blob: 1bd31a38c51edfe699f4e26a0607a882cd31f828 [file] [log] [blame]
George Zhang20259842013-01-08 15:55:59 -08001/*
2 * VMware VMCI Driver
3 *
4 * Copyright (C) 2012 VMware, Inc. All rights reserved.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the
8 * Free Software Foundation version 2 and no later version.
9 *
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
12 * or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * for more details.
14 */
15
16#ifndef _VMW_VMCI_DEF_H_
17#define _VMW_VMCI_DEF_H_
18
19#include <linux/atomic.h>
20
21/* Register offsets. */
22#define VMCI_STATUS_ADDR 0x00
23#define VMCI_CONTROL_ADDR 0x04
24#define VMCI_ICR_ADDR 0x08
25#define VMCI_IMR_ADDR 0x0c
26#define VMCI_DATA_OUT_ADDR 0x10
27#define VMCI_DATA_IN_ADDR 0x14
28#define VMCI_CAPS_ADDR 0x18
29#define VMCI_RESULT_LOW_ADDR 0x1c
30#define VMCI_RESULT_HIGH_ADDR 0x20
31
32/* Max number of devices. */
33#define VMCI_MAX_DEVICES 1
34
35/* Status register bits. */
36#define VMCI_STATUS_INT_ON 0x1
37
38/* Control register bits. */
39#define VMCI_CONTROL_RESET 0x1
40#define VMCI_CONTROL_INT_ENABLE 0x2
41#define VMCI_CONTROL_INT_DISABLE 0x4
42
43/* Capabilities register bits. */
44#define VMCI_CAPS_HYPERCALL 0x1
45#define VMCI_CAPS_GUESTCALL 0x2
46#define VMCI_CAPS_DATAGRAM 0x4
47#define VMCI_CAPS_NOTIFICATIONS 0x8
48
49/* Interrupt Cause register bits. */
50#define VMCI_ICR_DATAGRAM 0x1
51#define VMCI_ICR_NOTIFICATION 0x2
52
53/* Interrupt Mask register bits. */
54#define VMCI_IMR_DATAGRAM 0x1
55#define VMCI_IMR_NOTIFICATION 0x2
56
57/* Interrupt type. */
58enum {
59 VMCI_INTR_TYPE_INTX = 0,
60 VMCI_INTR_TYPE_MSI = 1,
61 VMCI_INTR_TYPE_MSIX = 2,
62};
63
64/* Maximum MSI/MSI-X interrupt vectors in the device. */
65#define VMCI_MAX_INTRS 2
66
67/*
68 * Supported interrupt vectors. There is one for each ICR value above,
69 * but here they indicate the position in the vector array/message ID.
70 */
71enum {
72 VMCI_INTR_DATAGRAM = 0,
73 VMCI_INTR_NOTIFICATION = 1,
74};
75
76/*
77 * A single VMCI device has an upper limit of 128MB on the amount of
78 * memory that can be used for queue pairs.
79 */
80#define VMCI_MAX_GUEST_QP_MEMORY (128 * 1024 * 1024)
81
82/*
83 * Queues with pre-mapped data pages must be small, so that we don't pin
84 * too much kernel memory (especially on vmkernel). We limit a queuepair to
85 * 32 KB, or 16 KB per queue for symmetrical pairs.
86 */
87#define VMCI_MAX_PINNED_QP_MEMORY (32 * 1024)
88
89/*
90 * We have a fixed set of resource IDs available in the VMX.
91 * This allows us to have a very simple implementation since we statically
92 * know how many will create datagram handles. If a new caller arrives and
93 * we have run out of slots we can manually increment the maximum size of
94 * available resource IDs.
95 *
96 * VMCI reserved hypervisor datagram resource IDs.
97 */
98enum {
99 VMCI_RESOURCES_QUERY = 0,
100 VMCI_GET_CONTEXT_ID = 1,
101 VMCI_SET_NOTIFY_BITMAP = 2,
102 VMCI_DOORBELL_LINK = 3,
103 VMCI_DOORBELL_UNLINK = 4,
104 VMCI_DOORBELL_NOTIFY = 5,
105 /*
106 * VMCI_DATAGRAM_REQUEST_MAP and VMCI_DATAGRAM_REMOVE_MAP are
107 * obsoleted by the removal of VM to VM communication.
108 */
109 VMCI_DATAGRAM_REQUEST_MAP = 6,
110 VMCI_DATAGRAM_REMOVE_MAP = 7,
111 VMCI_EVENT_SUBSCRIBE = 8,
112 VMCI_EVENT_UNSUBSCRIBE = 9,
113 VMCI_QUEUEPAIR_ALLOC = 10,
114 VMCI_QUEUEPAIR_DETACH = 11,
115
116 /*
117 * VMCI_VSOCK_VMX_LOOKUP was assigned to 12 for Fusion 3.0/3.1,
118 * WS 7.0/7.1 and ESX 4.1
119 */
120 VMCI_HGFS_TRANSPORT = 13,
121 VMCI_UNITY_PBRPC_REGISTER = 14,
122 VMCI_RPC_PRIVILEGED = 15,
123 VMCI_RPC_UNPRIVILEGED = 16,
124 VMCI_RESOURCE_MAX = 17,
125};
126
127/*
128 * struct vmci_handle - Ownership information structure
129 * @context: The VMX context ID.
