xref: /linux/arch/x86/hyperv/hv_init.c (revision 5f054ef2e0f1ca7d32ac48e275d08e2ac29d84f3)
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/ibt.h>
21 #include <asm/hypervisor.h>
22 #include <hyperv/hvhdk.h>
23 #include <asm/mshyperv.h>
24 #include <asm/msr.h>
25 #include <asm/idtentry.h>
26 #include <asm/set_memory.h>
27 #include <linux/kexec.h>
28 #include <linux/version.h>
29 #include <linux/vmalloc.h>
30 #include <linux/mm.h>
31 #include <linux/slab.h>
32 #include <linux/kernel.h>
33 #include <linux/cpuhotplug.h>
34 #include <linux/syscore_ops.h>
35 #include <clocksource/hyperv_timer.h>
36 #include <linux/highmem.h>
37 #include <linux/export.h>
38 
39 void *hv_hypercall_pg;
40 EXPORT_SYMBOL_GPL(hv_hypercall_pg);
41 
42 union hv_ghcb * __percpu *hv_ghcb_pg;
43 
44 /* Storage to save the hypercall page temporarily for hibernation */
45 static void *hv_hypercall_pg_saved;
46 
47 struct hv_vp_assist_page **hv_vp_assist_page;
48 EXPORT_SYMBOL_GPL(hv_vp_assist_page);
49 
hyperv_init_ghcb(void)50 static int hyperv_init_ghcb(void)
51 {
52 	u64 ghcb_gpa;
53 	void *ghcb_va;
54 	void **ghcb_base;
55 
56 	if (!ms_hyperv.paravisor_present || !hv_isolation_type_snp())
57 		return 0;
58 
59 	if (!hv_ghcb_pg)
60 		return -EINVAL;
61 
62 	/*
63 	 * GHCB page is allocated by paravisor. The address
64 	 * returned by MSR_AMD64_SEV_ES_GHCB is above shared
65 	 * memory boundary and map it here.
66 	 */
67 	rdmsrq(MSR_AMD64_SEV_ES_GHCB, ghcb_gpa);
68 
69 	/* Mask out vTOM bit. ioremap_cache() maps decrypted */
70 	ghcb_gpa &= ~ms_hyperv.shared_gpa_boundary;
71 	ghcb_va = (void *)ioremap_cache(ghcb_gpa, HV_HYP_PAGE_SIZE);
72 	if (!ghcb_va)
73 		return -ENOMEM;
74 
75 	ghcb_base = (void **)this_cpu_ptr(hv_ghcb_pg);
76 	*ghcb_base = ghcb_va;
77 
78 	return 0;
79 }
80 
hv_cpu_init(unsigned int cpu)81 static int hv_cpu_init(unsigned int cpu)
82 {
83 	union hv_vp_assist_msr_contents msr = { 0 };
84 	struct hv_vp_assist_page **hvp;
85 	int ret;
86 
87 	ret = hv_common_cpu_init(cpu);
88 	if (ret)
89 		return ret;
90 
91 	if (!hv_vp_assist_page)
92 		return 0;
93 
94 	hvp = &hv_vp_assist_page[cpu];
95 	if (hv_root_partition()) {
96 		/*
97 		 * For root partition we get the hypervisor provided VP assist
98 		 * page, instead of allocating a new page.
99 		 */
100 		rdmsrq(HV_X64_MSR_VP_ASSIST_PAGE, msr.as_uint64);
101 		*hvp = memremap(msr.pfn << HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT,
102 				PAGE_SIZE, MEMREMAP_WB);
103 	} else {
104 		/*
105 		 * The VP assist page is an "overlay" page (see Hyper-V TLFS's
106 		 * Section 5.2.1 "GPA Overlay Pages"). Here it must be zeroed
107 		 * out to make sure we always write the EOI MSR in
108 		 * hv_apic_eoi_write() *after* the EOI optimization is disabled
109 		 * in hv_cpu_die(), otherwise a CPU may not be stopped in the
110 		 * case of CPU offlining and the VM will hang.
111 		 */
112 		if (!*hvp) {
113 			*hvp = __vmalloc(PAGE_SIZE, GFP_KERNEL | __GFP_ZERO);
114 
115 			/*
116 			 * Hyper-V should never specify a VM that is a Confidential
117 			 * VM and also running in the root partition. Root partition
118 			 * is blocked to run in Confidential VM. So only decrypt assist
119 			 * page in non-root partition here.
