace_vmbus_open(open_msg, err);
663
664         if (err != 0)
665                 goto error_clean_msglist;
666
667         wait_for_completion(&open_info->waitevent);
668
669         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
670         list_del(&open_info->msglistentry);
671         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
672
673         if (newchannel->rescind) {
674                 err = -ENODEV;
675                 goto error_free_info;
676         }
677
678         if (open_info->response.open_result.status) {
679                 err = -EAGAIN;
680                 goto error_free_info;
681         }
682
683         newchannel->state = CHANNEL_OPENED_STATE;
684         kfree(open_info);
685         return 0;
686
687 error_clean_msglist:
688         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
689         list_del(&open_info->msglistentry);
690         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
691 error_free_info:
692         kfree(open_info);
693 error_free_gpadl:
694         vmbus_teardown_gpadl(newchannel, newchannel->ringbuffer_gpadlhandle);
695         newchannel->ringbuffer_gpadlhandle = 0;
696 error_clean_ring:
697         hv_ringbuffer_cleanup(&newchannel->outbound);
698         hv_ringbuffer_cleanup(&newchannel->inbound);
699         vmbus_free_requestor(&newchannel->requestor);
700         newchannel->state = CHANNEL_OPEN_STATE;
701         return err;
702 }
703
704 /*
705  * vmbus_connect_ring - Open the channel but reuse ring buffer
706  */
707 int vmbus_connect_ring(struct vmbus_channel *newchannel,
708                        void (*onchannelcallback)(void *context), void *context)
709 {
710         return  __vmbus_open(newchannel, NULL, 0, onchannelcallback, context);
711 }
712 EXPORT_SYMBOL_GPL(vmbus_connect_ring);
713
714 /*
715  * vmbus_open - Open the specified channel.
716  */
717 int vmbus_open(struct vmbus_channel *newchannel,
718                u32 send_ringbuffer_size, u32 recv_ringbuffer_size,
719                void *userdata, u32 userdatalen,
720                void (*onchannelcallback)(void *context), void *context)
721 {
722         int err;
723
724         err = vmbus_alloc_ring(newchannel, send_ringbuffer_size,
725                                recv_ringbuffer_size);
726         if (err)
727                 return err;
728
729         err = __vmbus_open(newchannel, userdata, userdatalen,
730                            onchannelcallback, context);
731         if (err)
732                 vmbus_free_ring(newchannel);
733
734         return err;
735 }
736 EXPORT_SYMBOL_GPL(vmbus_open);
737
738 /*
739  * vmbus_teardown_gpadl -Teardown the specified GPADL handle
740  */
741 int vmbus_teardown_gpadl(struct vmbus_channel *channel, u32 gpadl_handle)
742 {
743         struct vmbus_channel_gpadl_teardown *msg;
744         struct vmbus_channel_msginfo *info;
745         unsigned long flags;
746         int ret;
747
748         info = kmalloc(sizeof(*info) +
749                        sizeof(struct vmbus_channel_gpadl_teardown), GFP_KERNEL);
750         if (!info)
751                 return -ENOMEM;
752
753         init_completion(&info->waitevent);
754         info->waiting_channel = channel;
755
756         msg = (struct vmbus_channel_gpadl_teardown *)info->msg;
757
758         msg->header.msgtype = CHANNELMSG_GPADL_TEARDOWN;
759         msg->child_relid = channel->offermsg.child_relid;
760         msg->gpadl = gpadl_handle;
761
762         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
763         list_add_tail(&info->msglistentry,
764                       &vmbus_connection.chn_msg_list);
765         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
766
767         if (channel->rescind)
768                 goto post_msg_err;
769
770         ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_gpadl_teardown),
771                              true);
772
773         trace_vmbus_teardown_gpadl(msg, ret);
774
775         if (ret)
776                 goto post_msg_err;
777
778         wait_for_completion(&info->waitevent);
779
780 post_msg_err:
781         /*
782          * If the channel has been rescinded;
783          * we will be awakened by the rescind
784          * handler; set the error code to zero so we don't leak memory.
