hyperv_vmbus.h

  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/mshyperv.h>
 19#include <linux/atomic.h>
 20#include <linux/hyperv.h>
 21#include <linux/interrupt.h>
 22#include <hyperv/hvhdk.h>
 23
 24#include "hv_trace.h"
 25
 26/*
 27 * Timeout for services such as KVP and fcopy.
 28 */
 29#define HV_UTIL_TIMEOUT 30
 30
 31/*
 32 * Timeout for guest-host handshake for services.
 33 */
 34#define HV_UTIL_NEGO_TIMEOUT 55
 35
 36void vmbus_isr(void);
 37
 38/* Definitions for the monitored notification facility */
 39union hv_monitor_trigger_group {
 40	u64 as_uint64;
 41	struct {
 42		u32 pending;
 43		u32 armed;
 44	};
 45};
 46
 47struct hv_monitor_parameter {
 48	union hv_connection_id connectionid;
 49	u16 flagnumber;
 50	u16 rsvdz;
 51};
 52
 53union hv_monitor_trigger_state {
 54	u32 asu32;
 55
 56	struct {
 57		u32 group_enable:4;
 58		u32 rsvdz:28;
 59	};
 60};
 61
 62/* struct hv_monitor_page Layout */
 63/* ------------------------------------------------------ */
 64/* | 0   | TriggerState (4 bytes) | Rsvd1 (4 bytes)     | */
 65/* | 8   | TriggerGroup[0]                              | */
 66/* | 10  | TriggerGroup[1]                              | */
 67/* | 18  | TriggerGroup[2]                              | */
 68/* | 20  | TriggerGroup[3]                              | */
 69/* | 28  | Rsvd2[0]                                     | */
 70/* | 30  | Rsvd2[1]                                     | */
 71/* | 38  | Rsvd2[2]                                     | */
 72/* | 40  | NextCheckTime[0][0]    | NextCheckTime[0][1] | */
 73/* | ...                                                | */
 74/* | 240 | Latency[0][0..3]                             | */
 75/* | 340 | Rsvz3[0]                                     | */
 76/* | 440 | Parameter[0][0]                              | */
 77/* | 448 | Parameter[0][1]                              | */
 78/* | ...                                                | */
 79/* | 840 | Rsvd4[0]                                     | */
 80/* ------------------------------------------------------ */
 81struct hv_monitor_page {
 82	union hv_monitor_trigger_state trigger_state;
 83	u32 rsvdz1;
 84
 85	union hv_monitor_trigger_group trigger_group[4];
 86	u64 rsvdz2[3];
 87
 88	s32 next_checktime[4][32];
 89
 90	u16 latency[4][32];
 91	u64 rsvdz3[32];
 92
 93	struct hv_monitor_parameter parameter[4][32];
 94
 95	u8 rsvdz4[1984];
 96};
 97
 98#define HV_HYPERCALL_PARAM_ALIGN	sizeof(u64)
 99
100/* Definition of the hv_post_message hypercall input structure. */
101struct hv_input_post_message {
102	union hv_connection_id connectionid;
103	u32 reserved;
104	u32 message_type;
105	u32 payload_size;
106	u64 payload[HV_MESSAGE_PAYLOAD_QWORD_COUNT];
107};
108
109
110enum {
111	VMBUS_MESSAGE_CONNECTION_ID	= 1,
112	VMBUS_MESSAGE_CONNECTION_ID_4	= 4,
113	VMBUS_MESSAGE_PORT_ID		= 1,
114	VMBUS_EVENT_CONNECTION_ID	= 2,
115	VMBUS_EVENT_PORT_ID		= 2,
116	VMBUS_MONITOR_CONNECTION_ID	= 3,
117	VMBUS_MONITOR_PORT_ID		= 3,
118	VMBUS_MESSAGE_SINT		= 2,
119};
120
121/*
122 * Per cpu state for channel handling
123 */
124struct hv_per_cpu_context {
125	/*
126	 * SynIC pages for communicating with the host.
