projects/linux·blob v6.13 · refs

318 lines · 9.1 KiB

1/* SPDX-License-Identifier: GPL-2.0 */
2
3/*
4 * Linux-specific definitions for managing interactions with Microsoft's
5 * Hyper-V hypervisor. The definitions in this file are architecture
6 * independent. See arch/<arch>/include/asm/mshyperv.h for definitions
7 * that are specific to architecture <arch>.
8 *
9 * Definitions that are specified in the Hyper-V Top Level Functional
10 * Spec (TLFS) should not go in this file, but should instead go in
11 * hyperv-tlfs.h.
12 *
13 * Copyright (C) 2019, Microsoft, Inc.
14 *
15 * Author : Michael Kelley <[email protected]>
16 */
17
18#ifndef _ASM_GENERIC_MSHYPERV_H
19#define _ASM_GENERIC_MSHYPERV_H
20
21#include <linux/types.h>
22#include <linux/atomic.h>
23#include <linux/bitops.h>
24#include <acpi/acpi_numa.h>
25#include <linux/cpumask.h>
26#include <linux/nmi.h>
27#include <asm/ptrace.h>
28#include <asm/hyperv-tlfs.h>
29
30#define VTPM_BASE_ADDRESS 0xfed40000
31
32struct ms_hyperv_info {
33 u32 features;
34 u32 priv_high;
35 u32 misc_features;
36 u32 hints;
37 u32 nested_features;
38 u32 max_vp_index;
39 u32 max_lp_index;
40 u8 vtl;
41 union {
42 u32 isolation_config_a;
43 struct {
44 u32 paravisor_present : 1;
45 u32 reserved_a1 : 31;
46 };
47 };
48 union {
49 u32 isolation_config_b;
50 struct {
51 u32 cvm_type : 4;
52 u32 reserved_b1 : 1;
53 u32 shared_gpa_boundary_active : 1;
54 u32 shared_gpa_boundary_bits : 6;
55 u32 reserved_b2 : 20;
56 };
57 };
58 u64 shared_gpa_boundary;
59};
60extern struct ms_hyperv_info ms_hyperv;
61extern bool hv_nested;
62
63extern void * __percpu *hyperv_pcpu_input_arg;
64extern void * __percpu *hyperv_pcpu_output_arg;
65
66extern u64 hv_do_hypercall(u64 control, void *inputaddr, void *outputaddr);
67extern u64 hv_do_fast_hypercall8(u16 control, u64 input8);
68bool hv_isolation_type_snp(void);
69bool hv_isolation_type_tdx(void);
70
71static inline struct hv_proximity_domain_info hv_numa_node_to_pxm_info(int node)
72{
73 struct hv_proximity_domain_info pxm_info = {};
74
75 if (node != NUMA_NO_NODE) {
76 pxm_info.domain_id = node_to_pxm(node);
77 pxm_info.flags.proximity_info_valid = 1;
78 pxm_info.flags.proximity_preferred = 1;
79 }
80
81 return pxm_info;
82}
83
84/* Helper functions that provide a consistent pattern for checking Hyper-V hypercall status. */
85static inline int hv_result(u64 status)
86{
87 return status & HV_HYPERCALL_RESULT_MASK;
88}
89
90static inline bool hv_result_success(u64 status)
91{
92 return hv_result(status) == HV_STATUS_SUCCESS;
93}
94
95static inline unsigned int hv_repcomp(u64 status)
96{
97 /* Bits [43:32] of status have 'Reps completed' data. */
98 return (status & HV_HYPERCALL_REP_COMP_MASK) >>
99 HV_HYPERCALL_REP_COMP_OFFSET;
100}
101
102/*
103 * Rep hypercalls. Callers of this functions are supposed to ensure that
104 * rep_count and varhead_size comply with Hyper-V hypercall definition.
