Loading...
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/efi.h>
11#include <linux/types.h>
12#include <linux/bitfield.h>
13#include <linux/io.h>
14#include <asm/apic.h>
15#include <asm/desc.h>
16#include <asm/sev.h>
17#include <asm/hypervisor.h>
18#include <asm/hyperv-tlfs.h>
19#include <asm/mshyperv.h>
20#include <asm/idtentry.h>
21#include <linux/kexec.h>
22#include <linux/version.h>
23#include <linux/vmalloc.h>
24#include <linux/mm.h>
25#include <linux/hyperv.h>
26#include <linux/slab.h>
27#include <linux/kernel.h>
28#include <linux/cpuhotplug.h>
29#include <linux/syscore_ops.h>
30#include <clocksource/hyperv_timer.h>
31#include <linux/highmem.h>
32#include <linux/swiotlb.h>
33
34int hyperv_init_cpuhp;
35u64 hv_current_partition_id = ~0ull;
36EXPORT_SYMBOL_GPL(hv_current_partition_id);
37
38void *hv_hypercall_pg;
39EXPORT_SYMBOL_GPL(hv_hypercall_pg);
40
41union hv_ghcb * __percpu *hv_ghcb_pg;
42
43/* Storage to save the hypercall page temporarily for hibernation */
44static void *hv_hypercall_pg_saved;
45
46struct hv_vp_assist_page **hv_vp_assist_page;
47EXPORT_SYMBOL_GPL(hv_vp_assist_page);
48
49static int hyperv_init_ghcb(void)
50{
51 u64 ghcb_gpa;
52 void *ghcb_va;
53 void **ghcb_base;
54
55 if (!hv_isolation_type_snp())
56 return 0;
57
58 if (!hv_ghcb_pg)
59 return -EINVAL;
60
61 /*
62 * GHCB page is allocated by paravisor. The address
63 * returned by MSR_AMD64_SEV_ES_GHCB is above shared
64 * memory boundary and map it here.
65 */
66 rdmsrl(MSR_AMD64_SEV_ES_GHCB, ghcb_gpa);
67 ghcb_va = memremap(ghcb_gpa, HV_HYP_PAGE_SIZE, MEMREMAP_WB);
68 if (!ghcb_va)
69 return -ENOMEM;
70
71 ghcb_base = (void **)this_cpu_ptr(hv_ghcb_pg);
72 *ghcb_base = ghcb_va;
73
74 return 0;
75}
76
77static int hv_cpu_init(unsigned int cpu)
78{
79 union hv_vp_assist_msr_contents msr = { 0 };
80 struct hv_vp_assist_page **hvp = &hv_vp_assist_page[cpu];
81 int ret;
82
83 ret = hv_common_cpu_init(cpu);
84 if (ret)
85 return ret;
86
87 if (!hv_vp_assist_page)
88 return 0;
89
90 if (hv_root_partition) {
91 /*
92 * For root partition we get the hypervisor provided VP assist
93 * page, instead of allocating a new page.
94 */
95 rdmsrl(HV_X64_MSR_VP_ASSIST_PAGE, msr.as_uint64);
96 *hvp = memremap(msr.pfn << HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT,
97 PAGE_SIZE, MEMREMAP_WB);
98 } else {
99 /*
100 * The VP assist page is an "overlay" page (see Hyper-V TLFS's
101 * Section 5.2.1 "GPA Overlay Pages"). Here it must be zeroed
102 * out to make sure we always write the EOI MSR in
103 * hv_apic_eoi_write() *after* the EOI optimization is disabled
104 * in hv_cpu_die(), otherwise a CPU may not be stopped in the
105 * case of CPU offlining and the VM will hang.
106 */
107 if (!*hvp)
108 *hvp = __vmalloc(PAGE_SIZE, GFP_KERNEL | __GFP_ZERO);
109 if (*hvp)
110 msr.pfn = vmalloc_to_pfn(*hvp);
111
112 }
113 if (!WARN_ON(!(*hvp))) {
114 msr.enable = 1;
115 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, msr.as_uint64);
116 }
117
118 return hyperv_init_ghcb();
119}
120
121static void (*hv_reenlightenment_cb)(void);
122
123static void hv_reenlightenment_notify(struct work_struct *dummy)
124{
125 struct hv_tsc_emulation_status emu_status;
126
127 rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
128
129 /* Don't issue the callback if TSC accesses are not emulated */
130 if (hv_reenlightenment_cb && emu_status.inprogress)
131 hv_reenlightenment_cb();
132}
133static DECLARE_DELAYED_WORK(hv_reenlightenment_work, hv_reenlightenment_notify);
134
135void hyperv_stop_tsc_emulation(void)
136{
137 u64 freq;
138 struct hv_tsc_emulation_status emu_status;
139
140 rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
141 emu_status.inprogress = 0;
142 wrmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
143
144 rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
145 tsc_khz = div64_u64(freq, 1000);
146}
147EXPORT_SYMBOL_GPL(hyperv_stop_tsc_emulation);
148
149static inline bool hv_reenlightenment_available(void)
150{
151 /*
152 * Check for required features and privileges to make TSC frequency
153 * change notifications work.