127	 *
128	 * These pages are accessible to the host partition and the hypervisor.
129	 * They may be used for exchanging data with the host partition and the
130	 * hypervisor even when they aren't trusted yet the guest partition
131	 * must be prepared to handle the malicious behavior.
132	 */
133	void *hyp_synic_message_page;
134	void *hyp_synic_event_page;
135	/*
136	 * SynIC pages for communicating with the paravisor.
137	 *
138	 * These pages may be accessed from within the guest partition only in
139	 * CoCo VMs. Neither the host partition nor the hypervisor can access
140	 * these pages in that case; they are used for exchanging data with the
141	 * paravisor.
142	 */
143	void *para_synic_message_page;
144	void *para_synic_event_page;
145
146	/*
147	 * The page is only used in hv_post_message() for a TDX VM (with the
148	 * paravisor) to post a messages to Hyper-V: when such a VM calls
149	 * HVCALL_POST_MESSAGE, it can't use the hyperv_pcpu_input_arg (which
150	 * is encrypted in such a VM) as the hypercall input page, because
151	 * the input page for HVCALL_POST_MESSAGE must be decrypted in such a
152	 * VM, so post_msg_page (which is decrypted in hv_synic_alloc()) is
153	 * introduced for this purpose. See hyperv_init() for more comments.
154	 */
155	void *post_msg_page;
156
157	/*
158	 * Starting with win8, we can take channel interrupts on any CPU;
159	 * we will manage the tasklet that handles events messages on a per CPU
160	 * basis.
161	 */
162	struct tasklet_struct msg_dpc;
163};
164
165struct hv_context {
166	/* We only support running on top of Hyper-V
167	 * So at this point this really can only contain the Hyper-V ID
168	 */
169	u64 guestid;
170
171	struct hv_per_cpu_context __percpu *cpu_context;
172
173	/*
174	 * To manage allocations in a NUMA node.
175	 * Array indexed by numa node ID.
176	 */
177	struct cpumask *hv_numa_map;
178};
179
180extern struct hv_context hv_context;
181
182/* Hv Interface */
183
184extern int hv_init(void);
185
186extern int hv_post_message(union hv_connection_id connection_id,
187			 enum hv_message_type message_type,
188			 void *payload, size_t payload_size);
189
190extern int hv_synic_alloc(void);
191
192extern void hv_synic_free(void);
193
194extern void hv_hyp_synic_enable_regs(unsigned int cpu);
195extern int hv_synic_init(unsigned int cpu);
196
197extern void hv_hyp_synic_disable_regs(unsigned int cpu);
198extern int hv_synic_cleanup(unsigned int cpu);
199
200/* Interface */
201
202void hv_ringbuffer_pre_init(struct vmbus_channel *channel);
203
204int hv_ringbuffer_init(struct hv_ring_buffer_info *ring_info,
205		       struct page *pages, u32 pagecnt, u32 max_pkt_size,
206			   bool confidential);
207
208void hv_ringbuffer_cleanup(struct hv_ring_buffer_info *ring_info);
209
210int hv_ringbuffer_write(struct vmbus_channel *channel,
211			const struct kvec *kv_list, u32 kv_count,
212			u64 requestid, u64 *trans_id);
213
214int hv_ringbuffer_read(struct vmbus_channel *channel,
215		       void *buffer, u32 buflen, u32 *buffer_actual_len,
216		       u64 *requestid, bool raw);
217
218/*
219 * The Maximum number of channels (16384) is determined by the size of the
220 * interrupt page, which is HV_HYP_PAGE_SIZE. 1/2 of HV_HYP_PAGE_SIZE is to
221 * send endpoint interrupts, and the other is to receive endpoint interrupts.