785          */
786         if (channel->rescind)
787                 ret = 0;
788
789         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
790         list_del(&info->msglistentry);
791         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
792
793         kfree(info);
794         return ret;
795 }
796 EXPORT_SYMBOL_GPL(vmbus_teardown_gpadl);
797
798 void vmbus_reset_channel_cb(struct vmbus_channel *channel)
799 {
800         unsigned long flags;
801
802         /*
803          * vmbus_on_event(), running in the per-channel tasklet, can race
804          * with vmbus_close_internal() in the case of SMP guest, e.g., when
805          * the former is accessing channel->inbound.ring_buffer, the latter
806          * could be freeing the ring_buffer pages, so here we must stop it
807          * first.
808          *
809          * vmbus_chan_sched() might call the netvsc driver callback function
810          * that ends up scheduling NAPI work that accesses the ring buffer.
811          * At this point, we have to ensure that any such work is completed
812          * and that the channel ring buffer is no longer being accessed, cf.
813          * the calls to napi_disable() in netvsc_device_remove().
814          */
815         tasklet_disable(&channel->callback_event);
816
817         /* See the inline comments in vmbus_chan_sched(). */
818         spin_lock_irqsave(&channel->sched_lock, flags);
819         channel->onchannel_callback = NULL;
820         spin_unlock_irqrestore(&channel->sched_lock, flags);
821
822         channel->sc_creation_callback = NULL;
823
824         /* Re-enable tasklet for use on re-open */
825         tasklet_enable(&channel->callback_event);
826 }
827
828 static int vmbus_close_internal(struct vmbus_channel *channel)
829 {
830         struct vmbus_channel_close_channel *msg;
831         int ret;
832
833         vmbus_reset_channel_cb(channel);
834
835         /*
836          * In case a device driver's probe() fails (e.g.,
837          * util_probe() -> vmbus_open() returns -ENOMEM) and the device is
838          * rescinded later (e.g., we dynamically disable an Integrated Service
839          * in Hyper-V Manager), the driver's remove() invokes vmbus_close():
840          * here we should skip most of the below cleanup work.
841          */
842         if (channel->state != CHANNEL_OPENED_STATE)
843                 return -EINVAL;
844
845         channel->state = CHANNEL_OPEN_STATE;
846
847         /* Send a closing message */
848
849         msg = &channel->close_msg.msg;
850
851         msg->header.msgtype = CHANNELMSG_CLOSECHANNEL;
852         msg->child_relid = channel->offermsg.child_relid;
853
854         ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_close_channel),
855                              true);
856
857         trace_vmbus_close_internal(msg, ret);
858
859         if (ret) {
860                 pr_err("Close failed: close post msg return is %d\n", ret);
861                 /*
862                  * If we failed to post the close msg,
863                  * it is perhaps better to leak memory.
864                  */
865         }
866
867         /* Tear down the gpadl for the channel's ring buffer */
868         else if (channel->ringbuffer_gpadlhandle) {
869                 ret = vmbus_teardown_gpadl(channel,
870                                            channel->ringbuffer_gpadlhandle);
871                 if (ret) {
872                         pr_err("Close failed: teardown gpadl return %d\n", ret);
873                         /*
874                          * If we failed to teardown gpadl,
875                          * it is perhaps better to leak memory.
876                          */
877                 }
878
879                 channel->ringbuffer_gpadlhandle = 0;
880         }
881
882         if (!ret)
883                 vmbus_free_requestor(&channel->requestor);
884
885         return ret;
886 }
887
888 /* disconnect ring - close all channels */
889 int vmbus_disconnect_ring(struct vmbus_channel *channel)
890 {
891         struct vmbus_channel *cur_channel, *tmp;
892         int ret;
893
894         if (channel->primary_channel != NULL)
895                 return -EINVAL;
896
897         list_for_each_entry_safe(cur_channel, tmp, &channel->sc_list, sc_list) {
898                 if (cur_channel->rescind)
899                         wait_for_completion(&cur_channel->rescind_event);
900
901                 mutex_lock(&vmbus_connection.channel_mutex);
902                 if (vmbus_close_internal(cur_channel) == 0) {
903                         vmbus_free_ring(cur_channel);
904
905                         if (cur_channel->rescind)
906                                 hv_process_channel_removal(cur_channel);
907                 }
908                 mutex_unlock(&vmbus_connection.channel_mutex);
909         }
910
911         /*
912          * Now close the primary.