130 * @resource: The resource ID (used for locating in resource hash).
131 *
132 * The vmci_handle structure is used to track resources used within
133 * vmw_vmci.
134 */
135struct vmci_handle {
136 u32 context;
137 u32 resource;
138};
139
140#define vmci_make_handle(_cid, _rid) \
141 (struct vmci_handle){ .context = _cid, .resource = _rid }
142
143static inline bool vmci_handle_is_equal(struct vmci_handle h1,
144 struct vmci_handle h2)
145{
146 return h1.context == h2.context && h1.resource == h2.resource;
147}
148
149#define VMCI_INVALID_ID ~0
150static const struct vmci_handle VMCI_INVALID_HANDLE = {
151 .context = VMCI_INVALID_ID,
152 .resource = VMCI_INVALID_ID
153};
154
155static inline bool vmci_handle_is_invalid(struct vmci_handle h)
156{
157 return vmci_handle_is_equal(h, VMCI_INVALID_HANDLE);
158}
159
160/*
161 * The below defines can be used to send anonymous requests.
162 * This also indicates that no response is expected.
163 */
164#define VMCI_ANON_SRC_CONTEXT_ID VMCI_INVALID_ID
165#define VMCI_ANON_SRC_RESOURCE_ID VMCI_INVALID_ID
166static const struct vmci_handle VMCI_ANON_SRC_HANDLE = {
167 .context = VMCI_ANON_SRC_CONTEXT_ID,
168 .resource = VMCI_ANON_SRC_RESOURCE_ID
169};
170
171/* The lowest 16 context ids are reserved for internal use. */
172#define VMCI_RESERVED_CID_LIMIT ((u32) 16)
173
174/*
175 * Hypervisor context id, used for calling into hypervisor
176 * supplied services from the VM.
177 */
178#define VMCI_HYPERVISOR_CONTEXT_ID 0
179
180/*
181 * Well-known context id, a logical context that contains a set of
182 * well-known services. This context ID is now obsolete.
183 */
184#define VMCI_WELL_KNOWN_CONTEXT_ID 1
185
186/*
187 * Context ID used by host endpoints.
188 */
189#define VMCI_HOST_CONTEXT_ID 2
190
191#define VMCI_CONTEXT_IS_VM(_cid) (VMCI_INVALID_ID != (_cid) && \
192 (_cid) > VMCI_HOST_CONTEXT_ID)
193
194/*
195 * The VMCI_CONTEXT_RESOURCE_ID is used together with vmci_make_handle to make
196 * handles that refer to a specific context.
197 */
198#define VMCI_CONTEXT_RESOURCE_ID 0
199
200/*
201 * VMCI error codes.
202 */
203enum {
204 VMCI_SUCCESS_QUEUEPAIR_ATTACH = 5,
205 VMCI_SUCCESS_QUEUEPAIR_CREATE = 4,
206 VMCI_SUCCESS_LAST_DETACH = 3,
207 VMCI_SUCCESS_ACCESS_GRANTED = 2,
208 VMCI_SUCCESS_ENTRY_DEAD = 1,
209 VMCI_SUCCESS = 0,
210 VMCI_ERROR_INVALID_RESOURCE = (-1),
211 VMCI_ERROR_INVALID_ARGS = (-2),
212 VMCI_ERROR_NO_MEM = (-3),
213 VMCI_ERROR_DATAGRAM_FAILED = (-4),
214 VMCI_ERROR_MORE_DATA = (-5),
215 VMCI_ERROR_NO_MORE_DATAGRAMS = (-6),
216 VMCI_ERROR_NO_ACCESS = (-7),
217 VMCI_ERROR_NO_HANDLE = (-8),
218 VMCI_ERROR_DUPLICATE_ENTRY = (-9),
219 VMCI_ERROR_DST_UNREACHABLE = (-10),
220 VMCI_ERROR_PAYLOAD_TOO_LARGE = (-11),
221 VMCI_ERROR_INVALID_PRIV = (-12),
222 VMCI_ERROR_GENERIC = (-13),
223 VMCI_ERROR_PAGE_ALREADY_SHARED = (-14),
224 VMCI_ERROR_CANNOT_SHARE_PAGE = (-15),
225 VMCI_ERROR_CANNOT_UNSHARE_PAGE = (-16),
226 VMCI_ERROR_NO_PROCESS = (-17),
227 VMCI_ERROR_NO_DATAGRAM = (-18),
228 VMCI_ERROR_NO_RESOURCES = (-19),
229 VMCI_ERROR_UNAVAILABLE = (-20),
230 VMCI_ERROR_NOT_FOUND = (-21),
231 VMCI_ERROR_ALREADY_EXISTS = (-22),