154 */
155 return ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
156 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE &&
157 ms_hyperv.features & HV_ACCESS_REENLIGHTENMENT;
158}
159
160DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_reenlightenment)
161{
162 ack_APIC_irq();
163 inc_irq_stat(irq_hv_reenlightenment_count);
164 schedule_delayed_work(&hv_reenlightenment_work, HZ/10);
165}
166
167void set_hv_tscchange_cb(void (*cb)(void))
168{
169 struct hv_reenlightenment_control re_ctrl = {
170 .vector = HYPERV_REENLIGHTENMENT_VECTOR,
171 .enabled = 1,
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 if (!hv_vp_index)
181 return;
182
183 hv_reenlightenment_cb = cb;
184
185 /* Make sure callback is registered before we write to MSRs */
186 wmb();
187
188 re_ctrl.target_vp = hv_vp_index[get_cpu()];
189
190 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
191 wrmsrl(HV_X64_MSR_TSC_EMULATION_CONTROL, *((u64 *)&emu_ctrl));
192
193 put_cpu();
194}
195EXPORT_SYMBOL_GPL(set_hv_tscchange_cb);
196
197void clear_hv_tscchange_cb(void)
198{
199 struct hv_reenlightenment_control re_ctrl;
200
201 if (!hv_reenlightenment_available())
202 return;
203
204 rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
205 re_ctrl.enabled = 0;
206 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
207
208 hv_reenlightenment_cb = NULL;
209}
210EXPORT_SYMBOL_GPL(clear_hv_tscchange_cb);
211
212static int hv_cpu_die(unsigned int cpu)
213{
214 struct hv_reenlightenment_control re_ctrl;
215 unsigned int new_cpu;
216 void **ghcb_va;
217
218 if (hv_ghcb_pg) {
219 ghcb_va = (void **)this_cpu_ptr(hv_ghcb_pg);
220 if (*ghcb_va)
221 memunmap(*ghcb_va);
222 *ghcb_va = NULL;
223 }
224
225 hv_common_cpu_die(cpu);
226
227 if (hv_vp_assist_page && hv_vp_assist_page[cpu]) {
228 union hv_vp_assist_msr_contents msr = { 0 };
229 if (hv_root_partition) {
230 /*
231 * For root partition the VP assist page is mapped to
232 * hypervisor provided page, and thus we unmap the
233 * page here and nullify it, so that in future we have
234 * correct page address mapped in hv_cpu_init.
235 */
236 memunmap(hv_vp_assist_page[cpu]);
237 hv_vp_assist_page[cpu] = NULL;
238 rdmsrl(HV_X64_MSR_VP_ASSIST_PAGE, msr.as_uint64);
239 msr.enable = 0;
240 }
241 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, msr.as_uint64);
242 }
243
244 if (hv_reenlightenment_cb == NULL)
245 return 0;
246
247 rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
248 if (re_ctrl.target_vp == hv_vp_index[cpu]) {
249 /*
250 * Reassign reenlightenment notifications to some other online
251 * CPU or just disable the feature if there are no online CPUs
252 * left (happens on hibernation).
253 */
254 new_cpu = cpumask_any_but(cpu_online_mask, cpu);
255
256 if (new_cpu < nr_cpu_ids)
257 re_ctrl.target_vp = hv_vp_index[new_cpu];
258 else
259 re_ctrl.enabled = 0;
260
261 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
262 }
263
264 return 0;
265}
266
267static int __init hv_pci_init(void)
268{
269 int gen2vm = efi_enabled(EFI_BOOT);
270
271 /*
272 * For Generation-2 VM, we exit from pci_arch_init() by returning 0.
273 * The purpose is to suppress the harmless warning:
274 * "PCI: Fatal: No config space access function found"
275 */
276 if (gen2vm)
277 return 0;
278
279 /* For Generation-1 VM, we'll proceed in pci_arch_init(). */
280 return 1;
281}
282
283static int hv_suspend(void)
284{
285 union hv_x64_msr_hypercall_contents hypercall_msr;
286 int ret;
287
288 if (hv_root_partition)
289 return -EPERM;
290
291 /*
292 * Reset the hypercall page as it is going to be invalidated
293 * across hibernation. Setting hv_hypercall_pg to NULL ensures
294 * that any subsequent hypercall operation fails safely instead of
295 * crashing due to an access of an invalid page. The hypercall page
296 * pointer is restored on resume.
297 */
298 hv_hypercall_pg_saved = hv_hypercall_pg;
299 hv_hypercall_pg = NULL;
300
301 /* Disable the hypercall page in the hypervisor */
302 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
303 hypercall_msr.enable = 0;
304 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
305
306 ret = hv_cpu_die(0);
307 return ret;
308}
309
310static void hv_resume(void)
311{
312 union hv_x64_msr_hypercall_contents hypercall_msr;
313 int ret;
314
315 ret = hv_cpu_init(0);
316 WARN_ON(ret);
317
318 /* Re-enable the hypercall page */
319 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
320 hypercall_msr.enable = 1;
321 hypercall_msr.guest_physical_address =
322 vmalloc_to_pfn(hv_hypercall_pg_saved);
323 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
324
325 hv_hypercall_pg = hv_hypercall_pg_saved;
326 hv_hypercall_pg_saved = NULL;
327
328 /*
329 * Reenlightenment notifications are disabled by hv_cpu_die(0),
330 * reenable them here if hv_reenlightenment_cb was previously set.
