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/types.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>
29 #include <linux/clockchips.h>
30 #include <linux/hyperv.h>
31 #include <linux/slab.h>
32 #include <linux/cpuhotplug.h>
34 #ifdef CONFIG_HYPERV_TSCPAGE
36 static struct ms_hyperv_tsc_page *tsc_pg;
38 struct ms_hyperv_tsc_page *hv_get_tsc_page(void)
42 EXPORT_SYMBOL_GPL(hv_get_tsc_page);
44 static u64 read_hv_clock_tsc(struct clocksource *arg)
46 u64 current_tick = hv_read_tsc_page(tsc_pg);
48 if (current_tick == U64_MAX)
49 rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
54 static struct clocksource hyperv_cs_tsc = {
55 .name = "hyperv_clocksource_tsc_page",
57 .read = read_hv_clock_tsc,
58 .mask = CLOCKSOURCE_MASK(64),
59 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
63 static u64 read_hv_clock_msr(struct clocksource *arg)
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.
71 rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
75 static struct clocksource hyperv_cs_msr = {
76 .name = "hyperv_clocksource_msr",
78 .read = read_hv_clock_msr,
79 .mask = CLOCKSOURCE_MASK(64),
80 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
83 void *hv_hypercall_pg;
84 EXPORT_SYMBOL_GPL(hv_hypercall_pg);
85 struct clocksource *hyperv_cs;
86 EXPORT_SYMBOL_GPL(hyperv_cs);
89 EXPORT_SYMBOL_GPL(hv_vp_index);
91 struct hv_vp_assist_page **hv_vp_assist_page;
92 EXPORT_SYMBOL_GPL(hv_vp_assist_page);
96 static int hv_cpu_init(unsigned int cpu)
99 struct hv_vp_assist_page **hvp = &hv_vp_assist_page[smp_processor_id()];
101 hv_get_vp_index(msr_vp_index);
103 hv_vp_index[smp_processor_id()] = msr_vp_index;
105 if (msr_vp_index > hv_max_vp_index)
106 hv_max_vp_index = msr_vp_index;
108 if (!hv_vp_assist_page)
112 *hvp = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL);
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;
121 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, val);
127 static void (*hv_reenlightenment_cb)(void);
129 static void hv_reenlightenment_notify(struct work_struct *dummy)
131 struct hv_tsc_emulation_status emu_status;
133 rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
135 /* Don't issue the callback if TSC accesses are not emulated */
136 if (hv_reenlightenment_cb && emu_status.inprogress)
137 hv_reenlightenment_cb();
139 static DECLARE_DELAYED_WORK(hv_reenlightenment_work, hv_reenlightenment_notify);
141 void hyperv_stop_tsc_emulation(void)
144 struct hv_tsc_emulation_status emu_status;
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);
150 rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
151 tsc_khz = div64_u64(freq, 1000);
153 EXPORT_SYMBOL_GPL(hyperv_stop_tsc_emulation);
155 static inline bool hv_reenlightenment_available(void)
158 * Check for required features and priviliges to make TSC frequency
159 * change notifications work.
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;
166 __visible void __irq_entry hyperv_reenlightenment_intr(struct pt_regs *regs)
170 inc_irq_stat(irq_hv_reenlightenment_count);
172 schedule_delayed_work(&hv_reenlightenment_work, HZ/10);
177 void set_hv_tscchange_cb(void (*cb)(void))
179 struct hv_reenlightenment_control re_ctrl = {
180 .vector = HYPERV_REENLIGHTENMENT_VECTOR,
182 .target_vp = hv_vp_index[smp_processor_id()]
184 struct hv_tsc_emulation_control emu_ctrl = {.enabled = 1};
186 if (!hv_reenlightenment_available()) {
187 pr_warn("Hyper-V: reenlightenment support is unavailable\n");
191 hv_reenlightenment_cb = cb;
193 /* Make sure callback is registered before we write to MSRs */
196 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
197 wrmsrl(HV_X64_MSR_TSC_EMULATION_CONTROL, *((u64 *)&emu_ctrl));
199 EXPORT_SYMBOL_GPL(set_hv_tscchange_cb);
201 void clear_hv_tscchange_cb(void)
203 struct hv_reenlightenment_control re_ctrl;
205 if (!hv_reenlightenment_available())
208 rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
210 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
212 hv_reenlightenment_cb = NULL;
214 EXPORT_SYMBOL_GPL(clear_hv_tscchange_cb);
216 static int hv_cpu_die(unsigned int cpu)
218 struct hv_reenlightenment_control re_ctrl;
219 unsigned int new_cpu;
221 if (hv_vp_assist_page && hv_vp_assist_page[cpu])
222 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, 0);
224 if (hv_reenlightenment_cb == NULL)
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);
232 re_ctrl.target_vp = hv_vp_index[new_cpu];
233 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
240 * This function is to be invoked early in the boot sequence after the
241 * hypervisor has been detected.
243 * 1. Setup the hypercall page.
244 * 2. Register Hyper-V specific clocksource.