913          */
914         mutex_lock(&vmbus_connection.channel_mutex);
915         ret = vmbus_close_internal(channel);
916         mutex_unlock(&vmbus_connection.channel_mutex);
917
918         return ret;
919 }
920 EXPORT_SYMBOL_GPL(vmbus_disconnect_ring);
921
922 /*
923  * vmbus_close - Close the specified channel
924  */
925 void vmbus_close(struct vmbus_channel *channel)
926 {
927         if (vmbus_disconnect_ring(channel) == 0)
928                 vmbus_free_ring(channel);
929 }
930 EXPORT_SYMBOL_GPL(vmbus_close);
931
932 /**
933  * vmbus_sendpacket() - Send the specified buffer on the given channel
934  * @channel: Pointer to vmbus_channel structure
935  * @buffer: Pointer to the buffer you want to send the data from.
936  * @bufferlen: Maximum size of what the buffer holds.
937  * @requestid: Identifier of the request
938  * @type: Type of packet that is being sent e.g. negotiate, time
939  *        packet etc.
940  * @flags: 0 or VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
941  *
942  * Sends data in @buffer directly to Hyper-V via the vmbus.
943  * This will send the data unparsed to Hyper-V.
944  *
945  * Mainly used by Hyper-V drivers.
946  */
947 int vmbus_sendpacket(struct vmbus_channel *channel, void *buffer,
948                            u32 bufferlen, u64 requestid,
949                            enum vmbus_packet_type type, u32 flags)
950 {
951         struct vmpacket_descriptor desc;
952         u32 packetlen = sizeof(struct vmpacket_descriptor) + bufferlen;
953         u32 packetlen_aligned = ALIGN(packetlen, sizeof(u64));
954         struct kvec bufferlist[3];
955         u64 aligned_data = 0;
956         int num_vecs = ((bufferlen != 0) ? 3 : 1);
957
958
959         /* Setup the descriptor */
960         desc.type = type; /* VmbusPacketTypeDataInBand; */
961         desc.flags = flags; /* VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED; */
962         /* in 8-bytes granularity */
963         desc.offset8 = sizeof(struct vmpacket_descriptor) >> 3;
964         desc.len8 = (u16)(packetlen_aligned >> 3);
965         desc.trans_id = VMBUS_RQST_ERROR; /* will be updated in hv_ringbuffer_write() */
966
967         bufferlist[0].iov_base = &desc;
968         bufferlist[0].iov_len = sizeof(struct vmpacket_descriptor);
969         bufferlist[1].iov_base = buffer;
970         bufferlist[1].iov_len = bufferlen;
971         bufferlist[2].iov_base = &aligned_data;
972         bufferlist[2].iov_len = (packetlen_aligned - packetlen);
973
974         return hv_ringbuffer_write(channel, bufferlist, num_vecs, requestid);
975 }
976 EXPORT_SYMBOL(vmbus_sendpacket);
977
978 /*
979  * vmbus_sendpacket_pagebuffer - Send a range of single-page buffer
980  * packets using a GPADL Direct packet type. This interface allows you
981  * to control notifying the host. This will be useful for sending
982  * batched data. Also the sender can control the send flags
983  * explicitly.
984  */
985 int vmbus_sendpacket_pagebuffer(struct vmbus_channel *channel,
986                                 struct hv_page_buffer pagebuffers[],
987                                 u32 pagecount, void *buffer, u32 bufferlen,
988                                 u64 requestid)
989 {
990         int i;
991         struct vmbus_channel_packet_page_buffer desc;
992         u32 descsize;
993         u32 packetlen;
994         u32 packetlen_aligned;
995         struct kvec bufferlist[3];
996         u64 aligned_data = 0;
997
998         if (pagecount > MAX_PAGE_BUFFER_COUNT)
999                 return -EINVAL;
1000
1001         /*
1002          * Adjust the size down since vmbus_channel_packet_page_buffer is the
1003          * largest size we support
1004          */
1005         descsize = sizeof(struct vmbus_channel_packet_page_buffer) -
1006                           ((MAX_PAGE_BUFFER_COUNT - pagecount) *
1007                           sizeof(struct hv_page_buffer));
1008         packetlen = descsize + bufferlen;
1009         packetlen_aligned = ALIGN(packetlen, sizeof(u64));
1010
1011         /* Setup the descriptor */
1012         desc.type = VM_PKT_DATA_USING_GPA_DIRECT;
1013         desc.flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
1014         desc.dataoffset8 = descsize >> 3; /* in 8-bytes granularity */
1015         desc.length8 = (u16)(packetlen_aligned >> 3);
1016         desc.transactionid = VMBUS_RQST_ERROR; /* will be updated in hv_ringbuffer_write() */
1017         desc.reserved = 0;
1018         desc.rangecount = pagecount;
1019
1020         for (i = 0; i < pagecount; i++) {
1021                 desc.range[i].len = pagebuffers[i].len;
1022                 desc.range[i].offset = pagebuffers[i].offset;
1023                 desc.range[i].pfn        = pagebuffers[i].pfn;
1024         }
1025
1026         bufferlist[0].iov_base = &desc;
1027         bufferlist[0].iov_len = descsize;
1028         bufferlist[1].iov_base = buffer;
1029         bufferlist[1].iov_len = bufferlen;
1030         bufferlist[2].iov_base = &aligned_data;
1031         bufferlist[2].iov_len = (packetlen_aligned - packetlen);
1032
1033         return hv_ringbuffer_write(channel, bufferlist, 3, requestid);
1034 }
1035 EXPORT_SYMBOL_GPL(vmbus_sendpacket_pagebuffer);
1036
1037 /*
1038  * vmbus_sendpacket_multipagebuffer - Send a multi-page buffer packet
1039  * using a GPADL Direct packet type.
