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#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 <asm/hyperv-tlfs.h>
23#include <asm/mshyperv.h>
24#include <asm/idtentry.h>
25#include <asm/set_memory.h>
26#include <linux/kexec.h>
27#include <linux/version.h>
28#include <linux/vmalloc.h>
29#include <linux/mm.h>
30#include <linux/hyperv.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
38u64 hv_current_partition_id = ~0ull;
39EXPORT_SYMBOL_GPL(hv_current_partition_id);
40
41void *hv_hypercall_pg;
42EXPORT_SYMBOL_GPL(hv_hypercall_pg);
43
44union hv_ghcb * __percpu *hv_ghcb_pg;
45
46/* Storage to save the hypercall page temporarily for hibernation */
47static void *hv_hypercall_pg_saved;
48
49struct hv_vp_assist_page **hv_vp_assist_page;
50EXPORT_SYMBOL_GPL(hv_vp_assist_page);
51
52static int hyperv_init_ghcb(void)
53{
54 u64 ghcb_gpa;
55 void *ghcb_va;
56 void **ghcb_base;
57
58 if (!ms_hyperv.paravisor_present || !hv_isolation_type_snp())
59 return 0;
60
61 if (!hv_ghcb_pg)
62 return -EINVAL;
63
64 /*
65 * GHCB page is allocated by paravisor. The address
66 * returned by MSR_AMD64_SEV_ES_GHCB is above shared
67 * memory boundary and map it here.
68 */
69 rdmsrl(MSR_AMD64_SEV_ES_GHCB, ghcb_gpa);
70
71 /* Mask out vTOM bit. ioremap_cache() maps decrypted */
72 ghcb_gpa &= ~ms_hyperv.shared_gpa_boundary;
73 ghcb_va = (void *)ioremap_cache(ghcb_gpa, HV_HYP_PAGE_SIZE);
74 if (!ghcb_va)
75 return -ENOMEM;
76
77 ghcb_base = (void **)this_cpu_ptr(hv_ghcb_pg);
78 *ghcb_base = ghcb_va;
79
80 return 0;
81}
82
83static int hv_cpu_init(unsigned int cpu)
84{
85 union hv_vp_assist_msr_contents msr = { 0 };
86 struct hv_vp_assist_page **hvp;
87 int ret;
88
89 ret = hv_common_cpu_init(cpu);
90 if (ret)
91 return ret;
92
93 if (!hv_vp_assist_page)
94 return 0;
95
96 hvp = &hv_vp_assist_page[cpu];
97 if (hv_root_partition) {
98 /*
99 * For root partition we get the hypervisor provided VP assist
100 * page, instead of allocating a new page.
101 */
102 rdmsrl(HV_X64_MSR_VP_ASSIST_PAGE, msr.as_uint64);
103 *hvp = memremap(msr.pfn << HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT,
104 PAGE_SIZE, MEMREMAP_WB);
105 } else {
106 /*
107 * The VP assist page is an "overlay" page (see Hyper-V TLFS's
108 * Section 5.2.1 "GPA Overlay Pages"). Here it must be zeroed
109 * out to make sure we always write the EOI MSR in
110 * hv_apic_eoi_write() *after* the EOI optimization is disabled
111 * in hv_cpu_die(), otherwise a CPU may not be stopped in the
112 * case of CPU offlining and the VM will hang.
113 */
114 if (!*hvp) {
115 *hvp = __vmalloc(PAGE_SIZE, GFP_KERNEL | __GFP_ZERO);
116
117 /*
118 * Hyper-V should never specify a VM that is a Confidential
119 * VM and also running in the root partition. Root partition
120 * is blocked to run in Confidential VM. So only decrypt assist
121 * page in non-root partition here.
122 */
123 if (*hvp && !ms_hyperv.paravisor_present && hv_isolation_type_snp()) {
124 WARN_ON_ONCE(set_memory_decrypted((unsigned long)(*hvp), 1));
125 memset(*hvp, 0, PAGE_SIZE);
126 }
127 }
128
129 if (*hvp)
130 msr.pfn = vmalloc_to_pfn(*hvp);
131
132 }
133 if (!WARN_ON(!(*hvp))) {
134 msr.enable = 1;
135 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, msr.as_uint64);
136 }
137
138 return hyperv_init_ghcb();
139}
140
141static void (*hv_reenlightenment_cb)(void);
142
143static void hv_reenlightenment_notify(struct work_struct *dummy)
144{
145 struct hv_tsc_emulation_status emu_status;
146
147 rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
148
149 /* Don't issue the callback if TSC accesses are not emulated */
150 if (hv_reenlightenment_cb && emu_status.inprogress)
151 hv_reenlightenment_cb();
152}
153static DECLARE_DELAYED_WORK(hv_reenlightenment_work, hv_reenlightenment_notify);
154
155void hyperv_stop_tsc_emulation(void)
156{
157 u64 freq;
158 struct hv_tsc_emulation_status emu_status;
159
160 rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
161 emu_status.inprogress = 0;
162 wrmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
163
164 rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
165 tsc_khz = div64_u64(freq, 1000);
166}
167EXPORT_SYMBOL_GPL(hyperv_stop_tsc_emulation);
168
169static inline bool hv_reenlightenment_available(void)
170{
171 /*
172 * Check for required features and privileges to make TSC frequency
173 * change notifications work.