222 */
223#define MAX_NUM_CHANNELS	((HV_HYP_PAGE_SIZE >> 1) << 3)
224
225/* The value here must be in multiple of 32 */
226#define MAX_NUM_CHANNELS_SUPPORTED	256
227
228#define MAX_CHANNEL_RELIDS					\
229	max(MAX_NUM_CHANNELS_SUPPORTED, HV_EVENT_FLAGS_COUNT)
230
231enum vmbus_connect_state {
232	DISCONNECTED,
233	CONNECTING,
234	CONNECTED,
235	DISCONNECTING
236};
237
238#define MAX_SIZE_CHANNEL_MESSAGE	HV_MESSAGE_PAYLOAD_BYTE_COUNT
239
240/*
241 * The CPU that Hyper-V will interrupt for VMBUS messages, such as
242 * CHANNELMSG_OFFERCHANNEL and CHANNELMSG_RESCIND_CHANNELOFFER.
243 */
244#define VMBUS_CONNECT_CPU	0
245
246struct vmbus_connection {
247	u32 msg_conn_id;
248
249	atomic_t offer_in_progress;
250
251	enum vmbus_connect_state conn_state;
252
253	atomic_t next_gpadl_handle;
254
255	struct completion  unload_event;
256	/*
257	 * Represents channel interrupts. Each bit position represents a
258	 * channel.  When a channel sends an interrupt via VMBUS, it finds its
259	 * bit in the sendInterruptPage, set it and calls Hv to generate a port
260	 * event. The other end receives the port event and parse the
261	 * recvInterruptPage to see which bit is set
262	 */
263	void *int_page;
264	void *send_int_page;
265	void *recv_int_page;
266
267	/*
268	 * 2 pages - 1st page for parent->child notification and 2nd
269	 * is child->parent notification
270	 */
271	struct hv_monitor_page *monitor_pages[2];
272	struct list_head chn_msg_list;
273	spinlock_t channelmsg_lock;
274
275	/* List of channels */
276	struct list_head chn_list;
277	struct mutex channel_mutex;
278
279	/* Array of channels */
280	struct vmbus_channel **channels;
281
282	/*
283	 * An offer message is handled first on the work_queue, and then
284	 * is further handled on handle_primary_chan_wq or
285	 * handle_sub_chan_wq.
286	 */
287	struct workqueue_struct *work_queue;
288	struct workqueue_struct *handle_primary_chan_wq;
289	struct workqueue_struct *handle_sub_chan_wq;
290	struct workqueue_struct *rescind_work_queue;
291
292	/*
293	 * On suspension of the vmbus, the accumulated offer messages
294	 * must be dropped.
295	 */
296	bool ignore_any_offer_msg;
297
298	/*
299	 * The number of sub-channels and hv_sock channels that should be
300	 * cleaned up upon suspend: sub-channels will be re-created upon
301	 * resume, and hv_sock channels should not survive suspend.
302	 */
303	atomic_t nr_chan_close_on_suspend;
304	/*
305	 * vmbus_bus_suspend() waits for "nr_chan_close_on_suspend" to
306	 * drop to zero.
307	 */
308	struct completion ready_for_suspend_event;
309
310	/*
311	 * Completed once the host has offered all boot-time channels.
312	 * Note that some channels may still be under process on a workqueue.