120 			 */
121 			if (*hvp && !ms_hyperv.paravisor_present && hv_isolation_type_snp()) {
122 				WARN_ON_ONCE(set_memory_decrypted((unsigned long)(*hvp), 1));
123 				memset(*hvp, 0, PAGE_SIZE);
124 			}
125 		}
126 
127 		if (*hvp)
128 			msr.pfn = vmalloc_to_pfn(*hvp);
129 
130 	}
131 	if (!WARN_ON(!(*hvp))) {
132 		msr.enable = 1;
133 		wrmsrq(HV_X64_MSR_VP_ASSIST_PAGE, msr.as_uint64);
134 	}
135 
136 	return hyperv_init_ghcb();
137 }
138 
139 static void (*hv_reenlightenment_cb)(void);
140 
hv_reenlightenment_notify(struct work_struct * dummy)141 static void hv_reenlightenment_notify(struct work_struct *dummy)
142 {
143 	struct hv_tsc_emulation_status emu_status;
144 
145 	rdmsrq(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
146 
147 	/* Don't issue the callback if TSC accesses are not emulated */
148 	if (hv_reenlightenment_cb && emu_status.inprogress)
149 		hv_reenlightenment_cb();
150 }
151 static DECLARE_DELAYED_WORK(hv_reenlightenment_work, hv_reenlightenment_notify);
152 
hyperv_stop_tsc_emulation(void)153 void hyperv_stop_tsc_emulation(void)
154 {
155 	u64 freq;
156 	struct hv_tsc_emulation_status emu_status;
157 
158 	rdmsrq(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
159 	emu_status.inprogress = 0;
160 	wrmsrq(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
161 
162 	rdmsrq(HV_X64_MSR_TSC_FREQUENCY, freq);
163 	tsc_khz = div64_u64(freq, 1000);
164 }
165 EXPORT_SYMBOL_GPL(hyperv_stop_tsc_emulation);
166 
hv_reenlightenment_available(void)167 static inline bool hv_reenlightenment_available(void)
168 {
169 	/*
170 	 * Check for required features and privileges to make TSC frequency
171 	 * change notifications work.
172 	 */
173 	return ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
174 		ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE &&
175 		ms_hyperv.features & HV_ACCESS_REENLIGHTENMENT;
176 }
177 
DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_reenlightenment)178 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_reenlightenment)
179 {
180 	apic_eoi();
181 	inc_irq_stat(irq_hv_reenlightenment_count);
182 	schedule_delayed_work(&hv_reenlightenment_work, HZ/10);
183 }
184 
set_hv_tscchange_cb(void (* cb)(void))185 void set_hv_tscchange_cb(void (*cb)(void))
186 {
187 	struct hv_reenlightenment_control re_ctrl = {
188 		.vector = HYPERV_REENLIGHTENMENT_VECTOR,
189 		.enabled = 1,
190 	};
191 	struct hv_tsc_emulation_control emu_ctrl = {.enabled = 1};
192 
193 	if (!hv_reenlightenment_available()) {
194 		pr_warn("reenlightenment support is unavailable\n");
195 		return;
196 	}
197 
198 	if (!hv_vp_index)
199 		return;
200 
201 	hv_reenlightenment_cb = cb;
202 
203 	/* Make sure callback is registered before we write to MSRs */
204 	wmb();
205 
206 	re_ctrl.target_vp = hv_vp_index[get_cpu()];
207 
208 	wrmsrq(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
209 	wrmsrq(HV_X64_MSR_TSC_EMULATION_CONTROL, *((u64 *)&emu_ctrl));
210 
211 	put_cpu();
212 }
213 EXPORT_SYMBOL_GPL(set_hv_tscchange_cb);
214 
clear_hv_tscchange_cb(void)215 void clear_hv_tscchange_cb(void)
216 {
217 	struct hv_reenlightenment_control re_ctrl;
218 
219 	if (!hv_reenlightenment_available())
220 		return;
221 
222 	rdmsrq(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
223 	re_ctrl.enabled = 0;
224 	wrmsrq(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
225 
226 	hv_reenlightenment_cb = NULL;
227 }
228 EXPORT_SYMBOL_GPL(clear_hv_tscchange_cb);
229 
hv_cpu_die(unsigned int cpu)230 static int hv_cpu_die(unsigned int cpu)
231 {
232 	struct hv_reenlightenment_control re_ctrl;
233 	unsigned int new_cpu;
234 	void **ghcb_va;
235 
236 	if (hv_ghcb_pg) {
237 		ghcb_va = (void **)this_cpu_ptr(hv_ghcb_pg);
238 		if (*ghcb_va)
239 			iounmap(*ghcb_va);
240 		*ghcb_va = NULL;
241 	}
242 
243 	hv_common_cpu_die(cpu);
244 
245 	if (hv_vp_assist_page && hv_vp_assist_page[cpu]) {
246 		union hv_vp_assist_msr_contents msr = { 0 };
247 		if (hv_root_partition()) {
248 			/*
249 			 * For root partition the VP assist page is mapped to
250 			 * hypervisor provided page, and thus we unmap the
251 			 * page here and nullify it, so that in future we have
252 			 * correct page address mapped in hv_cpu_init.
