xref: /linux/arch/x86/hyperv/hv_init.c (revision feb06d2690bb826fd33798a99ce5cff8d07b38f9)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * X86 specific Hyper-V initialization code.
4  *
5  * Copyright (C) 2016, Microsoft, Inc.
6  *
7  * Author : K. Y. Srinivasan <kys@microsoft.com>
8  */
9 
10 #define pr_fmt(fmt)  "Hyper-V: " fmt
11 
12 #include <linux/efi.h>
13 #include <linux/types.h>
14 #include <linux/bitfield.h>
15 #include <linux/io.h>
16 #include <asm/apic.h>
17 #include <asm/desc.h>
18 #include <asm/e820/api.h>
19 #include <asm/sev.h>
20 #include <asm/hypervisor.h>
21 #include <hyperv/hvhdk.h>
22 #include <asm/mshyperv.h>
23 #include <asm/msr.h>
24 #include <asm/idtentry.h>
25 #include <asm/set_memory.h>
26 #include <linux/kexec.h>
27 #include <linux/version.h>
28 #include <linux/vmalloc.h>
29 #include <linux/mm.h>
30 #include <linux/slab.h>
31 #include <linux/kernel.h>
32 #include <linux/cpuhotplug.h>
33 #include <linux/syscore_ops.h>
34 #include <clocksource/hyperv_timer.h>
35 #include <linux/highmem.h>
36 #include <linux/export.h>
37 
38 void *hv_hypercall_pg;
39 
40 #ifdef CONFIG_X86_64
__hv_hyperfail(u64 control,u64 param1,u64 param2)41 static u64 __hv_hyperfail(u64 control, u64 param1, u64 param2)
42 {
43 	return U64_MAX;
44 }
45 
46 DEFINE_STATIC_CALL(__hv_hypercall, __hv_hyperfail);
47 
hv_std_hypercall(u64 control,u64 param1,u64 param2)48 u64 hv_std_hypercall(u64 control, u64 param1, u64 param2)
49 {
50 	u64 hv_status;
51 
52 	register u64 __r8 asm("r8") = param2;
53 	asm volatile ("call " STATIC_CALL_TRAMP_STR(__hv_hypercall)
54 		      : "=a" (hv_status), ASM_CALL_CONSTRAINT,
55 		        "+c" (control), "+d" (param1), "+r" (__r8)
56 		      : : "cc", "memory", "r9", "r10", "r11");
57 
58 	return hv_status;
59 }
60 
61 typedef u64 (*hv_hypercall_f)(u64 control, u64 param1, u64 param2);
62 
hv_set_hypercall_pg(void * ptr)63 static inline void hv_set_hypercall_pg(void *ptr)
64 {
65 	hv_hypercall_pg = ptr;
66 
67 	if (!ptr)
68 		ptr = &__hv_hyperfail;
69 	static_call_update(__hv_hypercall, (hv_hypercall_f)ptr);
70 }
71 #else
hv_set_hypercall_pg(void * ptr)72 static inline void hv_set_hypercall_pg(void *ptr)
73 {
74 	hv_hypercall_pg = ptr;
75 }
76 EXPORT_SYMBOL_GPL(hv_hypercall_pg);
77 #endif
78 
79 union hv_ghcb * __percpu *hv_ghcb_pg;
80 
81 /* Storage to save the hypercall page temporarily for hibernation */
82 static void *hv_hypercall_pg_saved;
83 
84 struct hv_vp_assist_page **hv_vp_assist_page;
85 EXPORT_SYMBOL_GPL(hv_vp_assist_page);
86 
hyperv_init_ghcb(void)87 static int hyperv_init_ghcb(void)
88 {
89 	u64 ghcb_gpa;
90 	void *ghcb_va;
91 	void **ghcb_base;
92 
93 	if (!ms_hyperv.paravisor_present || !hv_isolation_type_snp())
94 		return 0;
95 
96 	if (!hv_ghcb_pg)
97 		return -EINVAL;
98 
99 	/*
100 	 * GHCB page is allocated by paravisor. The address
101 	 * returned by MSR_AMD64_SEV_ES_GHCB is above shared
102 	 * memory boundary and map it here.
