1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3 *
4 * Copyright (c) 2011, Microsoft Corporation.
5 *
6 * Authors:
7 * Haiyang Zhang <haiyangz@microsoft.com>
8 * Hank Janssen <hjanssen@microsoft.com>
9 * K. Y. Srinivasan <kys@microsoft.com>
10 */
11
12 #ifndef _HYPERV_VMBUS_H
13 #define _HYPERV_VMBUS_H
14
15 #include <linux/list.h>
16 #include <linux/bitops.h>
17 #include <asm/sync_bitops.h>
18 #include <asm/hyperv-tlfs.h>
19 #include <linux/atomic.h>
20 #include <linux/hyperv.h>
21 #include <linux/interrupt.h>
22
23 #include "hv_trace.h"
24
25 /*
26 * Timeout for services such as KVP and fcopy.
27 */
28 #define HV_UTIL_TIMEOUT 30
29
30 /*
31 * Timeout for guest-host handshake for services.
32 */
33 #define HV_UTIL_NEGO_TIMEOUT 55
34
35
36 /* Definitions for the monitored notification facility */
37 union hv_monitor_trigger_group {
38 u64 as_uint64;
39 struct {
40 u32 pending;
41 u32 armed;
42 };
43 };
44
45 struct hv_monitor_parameter {
46 union hv_connection_id connectionid;
47 u16 flagnumber;
48 u16 rsvdz;
49 };
50
51 union hv_monitor_trigger_state {
52 u32 asu32;
53
54 struct {
55 u32 group_enable:4;
56 u32 rsvdz:28;
57 };
58 };
59
60 /* struct hv_monitor_page Layout */
61 /* ------------------------------------------------------ */
62 /* | 0 | TriggerState (4 bytes) | Rsvd1 (4 bytes) | */
63 /* | 8 | TriggerGroup[0] | */
64 /* | 10 | TriggerGroup[1] | */
65 /* | 18 | TriggerGroup[2] | */
66 /* | 20 | TriggerGroup[3] | */
67 /* | 28 | Rsvd2[0] | */
68 /* | 30 | Rsvd2[1] | */
69 /* | 38 | Rsvd2[2] | */
70 /* | 40 | NextCheckTime[0][0] | NextCheckTime[0][1] | */
71 /* | ... | */
72 /* | 240 | Latency[0][0..3] | */
73 /* | 340 | Rsvz3[0] | */
74 /* | 440 | Parameter[0][0] | */
75 /* | 448 | Parameter[0][1] | */
76 /* | ... | */
77 /* | 840 | Rsvd4[0] | */
78 /* ------------------------------------------------------ */
79 struct hv_monitor_page {
80 union hv_monitor_trigger_state trigger_state;
81 u32 rsvdz1;
82
83 union hv_monitor_trigger_group trigger_group[4];
84 u64 rsvdz2[3];
85
86 s32 next_checktime[4][32];
87
88 u16 latency[4][32];
89 u64 rsvdz3[32];
90
91 struct hv_monitor_parameter parameter[4][32];
92
93 u8 rsvdz4[1984];
94 };
95
96 #define HV_HYPERCALL_PARAM_ALIGN sizeof(u64)
97
98 /* Definition of the hv_post_message hypercall input structure. */
99 struct hv_input_post_message {
100 union hv_connection_id connectionid;
101 u32 reserved;
102 u32 message_type;
103 u32 payload_size;
104 u64 payload[HV_MESSAGE_PAYLOAD_QWORD_COUNT];
105 };
106
107
108 enum {
109 VMBUS_MESSAGE_CONNECTION_ID = 1,
110 VMBUS_MESSAGE_CONNECTION_ID_4 = 4,
111 VMBUS_MESSAGE_PORT_ID = 1,
112 VMBUS_EVENT_CONNECTION_ID = 2,
113 VMBUS_EVENT_PORT_ID = 2,
114 VMBUS_MONITOR_CONNECTION_ID = 3,
115 VMBUS_MONITOR_PORT_ID = 3,
116 VMBUS_MESSAGE_SINT = 2,
117 };
118
119 /*
120 * Per cpu state for channel handling
121 */
122 struct hv_per_cpu_context {
123 void *synic_message_page;
124 void *synic_event_page;
125 /*
126 * buffer to post messages to the host.
127 */
128 void *post_msg_page;
129
130 /*
131 * Starting with win8, we can take channel interrupts on any CPU;
132 * we will manage the tasklet that handles events messages on a per CPU
133 * basis.