232 VMCI_ERROR_NOT_PAGE_ALIGNED = (-23),
233 VMCI_ERROR_INVALID_SIZE = (-24),
234 VMCI_ERROR_REGION_ALREADY_SHARED = (-25),
235 VMCI_ERROR_TIMEOUT = (-26),
236 VMCI_ERROR_DATAGRAM_INCOMPLETE = (-27),
237 VMCI_ERROR_INCORRECT_IRQL = (-28),
238 VMCI_ERROR_EVENT_UNKNOWN = (-29),
239 VMCI_ERROR_OBSOLETE = (-30),
240 VMCI_ERROR_QUEUEPAIR_MISMATCH = (-31),
241 VMCI_ERROR_QUEUEPAIR_NOTSET = (-32),
242 VMCI_ERROR_QUEUEPAIR_NOTOWNER = (-33),
243 VMCI_ERROR_QUEUEPAIR_NOTATTACHED = (-34),
244 VMCI_ERROR_QUEUEPAIR_NOSPACE = (-35),
245 VMCI_ERROR_QUEUEPAIR_NODATA = (-36),
246 VMCI_ERROR_BUSMEM_INVALIDATION = (-37),
247 VMCI_ERROR_MODULE_NOT_LOADED = (-38),
248 VMCI_ERROR_DEVICE_NOT_FOUND = (-39),
249 VMCI_ERROR_QUEUEPAIR_NOT_READY = (-40),
250 VMCI_ERROR_WOULD_BLOCK = (-41),
251
252 /* VMCI clients should return error code within this range */
253 VMCI_ERROR_CLIENT_MIN = (-500),
254 VMCI_ERROR_CLIENT_MAX = (-550),
255
256 /* Internal error codes. */
257 VMCI_SHAREDMEM_ERROR_BAD_CONTEXT = (-1000),
258};
259
260/* VMCI reserved events. */
261enum {
262 /* Only applicable to guest endpoints */
263 VMCI_EVENT_CTX_ID_UPDATE = 0,
264
265 /* Applicable to guest and host */
266 VMCI_EVENT_CTX_REMOVED = 1,
267
268 /* Only applicable to guest endpoints */
269 VMCI_EVENT_QP_RESUMED = 2,
270
271 /* Applicable to guest and host */
272 VMCI_EVENT_QP_PEER_ATTACH = 3,
273
274 /* Applicable to guest and host */
275 VMCI_EVENT_QP_PEER_DETACH = 4,
276
277 /*
278 * Applicable to VMX and vmk. On vmk,
279 * this event has the Context payload type.
280 */
281 VMCI_EVENT_MEM_ACCESS_ON = 5,
282
283 /*
284 * Applicable to VMX and vmk. Same as
285 * above for the payload type.
286 */
287 VMCI_EVENT_MEM_ACCESS_OFF = 6,
288 VMCI_EVENT_MAX = 7,
289};
290
291/*
292 * Of the above events, a few are reserved for use in the VMX, and
293 * other endpoints (guest and host kernel) should not use them. For
294 * the rest of the events, we allow both host and guest endpoints to
295 * subscribe to them, to maintain the same API for host and guest
296 * endpoints.
297 */
298#define VMCI_EVENT_VALID_VMX(_event) ((_event) == VMCI_EVENT_MEM_ACCESS_ON || \
299 (_event) == VMCI_EVENT_MEM_ACCESS_OFF)
300
301#define VMCI_EVENT_VALID(_event) ((_event) < VMCI_EVENT_MAX && \
302 !VMCI_EVENT_VALID_VMX(_event))
303
304/* Reserved guest datagram resource ids. */
305#define VMCI_EVENT_HANDLER 0
306
307/*
308 * VMCI coarse-grained privileges (per context or host
309 * process/endpoint. An entity with the restricted flag is only
310 * allowed to interact with the hypervisor and trusted entities.
311 */
312enum {
313 VMCI_NO_PRIVILEGE_FLAGS = 0,
314 VMCI_PRIVILEGE_FLAG_RESTRICTED = 1,
315 VMCI_PRIVILEGE_FLAG_TRUSTED = 2,
316 VMCI_PRIVILEGE_ALL_FLAGS = (VMCI_PRIVILEGE_FLAG_RESTRICTED |
317 VMCI_PRIVILEGE_FLAG_TRUSTED),
318 VMCI_DEFAULT_PROC_PRIVILEGE_FLAGS = VMCI_NO_PRIVILEGE_FLAGS,
319 VMCI_LEAST_PRIVILEGE_FLAGS = VMCI_PRIVILEGE_FLAG_RESTRICTED,
320 VMCI_MAX_PRIVILEGE_FLAGS = VMCI_PRIVILEGE_FLAG_TRUSTED,
321};
322
323/* 0 through VMCI_RESERVED_RESOURCE_ID_MAX are reserved. */
324#define VMCI_RESERVED_RESOURCE_ID_MAX 1023
325
326/*
327 * Driver version.