103 	 */
104 	rdmsrq(MSR_AMD64_SEV_ES_GHCB, ghcb_gpa);
105 
106 	/* Mask out vTOM bit. ioremap_cache() maps decrypted */
107 	ghcb_gpa &= ~ms_hyperv.shared_gpa_boundary;
108 	ghcb_va = (void *)ioremap_cache(ghcb_gpa, HV_HYP_PAGE_SIZE);
109 	if (!ghcb_va)
110 		return -ENOMEM;
111 
112 	ghcb_base = (void **)this_cpu_ptr(hv_ghcb_pg);
113 	*ghcb_base = ghcb_va;
114 
115 	return 0;
116 }
117 
hv_cpu_init(unsigned int cpu)118 static int hv_cpu_init(unsigned int cpu)
119 {
120 	union hv_vp_assist_msr_contents msr = { 0 };
121 	struct hv_vp_assist_page **hvp;
122 	int ret;
123 
124 	ret = hv_common_cpu_init(cpu);
125 	if (ret)
126 		return ret;
127 
128 	if (!hv_vp_assist_page)
129 		return 0;
130 
131 	hvp = &hv_vp_assist_page[cpu];
132 	if (hv_root_partition()) {
133 		/*
134 		 * For root partition we get the hypervisor provided VP assist
135 		 * page, instead of allocating a new page.
136 		 */
137 		rdmsrq(HV_X64_MSR_VP_ASSIST_PAGE, msr.as_uint64);
138 		*hvp = memremap(msr.pfn << HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT,
139 				PAGE_SIZE, MEMREMAP_WB);
140 	} else {
141 		/*
142 		 * The VP assist page is an "overlay" page (see Hyper-V TLFS's
143 		 * Section 5.2.1 "GPA Overlay Pages"). Here it must be zeroed
144 		 * out to make sure we always write the EOI MSR in
145 		 * hv_apic_eoi_write() *after* the EOI optimization is disabled
146 		 * in hv_cpu_die(), otherwise a CPU may not be stopped in the
147 		 * case of CPU offlining and the VM will hang.
148 		 */
149 		if (!*hvp) {
150 			*hvp = __vmalloc(PAGE_SIZE, GFP_KERNEL | __GFP_ZERO);
151 
152 			/*
153 			 * Hyper-V should never specify a VM that is a Confidential
154 			 * VM and also running in the root partition. Root partition
155 			 * is blocked to run in Confidential VM. So only decrypt assist
156 			 * page in non-root partition here.
157 			 */
158 			if (*hvp && !ms_hyperv.paravisor_present && hv_isolation_type_snp()) {
159 				WARN_ON_ONCE(set_memory_decrypted((unsigned long)(*hvp), 1));
160 				memset(*hvp, 0, PAGE_SIZE);
161 			}
162 		}
163 
164 		if (*hvp)
165 			msr.pfn = vmalloc_to_pfn(*hvp);
166 
167 	}
168 	if (!WARN_ON(!(*hvp))) {
169 		msr.enable = 1;
170 		wrmsrq(HV_X64_MSR_VP_ASSIST_PAGE, msr.as_uint64);
171 	}
172 
173 	/* Allow Hyper-V stimer vector to be injected from Hypervisor. */
174 	if (ms_hyperv.misc_features & HV_STIMER_DIRECT_MODE_AVAILABLE)
175 		apic_update_vector(cpu, HYPERV_STIMER0_VECTOR, true);
176 
177 	return hyperv_init_ghcb();
178 }
179 
180 static void (*hv_reenlightenment_cb)(void);
181 
hv_reenlightenment_notify(struct work_struct * dummy)182 static void hv_reenlightenment_notify(struct work_struct *dummy)
183 {
184 	struct hv_tsc_emulation_status emu_status;
185 
186 	rdmsrq(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
187 
188 	/* Don't issue the callback if TSC accesses are not emulated */
189 	if (hv_reenlightenment_cb && emu_status.inprogress)
190 		hv_reenlightenment_cb();
191 }
192 static DECLARE_DELAYED_WORK(hv_reenlightenment_work, hv_reenlightenment_notify);
193 
hyperv_stop_tsc_emulation(void)194 void hyperv_stop_tsc_emulation(void)
195 {
196 	u64 freq;
197 	struct hv_tsc_emulation_status emu_status;
198 
199 	rdmsrq(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
200 	emu_status.inprogress = 0;
201 	wrmsrq(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
202 
203 	rdmsrq(HV_X64_MSR_TSC_FREQUENCY, freq);
204 	tsc_khz = div64_u64(freq, 1000);
205 }
206 EXPORT_SYMBOL_GPL(hyperv_stop_tsc_emulation);
207 
hv_reenlightenment_available(void)208 static inline bool hv_reenlightenment_available(void)
209 {
210 	/*
211 	 * Check for required features and privileges to make TSC frequency
212 	 * change notifications work.
