projects/linux·blob master · refs

1475 lines · 41.9 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (c) 2009, Microsoft Corporation.
4 *
5 * Authors:
6 * Haiyang Zhang <[email protected]>
7 * Hank Janssen <[email protected]>
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/log2.h>
17#include <linux/module.h>
18#include <linux/hyperv.h>
19#include <linux/uio.h>
20#include <linux/interrupt.h>
21#include <linux/set_memory.h>
22#include <linux/vmalloc.h>
23#include <linux/export.h>
24#include <asm/page.h>
25#include <asm/mshyperv.h>
26
27#include "hyperv_vmbus.h"
28
29/*
30 * hv_gpadl_size - Return the real size of a gpadl, the size that Hyper-V uses
31 *
32 * For BUFFER gpadl, Hyper-V uses the exact same size as the guest does.
33 *
34 * For RING gpadl, in each ring, the guest uses one PAGE_SIZE as the header
35 * (because of the alignment requirement), however, the hypervisor only
36 * uses the first HV_HYP_PAGE_SIZE as the header, therefore leaving a
37 * (PAGE_SIZE - HV_HYP_PAGE_SIZE) gap. And since there are two rings in a
38 * ringbuffer, the total size for a RING gpadl that Hyper-V uses is the
39 * total size that the guest uses minus twice of the gap size.
40 */
41static inline u32 hv_gpadl_size(enum hv_gpadl_type type, u32 size)
42{
43 switch (type) {
44 case HV_GPADL_BUFFER:
45 case HV_GPADL_BUFFER_DECRYPTED:
46 return size;
47 case HV_GPADL_RING:
48 /* The size of a ringbuffer must be page-aligned */
49 BUG_ON(size % PAGE_SIZE);
50 /*
51 * Two things to notice here:
52 * 1) We're processing two ring buffers as a unit
53 * 2) We're skipping any space larger than HV_HYP_PAGE_SIZE in
54 * the first guest-size page of each of the two ring buffers.
55 * So we effectively subtract out two guest-size pages, and add
56 * back two Hyper-V size pages.
57 */
58 return size - 2 * (PAGE_SIZE - HV_HYP_PAGE_SIZE);
59 }
60 BUG();
61 return 0;
62}
63
64/*
65 * hv_ring_gpadl_send_hvpgoffset - Calculate the send offset (in unit of
66 * HV_HYP_PAGE) in a ring gpadl based on the
67 * offset in the guest
68 *
69 * @offset: the offset (in bytes) where the send ringbuffer starts in the
70 * virtual address space of the guest
71 */
72static inline u32 hv_ring_gpadl_send_hvpgoffset(u32 offset)
73{
74
75 /*
76 * For RING gpadl, in each ring, the guest uses one PAGE_SIZE as the
77 * header (because of the alignment requirement), however, the
78 * hypervisor only uses the first HV_HYP_PAGE_SIZE as the header,
79 * therefore leaving a (PAGE_SIZE - HV_HYP_PAGE_SIZE) gap.
80 *
81 * And to calculate the effective send offset in gpadl, we need to
82 * subtract this gap.
83 */
84 return (offset - (PAGE_SIZE - HV_HYP_PAGE_SIZE)) >> HV_HYP_PAGE_SHIFT;
85}
86
87/*
88 * hv_gpadl_hvpfn - Return the Hyper-V page PFN of the @i th Hyper-V page in
89 * the gpadl
90 *
91 * @type: the type of the gpadl
92 * @kbuffer: the pointer to the gpadl in the guest
93 * @size: the total size (in bytes) of the gpadl
94 * @send_offset: the offset (in bytes) where the send ringbuffer starts in the
95 * virtual address space of the guest
96 * @i: the index
97 */
98static inline u64 hv_gpadl_hvpfn(enum hv_gpadl_type type, void *kbuffer,
99 u32 size, u32 send_offset, int i)
100{
101 int send_idx = hv_ring_gpadl_send_hvpgoffset(send_offset);
102 unsigned long delta = 0UL;
103
104 switch (type) {
105 case HV_GPADL_BUFFER:
106 case HV_GPADL_BUFFER_DECRYPTED:
107 break;
108 case HV_GPADL_RING:
109 if (i == 0)
110 delta = 0;
111 else if (i <= send_idx)
112 delta = PAGE_SIZE - HV_HYP_PAGE_SIZE;
113 else
114 delta = 2 * (PAGE_SIZE - HV_HYP_PAGE_SIZE);
115 break;
116 default:
117 BUG();
118 break;
119 }
120
121 return virt_to_hvpfn(kbuffer + delta + (HV_HYP_PAGE_SIZE * i));
122}
123
124/*
125 * vmbus_setevent- Trigger an event notification on the specified
126 * channel.
127 */
128void vmbus_setevent(struct vmbus_channel *channel)
129{
130 struct hv_monitor_page *monitorpage;
131
132 trace_vmbus_setevent(channel);
133
134 /*
135 * For channels marked as in "low latency" mode
136 * bypass the monitor page mechanism.