174 */
175 return ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
176 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE &&
177 ms_hyperv.features & HV_ACCESS_REENLIGHTENMENT;
178}
179
180DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_reenlightenment)
181{
182 apic_eoi();
183 inc_irq_stat(irq_hv_reenlightenment_count);
184 schedule_delayed_work(&hv_reenlightenment_work, HZ/10);
185}
186
187void set_hv_tscchange_cb(void (*cb)(void))
188{
189 struct hv_reenlightenment_control re_ctrl = {
190 .vector = HYPERV_REENLIGHTENMENT_VECTOR,
191 .enabled = 1,
192 };
193 struct hv_tsc_emulation_control emu_ctrl = {.enabled = 1};
194
195 if (!hv_reenlightenment_available()) {
196 pr_warn("reenlightenment support is unavailable\n");
197 return;
198 }
199
200 if (!hv_vp_index)
201 return;
202
203 hv_reenlightenment_cb = cb;
204
205 /* Make sure callback is registered before we write to MSRs */
206 wmb();
207
208 re_ctrl.target_vp = hv_vp_index[get_cpu()];
209
210 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
211 wrmsrl(HV_X64_MSR_TSC_EMULATION_CONTROL, *((u64 *)&emu_ctrl));
212
213 put_cpu();
214}
215EXPORT_SYMBOL_GPL(set_hv_tscchange_cb);
216
217void clear_hv_tscchange_cb(void)
218{
219 struct hv_reenlightenment_control re_ctrl;
220
221 if (!hv_reenlightenment_available())
222 return;
223
224 rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
225 re_ctrl.enabled = 0;
226 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
227
228 hv_reenlightenment_cb = NULL;
229}
230EXPORT_SYMBOL_GPL(clear_hv_tscchange_cb);
231
232static int hv_cpu_die(unsigned int cpu)
233{
234 struct hv_reenlightenment_control re_ctrl;
235 unsigned int new_cpu;
236 void **ghcb_va;
237
238 if (hv_ghcb_pg) {
239 ghcb_va = (void **)this_cpu_ptr(hv_ghcb_pg);
240 if (*ghcb_va)
241 iounmap(*ghcb_va);
242 *ghcb_va = NULL;
243 }
244
245 hv_common_cpu_die(cpu);
246
247 if (hv_vp_assist_page && hv_vp_assist_page[cpu]) {
248 union hv_vp_assist_msr_contents msr = { 0 };
249 if (hv_root_partition) {
250 /*
251 * For root partition the VP assist page is mapped to
252 * hypervisor provided page, and thus we unmap the
253 * page here and nullify it, so that in future we have
254 * correct page address mapped in hv_cpu_init.
255 */
256 memunmap(hv_vp_assist_page[cpu]);
257 hv_vp_assist_page[cpu] = NULL;
258 rdmsrl(HV_X64_MSR_VP_ASSIST_PAGE, msr.as_uint64);
259 msr.enable = 0;
260 }
261 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, msr.as_uint64);
262 }
263
264 if (hv_reenlightenment_cb == NULL)
265 return 0;
266
267 rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
268 if (re_ctrl.target_vp == hv_vp_index[cpu]) {
269 /*
270 * Reassign reenlightenment notifications to some other online
271 * CPU or just disable the feature if there are no online CPUs
272 * left (happens on hibernation).
273 */
274 new_cpu = cpumask_any_but(cpu_online_mask, cpu);
275
276 if (new_cpu < nr_cpu_ids)
277 re_ctrl.target_vp = hv_vp_index[new_cpu];
278 else
279 re_ctrl.enabled = 0;
280
281 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
282 }
283
284 return 0;
285}
286
287static int __init hv_pci_init(void)
288{
289 bool gen2vm = efi_enabled(EFI_BOOT);
290
291 /*
292 * A Generation-2 VM doesn't support legacy PCI/PCIe, so both
293 * raw_pci_ops and raw_pci_ext_ops are NULL, and pci_subsys_init() ->
294 * pcibios_init() doesn't call pcibios_resource_survey() ->
295 * e820__reserve_resources_late(); as a result, any emulated persistent
296 * memory of E820_TYPE_PRAM (12) via the kernel parameter
297 * memmap=nn[KMG]!ss is not added into iomem_resource and hence can't be
298 * detected by register_e820_pmem(). Fix this by directly calling
299 * e820__reserve_resources_late() here: e820__reserve_resources_late()
300 * depends on e820__reserve_resources(), which has been called earlier
301 * from setup_arch(). Note: e820__reserve_resources_late() also adds
302 * any memory of E820_TYPE_PMEM (7) into iomem_resource, and
303 * acpi_nfit_register_region() -> acpi_nfit_insert_resource() ->
304 * region_intersects() returns REGION_INTERSECTS, so the memory of
305 * E820_TYPE_PMEM won't get added twice.