213 	 */
214 	return ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
215 		ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE &&
216 		ms_hyperv.features & HV_ACCESS_REENLIGHTENMENT;
217 }
218 
DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_reenlightenment)219 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_reenlightenment)
220 {
221 	apic_eoi();
222 	inc_irq_stat(irq_hv_reenlightenment_count);
223 	schedule_delayed_work(&hv_reenlightenment_work, HZ/10);
224 }
225 
set_hv_tscchange_cb(void (* cb)(void))226 void set_hv_tscchange_cb(void (*cb)(void))
227 {
228 	struct hv_reenlightenment_control re_ctrl = {
229 		.vector = HYPERV_REENLIGHTENMENT_VECTOR,
230 		.enabled = 1,
231 	};
232 	struct hv_tsc_emulation_control emu_ctrl = {.enabled = 1};
233 
234 	if (!hv_reenlightenment_available()) {
235 		pr_warn("reenlightenment support is unavailable\n");
236 		return;
237 	}
238 
239 	if (!hv_vp_index)
240 		return;
241 
242 	hv_reenlightenment_cb = cb;
243 
244 	/* Make sure callback is registered before we write to MSRs */
245 	wmb();
246 
247 	re_ctrl.target_vp = hv_vp_index[get_cpu()];
248 
249 	wrmsrq(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
250 	wrmsrq(HV_X64_MSR_TSC_EMULATION_CONTROL, *((u64 *)&emu_ctrl));
251 
252 	put_cpu();
253 }
254 EXPORT_SYMBOL_GPL(set_hv_tscchange_cb);
255 
clear_hv_tscchange_cb(void)256 void clear_hv_tscchange_cb(void)
257 {
258 	struct hv_reenlightenment_control re_ctrl;
259 
260 	if (!hv_reenlightenment_available())
261 		return;
262 
263 	rdmsrq(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
264 	re_ctrl.enabled = 0;
265 	wrmsrq(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
266 
267 	hv_reenlightenment_cb = NULL;
268 }
269 EXPORT_SYMBOL_GPL(clear_hv_tscchange_cb);
270 
hv_cpu_die(unsigned int cpu)271 static int hv_cpu_die(unsigned int cpu)
272 {
273 	struct hv_reenlightenment_control re_ctrl;
274 	unsigned int new_cpu;
275 	void **ghcb_va;
276 
277 	if (hv_ghcb_pg) {
278 		ghcb_va = (void **)this_cpu_ptr(hv_ghcb_pg);
279 		if (*ghcb_va)
280 			iounmap(*ghcb_va);
281 		*ghcb_va = NULL;
282 	}
283 
284 	if (ms_hyperv.misc_features & HV_STIMER_DIRECT_MODE_AVAILABLE)
285 		apic_update_vector(cpu, HYPERV_STIMER0_VECTOR, false);
286 
287 	hv_common_cpu_die(cpu);
288 
289 	if (hv_vp_assist_page && hv_vp_assist_page[cpu]) {
290 		union hv_vp_assist_msr_contents msr = { 0 };
291 		if (hv_root_partition()) {
292 			/*
293 			 * For root partition the VP assist page is mapped to
294 			 * hypervisor provided page, and thus we unmap the
295 			 * page here and nullify it, so that in future we have
296 			 * correct page address mapped in hv_cpu_init.
