xref: /linux/arch/x86/hyperv/hv_init.c (revision a7f7f6248d9740d710fd6bd190293fe5e16410ac)
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 #include <linux/acpi.h>
11 #include <linux/efi.h>
12 #include <linux/types.h>
13 #include <asm/apic.h>
14 #include <asm/desc.h>
15 #include <asm/hypervisor.h>
16 #include <asm/hyperv-tlfs.h>
17 #include <asm/mshyperv.h>
18 #include <linux/version.h>
19 #include <linux/vmalloc.h>
20 #include <linux/mm.h>
21 #include <linux/hyperv.h>
22 #include <linux/slab.h>
23 #include <linux/kernel.h>
24 #include <linux/cpuhotplug.h>
25 #include <linux/syscore_ops.h>
26 #include <clocksource/hyperv_timer.h>
27 
28 void *hv_hypercall_pg;
29 EXPORT_SYMBOL_GPL(hv_hypercall_pg);
30 
31 /* Storage to save the hypercall page temporarily for hibernation */
32 static void *hv_hypercall_pg_saved;
33 
34 u32 *hv_vp_index;
35 EXPORT_SYMBOL_GPL(hv_vp_index);
36 
37 struct hv_vp_assist_page **hv_vp_assist_page;
38 EXPORT_SYMBOL_GPL(hv_vp_assist_page);
39 
40 void  __percpu **hyperv_pcpu_input_arg;
41 EXPORT_SYMBOL_GPL(hyperv_pcpu_input_arg);
42 
43 u32 hv_max_vp_index;
44 EXPORT_SYMBOL_GPL(hv_max_vp_index);
45 
46 void *hv_alloc_hyperv_page(void)
47 {
48 	BUILD_BUG_ON(PAGE_SIZE != HV_HYP_PAGE_SIZE);
49 
50 	return (void *)__get_free_page(GFP_KERNEL);
51 }
52 EXPORT_SYMBOL_GPL(hv_alloc_hyperv_page);
53 
54 void *hv_alloc_hyperv_zeroed_page(void)
55 {
56         BUILD_BUG_ON(PAGE_SIZE != HV_HYP_PAGE_SIZE);
57 
58         return (void *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
59 }
60 EXPORT_SYMBOL_GPL(hv_alloc_hyperv_zeroed_page);
61 
62 void hv_free_hyperv_page(unsigned long addr)
63 {
64 	free_page(addr);
65 }
66 EXPORT_SYMBOL_GPL(hv_free_hyperv_page);
67 
68 static int hv_cpu_init(unsigned int cpu)
69 {
70 	u64 msr_vp_index;
71 	struct hv_vp_assist_page **hvp = &hv_vp_assist_page[smp_processor_id()];
72 	void **input_arg;
73 	struct page *pg;
74 
75 	input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
76 	/* hv_cpu_init() can be called with IRQs disabled from hv_resume() */
77 	pg = alloc_page(irqs_disabled() ? GFP_ATOMIC : GFP_KERNEL);
78 	if (unlikely(!pg))
79 		return -ENOMEM;
80 	*input_arg = page_address(pg);
81 
82 	hv_get_vp_index(msr_vp_index);
83 
84 	hv_vp_index[smp_processor_id()] = msr_vp_index;
85 
86 	if (msr_vp_index > hv_max_vp_index)
87 		hv_max_vp_index = msr_vp_index;
88 
89 	if (!hv_vp_assist_page)
90 		return 0;
91 
92 	/*
93 	 * The VP ASSIST PAGE is an "overlay" page (see Hyper-V TLFS's Section
94 	 * 5.2.1 "GPA Overlay Pages"). Here it must be zeroed out to make sure
95 	 * we always write the EOI MSR in hv_apic_eoi_write() *after* the
96 	 * EOI optimization is disabled in hv_cpu_die(), otherwise a CPU may
97 	 * not be stopped in the case of CPU offlining and the VM will hang.