306 *
307 * We return 0 here so that pci_arch_init() won't print the warning:
308 * "PCI: Fatal: No config space access function found"
309 */
310 if (gen2vm) {
311 e820__reserve_resources_late();
312 return 0;
313 }
314
315 /* For Generation-1 VM, we'll proceed in pci_arch_init(). */
316 return 1;
317}
318
319static int hv_suspend(void)
320{
321 union hv_x64_msr_hypercall_contents hypercall_msr;
322 int ret;
323
324 if (hv_root_partition)
325 return -EPERM;
326
327 /*
328 * Reset the hypercall page as it is going to be invalidated
329 * across hibernation. Setting hv_hypercall_pg to NULL ensures
330 * that any subsequent hypercall operation fails safely instead of
331 * crashing due to an access of an invalid page. The hypercall page
332 * pointer is restored on resume.
333 */
334 hv_hypercall_pg_saved = hv_hypercall_pg;
335 hv_hypercall_pg = NULL;
336
337 /* Disable the hypercall page in the hypervisor */
338 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
339 hypercall_msr.enable = 0;
340 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
341
342 ret = hv_cpu_die(0);
343 return ret;
344}
345
346static void hv_resume(void)
347{
348 union hv_x64_msr_hypercall_contents hypercall_msr;
349 int ret;
350
351 ret = hv_cpu_init(0);
352 WARN_ON(ret);
353
354 /* Re-enable the hypercall page */
355 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
356 hypercall_msr.enable = 1;
357 hypercall_msr.guest_physical_address =
358 vmalloc_to_pfn(hv_hypercall_pg_saved);
359 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
360
361 hv_hypercall_pg = hv_hypercall_pg_saved;
362 hv_hypercall_pg_saved = NULL;
363
364 /*
365 * Reenlightenment notifications are disabled by hv_cpu_die(0),
366 * reenable them here if hv_reenlightenment_cb was previously set.
367 */
368 if (hv_reenlightenment_cb)
369 set_hv_tscchange_cb(hv_reenlightenment_cb);
370}
371
372/* Note: when the ops are called, only CPU0 is online and IRQs are disabled. */
373static struct syscore_ops hv_syscore_ops = {
374 .suspend = hv_suspend,
375 .resume = hv_resume,
376};
377
378static void (* __initdata old_setup_percpu_clockev)(void);
379
380static void __init hv_stimer_setup_percpu_clockev(void)
381{
382 /*
383 * Ignore any errors in setting up stimer clockevents
384 * as we can run with the LAPIC timer as a fallback.
385 */
386 (void)hv_stimer_alloc(false);
387
388 /*
389 * Still register the LAPIC timer, because the direct-mode STIMER is
390 * not supported by old versions of Hyper-V. This also allows users
391 * to switch to LAPIC timer via /sys, if they want to.
392 */
393 if (old_setup_percpu_clockev)
394 old_setup_percpu_clockev();
395}
396
397static void __init hv_get_partition_id(void)
398{
399 struct hv_get_partition_id *output_page;
400 u64 status;
401 unsigned long flags;
402
403 local_irq_save(flags);
404 output_page = *this_cpu_ptr(hyperv_pcpu_output_arg);
405 status = hv_do_hypercall(HVCALL_GET_PARTITION_ID, NULL, output_page);
406 if (!hv_result_success(status)) {
407 /* No point in proceeding if this failed */
408 pr_err("Failed to get partition ID: %lld\n", status);
409 BUG();
410 }
411 hv_current_partition_id = output_page->partition_id;
412 local_irq_restore(flags);
413}
414
415#if IS_ENABLED(CONFIG_HYPERV_VTL_MODE)
416static u8 __init get_vtl(void)
417{
418 u64 control = HV_HYPERCALL_REP_COMP_1 | HVCALL_GET_VP_REGISTERS;
419 struct hv_get_vp_registers_input *input;
420 struct hv_get_vp_registers_output *output;
421 unsigned long flags;
422 u64 ret;
423
424 local_irq_save(flags);
425 input = *this_cpu_ptr(hyperv_pcpu_input_arg);
426 output = (struct hv_get_vp_registers_output *)input;
427
428 memset(input, 0, struct_size(input, element, 1));
429 input->header.partitionid = HV_PARTITION_ID_SELF;
430 input->header.vpindex = HV_VP_INDEX_SELF;
431 input->header.inputvtl = 0;
432 input->element[0].name0 = HV_X64_REGISTER_VSM_VP_STATUS;
433
434 ret = hv_do_hypercall(control, input, output);
435 if (hv_result_success(ret)) {
436 ret = output->as64.low & HV_X64_VTL_MASK;
437 } else {
438 pr_err("Failed to get VTL(error: %lld) exiting...\n", ret);
439 BUG();
440 }
441
442 local_irq_restore(flags);
443 return ret;
444}
445#else
446static inline u8 get_vtl(void) { return 0; }
447#endif
448
449/*
450 * This function is to be invoked early in the boot sequence after the
451 * hypervisor has been detected.