253 			 */
254 			memunmap(hv_vp_assist_page[cpu]);
255 			hv_vp_assist_page[cpu] = NULL;
256 			rdmsrq(HV_X64_MSR_VP_ASSIST_PAGE, msr.as_uint64);
257 			msr.enable = 0;
258 		}
259 		wrmsrq(HV_X64_MSR_VP_ASSIST_PAGE, msr.as_uint64);
260 	}
261 
262 	if (hv_reenlightenment_cb == NULL)
263 		return 0;
264 
265 	rdmsrq(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
266 	if (re_ctrl.target_vp == hv_vp_index[cpu]) {
267 		/*
268 		 * Reassign reenlightenment notifications to some other online
269 		 * CPU or just disable the feature if there are no online CPUs
270 		 * left (happens on hibernation).
271 		 */
272 		new_cpu = cpumask_any_but(cpu_online_mask, cpu);
273 
274 		if (new_cpu < nr_cpu_ids)
275 			re_ctrl.target_vp = hv_vp_index[new_cpu];
276 		else
277 			re_ctrl.enabled = 0;
278 
279 		wrmsrq(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
280 	}
281 
282 	return 0;
283 }
284 
hv_pci_init(void)285 static int __init hv_pci_init(void)
286 {
287 	bool gen2vm = efi_enabled(EFI_BOOT);
288 
289 	/*
290 	 * A Generation-2 VM doesn't support legacy PCI/PCIe, so both
291 	 * raw_pci_ops and raw_pci_ext_ops are NULL, and pci_subsys_init() ->
292 	 * pcibios_init() doesn't call pcibios_resource_survey() ->
293 	 * e820__reserve_resources_late(); as a result, any emulated persistent
294 	 * memory of E820_TYPE_PRAM (12) via the kernel parameter
295 	 * memmap=nn[KMG]!ss is not added into iomem_resource and hence can't be
296 	 * detected by register_e820_pmem(). Fix this by directly calling
297 	 * e820__reserve_resources_late() here: e820__reserve_resources_late()
298 	 * depends on e820__reserve_resources(), which has been called earlier
299 	 * from setup_arch(). Note: e820__reserve_resources_late() also adds
300 	 * any memory of E820_TYPE_PMEM (7) into iomem_resource, and
301 	 * acpi_nfit_register_region() -> acpi_nfit_insert_resource() ->
302 	 * region_intersects() returns REGION_INTERSECTS, so the memory of
303 	 * E820_TYPE_PMEM won't get added twice.
304 	 *
305 	 * We return 0 here so that pci_arch_init() won't print the warning:
306 	 * "PCI: Fatal: No config space access function found"
307 	 */
308 	if (gen2vm) {
309 		e820__reserve_resources_late();
310 		return 0;
311 	}
312 
313 	/* For Generation-1 VM, we'll proceed in pci_arch_init().  */
314 	return 1;
315 }
316 
hv_suspend(void)317 static int hv_suspend(void)
318 {
319 	union hv_x64_msr_hypercall_contents hypercall_msr;
320 	int ret;
321 
322 	if (hv_root_partition())
323 		return -EPERM;
324 
325 	/*
326 	 * Reset the hypercall page as it is going to be invalidated
327 	 * across hibernation. Setting hv_hypercall_pg to NULL ensures
328 	 * that any subsequent hypercall operation fails safely instead of
329 	 * crashing due to an access of an invalid page. The hypercall page
330 	 * pointer is restored on resume.
