drivers/hv/hyperv_vmbus.h

0001 /* SPDX-License-Identifier: GPL-2.0-only */
0002 /*
0003  *
0004  * Copyright (c) 2011, Microsoft Corporation.
0005  *
0006  * Authors:
0007  *   Haiyang Zhang <haiyangz@microsoft.com>
0008  *   Hank Janssen  <hjanssen@microsoft.com>
0009  *   K. Y. Srinivasan <kys@microsoft.com>
0010  */
0011
0012 #ifndef _HYPERV_VMBUS_H
0013 #define _HYPERV_VMBUS_H
0014
0015 #include <linux/list.h>
0016 #include <linux/bitops.h>
0017 #include <asm/sync_bitops.h>
0018 #include <asm/hyperv-tlfs.h>
0019 #include <linux/atomic.h>
0020 #include <linux/hyperv.h>
0021 #include <linux/interrupt.h>
0022
0023 #include "hv_trace.h"
0024
0025 /*
0026  * Timeout for services such as KVP and fcopy.
0027  */
0028 #define HV_UTIL_TIMEOUT 30
0029
0030 /*
0031  * Timeout for guest-host handshake for services.
0032  */
0033 #define HV_UTIL_NEGO_TIMEOUT 55
0034
0035
0036 /* Definitions for the monitored notification facility */
0037 union hv_monitor_trigger_group {
0038     u64 as_uint64;
0039     struct {
0040         u32 pending;
0041         u32 armed;
0042     };
0043 };
0044
0045 struct hv_monitor_parameter {
0046     union hv_connection_id connectionid;
0047     u16 flagnumber;
0048     u16 rsvdz;
0049 };
0050
0051 union hv_monitor_trigger_state {
0052     u32 asu32;
0053
0054     struct {
0055         u32 group_enable:4;
0056         u32 rsvdz:28;
0057     };
0058 };
0059
0060 /* struct hv_monitor_page Layout */
0061 /* ------------------------------------------------------ */
0062 /* | 0   | TriggerState (4 bytes) | Rsvd1 (4 bytes)     | */
0063 /* | 8   | TriggerGroup[0]                              | */
0064 /* | 10  | TriggerGroup[1]                              | */
0065 /* | 18  | TriggerGroup[2]                              | */
0066 /* | 20  | TriggerGroup[3]                              | */
0067 /* | 28  | Rsvd2[0]                                     | */
0068 /* | 30  | Rsvd2[1]                                     | */
0069 /* | 38  | Rsvd2[2]                                     | */
0070 /* | 40  | NextCheckTime[0][0]    | NextCheckTime[0][1] | */
0071 /* | ...                                                | */
0072 /* | 240 | Latency[0][0..3]                             | */
0073 /* | 340 | Rsvz3[0]                                     | */
0074 /* | 440 | Parameter[0][0]                              | */
0075 /* | 448 | Parameter[0][1]                              | */
0076 /* | ...                                                | */
0077 /* | 840 | Rsvd4[0]                                     | */
0078 /* ------------------------------------------------------ */
0079 struct hv_monitor_page {
0080     union hv_monitor_trigger_state trigger_state;
0081     u32 rsvdz1;
0082
0083     union hv_monitor_trigger_group trigger_group[4];
0084     u64 rsvdz2[3];
0085
0086     s32 next_checktime[4][32];
0087
0088     u16 latency[4][32];
0089     u64 rsvdz3[32];
0090
0091     struct hv_monitor_parameter parameter[4][32];
0092
0093     u8 rsvdz4[1984];
0094 };
0095
0096 #define HV_HYPERCALL_PARAM_ALIGN    sizeof(u64)
0097
0098 /* Definition of the hv_post_message hypercall input structure. */
0099 struct hv_input_post_message {
0100     union hv_connection_id connectionid;
0101     u32 reserved;
0102     u32 message_type;
0103     u32 payload_size;
0104     u64 payload[HV_MESSAGE_PAYLOAD_QWORD_COUNT];
0105 };
0106
0107
0108 enum {
0109     VMBUS_MESSAGE_CONNECTION_ID = 1,
0110     VMBUS_MESSAGE_CONNECTION_ID_4   = 4,
0111     VMBUS_MESSAGE_PORT_ID       = 1,
0112     VMBUS_EVENT_CONNECTION_ID   = 2,
0113     VMBUS_EVENT_PORT_ID     = 2,
0114     VMBUS_MONITOR_CONNECTION_ID = 3,
0115     VMBUS_MONITOR_PORT_ID       = 3,
0116     VMBUS_MESSAGE_SINT      = 2,
0117 };
0118
0119 /*
0120  * Per cpu state for channel handling
0121  */
0122 struct hv_per_cpu_context {
0123     void *synic_message_page;
0124     void *synic_event_page;
0125     /*
0126      * buffer to post messages to the host.