134 */
135 struct tasklet_struct msg_dpc;
136 };
137
138 struct hv_context {
139 /* We only support running on top of Hyper-V
140 * So at this point this really can only contain the Hyper-V ID
141 */
142 u64 guestid;
143
144 struct hv_per_cpu_context __percpu *cpu_context;
145
146 /*
147 * To manage allocations in a NUMA node.
148 * Array indexed by numa node ID.
149 */
150 struct cpumask *hv_numa_map;
151 };
152
153 extern struct hv_context hv_context;
154
155 /* Hv Interface */
156
157 extern int hv_init(void);
158
159 extern int hv_post_message(union hv_connection_id connection_id,
160 enum hv_message_type message_type,
161 void *payload, size_t payload_size);
162
163 extern int hv_synic_alloc(void);
164
165 extern void hv_synic_free(void);
166
167 extern void hv_synic_enable_regs(unsigned int cpu);
168 extern int hv_synic_init(unsigned int cpu);
169
170 extern void hv_synic_disable_regs(unsigned int cpu);
171 extern int hv_synic_cleanup(unsigned int cpu);
172
173 /* Interface */
174
175 void hv_ringbuffer_pre_init(struct vmbus_channel *channel);
176
177 int hv_ringbuffer_init(struct hv_ring_buffer_info *ring_info,
178 struct page *pages, u32 pagecnt, u32 max_pkt_size);
179
180 void hv_ringbuffer_cleanup(struct hv_ring_buffer_info *ring_info);
181
182 int hv_ringbuffer_write(struct vmbus_channel *channel,
183 const struct kvec *kv_list, u32 kv_count,
184 u64 requestid, u64 *trans_id);
185
186 int hv_ringbuffer_read(struct vmbus_channel *channel,
187 void *buffer, u32 buflen, u32 *buffer_actual_len,
188 u64 *requestid, bool raw);
189
190 /*
191 * The Maximum number of channels (16384) is determined by the size of the
192 * interrupt page, which is HV_HYP_PAGE_SIZE. 1/2 of HV_HYP_PAGE_SIZE is to
193 * send endpoint interrupts, and the other is to receive endpoint interrupts.
194 */
195 #define MAX_NUM_CHANNELS ((HV_HYP_PAGE_SIZE >> 1) << 3)
196
197 /* The value here must be in multiple of 32 */
198 #define MAX_NUM_CHANNELS_SUPPORTED 256
199
200 #define MAX_CHANNEL_RELIDS \
201 max(MAX_NUM_CHANNELS_SUPPORTED, HV_EVENT_FLAGS_COUNT)
202
203 enum vmbus_connect_state {
204 DISCONNECTED,
205 CONNECTING,
206 CONNECTED,
207 DISCONNECTING
208 };
209
210 #define MAX_SIZE_CHANNEL_MESSAGE HV_MESSAGE_PAYLOAD_BYTE_COUNT
211
212 /*
213 * The CPU that Hyper-V will interrupt for VMBUS messages, such as
214 * CHANNELMSG_OFFERCHANNEL and CHANNELMSG_RESCIND_CHANNELOFFER.
215 */
216 #define VMBUS_CONNECT_CPU 0
217
218 struct vmbus_connection {
219 u32 msg_conn_id;
220
221 atomic_t offer_in_progress;
222
223 enum vmbus_connect_state conn_state;
224
225 atomic_t next_gpadl_handle;
226
227 struct completion unload_event;
228 /*
229 * Represents channel interrupts. Each bit position represents a
230 * channel. When a channel sends an interrupt via VMBUS, it finds its
231 * bit in the sendInterruptPage, set it and calls Hv to generate a port
232 * event. The other end receives the port event and parse the
233 * recvInterruptPage to see which bit is set
234 */
235 void *int_page;
236 void *send_int_page;
237 void *recv_int_page;
238
239 /*
240 * 2 pages - 1st page for parent->child notification and 2nd
241 * is child->parent notification
242 */
243 struct hv_monitor_page *monitor_pages[2];
244 void *monitor_pages_original[2];
245 phys_addr_t monitor_pages_pa[2];
246 struct list_head chn_msg_list;
247 spinlock_t channelmsg_lock;
248
249 /* List of channels */
250 struct list_head chn_list;
251 struct mutex channel_mutex;
252
253 /* Array of channels */
254 struct vmbus_channel **channels;
255
256 /*
257 * An offer message is handled first on the work_queue, and then
258 * is further handled on handle_primary_chan_wq or
259 * handle_sub_chan_wq.