137 */
138 if (channel->offermsg.monitor_allocated && !channel->low_latency) {
139 vmbus_send_interrupt(channel->offermsg.child_relid);
140
141 /* Get the child to parent monitor page */
142 monitorpage = vmbus_connection.monitor_pages[1];
143
144 sync_set_bit(channel->monitor_bit,
145 (unsigned long *)&monitorpage->trigger_group
146 [channel->monitor_grp].pending);
147
148 } else {
149 vmbus_set_event(channel);
150 }
151}
152EXPORT_SYMBOL_GPL(vmbus_setevent);
153
154/* vmbus_free_ring - drop mapping of ring buffer */
155void vmbus_free_ring(struct vmbus_channel *channel)
156{
157 hv_ringbuffer_cleanup(&channel->outbound);
158 hv_ringbuffer_cleanup(&channel->inbound);
159
160 if (channel->ringbuffer_page) {
161 /* In a CoCo VM leak the memory if it didn't get re-encrypted */
162 if (!channel->ringbuffer_gpadlhandle.decrypted)
163 __free_pages(channel->ringbuffer_page,
164 get_order(channel->ringbuffer_pagecount
165 << PAGE_SHIFT));
166 channel->ringbuffer_page = NULL;
167 }
168}
169EXPORT_SYMBOL_GPL(vmbus_free_ring);
170
171/* vmbus_alloc_ring - allocate and map pages for ring buffer */
172int vmbus_alloc_ring(struct vmbus_channel *newchannel,
173 u32 send_size, u32 recv_size)
174{
175 struct page *page;
176 int order;
177
178 if (send_size % PAGE_SIZE || recv_size % PAGE_SIZE)
179 return -EINVAL;
180
181 /* Allocate the ring buffer */
182 order = get_order(send_size + recv_size);
183 page = alloc_pages_node(cpu_to_node(newchannel->target_cpu),
184 GFP_KERNEL|__GFP_ZERO, order);
185
186 if (!page)
187 page = alloc_pages(GFP_KERNEL|__GFP_ZERO, order);
188
189 if (!page)
190 return -ENOMEM;
191
192 newchannel->ringbuffer_page = page;
193 newchannel->ringbuffer_pagecount = (send_size + recv_size) >> PAGE_SHIFT;
194 newchannel->ringbuffer_send_offset = send_size >> PAGE_SHIFT;
195
196 return 0;
197}
198EXPORT_SYMBOL_GPL(vmbus_alloc_ring);
199
200/* Used for Hyper-V Socket: a guest client's connect() to the host */
201int vmbus_send_tl_connect_request(const guid_t *shv_guest_servie_id,
202 const guid_t *shv_host_servie_id)
203{
204 struct vmbus_channel_tl_connect_request conn_msg;
205 int ret;
206
207 memset(&conn_msg, 0, sizeof(conn_msg));
208 conn_msg.header.msgtype = CHANNELMSG_TL_CONNECT_REQUEST;
209 conn_msg.guest_endpoint_id = *shv_guest_servie_id;
210 conn_msg.host_service_id = *shv_host_servie_id;
211
212 ret = vmbus_post_msg(&conn_msg, sizeof(conn_msg), true);
213
214 trace_vmbus_send_tl_connect_request(&conn_msg, ret);
215
216 return ret;
217}
218EXPORT_SYMBOL_GPL(vmbus_send_tl_connect_request);
219
220static int send_modifychannel_without_ack(struct vmbus_channel *channel, u32 target_vp)
221{
222 struct vmbus_channel_modifychannel msg;
223 int ret;
224
225 memset(&msg, 0, sizeof(msg));
226 msg.header.msgtype = CHANNELMSG_MODIFYCHANNEL;
227 msg.child_relid = channel->offermsg.child_relid;
228 msg.target_vp = target_vp;
229
230 ret = vmbus_post_msg(&msg, sizeof(msg), true);
231 trace_vmbus_send_modifychannel(&msg, ret);
232
233 return ret;
234}
235
236static int send_modifychannel_with_ack(struct vmbus_channel *channel, u32 target_vp)
237{
238 struct vmbus_channel_modifychannel *msg;
239 struct vmbus_channel_msginfo *info;
240 unsigned long flags;
241 int ret;
242
243 info = kzalloc(sizeof(struct vmbus_channel_msginfo) +
244 sizeof(struct vmbus_channel_modifychannel),
245 GFP_KERNEL);
246 if (!info)
247 return -ENOMEM;
248
249 init_completion(&info->waitevent);
250 info->waiting_channel = channel;
251
252 msg = (struct vmbus_channel_modifychannel *)info->msg;
253 msg->header.msgtype = CHANNELMSG_MODIFYCHANNEL;
254 msg->child_relid = channel->offermsg.child_relid;
255 msg->target_vp = target_vp;
256
257 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
258 list_add_tail(&info->msglistentry, &vmbus_connection.chn_msg_list);
259 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
260
261 ret = vmbus_post_msg(msg, sizeof(*msg), true);
262 trace_vmbus_send_modifychannel(msg, ret);
263 if (ret != 0) {
264 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
265 list_del(&info->msglistentry);
266 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
267 goto free_info;
268 }
269
270 /*
271 * Release channel_mutex; otherwise, vmbus_onoffer_rescind() could block on
272 * the mutex and be unable to signal the completion.
273 *
274 * See the caller target_cpu_store() for information about the usage of the
275 * mutex.
276 */
277 mutex_unlock(&vmbus_connection.channel_mutex);
278 wait_for_completion(&info->waitevent);
279 mutex_lock(&vmbus_connection.channel_mutex);
280
281 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
282 list_del(&info->msglistentry);
283 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
284
285 if (info->response.modify_response.status)
286 ret = -EAGAIN;
287
288free_info:
289 kfree(info);
290 return ret;
291}
292
293/*
294 * Set/change the vCPU (@target_vp) the channel (@child_relid) will interrupt.
295 *
296 * CHANNELMSG_MODIFYCHANNEL messages are aynchronous. When VMbus version 5.3
297 * or later is negotiated, Hyper-V always sends an ACK in response to such a
298 * message. For VMbus version 5.2 and earlier, it never sends an ACK. With-
299 * out an ACK, we can not know when the host will stop interrupting the "old"
300 * vCPU and start interrupting the "new" vCPU for the given channel.
301 *
302 * The CHANNELMSG_MODIFYCHANNEL message type is supported since VMBus version
303 * VERSION_WIN10_V4_1.
304 */
305int vmbus_send_modifychannel(struct vmbus_channel *channel, u32 target_vp)
306{
307 if (vmbus_proto_version >= VERSION_WIN10_V5_3)
308 return send_modifychannel_with_ack(channel, target_vp);
309 return send_modifychannel_without_ack(channel, target_vp);
310}
311EXPORT_SYMBOL_GPL(vmbus_send_modifychannel);
312
313/*
314 * create_gpadl_header - Creates a gpadl for the specified buffer
315 */
316static int create_gpadl_header(enum hv_gpadl_type type, void *kbuffer,
317 u32 size, u32 send_offset,
318 struct vmbus_channel_msginfo **msginfo)
319{
320 int i;
321 int pagecount;
322 struct vmbus_channel_gpadl_header *gpadl_header;
323 struct vmbus_channel_gpadl_body *gpadl_body;
324 struct vmbus_channel_msginfo *msgheader;
325 struct vmbus_channel_msginfo *msgbody = NULL;
326 u32 msgsize;
327
328 int pfnsum, pfncount, pfnleft, pfncurr, pfnsize;
329
330 pagecount = hv_gpadl_size(type, size) >> HV_HYP_PAGE_SHIFT;
331
332 pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
333 sizeof(struct vmbus_channel_gpadl_header) -
334 sizeof(struct gpa_range);
335 pfncount = umin(pagecount, pfnsize / sizeof(u64));
336
337 msgsize = sizeof(struct vmbus_channel_msginfo) +
338 sizeof(struct vmbus_channel_gpadl_header) +
339 sizeof(struct gpa_range) + pfncount * sizeof(u64);
340 msgheader = kzalloc(msgsize, GFP_KERNEL);
341 if (!msgheader)
342 return -ENOMEM;
343
344 INIT_LIST_HEAD(&msgheader->submsglist);
345 msgheader->msgsize = msgsize;
346
347 gpadl_header = (struct vmbus_channel_gpadl_header *)
348 msgheader->msg;
349 gpadl_header->rangecount = 1;
350 gpadl_header->range_buflen = sizeof(struct gpa_range) +
351 pagecount * sizeof(u64);
352 gpadl_header->range[0].byte_offset = 0;
353 gpadl_header->range[0].byte_count = hv_gpadl_size(type, size);
354 for (i = 0; i < pfncount; i++)
355 gpadl_header->range[0].pfn_array[i] = hv_gpadl_hvpfn(
356 type, kbuffer, size, send_offset, i);
357 *msginfo = msgheader;
358
359 pfnsum = pfncount;
360 pfnleft = pagecount - pfncount;
361
362 /* how many pfns can we fit in a body message */
363 pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
364 sizeof(struct vmbus_channel_gpadl_body);
365 pfncount = pfnsize / sizeof(u64);
366
367 /*
368 * If pfnleft is zero, everything fits in the header and no body
369 * messages are needed
370 */
371 while (pfnleft) {
372 pfncurr = umin(pfncount, pfnleft);
373 msgsize = sizeof(struct vmbus_channel_msginfo) +
374 sizeof(struct vmbus_channel_gpadl_body) +
375 pfncurr * sizeof(u64);
376 msgbody = kzalloc(msgsize, GFP_KERNEL);
377
378 if (!msgbody) {
379 struct vmbus_channel_msginfo *pos = NULL;
380 struct vmbus_channel_msginfo *tmp = NULL;
381 /*
382 * Free up all the allocated messages.