0127      */
0128     void *post_msg_page;
0129
0130     /*
0131      * Starting with win8, we can take channel interrupts on any CPU;
0132      * we will manage the tasklet that handles events messages on a per CPU
0133      * basis.
0134      */
0135     struct tasklet_struct msg_dpc;
0136 };
0137
0138 struct hv_context {
0139     /* We only support running on top of Hyper-V
0140      * So at this point this really can only contain the Hyper-V ID
0141      */
0142     u64 guestid;
0143
0144     struct hv_per_cpu_context __percpu *cpu_context;
0145
0146     /*
0147      * To manage allocations in a NUMA node.
0148      * Array indexed by numa node ID.
0149      */
0150     struct cpumask *hv_numa_map;
0151 };
0152
0153 extern struct hv_context hv_context;
0154
0155 /* Hv Interface */
0156
0157 extern int hv_init(void);
0158
0159 extern int hv_post_message(union hv_connection_id connection_id,
0160              enum hv_message_type message_type,
0161              void *payload, size_t payload_size);
0162
0163 extern int hv_synic_alloc(void);
0164
0165 extern void hv_synic_free(void);
0166
0167 extern void hv_synic_enable_regs(unsigned int cpu);
0168 extern int hv_synic_init(unsigned int cpu);
0169
0170 extern void hv_synic_disable_regs(unsigned int cpu);
0171 extern int hv_synic_cleanup(unsigned int cpu);
0172
0173 /* Interface */
0174
0175 void hv_ringbuffer_pre_init(struct vmbus_channel *channel);
0176
0177 int hv_ringbuffer_init(struct hv_ring_buffer_info *ring_info,
0178                struct page *pages, u32 pagecnt, u32 max_pkt_size);
0179
0180 void hv_ringbuffer_cleanup(struct hv_ring_buffer_info *ring_info);
0181
0182 int hv_ringbuffer_write(struct vmbus_channel *channel,
0183             const struct kvec *kv_list, u32 kv_count,
0184             u64 requestid, u64 *trans_id);
0185
0186 int hv_ringbuffer_read(struct vmbus_channel *channel,
0187                void *buffer, u32 buflen, u32 *buffer_actual_len,
0188                u64 *requestid, bool raw);
0189
0190 /*
0191  * The Maximum number of channels (16384) is determined by the size of the
0192  * interrupt page, which is HV_HYP_PAGE_SIZE. 1/2 of HV_HYP_PAGE_SIZE is to
0193  * send endpoint interrupts, and the other is to receive endpoint interrupts.
0194  */
0195 #define MAX_NUM_CHANNELS    ((HV_HYP_PAGE_SIZE >> 1) << 3)
0196
0197 /* The value here must be in multiple of 32 */
0198 #define MAX_NUM_CHANNELS_SUPPORTED  256
0199
0200 #define MAX_CHANNEL_RELIDS                  \
0201     max(MAX_NUM_CHANNELS_SUPPORTED, HV_EVENT_FLAGS_COUNT)
0202
0203 enum vmbus_connect_state {
0204     DISCONNECTED,
0205     CONNECTING,
0206     CONNECTED,
0207     DISCONNECTING
0208 };
0209
0210 #define MAX_SIZE_CHANNEL_MESSAGE    HV_MESSAGE_PAYLOAD_BYTE_COUNT
0211
0212 /*
0213  * The CPU that Hyper-V will interrupt for VMBUS messages, such as
0214  * CHANNELMSG_OFFERCHANNEL and CHANNELMSG_RESCIND_CHANNELOFFER.
0215  */
0216 #define VMBUS_CONNECT_CPU   0
0217
0218 struct vmbus_connection {
0219     u32 msg_conn_id;
0220
0221     atomic_t offer_in_progress;
0222
0223     enum vmbus_connect_state conn_state;
0224
0225     atomic_t next_gpadl_handle;
0226
0227     struct completion  unload_event;
0228     /*
0229      * Represents channel interrupts. Each bit position represents a
0230      * channel.  When a channel sends an interrupt via VMBUS, it finds its
0231      * bit in the sendInterruptPage, set it and calls Hv to generate a port
0232      * event. The other end receives the port event and parse the
0233      * recvInterruptPage to see which bit is set
0234      */
0235     void *int_page;
0236     void *send_int_page;
0237     void *recv_int_page;
0238
0239     /*
0240      * 2 pages - 1st page for parent->child notification and 2nd
0241      * is child->parent notification
0242      */
0243     struct hv_monitor_page *monitor_pages[2];
0244     void *monitor_pages_original[2];
0245     phys_addr_t monitor_pages_pa[2];
0246     struct list_head chn_msg_list;
0247     spinlock_t channelmsg_lock;
0248
0249     /* List of channels */
0250     struct list_head chn_list;
0251     struct mutex channel_mutex;
0252
0253     /* Array of channels */
0254     struct vmbus_channel **channels;
0255
0256     /*
0257      * An offer message is handled first on the work_queue, and then
0258      * is further handled on handle_primary_chan_wq or
0259      * handle_sub_chan_wq.