297 			 */
298 			memunmap(hv_vp_assist_page[cpu]);
299 			hv_vp_assist_page[cpu] = NULL;
300 			rdmsrq(HV_X64_MSR_VP_ASSIST_PAGE, msr.as_uint64);
301 			msr.enable = 0;
302 		}
303 		wrmsrq(HV_X64_MSR_VP_ASSIST_PAGE, msr.as_uint64);
304 	}
305 
306 	if (hv_reenlightenment_cb == NULL)
307 		return 0;
308 
309 	rdmsrq(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
310 	if (re_ctrl.target_vp == hv_vp_index[cpu]) {
311 		/*
312 		 * Reassign reenlightenment notifications to some other online
313 		 * CPU or just disable the feature if there are no online CPUs
314 		 * left (happens on hibernation).
315 		 */
316 		new_cpu = cpumask_any_but(cpu_online_mask, cpu);
317 
318 		if (new_cpu < nr_cpu_ids)
319 			re_ctrl.target_vp = hv_vp_index[new_cpu];
320 		else
321 			re_ctrl.enabled = 0;
322 
323 		wrmsrq(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
324 	}
325 
326 	return 0;
327 }
328 
hv_pci_init(void)329 static int __init hv_pci_init(void)
330 {
331 	bool gen2vm = efi_enabled(EFI_BOOT);
332 
333 	/*
334 	 * A Generation-2 VM doesn't support legacy PCI/PCIe, so both
335 	 * raw_pci_ops and raw_pci_ext_ops are NULL, and pci_subsys_init() ->
336 	 * pcibios_init() doesn't call pcibios_resource_survey() ->
337 	 * e820__reserve_resources_late(); as a result, any emulated persistent
338 	 * memory of E820_TYPE_PRAM (12) via the kernel parameter
339 	 * memmap=nn[KMG]!ss is not added into iomem_resource and hence can't be
340 	 * detected by register_e820_pmem(). Fix this by directly calling
341 	 * e820__reserve_resources_late() here: e820__reserve_resources_late()
342 	 * depends on e820__reserve_resources(), which has been called earlier
343 	 * from setup_arch(). Note: e820__reserve_resources_late() also adds
344 	 * any memory of E820_TYPE_PMEM (7) into iomem_resource, and
345 	 * acpi_nfit_register_region() -> acpi_nfit_insert_resource() ->
346 	 * region_intersects() returns REGION_INTERSECTS, so the memory of
347 	 * E820_TYPE_PMEM won't get added twice.
348 	 *
349 	 * We return 0 here so that pci_arch_init() won't print the warning:
350 	 * "PCI: Fatal: No config space access function found"
351 	 */
352 	if (gen2vm) {
353 		e820__reserve_resources_late();
354 		return 0;
355 	}
356 
357 	/* For Generation-1 VM, we'll proceed in pci_arch_init().  */
358 	return 1;
359 }
360 
hv_suspend(void * data)361 static int hv_suspend(void *data)
362 {
363 	union hv_x64_msr_hypercall_contents hypercall_msr;
364 	int ret;
365 
366 	if (hv_root_partition())
367 		return -EPERM;
368 
369 	/*
370 	 * Reset the hypercall page as it is going to be invalidated
371 	 * across hibernation. Setting hv_hypercall_pg to NULL ensures
372 	 * that any subsequent hypercall operation fails safely instead of
373 	 * crashing due to an access of an invalid page. The hypercall page
374 	 * pointer is restored on resume.