452 *
453 * 1. Setup the hypercall page.
454 * 2. Register Hyper-V specific clocksource.
455 * 3. Setup Hyper-V specific APIC entry points.
456 */
457void __init hyperv_init(void)
458{
459 u64 guest_id;
460 union hv_x64_msr_hypercall_contents hypercall_msr;
461 int cpuhp;
462
463 if (x86_hyper_type != X86_HYPER_MS_HYPERV)
464 return;
465
466 if (hv_common_init())
467 return;
468
469 /*
470 * The VP assist page is useless to a TDX guest: the only use we
471 * would have for it is lazy EOI, which can not be used with TDX.
472 */
473 if (hv_isolation_type_tdx())
474 hv_vp_assist_page = NULL;
475 else
476 hv_vp_assist_page = kcalloc(num_possible_cpus(),
477 sizeof(*hv_vp_assist_page),
478 GFP_KERNEL);
479 if (!hv_vp_assist_page) {
480 ms_hyperv.hints &= ~HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
481
482 if (!hv_isolation_type_tdx())
483 goto common_free;
484 }
485
486 if (ms_hyperv.paravisor_present && hv_isolation_type_snp()) {
487 /* Negotiate GHCB Version. */
488 if (!hv_ghcb_negotiate_protocol())
489 hv_ghcb_terminate(SEV_TERM_SET_GEN,
490 GHCB_SEV_ES_PROT_UNSUPPORTED);
491
492 hv_ghcb_pg = alloc_percpu(union hv_ghcb *);
493 if (!hv_ghcb_pg)
494 goto free_vp_assist_page;
495 }
496
497 cpuhp = cpuhp_setup_state(CPUHP_AP_HYPERV_ONLINE, "x86/hyperv_init:online",
498 hv_cpu_init, hv_cpu_die);
499 if (cpuhp < 0)
500 goto free_ghcb_page;
501
502 /*
503 * Setup the hypercall page and enable hypercalls.
504 * 1. Register the guest ID
505 * 2. Enable the hypercall and register the hypercall page
506 *
507 * A TDX VM with no paravisor only uses TDX GHCI rather than hv_hypercall_pg:
508 * when the hypercall input is a page, such a VM must pass a decrypted
509 * page to Hyper-V, e.g. hv_post_message() uses the per-CPU page
510 * hyperv_pcpu_input_arg, which is decrypted if no paravisor is present.
511 *
512 * A TDX VM with the paravisor uses hv_hypercall_pg for most hypercalls,
513 * which are handled by the paravisor and the VM must use an encrypted
514 * input page: in such a VM, the hyperv_pcpu_input_arg is encrypted and
515 * used in the hypercalls, e.g. see hv_mark_gpa_visibility() and
516 * hv_arch_irq_unmask(). Such a VM uses TDX GHCI for two hypercalls:
517 * 1. HVCALL_SIGNAL_EVENT: see vmbus_set_event() and _hv_do_fast_hypercall8().
518 * 2. HVCALL_POST_MESSAGE: the input page must be a decrypted page, i.e.
519 * hv_post_message() in such a VM can't use the encrypted hyperv_pcpu_input_arg;
520 * instead, hv_post_message() uses the post_msg_page, which is decrypted
521 * in such a VM and is only used in such a VM.
522 */
523 guest_id = hv_generate_guest_id(LINUX_VERSION_CODE);
524 wrmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
525
526 /* With the paravisor, the VM must also write the ID via GHCB/GHCI */
527 hv_ivm_msr_write(HV_X64_MSR_GUEST_OS_ID, guest_id);
528
529 /* A TDX VM with no paravisor only uses TDX GHCI rather than hv_hypercall_pg */
530 if (hv_isolation_type_tdx() && !ms_hyperv.paravisor_present)
531 goto skip_hypercall_pg_init;
532
533 hv_hypercall_pg = __vmalloc_node_range(PAGE_SIZE, 1, VMALLOC_START,
534 VMALLOC_END, GFP_KERNEL, PAGE_KERNEL_ROX,
535 VM_FLUSH_RESET_PERMS, NUMA_NO_NODE,
536 __builtin_return_address(0));
537 if (hv_hypercall_pg == NULL)
538 goto clean_guest_os_id;
539
540 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
541 hypercall_msr.enable = 1;
542
543 if (hv_root_partition) {
544 struct page *pg;
545 void *src;
546
547 /*
548 * For the root partition, the hypervisor will set up its
549 * hypercall page. The hypervisor guarantees it will not show
550 * up in the root's address space. The root can't change the
551 * location of the hypercall page.
552 *
553 * Order is important here. We must enable the hypercall page
554 * so it is populated with code, then copy the code to an
555 * executable page.