98 	 */
99 	if (!*hvp) {
100 		*hvp = __vmalloc(PAGE_SIZE, GFP_KERNEL | __GFP_ZERO);
101 	}
102 
103 	if (*hvp) {
104 		u64 val;
105 
106 		val = vmalloc_to_pfn(*hvp);
107 		val = (val << HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT) |
108 			HV_X64_MSR_VP_ASSIST_PAGE_ENABLE;
109 
110 		wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, val);
111 	}
112 
113 	return 0;
114 }
115 
116 static void (*hv_reenlightenment_cb)(void);
117 
118 static void hv_reenlightenment_notify(struct work_struct *dummy)
119 {
120 	struct hv_tsc_emulation_status emu_status;
121 
122 	rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
123 
124 	/* Don't issue the callback if TSC accesses are not emulated */
125 	if (hv_reenlightenment_cb && emu_status.inprogress)
126 		hv_reenlightenment_cb();
127 }
128 static DECLARE_DELAYED_WORK(hv_reenlightenment_work, hv_reenlightenment_notify);
129 
130 void hyperv_stop_tsc_emulation(void)
131 {
132 	u64 freq;
133 	struct hv_tsc_emulation_status emu_status;
134 
135 	rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
136 	emu_status.inprogress = 0;
137 	wrmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
138 
139 	rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
140 	tsc_khz = div64_u64(freq, 1000);
141 }
142 EXPORT_SYMBOL_GPL(hyperv_stop_tsc_emulation);
143 
144 static inline bool hv_reenlightenment_available(void)
145 {
146 	/*
147 	 * Check for required features and priviliges to make TSC frequency
148 	 * change notifications work.
149 	 */
150 	return ms_hyperv.features & HV_X64_ACCESS_FREQUENCY_MSRS &&
151 		ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE &&
152 		ms_hyperv.features & HV_X64_ACCESS_REENLIGHTENMENT;
153 }
154 
155 __visible void __irq_entry hyperv_reenlightenment_intr(struct pt_regs *regs)
156 {
157 	entering_ack_irq();
158 
159 	inc_irq_stat(irq_hv_reenlightenment_count);
160 
161 	schedule_delayed_work(&hv_reenlightenment_work, HZ/10);
162 
163 	exiting_irq();
164 }
165 
166 void set_hv_tscchange_cb(void (*cb)(void))
167 {
168 	struct hv_reenlightenment_control re_ctrl = {
169 		.vector = HYPERV_REENLIGHTENMENT_VECTOR,
170 		.enabled = 1,
171 		.target_vp = hv_vp_index[smp_processor_id()]
172 	};
173 	struct hv_tsc_emulation_control emu_ctrl = {.enabled = 1};
174 
175 	if (!hv_reenlightenment_available()) {
176 		pr_warn("Hyper-V: reenlightenment support is unavailable\n");
177 		return;
178 	}
179 
180 	hv_reenlightenment_cb = cb;
181 
182 	/* Make sure callback is registered before we write to MSRs */
183 	wmb();
184 
185 	wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
186 	wrmsrl(HV_X64_MSR_TSC_EMULATION_CONTROL, *((u64 *)&emu_ctrl));
187 }
188 EXPORT_SYMBOL_GPL(set_hv_tscchange_cb);
189 
190 void clear_hv_tscchange_cb(void)
191 {
192 	struct hv_reenlightenment_control re_ctrl;
193 
194 	if (!hv_reenlightenment_available())
195 		return;
196 
197 	rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
198 	re_ctrl.enabled = 0;
199 	wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
200 
201 	hv_reenlightenment_cb = NULL;
202 }
203 EXPORT_SYMBOL_GPL(clear_hv_tscchange_cb);
204 
205 static int hv_cpu_die(unsigned int cpu)
206 {
207 	struct hv_reenlightenment_control re_ctrl;
208 	unsigned int new_cpu;
209 	unsigned long flags;
210 	void **input_arg;
211 	void *input_pg = NULL;
212 
213 	local_irq_save(flags);
214 	input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
215 	input_pg = *input_arg;
216 	*input_arg = NULL;
217 	local_irq_restore(flags);
218 	free_page((unsigned long)input_pg);
219 
220 	if (hv_vp_assist_page && hv_vp_assist_page[cpu])
221 		wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, 0);
222 
223 	if (hv_reenlightenment_cb == NULL)
224 		return 0;
225 
226 	rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
227 	if (re_ctrl.target_vp == hv_vp_index[cpu]) {
228 		/*
229 		 * Reassign reenlightenment notifications to some other online
230 		 * CPU or just disable the feature if there are no online CPUs
231 		 * left (happens on hibernation).