383 */
384 list_for_each_entry_safe(pos, tmp,
385 &msgheader->submsglist,
386 msglistentry) {
387
388 list_del(&pos->msglistentry);
389 kfree(pos);
390 }
391 kfree(msgheader);
392 return -ENOMEM;
393 }
394
395 msgbody->msgsize = msgsize;
396 gpadl_body = (struct vmbus_channel_gpadl_body *)msgbody->msg;
397
398 /*
399 * Gpadl is u32 and we are using a pointer which could
400 * be 64-bit
401 * This is governed by the guest/host protocol and
402 * so the hypervisor guarantees that this is ok.
403 */
404 for (i = 0; i < pfncurr; i++)
405 gpadl_body->pfn[i] = hv_gpadl_hvpfn(type,
406 kbuffer, size, send_offset, pfnsum + i);
407
408 /* add to msg header */
409 list_add_tail(&msgbody->msglistentry, &msgheader->submsglist);
410 pfnsum += pfncurr;
411 pfnleft -= pfncurr;
412 }
413
414 return 0;
415}
416
417static void vmbus_free_channel_msginfo(struct vmbus_channel_msginfo *msginfo)
418{
419 struct vmbus_channel_msginfo *submsginfo, *tmp;
420
421 if (!msginfo)
422 return;
423
424 list_for_each_entry_safe(submsginfo, tmp, &msginfo->submsglist,
425 msglistentry) {
426 kfree(submsginfo);
427 }
428
429 kfree(msginfo);
430}
431
432/*
433 * __vmbus_establish_gpadl - Establish a GPADL for a buffer or ringbuffer
434 *
435 * @channel: a channel
436 * @type: the type of the corresponding GPADL, only meaningful for the guest.
437 * @kbuffer: from kmalloc or vmalloc
438 * @size: page-size multiple
439 * @send_offset: the offset (in bytes) where the send ring buffer starts,
440 * should be 0 for BUFFER type gpadl
441 * @gpadl_handle: some funky thing
442 */
443static int __vmbus_establish_gpadl(struct vmbus_channel *channel,
444 enum hv_gpadl_type type, void *kbuffer,
445 u32 size, u32 send_offset,
446 struct vmbus_gpadl *gpadl)
447{
448 struct vmbus_channel_gpadl_header *gpadlmsg;
449 struct vmbus_channel_gpadl_body *gpadl_body;
450 struct vmbus_channel_msginfo *msginfo = NULL;
451 struct vmbus_channel_msginfo *submsginfo;
452 struct list_head *curr;
453 u32 next_gpadl_handle;
454 unsigned long flags;
455 int ret = 0;
456
457 next_gpadl_handle =
458 (atomic_inc_return(&vmbus_connection.next_gpadl_handle) - 1);
459
460 ret = create_gpadl_header(type, kbuffer, size, send_offset, &msginfo);
461 if (ret) {
462 gpadl->decrypted = false;
463 return ret;
464 }
465
466 gpadl->decrypted = !((channel->co_external_memory && type == HV_GPADL_BUFFER) ||
467 (channel->co_ring_buffer && type == HV_GPADL_RING) ||
468 (type == HV_GPADL_BUFFER_DECRYPTED));
469 if (gpadl->decrypted) {
470 /*
471 * The "decrypted" flag being true assumes that set_memory_decrypted() succeeds.
472 * But if it fails, the encryption state of the memory is unknown. In that case,
473 * leave "decrypted" as true to ensure the memory is leaked instead of going back
474 * on the free list.
475 */
476 ret = set_memory_decrypted((unsigned long)kbuffer,
477 PFN_UP(size));
478 if (ret) {
479 dev_warn(&channel->device_obj->device,
480 "Failed to set host visibility for new GPADL %d.\n",
481 ret);
482 vmbus_free_channel_msginfo(msginfo);
483 return ret;
484 }
485 }
486
487 init_completion(&msginfo->waitevent);
488 msginfo->waiting_channel = channel;
489
490 gpadlmsg = (struct vmbus_channel_gpadl_header *)msginfo->msg;
491 gpadlmsg->header.msgtype = CHANNELMSG_GPADL_HEADER;
492 gpadlmsg->child_relid = channel->offermsg.child_relid;
493 gpadlmsg->gpadl = next_gpadl_handle;
494
495
496 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
497 list_add_tail(&msginfo->msglistentry,
498 &vmbus_connection.chn_msg_list);
499
500 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
501
502 if (channel->rescind) {
503 ret = -ENODEV;
504 goto cleanup;
505 }
506
507 ret = vmbus_post_msg(gpadlmsg, msginfo->msgsize -
508 sizeof(*msginfo), true);
509
510 trace_vmbus_establish_gpadl_header(gpadlmsg, ret);
511
512 if (ret != 0)
513 goto cleanup;
514
515 list_for_each(curr, &msginfo->submsglist) {
516 submsginfo = (struct vmbus_channel_msginfo *)curr;
517 gpadl_body =
518 (struct vmbus_channel_gpadl_body *)submsginfo->msg;
519
520 gpadl_body->header.msgtype =
521 CHANNELMSG_GPADL_BODY;
522 gpadl_body->gpadl = next_gpadl_handle;
523
524 ret = vmbus_post_msg(gpadl_body,
525 submsginfo->msgsize - sizeof(*submsginfo),
526 true);
527
528 trace_vmbus_establish_gpadl_body(gpadl_body, ret);
529
530 if (ret != 0)
531 goto cleanup;
532
533 }
534 wait_for_completion(&msginfo->waitevent);
535
536 if (msginfo->response.gpadl_created.creation_status != 0) {
537 pr_err("Failed to establish GPADL: err = 0x%x\n",
538 msginfo->response.gpadl_created.creation_status);
539
540 ret = -EDQUOT;
541 goto cleanup;
542 }
543
544 if (channel->rescind) {
545 ret = -ENODEV;
546 goto cleanup;
547 }
548
549 /* At this point, we received the gpadl created msg */
550 gpadl->gpadl_handle = gpadlmsg->gpadl;
551 gpadl->buffer = kbuffer;
552 gpadl->size = size;
553
554
555cleanup:
556 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
557 list_del(&msginfo->msglistentry);
558 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
559
560 vmbus_free_channel_msginfo(msginfo);
561
562 if (ret) {
563 /*
564 * If set_memory_encrypted() fails, the decrypted flag is
565 * left as true so the memory is leaked instead of being
566 * put back on the free list.