556 */
557 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
558
559 pg = vmalloc_to_page(hv_hypercall_pg);
560 src = memremap(hypercall_msr.guest_physical_address << PAGE_SHIFT, PAGE_SIZE,
561 MEMREMAP_WB);
562 BUG_ON(!src);
563 memcpy_to_page(pg, 0, src, HV_HYP_PAGE_SIZE);
564 memunmap(src);
565
566 hv_remap_tsc_clocksource();
567 } else {
568 hypercall_msr.guest_physical_address = vmalloc_to_pfn(hv_hypercall_pg);
569 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
570 }
571
572skip_hypercall_pg_init:
573 /*
574 * Some versions of Hyper-V that provide IBT in guest VMs have a bug
575 * in that there's no ENDBR64 instruction at the entry to the
576 * hypercall page. Because hypercalls are invoked via an indirect call
577 * to the hypercall page, all hypercall attempts fail when IBT is
578 * enabled, and Linux panics. For such buggy versions, disable IBT.
579 *
580 * Fixed versions of Hyper-V always provide ENDBR64 on the hypercall
581 * page, so if future Linux kernel versions enable IBT for 32-bit
582 * builds, additional hypercall page hackery will be required here
583 * to provide an ENDBR32.
584 */
585#ifdef CONFIG_X86_KERNEL_IBT
586 if (cpu_feature_enabled(X86_FEATURE_IBT) &&
587 *(u32 *)hv_hypercall_pg != gen_endbr()) {
588 setup_clear_cpu_cap(X86_FEATURE_IBT);
589 pr_warn("Disabling IBT because of Hyper-V bug\n");
590 }
591#endif
592
593 /*
594 * hyperv_init() is called before LAPIC is initialized: see
595 * apic_intr_mode_init() -> x86_platform.apic_post_init() and
596 * apic_bsp_setup() -> setup_local_APIC(). The direct-mode STIMER
597 * depends on LAPIC, so hv_stimer_alloc() should be called from
598 * x86_init.timers.setup_percpu_clockev.
599 */
600 old_setup_percpu_clockev = x86_init.timers.setup_percpu_clockev;
601 x86_init.timers.setup_percpu_clockev = hv_stimer_setup_percpu_clockev;
602
603 hv_apic_init();
604
605 x86_init.pci.arch_init = hv_pci_init;
606
607 register_syscore_ops(&hv_syscore_ops);
608
609 if (cpuid_ebx(HYPERV_CPUID_FEATURES) & HV_ACCESS_PARTITION_ID)
610 hv_get_partition_id();
611
612 BUG_ON(hv_root_partition && hv_current_partition_id == ~0ull);
613
614#ifdef CONFIG_PCI_MSI
615 /*
616 * If we're running as root, we want to create our own PCI MSI domain.
617 * We can't set this in hv_pci_init because that would be too late.
618 */
619 if (hv_root_partition)
620 x86_init.irqs.create_pci_msi_domain = hv_create_pci_msi_domain;
621#endif
622
623 /* Query the VMs extended capability once, so that it can be cached. */
624 hv_query_ext_cap(0);
625
626 /* Find the VTL */
627 ms_hyperv.vtl = get_vtl();
628
629 if (ms_hyperv.vtl > 0) /* non default VTL */
630 hv_vtl_early_init();
631
632 return;
633
634clean_guest_os_id:
635 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
636 hv_ivm_msr_write(HV_X64_MSR_GUEST_OS_ID, 0);
637 cpuhp_remove_state(CPUHP_AP_HYPERV_ONLINE);
638free_ghcb_page:
639 free_percpu(hv_ghcb_pg);
640free_vp_assist_page:
641 kfree(hv_vp_assist_page);
642 hv_vp_assist_page = NULL;
643common_free:
644 hv_common_free();
645}
646
647/*
648 * This routine is called before kexec/kdump, it does the required cleanup.
649 */
650void hyperv_cleanup(void)
651{
652 union hv_x64_msr_hypercall_contents hypercall_msr;
653 union hv_reference_tsc_msr tsc_msr;
654
655 /* Reset our OS id */
656 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
657 hv_ivm_msr_write(HV_X64_MSR_GUEST_OS_ID, 0);
658
659 /*
660 * Reset hypercall page reference before reset the page,
661 * let hypercall operations fail safely rather than
662 * panic the kernel for using invalid hypercall page
663 */
664 hv_hypercall_pg = NULL;
665
666 /* Reset the hypercall page */
667 hypercall_msr.as_uint64 = hv_get_msr(HV_X64_MSR_HYPERCALL);
668 hypercall_msr.enable = 0;
669 hv_set_msr(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
670
671 /* Reset the TSC page */
672 tsc_msr.as_uint64 = hv_get_msr(HV_X64_MSR_REFERENCE_TSC);
673 tsc_msr.enable = 0;
674 hv_set_msr(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
675}
676
677void hyperv_report_panic(struct pt_regs *regs, long err, bool in_die)
678{
679 static bool panic_reported;
680 u64 guest_id;
681
682 if (in_die && !panic_on_oops)
683 return;
684
685 /*
686 * We prefer to report panic on 'die' chain as we have proper
687 * registers to report, but if we miss it (e.g. on BUG()) we need
688 * to report it on 'panic'.
