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