331 */
332 if (hv_reenlightenment_cb)
333 set_hv_tscchange_cb(hv_reenlightenment_cb);
334}
335
336/* Note: when the ops are called, only CPU0 is online and IRQs are disabled. */
337static struct syscore_ops hv_syscore_ops = {
338 .suspend = hv_suspend,
339 .resume = hv_resume,
340};
341
342static void (* __initdata old_setup_percpu_clockev)(void);
343
344static void __init hv_stimer_setup_percpu_clockev(void)
345{
346 /*
347 * Ignore any errors in setting up stimer clockevents
348 * as we can run with the LAPIC timer as a fallback.
349 */
350 (void)hv_stimer_alloc(false);
351
352 /*
353 * Still register the LAPIC timer, because the direct-mode STIMER is
354 * not supported by old versions of Hyper-V. This also allows users
355 * to switch to LAPIC timer via /sys, if they want to.
356 */
357 if (old_setup_percpu_clockev)
358 old_setup_percpu_clockev();
359}
360
361static void __init hv_get_partition_id(void)
362{
363 struct hv_get_partition_id *output_page;
364 u64 status;
365 unsigned long flags;
366
367 local_irq_save(flags);
368 output_page = *this_cpu_ptr(hyperv_pcpu_output_arg);
369 status = hv_do_hypercall(HVCALL_GET_PARTITION_ID, NULL, output_page);
370 if (!hv_result_success(status)) {
371 /* No point in proceeding if this failed */
372 pr_err("Failed to get partition ID: %lld\n", status);
373 BUG();
374 }
375 hv_current_partition_id = output_page->partition_id;
376 local_irq_restore(flags);
377}
378
379/*
380 * This function is to be invoked early in the boot sequence after the
381 * hypervisor has been detected.
382 *
383 * 1. Setup the hypercall page.
384 * 2. Register Hyper-V specific clocksource.
385 * 3. Setup Hyper-V specific APIC entry points.
386 */
387void __init hyperv_init(void)
388{
389 u64 guest_id;
390 union hv_x64_msr_hypercall_contents hypercall_msr;
391 int cpuhp;
392
393 if (x86_hyper_type != X86_HYPER_MS_HYPERV)
394 return;
395
396 if (hv_common_init())
397 return;
398
399 hv_vp_assist_page = kcalloc(num_possible_cpus(),
400 sizeof(*hv_vp_assist_page), GFP_KERNEL);
401 if (!hv_vp_assist_page) {
402 ms_hyperv.hints &= ~HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
403 goto common_free;
404 }
405
406 if (hv_isolation_type_snp()) {
407 /* Negotiate GHCB Version. */
408 if (!hv_ghcb_negotiate_protocol())
409 hv_ghcb_terminate(SEV_TERM_SET_GEN,
410 GHCB_SEV_ES_PROT_UNSUPPORTED);
411
412 hv_ghcb_pg = alloc_percpu(union hv_ghcb *);
413 if (!hv_ghcb_pg)
414 goto free_vp_assist_page;
415 }
416
417 cpuhp = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/hyperv_init:online",
418 hv_cpu_init, hv_cpu_die);
419 if (cpuhp < 0)
420 goto free_ghcb_page;
421
422 /*
423 * Setup the hypercall page and enable hypercalls.
424 * 1. Register the guest ID
425 * 2. Enable the hypercall and register the hypercall page
426 */
427 guest_id = hv_generate_guest_id(LINUX_VERSION_CODE);
428 wrmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
429
430 /* Hyper-V requires to write guest os id via ghcb in SNP IVM. */
431 hv_ghcb_msr_write(HV_X64_MSR_GUEST_OS_ID, guest_id);
432
433 hv_hypercall_pg = __vmalloc_node_range(PAGE_SIZE, 1, VMALLOC_START,
434 VMALLOC_END, GFP_KERNEL, PAGE_KERNEL_ROX,
435 VM_FLUSH_RESET_PERMS, NUMA_NO_NODE,
436 __builtin_return_address(0));
437 if (hv_hypercall_pg == NULL)
438 goto clean_guest_os_id;
439
440 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
441 hypercall_msr.enable = 1;
442
443 if (hv_root_partition) {
444 struct page *pg;
445 void *src;
446
447 /*
448 * For the root partition, the hypervisor will set up its
449 * hypercall page. The hypervisor guarantees it will not show
450 * up in the root's address space. The root can't change the
451 * location of the hypercall page.
452 *
453 * Order is important here. We must enable the hypercall page
454 * so it is populated with code, then copy the code to an
455 * executable page.
456 */
457 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
458
459 pg = vmalloc_to_page(hv_hypercall_pg);
460 src = memremap(hypercall_msr.guest_physical_address << PAGE_SHIFT, PAGE_SIZE,
461 MEMREMAP_WB);
462 BUG_ON(!src);
463 memcpy_to_page(pg, 0, src, HV_HYP_PAGE_SIZE);
464 memunmap(src);
465
466 hv_remap_tsc_clocksource();
467 } else {
468 hypercall_msr.guest_physical_address = vmalloc_to_pfn(hv_hypercall_pg);
469 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
470 }
471
472 /*
473 * hyperv_init() is called before LAPIC is initialized: see
474 * apic_intr_mode_init() -> x86_platform.apic_post_init() and
475 * apic_bsp_setup() -> setup_local_APIC(). The direct-mode STIMER
476 * depends on LAPIC, so hv_stimer_alloc() should be called from
477 * x86_init.timers.setup_percpu_clockev.