232 		 */
233 		new_cpu = cpumask_any_but(cpu_online_mask, cpu);
234 
235 		if (new_cpu < nr_cpu_ids)
236 			re_ctrl.target_vp = hv_vp_index[new_cpu];
237 		else
238 			re_ctrl.enabled = 0;
239 
240 		wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
241 	}
242 
243 	return 0;
244 }
245 
246 static int __init hv_pci_init(void)
247 {
248 	int gen2vm = efi_enabled(EFI_BOOT);
249 
250 	/*
251 	 * For Generation-2 VM, we exit from pci_arch_init() by returning 0.
252 	 * The purpose is to suppress the harmless warning:
253 	 * "PCI: Fatal: No config space access function found"
254 	 */
255 	if (gen2vm)
256 		return 0;
257 
258 	/* For Generation-1 VM, we'll proceed in pci_arch_init().  */
259 	return 1;
260 }
261 
262 static int hv_suspend(void)
263 {
264 	union hv_x64_msr_hypercall_contents hypercall_msr;
265 	int ret;
266 
267 	/*
268 	 * Reset the hypercall page as it is going to be invalidated
269 	 * accross hibernation. Setting hv_hypercall_pg to NULL ensures
270 	 * that any subsequent hypercall operation fails safely instead of
271 	 * crashing due to an access of an invalid page. The hypercall page
272 	 * pointer is restored on resume.
273 	 */
274 	hv_hypercall_pg_saved = hv_hypercall_pg;
275 	hv_hypercall_pg = NULL;
276 
277 	/* Disable the hypercall page in the hypervisor */
278 	rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
279 	hypercall_msr.enable = 0;
280 	wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
281 
282 	ret = hv_cpu_die(0);
283 	return ret;
284 }
285 
286 static void hv_resume(void)
287 {
288 	union hv_x64_msr_hypercall_contents hypercall_msr;
289 	int ret;
290 
291 	ret = hv_cpu_init(0);
292 	WARN_ON(ret);
293 
294 	/* Re-enable the hypercall page */
295 	rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
296 	hypercall_msr.enable = 1;
297 	hypercall_msr.guest_physical_address =
298 		vmalloc_to_pfn(hv_hypercall_pg_saved);
299 	wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
300 
301 	hv_hypercall_pg = hv_hypercall_pg_saved;
302 	hv_hypercall_pg_saved = NULL;
303 
304 	/*
305 	 * Reenlightenment notifications are disabled by hv_cpu_die(0),
306 	 * reenable them here if hv_reenlightenment_cb was previously set.
307 	 */
308 	if (hv_reenlightenment_cb)
309 		set_hv_tscchange_cb(hv_reenlightenment_cb);
310 }
311 
312 /* Note: when the ops are called, only CPU0 is online and IRQs are disabled. */
313 static struct syscore_ops hv_syscore_ops = {
314 	.suspend	= hv_suspend,
315 	.resume		= hv_resume,
316 };
317 
318 /*
319  * This function is to be invoked early in the boot sequence after the
320  * hypervisor has been detected.
321  *
322  * 1. Setup the hypercall page.
323  * 2. Register Hyper-V specific clocksource.
324  * 3. Setup Hyper-V specific APIC entry points.