105 */
106static inline u64 hv_do_rep_hypercall(u16 code, u16 rep_count, u16 varhead_size,
107 void *input, void *output)
108{
109 u64 control = code;
110 u64 status;
111 u16 rep_comp;
112
113 control |= (u64)varhead_size << HV_HYPERCALL_VARHEAD_OFFSET;
114 control |= (u64)rep_count << HV_HYPERCALL_REP_COMP_OFFSET;
115
116 do {
117 status = hv_do_hypercall(control, input, output);
118 if (!hv_result_success(status))
119 return status;
120
121 rep_comp = hv_repcomp(status);
122
123 control &= ~HV_HYPERCALL_REP_START_MASK;
124 control |= (u64)rep_comp << HV_HYPERCALL_REP_START_OFFSET;
125
126 touch_nmi_watchdog();
127 } while (rep_comp < rep_count);
128
129 return status;
130}
131
132/* Generate the guest OS identifier as described in the Hyper-V TLFS */
133static inline u64 hv_generate_guest_id(u64 kernel_version)
134{
135 u64 guest_id;
136
137 guest_id = (((u64)HV_LINUX_VENDOR_ID) << 48);
138 guest_id |= (kernel_version << 16);
139
140 return guest_id;
141}
142
143/* Free the message slot and signal end-of-message if required */
144static inline void vmbus_signal_eom(struct hv_message *msg, u32 old_msg_type)
145{
146 /*
147 * On crash we're reading some other CPU's message page and we need
148 * to be careful: this other CPU may already had cleared the header
149 * and the host may already had delivered some other message there.
150 * In case we blindly write msg->header.message_type we're going
151 * to lose it. We can still lose a message of the same type but
152 * we count on the fact that there can only be one
153 * CHANNELMSG_UNLOAD_RESPONSE and we don't care about other messages
154 * on crash.
155 */
156 if (cmpxchg(&msg->header.message_type, old_msg_type,
157 HVMSG_NONE) != old_msg_type)
158 return;
159
160 /*
161 * The cmxchg() above does an implicit memory barrier to
162 * ensure the write to MessageType (ie set to
163 * HVMSG_NONE) happens before we read the
164 * MessagePending and EOMing. Otherwise, the EOMing
165 * will not deliver any more messages since there is
166 * no empty slot
167 */
168 if (msg->header.message_flags.msg_pending) {
169 /*
170 * This will cause message queue rescan to
171 * possibly deliver another msg from the
172 * hypervisor
173 */
174 hv_set_msr(HV_MSR_EOM, 0);
175 }
176}
177
178int hv_get_hypervisor_version(union hv_hypervisor_version_info *info);
179
180void hv_setup_vmbus_handler(void (*handler)(void));
181void hv_remove_vmbus_handler(void);
182void hv_setup_stimer0_handler(void (*handler)(void));
183void hv_remove_stimer0_handler(void);
184
185void hv_setup_kexec_handler(void (*handler)(void));
186void hv_remove_kexec_handler(void);
187void hv_setup_crash_handler(void (*handler)(struct pt_regs *regs));
188void hv_remove_crash_handler(void);
189
190extern int vmbus_interrupt;
191extern int vmbus_irq;
192
193extern bool hv_root_partition;
194
195#if IS_ENABLED(CONFIG_HYPERV)
196/*
197 * Hypervisor's notion of virtual processor ID is different from
198 * Linux' notion of CPU ID. This information can only be retrieved
199 * in the context of the calling CPU. Setup a map for easy access
200 * to this information.
201 */
202extern u32 *hv_vp_index;
203extern u32 hv_max_vp_index;
204
205extern u64 (*hv_read_reference_counter)(void);
206
207/* Sentinel value for an uninitialized entry in hv_vp_index array */
208#define VP_INVAL U32_MAX
209
210int __init hv_common_init(void);
211void __init hv_common_free(void);
212void __init ms_hyperv_late_init(void);
213int hv_common_cpu_init(unsigned int cpu);
214int hv_common_cpu_die(unsigned int cpu);
215
216void *hv_alloc_hyperv_page(void);
217void *hv_alloc_hyperv_zeroed_page(void);
218void hv_free_hyperv_page(void *addr);
219
220/**
221 * hv_cpu_number_to_vp_number() - Map CPU to VP.
222 * @cpu_number: CPU number in Linux terms
223 *
224 * This function returns the mapping between the Linux processor
225 * number and the hypervisor's virtual processor number, useful
226 * in making hypercalls and such that talk about specific
227 * processors.