1040  * The buffer includes the vmbus descriptor.
1041  */
1042 int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
1043                               struct vmbus_packet_mpb_array *desc,
1044                               u32 desc_size,
1045                               void *buffer, u32 bufferlen, u64 requestid)
1046 {
1047         u32 packetlen;
1048         u32 packetlen_aligned;
1049         struct kvec bufferlist[3];
1050         u64 aligned_data = 0;
1051
1052         packetlen = desc_size + bufferlen;
1053         packetlen_aligned = ALIGN(packetlen, sizeof(u64));
1054
1055         /* Setup the descriptor */
1056         desc->type = VM_PKT_DATA_USING_GPA_DIRECT;
1057         desc->flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
1058         desc->dataoffset8 = desc_size >> 3; /* in 8-bytes granularity */
1059         desc->length8 = (u16)(packetlen_aligned >> 3);
1060         desc->transactionid = VMBUS_RQST_ERROR; /* will be updated in hv_ringbuffer_write() */
1061         desc->reserved = 0;
1062         desc->rangecount = 1;
1063
1064         bufferlist[0].iov_base = desc;
1065         bufferlist[0].iov_len = desc_size;
1066         bufferlist[1].iov_base = buffer;
1067         bufferlist[1].iov_len = bufferlen;
1068         bufferlist[2].iov_base = &aligned_data;
1069         bufferlist[2].iov_len = (packetlen_aligned - packetlen);
1070
1071         return hv_ringbuffer_write(channel, bufferlist, 3, requestid);
1072 }
1073 EXPORT_SYMBOL_GPL(vmbus_sendpacket_mpb_desc);
1074
1075 /**
1076  * __vmbus_recvpacket() - Retrieve the user packet on the specified channel
1077  * @channel: Pointer to vmbus_channel structure
1078  * @buffer: Pointer to the buffer you want to receive the data into.
1079  * @bufferlen: Maximum size of what the buffer can hold.
1080  * @buffer_actual_len: The actual size of the data after it was received.
1081  * @requestid: Identifier of the request
1082  * @raw: true means keep the vmpacket_descriptor header in the received data.
1083  *
1084  * Receives directly from the hyper-v vmbus and puts the data it received
1085  * into Buffer. This will receive the data unparsed from hyper-v.
1086  *
1087  * Mainly used by Hyper-V drivers.
1088  */
1089 static inline int
1090 __vmbus_recvpacket(struct vmbus_channel *channel, void *buffer,
1091                    u32 bufferlen, u32 *buffer_actual_len, u64 *requestid,
1092                    bool raw)
1093 {
1094         return hv_ringbuffer_read(channel, buffer, bufferlen,
1095                                   buffer_actual_len, requestid, raw);
1096
1097 }
1098
1099 int vmbus_recvpacket(struct vmbus_channel *channel, void *buffer,
1100                      u32 bufferlen, u32 *buffer_actual_len,
1101                      u64 *requestid)
1102 {
1103         return __vmbus_recvpacket(channel, buffer, bufferlen,
1104                                   buffer_actual_len, requestid, false);
1105 }
1106 EXPORT_SYMBOL(vmbus_recvpacket);
1107
1108 /*
1109  * vmbus_recvpacket_raw - Retrieve the raw packet on the specified channel
1110  */
1111 int vmbus_recvpacket_raw(struct vmbus_channel *channel, void *buffer,
1112                               u32 bufferlen, u32 *buffer_actual_len,
1113                               u64 *requestid)
1114 {
1115         return __vmbus_recvpacket(channel, buffer, bufferlen,
1116                                   buffer_actual_len, requestid, true);
1117 }
1118 EXPORT_SYMBOL_GPL(vmbus_recvpacket_raw);
1119
1120 /*
1121  * vmbus_next_request_id - Returns a new request id. It is also
1122  * the index at which the guest memory address is stored.