328 *
329 * Increment major version when you make an incompatible change.
330 * Compatibility goes both ways (old driver with new executable
331 * as well as new driver with old executable).
332 */
333
334/* Never change VMCI_VERSION_SHIFT_WIDTH */
335#define VMCI_VERSION_SHIFT_WIDTH 16
336#define VMCI_MAKE_VERSION(_major, _minor) \
337 ((_major) << VMCI_VERSION_SHIFT_WIDTH | (u16) (_minor))
338
339#define VMCI_VERSION_MAJOR(v) ((u32) (v) >> VMCI_VERSION_SHIFT_WIDTH)
340#define VMCI_VERSION_MINOR(v) ((u16) (v))
341
342/*
343 * VMCI_VERSION is always the current version. Subsequently listed
344 * versions are ways of detecting previous versions of the connecting
345 * application (i.e., VMX).
346 *
347 * VMCI_VERSION_NOVMVM: This version removed support for VM to VM
348 * communication.
349 *
350 * VMCI_VERSION_NOTIFY: This version introduced doorbell notification
351 * support.
352 *
353 * VMCI_VERSION_HOSTQP: This version introduced host end point support
354 * for hosted products.
355 *
356 * VMCI_VERSION_PREHOSTQP: This is the version prior to the adoption of
357 * support for host end-points.
358 *
359 * VMCI_VERSION_PREVERS2: This fictional version number is intended to
360 * represent the version of a VMX which doesn't call into the driver
361 * with ioctl VERSION2 and thus doesn't establish its version with the
362 * driver.
363 */
364
365#define VMCI_VERSION VMCI_VERSION_NOVMVM
366#define VMCI_VERSION_NOVMVM VMCI_MAKE_VERSION(11, 0)
367#define VMCI_VERSION_NOTIFY VMCI_MAKE_VERSION(10, 0)
368#define VMCI_VERSION_HOSTQP VMCI_MAKE_VERSION(9, 0)
369#define VMCI_VERSION_PREHOSTQP VMCI_MAKE_VERSION(8, 0)
370#define VMCI_VERSION_PREVERS2 VMCI_MAKE_VERSION(1, 0)
371
372#define VMCI_SOCKETS_MAKE_VERSION(_p) \
373 ((((_p)[0] & 0xFF) << 24) | (((_p)[1] & 0xFF) << 16) | ((_p)[2]))
374
375/*
376 * The VMCI IOCTLs. We use identity code 7, as noted in ioctl-number.h, and
377 * we start at sequence 9f. This gives us the same values that our shipping
378 * products use, starting at 1951, provided we leave out the direction and
379 * structure size. Note that VMMon occupies the block following us, starting
380 * at 2001.
381 */
382#define IOCTL_VMCI_VERSION _IO(7, 0x9f) /* 1951 */
383#define IOCTL_VMCI_INIT_CONTEXT _IO(7, 0xa0)
384#define IOCTL_VMCI_QUEUEPAIR_SETVA _IO(7, 0xa4)
385#define IOCTL_VMCI_NOTIFY_RESOURCE _IO(7, 0xa5)
386#define IOCTL_VMCI_NOTIFICATIONS_RECEIVE _IO(7, 0xa6)
387#define IOCTL_VMCI_VERSION2 _IO(7, 0xa7)
388#define IOCTL_VMCI_QUEUEPAIR_ALLOC _IO(7, 0xa8)
389#define IOCTL_VMCI_QUEUEPAIR_SETPAGEFILE _IO(7, 0xa9)
390#define IOCTL_VMCI_QUEUEPAIR_DETACH _IO(7, 0xaa)
391#define IOCTL_VMCI_DATAGRAM_SEND _IO(7, 0xab)
392#define IOCTL_VMCI_DATAGRAM_RECEIVE _IO(7, 0xac)
393#define IOCTL_VMCI_CTX_ADD_NOTIFICATION _IO(7, 0xaf)
394#define IOCTL_VMCI_CTX_REMOVE_NOTIFICATION _IO(7, 0xb0)
395#define IOCTL_VMCI_CTX_GET_CPT_STATE _IO(7, 0xb1)
396#define IOCTL_VMCI_CTX_SET_CPT_STATE _IO(7, 0xb2)
397#define IOCTL_VMCI_GET_CONTEXT_ID _IO(7, 0xb3)
398#define IOCTL_VMCI_SOCKETS_VERSION _IO(7, 0xb4)
399#define IOCTL_VMCI_SOCKETS_GET_AF_VALUE _IO(7, 0xb8)
400#define IOCTL_VMCI_SOCKETS_GET_LOCAL_CID _IO(7, 0xb9)
401#define IOCTL_VMCI_SET_NOTIFY _IO(7, 0xcb) /* 1995 */
402/*IOCTL_VMMON_START _IO(7, 0xd1)*/ /* 2001 */
403
404/*
405 * struct vmci_queue_header - VMCI Queue Header information.