313	 */
314	struct completion all_offers_delivered_event;
315};
316
317
318struct vmbus_msginfo {
319	/* Bookkeeping stuff */
320	struct list_head msglist_entry;
321
322	/* The message itself */
323	unsigned char msg[];
324};
325
326
327extern struct vmbus_connection vmbus_connection;
328
329int vmbus_negotiate_version(struct vmbus_channel_msginfo *msginfo, u32 version);
330
331static inline void vmbus_send_interrupt(u32 relid)
332{
333	sync_set_bit(relid, vmbus_connection.send_int_page);
334}
335
336enum vmbus_message_handler_type {
337	/* The related handler can sleep. */
338	VMHT_BLOCKING = 0,
339
340	/* The related handler must NOT sleep. */
341	VMHT_NON_BLOCKING = 1,
342};
343
344struct vmbus_channel_message_table_entry {
345	enum vmbus_channel_message_type message_type;
346	enum vmbus_message_handler_type handler_type;
347	void (*message_handler)(struct vmbus_channel_message_header *msg);
348	u32 min_payload_len;
349};
350
351extern const struct vmbus_channel_message_table_entry
352	channel_message_table[CHANNELMSG_COUNT];
353
354
355/* General vmbus interface */
356
357bool vmbus_is_confidential(void);
358
359#if IS_ENABLED(CONFIG_HYPERV_VMBUS)
360/* Free the message slot and signal end-of-message if required */
361static inline void vmbus_signal_eom(struct hv_message *msg, u32 old_msg_type)
362{
363	/*
364	 * On crash we're reading some other CPU's message page and we need
365	 * to be careful: this other CPU may already had cleared the header
366	 * and the host may already had delivered some other message there.
367	 * In case we blindly write msg->header.message_type we're going
368	 * to lose it. We can still lose a message of the same type but
369	 * we count on the fact that there can only be one
370	 * CHANNELMSG_UNLOAD_RESPONSE and we don't care about other messages
371	 * on crash.
372	 */
373	if (cmpxchg(&msg->header.message_type, old_msg_type,
374		    HVMSG_NONE) != old_msg_type)
375		return;
376
377	/*
378	 * The cmpxchg() above does an implicit memory barrier to
379	 * ensure the write to MessageType (ie set to
380	 * HVMSG_NONE) happens before we read the
381	 * MessagePending and EOMing. Otherwise, the EOMing
382	 * will not deliver any more messages since there is
383	 * no empty slot
384	 */
385	if (msg->header.message_flags.msg_pending) {
386		/*
387		 * This will cause message queue rescan to
388		 * possibly deliver another msg from the
389		 * hypervisor
390		 */
391		if (vmbus_is_confidential())
392			hv_para_set_synic_register(HV_MSR_EOM, 0);
393		else
394			hv_set_msr(HV_MSR_EOM, 0);
395	}
396}
397
398extern int vmbus_interrupt;
399extern int vmbus_irq;
400#endif /* CONFIG_HYPERV_VMBUS */
401
402struct hv_device *vmbus_device_create(const guid_t *type,
403				      const guid_t *instance,
404				      struct vmbus_channel *channel);
405
406int vmbus_device_register(struct hv_device *child_device_obj);
407void vmbus_device_unregister(struct hv_device *device_obj);
408int vmbus_add_channel_kobj(struct hv_device *device_obj,
409			   struct vmbus_channel *channel);
410
411void vmbus_remove_channel_attr_group(struct vmbus_channel *channel);
412
413void vmbus_channel_map_relid(struct vmbus_channel *channel);
414void vmbus_channel_unmap_relid(struct vmbus_channel *channel);
415
416struct vmbus_channel *relid2channel(u32 relid);
417
418void vmbus_free_channels(void);
419
420/* Connection interface */
421
422int vmbus_connect(void);
423void vmbus_disconnect(void);
424
425int vmbus_post_msg(void *buffer, size_t buflen, bool can_sleep);
426
427void vmbus_on_event(unsigned long data);
428void vmbus_on_msg_dpc(unsigned long data);
429
430int hv_kvp_init(struct hv_util_service *srv);
431int hv_kvp_init_transport(void);