260 */
261 struct workqueue_struct *work_queue;
262 struct workqueue_struct *handle_primary_chan_wq;
263 struct workqueue_struct *handle_sub_chan_wq;
264 struct workqueue_struct *rescind_work_queue;
265
266 /*
267 * On suspension of the vmbus, the accumulated offer messages
268 * must be dropped.
269 */
270 bool ignore_any_offer_msg;
271
272 /*
273 * The number of sub-channels and hv_sock channels that should be
274 * cleaned up upon suspend: sub-channels will be re-created upon
275 * resume, and hv_sock channels should not survive suspend.
276 */
277 atomic_t nr_chan_close_on_suspend;
278 /*
279 * vmbus_bus_suspend() waits for "nr_chan_close_on_suspend" to
280 * drop to zero.
281 */
282 struct completion ready_for_suspend_event;
283
284 /*
285 * The number of primary channels that should be "fixed up"
286 * upon resume: these channels are re-offered upon resume, and some
287 * fields of the channel offers (i.e. child_relid and connection_id)
288 * can change, so the old offermsg must be fixed up, before the resume
289 * callbacks of the VSC drivers start to further touch the channels.
290 */
291 atomic_t nr_chan_fixup_on_resume;
292 /*
293 * vmbus_bus_resume() waits for "nr_chan_fixup_on_resume" to
294 * drop to zero.
295 */
296 struct completion ready_for_resume_event;
297 };
298
299
300 struct vmbus_msginfo {
301 /* Bookkeeping stuff */
302 struct list_head msglist_entry;
303
304 /* The message itself */
305 unsigned char msg[];
306 };
307
308
309 extern struct vmbus_connection vmbus_connection;
310
311 int vmbus_negotiate_version(struct vmbus_channel_msginfo *msginfo, u32 version);
312
vmbus_send_interrupt(u32 relid)313 static inline void vmbus_send_interrupt(u32 relid)
314 {
315 sync_set_bit(relid, vmbus_connection.send_int_page);
316 }
317
318 enum vmbus_message_handler_type {
319 /* The related handler can sleep. */
320 VMHT_BLOCKING = 0,
321
322 /* The related handler must NOT sleep. */
323 VMHT_NON_BLOCKING = 1,
324 };
325
326 struct vmbus_channel_message_table_entry {
327 enum vmbus_channel_message_type message_type;
328 enum vmbus_message_handler_type handler_type;
329 void (*message_handler)(struct vmbus_channel_message_header *msg);
330 u32 min_payload_len;
331 };
332
333 extern const struct vmbus_channel_message_table_entry
334 channel_message_table[CHANNELMSG_COUNT];
335
336
337 /* General vmbus interface */
338
339 struct hv_device *vmbus_device_create(const guid_t *type,
340 const guid_t *instance,
341 struct vmbus_channel *channel);
342
343 int vmbus_device_register(struct hv_device *child_device_obj);
344 void vmbus_device_unregister(struct hv_device *device_obj);
345 int vmbus_add_channel_kobj(struct hv_device *device_obj,
346 struct vmbus_channel *channel);
347
348 void vmbus_remove_channel_attr_group(struct vmbus_channel *channel);
349
350 void vmbus_channel_map_relid(struct vmbus_channel *channel);
351 void vmbus_channel_unmap_relid(struct vmbus_channel *channel);
352
353 struct vmbus_channel *relid2channel(u32 relid);
354
355 void vmbus_free_channels(void);
356
357 /* Connection interface */
358
359 int vmbus_connect(void);
360 void vmbus_disconnect(void);
361
362 int vmbus_post_msg(void *buffer, size_t buflen, bool can_sleep);
363
364 void vmbus_on_event(unsigned long data);
365 void vmbus_on_msg_dpc(unsigned long data);
366
367 int hv_kvp_init(struct hv_util_service *srv);
368 void hv_kvp_deinit(void);
369 int hv_kvp_pre_suspend(void);
370 int hv_kvp_pre_resume(void);
371 void hv_kvp_onchannelcallback(void *context);
372
373 int hv_vss_init(struct hv_util_service *srv);
374 void hv_vss_deinit(void);
375 int hv_vss_pre_suspend(void);