406 *
407 * A Queue cannot stand by itself as designed. Each Queue's header
408 * contains a pointer into itself (the producer_tail) and into its peer
409 * (consumer_head). The reason for the separation is one of
410 * accessibility: Each end-point can modify two things: where the next
411 * location to enqueue is within its produce_q (producer_tail); and
412 * where the next dequeue location is in its consume_q (consumer_head).
413 *
414 * An end-point cannot modify the pointers of its peer (guest to
415 * guest; NOTE that in the host both queue headers are mapped r/w).
416 * But, each end-point needs read access to both Queue header
417 * structures in order to determine how much space is used (or left)
418 * in the Queue. This is because for an end-point to know how full
419 * its produce_q is, it needs to use the consumer_head that points into
420 * the produce_q but -that- consumer_head is in the Queue header for
421 * that end-points consume_q.
422 *
423 * Thoroughly confused? Sorry.
424 *
425 * producer_tail: the point to enqueue new entrants. When you approach
426 * a line in a store, for example, you walk up to the tail.
427 *
428 * consumer_head: the point in the queue from which the next element is
429 * dequeued. In other words, who is next in line is he who is at the
430 * head of the line.
431 *
432 * Also, producer_tail points to an empty byte in the Queue, whereas
433 * consumer_head points to a valid byte of data (unless producer_tail ==
434 * consumer_head in which case consumer_head does not point to a valid
435 * byte of data).
436 *
437 * For a queue of buffer 'size' bytes, the tail and head pointers will be in
438 * the range [0, size-1].
439 *
440 * If produce_q_header->producer_tail == consume_q_header->consumer_head
441 * then the produce_q is empty.
442 */
443struct vmci_queue_header {
444 /* All fields are 64bit and aligned. */
445 struct vmci_handle handle; /* Identifier. */
446 atomic64_t producer_tail; /* Offset in this queue. */
447 atomic64_t consumer_head; /* Offset in peer queue. */
448};
449
450/*
451 * struct vmci_datagram - Base struct for vmci datagrams.
452 * @dst: A vmci_handle that tracks the destination of the datagram.
453 * @src: A vmci_handle that tracks the source of the datagram.
454 * @payload_size: The size of the payload.
455 *
456 * vmci_datagram structs are used when sending vmci datagrams. They include
457 * the necessary source and destination information to properly route
458 * the information along with the size of the package.
459 */
460struct vmci_datagram {
461 struct vmci_handle dst;
462 struct vmci_handle src;
463 u64 payload_size;
464};
465
466/*
467 * Second flag is for creating a well-known handle instead of a per context
468 * handle. Next flag is for deferring datagram delivery, so that the
469 * datagram callback is invoked in a delayed context (not interrupt context).
470 */
471#define VMCI_FLAG_DG_NONE 0
472#define VMCI_FLAG_WELLKNOWN_DG_HND 0x1
473#define VMCI_FLAG_ANYCID_DG_HND 0x2
474#define VMCI_FLAG_DG_DELAYED_CB 0x4
475
476/*
477 * Maximum supported size of a VMCI datagram for routable datagrams.
478 * Datagrams going to the hypervisor are allowed to be larger.
479 */
480#define VMCI_MAX_DG_SIZE (17 * 4096)
481#define VMCI_MAX_DG_PAYLOAD_SIZE (VMCI_MAX_DG_SIZE - \
482 sizeof(struct vmci_datagram))
483#define VMCI_DG_PAYLOAD(_dg) (void *)((char *)(_dg) + \
484 sizeof(struct vmci_datagram))
485#define VMCI_DG_HEADERSIZE sizeof(struct vmci_datagram)
486#define VMCI_DG_SIZE(_dg) (VMCI_DG_HEADERSIZE + (size_t)(_dg)->payload_size)
487#define VMCI_DG_SIZE_ALIGNED(_dg) ((VMCI_DG_SIZE(_dg) + 7) & (~((size_t) 0x7)))
488#define VMCI_MAX_DATAGRAM_QUEUE_SIZE (VMCI_MAX_DG_SIZE * 2)
489
490struct vmci_event_payload_qp {
491 struct vmci_handle handle; /* queue_pair handle. */
492 u32 peer_id; /* Context id of attaching/detaching VM. */
493 u32 _pad;
494};
495
496/* Flags for VMCI queue_pair API. */
497enum {
498 /* Fail alloc if QP not created by peer. */
499 VMCI_QPFLAG_ATTACH_ONLY = 1 << 0,
500
501 /* Only allow attaches from local context. */
502 VMCI_QPFLAG_LOCAL = 1 << 1,
503
504 /* Host won't block when guest is quiesced. */
505 VMCI_QPFLAG_NONBLOCK = 1 << 2,
506
507 /* Pin data pages in ESX. Used with NONBLOCK */
508 VMCI_QPFLAG_PINNED = 1 << 3,
509
510 /* Update the following flag when adding new flags. */
511 VMCI_QP_ALL_FLAGS = (VMCI_QPFLAG_ATTACH_ONLY | VMCI_QPFLAG_LOCAL |
512 VMCI_QPFLAG_NONBLOCK | VMCI_QPFLAG_PINNED),
513
514 /* Convenience flags */
515 VMCI_QP_ASYMM = (VMCI_QPFLAG_NONBLOCK | VMCI_QPFLAG_PINNED),
516 VMCI_QP_ASYMM_PEER = (VMCI_QPFLAG_ATTACH_ONLY | VMCI_QP_ASYMM),
517};
518
519/*
520 * We allow at least 1024 more event datagrams from the hypervisor past the
521 * normally allowed datagrams pending for a given context. We define this
522 * limit on event datagrams from the hypervisor to guard against DoS attack
523 * from a malicious VM which could repeatedly attach to and detach from a queue
524 * pair, causing events to be queued at the destination VM. However, the rate
525 * at which such events can be generated is small since it requires a VM exit
526 * and handling of queue pair attach/detach call at the hypervisor. Event
527 * datagrams may be queued up at the destination VM if it has interrupts
528 * disabled or if it is not draining events for some other reason. 1024
529 * datagrams is a grossly conservative estimate of the time for which
530 * interrupts may be disabled in the destination VM, but at the same time does
531 * not exacerbate the memory pressure problem on the host by much (size of each
532 * event datagram is small).