478 */
479 old_setup_percpu_clockev = x86_init.timers.setup_percpu_clockev;
480 x86_init.timers.setup_percpu_clockev = hv_stimer_setup_percpu_clockev;
481
482 hv_apic_init();
483
484 x86_init.pci.arch_init = hv_pci_init;
485
486 register_syscore_ops(&hv_syscore_ops);
487
488 hyperv_init_cpuhp = cpuhp;
489
490 if (cpuid_ebx(HYPERV_CPUID_FEATURES) & HV_ACCESS_PARTITION_ID)
491 hv_get_partition_id();
492
493 BUG_ON(hv_root_partition && hv_current_partition_id == ~0ull);
494
495#ifdef CONFIG_PCI_MSI
496 /*
497 * If we're running as root, we want to create our own PCI MSI domain.
498 * We can't set this in hv_pci_init because that would be too late.
499 */
500 if (hv_root_partition)
501 x86_init.irqs.create_pci_msi_domain = hv_create_pci_msi_domain;
502#endif
503
504 /* Query the VMs extended capability once, so that it can be cached. */
505 hv_query_ext_cap(0);
506
507#ifdef CONFIG_SWIOTLB
508 /*
509 * Swiotlb bounce buffer needs to be mapped in extra address
510 * space. Map function doesn't work in the early place and so
511 * call swiotlb_update_mem_attributes() here.
512 */
513 if (hv_is_isolation_supported())
514 swiotlb_update_mem_attributes();
515#endif
516
517 return;
518
519clean_guest_os_id:
520 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
521 hv_ghcb_msr_write(HV_X64_MSR_GUEST_OS_ID, 0);
522 cpuhp_remove_state(cpuhp);
523free_ghcb_page:
524 free_percpu(hv_ghcb_pg);
525free_vp_assist_page:
526 kfree(hv_vp_assist_page);
527 hv_vp_assist_page = NULL;
528common_free:
529 hv_common_free();
530}
531
532/*
533 * This routine is called before kexec/kdump, it does the required cleanup.
534 */
535void hyperv_cleanup(void)
536{
537 union hv_x64_msr_hypercall_contents hypercall_msr;
538 union hv_reference_tsc_msr tsc_msr;
539
540 /* Reset our OS id */
541 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
542 hv_ghcb_msr_write(HV_X64_MSR_GUEST_OS_ID, 0);
543
544 /*
545 * Reset hypercall page reference before reset the page,
546 * let hypercall operations fail safely rather than
547 * panic the kernel for using invalid hypercall page
548 */
549 hv_hypercall_pg = NULL;
550
551 /* Reset the hypercall page */
552 hypercall_msr.as_uint64 = hv_get_register(HV_X64_MSR_HYPERCALL);
553 hypercall_msr.enable = 0;
554 hv_set_register(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
555
556 /* Reset the TSC page */
557 tsc_msr.as_uint64 = hv_get_register(HV_X64_MSR_REFERENCE_TSC);
558 tsc_msr.enable = 0;
559 hv_set_register(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
560}
561
562void hyperv_report_panic(struct pt_regs *regs, long err, bool in_die)
563{
564 static bool panic_reported;
565 u64 guest_id;
566
567 if (in_die && !panic_on_oops)
568 return;
569
570 /*
571 * We prefer to report panic on 'die' chain as we have proper
572 * registers to report, but if we miss it (e.g. on BUG()) we need
573 * to report it on 'panic'.
