1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_X86_MSHYPER_H
3 #define _ASM_X86_MSHYPER_H
4
5 #include <linux/types.h>
6 #include <linux/atomic.h>
7 #include <linux/nmi.h>
8 #include <asm/io.h>
9 #include <asm/hyperv.h>
10 #include <asm/nospec-branch.h>
11
12 /*
13 * The below CPUID leaves are present if VersionAndFeatures.HypervisorPresent
14 * is set by CPUID(HVCPUID_VERSION_FEATURES).
15 */
16 enum hv_cpuid_function {
17 HVCPUID_VERSION_FEATURES = 0x00000001,
18 HVCPUID_VENDOR_MAXFUNCTION = 0x40000000,
19 HVCPUID_INTERFACE = 0x40000001,
20
21 /*
22 * The remaining functions depend on the value of
23 * HVCPUID_INTERFACE
24 */
25 HVCPUID_VERSION = 0x40000002,
26 HVCPUID_FEATURES = 0x40000003,
27 HVCPUID_ENLIGHTENMENT_INFO = 0x40000004,
28 HVCPUID_IMPLEMENTATION_LIMITS = 0x40000005,
29 };
30
31 struct ms_hyperv_info {
32 u32 features;
33 u32 misc_features;
34 u32 hints;
35 u32 max_vp_index;
36 u32 max_lp_index;
37 };
38
39 extern struct ms_hyperv_info ms_hyperv;
40
41 /*
42 * Declare the MSR used to setup pages used to communicate with the hypervisor.
43 */
44 union hv_x64_msr_hypercall_contents {
45 u64 as_uint64;
46 struct {
47 u64 enable:1;
48 u64 reserved:11;
49 u64 guest_physical_address:52;
50 };
51 };
52
53 /*
54 * TSC page layout.
55 */
56
57 struct ms_hyperv_tsc_page {
58 volatile u32 tsc_sequence;
59 u32 reserved1;
60 volatile u64 tsc_scale;
61 volatile s64 tsc_offset;
62 u64 reserved2[509];
63 };
64
65 /*
66 * The guest OS needs to register the guest ID with the hypervisor.
67 * The guest ID is a 64 bit entity and the structure of this ID is
68 * specified in the Hyper-V specification:
69 *
70 * msdn.microsoft.com/en-us/library/windows/hardware/ff542653%28v=vs.85%29.aspx
71 *
72 * While the current guideline does not specify how Linux guest ID(s)
73 * need to be generated, our plan is to publish the guidelines for
74 * Linux and other guest operating systems that currently are hosted
75 * on Hyper-V. The implementation here conforms to this yet
76 * unpublished guidelines.
77 *
78 *
79 * Bit(s)
80 * 63 - Indicates if the OS is Open Source or not; 1 is Open Source
81 * 62:56 - Os Type; Linux is 0x100
82 * 55:48 - Distro specific identification
83 * 47:16 - Linux kernel version number
84 * 15:0 - Distro specific identification
85 *
86 *
87 */
88
89 #define HV_LINUX_VENDOR_ID 0x8100
90
91 /*
92 * Generate the guest ID based on the guideline described above.
93 */
94
generate_guest_id(__u64 d_info1,__u64 kernel_version,__u64 d_info2)95 static inline __u64 generate_guest_id(__u64 d_info1, __u64 kernel_version,
96 __u64 d_info2)
97 {
98 __u64 guest_id = 0;
99
100 guest_id = (((__u64)HV_LINUX_VENDOR_ID) << 48);
101 guest_id |= (d_info1 << 48);
102 guest_id |= (kernel_version << 16);
103 guest_id |= d_info2;
104
105 return guest_id;
106 }
107
108
109 /* Free the message slot and signal end-of-message if required */
vmbus_signal_eom(struct hv_message * msg,u32 old_msg_type)110 static inline void vmbus_signal_eom(struct hv_message *msg, u32 old_msg_type)
111 {
112 /*
113 * On crash we're reading some other CPU's message page and we need
114 * to be careful: this other CPU may already had cleared the header
115 * and the host may already had delivered some other message there.