689 */
690 if (panic_reported)
691 return;
692 panic_reported = true;
693
694 rdmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
695
696 wrmsrl(HV_X64_MSR_CRASH_P0, err);
697 wrmsrl(HV_X64_MSR_CRASH_P1, guest_id);
698 wrmsrl(HV_X64_MSR_CRASH_P2, regs->ip);
699 wrmsrl(HV_X64_MSR_CRASH_P3, regs->ax);
700 wrmsrl(HV_X64_MSR_CRASH_P4, regs->sp);
701
702 /*
703 * Let Hyper-V know there is crash data available
704 */
705 wrmsrl(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY);
706}
707EXPORT_SYMBOL_GPL(hyperv_report_panic);
708
709bool hv_is_hyperv_initialized(void)
710{
711 union hv_x64_msr_hypercall_contents hypercall_msr;
712
713 /*
714 * Ensure that we're really on Hyper-V, and not a KVM or Xen
715 * emulation of Hyper-V
716 */
717 if (x86_hyper_type != X86_HYPER_MS_HYPERV)
718 return false;
719
720 /* A TDX VM with no paravisor uses TDX GHCI call rather than hv_hypercall_pg */
721 if (hv_isolation_type_tdx() && !ms_hyperv.paravisor_present)
722 return true;
723 /*
724 * Verify that earlier initialization succeeded by checking
725 * that the hypercall page is setup
726 */
727 hypercall_msr.as_uint64 = 0;
728 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
729
730 return hypercall_msr.enable;
731}
732EXPORT_SYMBOL_GPL(hv_is_hyperv_initialized);
1/*
2 * X86 specific Hyper-V initialization code.
3 *
4 * Copyright (C) 2016, Microsoft, Inc.
5 *
6 * Author : K. Y. Srinivasan <kys@microsoft.com>
7 *
8 * This program is free software; you can redistribute it and/or modify it
9 * under the terms of the GNU General Public License version 2 as published
10 * by the Free Software Foundation.
11 *
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
15 * NON INFRINGEMENT. See the GNU General Public License for more
16 * details.
17 *
18 */
19
20#include <linux/types.h>
21#include <asm/apic.h>
22#include <asm/desc.h>
23#include <asm/hypervisor.h>
24#include <asm/hyperv-tlfs.h>
25#include <asm/mshyperv.h>
26#include <linux/version.h>
27#include <linux/vmalloc.h>
28#include <linux/mm.h>
29#include <linux/clockchips.h>
30#include <linux/hyperv.h>
31#include <linux/slab.h>
32#include <linux/cpuhotplug.h>
33
34#ifdef CONFIG_HYPERV_TSCPAGE
35
36static struct ms_hyperv_tsc_page *tsc_pg;
37
38struct ms_hyperv_tsc_page *hv_get_tsc_page(void)
39{
40 return tsc_pg;
41}
42EXPORT_SYMBOL_GPL(hv_get_tsc_page);
43
44static u64 read_hv_clock_tsc(struct clocksource *arg)
45{
46 u64 current_tick = hv_read_tsc_page(tsc_pg);
47
48 if (current_tick == U64_MAX)
49 rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
50
51 return current_tick;
52}
53
54static struct clocksource hyperv_cs_tsc = {
55 .name = "hyperv_clocksource_tsc_page",
56 .rating = 400,
57 .read = read_hv_clock_tsc,
58 .mask = CLOCKSOURCE_MASK(64),
59 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
60};
61#endif
62
63static u64 read_hv_clock_msr(struct clocksource *arg)
64{
65 u64 current_tick;
66 /*
67 * Read the partition counter to get the current tick count. This count
68 * is set to 0 when the partition is created and is incremented in
69 * 100 nanosecond units.