331 	 */
332 	hv_hypercall_pg_saved = hv_hypercall_pg;
333 	hv_hypercall_pg = NULL;
334 
335 	/* Disable the hypercall page in the hypervisor */
336 	rdmsrq(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
337 	hypercall_msr.enable = 0;
338 	wrmsrq(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
339 
340 	ret = hv_cpu_die(0);
341 	return ret;
342 }
343 
hv_resume(void)344 static void hv_resume(void)
345 {
346 	union hv_x64_msr_hypercall_contents hypercall_msr;
347 	int ret;
348 
349 	ret = hv_cpu_init(0);
350 	WARN_ON(ret);
351 
352 	/* Re-enable the hypercall page */
353 	rdmsrq(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
354 	hypercall_msr.enable = 1;
355 	hypercall_msr.guest_physical_address =
356 		vmalloc_to_pfn(hv_hypercall_pg_saved);
357 	wrmsrq(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
358 
359 	hv_hypercall_pg = hv_hypercall_pg_saved;
360 	hv_hypercall_pg_saved = NULL;
361 
362 	/*
363 	 * Reenlightenment notifications are disabled by hv_cpu_die(0),
364 	 * reenable them here if hv_reenlightenment_cb was previously set.
365 	 */
366 	if (hv_reenlightenment_cb)
367 		set_hv_tscchange_cb(hv_reenlightenment_cb);
368 }
369 
370 /* Note: when the ops are called, only CPU0 is online and IRQs are disabled. */
371 static struct syscore_ops hv_syscore_ops = {
372 	.suspend	= hv_suspend,
373 	.resume		= hv_resume,
374 };
375 
376 static void (* __initdata old_setup_percpu_clockev)(void);
377 
hv_stimer_setup_percpu_clockev(void)378 static void __init hv_stimer_setup_percpu_clockev(void)
379 {
380 	/*
381 	 * Ignore any errors in setting up stimer clockevents
382 	 * as we can run with the LAPIC timer as a fallback.
383 	 */
384 	(void)hv_stimer_alloc(false);
385 
386 	/*
387 	 * Still register the LAPIC timer, because the direct-mode STIMER is
388 	 * not supported by old versions of Hyper-V. This also allows users
389 	 * to switch to LAPIC timer via /sys, if they want to.
390 	 */
391 	if (old_setup_percpu_clockev)
392 		old_setup_percpu_clockev();
393 }
394 
395 /*
396  * This function is to be invoked early in the boot sequence after the
397  * hypervisor has been detected.
398  *
399  * 1. Setup the hypercall page.
400  * 2. Register Hyper-V specific clocksource.
401  * 3. Setup Hyper-V specific APIC entry points.
402  */
hyperv_init(void)403 void __init hyperv_init(void)
404 {
405 	u64 guest_id;
406 	union hv_x64_msr_hypercall_contents hypercall_msr;
407 	int cpuhp;
408 
409 	if (x86_hyper_type != X86_HYPER_MS_HYPERV)
410 		return;
411 
412 	if (hv_common_init())
413 		return;
414 
415 	/*
416 	 * The VP assist page is useless to a TDX guest: the only use we
417 	 * would have for it is lazy EOI, which can not be used with TDX.
418 	 */
419 	if (hv_isolation_type_tdx())
420 		hv_vp_assist_page = NULL;
421 	else
422 		hv_vp_assist_page = kcalloc(nr_cpu_ids,
423 					    sizeof(*hv_vp_assist_page),
424 					    GFP_KERNEL);
425 	if (!hv_vp_assist_page) {
426 		ms_hyperv.hints &= ~HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
427 
428 		if (!hv_isolation_type_tdx())
429 			goto common_free;
430 	}
431 
432 	if (ms_hyperv.paravisor_present && hv_isolation_type_snp()) {
433 		/* Negotiate GHCB Version. */
434 		if (!hv_ghcb_negotiate_protocol())
435 			hv_ghcb_terminate(SEV_TERM_SET_GEN,
436 					  GHCB_SEV_ES_PROT_UNSUPPORTED);
437 
438 		hv_ghcb_pg = alloc_percpu(union hv_ghcb *);
439 		if (!hv_ghcb_pg)
440 			goto free_vp_assist_page;
441 	}
442 
443 	cpuhp = cpuhp_setup_state(CPUHP_AP_HYPERV_ONLINE, "x86/hyperv_init:online",
444 				  hv_cpu_init, hv_cpu_die);
445 	if (cpuhp < 0)
446 		goto free_ghcb_page;
447 
448 	/*
449 	 * Setup the hypercall page and enable hypercalls.
