xref: /linux/arch/x86/include/asm/mshyperv.h (revision 43986798fd504c8cd846e3c65bcf732e3f971938)
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-tlfs.h>
10 #include <asm/nospec-branch.h>
11 
12 struct ms_hyperv_info {
13 	u32 features;
14 	u32 misc_features;
15 	u32 hints;
16 	u32 nested_features;
17 	u32 max_vp_index;
18 	u32 max_lp_index;
19 };
20 
21 extern struct ms_hyperv_info ms_hyperv;
22 
23 
24 /*
25  * Generate the guest ID.
26  */
27 
28 static inline  __u64 generate_guest_id(__u64 d_info1, __u64 kernel_version,
29 				       __u64 d_info2)
30 {
31 	__u64 guest_id = 0;
32 
33 	guest_id = (((__u64)HV_LINUX_VENDOR_ID) << 48);
34 	guest_id |= (d_info1 << 48);
35 	guest_id |= (kernel_version << 16);
36 	guest_id |= d_info2;
37 
38 	return guest_id;
39 }
40 
41 
42 /* Free the message slot and signal end-of-message if required */
43 static inline void vmbus_signal_eom(struct hv_message *msg, u32 old_msg_type)
44 {
45 	/*
46 	 * On crash we're reading some other CPU's message page and we need
47 	 * to be careful: this other CPU may already had cleared the header
48 	 * and the host may already had delivered some other message there.
49 	 * In case we blindly write msg->header.message_type we're going
50 	 * to lose it. We can still lose a message of the same type but
51 	 * we count on the fact that there can only be one
52 	 * CHANNELMSG_UNLOAD_RESPONSE and we don't care about other messages
53 	 * on crash.
54 	 */
55 	if (cmpxchg(&msg->header.message_type, old_msg_type,
56 		    HVMSG_NONE) != old_msg_type)
57 		return;
58 
59 	/*
60 	 * Make sure the write to MessageType (ie set to
61 	 * HVMSG_NONE) happens before we read the
62 	 * MessagePending and EOMing. Otherwise, the EOMing
63 	 * will not deliver any more messages since there is
64 	 * no empty slot
65 	 */
66 	mb();
67 
68 	if (msg->header.message_flags.msg_pending) {
69 		/*
70 		 * This will cause message queue rescan to
71 		 * possibly deliver another msg from the
72 		 * hypervisor
73 		 */
74 		wrmsrl(HV_X64_MSR_EOM, 0);
75 	}
76 }
77 
78 #define hv_init_timer(timer, tick) \
79 	wrmsrl(HV_X64_MSR_STIMER0_COUNT + (2*timer), tick)
80 #define hv_init_timer_config(timer, val) \
81 	wrmsrl(HV_X64_MSR_STIMER0_CONFIG + (2*timer), val)
82 
83 #define hv_get_simp(val) rdmsrl(HV_X64_MSR_SIMP, val)
84 #define hv_set_simp(val) wrmsrl(HV_X64_MSR_SIMP, val)
85 
86 #define hv_get_siefp(val) rdmsrl(HV_X64_MSR_SIEFP, val)
87 #define hv_set_siefp(val) wrmsrl(HV_X64_MSR_SIEFP, val)
88 
89 #define hv_get_synic_state(val) rdmsrl(HV_X64_MSR_SCONTROL, val)
90 #define hv_set_synic_state(val) wrmsrl(HV_X64_MSR_SCONTROL, val)
91 
92 #define hv_get_vp_index(index) rdmsrl(HV_X64_MSR_VP_INDEX, index)
93 
94 #define hv_get_synint_state(int_num, val) \
95 	rdmsrl(HV_X64_MSR_SINT0 + int_num, val)
96 #define hv_set_synint_state(int_num, val) \
97 	wrmsrl(HV_X64_MSR_SINT0 + int_num, val)
98 
99 void hyperv_callback_vector(void);
100 void hyperv_reenlightenment_vector(void);
101 #ifdef CONFIG_TRACING
102 #define trace_hyperv_callback_vector hyperv_callback_vector
103 #endif
104 void hyperv_vector_handler(struct pt_regs *regs);
105 void hv_setup_vmbus_irq(void (*handler)(void));
106 void hv_remove_vmbus_irq(void);
107 
108 void hv_setup_kexec_handler(void (*handler)(void));
109 void hv_remove_kexec_handler(void);
110 void hv_setup_crash_handler(void (*handler)(struct pt_regs *regs));
111 void hv_remove_crash_handler(void);
112 
113 /*
114  * Routines for stimer0 Direct Mode handling.