533 */
534#define VMCI_MAX_DATAGRAM_AND_EVENT_QUEUE_SIZE \
535 (VMCI_MAX_DATAGRAM_QUEUE_SIZE + \
536 1024 * (sizeof(struct vmci_datagram) + \
537 sizeof(struct vmci_event_data_max)))
538
539/*
540 * Struct used for querying, via VMCI_RESOURCES_QUERY, the availability of
541 * hypervisor resources. Struct size is 16 bytes. All fields in struct are
542 * aligned to their natural alignment.
543 */
544struct vmci_resource_query_hdr {
545 struct vmci_datagram hdr;
546 u32 num_resources;
547 u32 _padding;
548};
549
550/*
551 * Convenience struct for negotiating vectors. Must match layout of
552 * VMCIResourceQueryHdr minus the struct vmci_datagram header.
553 */
554struct vmci_resource_query_msg {
555 u32 num_resources;
556 u32 _padding;
557 u32 resources[1];
558};
559
560/*
561 * The maximum number of resources that can be queried using
562 * VMCI_RESOURCE_QUERY is 31, as the result is encoded in the lower 31
563 * bits of a positive return value. Negative values are reserved for
564 * errors.
565 */
566#define VMCI_RESOURCE_QUERY_MAX_NUM 31
567
568/* Maximum size for the VMCI_RESOURCE_QUERY request. */
569#define VMCI_RESOURCE_QUERY_MAX_SIZE \
570 (sizeof(struct vmci_resource_query_hdr) + \
571 sizeof(u32) * VMCI_RESOURCE_QUERY_MAX_NUM)
572
573/*
574 * Struct used for setting the notification bitmap. All fields in
575 * struct are aligned to their natural alignment.
576 */
577struct vmci_notify_bm_set_msg {
578 struct vmci_datagram hdr;
579 u32 bitmap_ppn;
580 u32 _pad;
581};
582
583/*
584 * Struct used for linking a doorbell handle with an index in the
585 * notify bitmap. All fields in struct are aligned to their natural
586 * alignment.
587 */
588struct vmci_doorbell_link_msg {
589 struct vmci_datagram hdr;
590 struct vmci_handle handle;
591 u64 notify_idx;
592};
593
594/*
595 * Struct used for unlinking a doorbell handle from an index in the
596 * notify bitmap. All fields in struct are aligned to their natural
597 * alignment.
598 */
599struct vmci_doorbell_unlink_msg {
600 struct vmci_datagram hdr;
601 struct vmci_handle handle;
602};
603
604/*
605 * Struct used for generating a notification on a doorbell handle. All
606 * fields in struct are aligned to their natural alignment.
607 */
608struct vmci_doorbell_notify_msg {
609 struct vmci_datagram hdr;
610 struct vmci_handle handle;
611};
612
613/*
614 * This struct is used to contain data for events. Size of this struct is a
615 * multiple of 8 bytes, and all fields are aligned to their natural alignment.
616 */
617struct vmci_event_data {
618 u32 event; /* 4 bytes. */
619 u32 _pad;
620 /* Event payload is put here. */
621};
622
623/*
624 * Define the different VMCI_EVENT payload data types here. All structs must
625 * be a multiple of 8 bytes, and fields must be aligned to their natural
626 * alignment.
627 */
628struct vmci_event_payld_ctx {
629 u32 context_id; /* 4 bytes. */
630 u32 _pad;
631};
632
633struct vmci_event_payld_qp {
634 struct vmci_handle handle; /* queue_pair handle. */
635 u32 peer_id; /* Context id of attaching/detaching VM. */
636 u32 _pad;
637};
638
639/*
640 * We define the following struct to get the size of the maximum event
641 * data the hypervisor may send to the guest. If adding a new event
642 * payload type above, add it to the following struct too (inside the
643 * union).