567 */
568 if (gpadl->decrypted) {
569 if (!set_memory_encrypted((unsigned long)kbuffer, PFN_UP(size)))
570 gpadl->decrypted = false;
571 }
572 }
573
574 return ret;
575}
576
577/*
578 * vmbus_establish_gpadl - Establish a GPADL for the specified buffer
579 *
580 * @channel: a channel
581 * @kbuffer: from kmalloc or vmalloc
582 * @size: page-size multiple
583 * @gpadl: output gpadl
584 */
585int vmbus_establish_gpadl(struct vmbus_channel *channel, void *kbuffer,
586 u32 size, struct vmbus_gpadl *gpadl)
587{
588 return __vmbus_establish_gpadl(channel, HV_GPADL_BUFFER, kbuffer, size,
589 0U, gpadl);
590}
591EXPORT_SYMBOL_GPL(vmbus_establish_gpadl);
592
593/*
594 * vmbus_establish_gpadl_caller_decrypted - Establish a GPADL for a buffer
595 * that has already been decrypted by the caller.
596 *
597 * @channel: a channel
598 * @kbuffer: from kmalloc or vmalloc; must already be decrypted by the caller
599 * @size: page-size multiple
600 * @gpadl: output gpadl
601 *
602 * The caller is responsible for re-encrypting the buffer before freeing it.
603 */
604int vmbus_establish_gpadl_caller_decrypted(struct vmbus_channel *channel,
605 void *kbuffer, u32 size,
606 struct vmbus_gpadl *gpadl)
607{
608 return __vmbus_establish_gpadl(channel, HV_GPADL_BUFFER_DECRYPTED,
609 kbuffer, size, 0U, gpadl);
610}
611EXPORT_SYMBOL_GPL(vmbus_establish_gpadl_caller_decrypted);
612
613/**
614 * vmbus_free_buffer - release a buffer allocated by vmbus_alloc_buffer().
615 *
616 * @addr: buffer address, or NULL if none was allocated (e.g. cleanup from a
617 * failed allocation)
618 * @chunks: chunks array from vmbus_alloc_buffer(), or NULL
619 * @chunk_cnt: number of entries in @chunks
620 *
621 * When @chunks is NULL the buffer is a plain vzalloc() allocation.
622 *
623 * Otherwise tear down the vmap, and for each chunk re-encrypt and free
624 * the underlying pages. Any chunk that cannot be re-encrypted is leaked.
625 */
626void vmbus_free_buffer(void *addr, struct page **chunks, u32 chunk_cnt)
627{
628 u32 i;
629
630 if (!chunks) {
631 vfree(addr);
632 return;
633 }
634
635 vunmap(addr);
636
637 for (i = 0; i < chunk_cnt; i++) {
638 unsigned long vaddr =
639 (unsigned long)page_address(chunks[i]);
640 unsigned int order = folio_order(page_folio(chunks[i]));
641
642 if (set_memory_encrypted(vaddr, 1U << order))
643 continue;
644 __free_pages(chunks[i], order);
645 }
646
647 kvfree(chunks);
648}
649EXPORT_SYMBOL_GPL(vmbus_free_buffer);
650
651/**
652 * vmbus_alloc_buffer - allocate a host-visible, virtually-contiguous buffer.
653 *
654 * @channel: the channel the buffer will be attached to
655 * @size: requested buffer size in bytes (will be rounded up to PAGE_SIZE)
656 * @chunks_out: on success, set to the array of underlying chunks, or NULL when
657 * the buffer was allocated with vzalloc()
658 * @chunk_cnt_out: on success, set to the number of chunks
659 *
660 * Buffers not requiring decryption are allocated with vzalloc().
661 *
662 * Buffers requiring decryption are allocated as a series of
663 * physically-contiguous chunks, starting at MAX_PAGE_ORDER and falling back to
664 * smaller orders on allocation failure. Each chunk is transitioned to
665 * host-visible via set_memory_decrypted() on its direct-map address, then all
666 * chunks are combined into a virtually-contiguous range via vmap().
667 *
668 * Return: the buffer's virtual address, or NULL on failure.
669 */
670void *vmbus_alloc_buffer(struct vmbus_channel *channel,
671 u32 size,
672 struct page ***chunks_out,
673 u32 *chunk_cnt_out)
674{
675 unsigned long nr_pages = PFN_UP(size);
676 unsigned long remaining = nr_pages;
677 unsigned long page_idx = 0;
678 struct page **chunks = NULL;
679 struct page **pages = NULL;
680 int order = MAX_PAGE_ORDER;
681 u32 chunk_cnt = 0;
682 void *addr;
683 u32 i;
684 int ret;
685
686 *chunks_out = NULL;
687 *chunk_cnt_out = 0;
688
689 if (!nr_pages)
690 return NULL;
691
692 /* If the buffer does not need to be decrypted, just use vzalloc() */
693 if (!hv_is_isolation_supported() || channel->co_external_memory)
694 return vzalloc(nr_pages << PAGE_SHIFT);
695
696 /* Worst case: every chunk is a single page. */
697 chunks = kvmalloc_objs(*chunks, nr_pages, GFP_KERNEL | __GFP_ZERO);
698 if (!chunks)
699 goto err;
700
701 pages = kvmalloc_objs(*pages, nr_pages);
702 if (!pages)
703 goto err;
704
705 while (remaining) {
706 struct page *page;
707 gfp_t gfp;
708
709 order = min(order, ilog2(remaining));
710
711 /*
712 * Use __GFP_NORETRY | __GFP_NOWARN to avoid OOM-killing,
713 * but try harder at order 0 since that is the final
714 * fallback.
715 * __GFP_COMP stores order information in the page folio.
716 */
717 gfp = GFP_KERNEL | __GFP_ZERO;
718 if (order)
719 gfp |= __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN;
720
721 page = alloc_pages_node(cpu_to_node(channel->target_cpu),
722 gfp, order);
723 if (!page) {
724 if (!order--)
725 goto err;
726 continue;
727 }
728
729 ret = set_memory_decrypted((unsigned long)page_address(page),
730 1U << order);
731 if (ret) {
732 /*
733 * set_memory_decrypted() failed; the page state is
734 * unknown so it must be leaked rather than freed.
735 */
736 goto err;
737 }
738
739 chunks[chunk_cnt++] = page;
740
741 for (i = 0; i < (1U << order); i++)
742 pages[page_idx++] = page + i;
743
744 remaining -= 1U << order;
745 }
746
747 addr = vmap(pages, nr_pages, VM_MAP, pgprot_decrypted(PAGE_KERNEL));
748 if (!addr)
749 goto err;
750
751 memset(addr, 0, nr_pages << PAGE_SHIFT);
752
753 kvfree(pages);
754 *chunks_out = chunks;
755 *chunk_cnt_out = chunk_cnt;
756 return addr;
757
758err:
759 kvfree(pages);
760 vmbus_free_buffer(NULL, chunks, chunk_cnt);
761 return NULL;
762}
763EXPORT_SYMBOL_GPL(vmbus_alloc_buffer);
764
765/**
766 * request_arr_init - Allocates memory for the requestor array. Each slot
767 * keeps track of the next available slot in the array. Initially, each
768 * slot points to the next one (as in a Linked List). The last slot
769 * does not point to anything, so its value is U64_MAX by default.
