2 * X86 specific Hyper-V initialization code.
4 * Copyright (C) 2016, Microsoft, Inc.
6 * Author : K. Y. Srinivasan <kys@microsoft.com>
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.
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
20 #include <linux/efi.h>
21 #include <linux/types.h>
24 #include <asm/hypervisor.h>
25 #include <asm/hyperv-tlfs.h>
26 #include <asm/mshyperv.h>
27 #include <linux/version.h>
28 #include <linux/vmalloc.h>
30 #include <linux/clockchips.h>
31 #include <linux/hyperv.h>
32 #include <linux/slab.h>
33 #include <linux/kernel.h>
34 #include <linux/cpuhotplug.h>
36 #ifdef CONFIG_HYPERV_TSCPAGE
38 static struct ms_hyperv_tsc_page *tsc_pg;
40 struct ms_hyperv_tsc_page *hv_get_tsc_page(void)
44 EXPORT_SYMBOL_GPL(hv_get_tsc_page);
46 static u64 read_hv_clock_tsc(struct clocksource *arg)
48 u64 current_tick = hv_read_tsc_page(tsc_pg);
50 if (current_tick == U64_MAX)
51 rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
56 static struct clocksource hyperv_cs_tsc = {
57 .name = "hyperv_clocksource_tsc_page",
59 .read = read_hv_clock_tsc,
60 .mask = CLOCKSOURCE_MASK(64),
61 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
65 static u64 read_hv_clock_msr(struct clocksource *arg)
69 * Read the partition counter to get the current tick count. This count
70 * is set to 0 when the partition is created and is incremented in
71 * 100 nanosecond units.
73 rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
77 static struct clocksource hyperv_cs_msr = {
78 .name = "hyperv_clocksource_msr",
80 .read = read_hv_clock_msr,
81 .mask = CLOCKSOURCE_MASK(64),
82 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
85 void *hv_hypercall_pg;
86 EXPORT_SYMBOL_GPL(hv_hypercall_pg);
87 struct clocksource *hyperv_cs;
88 EXPORT_SYMBOL_GPL(hyperv_cs);
91 EXPORT_SYMBOL_GPL(hv_vp_index);
93 struct hv_vp_assist_page **hv_vp_assist_page;
94 EXPORT_SYMBOL_GPL(hv_vp_assist_page);
96 void __percpu **hyperv_pcpu_input_arg;
97 EXPORT_SYMBOL_GPL(hyperv_pcpu_input_arg);
101 static int hv_cpu_init(unsigned int cpu)
104 struct hv_vp_assist_page **hvp = &hv_vp_assist_page[smp_processor_id()];
108 input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
109 pg = alloc_page(GFP_KERNEL);
112 *input_arg = page_address(pg);
114 hv_get_vp_index(msr_vp_index);
116 hv_vp_index[smp_processor_id()] = msr_vp_index;
118 if (msr_vp_index > hv_max_vp_index)
119 hv_max_vp_index = msr_vp_index;
121 if (!hv_vp_assist_page)
125 *hvp = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL);
130 val = vmalloc_to_pfn(*hvp);
131 val = (val << HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT) |
132 HV_X64_MSR_VP_ASSIST_PAGE_ENABLE;
134 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, val);
140 static void (*hv_reenlightenment_cb)(void);
142 static void hv_reenlightenment_notify(struct work_struct *dummy)
144 struct hv_tsc_emulation_status emu_status;
146 rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
148 /* Don't issue the callback if TSC accesses are not emulated */
149 if (hv_reenlightenment_cb && emu_status.inprogress)
150 hv_reenlightenment_cb();
152 static DECLARE_DELAYED_WORK(hv_reenlightenment_work, hv_reenlightenment_notify);
154 void hyperv_stop_tsc_emulation(void)
157 struct hv_tsc_emulation_status emu_status;
159 rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
160 emu_status.inprogress = 0;
161 wrmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
163 rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
164 tsc_khz = div64_u64(freq, 1000);
166 EXPORT_SYMBOL_GPL(hyperv_stop_tsc_emulation);
168 static inline bool hv_reenlightenment_available(void)
171 * Check for required features and priviliges to make TSC frequency
172 * change notifications work.