644 */
645struct vmci_event_data_max {
646 struct vmci_event_data event_data;
647 union {
648 struct vmci_event_payld_ctx context_payload;
649 struct vmci_event_payld_qp qp_payload;
650 } ev_data_payload;
651};
652
653/*
654 * Struct used for VMCI_EVENT_SUBSCRIBE/UNSUBSCRIBE and
655 * VMCI_EVENT_HANDLER messages. Struct size is 32 bytes. All fields
656 * in struct are aligned to their natural alignment.
657 */
658struct vmci_event_msg {
659 struct vmci_datagram hdr;
660
661 /* Has event type and payload. */
662 struct vmci_event_data event_data;
663
664 /* Payload gets put here. */
665};
666
667/* Event with context payload. */
668struct vmci_event_ctx {
669 struct vmci_event_msg msg;
670 struct vmci_event_payld_ctx payload;
671};
672
673/* Event with QP payload. */
674struct vmci_event_qp {
675 struct vmci_event_msg msg;
676 struct vmci_event_payld_qp payload;
677};
678
679/*
680 * Structs used for queue_pair alloc and detach messages. We align fields of
681 * these structs to 64bit boundaries.
682 */
683struct vmci_qp_alloc_msg {
684 struct vmci_datagram hdr;
685 struct vmci_handle handle;
686 u32 peer;
687 u32 flags;
688 u64 produce_size;
689 u64 consume_size;
690 u64 num_ppns;
691
692 /* List of PPNs placed here. */
693};
694
695struct vmci_qp_detach_msg {
696 struct vmci_datagram hdr;
697 struct vmci_handle handle;
698};
699
700/* VMCI Doorbell API. */
701#define VMCI_FLAG_DELAYED_CB 0x01
702
703typedef void (*vmci_callback) (void *client_data);
704
705/*
706 * struct vmci_qp - A vmw_vmci queue pair handle.
707 *
708 * This structure is used as a handle to a queue pair created by
709 * VMCI. It is intentionally left opaque to clients.
710 */
711struct vmci_qp;
712
713/* Callback needed for correctly waiting on events. */
714typedef int (*vmci_datagram_recv_cb) (void *client_data,
715 struct vmci_datagram *msg);
716
717/* VMCI Event API. */
718typedef void (*vmci_event_cb) (u32 sub_id, const struct vmci_event_data *ed,
719 void *client_data);
720
721/*
722 * We use the following inline function to access the payload data
723 * associated with an event data.
724 */
725static inline const void *
726vmci_event_data_const_payload(const struct vmci_event_data *ev_data)
727{
728 return (const char *)ev_data + sizeof(*ev_data);
729}
730
731static inline void *vmci_event_data_payload(struct vmci_event_data *ev_data)
732{
733 return (void *)vmci_event_data_const_payload(ev_data);
734}
735
736/*
Jorgen Hansenf42a0fd2015-11-12 01:29:32 -0800737 * Helper to read a value from a head or tail pointer. For X86_32, the
738 * pointer is treated as a 32bit value, since the pointer value
739 * never exceeds a 32bit value in this case. Also, doing an
740 * atomic64_read on X86_32 uniprocessor systems may be implemented
741 * as a non locked cmpxchg8b, that may end up overwriting updates done
742 * by the VMCI device to the memory location. On 32bit SMP, the lock
743 * prefix will be used, so correctness isn't an issue, but using a
744 * 64bit operation still adds unnecessary overhead.
745 */
746static inline u64 vmci_q_read_pointer(atomic64_t *var)
747{
748#if defined(CONFIG_X86_32)
749 return atomic_read((atomic_t *)var);
750#else
751 return atomic64_read(var);
752#endif
753}
754
755/*
756 * Helper to set the value of a head or tail pointer. For X86_32, the
757 * pointer is treated as a 32bit value, since the pointer value
758 * never exceeds a 32bit value in this case. On 32bit SMP, using a
759 * locked cmpxchg8b adds unnecessary overhead.
760 */
761static inline void vmci_q_set_pointer(atomic64_t *var,
762 u64 new_val)
763{
764#if defined(CONFIG_X86_32)
765 return atomic_set((atomic_t *)var, (u32)new_val);
766#else
767 return atomic64_set(var, new_val);
768#endif
769}
770
771/*
George Zhang20259842013-01-08 15:55:59 -0800772 * Helper to add a given offset to a head or tail pointer. Wraps the
773 * value of the pointer around the max size of the queue.