325  */
326 void __init hyperv_init(void)
327 {
328 	u64 guest_id, required_msrs;
329 	union hv_x64_msr_hypercall_contents hypercall_msr;
330 	int cpuhp, i;
331 
332 	if (x86_hyper_type != X86_HYPER_MS_HYPERV)
333 		return;
334 
335 	/* Absolutely required MSRs */
336 	required_msrs = HV_X64_MSR_HYPERCALL_AVAILABLE |
337 		HV_X64_MSR_VP_INDEX_AVAILABLE;
338 
339 	if ((ms_hyperv.features & required_msrs) != required_msrs)
340 		return;
341 
342 	/*
343 	 * Allocate the per-CPU state for the hypercall input arg.
344 	 * If this allocation fails, we will not be able to setup
345 	 * (per-CPU) hypercall input page and thus this failure is
346 	 * fatal on Hyper-V.
347 	 */
348 	hyperv_pcpu_input_arg = alloc_percpu(void  *);
349 
350 	BUG_ON(hyperv_pcpu_input_arg == NULL);
351 
352 	/* Allocate percpu VP index */
353 	hv_vp_index = kmalloc_array(num_possible_cpus(), sizeof(*hv_vp_index),
354 				    GFP_KERNEL);
355 	if (!hv_vp_index)
356 		return;
357 
358 	for (i = 0; i < num_possible_cpus(); i++)
359 		hv_vp_index[i] = VP_INVAL;
360 
361 	hv_vp_assist_page = kcalloc(num_possible_cpus(),
362 				    sizeof(*hv_vp_assist_page), GFP_KERNEL);
363 	if (!hv_vp_assist_page) {
364 		ms_hyperv.hints &= ~HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
365 		goto free_vp_index;
366 	}
367 
368 	cpuhp = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/hyperv_init:online",
369 				  hv_cpu_init, hv_cpu_die);
370 	if (cpuhp < 0)
371 		goto free_vp_assist_page;
372 
373 	/*
374 	 * Setup the hypercall page and enable hypercalls.
375 	 * 1. Register the guest ID
376 	 * 2. Enable the hypercall and register the hypercall page
377 	 */
378 	guest_id = generate_guest_id(0, LINUX_VERSION_CODE, 0);
379 	wrmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
380 
381 	hv_hypercall_pg = vmalloc_exec(PAGE_SIZE);
382 	if (hv_hypercall_pg == NULL) {
383 		wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
384 		goto remove_cpuhp_state;
385 	}
386 
387 	rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
388 	hypercall_msr.enable = 1;
389 	hypercall_msr.guest_physical_address = vmalloc_to_pfn(hv_hypercall_pg);
390 	wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
391 
392 	/*
393 	 * Ignore any errors in setting up stimer clockevents
394 	 * as we can run with the LAPIC timer as a fallback.
395 	 */
396 	(void)hv_stimer_alloc();
397 
398 	hv_apic_init();
399 
400 	x86_init.pci.arch_init = hv_pci_init;
401 
402 	register_syscore_ops(&hv_syscore_ops);
403 
404 	return;
405 
406 remove_cpuhp_state:
407 	cpuhp_remove_state(cpuhp);
408 free_vp_assist_page:
409 	kfree(hv_vp_assist_page);
410 	hv_vp_assist_page = NULL;
411 free_vp_index:
412 	kfree(hv_vp_index);
413 	hv_vp_index = NULL;
414 }
415 
416 /*
417  * This routine is called before kexec/kdump, it does the required cleanup.