115  * On x86/x64, there are no percpu actions to take.
116  */
117 void hv_stimer0_vector_handler(struct pt_regs *regs);
118 void hv_stimer0_callback_vector(void);
119 int hv_setup_stimer0_irq(int *irq, int *vector, void (*handler)(void));
120 void hv_remove_stimer0_irq(int irq);
121 
122 static inline void hv_enable_stimer0_percpu_irq(int irq) {}
123 static inline void hv_disable_stimer0_percpu_irq(int irq) {}
124 
125 
126 #if IS_ENABLED(CONFIG_HYPERV)
127 extern struct clocksource *hyperv_cs;
128 extern void *hv_hypercall_pg;
129 extern void  __percpu  **hyperv_pcpu_input_arg;
130 
131 static inline u64 hv_do_hypercall(u64 control, void *input, void *output)
132 {
133 	u64 input_address = input ? virt_to_phys(input) : 0;
134 	u64 output_address = output ? virt_to_phys(output) : 0;
135 	u64 hv_status;
136 
137 #ifdef CONFIG_X86_64
138 	if (!hv_hypercall_pg)
139 		return U64_MAX;
140 
141 	__asm__ __volatile__("mov %4, %%r8\n"
142 			     CALL_NOSPEC
143 			     : "=a" (hv_status), ASM_CALL_CONSTRAINT,
144 			       "+c" (control), "+d" (input_address)
145 			     :  "r" (output_address),
146 				THUNK_TARGET(hv_hypercall_pg)
147 			     : "cc", "memory", "r8", "r9", "r10", "r11");
148 #else
149 	u32 input_address_hi = upper_32_bits(input_address);
150 	u32 input_address_lo = lower_32_bits(input_address);
151 	u32 output_address_hi = upper_32_bits(output_address);
152 	u32 output_address_lo = lower_32_bits(output_address);
153 
154 	if (!hv_hypercall_pg)
155 		return U64_MAX;
156 
157 	__asm__ __volatile__(CALL_NOSPEC
158 			     : "=A" (hv_status),
159 			       "+c" (input_address_lo), ASM_CALL_CONSTRAINT
160 			     : "A" (control),
161 			       "b" (input_address_hi),
162 			       "D"(output_address_hi), "S"(output_address_lo),
163 			       THUNK_TARGET(hv_hypercall_pg)
164 			     : "cc", "memory");
165 #endif /* !x86_64 */
166 	return hv_status;
167 }
168 
169 /* Fast hypercall with 8 bytes of input and no output */
170 static inline u64 hv_do_fast_hypercall8(u16 code, u64 input1)
171 {
172 	u64 hv_status, control = (u64)code | HV_HYPERCALL_FAST_BIT;
173 
174 #ifdef CONFIG_X86_64
175 	{
176 		__asm__ __volatile__(CALL_NOSPEC
177 				     : "=a" (hv_status), ASM_CALL_CONSTRAINT,
178 				       "+c" (control), "+d" (input1)
179 				     : THUNK_TARGET(hv_hypercall_pg)
180 				     : "cc", "r8", "r9", "r10", "r11");
181 	}
182 #else
183 	{
184 		u32 input1_hi = upper_32_bits(input1);
185 		u32 input1_lo = lower_32_bits(input1);
186 
187 		__asm__ __volatile__ (CALL_NOSPEC
188 				      : "=A"(hv_status),
189 					"+c"(input1_lo),
190 					ASM_CALL_CONSTRAINT
191 				      :	"A" (control),
192 					"b" (input1_hi),
193 					THUNK_TARGET(hv_hypercall_pg)
194 				      : "cc", "edi", "esi");
195 	}
196 #endif
197 		return hv_status;
198 }
199 
200 /*
201  * Rep hypercalls. Callers of this functions are supposed to ensure that
202  * rep_count and varhead_size comply with Hyper-V hypercall definition.