116 * In case we blindly write msg->header.message_type we're going
117 * to lose it. We can still lose a message of the same type but
118 * we count on the fact that there can only be one
119 * CHANNELMSG_UNLOAD_RESPONSE and we don't care about other messages
120 * on crash.
121 */
122 if (cmpxchg(&msg->header.message_type, old_msg_type,
123 HVMSG_NONE) != old_msg_type)
124 return;
125
126 /*
127 * Make sure the write to MessageType (ie set to
128 * HVMSG_NONE) happens before we read the
129 * MessagePending and EOMing. Otherwise, the EOMing
130 * will not deliver any more messages since there is
131 * no empty slot
132 */
133 mb();
134
135 if (msg->header.message_flags.msg_pending) {
136 /*
137 * This will cause message queue rescan to
138 * possibly deliver another msg from the
139 * hypervisor
140 */
141 wrmsrl(HV_X64_MSR_EOM, 0);
142 }
143 }
144
145 #define hv_init_timer(timer, tick) wrmsrl(timer, tick)
146 #define hv_init_timer_config(config, val) wrmsrl(config, val)
147
148 #define hv_get_simp(val) rdmsrl(HV_X64_MSR_SIMP, val)
149 #define hv_set_simp(val) wrmsrl(HV_X64_MSR_SIMP, val)
150
151 #define hv_get_siefp(val) rdmsrl(HV_X64_MSR_SIEFP, val)
152 #define hv_set_siefp(val) wrmsrl(HV_X64_MSR_SIEFP, val)
153
154 #define hv_get_synic_state(val) rdmsrl(HV_X64_MSR_SCONTROL, val)
155 #define hv_set_synic_state(val) wrmsrl(HV_X64_MSR_SCONTROL, val)
156
157 #define hv_get_vp_index(index) rdmsrl(HV_X64_MSR_VP_INDEX, index)
158
159 #define hv_get_synint_state(int_num, val) rdmsrl(int_num, val)
160 #define hv_set_synint_state(int_num, val) wrmsrl(int_num, val)
161
162 void hyperv_callback_vector(void);
163 #ifdef CONFIG_TRACING
164 #define trace_hyperv_callback_vector hyperv_callback_vector
165 #endif
166 void hyperv_vector_handler(struct pt_regs *regs);
167 void hv_setup_vmbus_irq(void (*handler)(void));
168 void hv_remove_vmbus_irq(void);
169
170 void hv_setup_kexec_handler(void (*handler)(void));
171 void hv_remove_kexec_handler(void);
172 void hv_setup_crash_handler(void (*handler)(struct pt_regs *regs));
173 void hv_remove_crash_handler(void);
174
175 #if IS_ENABLED(CONFIG_HYPERV)
176 extern struct clocksource *hyperv_cs;
177 extern void *hv_hypercall_pg;
178
hv_do_hypercall(u64 control,void * input,void * output)179 static inline u64 hv_do_hypercall(u64 control, void *input, void *output)
180 {
181 u64 input_address = input ? virt_to_phys(input) : 0;
182 u64 output_address = output ? virt_to_phys(output) : 0;
183 u64 hv_status;
184
185 #ifdef CONFIG_X86_64
186 if (!hv_hypercall_pg)
187 return U64_MAX;
188
189 __asm__ __volatile__("mov %4, %%r8\n"
190 CALL_NOSPEC
191 : "=a" (hv_status), ASM_CALL_CONSTRAINT,
192 "+c" (control), "+d" (input_address)
193 : "r" (output_address),
194 THUNK_TARGET(hv_hypercall_pg)
195 : "cc", "memory", "r8", "r9", "r10", "r11");
196 #else
197 u32 input_address_hi = upper_32_bits(input_address);
198 u32 input_address_lo = lower_32_bits(input_address);
199 u32 output_address_hi = upper_32_bits(output_address);