228 *
229 * Return: Virtual processor number in Hyper-V terms
230 */
231static inline int hv_cpu_number_to_vp_number(int cpu_number)
232{
233 return hv_vp_index[cpu_number];
234}
235
236static inline int __cpumask_to_vpset(struct hv_vpset *vpset,
237 const struct cpumask *cpus,
238 bool (*func)(int cpu))
239{
240 int cpu, vcpu, vcpu_bank, vcpu_offset, nr_bank = 1;
241 int max_vcpu_bank = hv_max_vp_index / HV_VCPUS_PER_SPARSE_BANK;
242
243 /* vpset.valid_bank_mask can represent up to HV_MAX_SPARSE_VCPU_BANKS banks */
244 if (max_vcpu_bank >= HV_MAX_SPARSE_VCPU_BANKS)
245 return 0;
246
247 /*
248 * Clear all banks up to the maximum possible bank as hv_tlb_flush_ex
249 * structs are not cleared between calls, we risk flushing unneeded
250 * vCPUs otherwise.
251 */
252 for (vcpu_bank = 0; vcpu_bank <= max_vcpu_bank; vcpu_bank++)
253 vpset->bank_contents[vcpu_bank] = 0;
254
255 /*
256 * Some banks may end up being empty but this is acceptable.
257 */
258 for_each_cpu(cpu, cpus) {
259 if (func && func(cpu))
260 continue;
261 vcpu = hv_cpu_number_to_vp_number(cpu);
262 if (vcpu == VP_INVAL)
263 return -1;
264 vcpu_bank = vcpu / HV_VCPUS_PER_SPARSE_BANK;
265 vcpu_offset = vcpu % HV_VCPUS_PER_SPARSE_BANK;
266 __set_bit(vcpu_offset, (unsigned long *)
267 &vpset->bank_contents[vcpu_bank]);
268 if (vcpu_bank >= nr_bank)
269 nr_bank = vcpu_bank + 1;
270 }
271 vpset->valid_bank_mask = GENMASK_ULL(nr_bank - 1, 0);
272 return nr_bank;
273}
274
275/*
276 * Convert a Linux cpumask into a Hyper-V VPset. In the _skip variant,
277 * 'func' is called for each CPU present in cpumask. If 'func' returns
278 * true, that CPU is skipped -- i.e., that CPU from cpumask is *not*
279 * added to the Hyper-V VPset. If 'func' is NULL, no CPUs are
280 * skipped.
281 */
282static inline int cpumask_to_vpset(struct hv_vpset *vpset,
283 const struct cpumask *cpus)
284{
285 return __cpumask_to_vpset(vpset, cpus, NULL);
286}
287
288static inline int cpumask_to_vpset_skip(struct hv_vpset *vpset,
289 const struct cpumask *cpus,
290 bool (*func)(int cpu))
291{
292 return __cpumask_to_vpset(vpset, cpus, func);
293}
294
295void hyperv_report_panic(struct pt_regs *regs, long err, bool in_die);
296bool hv_is_hyperv_initialized(void);
297bool hv_is_hibernation_supported(void);
298enum hv_isolation_type hv_get_isolation_type(void);
299bool hv_is_isolation_supported(void);
300bool hv_isolation_type_snp(void);
301u64 hv_ghcb_hypercall(u64 control, void *input, void *output, u32 input_size);
302u64 hv_tdx_hypercall(u64 control, u64 param1, u64 param2);
303void hyperv_cleanup(void);
304bool hv_query_ext_cap(u64 cap_query);
305void hv_setup_dma_ops(struct device *dev, bool coherent);
306#else /* CONFIG_HYPERV */
307static inline bool hv_is_hyperv_initialized(void) { return false; }
308static inline bool hv_is_hibernation_supported(void) { return false; }
309static inline void hyperv_cleanup(void) {}
310static inline void ms_hyperv_late_init(void) {}
311static inline bool hv_is_isolation_supported(void) { return false; }
312static inline enum hv_isolation_type hv_get_isolation_type(void)
313{
314 return HV_ISOLATION_TYPE_NONE;
315}
316#endif /* CONFIG_HYPERV */
317
318#endif