0260      */
0261     struct workqueue_struct *work_queue;
0262     struct workqueue_struct *handle_primary_chan_wq;
0263     struct workqueue_struct *handle_sub_chan_wq;
0264     struct workqueue_struct *rescind_work_queue;
0265
0266     /*
0267      * On suspension of the vmbus, the accumulated offer messages
0268      * must be dropped.
0269      */
0270     bool ignore_any_offer_msg;
0271
0272     /*
0273      * The number of sub-channels and hv_sock channels that should be
0274      * cleaned up upon suspend: sub-channels will be re-created upon
0275      * resume, and hv_sock channels should not survive suspend.
0276      */
0277     atomic_t nr_chan_close_on_suspend;
0278     /*
0279      * vmbus_bus_suspend() waits for "nr_chan_close_on_suspend" to
0280      * drop to zero.
0281      */
0282     struct completion ready_for_suspend_event;
0283
0284     /*
0285      * The number of primary channels that should be "fixed up"
0286      * upon resume: these channels are re-offered upon resume, and some
0287      * fields of the channel offers (i.e. child_relid and connection_id)
0288      * can change, so the old offermsg must be fixed up, before the resume
0289      * callbacks of the VSC drivers start to further touch the channels.
0290      */
0291     atomic_t nr_chan_fixup_on_resume;
0292     /*
0293      * vmbus_bus_resume() waits for "nr_chan_fixup_on_resume" to
0294      * drop to zero.
0295      */
0296     struct completion ready_for_resume_event;
0297 };
0298
0299
0300 struct vmbus_msginfo {
0301     /* Bookkeeping stuff */
0302     struct list_head msglist_entry;
0303
0304     /* The message itself */
0305     unsigned char msg[];
0306 };
0307
0308
0309 extern struct vmbus_connection vmbus_connection;
0310
0311 int vmbus_negotiate_version(struct vmbus_channel_msginfo *msginfo, u32 version);
0312
0313 static inline void vmbus_send_interrupt(u32 relid)
0314 {
0315     sync_set_bit(relid, vmbus_connection.send_int_page);
0316 }
0317
0318 enum vmbus_message_handler_type {
0319     /* The related handler can sleep. */
0320     VMHT_BLOCKING = 0,
0321
0322     /* The related handler must NOT sleep. */
0323     VMHT_NON_BLOCKING = 1,
0324 };
0325
0326 struct vmbus_channel_message_table_entry {
0327     enum vmbus_channel_message_type message_type;
0328     enum vmbus_message_handler_type handler_type;
0329     void (*message_handler)(struct vmbus_channel_message_header *msg);
0330     u32 min_payload_len;
0331 };
0332
0333 extern const struct vmbus_channel_message_table_entry
0334     channel_message_table[CHANNELMSG_COUNT];
0335
0336
0337 /* General vmbus interface */
0338
0339 struct hv_device *vmbus_device_create(const guid_t *type,
0340                       const guid_t *instance,
0341                       struct vmbus_channel *channel);
0342
0343 int vmbus_device_register(struct hv_device *child_device_obj);
0344 void vmbus_device_unregister(struct hv_device *device_obj);
0345 int vmbus_add_channel_kobj(struct hv_device *device_obj,
0346                struct vmbus_channel *channel);
0347
0348 void vmbus_remove_channel_attr_group(struct vmbus_channel *channel);
0349
0350 void vmbus_channel_map_relid(struct vmbus_channel *channel);
0351 void vmbus_channel_unmap_relid(struct vmbus_channel *channel);
0352
0353 struct vmbus_channel *relid2channel(u32 relid);
0354
0355 void vmbus_free_channels(void);
0356
0357 /* Connection interface */
0358
0359 int vmbus_connect(void);
0360 void vmbus_disconnect(void);
0361
0362 int vmbus_post_msg(void *buffer, size_t buflen, bool can_sleep);
0363
0364 void vmbus_on_event(unsigned long data);
0365 void vmbus_on_msg_dpc(unsigned long data);
0366
0367 int hv_kvp_init(struct hv_util_service *srv);
0368 void hv_kvp_deinit(void);
0369 int hv_kvp_pre_suspend(void);