376 int hv_vss_pre_resume(void);
377 void hv_vss_onchannelcallback(void *context);
378
379 int hv_fcopy_init(struct hv_util_service *srv);
380 void hv_fcopy_deinit(void);
381 int hv_fcopy_pre_suspend(void);
382 int hv_fcopy_pre_resume(void);
383 void hv_fcopy_onchannelcallback(void *context);
384 void vmbus_initiate_unload(bool crash);
385
hv_poll_channel(struct vmbus_channel * channel,void (* cb)(void *))386 static inline void hv_poll_channel(struct vmbus_channel *channel,
387 void (*cb)(void *))
388 {
389 if (!channel)
390 return;
391 cb(channel);
392 }
393
394 enum hvutil_device_state {
395 HVUTIL_DEVICE_INIT = 0, /* driver is loaded, waiting for userspace */
396 HVUTIL_READY, /* userspace is registered */
397 HVUTIL_HOSTMSG_RECEIVED, /* message from the host was received */
398 HVUTIL_USERSPACE_REQ, /* request to userspace was sent */
399 HVUTIL_USERSPACE_RECV, /* reply from userspace was received */
400 HVUTIL_DEVICE_DYING, /* driver unload is in progress */
401 };
402
403 enum delay {
404 INTERRUPT_DELAY = 0,
405 MESSAGE_DELAY = 1,
406 };
407
408 extern const struct vmbus_device vmbus_devs[];
409
hv_is_perf_channel(struct vmbus_channel * channel)410 static inline bool hv_is_perf_channel(struct vmbus_channel *channel)
411 {
412 return vmbus_devs[channel->device_id].perf_device;
413 }
414
hv_is_allocated_cpu(unsigned int cpu)415 static inline bool hv_is_allocated_cpu(unsigned int cpu)
416 {
417 struct vmbus_channel *channel, *sc;
418
419 lockdep_assert_held(&vmbus_connection.channel_mutex);
420 /*
421 * List additions/deletions as well as updates of the target CPUs are
422 * protected by channel_mutex.
423 */
424 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
425 if (!hv_is_perf_channel(channel))
426 continue;
427 if (channel->target_cpu == cpu)
428 return true;
429 list_for_each_entry(sc, &channel->sc_list, sc_list) {
430 if (sc->target_cpu == cpu)
431 return true;
432 }
433 }
434 return false;
435 }
436
hv_set_allocated_cpu(unsigned int cpu)437 static inline void hv_set_allocated_cpu(unsigned int cpu)
438 {
439 cpumask_set_cpu(cpu, &hv_context.hv_numa_map[cpu_to_node(cpu)]);
440 }
441
hv_clear_allocated_cpu(unsigned int cpu)442 static inline void hv_clear_allocated_cpu(unsigned int cpu)
443 {
444 if (hv_is_allocated_cpu(cpu))
445 return;
446 cpumask_clear_cpu(cpu, &hv_context.hv_numa_map[cpu_to_node(cpu)]);
447 }
448
hv_update_allocated_cpus(unsigned int old_cpu,unsigned int new_cpu)449 static inline void hv_update_allocated_cpus(unsigned int old_cpu,
450 unsigned int new_cpu)
451 {
452 hv_set_allocated_cpu(new_cpu);
453 hv_clear_allocated_cpu(old_cpu);
454 }
455
456 #ifdef CONFIG_HYPERV_TESTING
457
458 int hv_debug_add_dev_dir(struct hv_device *dev);
459 void hv_debug_rm_dev_dir(struct hv_device *dev);
460 void hv_debug_rm_all_dir(void);
461 int hv_debug_init(void);
462 void hv_debug_delay_test(struct vmbus_channel *channel, enum delay delay_type);
463
464 #else /* CONFIG_HYPERV_TESTING */
465
hv_debug_rm_dev_dir(struct hv_device * dev)466 static inline void hv_debug_rm_dev_dir(struct hv_device *dev) {};
hv_debug_rm_all_dir(void)467 static inline void hv_debug_rm_all_dir(void) {};
hv_debug_delay_test(struct vmbus_channel * channel,enum delay delay_type)468 static inline void hv_debug_delay_test(struct vmbus_channel *channel,
469 enum delay delay_type) {};
hv_debug_init(void)470 static inline int hv_debug_init(void)
471 {
472 return -1;
473 }
474
hv_debug_add_dev_dir(struct hv_device * dev)475 static inline int hv_debug_add_dev_dir(struct hv_device *dev)
476 {
477 return -1;
478 }
479
480 #endif /* CONFIG_HYPERV_TESTING */
481
482 #endif /* _HYPERV_VMBUS_H */
483