375 	 */
376 	hv_hypercall_pg_saved = hv_hypercall_pg;
377 	hv_set_hypercall_pg(NULL);
378 
379 	/* Disable the hypercall page in the hypervisor */
380 	rdmsrq(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
381 	hypercall_msr.enable = 0;
382 	wrmsrq(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
383 
384 	ret = hv_cpu_die(0);
385 	return ret;
386 }
387 
hv_resume(void * data)388 static void hv_resume(void *data)
389 {
390 	union hv_x64_msr_hypercall_contents hypercall_msr;
391 	int ret;
392 
393 	ret = hv_cpu_init(0);
394 	WARN_ON(ret);
395 
396 	/* Re-enable the hypercall page */
397 	rdmsrq(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
398 	hypercall_msr.enable = 1;
399 	hypercall_msr.guest_physical_address =
400 		vmalloc_to_pfn(hv_hypercall_pg_saved);
401 	wrmsrq(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
402 
403 	hv_set_hypercall_pg(hv_hypercall_pg_saved);
404 	hv_hypercall_pg_saved = NULL;
405 
406 	/*
407 	 * Reenlightenment notifications are disabled by hv_cpu_die(0),
408 	 * reenable them here if hv_reenlightenment_cb was previously set.
409 	 */
410 	if (hv_reenlightenment_cb)
411 		set_hv_tscchange_cb(hv_reenlightenment_cb);
412 }
413 
414 /* Note: when the ops are called, only CPU0 is online and IRQs are disabled. */
415 static const struct syscore_ops hv_syscore_ops = {
416 	.suspend	= hv_suspend,
417 	.resume		= hv_resume,
418 };
419 
420 static struct syscore hv_syscore = {
421 	.ops = &hv_syscore_ops,
422 };
423 
424 static void (* __initdata old_setup_percpu_clockev)(void);
425 
hv_stimer_setup_percpu_clockev(void)426 static void __init hv_stimer_setup_percpu_clockev(void)
427 {
428 	/*
429 	 * Ignore any errors in setting up stimer clockevents
430 	 * as we can run with the LAPIC timer as a fallback.
431 	 */
432 	(void)hv_stimer_alloc(false);
433 
434 	/*
435 	 * Still register the LAPIC timer, because the direct-mode STIMER is
436 	 * not supported by old versions of Hyper-V. This also allows users
437 	 * to switch to LAPIC timer via /sys, if they want to.
438 	 */
439 	if (old_setup_percpu_clockev)
440 		old_setup_percpu_clockev();
441 }
442 
443 /*
444  * This function is to be invoked early in the boot sequence after the
445  * hypervisor has been detected.
446  *
447  * 1. Setup the hypercall page.
448  * 2. Register Hyper-V specific clocksource.
449  * 3. Setup Hyper-V specific APIC entry points.
450  */
hyperv_init(void)451 void __init hyperv_init(void)
452 {
453 	u64 guest_id;
454 	union hv_x64_msr_hypercall_contents hypercall_msr;
455 	int cpuhp;
456 
457 	if (x86_hyper_type != X86_HYPER_MS_HYPERV)
458 		return;
459 
460 	if (hv_common_init())
461 		return;
462 
463 	/*
464 	 * The VP assist page is useless to a TDX guest: the only use we
465 	 * would have for it is lazy EOI, which can not be used with TDX.