770 * @size: The size of the array
771 */
772static u64 *request_arr_init(u32 size)
773{
774 int i;
775 u64 *req_arr;
776
777 req_arr = kcalloc(size, sizeof(u64), GFP_KERNEL);
778 if (!req_arr)
779 return NULL;
780
781 for (i = 0; i < size - 1; i++)
782 req_arr[i] = i + 1;
783
784 /* Last slot (no more available slots) */
785 req_arr[i] = U64_MAX;
786
787 return req_arr;
788}
789
790/*
791 * vmbus_alloc_requestor - Initializes @rqstor's fields.
792 * Index 0 is the first free slot
793 * @size: Size of the requestor array
794 */
795static int vmbus_alloc_requestor(struct vmbus_requestor *rqstor, u32 size)
796{
797 u64 *rqst_arr;
798 unsigned long *bitmap;
799
800 rqst_arr = request_arr_init(size);
801 if (!rqst_arr)
802 return -ENOMEM;
803
804 bitmap = bitmap_zalloc(size, GFP_KERNEL);
805 if (!bitmap) {
806 kfree(rqst_arr);
807 return -ENOMEM;
808 }
809
810 rqstor->req_arr = rqst_arr;
811 rqstor->req_bitmap = bitmap;
812 rqstor->size = size;
813 rqstor->next_request_id = 0;
814 spin_lock_init(&rqstor->req_lock);
815
816 return 0;
817}
818
819/*
820 * vmbus_free_requestor - Frees memory allocated for @rqstor
821 * @rqstor: Pointer to the requestor struct
822 */
823static void vmbus_free_requestor(struct vmbus_requestor *rqstor)
824{
825 kfree(rqstor->req_arr);
826 bitmap_free(rqstor->req_bitmap);
827}
828
829static int __vmbus_open(struct vmbus_channel *newchannel,
830 void *userdata, u32 userdatalen,
831 void (*onchannelcallback)(void *context), void *context)
832{
833 struct vmbus_channel_open_channel *open_msg;
834 struct vmbus_channel_msginfo *open_info = NULL;
835 struct page *page = newchannel->ringbuffer_page;
836 u32 send_pages, recv_pages;
837 unsigned long flags;
838 int err;
839
840 if (userdatalen > MAX_USER_DEFINED_BYTES)
841 return -EINVAL;
842
843 send_pages = newchannel->ringbuffer_send_offset;
844 recv_pages = newchannel->ringbuffer_pagecount - send_pages;
845
846 if (newchannel->state != CHANNEL_OPEN_STATE)
847 return -EINVAL;
848
849 /* Create and init requestor */
850 if (newchannel->rqstor_size) {
851 if (vmbus_alloc_requestor(&newchannel->requestor, newchannel->rqstor_size))
852 return -ENOMEM;
853 }
854
855 newchannel->state = CHANNEL_OPENING_STATE;
856 newchannel->onchannel_callback = onchannelcallback;
857 newchannel->channel_callback_context = context;
858
859 if (!newchannel->max_pkt_size)
860 newchannel->max_pkt_size = VMBUS_DEFAULT_MAX_PKT_SIZE;
861
862 /* Establish the gpadl for the ring buffer */
863 newchannel->ringbuffer_gpadlhandle.gpadl_handle = 0;
864
865 err = __vmbus_establish_gpadl(newchannel, HV_GPADL_RING,
866 page_address(newchannel->ringbuffer_page),
867 (send_pages + recv_pages) << PAGE_SHIFT,
868 newchannel->ringbuffer_send_offset << PAGE_SHIFT,
869 &newchannel->ringbuffer_gpadlhandle);
870 if (err)
871 goto error_clean_ring;
872
873 err = hv_ringbuffer_init(&newchannel->outbound,
874 page, send_pages, 0, newchannel->co_ring_buffer);
875 if (err)
876 goto error_free_gpadl;
877
878 err = hv_ringbuffer_init(&newchannel->inbound, &page[send_pages],
879 recv_pages, newchannel->max_pkt_size,
880 newchannel->co_ring_buffer);
881 if (err)
882 goto error_free_gpadl;
883
884 /* Create and init the channel open message */
885 open_info = kzalloc(sizeof(*open_info) +
886 sizeof(struct vmbus_channel_open_channel),
887 GFP_KERNEL);
888 if (!open_info) {
889 err = -ENOMEM;
890 goto error_free_gpadl;
891 }
892
893 init_completion(&open_info->waitevent);
894 open_info->waiting_channel = newchannel;
895
896 open_msg = (struct vmbus_channel_open_channel *)open_info->msg;
897 open_msg->header.msgtype = CHANNELMSG_OPENCHANNEL;
898 open_msg->openid = newchannel->offermsg.child_relid;
899 open_msg->child_relid = newchannel->offermsg.child_relid;
900 open_msg->ringbuffer_gpadlhandle
901 = newchannel->ringbuffer_gpadlhandle.gpadl_handle;
902 /*
903 * The unit of ->downstream_ringbuffer_pageoffset is HV_HYP_PAGE and
904 * the unit of ->ringbuffer_send_offset (i.e. send_pages) is PAGE, so
905 * here we calculate it into HV_HYP_PAGE.
906 */
907 open_msg->downstream_ringbuffer_pageoffset =
908 hv_ring_gpadl_send_hvpgoffset(send_pages << PAGE_SHIFT);
909 open_msg->target_vp = hv_cpu_number_to_vp_number(newchannel->target_cpu);
910
911 if (userdatalen)
912 memcpy(open_msg->userdata, userdata, userdatalen);
913
914 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
915 list_add_tail(&open_info->msglistentry,
916 &vmbus_connection.chn_msg_list);
917 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
918
919 if (newchannel->rescind) {
920 err = -ENODEV;
921 goto error_clean_msglist;
922 }
923
924 err = vmbus_post_msg(open_msg,
925 sizeof(struct vmbus_channel_open_channel), true);
926
927 trace_vmbus_open(open_msg, err);
928
929 if (err != 0)
930 goto error_clean_msglist;
931
932 wait_for_completion(&open_info->waitevent);
933
934 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
935 list_del(&open_info->msglistentry);
936 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
937
938 if (newchannel->rescind) {
939 err = -ENODEV;
940 goto error_free_info;
941 }
942
943 if (open_info->response.open_result.status) {
944 err = -EAGAIN;
945 goto error_free_info;
946 }
947
948 newchannel->state = CHANNEL_OPENED_STATE;
949 kfree(open_info);
950 return 0;
951
952error_clean_msglist:
953 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
954 list_del(&open_info->msglistentry);
955 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
956error_free_info:
957 kfree(open_info);
958error_free_gpadl:
959 vmbus_teardown_gpadl(newchannel, &newchannel->ringbuffer_gpadlhandle);
960error_clean_ring:
961 hv_ringbuffer_cleanup(&newchannel->outbound);
962 hv_ringbuffer_cleanup(&newchannel->inbound);
963 vmbus_free_requestor(&newchannel->requestor);
964 newchannel->state = CHANNEL_OPEN_STATE;
965 return err;
966}
967
968/*
969 * vmbus_connect_ring - Open the channel but reuse ring buffer
970 */
971int vmbus_connect_ring(struct vmbus_channel *newchannel,
972 void (*onchannelcallback)(void *context), void *context)
973{
974 return __vmbus_open(newchannel, NULL, 0, onchannelcallback, context);
975}
976EXPORT_SYMBOL_GPL(vmbus_connect_ring);
977
978/*
979 * vmbus_open - Open the specified channel.