450 	 * 1. Register the guest ID
451 	 * 2. Enable the hypercall and register the hypercall page
452 	 *
453 	 * A TDX VM with no paravisor only uses TDX GHCI rather than hv_hypercall_pg:
454 	 * when the hypercall input is a page, such a VM must pass a decrypted
455 	 * page to Hyper-V, e.g. hv_post_message() uses the per-CPU page
456 	 * hyperv_pcpu_input_arg, which is decrypted if no paravisor is present.
457 	 *
458 	 * A TDX VM with the paravisor uses hv_hypercall_pg for most hypercalls,
459 	 * which are handled by the paravisor and the VM must use an encrypted
460 	 * input page: in such a VM, the hyperv_pcpu_input_arg is encrypted and
461 	 * used in the hypercalls, e.g. see hv_mark_gpa_visibility() and
462 	 * hv_arch_irq_unmask(). Such a VM uses TDX GHCI for two hypercalls:
463 	 * 1. HVCALL_SIGNAL_EVENT: see vmbus_set_event() and _hv_do_fast_hypercall8().
464 	 * 2. HVCALL_POST_MESSAGE: the input page must be a decrypted page, i.e.
465 	 * hv_post_message() in such a VM can't use the encrypted hyperv_pcpu_input_arg;
466 	 * instead, hv_post_message() uses the post_msg_page, which is decrypted
467 	 * in such a VM and is only used in such a VM.
468 	 */
469 	guest_id = hv_generate_guest_id(LINUX_VERSION_CODE);
470 	wrmsrq(HV_X64_MSR_GUEST_OS_ID, guest_id);
471 
472 	/* With the paravisor, the VM must also write the ID via GHCB/GHCI */
473 	hv_ivm_msr_write(HV_X64_MSR_GUEST_OS_ID, guest_id);
474 
475 	/* A TDX VM with no paravisor only uses TDX GHCI rather than hv_hypercall_pg */
476 	if (hv_isolation_type_tdx() && !ms_hyperv.paravisor_present)
477 		goto skip_hypercall_pg_init;
478 
479 	hv_hypercall_pg = __vmalloc_node_range(PAGE_SIZE, 1, VMALLOC_START,
480 			VMALLOC_END, GFP_KERNEL, PAGE_KERNEL_ROX,
481 			VM_FLUSH_RESET_PERMS, NUMA_NO_NODE,
482 			__builtin_return_address(0));
483 	if (hv_hypercall_pg == NULL)
484 		goto clean_guest_os_id;
485 
486 	rdmsrq(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
487 	hypercall_msr.enable = 1;
488 
489 	if (hv_root_partition()) {
490 		struct page *pg;
491 		void *src;
492 
493 		/*
494 		 * For the root partition, the hypervisor will set up its
495 		 * hypercall page. The hypervisor guarantees it will not show
496 		 * up in the root's address space. The root can't change the
497 		 * location of the hypercall page.
498 		 *
499 		 * Order is important here. We must enable the hypercall page
500 		 * so it is populated with code, then copy the code to an
501 		 * executable page.