466 	 */
467 	if (hv_isolation_type_tdx())
468 		hv_vp_assist_page = NULL;
469 	else
470 		hv_vp_assist_page = kcalloc(nr_cpu_ids,
471 					    sizeof(*hv_vp_assist_page),
472 					    GFP_KERNEL);
473 	if (!hv_vp_assist_page) {
474 		ms_hyperv.hints &= ~HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
475 
476 		if (!hv_isolation_type_tdx())
477 			goto common_free;
478 	}
479 
480 	if (ms_hyperv.paravisor_present && hv_isolation_type_snp()) {
481 		/* Negotiate GHCB Version. */
482 		if (!hv_ghcb_negotiate_protocol())
483 			hv_ghcb_terminate(SEV_TERM_SET_GEN,
484 					  GHCB_SEV_ES_PROT_UNSUPPORTED);
485 
486 		hv_ghcb_pg = alloc_percpu(union hv_ghcb *);
487 		if (!hv_ghcb_pg)
488 			goto free_vp_assist_page;
489 	}
490 
491 	cpuhp = cpuhp_setup_state(CPUHP_AP_HYPERV_ONLINE, "x86/hyperv_init:online",
492 				  hv_cpu_init, hv_cpu_die);
493 	if (cpuhp < 0)
494 		goto free_ghcb_page;
495 
496 	/*
497 	 * Setup the hypercall page and enable hypercalls.
498 	 * 1. Register the guest ID
499 	 * 2. Enable the hypercall and register the hypercall page
500 	 *
501 	 * A TDX VM with no paravisor only uses TDX GHCI rather than hv_hypercall_pg:
502 	 * when the hypercall input is a page, such a VM must pass a decrypted
503 	 * page to Hyper-V, e.g. hv_post_message() uses the per-CPU page
504 	 * hyperv_pcpu_input_arg, which is decrypted if no paravisor is present.
505 	 *
506 	 * A TDX VM with the paravisor uses hv_hypercall_pg for most hypercalls,
507 	 * which are handled by the paravisor and the VM must use an encrypted
508 	 * input page: in such a VM, the hyperv_pcpu_input_arg is encrypted and
509 	 * used in the hypercalls, e.g. see hv_mark_gpa_visibility() and
510 	 * hv_arch_irq_unmask(). Such a VM uses TDX GHCI for two hypercalls:
511 	 * 1. HVCALL_SIGNAL_EVENT: see vmbus_set_event() and _hv_do_fast_hypercall8().
512 	 * 2. HVCALL_POST_MESSAGE: the input page must be a decrypted page, i.e.
513 	 * hv_post_message() in such a VM can't use the encrypted hyperv_pcpu_input_arg;
514 	 * instead, hv_post_message() uses the post_msg_page, which is decrypted
515 	 * in such a VM and is only used in such a VM.
516 	 */
517 	guest_id = hv_generate_guest_id(LINUX_VERSION_CODE);
518 	wrmsrq(HV_X64_MSR_GUEST_OS_ID, guest_id);
519 
520 	/* With the paravisor, the VM must also write the ID via GHCB/GHCI */
521 	hv_ivm_msr_write(HV_X64_MSR_GUEST_OS_ID, guest_id);
522 
523 	/* A TDX VM with no paravisor only uses TDX GHCI rather than hv_hypercall_pg */
524 	if (hv_isolation_type_tdx() && !ms_hyperv.paravisor_present)
525 		goto skip_hypercall_pg_init;
526 
527 	hv_hypercall_pg = __vmalloc_node_range(PAGE_SIZE, 1, MODULES_VADDR,
528 			MODULES_END, GFP_KERNEL, PAGE_KERNEL_ROX,
529 			VM_FLUSH_RESET_PERMS, NUMA_NO_NODE,
530 			__builtin_return_address(0));
531 	if (hv_hypercall_pg == NULL)
532 		goto clean_guest_os_id;
533 
534 	rdmsrq(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
535 	hypercall_msr.enable = 1;
536 
537 	if (hv_root_partition()) {
538 		struct page *pg;
539 		void *src;
540 
541 		/*
542 		 * For the root partition, the hypervisor will set up its
543 		 * hypercall page. The hypervisor guarantees it will not show
544 		 * up in the root's address space. The root can't change the
545 		 * location of the hypercall page.
546 		 *
547 		 * Order is important here. We must enable the hypercall page
548 		 * so it is populated with code, then copy the code to an
549 		 * executable page.