980 */
981int vmbus_open(struct vmbus_channel *newchannel,
982 u32 send_ringbuffer_size, u32 recv_ringbuffer_size,
983 void *userdata, u32 userdatalen,
984 void (*onchannelcallback)(void *context), void *context)
985{
986 int err;
987
988 err = vmbus_alloc_ring(newchannel, send_ringbuffer_size,
989 recv_ringbuffer_size);
990 if (err)
991 return err;
992
993 err = __vmbus_open(newchannel, userdata, userdatalen,
994 onchannelcallback, context);
995 if (err)
996 vmbus_free_ring(newchannel);
997
998 return err;
999}
1000EXPORT_SYMBOL_GPL(vmbus_open);
1001
1002/*
1003 * vmbus_teardown_gpadl -Teardown the specified GPADL handle
1004 */
1005int vmbus_teardown_gpadl(struct vmbus_channel *channel, struct vmbus_gpadl *gpadl)
1006{
1007 struct vmbus_channel_gpadl_teardown *msg;
1008 struct vmbus_channel_msginfo *info;
1009 unsigned long flags;
1010 int ret;
1011
1012 info = kzalloc(sizeof(*info) +
1013 sizeof(struct vmbus_channel_gpadl_teardown), GFP_KERNEL);
1014 if (!info)
1015 return -ENOMEM;
1016
1017 init_completion(&info->waitevent);
1018 info->waiting_channel = channel;
1019
1020 msg = (struct vmbus_channel_gpadl_teardown *)info->msg;
1021
1022 msg->header.msgtype = CHANNELMSG_GPADL_TEARDOWN;
1023 msg->child_relid = channel->offermsg.child_relid;
1024 msg->gpadl = gpadl->gpadl_handle;
1025
1026 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1027 list_add_tail(&info->msglistentry,
1028 &vmbus_connection.chn_msg_list);
1029 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1030
1031 if (channel->rescind)
1032 goto post_msg_err;
1033
1034 ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_gpadl_teardown),
1035 true);
1036
1037 trace_vmbus_teardown_gpadl(msg, ret);
1038
1039 if (ret)
1040 goto post_msg_err;
1041
1042 wait_for_completion(&info->waitevent);
1043
1044 gpadl->gpadl_handle = 0;
1045
1046post_msg_err:
1047 /*
1048 * If the channel has been rescinded;
1049 * we will be awakened by the rescind
1050 * handler; set the error code to zero so we don't leak memory.
1051 */
1052 if (channel->rescind)
1053 ret = 0;
1054
1055 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1056 list_del(&info->msglistentry);
1057 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1058
1059 kfree(info);
1060
1061 if (gpadl->decrypted)
1062 ret = set_memory_encrypted((unsigned long)gpadl->buffer,
1063 PFN_UP(gpadl->size));
1064 else
1065 ret = 0;
1066 if (ret)
1067 pr_warn("Fail to set mem host visibility in GPADL teardown %d.\n", ret);
1068
1069 gpadl->decrypted = ret;
1070
1071 return ret;
1072}
1073EXPORT_SYMBOL_GPL(vmbus_teardown_gpadl);
1074
1075void vmbus_reset_channel_cb(struct vmbus_channel *channel)
1076{
1077 unsigned long flags;
1078
1079 /*
1080 * vmbus_on_event(), running in the per-channel tasklet, can race
1081 * with vmbus_close_internal() in the case of SMP guest, e.g., when
1082 * the former is accessing channel->inbound.ring_buffer, the latter
1083 * could be freeing the ring_buffer pages, so here we must stop it
1084 * first.
1085 *
1086 * vmbus_chan_sched() might call the netvsc driver callback function
1087 * that ends up scheduling NAPI work that accesses the ring buffer.
1088 * At this point, we have to ensure that any such work is completed
1089 * and that the channel ring buffer is no longer being accessed, cf.
1090 * the calls to napi_disable() in netvsc_device_remove().
1091 */
1092 tasklet_disable(&channel->callback_event);
1093
1094 /* See the inline comments in vmbus_chan_sched(). */
1095 spin_lock_irqsave(&channel->sched_lock, flags);
1096 channel->onchannel_callback = NULL;
1097 spin_unlock_irqrestore(&channel->sched_lock, flags);
1098
1099 channel->sc_creation_callback = NULL;
1100
1101 /* Re-enable tasklet for use on re-open */
1102 tasklet_enable(&channel->callback_event);
1103}
1104
1105static int vmbus_close_internal(struct vmbus_channel *channel)
1106{
1107 struct vmbus_channel_close_channel *msg;
1108 int ret;
1109
1110 vmbus_reset_channel_cb(channel);
1111
1112 /*
1113 * In case a device driver's probe() fails (e.g.,
1114 * util_probe() -> vmbus_open() returns -ENOMEM) and the device is
1115 * rescinded later (e.g., we dynamically disable an Integrated Service
1116 * in Hyper-V Manager), the driver's remove() invokes vmbus_close():
1117 * here we should skip most of the below cleanup work.
1118 */
1119 if (channel->state != CHANNEL_OPENED_STATE)
1120 return -EINVAL;
1121
1122 channel->state = CHANNEL_OPEN_STATE;
1123
1124 /* Send a closing message */
1125
1126 msg = &channel->close_msg;
1127
1128 msg->header.msgtype = CHANNELMSG_CLOSECHANNEL;
1129 msg->child_relid = channel->offermsg.child_relid;
1130
1131 ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_close_channel),
1132 true);
1133
1134 trace_vmbus_close_internal(msg, ret);
1135
1136 if (ret) {
1137 pr_err("Close failed: close post msg return is %d\n", ret);
1138 /*
1139 * If we failed to post the close msg,
1140 * it is perhaps better to leak memory.
1141 */
1142 }
1143
1144 /* Tear down the gpadl for the channel's ring buffer */
1145 else if (channel->ringbuffer_gpadlhandle.gpadl_handle) {
1146 ret = vmbus_teardown_gpadl(channel, &channel->ringbuffer_gpadlhandle);
1147 if (ret) {
1148 pr_err("Close failed: teardown gpadl return %d\n", ret);
1149 /*
1150 * If we failed to teardown gpadl,
1151 * it is perhaps better to leak memory.