70 */
71 rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
72 return current_tick;
73}
74
75static struct clocksource hyperv_cs_msr = {
76 .name = "hyperv_clocksource_msr",
77 .rating = 400,
78 .read = read_hv_clock_msr,
79 .mask = CLOCKSOURCE_MASK(64),
80 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
81};
82
83void *hv_hypercall_pg;
84EXPORT_SYMBOL_GPL(hv_hypercall_pg);
85struct clocksource *hyperv_cs;
86EXPORT_SYMBOL_GPL(hyperv_cs);
87
88u32 *hv_vp_index;
89EXPORT_SYMBOL_GPL(hv_vp_index);
90
91struct hv_vp_assist_page **hv_vp_assist_page;
92EXPORT_SYMBOL_GPL(hv_vp_assist_page);
93
94u32 hv_max_vp_index;
95
96static int hv_cpu_init(unsigned int cpu)
97{
98 u64 msr_vp_index;
99 struct hv_vp_assist_page **hvp = &hv_vp_assist_page[smp_processor_id()];
100
101 hv_get_vp_index(msr_vp_index);
102
103 hv_vp_index[smp_processor_id()] = msr_vp_index;
104
105 if (msr_vp_index > hv_max_vp_index)
106 hv_max_vp_index = msr_vp_index;
107
108 if (!hv_vp_assist_page)
109 return 0;
110
111 if (!*hvp)
112 *hvp = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL);
113
114 if (*hvp) {
115 u64 val;
116
117 val = vmalloc_to_pfn(*hvp);
118 val = (val << HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT) |
119 HV_X64_MSR_VP_ASSIST_PAGE_ENABLE;
120
121 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, val);
122 }
123
124 return 0;
125}
126
127static void (*hv_reenlightenment_cb)(void);
128
129static void hv_reenlightenment_notify(struct work_struct *dummy)
130{
131 struct hv_tsc_emulation_status emu_status;
132
133 rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
134
135 /* Don't issue the callback if TSC accesses are not emulated */
136 if (hv_reenlightenment_cb && emu_status.inprogress)
137 hv_reenlightenment_cb();
138}
139static DECLARE_DELAYED_WORK(hv_reenlightenment_work, hv_reenlightenment_notify);
140
141void hyperv_stop_tsc_emulation(void)
142{
143 u64 freq;
144 struct hv_tsc_emulation_status emu_status;
145
146 rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
147 emu_status.inprogress = 0;
148 wrmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
149
150 rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
151 tsc_khz = div64_u64(freq, 1000);
152}
153EXPORT_SYMBOL_GPL(hyperv_stop_tsc_emulation);
154
155static inline bool hv_reenlightenment_available(void)
156{
157 /*
158 * Check for required features and priviliges to make TSC frequency
159 * change notifications work.
160 */
161 return ms_hyperv.features & HV_X64_ACCESS_FREQUENCY_MSRS &&
162 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE &&
163 ms_hyperv.features & HV_X64_ACCESS_REENLIGHTENMENT;
164}
165
166__visible void __irq_entry hyperv_reenlightenment_intr(struct pt_regs *regs)
167{
168 entering_ack_irq();
169
170 inc_irq_stat(irq_hv_reenlightenment_count);
171
172 schedule_delayed_work(&hv_reenlightenment_work, HZ/10);
173
174 exiting_irq();
175}
176
177void set_hv_tscchange_cb(void (*cb)(void))
178{
179 struct hv_reenlightenment_control re_ctrl = {
180 .vector = HYPERV_REENLIGHTENMENT_VECTOR,
181 .enabled = 1,
182 .target_vp = hv_vp_index[smp_processor_id()]
183 };
184 struct hv_tsc_emulation_control emu_ctrl = {.enabled = 1};
185
186 if (!hv_reenlightenment_available()) {
187 pr_warn("Hyper-V: reenlightenment support is unavailable\n");
188 return;
189 }
190
191 hv_reenlightenment_cb = cb;
192
193 /* Make sure callback is registered before we write to MSRs */
194 wmb();
195
196 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
197 wrmsrl(HV_X64_MSR_TSC_EMULATION_CONTROL, *((u64 *)&emu_ctrl));
198}
199EXPORT_SYMBOL_GPL(set_hv_tscchange_cb);
200
201void clear_hv_tscchange_cb(void)
202{
203 struct hv_reenlightenment_control re_ctrl;
204
205 if (!hv_reenlightenment_available())
206 return;
207
208 rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
209 re_ctrl.enabled = 0;
210 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
211
212 hv_reenlightenment_cb = NULL;
213}
214EXPORT_SYMBOL_GPL(clear_hv_tscchange_cb);
215
216static int hv_cpu_die(unsigned int cpu)
217{
218 struct hv_reenlightenment_control re_ctrl;
219 unsigned int new_cpu;
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 /* Reassign to some other online CPU */
230 new_cpu = cpumask_any_but(cpu_online_mask, cpu);
231
232 re_ctrl.target_vp = hv_vp_index[new_cpu];
233 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
234 }
235
236 return 0;
237}
238
239/*
240 * This function is to be invoked early in the boot sequence after the
241 * hypervisor has been detected.
242 *
243 * 1. Setup the hypercall page.
244 * 2. Register Hyper-V specific clocksource.
245 */
246void hyperv_init(void)
247{
248 u64 guest_id, required_msrs;
249 union hv_x64_msr_hypercall_contents hypercall_msr;
250 int cpuhp;
251
252 if (x86_hyper_type != X86_HYPER_MS_HYPERV)
253 return;
254
255 /* Absolutely required MSRs */
256 required_msrs = HV_X64_MSR_HYPERCALL_AVAILABLE |
257 HV_X64_MSR_VP_INDEX_AVAILABLE;
258
259 if ((ms_hyperv.features & required_msrs) != required_msrs)
260 return;
261
262 /* Allocate percpu VP index */
263 hv_vp_index = kmalloc_array(num_possible_cpus(), sizeof(*hv_vp_index),
264 GFP_KERNEL);
265 if (!hv_vp_index)
266 return;
267
268 hv_vp_assist_page = kcalloc(num_possible_cpus(),
269 sizeof(*hv_vp_assist_page), GFP_KERNEL);
270 if (!hv_vp_assist_page) {
271 ms_hyperv.hints &= ~HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
272 goto free_vp_index;
273 }
274
275 cpuhp = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/hyperv_init:online",
276 hv_cpu_init, hv_cpu_die);
277 if (cpuhp < 0)
278 goto free_vp_assist_page;
279
280 /*
281 * Setup the hypercall page and enable hypercalls.
