1 // SPDX-License-Identifier: GPL-2.0-only
3 * X86 specific Hyper-V initialization code.
5 * Copyright (C) 2016, Microsoft, Inc.
7 * Author : K. Y. Srinivasan <kys@microsoft.com>
10 #include <linux/efi.h>
11 #include <linux/types.h>
14 #include <asm/hypervisor.h>
15 #include <asm/hyperv-tlfs.h>
16 #include <asm/mshyperv.h>
17 #include <linux/version.h>
18 #include <linux/vmalloc.h>
20 #include <linux/clockchips.h>
21 #include <linux/hyperv.h>
22 #include <linux/slab.h>
23 #include <linux/cpuhotplug.h>
25 #ifdef CONFIG_HYPERV_TSCPAGE
27 static struct ms_hyperv_tsc_page *tsc_pg;
29 struct ms_hyperv_tsc_page *hv_get_tsc_page(void)
33 EXPORT_SYMBOL_GPL(hv_get_tsc_page);
35 static u64 read_hv_clock_tsc(struct clocksource *arg)
37 u64 current_tick = hv_read_tsc_page(tsc_pg);
39 if (current_tick == U64_MAX)
40 rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
45 static struct clocksource hyperv_cs_tsc = {
46 .name = "hyperv_clocksource_tsc_page",
48 .read = read_hv_clock_tsc,
49 .mask = CLOCKSOURCE_MASK(64),
50 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
54 static u64 read_hv_clock_msr(struct clocksource *arg)
58 * Read the partition counter to get the current tick count. This count
59 * is set to 0 when the partition is created and is incremented in
60 * 100 nanosecond units.
62 rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
66 static struct clocksource hyperv_cs_msr = {
67 .name = "hyperv_clocksource_msr",
69 .read = read_hv_clock_msr,
70 .mask = CLOCKSOURCE_MASK(64),
71 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
74 void *hv_hypercall_pg;
75 EXPORT_SYMBOL_GPL(hv_hypercall_pg);
76 struct clocksource *hyperv_cs;
77 EXPORT_SYMBOL_GPL(hyperv_cs);
80 EXPORT_SYMBOL_GPL(hv_vp_index);
82 struct hv_vp_assist_page **hv_vp_assist_page;
83 EXPORT_SYMBOL_GPL(hv_vp_assist_page);
85 void __percpu **hyperv_pcpu_input_arg;
86 EXPORT_SYMBOL_GPL(hyperv_pcpu_input_arg);
89 EXPORT_SYMBOL_GPL(hv_max_vp_index);
91 static int hv_cpu_init(unsigned int cpu)
94 struct hv_vp_assist_page **hvp = &hv_vp_assist_page[smp_processor_id()];
98 input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
99 pg = alloc_page(GFP_KERNEL);
102 *input_arg = page_address(pg);
104 hv_get_vp_index(msr_vp_index);
106 hv_vp_index[smp_processor_id()] = msr_vp_index;
108 if (msr_vp_index > hv_max_vp_index)
109 hv_max_vp_index = msr_vp_index;
111 if (!hv_vp_assist_page)
115 *hvp = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL);
120 val = vmalloc_to_pfn(*hvp);
121 val = (val << HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT) |
122 HV_X64_MSR_VP_ASSIST_PAGE_ENABLE;
124 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, val);
130 static void (*hv_reenlightenment_cb)(void);
132 static void hv_reenlightenment_notify(struct work_struct *dummy)
134 struct hv_tsc_emulation_status emu_status;
136 rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
138 /* Don't issue the callback if TSC accesses are not emulated */
139 if (hv_reenlightenment_cb && emu_status.inprogress)
140 hv_reenlightenment_cb();
142 static DECLARE_DELAYED_WORK(hv_reenlightenment_work, hv_reenlightenment_notify);
144 void hyperv_stop_tsc_emulation(void)
147 struct hv_tsc_emulation_status emu_status;
149 rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
150 emu_status.inprogress = 0;
151 wrmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
153 rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
154 tsc_khz = div64_u64(freq, 1000);
156 EXPORT_SYMBOL_GPL(hyperv_stop_tsc_emulation);
158 static inline bool hv_reenlightenment_available(void)
161 * Check for required features and priviliges to make TSC frequency
162 * change notifications work.
