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 36 static struct ms_hyperv_tsc_page *tsc_pg; 37 38 struct ms_hyperv_tsc_page *hv_get_tsc_page(void) 39 { 40 return tsc_pg; 41 } 42 EXPORT_SYMBOL_GPL(hv_get_tsc_page); 43 44 static 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 54 static 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 63 static 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 75 static 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 83 void *hv_hypercall_pg; 84 EXPORT_SYMBOL_GPL(hv_hypercall_pg); 85 struct clocksource *hyperv_cs; 86 EXPORT_SYMBOL_GPL(hyperv_cs); 87 88 u32 *hv_vp_index; 89 EXPORT_SYMBOL_GPL(hv_vp_index); 90 91 struct hv_vp_assist_page **hv_vp_assist_page; 92 EXPORT_SYMBOL_GPL(hv_vp_assist_page); 93 94 u32 hv_max_vp_index; 95 96 static 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 } 126 127 static void (*hv_reenlightenment_cb)(void); 128 129 static 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 } 139 static DECLARE_DELAYED_WORK(hv_reenlightenment_work, hv_reenlightenment_notify); 140 141 void 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 } 153 EXPORT_SYMBOL_GPL(hyperv_stop_tsc_emulation); 154 155 static 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 } 176 177 void 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 } 199 EXPORT_SYMBOL_GPL(set_hv_tscchange_cb); 200 201 void 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 } 214 EXPORT_SYMBOL_GPL(clear_hv_tscchange_cb); 215 216 static 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 */ 246 void 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 } 326 register_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; 338 339 remove_cpuhp_state: 340 cpuhp_remove_state(cpuhp); 341 free_vp_assist_page: 342 kfree(hv_vp_assist_page); 343 hv_vp_assist_page = NULL; 344 free_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 */ 352 void 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 } 367 EXPORT_SYMBOL_GPL(hyperv_cleanup); 368 369 void 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 } 396 EXPORT_SYMBOL_GPL(hyperv_report_panic); 397 398 bool 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 } 418 EXPORT_SYMBOL_GPL(hv_is_hyperv_initialized); 419