282 * 1. Register the guest ID
283 * 2. Enable the hypercall and register the hypercall page
284 */
285 guest_id = generate_guest_id(0, LINUX_VERSION_CODE, 0);
286 wrmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
287
288 hv_hypercall_pg = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_RX);
289 if (hv_hypercall_pg == NULL) {
290 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
291 goto remove_cpuhp_state;
292 }
293
294 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
295 hypercall_msr.enable = 1;
296 hypercall_msr.guest_physical_address = vmalloc_to_pfn(hv_hypercall_pg);
297 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
298
299 hyper_alloc_mmu();
300
301 /*
302 * Register Hyper-V specific clocksource.
303 */
304#ifdef CONFIG_HYPERV_TSCPAGE
305 if (ms_hyperv.features & HV_X64_MSR_REFERENCE_TSC_AVAILABLE) {
306 union hv_x64_msr_hypercall_contents tsc_msr;
307
308 tsc_pg = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL);
309 if (!tsc_pg)
310 goto register_msr_cs;
311
312 hyperv_cs = &hyperv_cs_tsc;
313
314 rdmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
315
316 tsc_msr.enable = 1;
317 tsc_msr.guest_physical_address = vmalloc_to_pfn(tsc_pg);
318
319 wrmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
320
321 hyperv_cs_tsc.archdata.vclock_mode = VCLOCK_HVCLOCK;
322
323 clocksource_register_hz(&hyperv_cs_tsc, NSEC_PER_SEC/100);
324 return;
325 }
326register_msr_cs:
327#endif
328 /*
329 * For 32 bit guests just use the MSR based mechanism for reading
330 * the partition counter.
331 */
332
333 hyperv_cs = &hyperv_cs_msr;
334 if (ms_hyperv.features & HV_X64_MSR_TIME_REF_COUNT_AVAILABLE)
335 clocksource_register_hz(&hyperv_cs_msr, NSEC_PER_SEC/100);
336
337 return;
338
339remove_cpuhp_state:
340 cpuhp_remove_state(cpuhp);
341free_vp_assist_page:
342 kfree(hv_vp_assist_page);
343 hv_vp_assist_page = NULL;
344free_vp_index:
345 kfree(hv_vp_index);
346 hv_vp_index = NULL;
347}
348
349/*
350 * This routine is called before kexec/kdump, it does the required cleanup.
351 */
352void hyperv_cleanup(void)
353{
354 union hv_x64_msr_hypercall_contents hypercall_msr;
355
356 /* Reset our OS id */
357 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
358
359 /* Reset the hypercall page */
360 hypercall_msr.as_uint64 = 0;
361 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
362
363 /* Reset the TSC page */
364 hypercall_msr.as_uint64 = 0;
365 wrmsrl(HV_X64_MSR_REFERENCE_TSC, hypercall_msr.as_uint64);
366}
367EXPORT_SYMBOL_GPL(hyperv_cleanup);
368
369void hyperv_report_panic(struct pt_regs *regs, long err)
370{
371 static bool panic_reported;
372 u64 guest_id;
373
374 /*
375 * We prefer to report panic on 'die' chain as we have proper
376 * registers to report, but if we miss it (e.g. on BUG()) we need
377 * to report it on 'panic'.
378 */
379 if (panic_reported)
380 return;
381 panic_reported = true;
382
383 rdmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
384
385 wrmsrl(HV_X64_MSR_CRASH_P0, err);
386 wrmsrl(HV_X64_MSR_CRASH_P1, guest_id);
387 wrmsrl(HV_X64_MSR_CRASH_P2, regs->ip);
388 wrmsrl(HV_X64_MSR_CRASH_P3, regs->ax);
389 wrmsrl(HV_X64_MSR_CRASH_P4, regs->sp);
390
391 /*
392 * Let Hyper-V know there is crash data available
393 */
394 wrmsrl(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY);
395}
396EXPORT_SYMBOL_GPL(hyperv_report_panic);
397
398bool hv_is_hyperv_initialized(void)
399{
400 union hv_x64_msr_hypercall_contents hypercall_msr;
401
402 /*
403 * Ensure that we're really on Hyper-V, and not a KVM or Xen
404 * emulation of Hyper-V
405 */
406 if (x86_hyper_type != X86_HYPER_MS_HYPERV)
407 return false;
408
409 /*
410 * Verify that earlier initialization succeeded by checking
411 * that the hypercall page is setup
412 */
413 hypercall_msr.as_uint64 = 0;
414 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
415
416 return hypercall_msr.enable;
417}
418EXPORT_SYMBOL_GPL(hv_is_hyperv_initialized);