174 return ms_hyperv.features & HV_X64_ACCESS_FREQUENCY_MSRS &&
175 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE &&
176 ms_hyperv.features & HV_X64_ACCESS_REENLIGHTENMENT;
179 __visible void __irq_entry hyperv_reenlightenment_intr(struct pt_regs *regs)
183 inc_irq_stat(irq_hv_reenlightenment_count);
185 schedule_delayed_work(&hv_reenlightenment_work, HZ/10);
190 void set_hv_tscchange_cb(void (*cb)(void))
192 struct hv_reenlightenment_control re_ctrl = {
193 .vector = HYPERV_REENLIGHTENMENT_VECTOR,
196 struct hv_tsc_emulation_control emu_ctrl = {.enabled = 1};
198 if (!hv_reenlightenment_available()) {
199 pr_warn("Hyper-V: reenlightenment support is unavailable\n");
206 hv_reenlightenment_cb = cb;
208 /* Make sure callback is registered before we write to MSRs */
211 re_ctrl.target_vp = hv_vp_index[get_cpu()];
213 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
214 wrmsrl(HV_X64_MSR_TSC_EMULATION_CONTROL, *((u64 *)&emu_ctrl));
218 EXPORT_SYMBOL_GPL(set_hv_tscchange_cb);
220 void clear_hv_tscchange_cb(void)
222 struct hv_reenlightenment_control re_ctrl;
224 if (!hv_reenlightenment_available())
227 rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
229 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
231 hv_reenlightenment_cb = NULL;
233 EXPORT_SYMBOL_GPL(clear_hv_tscchange_cb);
235 static int hv_cpu_die(unsigned int cpu)
237 struct hv_reenlightenment_control re_ctrl;
238 unsigned int new_cpu;
241 void *input_pg = NULL;
243 local_irq_save(flags);
244 input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
245 input_pg = *input_arg;
247 local_irq_restore(flags);
248 free_page((unsigned long)input_pg);
250 if (hv_vp_assist_page && hv_vp_assist_page[cpu])
251 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, 0);
253 if (hv_reenlightenment_cb == NULL)
256 rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
257 if (re_ctrl.target_vp == hv_vp_index[cpu]) {
258 /* Reassign to some other online CPU */
259 new_cpu = cpumask_any_but(cpu_online_mask, cpu);
261 re_ctrl.target_vp = hv_vp_index[new_cpu];
262 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
268 static int __init hv_pci_init(void)
270 int gen2vm = efi_enabled(EFI_BOOT);
273 * For Generation-2 VM, we exit from pci_arch_init() by returning 0.
274 * The purpose is to suppress the harmless warning:
275 * "PCI: Fatal: No config space access function found"
280 /* For Generation-1 VM, we'll proceed in pci_arch_init(). */
285 * This function is to be invoked early in the boot sequence after the
286 * hypervisor has been detected.
288 * 1. Setup the hypercall page.
289 * 2. Register Hyper-V specific clocksource.
290 * 3. Setup Hyper-V specific APIC entry points.
292 void __init hyperv_init(void)
294 u64 guest_id, required_msrs;
295 union hv_x64_msr_hypercall_contents hypercall_msr;
298 if (x86_hyper_type != X86_HYPER_MS_HYPERV)
301 /* Absolutely required MSRs */
302 required_msrs = HV_X64_MSR_HYPERCALL_AVAILABLE |
303 HV_X64_MSR_VP_INDEX_AVAILABLE;
305 if ((ms_hyperv.features & required_msrs) != required_msrs)
309 * Allocate the per-CPU state for the hypercall input arg.
310 * If this allocation fails, we will not be able to setup
311 * (per-CPU) hypercall input page and thus this failure is
314 hyperv_pcpu_input_arg = alloc_percpu(void *);
316 BUG_ON(hyperv_pcpu_input_arg == NULL);
318 /* Allocate percpu VP index */
319 hv_vp_index = kmalloc_array(num_possible_cpus(), sizeof(*hv_vp_index),
324 for (i = 0; i < num_possible_cpus(); i++)
325 hv_vp_index[i] = VP_INVAL;
327 hv_vp_assist_page = kcalloc(num_possible_cpus(),
328 sizeof(*hv_vp_assist_page), GFP_KERNEL);
329 if (!hv_vp_assist_page) {
330 ms_hyperv.hints &= ~HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
334 cpuhp = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/hyperv_init:online",
335 hv_cpu_init, hv_cpu_die);
337 goto free_vp_assist_page;
340 * Setup the hypercall page and enable hypercalls.
341 * 1. Register the guest ID
342 * 2. Enable the hypercall and register the hypercall page
344 guest_id = generate_guest_id(0, LINUX_VERSION_CODE, 0);
345 wrmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
347 hv_hypercall_pg = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_RX);
348 if (hv_hypercall_pg == NULL) {
349 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
350 goto remove_cpuhp_state;
353 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
354 hypercall_msr.enable = 1;
355 hypercall_msr.guest_physical_address = vmalloc_to_pfn(hv_hypercall_pg);
356 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
360 x86_init.pci.arch_init = hv_pci_init;
363 * Register Hyper-V specific clocksource.