0370 int hv_kvp_pre_resume(void);
0371 void hv_kvp_onchannelcallback(void *context);
0372
0373 int hv_vss_init(struct hv_util_service *srv);
0374 void hv_vss_deinit(void);
0375 int hv_vss_pre_suspend(void);
0376 int hv_vss_pre_resume(void);
0377 void hv_vss_onchannelcallback(void *context);
0378
0379 int hv_fcopy_init(struct hv_util_service *srv);
0380 void hv_fcopy_deinit(void);
0381 int hv_fcopy_pre_suspend(void);
0382 int hv_fcopy_pre_resume(void);
0383 void hv_fcopy_onchannelcallback(void *context);
0384 void vmbus_initiate_unload(bool crash);
0385
0386 static inline void hv_poll_channel(struct vmbus_channel *channel,
0387                    void (*cb)(void *))
0388 {
0389     if (!channel)
0390         return;
0391     cb(channel);
0392 }
0393
0394 enum hvutil_device_state {
0395     HVUTIL_DEVICE_INIT = 0,  /* driver is loaded, waiting for userspace */
0396     HVUTIL_READY,            /* userspace is registered */
0397     HVUTIL_HOSTMSG_RECEIVED, /* message from the host was received */
0398     HVUTIL_USERSPACE_REQ,    /* request to userspace was sent */
0399     HVUTIL_USERSPACE_RECV,   /* reply from userspace was received */
0400     HVUTIL_DEVICE_DYING,     /* driver unload is in progress */
0401 };
0402
0403 enum delay {
0404     INTERRUPT_DELAY = 0,
0405     MESSAGE_DELAY   = 1,
0406 };
0407
0408 extern const struct vmbus_device vmbus_devs[];
0409
0410 static inline bool hv_is_perf_channel(struct vmbus_channel *channel)
0411 {
0412     return vmbus_devs[channel->device_id].perf_device;
0413 }
0414
0415 static inline bool hv_is_allocated_cpu(unsigned int cpu)
0416 {
0417     struct vmbus_channel *channel, *sc;
0418
0419     lockdep_assert_held(&vmbus_connection.channel_mutex);
0420     /*
0421      * List additions/deletions as well as updates of the target CPUs are
0422      * protected by channel_mutex.
0423      */
0424     list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
0425         if (!hv_is_perf_channel(channel))
0426             continue;
0427         if (channel->target_cpu == cpu)
0428             return true;
0429         list_for_each_entry(sc, &channel->sc_list, sc_list) {
0430             if (sc->target_cpu == cpu)
0431                 return true;
0432         }
0433     }
0434     return false;
0435 }
0436
0437 static inline void hv_set_allocated_cpu(unsigned int cpu)
0438 {
0439     cpumask_set_cpu(cpu, &hv_context.hv_numa_map[cpu_to_node(cpu)]);
0440 }
0441
0442 static inline void hv_clear_allocated_cpu(unsigned int cpu)
0443 {
0444     if (hv_is_allocated_cpu(cpu))
0445         return;
0446     cpumask_clear_cpu(cpu, &hv_context.hv_numa_map[cpu_to_node(cpu)]);
0447 }
0448
0449 static inline void hv_update_allocated_cpus(unsigned int old_cpu,
0450                       unsigned int new_cpu)
0451 {
0452     hv_set_allocated_cpu(new_cpu);
0453     hv_clear_allocated_cpu(old_cpu);
0454 }
0455
0456 #ifdef CONFIG_HYPERV_TESTING
0457
0458 int hv_debug_add_dev_dir(struct hv_device *dev);
0459 void hv_debug_rm_dev_dir(struct hv_device *dev);
0460 void hv_debug_rm_all_dir(void);
0461 int hv_debug_init(void);
0462 void hv_debug_delay_test(struct vmbus_channel *channel, enum delay delay_type);
0463
0464 #else /* CONFIG_HYPERV_TESTING */
0465
0466 static inline void hv_debug_rm_dev_dir(struct hv_device *dev) {};
0467 static inline void hv_debug_rm_all_dir(void) {};
0468 static inline void hv_debug_delay_test(struct vmbus_channel *channel,
0469                        enum delay delay_type) {};
0470 static inline int hv_debug_init(void)
0471 {
0472     return -1;
0473 }
0474
0475 static inline int hv_debug_add_dev_dir(struct hv_device *dev)
0476 {
0477     return -1;
0478 }
0479
0480 #endif /* CONFIG_HYPERV_TESTING */
0481
0482 #endif /* _HYPERV_VMBUS_H */