418  */
419 void hyperv_cleanup(void)
420 {
421 	union hv_x64_msr_hypercall_contents hypercall_msr;
422 
423 	unregister_syscore_ops(&hv_syscore_ops);
424 
425 	/* Reset our OS id */
426 	wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
427 
428 	/*
429 	 * Reset hypercall page reference before reset the page,
430 	 * let hypercall operations fail safely rather than
431 	 * panic the kernel for using invalid hypercall page
432 	 */
433 	hv_hypercall_pg = NULL;
434 
435 	/* Reset the hypercall page */
436 	hypercall_msr.as_uint64 = 0;
437 	wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
438 
439 	/* Reset the TSC page */
440 	hypercall_msr.as_uint64 = 0;
441 	wrmsrl(HV_X64_MSR_REFERENCE_TSC, hypercall_msr.as_uint64);
442 }
443 EXPORT_SYMBOL_GPL(hyperv_cleanup);
444 
445 void hyperv_report_panic(struct pt_regs *regs, long err, bool in_die)
446 {
447 	static bool panic_reported;
448 	u64 guest_id;
449 
450 	if (in_die && !panic_on_oops)
451 		return;
452 
453 	/*
454 	 * We prefer to report panic on 'die' chain as we have proper
455 	 * registers to report, but if we miss it (e.g. on BUG()) we need
456 	 * to report it on 'panic'.
457 	 */
458 	if (panic_reported)
459 		return;
460 	panic_reported = true;
461 
462 	rdmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
463 
464 	wrmsrl(HV_X64_MSR_CRASH_P0, err);
465 	wrmsrl(HV_X64_MSR_CRASH_P1, guest_id);
466 	wrmsrl(HV_X64_MSR_CRASH_P2, regs->ip);
467 	wrmsrl(HV_X64_MSR_CRASH_P3, regs->ax);
468 	wrmsrl(HV_X64_MSR_CRASH_P4, regs->sp);
469 
470 	/*
471 	 * Let Hyper-V know there is crash data available
472 	 */
473 	wrmsrl(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY);
474 }
475 EXPORT_SYMBOL_GPL(hyperv_report_panic);
476 
477 /**
478  * hyperv_report_panic_msg - report panic message to Hyper-V
479  * @pa: physical address of the panic page containing the message
480  * @size: size of the message in the page
481  */
482 void hyperv_report_panic_msg(phys_addr_t pa, size_t size)
483 {
484 	/*
485 	 * P3 to contain the physical address of the panic page & P4 to
486 	 * contain the size of the panic data in that page. Rest of the
487 	 * registers are no-op when the NOTIFY_MSG flag is set.
488 	 */
489 	wrmsrl(HV_X64_MSR_CRASH_P0, 0);
490 	wrmsrl(HV_X64_MSR_CRASH_P1, 0);
491 	wrmsrl(HV_X64_MSR_CRASH_P2, 0);
492 	wrmsrl(HV_X64_MSR_CRASH_P3, pa);
493 	wrmsrl(HV_X64_MSR_CRASH_P4, size);
494 
495 	/*
496 	 * Let Hyper-V know there is crash data available along with
497 	 * the panic message.
498 	 */
499 	wrmsrl(HV_X64_MSR_CRASH_CTL,
500 	       (HV_CRASH_CTL_CRASH_NOTIFY | HV_CRASH_CTL_CRASH_NOTIFY_MSG));
501 }
502 EXPORT_SYMBOL_GPL(hyperv_report_panic_msg);
503 
504 bool hv_is_hyperv_initialized(void)
505 {
506 	union hv_x64_msr_hypercall_contents hypercall_msr;
507 
508 	/*
509 	 * Ensure that we're really on Hyper-V, and not a KVM or Xen
510 	 * emulation of Hyper-V
511 	 */
512 	if (x86_hyper_type != X86_HYPER_MS_HYPERV)
513 		return false;
514 
515 	/*
516 	 * Verify that earlier initialization succeeded by checking
517 	 * that the hypercall page is setup
518 	 */
519 	hypercall_msr.as_uint64 = 0;
520 	rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
521 
522 	return hypercall_msr.enable;
523 }
524 EXPORT_SYMBOL_GPL(hv_is_hyperv_initialized);
525 
526 bool hv_is_hibernation_supported(void)
527 {
528 	return acpi_sleep_state_supported(ACPI_STATE_S4);
529 }
530 EXPORT_SYMBOL_GPL(hv_is_hibernation_supported);
531