164 return ms_hyperv.features & HV_X64_ACCESS_FREQUENCY_MSRS &&
165 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE &&
166 ms_hyperv.features & HV_X64_ACCESS_REENLIGHTENMENT;
169 __visible void __irq_entry hyperv_reenlightenment_intr(struct pt_regs *regs)
173 inc_irq_stat(irq_hv_reenlightenment_count);
175 schedule_delayed_work(&hv_reenlightenment_work, HZ/10);
180 void set_hv_tscchange_cb(void (*cb)(void))
182 struct hv_reenlightenment_control re_ctrl = {
183 .vector = HYPERV_REENLIGHTENMENT_VECTOR,
185 .target_vp = hv_vp_index[smp_processor_id()]
187 struct hv_tsc_emulation_control emu_ctrl = {.enabled = 1};
189 if (!hv_reenlightenment_available()) {
190 pr_warn("Hyper-V: reenlightenment support is unavailable\n");
194 hv_reenlightenment_cb = cb;
196 /* Make sure callback is registered before we write to MSRs */
199 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
200 wrmsrl(HV_X64_MSR_TSC_EMULATION_CONTROL, *((u64 *)&emu_ctrl));
202 EXPORT_SYMBOL_GPL(set_hv_tscchange_cb);
204 void clear_hv_tscchange_cb(void)
206 struct hv_reenlightenment_control re_ctrl;
208 if (!hv_reenlightenment_available())
211 rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
213 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
215 hv_reenlightenment_cb = NULL;
217 EXPORT_SYMBOL_GPL(clear_hv_tscchange_cb);
219 static int hv_cpu_die(unsigned int cpu)
221 struct hv_reenlightenment_control re_ctrl;
222 unsigned int new_cpu;
225 void *input_pg = NULL;
227 local_irq_save(flags);
228 input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
229 input_pg = *input_arg;
231 local_irq_restore(flags);
232 free_page((unsigned long)input_pg);
234 if (hv_vp_assist_page && hv_vp_assist_page[cpu])
235 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, 0);
237 if (hv_reenlightenment_cb == NULL)
240 rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
241 if (re_ctrl.target_vp == hv_vp_index[cpu]) {
242 /* Reassign to some other online CPU */
243 new_cpu = cpumask_any_but(cpu_online_mask, cpu);
245 re_ctrl.target_vp = hv_vp_index[new_cpu];
246 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
252 static int __init hv_pci_init(void)
254 int gen2vm = efi_enabled(EFI_BOOT);
257 * For Generation-2 VM, we exit from pci_arch_init() by returning 0.
258 * The purpose is to suppress the harmless warning:
259 * "PCI: Fatal: No config space access function found"
264 /* For Generation-1 VM, we'll proceed in pci_arch_init(). */
269 * This function is to be invoked early in the boot sequence after the
270 * hypervisor has been detected.
272 * 1. Setup the hypercall page.
273 * 2. Register Hyper-V specific clocksource.
274 * 3. Setup Hyper-V specific APIC entry points.
276 void __init hyperv_init(void)
278 u64 guest_id, required_msrs;
279 union hv_x64_msr_hypercall_contents hypercall_msr;
282 if (x86_hyper_type != X86_HYPER_MS_HYPERV)
285 /* Absolutely required MSRs */
286 required_msrs = HV_X64_MSR_HYPERCALL_AVAILABLE |
287 HV_X64_MSR_VP_INDEX_AVAILABLE;
289 if ((ms_hyperv.features & required_msrs) != required_msrs)
293 * Allocate the per-CPU state for the hypercall input arg.
294 * If this allocation fails, we will not be able to setup
295 * (per-CPU) hypercall input page and thus this failure is
298 hyperv_pcpu_input_arg = alloc_percpu(void *);
300 BUG_ON(hyperv_pcpu_input_arg == NULL);
302 /* Allocate percpu VP index */
303 hv_vp_index = kmalloc_array(num_possible_cpus(), sizeof(*hv_vp_index),
308 for (i = 0; i < num_possible_cpus(); i++)
309 hv_vp_index[i] = VP_INVAL;
311 hv_vp_assist_page = kcalloc(num_possible_cpus(),
312 sizeof(*hv_vp_assist_page), GFP_KERNEL);
313 if (!hv_vp_assist_page) {
314 ms_hyperv.hints &= ~HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
318 cpuhp = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/hyperv_init:online",
319 hv_cpu_init, hv_cpu_die);
321 goto free_vp_assist_page;
324 * Setup the hypercall page and enable hypercalls.
325 * 1. Register the guest ID
326 * 2. Enable the hypercall and register the hypercall page
328 guest_id = generate_guest_id(0, LINUX_VERSION_CODE, 0);
329 wrmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
331 hv_hypercall_pg = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_RX);
332 if (hv_hypercall_pg == NULL) {
333 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
334 goto remove_cpuhp_state;
337 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
338 hypercall_msr.enable = 1;
339 hypercall_msr.guest_physical_address = vmalloc_to_pfn(hv_hypercall_pg);
340 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
344 x86_init.pci.arch_init = hv_pci_init;
347 * Register Hyper-V specific clocksource.