203  */
204 static inline u64 hv_do_rep_hypercall(u16 code, u16 rep_count, u16 varhead_size,
205 				      void *input, void *output)
206 {
207 	u64 control = code;
208 	u64 status;
209 	u16 rep_comp;
210 
211 	control |= (u64)varhead_size << HV_HYPERCALL_VARHEAD_OFFSET;
212 	control |= (u64)rep_count << HV_HYPERCALL_REP_COMP_OFFSET;
213 
214 	do {
215 		status = hv_do_hypercall(control, input, output);
216 		if ((status & HV_HYPERCALL_RESULT_MASK) != HV_STATUS_SUCCESS)
217 			return status;
218 
219 		/* Bits 32-43 of status have 'Reps completed' data. */
220 		rep_comp = (status & HV_HYPERCALL_REP_COMP_MASK) >>
221 			HV_HYPERCALL_REP_COMP_OFFSET;
222 
223 		control &= ~HV_HYPERCALL_REP_START_MASK;
224 		control |= (u64)rep_comp << HV_HYPERCALL_REP_START_OFFSET;
225 
226 		touch_nmi_watchdog();
227 	} while (rep_comp < rep_count);
228 
229 	return status;
230 }
231 
232 /*
233  * Hypervisor's notion of virtual processor ID is different from
234  * Linux' notion of CPU ID. This information can only be retrieved
235  * in the context of the calling CPU. Setup a map for easy access
236  * to this information.
237  */
238 extern u32 *hv_vp_index;
239 extern u32 hv_max_vp_index;
240 extern struct hv_vp_assist_page **hv_vp_assist_page;
241 
242 static inline struct hv_vp_assist_page *hv_get_vp_assist_page(unsigned int cpu)
243 {
244 	if (!hv_vp_assist_page)
245 		return NULL;
246 
247 	return hv_vp_assist_page[cpu];
248 }
249 
250 /**
251  * hv_cpu_number_to_vp_number() - Map CPU to VP.
252  * @cpu_number: CPU number in Linux terms
253  *
254  * This function returns the mapping between the Linux processor
255  * number and the hypervisor's virtual processor number, useful
256  * in making hypercalls and such that talk about specific
257  * processors.
258  *
259  * Return: Virtual processor number in Hyper-V terms
260  */
261 static inline int hv_cpu_number_to_vp_number(int cpu_number)
262 {
263 	return hv_vp_index[cpu_number];
264 }
265 
266 static inline int cpumask_to_vpset(struct hv_vpset *vpset,
267 				    const struct cpumask *cpus)
268 {
269 	int cpu, vcpu, vcpu_bank, vcpu_offset, nr_bank = 1;
270 
271 	/* valid_bank_mask can represent up to 64 banks */
272 	if (hv_max_vp_index / 64 >= 64)
273 		return 0;
274 
275 	/*
276 	 * Clear all banks up to the maximum possible bank as hv_tlb_flush_ex
277 	 * structs are not cleared between calls, we risk flushing unneeded
278 	 * vCPUs otherwise.
279 	 */
280 	for (vcpu_bank = 0; vcpu_bank <= hv_max_vp_index / 64; vcpu_bank++)
281 		vpset->bank_contents[vcpu_bank] = 0;
282 
283 	/*
284 	 * Some banks may end up being empty but this is acceptable.
