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