574 */
575 if (panic_reported)
576 return;
577 panic_reported = true;
578
579 rdmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
580
581 wrmsrl(HV_X64_MSR_CRASH_P0, err);
582 wrmsrl(HV_X64_MSR_CRASH_P1, guest_id);
583 wrmsrl(HV_X64_MSR_CRASH_P2, regs->ip);
584 wrmsrl(HV_X64_MSR_CRASH_P3, regs->ax);
585 wrmsrl(HV_X64_MSR_CRASH_P4, regs->sp);
586
587 /*
588 * Let Hyper-V know there is crash data available
589 */
590 wrmsrl(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY);
591}
592EXPORT_SYMBOL_GPL(hyperv_report_panic);
593
594bool hv_is_hyperv_initialized(void)
595{
596 union hv_x64_msr_hypercall_contents hypercall_msr;
597
598 /*
599 * Ensure that we're really on Hyper-V, and not a KVM or Xen
600 * emulation of Hyper-V
601 */
602 if (x86_hyper_type != X86_HYPER_MS_HYPERV)
603 return false;
604
605 /*
606 * Verify that earlier initialization succeeded by checking
607 * that the hypercall page is setup
608 */
609 hypercall_msr.as_uint64 = 0;
610 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
611
612 return hypercall_msr.enable;
613}
614EXPORT_SYMBOL_GPL(hv_is_hyperv_initialized);
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 <linux/bitfield.h>
14#include <asm/apic.h>
15#include <asm/desc.h>
16#include <asm/hypervisor.h>
17#include <asm/hyperv-tlfs.h>
18#include <asm/mshyperv.h>
19#include <asm/idtentry.h>
20#include <linux/kexec.h>
21#include <linux/version.h>
22#include <linux/vmalloc.h>
23#include <linux/mm.h>
24#include <linux/hyperv.h>
25#include <linux/slab.h>
26#include <linux/kernel.h>
27#include <linux/cpuhotplug.h>
28#include <linux/syscore_ops.h>
29#include <clocksource/hyperv_timer.h>
30#include <linux/highmem.h>
31
32int hyperv_init_cpuhp;
33u64 hv_current_partition_id = ~0ull;
34EXPORT_SYMBOL_GPL(hv_current_partition_id);
35
36void *hv_hypercall_pg;
37EXPORT_SYMBOL_GPL(hv_hypercall_pg);
38
39/* Storage to save the hypercall page temporarily for hibernation */
40static void *hv_hypercall_pg_saved;
41
42u32 *hv_vp_index;
43EXPORT_SYMBOL_GPL(hv_vp_index);
44
45struct hv_vp_assist_page **hv_vp_assist_page;
46EXPORT_SYMBOL_GPL(hv_vp_assist_page);
47
48void __percpu **hyperv_pcpu_input_arg;
49EXPORT_SYMBOL_GPL(hyperv_pcpu_input_arg);
50
51void __percpu **hyperv_pcpu_output_arg;
52EXPORT_SYMBOL_GPL(hyperv_pcpu_output_arg);
53
54u32 hv_max_vp_index;
55EXPORT_SYMBOL_GPL(hv_max_vp_index);
56
57static int hv_cpu_init(unsigned int cpu)
58{
59 u64 msr_vp_index;
60 struct hv_vp_assist_page **hvp = &hv_vp_assist_page[smp_processor_id()];
61 void **input_arg;
62 struct page *pg;
63
64 /* hv_cpu_init() can be called with IRQs disabled from hv_resume() */
65 pg = alloc_pages(irqs_disabled() ? GFP_ATOMIC : GFP_KERNEL, hv_root_partition ? 1 : 0);
66 if (unlikely(!pg))
67 return -ENOMEM;
68
69 input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
70 *input_arg = page_address(pg);
71 if (hv_root_partition) {
72 void **output_arg;
73
74 output_arg = (void **)this_cpu_ptr(hyperv_pcpu_output_arg);
75 *output_arg = page_address(pg + 1);
76 }
77
78 msr_vp_index = hv_get_register(HV_REGISTER_VP_INDEX);
79
80 hv_vp_index[smp_processor_id()] = msr_vp_index;
81
82 if (msr_vp_index > hv_max_vp_index)
83 hv_max_vp_index = msr_vp_index;
84
85 if (!hv_vp_assist_page)
86 return 0;
87
88 /*
89 * The VP ASSIST PAGE is an "overlay" page (see Hyper-V TLFS's Section
90 * 5.2.1 "GPA Overlay Pages"). Here it must be zeroed out to make sure
91 * we always write the EOI MSR in hv_apic_eoi_write() *after* the
92 * EOI optimization is disabled in hv_cpu_die(), otherwise a CPU may
93 * not be stopped in the case of CPU offlining and the VM will hang.
94 */
95 if (!*hvp) {
96 *hvp = __vmalloc(PAGE_SIZE, GFP_KERNEL | __GFP_ZERO);
97 }
98
99 if (*hvp) {
100 u64 val;
101
102 val = vmalloc_to_pfn(*hvp);
103 val = (val << HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT) |
104 HV_X64_MSR_VP_ASSIST_PAGE_ENABLE;
105
106 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, val);
107 }
108
109 return 0;
110}
111
112static void (*hv_reenlightenment_cb)(void);
113
114static void hv_reenlightenment_notify(struct work_struct *dummy)
115{
116 struct hv_tsc_emulation_status emu_status;
117
118 rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
119
120 /* Don't issue the callback if TSC accesses are not emulated */
121 if (hv_reenlightenment_cb && emu_status.inprogress)
122 hv_reenlightenment_cb();
123}
124static DECLARE_DELAYED_WORK(hv_reenlightenment_work, hv_reenlightenment_notify);
125
126void hyperv_stop_tsc_emulation(void)
127{
128 u64 freq;
129 struct hv_tsc_emulation_status emu_status;
130
131 rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
132 emu_status.inprogress = 0;
133 wrmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
134
135 rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
136 tsc_khz = div64_u64(freq, 1000);
137}
138EXPORT_SYMBOL_GPL(hyperv_stop_tsc_emulation);
139
140static inline bool hv_reenlightenment_available(void)
141{
142 /*
143 * Check for required features and privileges to make TSC frequency
144 * change notifications work.