432void hv_kvp_deinit(void);
433int hv_kvp_pre_suspend(void);
434int hv_kvp_pre_resume(void);
435void hv_kvp_onchannelcallback(void *context);
436
437int hv_vss_init(struct hv_util_service *srv);
438int hv_vss_init_transport(void);
439void hv_vss_deinit(void);
440int hv_vss_pre_suspend(void);
441int hv_vss_pre_resume(void);
442void hv_vss_onchannelcallback(void *context);
443void vmbus_initiate_unload(bool crash);
444
445static inline void hv_poll_channel(struct vmbus_channel *channel,
446				   void (*cb)(void *))
447{
448	if (!channel)
449		return;
450	cb(channel);
451}
452
453enum hvutil_device_state {
454	HVUTIL_DEVICE_INIT = 0,  /* driver is loaded, waiting for userspace */
455	HVUTIL_READY,            /* userspace is registered */
456	HVUTIL_HOSTMSG_RECEIVED, /* message from the host was received */
457	HVUTIL_USERSPACE_REQ,    /* request to userspace was sent */
458	HVUTIL_USERSPACE_RECV,   /* reply from userspace was received */
459	HVUTIL_DEVICE_DYING,     /* driver unload is in progress */
460};
461
462enum delay {
463	INTERRUPT_DELAY = 0,
464	MESSAGE_DELAY   = 1,
465};
466
467extern const struct vmbus_device vmbus_devs[];
468
469static inline bool hv_is_perf_channel(struct vmbus_channel *channel)
470{
471	return vmbus_devs[channel->device_id].perf_device;
472}
473
474static inline size_t hv_dev_ring_size(struct vmbus_channel *channel)
475{
476	return vmbus_devs[channel->device_id].pref_ring_size;
477}
478
479static inline bool hv_is_allocated_cpu(unsigned int cpu)
480{
481	struct vmbus_channel *channel, *sc;
482
483	lockdep_assert_held(&vmbus_connection.channel_mutex);
484	/*
485	 * List additions/deletions as well as updates of the target CPUs are
486	 * protected by channel_mutex.
487	 */
488	list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
489		if (!hv_is_perf_channel(channel))
490			continue;
491		if (channel->target_cpu == cpu)
492			return true;
493		list_for_each_entry(sc, &channel->sc_list, sc_list) {
494			if (sc->target_cpu == cpu)
495				return true;
496		}
497	}
498	return false;
499}
500
501static inline void hv_set_allocated_cpu(unsigned int cpu)
502{
503	cpumask_set_cpu(cpu, &hv_context.hv_numa_map[cpu_to_node(cpu)]);
504}
505
506static inline void hv_clear_allocated_cpu(unsigned int cpu)
507{
508	if (hv_is_allocated_cpu(cpu))
509		return;
510	cpumask_clear_cpu(cpu, &hv_context.hv_numa_map[cpu_to_node(cpu)]);
511}
512
513static inline void hv_update_allocated_cpus(unsigned int old_cpu,
514					  unsigned int new_cpu)
515{
516	hv_set_allocated_cpu(new_cpu);
517	hv_clear_allocated_cpu(old_cpu);
518}
519
520#ifdef CONFIG_HYPERV_TESTING
521
522int hv_debug_add_dev_dir(struct hv_device *dev);
523void hv_debug_rm_dev_dir(struct hv_device *dev);
524void hv_debug_rm_all_dir(void);
525int hv_debug_init(void);
526void hv_debug_delay_test(struct vmbus_channel *channel, enum delay delay_type);
527
528#else /* CONFIG_HYPERV_TESTING */
529
530static inline void hv_debug_rm_dev_dir(struct hv_device *dev) {};
531static inline void hv_debug_rm_all_dir(void) {};
532static inline void hv_debug_delay_test(struct vmbus_channel *channel,
533				       enum delay delay_type) {};
534static inline int hv_debug_init(void)
535{
536	return -1;
537}
538
539static inline int hv_debug_add_dev_dir(struct hv_device *dev)
540{
541	return -1;
542}
543
544#endif /* CONFIG_HYPERV_TESTING */
545
546/* Create and remove sysfs entry for memory mapped ring buffers for a channel */
547int hv_create_ring_sysfs(struct vmbus_channel *channel,
548			 int (*hv_mmap_ring_buffer)(struct vmbus_channel *channel,
549						    struct vm_area_struct *vma));
550int hv_remove_ring_sysfs(struct vmbus_channel *channel);
551
552#endif /* _HYPERV_VMBUS_H */