774 */
775static inline void vmci_qp_add_pointer(atomic64_t *var,
776 size_t add,
777 u64 size)
778{
Jorgen Hansenf42a0fd2015-11-12 01:29:32 -0800779 u64 new_val = vmci_q_read_pointer(var);
George Zhang20259842013-01-08 15:55:59 -0800780
781 if (new_val >= size - add)
782 new_val -= size;
783
784 new_val += add;
785
Jorgen Hansenf42a0fd2015-11-12 01:29:32 -0800786 vmci_q_set_pointer(var, new_val);
George Zhang20259842013-01-08 15:55:59 -0800787}
788
789/*
790 * Helper routine to get the Producer Tail from the supplied queue.
791 */
792static inline u64
793vmci_q_header_producer_tail(const struct vmci_queue_header *q_header)
794{
795 struct vmci_queue_header *qh = (struct vmci_queue_header *)q_header;
Jorgen Hansenf42a0fd2015-11-12 01:29:32 -0800796 return vmci_q_read_pointer(&qh->producer_tail);
George Zhang20259842013-01-08 15:55:59 -0800797}
798
799/*
800 * Helper routine to get the Consumer Head from the supplied queue.
801 */
802static inline u64
803vmci_q_header_consumer_head(const struct vmci_queue_header *q_header)
804{
805 struct vmci_queue_header *qh = (struct vmci_queue_header *)q_header;
Jorgen Hansenf42a0fd2015-11-12 01:29:32 -0800806 return vmci_q_read_pointer(&qh->consumer_head);
George Zhang20259842013-01-08 15:55:59 -0800807}
808
809/*
810 * Helper routine to increment the Producer Tail. Fundamentally,
811 * vmci_qp_add_pointer() is used to manipulate the tail itself.
812 */
813static inline void
814vmci_q_header_add_producer_tail(struct vmci_queue_header *q_header,
815 size_t add,
816 u64 queue_size)
817{
818 vmci_qp_add_pointer(&q_header->producer_tail, add, queue_size);
819}
820
821/*
822 * Helper routine to increment the Consumer Head. Fundamentally,
823 * vmci_qp_add_pointer() is used to manipulate the head itself.
824 */
825static inline void
826vmci_q_header_add_consumer_head(struct vmci_queue_header *q_header,
827 size_t add,
828 u64 queue_size)
829{
830 vmci_qp_add_pointer(&q_header->consumer_head, add, queue_size);
831}
832
833/*
834 * Helper routine for getting the head and the tail pointer for a queue.
835 * Both the VMCIQueues are needed to get both the pointers for one queue.
836 */
837static inline void
838vmci_q_header_get_pointers(const struct vmci_queue_header *produce_q_header,
839 const struct vmci_queue_header *consume_q_header,
840 u64 *producer_tail,
841 u64 *consumer_head)
842{
843 if (producer_tail)
844 *producer_tail = vmci_q_header_producer_tail(produce_q_header);
845
846 if (consumer_head)
847 *consumer_head = vmci_q_header_consumer_head(consume_q_header);
848}
849
850static inline void vmci_q_header_init(struct vmci_queue_header *q_header,
851 const struct vmci_handle handle)
852{
853 q_header->handle = handle;
854 atomic64_set(&q_header->producer_tail, 0);
855 atomic64_set(&q_header->consumer_head, 0);
856}
857
858/*
859 * Finds available free space in a produce queue to enqueue more
860 * data or reports an error if queue pair corruption is detected.
861 */
862static s64
863vmci_q_header_free_space(const struct vmci_queue_header *produce_q_header,
864 const struct vmci_queue_header *consume_q_header,
865 const u64 produce_q_size)
866{
867 u64 tail;
868 u64 head;
869 u64 free_space;
870
871 tail = vmci_q_header_producer_tail(produce_q_header);
872 head = vmci_q_header_consumer_head(consume_q_header);
873
874 if (tail >= produce_q_size || head >= produce_q_size)
875 return VMCI_ERROR_INVALID_SIZE;
876
877 /*
878 * Deduct 1 to avoid tail becoming equal to head which causes
879 * ambiguity. If head and tail are equal it means that the
880 * queue is empty.
881 */
882 if (tail >= head)
883 free_space = produce_q_size - (tail - head) - 1;
884 else
885 free_space = head - tail - 1;
886
887 return free_space;
888}
889
890/*
891 * vmci_q_header_free_space() does all the heavy lifting of
892 * determing the number of free bytes in a Queue. This routine,
893 * then subtracts that size from the full size of the Queue so
894 * the caller knows how many bytes are ready to be dequeued.
895 * Results:
896 * On success, available data size in bytes (up to MAX_INT64).
897 * On failure, appropriate error code.
898 */
899static inline s64
900vmci_q_header_buf_ready(const struct vmci_queue_header *consume_q_header,
901 const struct vmci_queue_header *produce_q_header,
902 const u64 consume_q_size)
903{
904 s64 free_space;
905
906 free_space = vmci_q_header_free_space(consume_q_header,
907 produce_q_header, consume_q_size);
908 if (free_space < VMCI_SUCCESS)
909 return free_space;
910
911 return consume_q_size - free_space - 1;
912}
913
914
915#endif /* _VMW_VMCI_DEF_H_ */