1123  * Uses a spin lock to avoid race conditions.
1124  * @rqstor: Pointer to the requestor struct
1125  * @rqst_add: Guest memory address to be stored in the array
1126  */
1127 u64 vmbus_next_request_id(struct vmbus_requestor *rqstor, u64 rqst_addr)
1128 {
1129         unsigned long flags;
1130         u64 current_id;
1131         const struct vmbus_channel *channel =
1132                 container_of(rqstor, const struct vmbus_channel, requestor);
1133
1134         /* Check rqstor has been initialized */
1135         if (!channel->rqstor_size)
1136                 return VMBUS_NO_RQSTOR;
1137
1138         spin_lock_irqsave(&rqstor->req_lock, flags);
1139         current_id = rqstor->next_request_id;
1140
1141         /* Requestor array is full */
1142         if (current_id >= rqstor->size) {
1143                 spin_unlock_irqrestore(&rqstor->req_lock, flags);
1144                 return VMBUS_RQST_ERROR;
1145         }
1146
1147         rqstor->next_request_id = rqstor->req_arr[current_id];
1148         rqstor->req_arr[current_id] = rqst_addr;
1149
1150         /* The already held spin lock provides atomicity */
1151         bitmap_set(rqstor->req_bitmap, current_id, 1);
1152
1153         spin_unlock_irqrestore(&rqstor->req_lock, flags);
1154
1155         /*
1156          * Cannot return an ID of 0, which is reserved for an unsolicited
1157          * message from Hyper-V.
1158          */
1159         return current_id + 1;
1160 }
1161 EXPORT_SYMBOL_GPL(vmbus_next_request_id);
1162
1163 /*
1164  * vmbus_request_addr - Returns the memory address stored at @trans_id
1165  * in @rqstor. Uses a spin lock to avoid race conditions.
1166  * @rqstor: Pointer to the requestor struct
1167  * @trans_id: Request id sent back from Hyper-V. Becomes the requestor's
1168  * next request id.
1169  */
1170 u64 vmbus_request_addr(struct vmbus_requestor *rqstor, u64 trans_id)
1171 {
1172         unsigned long flags;
1173         u64 req_addr;
1174         const struct vmbus_channel *channel =
1175                 container_of(rqstor, const struct vmbus_channel, requestor);
1176
1177         /* Check rqstor has been initialized */
1178         if (!channel->rqstor_size)
1179                 return VMBUS_NO_RQSTOR;
1180
1181         /* Hyper-V can send an unsolicited message with ID of 0 */
1182         if (!trans_id)
1183                 return trans_id;
1184
1185         spin_lock_irqsave(&rqstor->req_lock, flags);
1186
1187         /* Data corresponding to trans_id is stored at trans_id - 1 */
1188         trans_id--;
1189
1190         /* Invalid trans_id */
1191         if (trans_id >= rqstor->size || !test_bit(trans_id, rqstor->req_bitmap)) {
1192                 spin_unlock_irqrestore(&rqstor->req_lock, flags);
1193                 return VMBUS_RQST_ERROR;
1194         }
1195
1196         req_addr = rqstor->req_arr[trans_id];
1197         rqstor->req_arr[trans_id] = rqstor->next_request_id;
1198         rqstor->next_request_id = trans_id;
1199
1200         /* The already held spin lock provides atomicity */
1201         bitmap_clear(rqstor->req_bitmap, trans_id, 1);
1202
1203         spin_unlock_irqrestore(&rqstor->req_lock, flags);
1204         return req_addr;
1205 }
1206 EXPORT_SYMBOL_GPL(vmbus_request_addr);