Merge branch 'core-rcu-2021.07.04' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux-2.6-microblaze.git] / arch / x86 / hyperv / hv_init.c
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
32 int hyperv_init_cpuhp;
33 u64 hv_current_partition_id = ~0ull;
34 EXPORT_SYMBOL_GPL(hv_current_partition_id);
35
36 void *hv_hypercall_pg;
37 EXPORT_SYMBOL_GPL(hv_hypercall_pg);
38
39 /* Storage to save the hypercall page temporarily for hibernation */
40 static void *hv_hypercall_pg_saved;
41
42 u32 *hv_vp_index;
43 EXPORT_SYMBOL_GPL(hv_vp_index);
44
45 struct hv_vp_assist_page **hv_vp_assist_page;
46 EXPORT_SYMBOL_GPL(hv_vp_assist_page);
47
48 void  __percpu **hyperv_pcpu_input_arg;
49 EXPORT_SYMBOL_GPL(hyperv_pcpu_input_arg);
50
51 void  __percpu **hyperv_pcpu_output_arg;
52 EXPORT_SYMBOL_GPL(hyperv_pcpu_output_arg);
53
54 u32 hv_max_vp_index;
55 EXPORT_SYMBOL_GPL(hv_max_vp_index);
56
57 static 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
112 static void (*hv_reenlightenment_cb)(void);
113
114 static 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 }
124 static DECLARE_DELAYED_WORK(hv_reenlightenment_work, hv_reenlightenment_notify);
125
126 void 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 }
138 EXPORT_SYMBOL_GPL(hyperv_stop_tsc_emulation);
139
140 static 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
151 DEFINE_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
158 void 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 }
180 EXPORT_SYMBOL_GPL(set_hv_tscchange_cb);
181
182 void 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 }
195 EXPORT_SYMBOL_GPL(clear_hv_tscchange_cb);
196
197 static 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
247 static 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
263 static 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
290 static 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. */
317 static struct syscore_ops hv_syscore_ops = {
318         .suspend        = hv_suspend,
319         .resume         = hv_resume,
320 };
321
322 static void (* __initdata old_setup_percpu_clockev)(void);
323
324 static 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
341 static 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  */
367 void __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
505 remove_cpuhp_state:
506         cpuhp_remove_state(cpuhp);
507 free_vp_assist_page:
508         kfree(hv_vp_assist_page);
509         hv_vp_assist_page = NULL;
510 free_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  */
518 void 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 }
542 EXPORT_SYMBOL_GPL(hyperv_cleanup);
543
544 void 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 }
574 EXPORT_SYMBOL_GPL(hyperv_report_panic);
575
576 bool 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 }
596 EXPORT_SYMBOL_GPL(hv_is_hyperv_initialized);
597
598 bool hv_is_hibernation_supported(void)
599 {
600         return !hv_root_partition && acpi_sleep_state_supported(ACPI_STATE_S4);
601 }
602 EXPORT_SYMBOL_GPL(hv_is_hibernation_supported);
603
604 enum 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 }
610 EXPORT_SYMBOL_GPL(hv_get_isolation_type);
611
612 bool hv_is_isolation_supported(void)
613 {
614         return hv_get_isolation_type() != HV_ISOLATION_TYPE_NONE;
615 }
616 EXPORT_SYMBOL_GPL(hv_is_isolation_supported);