1152 */
1153 }
1154 }
1155
1156 if (!ret)
1157 vmbus_free_requestor(&channel->requestor);
1158
1159 return ret;
1160}
1161
1162/* disconnect ring - close all channels */
1163int vmbus_disconnect_ring(struct vmbus_channel *channel)
1164{
1165 struct vmbus_channel *cur_channel, *tmp;
1166 int ret;
1167
1168 if (channel->primary_channel != NULL)
1169 return -EINVAL;
1170
1171 list_for_each_entry_safe(cur_channel, tmp, &channel->sc_list, sc_list) {
1172 if (cur_channel->rescind)
1173 wait_for_completion(&cur_channel->rescind_event);
1174
1175 mutex_lock(&vmbus_connection.channel_mutex);
1176 if (vmbus_close_internal(cur_channel) == 0) {
1177 vmbus_free_ring(cur_channel);
1178
1179 if (cur_channel->rescind)
1180 hv_process_channel_removal(cur_channel);
1181 }
1182 mutex_unlock(&vmbus_connection.channel_mutex);
1183 }
1184
1185 /*
1186 * Now close the primary.
1187 */
1188 mutex_lock(&vmbus_connection.channel_mutex);
1189 ret = vmbus_close_internal(channel);
1190 mutex_unlock(&vmbus_connection.channel_mutex);
1191
1192 return ret;
1193}
1194EXPORT_SYMBOL_GPL(vmbus_disconnect_ring);
1195
1196/*
1197 * vmbus_close - Close the specified channel
1198 */
1199void vmbus_close(struct vmbus_channel *channel)
1200{
1201 if (vmbus_disconnect_ring(channel) == 0)
1202 vmbus_free_ring(channel);
1203}
1204EXPORT_SYMBOL_GPL(vmbus_close);
1205
1206/**
1207 * vmbus_sendpacket_getid() - Send the specified buffer on the given channel
1208 * @channel: Pointer to vmbus_channel structure
1209 * @buffer: Pointer to the buffer you want to send the data from.
1210 * @bufferlen: Maximum size of what the buffer holds.
1211 * @requestid: Identifier of the request
1212 * @trans_id: Identifier of the transaction associated to this request, if
1213 * the send is successful; undefined, otherwise.
1214 * @type: Type of packet that is being sent e.g. negotiate, time
1215 * packet etc.
1216 * @flags: 0 or VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
1217 *
1218 * Sends data in @buffer directly to Hyper-V via the vmbus.
1219 * This will send the data unparsed to Hyper-V.
1220 *
1221 * Mainly used by Hyper-V drivers.
1222 */
1223int vmbus_sendpacket_getid(struct vmbus_channel *channel, void *buffer,
1224 u32 bufferlen, u64 requestid, u64 *trans_id,
1225 enum vmbus_packet_type type, u32 flags)
1226{
1227 struct vmpacket_descriptor desc;
1228 u32 packetlen = sizeof(struct vmpacket_descriptor) + bufferlen;
1229 u32 packetlen_aligned = ALIGN(packetlen, sizeof(u64));
1230 struct kvec bufferlist[3];
1231 u64 aligned_data = 0;
1232 int num_vecs = ((bufferlen != 0) ? 3 : 1);
1233
1234
1235 /* Setup the descriptor */
1236 desc.type = type; /* VmbusPacketTypeDataInBand; */
1237 desc.flags = flags; /* VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED; */
1238 /* in 8-bytes granularity */
1239 desc.offset8 = sizeof(struct vmpacket_descriptor) >> 3;
1240 desc.len8 = (u16)(packetlen_aligned >> 3);
1241 desc.trans_id = VMBUS_RQST_ERROR; /* will be updated in hv_ringbuffer_write() */
1242
1243 bufferlist[0].iov_base = &desc;
1244 bufferlist[0].iov_len = sizeof(struct vmpacket_descriptor);
1245 bufferlist[1].iov_base = buffer;
1246 bufferlist[1].iov_len = bufferlen;
1247 bufferlist[2].iov_base = &aligned_data;
1248 bufferlist[2].iov_len = (packetlen_aligned - packetlen);
1249
1250 return hv_ringbuffer_write(channel, bufferlist, num_vecs, requestid, trans_id);
1251}
1252EXPORT_SYMBOL(vmbus_sendpacket_getid);
1253
1254/**
1255 * vmbus_sendpacket() - Send the specified buffer on the given channel
1256 * @channel: Pointer to vmbus_channel structure
1257 * @buffer: Pointer to the buffer you want to send the data from.
1258 * @bufferlen: Maximum size of what the buffer holds.
1259 * @requestid: Identifier of the request
1260 * @type: Type of packet that is being sent e.g. negotiate, time
1261 * packet etc.
1262 * @flags: 0 or VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
1263 *
1264 * Sends data in @buffer directly to Hyper-V via the vmbus.
1265 * This will send the data unparsed to Hyper-V.
1266 *
1267 * Mainly used by Hyper-V drivers.
1268 */
1269int vmbus_sendpacket(struct vmbus_channel *channel, void *buffer,
1270 u32 bufferlen, u64 requestid,
1271 enum vmbus_packet_type type, u32 flags)
1272{
1273 return vmbus_sendpacket_getid(channel, buffer, bufferlen,
1274 requestid, NULL, type, flags);
1275}
1276EXPORT_SYMBOL(vmbus_sendpacket);
1277
1278/*
1279 * vmbus_sendpacket_mpb_desc - Send one or more multi-page buffer packets
1280 * using a GPADL Direct packet type.
1281 * The desc argument must include space for the VMBus descriptor. The
1282 * rangecount field must already be set.
1283 */
1284int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
1285 struct vmbus_packet_mpb_array *desc,
1286 u32 desc_size,
1287 void *buffer, u32 bufferlen, u64 requestid)
1288{
1289 u32 packetlen;
1290 u32 packetlen_aligned;
1291 struct kvec bufferlist[3];
1292 u64 aligned_data = 0;
1293
1294 packetlen = desc_size + bufferlen;
1295 packetlen_aligned = ALIGN(packetlen, sizeof(u64));
1296
1297 /* Setup the descriptor */
1298 desc->type = VM_PKT_DATA_USING_GPA_DIRECT;
1299 desc->flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
1300 desc->dataoffset8 = desc_size >> 3; /* in 8-bytes granularity */
1301 desc->length8 = (u16)(packetlen_aligned >> 3);
1302 desc->transactionid = VMBUS_RQST_ERROR; /* will be updated in hv_ringbuffer_write() */
1303 desc->reserved = 0;
1304
1305 bufferlist[0].iov_base = desc;
1306 bufferlist[0].iov_len = desc_size;
1307 bufferlist[1].iov_base = buffer;
1308 bufferlist[1].iov_len = bufferlen;
1309 bufferlist[2].iov_base = &aligned_data;
1310 bufferlist[2].iov_len = (packetlen_aligned - packetlen);
1311
1312 return hv_ringbuffer_write(channel, bufferlist, 3, requestid, NULL);
1313}
1314EXPORT_SYMBOL_GPL(vmbus_sendpacket_mpb_desc);
1315
1316/**
1317 * __vmbus_recvpacket() - Retrieve the user packet on the specified channel
1318 * @channel: Pointer to vmbus_channel structure
1319 * @buffer: Pointer to the buffer you want to receive the data into.