502 		 */
503 		wrmsrq(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
504 
505 		pg = vmalloc_to_page(hv_hypercall_pg);
506 		src = memremap(hypercall_msr.guest_physical_address << PAGE_SHIFT, PAGE_SIZE,
507 				MEMREMAP_WB);
508 		BUG_ON(!src);
509 		memcpy_to_page(pg, 0, src, HV_HYP_PAGE_SIZE);
510 		memunmap(src);
511 
512 		hv_remap_tsc_clocksource();
513 	} else {
514 		hypercall_msr.guest_physical_address = vmalloc_to_pfn(hv_hypercall_pg);
515 		wrmsrq(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
516 	}
517 
518 skip_hypercall_pg_init:
519 	/*
520 	 * Some versions of Hyper-V that provide IBT in guest VMs have a bug
521 	 * in that there's no ENDBR64 instruction at the entry to the
522 	 * hypercall page. Because hypercalls are invoked via an indirect call
523 	 * to the hypercall page, all hypercall attempts fail when IBT is
524 	 * enabled, and Linux panics. For such buggy versions, disable IBT.
525 	 *
526 	 * Fixed versions of Hyper-V always provide ENDBR64 on the hypercall
527 	 * page, so if future Linux kernel versions enable IBT for 32-bit
528 	 * builds, additional hypercall page hackery will be required here
529 	 * to provide an ENDBR32.
530 	 */
531 #ifdef CONFIG_X86_KERNEL_IBT
532 	if (cpu_feature_enabled(X86_FEATURE_IBT) &&
533 	    *(u32 *)hv_hypercall_pg != gen_endbr()) {
534 		setup_clear_cpu_cap(X86_FEATURE_IBT);
535 		pr_warn("Disabling IBT because of Hyper-V bug\n");
536 	}
537 #endif
538 
539 	/*
540 	 * hyperv_init() is called before LAPIC is initialized: see
541 	 * apic_intr_mode_init() -> x86_platform.apic_post_init() and
542 	 * apic_bsp_setup() -> setup_local_APIC(). The direct-mode STIMER
543 	 * depends on LAPIC, so hv_stimer_alloc() should be called from
544 	 * x86_init.timers.setup_percpu_clockev.
545 	 */
546 	old_setup_percpu_clockev = x86_init.timers.setup_percpu_clockev;
547 	x86_init.timers.setup_percpu_clockev = hv_stimer_setup_percpu_clockev;
548 
549 	hv_apic_init();
550 
551 	x86_init.pci.arch_init = hv_pci_init;
552 
553 	register_syscore_ops(&hv_syscore_ops);
554 
555 	if (ms_hyperv.priv_high & HV_ACCESS_PARTITION_ID)
556 		hv_get_partition_id();
557 
558 #ifdef CONFIG_PCI_MSI
559 	/*
560 	 * If we're running as root, we want to create our own PCI MSI domain.
561 	 * We can't set this in hv_pci_init because that would be too late.
562 	 */
563 	if (hv_root_partition())
564 		x86_init.irqs.create_pci_msi_domain = hv_create_pci_msi_domain;
565 #endif
566 
567 	/* Query the VMs extended capability once, so that it can be cached. */
568 	hv_query_ext_cap(0);
569 
570 	/* Find the VTL */
571 	ms_hyperv.vtl = get_vtl();
572 
573 	if (ms_hyperv.vtl > 0) /* non default VTL */
574 		hv_vtl_early_init();
575 
576 	return;
577 
578 clean_guest_os_id:
579 	wrmsrq(HV_X64_MSR_GUEST_OS_ID, 0);
580 	hv_ivm_msr_write(HV_X64_MSR_GUEST_OS_ID, 0);
581 	cpuhp_remove_state(CPUHP_AP_HYPERV_ONLINE);
582 free_ghcb_page:
583 	free_percpu(hv_ghcb_pg);
584 free_vp_assist_page:
585 	kfree(hv_vp_assist_page);
586 	hv_vp_assist_page = NULL;
587 common_free:
588 	hv_common_free();
589 }
590 
591 /*
592  * This routine is called before kexec/kdump, it does the required cleanup.