365 #ifdef CONFIG_HYPERV_TSCPAGE
366 if (ms_hyperv.features & HV_MSR_REFERENCE_TSC_AVAILABLE) {
367 union hv_x64_msr_hypercall_contents tsc_msr;
369 tsc_pg = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL);
371 goto register_msr_cs;
373 hyperv_cs = &hyperv_cs_tsc;
375 rdmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
378 tsc_msr.guest_physical_address = vmalloc_to_pfn(tsc_pg);
380 wrmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
382 hyperv_cs_tsc.archdata.vclock_mode = VCLOCK_HVCLOCK;
384 clocksource_register_hz(&hyperv_cs_tsc, NSEC_PER_SEC/100);
390 * For 32 bit guests just use the MSR based mechanism for reading
391 * the partition counter.
394 hyperv_cs = &hyperv_cs_msr;
395 if (ms_hyperv.features & HV_MSR_TIME_REF_COUNT_AVAILABLE)
396 clocksource_register_hz(&hyperv_cs_msr, NSEC_PER_SEC/100);
401 cpuhp_remove_state(cpuhp);
403 kfree(hv_vp_assist_page);
404 hv_vp_assist_page = NULL;
411 * This routine is called before kexec/kdump, it does the required cleanup.
413 void hyperv_cleanup(void)
415 union hv_x64_msr_hypercall_contents hypercall_msr;
417 /* Reset our OS id */
418 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
421 * Reset hypercall page reference before reset the page,
422 * let hypercall operations fail safely rather than
423 * panic the kernel for using invalid hypercall page
425 hv_hypercall_pg = NULL;
427 /* Reset the hypercall page */
428 hypercall_msr.as_uint64 = 0;
429 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
431 /* Reset the TSC page */
432 hypercall_msr.as_uint64 = 0;
433 wrmsrl(HV_X64_MSR_REFERENCE_TSC, hypercall_msr.as_uint64);
435 EXPORT_SYMBOL_GPL(hyperv_cleanup);
437 void hyperv_report_panic(struct pt_regs *regs, long err, bool in_die)
439 static bool panic_reported;
442 if (in_die && !panic_on_oops)
446 * We prefer to report panic on 'die' chain as we have proper
447 * registers to report, but if we miss it (e.g. on BUG()) we need
448 * to report it on 'panic'.
452 panic_reported = true;
454 rdmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
456 wrmsrl(HV_X64_MSR_CRASH_P0, err);
457 wrmsrl(HV_X64_MSR_CRASH_P1, guest_id);
458 wrmsrl(HV_X64_MSR_CRASH_P2, regs->ip);
459 wrmsrl(HV_X64_MSR_CRASH_P3, regs->ax);
460 wrmsrl(HV_X64_MSR_CRASH_P4, regs->sp);
463 * Let Hyper-V know there is crash data available
465 wrmsrl(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY);
467 EXPORT_SYMBOL_GPL(hyperv_report_panic);
470 * hyperv_report_panic_msg - report panic message to Hyper-V
471 * @pa: physical address of the panic page containing the message
472 * @size: size of the message in the page
474 void hyperv_report_panic_msg(phys_addr_t pa, size_t size)
477 * P3 to contain the physical address of the panic page & P4 to
478 * contain the size of the panic data in that page. Rest of the
479 * registers are no-op when the NOTIFY_MSG flag is set.
481 wrmsrl(HV_X64_MSR_CRASH_P0, 0);
482 wrmsrl(HV_X64_MSR_CRASH_P1, 0);
483 wrmsrl(HV_X64_MSR_CRASH_P2, 0);
484 wrmsrl(HV_X64_MSR_CRASH_P3, pa);
485 wrmsrl(HV_X64_MSR_CRASH_P4, size);
488 * Let Hyper-V know there is crash data available along with
491 wrmsrl(HV_X64_MSR_CRASH_CTL,
492 (HV_CRASH_CTL_CRASH_NOTIFY | HV_CRASH_CTL_CRASH_NOTIFY_MSG));
494 EXPORT_SYMBOL_GPL(hyperv_report_panic_msg);
496 bool hv_is_hyperv_initialized(void)
498 union hv_x64_msr_hypercall_contents hypercall_msr;
501 * Ensure that we're really on Hyper-V, and not a KVM or Xen
502 * emulation of Hyper-V
504 if (x86_hyper_type != X86_HYPER_MS_HYPERV)
508 * Verify that earlier initialization succeeded by checking
509 * that the hypercall page is setup
511 hypercall_msr.as_uint64 = 0;
512 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
514 return hypercall_msr.enable;
516 EXPORT_SYMBOL_GPL(hv_is_hyperv_initialized);