550 		 */
551 		wrmsrq(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
552 
553 		pg = vmalloc_to_page(hv_hypercall_pg);
554 		src = memremap(hypercall_msr.guest_physical_address << PAGE_SHIFT, PAGE_SIZE,
555 				MEMREMAP_WB);
556 		BUG_ON(!src);
557 		memcpy_to_page(pg, 0, src, HV_HYP_PAGE_SIZE);
558 		memunmap(src);
559 
560 		hv_remap_tsc_clocksource();
561 		hv_root_crash_init();
562 		hv_sleep_notifiers_register();
563 	} else {
564 		hypercall_msr.guest_physical_address = vmalloc_to_pfn(hv_hypercall_pg);
565 		wrmsrq(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
566 	}
567 
568 	hv_set_hypercall_pg(hv_hypercall_pg);
569 
570 skip_hypercall_pg_init:
571 	/*
572 	 * hyperv_init() is called before LAPIC is initialized: see
573 	 * apic_intr_mode_init() -> x86_platform.apic_post_init() and
574 	 * apic_bsp_setup() -> setup_local_APIC(). The direct-mode STIMER
575 	 * depends on LAPIC, so hv_stimer_alloc() should be called from
576 	 * x86_init.timers.setup_percpu_clockev.
577 	 */
578 	old_setup_percpu_clockev = x86_init.timers.setup_percpu_clockev;
579 	x86_init.timers.setup_percpu_clockev = hv_stimer_setup_percpu_clockev;
580 
581 	hv_apic_init();
582 
583 	x86_init.pci.arch_init = hv_pci_init;
584 
585 	register_syscore(&hv_syscore);
586 
587 	if (ms_hyperv.priv_high & HV_ACCESS_PARTITION_ID)
588 		hv_get_partition_id();
589 
590 #ifdef CONFIG_PCI_MSI
591 	/*
592 	 * If we're running as root, we want to create our own PCI MSI domain.
593 	 * We can't set this in hv_pci_init because that would be too late.
594 	 */
595 	if (hv_root_partition())
596 		x86_init.irqs.create_pci_msi_domain = hv_create_pci_msi_domain;
597 #endif
598 
599 	/* Query the VMs extended capability once, so that it can be cached. */
600 	hv_query_ext_cap(0);
601 
602 	/* Find the VTL */
603 	ms_hyperv.vtl = get_vtl();
604 
605 	if (ms_hyperv.vtl > 0) /* non default VTL */
606 		hv_vtl_early_init();
607 
608 	return;
609 
610 clean_guest_os_id:
611 	wrmsrq(HV_X64_MSR_GUEST_OS_ID, 0);
612 	hv_ivm_msr_write(HV_X64_MSR_GUEST_OS_ID, 0);
613 	cpuhp_remove_state(CPUHP_AP_HYPERV_ONLINE);
614 free_ghcb_page:
615 	free_percpu(hv_ghcb_pg);
616 free_vp_assist_page:
617 	kfree(hv_vp_assist_page);
618 	hv_vp_assist_page = NULL;
619 common_free:
620 	hv_common_free();
621 }
622 
623 /*
624  * This routine is called before kexec/kdump, it does the required cleanup.
625  */
hyperv_cleanup(void)626 void hyperv_cleanup(void)
627 {
628 	union hv_x64_msr_hypercall_contents hypercall_msr;
629 	union hv_reference_tsc_msr tsc_msr;
630 
631 	/* Reset our OS id */
632 	wrmsrq(HV_X64_MSR_GUEST_OS_ID, 0);
633 	hv_ivm_msr_write(HV_X64_MSR_GUEST_OS_ID, 0);
634 
635 	/*
636 	 * Reset hypercall page reference before reset the page,
637 	 * let hypercall operations fail safely rather than
638 	 * panic the kernel for using invalid hypercall page
639 	 */
640 	hv_hypercall_pg = NULL;
641 
642 	/* Reset the hypercall page */
643 	hypercall_msr.as_uint64 = hv_get_msr(HV_X64_MSR_HYPERCALL);
644 	hypercall_msr.enable = 0;
645 	hv_set_msr(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
646 
647 	/* Reset the TSC page */
648 	tsc_msr.as_uint64 = hv_get_msr(HV_X64_MSR_REFERENCE_TSC);
649 	tsc_msr.enable = 0;
650 	hv_set_msr(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
651 }
652 
hyperv_report_panic(struct pt_regs * regs,long err,bool in_die)653 void hyperv_report_panic(struct pt_regs *regs, long err, bool in_die)
654 {
655 	static bool panic_reported;
656 	u64 guest_id;
657 
658 	if (in_die && !panic_on_oops)
659 		return;
660 
661 	/*
662 	 * We prefer to report panic on 'die' chain as we have proper
663 	 * registers to report, but if we miss it (e.g. on BUG()) we need
664 	 * to report it on 'panic'.