145 */
146 return ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
147 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE &&
148 ms_hyperv.features & HV_ACCESS_REENLIGHTENMENT;
149}
150
151DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_reenlightenment)
152{
153 ack_APIC_irq();
154 inc_irq_stat(irq_hv_reenlightenment_count);
155 schedule_delayed_work(&hv_reenlightenment_work, HZ/10);
156}
157
158void set_hv_tscchange_cb(void (*cb)(void))
159{
160 struct hv_reenlightenment_control re_ctrl = {
161 .vector = HYPERV_REENLIGHTENMENT_VECTOR,
162 .enabled = 1,
163 .target_vp = hv_vp_index[smp_processor_id()]
164 };
165 struct hv_tsc_emulation_control emu_ctrl = {.enabled = 1};
166
167 if (!hv_reenlightenment_available()) {
168 pr_warn("Hyper-V: reenlightenment support is unavailable\n");
169 return;
170 }
171
172 hv_reenlightenment_cb = cb;
173
174 /* Make sure callback is registered before we write to MSRs */
175 wmb();
176
177 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
178 wrmsrl(HV_X64_MSR_TSC_EMULATION_CONTROL, *((u64 *)&emu_ctrl));
179}
180EXPORT_SYMBOL_GPL(set_hv_tscchange_cb);
181
182void clear_hv_tscchange_cb(void)
183{
184 struct hv_reenlightenment_control re_ctrl;
185
186 if (!hv_reenlightenment_available())
187 return;
188
189 rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
190 re_ctrl.enabled = 0;
191 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
192
193 hv_reenlightenment_cb = NULL;
194}
195EXPORT_SYMBOL_GPL(clear_hv_tscchange_cb);
196
197static int hv_cpu_die(unsigned int cpu)
198{
199 struct hv_reenlightenment_control re_ctrl;
200 unsigned int new_cpu;
201 unsigned long flags;
202 void **input_arg;
203 void *pg;
204
205 local_irq_save(flags);
206 input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
207 pg = *input_arg;
208 *input_arg = NULL;
209
210 if (hv_root_partition) {
211 void **output_arg;
212
213 output_arg = (void **)this_cpu_ptr(hyperv_pcpu_output_arg);
214 *output_arg = NULL;
215 }
216
217 local_irq_restore(flags);
218
219 free_pages((unsigned long)pg, hv_root_partition ? 1 : 0);
220
221 if (hv_vp_assist_page && hv_vp_assist_page[cpu])
222 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, 0);
223
224 if (hv_reenlightenment_cb == NULL)
225 return 0;
226
227 rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
228 if (re_ctrl.target_vp == hv_vp_index[cpu]) {
229 /*
230 * Reassign reenlightenment notifications to some other online
231 * CPU or just disable the feature if there are no online CPUs
232 * left (happens on hibernation).
233 */
234 new_cpu = cpumask_any_but(cpu_online_mask, cpu);
235
236 if (new_cpu < nr_cpu_ids)
237 re_ctrl.target_vp = hv_vp_index[new_cpu];
238 else
239 re_ctrl.enabled = 0;
240
241 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
242 }
243
244 return 0;
245}
246
247static int __init hv_pci_init(void)
248{
249 int gen2vm = efi_enabled(EFI_BOOT);
250
251 /*
252 * For Generation-2 VM, we exit from pci_arch_init() by returning 0.
253 * The purpose is to suppress the harmless warning:
254 * "PCI: Fatal: No config space access function found"
255 */
256 if (gen2vm)
257 return 0;
258
259 /* For Generation-1 VM, we'll proceed in pci_arch_init(). */
260 return 1;
261}
262
263static int hv_suspend(void)
264{
265 union hv_x64_msr_hypercall_contents hypercall_msr;
266 int ret;
267
268 if (hv_root_partition)
269 return -EPERM;
270
271 /*
272 * Reset the hypercall page as it is going to be invalidated
273 * across hibernation. Setting hv_hypercall_pg to NULL ensures
274 * that any subsequent hypercall operation fails safely instead of
275 * crashing due to an access of an invalid page. The hypercall page
276 * pointer is restored on resume.
277 */
278 hv_hypercall_pg_saved = hv_hypercall_pg;
279 hv_hypercall_pg = NULL;
280
281 /* Disable the hypercall page in the hypervisor */
282 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
283 hypercall_msr.enable = 0;
284 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
285
286 ret = hv_cpu_die(0);
287 return ret;
288}
289
290static void hv_resume(void)
291{
292 union hv_x64_msr_hypercall_contents hypercall_msr;
293 int ret;
294
295 ret = hv_cpu_init(0);
296 WARN_ON(ret);
297
298 /* Re-enable the hypercall page */
299 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
300 hypercall_msr.enable = 1;
301 hypercall_msr.guest_physical_address =
302 vmalloc_to_pfn(hv_hypercall_pg_saved);
303 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
304
305 hv_hypercall_pg = hv_hypercall_pg_saved;
306 hv_hypercall_pg_saved = NULL;
307
308 /*
309 * Reenlightenment notifications are disabled by hv_cpu_die(0),
310 * reenable them here if hv_reenlightenment_cb was previously set.
311 */
312 if (hv_reenlightenment_cb)
313 set_hv_tscchange_cb(hv_reenlightenment_cb);
314}
315
316/* Note: when the ops are called, only CPU0 is online and IRQs are disabled. */
317static struct syscore_ops hv_syscore_ops = {
318 .suspend = hv_suspend,
319 .resume = hv_resume,
320};
321
322static void (* __initdata old_setup_percpu_clockev)(void);
323
324static void __init hv_stimer_setup_percpu_clockev(void)
325{
326 /*
327 * Ignore any errors in setting up stimer clockevents
328 * as we can run with the LAPIC timer as a fallback.