349 #ifdef CONFIG_HYPERV_TSCPAGE
350 if (ms_hyperv.features & HV_MSR_REFERENCE_TSC_AVAILABLE) {
351 union hv_x64_msr_hypercall_contents tsc_msr;
353 tsc_pg = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL);
355 goto register_msr_cs;
357 hyperv_cs = &hyperv_cs_tsc;
359 rdmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
362 tsc_msr.guest_physical_address = vmalloc_to_pfn(tsc_pg);
364 wrmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
366 hyperv_cs_tsc.archdata.vclock_mode = VCLOCK_HVCLOCK;
368 clocksource_register_hz(&hyperv_cs_tsc, NSEC_PER_SEC/100);
374 * For 32 bit guests just use the MSR based mechanism for reading
375 * the partition counter.
378 hyperv_cs = &hyperv_cs_msr;
379 if (ms_hyperv.features & HV_MSR_TIME_REF_COUNT_AVAILABLE)
380 clocksource_register_hz(&hyperv_cs_msr, NSEC_PER_SEC/100);
385 cpuhp_remove_state(cpuhp);
387 kfree(hv_vp_assist_page);
388 hv_vp_assist_page = NULL;
395 * This routine is called before kexec/kdump, it does the required cleanup.
397 void hyperv_cleanup(void)
399 union hv_x64_msr_hypercall_contents hypercall_msr;
401 /* Reset our OS id */
402 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
405 * Reset hypercall page reference before reset the page,
406 * let hypercall operations fail safely rather than
407 * panic the kernel for using invalid hypercall page
409 hv_hypercall_pg = NULL;
411 /* Reset the hypercall page */
412 hypercall_msr.as_uint64 = 0;
413 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
415 /* Reset the TSC page */
416 hypercall_msr.as_uint64 = 0;
417 wrmsrl(HV_X64_MSR_REFERENCE_TSC, hypercall_msr.as_uint64);
419 EXPORT_SYMBOL_GPL(hyperv_cleanup);
421 void hyperv_report_panic(struct pt_regs *regs, long err)
423 static bool panic_reported;
427 * We prefer to report panic on 'die' chain as we have proper
428 * registers to report, but if we miss it (e.g. on BUG()) we need
429 * to report it on 'panic'.
433 panic_reported = true;
435 rdmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
437 wrmsrl(HV_X64_MSR_CRASH_P0, err);
438 wrmsrl(HV_X64_MSR_CRASH_P1, guest_id);
439 wrmsrl(HV_X64_MSR_CRASH_P2, regs->ip);
440 wrmsrl(HV_X64_MSR_CRASH_P3, regs->ax);
441 wrmsrl(HV_X64_MSR_CRASH_P4, regs->sp);
444 * Let Hyper-V know there is crash data available
446 wrmsrl(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY);
448 EXPORT_SYMBOL_GPL(hyperv_report_panic);
451 * hyperv_report_panic_msg - report panic message to Hyper-V
452 * @pa: physical address of the panic page containing the message
453 * @size: size of the message in the page
455 void hyperv_report_panic_msg(phys_addr_t pa, size_t size)
458 * P3 to contain the physical address of the panic page & P4 to
459 * contain the size of the panic data in that page. Rest of the
460 * registers are no-op when the NOTIFY_MSG flag is set.
462 wrmsrl(HV_X64_MSR_CRASH_P0, 0);
463 wrmsrl(HV_X64_MSR_CRASH_P1, 0);
464 wrmsrl(HV_X64_MSR_CRASH_P2, 0);
465 wrmsrl(HV_X64_MSR_CRASH_P3, pa);
466 wrmsrl(HV_X64_MSR_CRASH_P4, size);
469 * Let Hyper-V know there is crash data available along with
472 wrmsrl(HV_X64_MSR_CRASH_CTL,
473 (HV_CRASH_CTL_CRASH_NOTIFY | HV_CRASH_CTL_CRASH_NOTIFY_MSG));
475 EXPORT_SYMBOL_GPL(hyperv_report_panic_msg);
477 bool hv_is_hyperv_initialized(void)
479 union hv_x64_msr_hypercall_contents hypercall_msr;
482 * Ensure that we're really on Hyper-V, and not a KVM or Xen
483 * emulation of Hyper-V
485 if (x86_hyper_type != X86_HYPER_MS_HYPERV)
489 * Verify that earlier initialization succeeded by checking
490 * that the hypercall page is setup
492 hypercall_msr.as_uint64 = 0;
493 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
495 return hypercall_msr.enable;
497 EXPORT_SYMBOL_GPL(hv_is_hyperv_initialized);