285 	 */
286 	for_each_cpu(cpu, cpus) {
287 		vcpu = hv_cpu_number_to_vp_number(cpu);
288 		vcpu_bank = vcpu / 64;
289 		vcpu_offset = vcpu % 64;
290 		__set_bit(vcpu_offset, (unsigned long *)
291 			  &vpset->bank_contents[vcpu_bank]);
292 		if (vcpu_bank >= nr_bank)
293 			nr_bank = vcpu_bank + 1;
294 	}
295 	vpset->valid_bank_mask = GENMASK_ULL(nr_bank - 1, 0);
296 	return nr_bank;
297 }
298 
299 void __init hyperv_init(void);
300 void hyperv_setup_mmu_ops(void);
301 void hyperv_report_panic(struct pt_regs *regs, long err);
302 bool hv_is_hyperv_initialized(void);
303 void hyperv_cleanup(void);
304 
305 void hyperv_reenlightenment_intr(struct pt_regs *regs);
306 void set_hv_tscchange_cb(void (*cb)(void));
307 void clear_hv_tscchange_cb(void);
308 void hyperv_stop_tsc_emulation(void);
309 
310 #ifdef CONFIG_X86_64
311 void hv_apic_init(void);
312 #else
313 static inline void hv_apic_init(void) {}
314 #endif
315 
316 #else /* CONFIG_HYPERV */
317 static inline void hyperv_init(void) {}
318 static inline bool hv_is_hyperv_initialized(void) { return false; }
319 static inline void hyperv_cleanup(void) {}
320 static inline void hyperv_setup_mmu_ops(void) {}
321 static inline void set_hv_tscchange_cb(void (*cb)(void)) {}
322 static inline void clear_hv_tscchange_cb(void) {}
323 static inline void hyperv_stop_tsc_emulation(void) {};
324 static inline struct hv_vp_assist_page *hv_get_vp_assist_page(unsigned int cpu)
325 {
326 	return NULL;
327 }
328 #endif /* CONFIG_HYPERV */
329 
330 #ifdef CONFIG_HYPERV_TSCPAGE
331 struct ms_hyperv_tsc_page *hv_get_tsc_page(void);
332 static inline u64 hv_read_tsc_page_tsc(const struct ms_hyperv_tsc_page *tsc_pg,
333 				       u64 *cur_tsc)
334 {
335 	u64 scale, offset;
336 	u32 sequence;
337 
338 	/*
339 	 * The protocol for reading Hyper-V TSC page is specified in Hypervisor
340 	 * Top-Level Functional Specification ver. 3.0 and above. To get the
341 	 * reference time we must do the following:
342 	 * - READ ReferenceTscSequence
343 	 *   A special '0' value indicates the time source is unreliable and we
344 	 *   need to use something else. The currently published specification
345 	 *   versions (up to 4.0b) contain a mistake and wrongly claim '-1'
346 	 *   instead of '0' as the special value, see commit c35b82ef0294.
347 	 * - ReferenceTime =
348 	 *        ((RDTSC() * ReferenceTscScale) >> 64) + ReferenceTscOffset
349 	 * - READ ReferenceTscSequence again. In case its value has changed
350 	 *   since our first reading we need to discard ReferenceTime and repeat
351 	 *   the whole sequence as the hypervisor was updating the page in
352 	 *   between.
353 	 */
354 	do {
355 		sequence = READ_ONCE(tsc_pg->tsc_sequence);
356 		if (!sequence)
357 			return U64_MAX;
358 		/*
359 		 * Make sure we read sequence before we read other values from
360 		 * TSC page.
361 		 */
362 		smp_rmb();
363 
364 		scale = READ_ONCE(tsc_pg->tsc_scale);
365 		offset = READ_ONCE(tsc_pg->tsc_offset);
366 		*cur_tsc = rdtsc_ordered();
367 
368 		/*
369 		 * Make sure we read sequence after we read all other values
370 		 * from TSC page.
371 		 */
372 		smp_rmb();
373 
374 	} while (READ_ONCE(tsc_pg->tsc_sequence) != sequence);
375 
376 	return mul_u64_u64_shr(*cur_tsc, scale, 64) + offset;
377 }
378 
379 static inline u64 hv_read_tsc_page(const struct ms_hyperv_tsc_page *tsc_pg)
380 {
381 	u64 cur_tsc;
382 
383 	return hv_read_tsc_page_tsc(tsc_pg, &cur_tsc);
384 }
385 
386 #else
387 static inline struct ms_hyperv_tsc_page *hv_get_tsc_page(void)
388 {
389 	return NULL;
390 }
391 
392 static inline u64 hv_read_tsc_page_tsc(const struct ms_hyperv_tsc_page *tsc_pg,
393 				       u64 *cur_tsc)
394 {
395 	BUG();
396 	return U64_MAX;
397 }
398 #endif
399 #endif
400