329 */
330 (void)hv_stimer_alloc(false);
331
332 /*
333 * Still register the LAPIC timer, because the direct-mode STIMER is
334 * not supported by old versions of Hyper-V. This also allows users
335 * to switch to LAPIC timer via /sys, if they want to.
336 */
337 if (old_setup_percpu_clockev)
338 old_setup_percpu_clockev();
339}
340
341static void __init hv_get_partition_id(void)
342{
343 struct hv_get_partition_id *output_page;
344 u64 status;
345 unsigned long flags;
346
347 local_irq_save(flags);
348 output_page = *this_cpu_ptr(hyperv_pcpu_output_arg);
349 status = hv_do_hypercall(HVCALL_GET_PARTITION_ID, NULL, output_page);
350 if (!hv_result_success(status)) {
351 /* No point in proceeding if this failed */
352 pr_err("Failed to get partition ID: %lld\n", status);
353 BUG();
354 }
355 hv_current_partition_id = output_page->partition_id;
356 local_irq_restore(flags);
357}
358
359/*
360 * This function is to be invoked early in the boot sequence after the
361 * hypervisor has been detected.
362 *
363 * 1. Setup the hypercall page.
364 * 2. Register Hyper-V specific clocksource.
365 * 3. Setup Hyper-V specific APIC entry points.
366 */
367void __init hyperv_init(void)
368{
369 u64 guest_id, required_msrs;
370 union hv_x64_msr_hypercall_contents hypercall_msr;
371 int cpuhp, i;
372
373 if (x86_hyper_type != X86_HYPER_MS_HYPERV)
374 return;
375
376 /* Absolutely required MSRs */
377 required_msrs = HV_MSR_HYPERCALL_AVAILABLE |
378 HV_MSR_VP_INDEX_AVAILABLE;
379
380 if ((ms_hyperv.features & required_msrs) != required_msrs)
381 return;
382
383 /*
384 * Allocate the per-CPU state for the hypercall input arg.
385 * If this allocation fails, we will not be able to setup
386 * (per-CPU) hypercall input page and thus this failure is
387 * fatal on Hyper-V.
388 */
389 hyperv_pcpu_input_arg = alloc_percpu(void *);
390
391 BUG_ON(hyperv_pcpu_input_arg == NULL);
392
393 /* Allocate the per-CPU state for output arg for root */
394 if (hv_root_partition) {
395 hyperv_pcpu_output_arg = alloc_percpu(void *);
396 BUG_ON(hyperv_pcpu_output_arg == NULL);
397 }
398
399 /* Allocate percpu VP index */
400 hv_vp_index = kmalloc_array(num_possible_cpus(), sizeof(*hv_vp_index),
401 GFP_KERNEL);
402 if (!hv_vp_index)
403 return;
404
405 for (i = 0; i < num_possible_cpus(); i++)
406 hv_vp_index[i] = VP_INVAL;
407
408 hv_vp_assist_page = kcalloc(num_possible_cpus(),
409 sizeof(*hv_vp_assist_page), GFP_KERNEL);
410 if (!hv_vp_assist_page) {
411 ms_hyperv.hints &= ~HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
412 goto free_vp_index;
413 }
414
415 cpuhp = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/hyperv_init:online",
416 hv_cpu_init, hv_cpu_die);
417 if (cpuhp < 0)
418 goto free_vp_assist_page;
419
420 /*
421 * Setup the hypercall page and enable hypercalls.
422 * 1. Register the guest ID
423 * 2. Enable the hypercall and register the hypercall page
424 */
425 guest_id = generate_guest_id(0, LINUX_VERSION_CODE, 0);
426 wrmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
427
428 hv_hypercall_pg = __vmalloc_node_range(PAGE_SIZE, 1, VMALLOC_START,
429 VMALLOC_END, GFP_KERNEL, PAGE_KERNEL_ROX,
430 VM_FLUSH_RESET_PERMS, NUMA_NO_NODE,
431 __builtin_return_address(0));
432 if (hv_hypercall_pg == NULL) {
433 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
434 goto remove_cpuhp_state;
435 }
436
437 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
438 hypercall_msr.enable = 1;
439
440 if (hv_root_partition) {
441 struct page *pg;
442 void *src, *dst;
443
444 /*
445 * For the root partition, the hypervisor will set up its
446 * hypercall page. The hypervisor guarantees it will not show
447 * up in the root's address space. The root can't change the
448 * location of the hypercall page.
449 *
450 * Order is important here. We must enable the hypercall page
451 * so it is populated with code, then copy the code to an
452 * executable page.
453 */
454 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
455
456 pg = vmalloc_to_page(hv_hypercall_pg);
457 dst = kmap(pg);
458 src = memremap(hypercall_msr.guest_physical_address << PAGE_SHIFT, PAGE_SIZE,
459 MEMREMAP_WB);
460 BUG_ON(!(src && dst));
461 memcpy(dst, src, HV_HYP_PAGE_SIZE);
462 memunmap(src);
463 kunmap(pg);
464 } else {
465 hypercall_msr.guest_physical_address = vmalloc_to_pfn(hv_hypercall_pg);
466 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
467 }
468
469 /*
470 * hyperv_init() is called before LAPIC is initialized: see
471 * apic_intr_mode_init() -> x86_platform.apic_post_init() and
472 * apic_bsp_setup() -> setup_local_APIC(). The direct-mode STIMER
473 * depends on LAPIC, so hv_stimer_alloc() should be called from
474 * x86_init.timers.setup_percpu_clockev.