593  */
hyperv_cleanup(void)594 void hyperv_cleanup(void)
595 {
596 	union hv_x64_msr_hypercall_contents hypercall_msr;
597 	union hv_reference_tsc_msr tsc_msr;
598 
599 	/* Reset our OS id */
600 	wrmsrq(HV_X64_MSR_GUEST_OS_ID, 0);
601 	hv_ivm_msr_write(HV_X64_MSR_GUEST_OS_ID, 0);
602 
603 	/*
604 	 * Reset hypercall page reference before reset the page,
605 	 * let hypercall operations fail safely rather than
606 	 * panic the kernel for using invalid hypercall page
607 	 */
608 	hv_hypercall_pg = NULL;
609 
610 	/* Reset the hypercall page */
611 	hypercall_msr.as_uint64 = hv_get_msr(HV_X64_MSR_HYPERCALL);
612 	hypercall_msr.enable = 0;
613 	hv_set_msr(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
614 
615 	/* Reset the TSC page */
616 	tsc_msr.as_uint64 = hv_get_msr(HV_X64_MSR_REFERENCE_TSC);
617 	tsc_msr.enable = 0;
618 	hv_set_msr(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
619 }
620 
hyperv_report_panic(struct pt_regs * regs,long err,bool in_die)621 void hyperv_report_panic(struct pt_regs *regs, long err, bool in_die)
622 {
623 	static bool panic_reported;
624 	u64 guest_id;
625 
626 	if (in_die && !panic_on_oops)
627 		return;
628 
629 	/*
630 	 * We prefer to report panic on 'die' chain as we have proper
631 	 * registers to report, but if we miss it (e.g. on BUG()) we need
632 	 * to report it on 'panic'.
633 	 */
634 	if (panic_reported)
635 		return;
636 	panic_reported = true;
637 
638 	rdmsrq(HV_X64_MSR_GUEST_OS_ID, guest_id);
639 
640 	wrmsrq(HV_X64_MSR_CRASH_P0, err);
641 	wrmsrq(HV_X64_MSR_CRASH_P1, guest_id);
642 	wrmsrq(HV_X64_MSR_CRASH_P2, regs->ip);
643 	wrmsrq(HV_X64_MSR_CRASH_P3, regs->ax);
644 	wrmsrq(HV_X64_MSR_CRASH_P4, regs->sp);
645 
646 	/*
647 	 * Let Hyper-V know there is crash data available
648 	 */
649 	wrmsrq(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY);
650 }
651 EXPORT_SYMBOL_GPL(hyperv_report_panic);
652 
hv_is_hyperv_initialized(void)653 bool hv_is_hyperv_initialized(void)
654 {
655 	union hv_x64_msr_hypercall_contents hypercall_msr;
656 
657 	/*
658 	 * Ensure that we're really on Hyper-V, and not a KVM or Xen
659 	 * emulation of Hyper-V
660 	 */
661 	if (x86_hyper_type != X86_HYPER_MS_HYPERV)
662 		return false;
663 
664 	/* A TDX VM with no paravisor uses TDX GHCI call rather than hv_hypercall_pg */
665 	if (hv_isolation_type_tdx() && !ms_hyperv.paravisor_present)
666 		return true;
667 	/*
668 	 * Verify that earlier initialization succeeded by checking
669 	 * that the hypercall page is setup
670 	 */
671 	hypercall_msr.as_uint64 = 0;
672 	rdmsrq(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
673 
674 	return hypercall_msr.enable;
675 }
676 EXPORT_SYMBOL_GPL(hv_is_hyperv_initialized);
677 
hv_apicid_to_vp_index(u32 apic_id)678 int hv_apicid_to_vp_index(u32 apic_id)
679 {
680 	u64 control;
681 	u64 status;
682 	unsigned long irq_flags;
683 	struct hv_get_vp_from_apic_id_in *input;
684 	u32 *output, ret;
685 
686 	local_irq_save(irq_flags);
687 
688 	input = *this_cpu_ptr(hyperv_pcpu_input_arg);
689 	memset(input, 0, sizeof(*input));
690 	input->partition_id = HV_PARTITION_ID_SELF;
691 	input->apic_ids[0] = apic_id;
692 
693 	output = *this_cpu_ptr(hyperv_pcpu_output_arg);
694 
695 	control = HV_HYPERCALL_REP_COMP_1 | HVCALL_GET_VP_INDEX_FROM_APIC_ID;
696 	status = hv_do_hypercall(control, input, output);
697 	ret = output[0];
698 
699 	local_irq_restore(irq_flags);
700 
701 	if (!hv_result_success(status)) {
702 		pr_err("failed to get vp index from apic id %d, status %#llx\n",
703 		       apic_id, status);
704 		return -EINVAL;
705 	}
706 
707 	return ret;
708 }
709 EXPORT_SYMBOL_GPL(hv_apicid_to_vp_index);
710