200 u32 output_address_lo = lower_32_bits(output_address);
201
202 if (!hv_hypercall_pg)
203 return U64_MAX;
204
205 __asm__ __volatile__(CALL_NOSPEC
206 : "=A" (hv_status),
207 "+c" (input_address_lo), ASM_CALL_CONSTRAINT
208 : "A" (control),
209 "b" (input_address_hi),
210 "D"(output_address_hi), "S"(output_address_lo),
211 THUNK_TARGET(hv_hypercall_pg)
212 : "cc", "memory");
213 #endif /* !x86_64 */
214 return hv_status;
215 }
216
217 #define HV_HYPERCALL_RESULT_MASK GENMASK_ULL(15, 0)
218 #define HV_HYPERCALL_FAST_BIT BIT(16)
219 #define HV_HYPERCALL_VARHEAD_OFFSET 17
220 #define HV_HYPERCALL_REP_COMP_OFFSET 32
221 #define HV_HYPERCALL_REP_COMP_MASK GENMASK_ULL(43, 32)
222 #define HV_HYPERCALL_REP_START_OFFSET 48
223 #define HV_HYPERCALL_REP_START_MASK GENMASK_ULL(59, 48)
224
225 /* Fast hypercall with 8 bytes of input and no output */
hv_do_fast_hypercall8(u16 code,u64 input1)226 static inline u64 hv_do_fast_hypercall8(u16 code, u64 input1)
227 {
228 u64 hv_status, control = (u64)code | HV_HYPERCALL_FAST_BIT;
229
230 #ifdef CONFIG_X86_64
231 {
232 __asm__ __volatile__(CALL_NOSPEC
233 : "=a" (hv_status), ASM_CALL_CONSTRAINT,
234 "+c" (control), "+d" (input1)
235 : THUNK_TARGET(hv_hypercall_pg)
236 : "cc", "r8", "r9", "r10", "r11");
237 }
238 #else
239 {
240 u32 input1_hi = upper_32_bits(input1);
241 u32 input1_lo = lower_32_bits(input1);
242
243 __asm__ __volatile__ (CALL_NOSPEC
244 : "=A"(hv_status),
245 "+c"(input1_lo),
246 ASM_CALL_CONSTRAINT
247 : "A" (control),
248 "b" (input1_hi),
249 THUNK_TARGET(hv_hypercall_pg)
250 : "cc", "edi", "esi");
251 }
252 #endif
253 return hv_status;
254 }
255
256 /*
257 * Rep hypercalls. Callers of this functions are supposed to ensure that
258 * rep_count and varhead_size comply with Hyper-V hypercall definition.
259 */
hv_do_rep_hypercall(u16 code,u16 rep_count,u16 varhead_size,void * input,void * output)260 static inline u64 hv_do_rep_hypercall(u16 code, u16 rep_count, u16 varhead_size,
261 void *input, void *output)
262 {
263 u64 control = code;
264 u64 status;
265 u16 rep_comp;
266
267 control |= (u64)varhead_size << HV_HYPERCALL_VARHEAD_OFFSET;
268 control |= (u64)rep_count << HV_HYPERCALL_REP_COMP_OFFSET;
269
270 do {
271 status = hv_do_hypercall(control, input, output);
272 if ((status & HV_HYPERCALL_RESULT_MASK) != HV_STATUS_SUCCESS)
273 return status;
274
275 /* Bits 32-43 of status have 'Reps completed' data. */
276 rep_comp = (status & HV_HYPERCALL_REP_COMP_MASK) >>
277 HV_HYPERCALL_REP_COMP_OFFSET;
278
279 control &= ~HV_HYPERCALL_REP_START_MASK;
280 control |= (u64)rep_comp << HV_HYPERCALL_REP_START_OFFSET;
281
282 touch_nmi_watchdog();
283 } while (rep_comp < rep_count);
284
285 return status;
286 }
287
288 /*
289 * Hypervisor's notion of virtual processor ID is different from
290 * Linux' notion of CPU ID. This information can only be retrieved
291 * in the context of the calling CPU. Setup a map for easy access
292 * to this information.