475 */
476 old_setup_percpu_clockev = x86_init.timers.setup_percpu_clockev;
477 x86_init.timers.setup_percpu_clockev = hv_stimer_setup_percpu_clockev;
478
479 hv_apic_init();
480
481 x86_init.pci.arch_init = hv_pci_init;
482
483 register_syscore_ops(&hv_syscore_ops);
484
485 hyperv_init_cpuhp = cpuhp;
486
487 if (cpuid_ebx(HYPERV_CPUID_FEATURES) & HV_ACCESS_PARTITION_ID)
488 hv_get_partition_id();
489
490 BUG_ON(hv_root_partition && hv_current_partition_id == ~0ull);
491
492#ifdef CONFIG_PCI_MSI
493 /*
494 * If we're running as root, we want to create our own PCI MSI domain.
495 * We can't set this in hv_pci_init because that would be too late.
496 */
497 if (hv_root_partition)
498 x86_init.irqs.create_pci_msi_domain = hv_create_pci_msi_domain;
499#endif
500
501 /* Query the VMs extended capability once, so that it can be cached. */
502 hv_query_ext_cap(0);
503 return;
504
505remove_cpuhp_state:
506 cpuhp_remove_state(cpuhp);
507free_vp_assist_page:
508 kfree(hv_vp_assist_page);
509 hv_vp_assist_page = NULL;
510free_vp_index:
511 kfree(hv_vp_index);
512 hv_vp_index = NULL;
513}
514
515/*
516 * This routine is called before kexec/kdump, it does the required cleanup.
517 */
518void hyperv_cleanup(void)
519{
520 union hv_x64_msr_hypercall_contents hypercall_msr;
521
522 unregister_syscore_ops(&hv_syscore_ops);
523
524 /* Reset our OS id */
525 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
526
527 /*
528 * Reset hypercall page reference before reset the page,
529 * let hypercall operations fail safely rather than
530 * panic the kernel for using invalid hypercall page
531 */
532 hv_hypercall_pg = NULL;
533
534 /* Reset the hypercall page */
535 hypercall_msr.as_uint64 = 0;
536 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
537
538 /* Reset the TSC page */
539 hypercall_msr.as_uint64 = 0;
540 wrmsrl(HV_X64_MSR_REFERENCE_TSC, hypercall_msr.as_uint64);
541}
542EXPORT_SYMBOL_GPL(hyperv_cleanup);
543
544void hyperv_report_panic(struct pt_regs *regs, long err, bool in_die)
545{
546 static bool panic_reported;
547 u64 guest_id;
548
549 if (in_die && !panic_on_oops)
550 return;
551
552 /*
553 * We prefer to report panic on 'die' chain as we have proper
554 * registers to report, but if we miss it (e.g. on BUG()) we need
555 * to report it on 'panic'.
556 */
557 if (panic_reported)
558 return;
559 panic_reported = true;
560
561 rdmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
562
563 wrmsrl(HV_X64_MSR_CRASH_P0, err);
564 wrmsrl(HV_X64_MSR_CRASH_P1, guest_id);
565 wrmsrl(HV_X64_MSR_CRASH_P2, regs->ip);
566 wrmsrl(HV_X64_MSR_CRASH_P3, regs->ax);
567 wrmsrl(HV_X64_MSR_CRASH_P4, regs->sp);
568
569 /*
570 * Let Hyper-V know there is crash data available
571 */
572 wrmsrl(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY);
573}
574EXPORT_SYMBOL_GPL(hyperv_report_panic);
575
576bool hv_is_hyperv_initialized(void)
577{
578 union hv_x64_msr_hypercall_contents hypercall_msr;
579
580 /*
581 * Ensure that we're really on Hyper-V, and not a KVM or Xen
582 * emulation of Hyper-V
583 */
584 if (x86_hyper_type != X86_HYPER_MS_HYPERV)
585 return false;
586
587 /*
588 * Verify that earlier initialization succeeded by checking
589 * that the hypercall page is setup
590 */
591 hypercall_msr.as_uint64 = 0;
592 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
593
594 return hypercall_msr.enable;
595}
596EXPORT_SYMBOL_GPL(hv_is_hyperv_initialized);
597
598bool hv_is_hibernation_supported(void)
599{
600 return !hv_root_partition && acpi_sleep_state_supported(ACPI_STATE_S4);
601}
602EXPORT_SYMBOL_GPL(hv_is_hibernation_supported);
603
604enum hv_isolation_type hv_get_isolation_type(void)
605{
606 if (!(ms_hyperv.priv_high & HV_ISOLATION))
607 return HV_ISOLATION_TYPE_NONE;
608 return FIELD_GET(HV_ISOLATION_TYPE, ms_hyperv.isolation_config_b);
609}
610EXPORT_SYMBOL_GPL(hv_get_isolation_type);
611
612bool hv_is_isolation_supported(void)
613{
614 return hv_get_isolation_type() != HV_ISOLATION_TYPE_NONE;
615}
616EXPORT_SYMBOL_GPL(hv_is_isolation_supported);