245 * 3. Setup Hyper-V specific APIC entry points.
247 void __init hyperv_init(void)
249 u64 guest_id, required_msrs;
250 union hv_x64_msr_hypercall_contents hypercall_msr;
253 if (x86_hyper_type != X86_HYPER_MS_HYPERV)
256 /* Absolutely required MSRs */
257 required_msrs = HV_X64_MSR_HYPERCALL_AVAILABLE |
258 HV_X64_MSR_VP_INDEX_AVAILABLE;
260 if ((ms_hyperv.features & required_msrs) != required_msrs)
263 /* Allocate percpu VP index */
264 hv_vp_index = kmalloc_array(num_possible_cpus(), sizeof(*hv_vp_index),
269 hv_vp_assist_page = kcalloc(num_possible_cpus(),
270 sizeof(*hv_vp_assist_page), GFP_KERNEL);
271 if (!hv_vp_assist_page) {
272 ms_hyperv.hints &= ~HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
276 cpuhp = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/hyperv_init:online",
277 hv_cpu_init, hv_cpu_die);
279 goto free_vp_assist_page;
282 * Setup the hypercall page and enable hypercalls.
283 * 1. Register the guest ID
284 * 2. Enable the hypercall and register the hypercall page
286 guest_id = generate_guest_id(0, LINUX_VERSION_CODE, 0);
287 wrmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
289 hv_hypercall_pg = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_RX);
290 if (hv_hypercall_pg == NULL) {
291 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
292 goto remove_cpuhp_state;
295 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
296 hypercall_msr.enable = 1;
297 hypercall_msr.guest_physical_address = vmalloc_to_pfn(hv_hypercall_pg);
298 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
305 * Register Hyper-V specific clocksource.
307 #ifdef CONFIG_HYPERV_TSCPAGE
308 if (ms_hyperv.features & HV_X64_MSR_REFERENCE_TSC_AVAILABLE) {
309 union hv_x64_msr_hypercall_contents tsc_msr;
311 tsc_pg = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL);
313 goto register_msr_cs;
315 hyperv_cs = &hyperv_cs_tsc;
317 rdmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
320 tsc_msr.guest_physical_address = vmalloc_to_pfn(tsc_pg);
322 wrmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
324 hyperv_cs_tsc.archdata.vclock_mode = VCLOCK_HVCLOCK;
326 clocksource_register_hz(&hyperv_cs_tsc, NSEC_PER_SEC/100);
332 * For 32 bit guests just use the MSR based mechanism for reading
333 * the partition counter.
336 hyperv_cs = &hyperv_cs_msr;
337 if (ms_hyperv.features & HV_X64_MSR_TIME_REF_COUNT_AVAILABLE)
338 clocksource_register_hz(&hyperv_cs_msr, NSEC_PER_SEC/100);
343 cpuhp_remove_state(cpuhp);
345 kfree(hv_vp_assist_page);
346 hv_vp_assist_page = NULL;
353 * This routine is called before kexec/kdump, it does the required cleanup.
355 void hyperv_cleanup(void)
357 union hv_x64_msr_hypercall_contents hypercall_msr;
359 /* Reset our OS id */
360 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
362 /* Reset the hypercall page */
363 hypercall_msr.as_uint64 = 0;
364 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
366 /* Reset the TSC page */
367 hypercall_msr.as_uint64 = 0;
368 wrmsrl(HV_X64_MSR_REFERENCE_TSC, hypercall_msr.as_uint64);
370 EXPORT_SYMBOL_GPL(hyperv_cleanup);
372 void hyperv_report_panic(struct pt_regs *regs, long err)
374 static bool panic_reported;
378 * We prefer to report panic on 'die' chain as we have proper
379 * registers to report, but if we miss it (e.g. on BUG()) we need
380 * to report it on 'panic'.
384 panic_reported = true;
386 rdmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
388 wrmsrl(HV_X64_MSR_CRASH_P0, err);
389 wrmsrl(HV_X64_MSR_CRASH_P1, guest_id);
390 wrmsrl(HV_X64_MSR_CRASH_P2, regs->ip);
391 wrmsrl(HV_X64_MSR_CRASH_P3, regs->ax);
392 wrmsrl(HV_X64_MSR_CRASH_P4, regs->sp);
395 * Let Hyper-V know there is crash data available
397 wrmsrl(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY);
399 EXPORT_SYMBOL_GPL(hyperv_report_panic);
401 bool hv_is_hyperv_initialized(void)
403 union hv_x64_msr_hypercall_contents hypercall_msr;
406 * Ensure that we're really on Hyper-V, and not a KVM or Xen
407 * emulation of Hyper-V
409 if (x86_hyper_type != X86_HYPER_MS_HYPERV)
413 * Verify that earlier initialization succeeded by checking
414 * that the hypercall page is setup
416 hypercall_msr.as_uint64 = 0;
417 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
419 return hypercall_msr.enable;
421 EXPORT_SYMBOL_GPL(hv_is_hyperv_initialized);