293 */
294 extern u32 *hv_vp_index;
295 extern u32 hv_max_vp_index;
296
297 /**
298 * hv_cpu_number_to_vp_number() - Map CPU to VP.
299 * @cpu_number: CPU number in Linux terms
300 *
301 * This function returns the mapping between the Linux processor
302 * number and the hypervisor's virtual processor number, useful
303 * in making hypercalls and such that talk about specific
304 * processors.
305 *
306 * Return: Virtual processor number in Hyper-V terms
307 */
hv_cpu_number_to_vp_number(int cpu_number)308 static inline int hv_cpu_number_to_vp_number(int cpu_number)
309 {
310 return hv_vp_index[cpu_number];
311 }
312
313 void hyperv_init(void);
314 void hyperv_setup_mmu_ops(void);
315 void hyper_alloc_mmu(void);
316 void hyperv_report_panic(struct pt_regs *regs);
317 bool hv_is_hypercall_page_setup(void);
318 void hyperv_cleanup(void);
319 #else /* CONFIG_HYPERV */
hyperv_init(void)320 static inline void hyperv_init(void) {}
hv_is_hypercall_page_setup(void)321 static inline bool hv_is_hypercall_page_setup(void) { return false; }
hyperv_cleanup(void)322 static inline void hyperv_cleanup(void) {}
hyperv_setup_mmu_ops(void)323 static inline void hyperv_setup_mmu_ops(void) {}
324 #endif /* CONFIG_HYPERV */
325
326 #ifdef CONFIG_HYPERV_TSCPAGE
327 struct ms_hyperv_tsc_page *hv_get_tsc_page(void);
hv_read_tsc_page(const struct ms_hyperv_tsc_page * tsc_pg)328 static inline u64 hv_read_tsc_page(const struct ms_hyperv_tsc_page *tsc_pg)
329 {
330 u64 scale, offset, cur_tsc;
331 u32 sequence;
332
333 /*
334 * The protocol for reading Hyper-V TSC page is specified in Hypervisor
335 * Top-Level Functional Specification ver. 3.0 and above. To get the
336 * reference time we must do the following:
337 * - READ ReferenceTscSequence
338 * A special '0' value indicates the time source is unreliable and we
339 * need to use something else. The currently published specification
340 * versions (up to 4.0b) contain a mistake and wrongly claim '-1'
341 * instead of '0' as the special value, see commit c35b82ef0294.
342 * - ReferenceTime =
343 * ((RDTSC() * ReferenceTscScale) >> 64) + ReferenceTscOffset
344 * - READ ReferenceTscSequence again. In case its value has changed
345 * since our first reading we need to discard ReferenceTime and repeat
346 * the whole sequence as the hypervisor was updating the page in
347 * between.
348 */
349 do {
350 sequence = READ_ONCE(tsc_pg->tsc_sequence);
351 if (!sequence)
352 return U64_MAX;
353 /*
354 * Make sure we read sequence before we read other values from
355 * TSC page.
356 */
357 smp_rmb();
358
359 scale = READ_ONCE(tsc_pg->tsc_scale);
360 offset = READ_ONCE(tsc_pg->tsc_offset);
361 cur_tsc = rdtsc_ordered();
362
363 /*
364 * Make sure we read sequence after we read all other values
365 * from TSC page.
366 */
367 smp_rmb();
368
369 } while (READ_ONCE(tsc_pg->tsc_sequence) != sequence);
370
371 return mul_u64_u64_shr(cur_tsc, scale, 64) + offset;
372 }
373
374 #else
hv_get_tsc_page(void)375 static inline struct ms_hyperv_tsc_page *hv_get_tsc_page(void)
376 {
377 return NULL;
378 }
379 #endif
380 #endif
381