665 	 */
666 	if (panic_reported)
667 		return;
668 	panic_reported = true;
669 
670 	rdmsrq(HV_X64_MSR_GUEST_OS_ID, guest_id);
671 
672 	wrmsrq(HV_X64_MSR_CRASH_P0, err);
673 	wrmsrq(HV_X64_MSR_CRASH_P1, guest_id);
674 	wrmsrq(HV_X64_MSR_CRASH_P2, regs->ip);
675 	wrmsrq(HV_X64_MSR_CRASH_P3, regs->ax);
676 	wrmsrq(HV_X64_MSR_CRASH_P4, regs->sp);
677 
678 	/*
679 	 * Let Hyper-V know there is crash data available
680 	 */
681 	wrmsrq(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY);
682 }
683 EXPORT_SYMBOL_GPL(hyperv_report_panic);
684 
hv_is_hyperv_initialized(void)685 bool hv_is_hyperv_initialized(void)
686 {
687 	union hv_x64_msr_hypercall_contents hypercall_msr;
688 
689 	/*
690 	 * Ensure that we're really on Hyper-V, and not a KVM or Xen
691 	 * emulation of Hyper-V
692 	 */
693 	if (x86_hyper_type != X86_HYPER_MS_HYPERV)
694 		return false;
695 
696 	/* A TDX VM with no paravisor uses TDX GHCI call rather than hv_hypercall_pg */
697 	if (hv_isolation_type_tdx() && !ms_hyperv.paravisor_present)
698 		return true;
699 	/*
700 	 * Verify that earlier initialization succeeded by checking
701 	 * that the hypercall page is setup
702 	 */
703 	hypercall_msr.as_uint64 = 0;
704 	rdmsrq(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
705 
706 	return hypercall_msr.enable;
707 }
708 EXPORT_SYMBOL_GPL(hv_is_hyperv_initialized);
709 
hv_apicid_to_vp_index(u32 apic_id)710 int hv_apicid_to_vp_index(u32 apic_id)
711 {
712 	u64 control;
713 	u64 status;
714 	unsigned long irq_flags;
715 	struct hv_get_vp_from_apic_id_in *input;
716 	u32 *output, ret;
717 
718 	local_irq_save(irq_flags);
719 
720 	input = *this_cpu_ptr(hyperv_pcpu_input_arg);
721 	memset(input, 0, sizeof(*input));
722 	input->partition_id = HV_PARTITION_ID_SELF;
723 	input->apic_ids[0] = apic_id;
724 
725 	output = *this_cpu_ptr(hyperv_pcpu_output_arg);
726 
727 	control = HV_HYPERCALL_REP_COMP_1 | HVCALL_GET_VP_INDEX_FROM_APIC_ID;
728 	status = hv_do_hypercall(control, input, output);
729 	ret = output[0];
730 
731 	local_irq_restore(irq_flags);
732 
733 	if (!hv_result_success(status)) {
734 		pr_err("failed to get vp index from apic id %d, status %#llx\n",
735 		       apic_id, status);
736 		return -EINVAL;
737 	}
738 
739 	return ret;
740 }
741 EXPORT_SYMBOL_GPL(hv_apicid_to_vp_index);
742