1320 * @bufferlen: Maximum size of what the buffer can hold.
1321 * @buffer_actual_len: The actual size of the data after it was received.
1322 * @requestid: Identifier of the request
1323 * @raw: true means keep the vmpacket_descriptor header in the received data.
1324 *
1325 * Receives directly from the hyper-v vmbus and puts the data it received
1326 * into Buffer. This will receive the data unparsed from hyper-v.
1327 *
1328 * Mainly used by Hyper-V drivers.
1329 */
1330static inline int
1331__vmbus_recvpacket(struct vmbus_channel *channel, void *buffer,
1332 u32 bufferlen, u32 *buffer_actual_len, u64 *requestid,
1333 bool raw)
1334{
1335 return hv_ringbuffer_read(channel, buffer, bufferlen,
1336 buffer_actual_len, requestid, raw);
1337
1338}
1339
1340int vmbus_recvpacket(struct vmbus_channel *channel, void *buffer,
1341 u32 bufferlen, u32 *buffer_actual_len,
1342 u64 *requestid)
1343{
1344 return __vmbus_recvpacket(channel, buffer, bufferlen,
1345 buffer_actual_len, requestid, false);
1346}
1347EXPORT_SYMBOL(vmbus_recvpacket);
1348
1349/*
1350 * vmbus_recvpacket_raw - Retrieve the raw packet on the specified channel
1351 */
1352int vmbus_recvpacket_raw(struct vmbus_channel *channel, void *buffer,
1353 u32 bufferlen, u32 *buffer_actual_len,
1354 u64 *requestid)
1355{
1356 return __vmbus_recvpacket(channel, buffer, bufferlen,
1357 buffer_actual_len, requestid, true);
1358}
1359EXPORT_SYMBOL_GPL(vmbus_recvpacket_raw);
1360
1361/*
1362 * vmbus_next_request_id - Returns a new request id. It is also
1363 * the index at which the guest memory address is stored.
1364 * Uses a spin lock to avoid race conditions.
1365 * @channel: Pointer to the VMbus channel struct
1366 * @rqst_add: Guest memory address to be stored in the array
1367 */
1368u64 vmbus_next_request_id(struct vmbus_channel *channel, u64 rqst_addr)
1369{
1370 struct vmbus_requestor *rqstor = &channel->requestor;
1371 unsigned long flags;
1372 u64 current_id;
1373
1374 /* Check rqstor has been initialized */
1375 if (!channel->rqstor_size)
1376 return VMBUS_NO_RQSTOR;
1377
1378 lock_requestor(channel, flags);
1379 current_id = rqstor->next_request_id;
1380
1381 /* Requestor array is full */
1382 if (current_id >= rqstor->size) {
1383 unlock_requestor(channel, flags);
1384 return VMBUS_RQST_ERROR;
1385 }
1386
1387 rqstor->next_request_id = rqstor->req_arr[current_id];
1388 rqstor->req_arr[current_id] = rqst_addr;
1389
1390 /* The already held spin lock provides atomicity */
1391 bitmap_set(rqstor->req_bitmap, current_id, 1);
1392
1393 unlock_requestor(channel, flags);
1394
1395 /*
1396 * Cannot return an ID of 0, which is reserved for an unsolicited
1397 * message from Hyper-V; Hyper-V does not acknowledge (respond to)
1398 * VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED requests with ID of
1399 * 0 sent by the guest.
1400 */
1401 return current_id + 1;
1402}
1403EXPORT_SYMBOL_GPL(vmbus_next_request_id);
1404
1405/* As in vmbus_request_addr_match() but without the requestor lock */
1406u64 __vmbus_request_addr_match(struct vmbus_channel *channel, u64 trans_id,
1407 u64 rqst_addr)
1408{
1409 struct vmbus_requestor *rqstor = &channel->requestor;
1410 u64 req_addr;
1411
1412 /* Check rqstor has been initialized */
1413 if (!channel->rqstor_size)
1414 return VMBUS_NO_RQSTOR;
1415
1416 /* Hyper-V can send an unsolicited message with ID of 0 */
1417 if (!trans_id)
1418 return VMBUS_RQST_ERROR;
1419
1420 /* Data corresponding to trans_id is stored at trans_id - 1 */
1421 trans_id--;
1422
1423 /* Invalid trans_id */
1424 if (trans_id >= rqstor->size || !test_bit(trans_id, rqstor->req_bitmap))
1425 return VMBUS_RQST_ERROR;
1426
1427 req_addr = rqstor->req_arr[trans_id];
1428 if (rqst_addr == VMBUS_RQST_ADDR_ANY || req_addr == rqst_addr) {
1429 rqstor->req_arr[trans_id] = rqstor->next_request_id;
1430 rqstor->next_request_id = trans_id;
1431
1432 /* The already held spin lock provides atomicity */
1433 bitmap_clear(rqstor->req_bitmap, trans_id, 1);
1434 }
1435
1436 return req_addr;
1437}
1438EXPORT_SYMBOL_GPL(__vmbus_request_addr_match);
1439
1440/*
1441 * vmbus_request_addr_match - Clears/removes @trans_id from the @channel's
1442 * requestor, provided the memory address stored at @trans_id equals @rqst_addr
1443 * (or provided @rqst_addr matches the sentinel value VMBUS_RQST_ADDR_ANY).
1444 *
1445 * Returns the memory address stored at @trans_id, or VMBUS_RQST_ERROR if
1446 * @trans_id is not contained in the requestor.
1447 *
1448 * Acquires and releases the requestor spin lock.
1449 */
1450u64 vmbus_request_addr_match(struct vmbus_channel *channel, u64 trans_id,
1451 u64 rqst_addr)
1452{
1453 unsigned long flags;
1454 u64 req_addr;
1455
1456 lock_requestor(channel, flags);
1457 req_addr = __vmbus_request_addr_match(channel, trans_id, rqst_addr);
1458 unlock_requestor(channel, flags);
1459
1460 return req_addr;
1461}
1462EXPORT_SYMBOL_GPL(vmbus_request_addr_match);
1463
1464/*
1465 * vmbus_request_addr - Returns the memory address stored at @trans_id
1466 * in @rqstor. Uses a spin lock to avoid race conditions.
1467 * @channel: Pointer to the VMbus channel struct
1468 * @trans_id: Request id sent back from Hyper-V. Becomes the requestor's
1469 * next request id.
1470 */
1471u64 vmbus_request_addr(struct vmbus_channel *channel, u64 trans_id)
1472{
1473 return vmbus_request_addr_match(channel, trans_id, VMBUS_RQST_ADDR_ANY);
1474}
1475EXPORT_SYMBOL_GPL(vmbus_request_addr);