1 /* 2 * Copyright (c) 2009, Microsoft Corporation. 3 * 4 * This program is free software; you can redistribute it and/or modify it 5 * under the terms and conditions of the GNU General Public License, 6 * version 2, as published by the Free Software Foundation. 7 * 8 * This program is distributed in the hope it will be useful, but WITHOUT 9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 11 * more details. 12 * 13 * You should have received a copy of the GNU General Public License along with 14 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple 15 * Place - Suite 330, Boston, MA 02111-1307 USA. 16 * 17 * Authors: 18 * Haiyang Zhang <haiyangz@microsoft.com> 19 * Hank Janssen <hjanssen@microsoft.com> 20 * 21 */ 22 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 23 24 #include <linux/kernel.h> 25 #include <linux/mm.h> 26 #include <linux/slab.h> 27 #include <linux/vmalloc.h> 28 #include <linux/hyperv.h> 29 #include <linux/version.h> 30 #include <linux/interrupt.h> 31 #include <linux/clockchips.h> 32 #include <asm/hyperv.h> 33 #include <asm/mshyperv.h> 34 #include "hyperv_vmbus.h" 35 36 /* The one and only */ 37 struct hv_context hv_context = { 38 .synic_initialized = false, 39 .hypercall_page = NULL, 40 }; 41 42 #define HV_TIMER_FREQUENCY (10 * 1000 * 1000) /* 100ns period */ 43 #define HV_MAX_MAX_DELTA_TICKS 0xffffffff 44 #define HV_MIN_DELTA_TICKS 1 45 46 /* 47 * query_hypervisor_info - Get version info of the windows hypervisor 48 */ 49 unsigned int host_info_eax; 50 unsigned int host_info_ebx; 51 unsigned int host_info_ecx; 52 unsigned int host_info_edx; 53 54 static int query_hypervisor_info(void) 55 { 56 unsigned int eax; 57 unsigned int ebx; 58 unsigned int ecx; 59 unsigned int edx; 60 unsigned int max_leaf; 61 unsigned int op; 62 63 /* 64 * Its assumed that this is called after confirming that Viridian 65 * is present. Query id and revision. 66 */ 67 eax = 0; 68 ebx = 0; 69 ecx = 0; 70 edx = 0; 71 op = HVCPUID_VENDOR_MAXFUNCTION; 72 cpuid(op, &eax, &ebx, &ecx, &edx); 73 74 max_leaf = eax; 75 76 if (max_leaf >= HVCPUID_VERSION) { 77 eax = 0; 78 ebx = 0; 79 ecx = 0; 80 edx = 0; 81 op = HVCPUID_VERSION; 82 cpuid(op, &eax, &ebx, &ecx, &edx); 83 host_info_eax = eax; 84 host_info_ebx = ebx; 85 host_info_ecx = ecx; 86 host_info_edx = edx; 87 } 88 return max_leaf; 89 } 90 91 /* 92 * hv_do_hypercall- Invoke the specified hypercall 93 */ 94 u64 hv_do_hypercall(u64 control, void *input, void *output) 95 { 96 u64 input_address = (input) ? virt_to_phys(input) : 0; 97 u64 output_address = (output) ? virt_to_phys(output) : 0; 98 void *hypercall_page = hv_context.hypercall_page; 99 #ifdef CONFIG_X86_64 100 u64 hv_status = 0; 101 102 if (!hypercall_page) 103 return (u64)ULLONG_MAX; 104 105 __asm__ __volatile__("mov %0, %%r8" : : "r" (output_address) : "r8"); 106 __asm__ __volatile__("call *%3" : "=a" (hv_status) : 107 "c" (control), "d" (input_address), 108 "m" (hypercall_page)); 109 110 return hv_status; 111 112 #else 113 114 u32 control_hi = control >> 32; 115 u32 control_lo = control & 0xFFFFFFFF; 116 u32 hv_status_hi = 1; 117 u32 hv_status_lo = 1; 118 u32 input_address_hi = input_address >> 32; 119 u32 input_address_lo = input_address & 0xFFFFFFFF; 120 u32 output_address_hi = output_address >> 32; 121 u32 output_address_lo = output_address & 0xFFFFFFFF; 122 123 if (!hypercall_page) 124 return (u64)ULLONG_MAX; 125 126 __asm__ __volatile__ ("call *%8" : "=d"(hv_status_hi), 127 "=a"(hv_status_lo) : "d" (control_hi), 128 "a" (control_lo), "b" (input_address_hi), 129 "c" (input_address_lo), "D"(output_address_hi), 130 "S"(output_address_lo), "m" (hypercall_page)); 131 132 return hv_status_lo | ((u64)hv_status_hi << 32); 133 #endif /* !x86_64 */ 134 } 135 EXPORT_SYMBOL_GPL(hv_do_hypercall); 136 137 #ifdef CONFIG_X86_64 138 static cycle_t read_hv_clock_tsc(struct clocksource *arg) 139 { 140 cycle_t current_tick; 141 struct ms_hyperv_tsc_page *tsc_pg = hv_context.tsc_page; 142 143 if (tsc_pg->tsc_sequence != 0) { 144 /* 145 * Use the tsc page to compute the value. 146 */ 147 148 while (1) { 149 cycle_t tmp; 150 u32 sequence = tsc_pg->tsc_sequence; 151 u64 cur_tsc; 152 u64 scale = tsc_pg->tsc_scale; 153 s64 offset = tsc_pg->tsc_offset; 154 155 rdtscll(cur_tsc); 156 /* current_tick = ((cur_tsc *scale) >> 64) + offset */ 157 asm("mulq %3" 158 : "=d" (current_tick), "=a" (tmp) 159 : "a" (cur_tsc), "r" (scale)); 160 161 current_tick += offset; 162 if (tsc_pg->tsc_sequence == sequence) 163 return current_tick; 164 165 if (tsc_pg->tsc_sequence != 0) 166 continue; 167 /* 168 * Fallback using MSR method. 169 */ 170 break; 171 } 172 } 173 rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick); 174 return current_tick; 175 } 176 177 static struct clocksource hyperv_cs_tsc = { 178 .name = "hyperv_clocksource_tsc_page", 179 .rating = 425, 180 .read = read_hv_clock_tsc, 181 .mask = CLOCKSOURCE_MASK(64), 182 .flags = CLOCK_SOURCE_IS_CONTINUOUS, 183 }; 184 #endif 185 186 187 /* 188 * hv_init - Main initialization routine. 189 * 190 * This routine must be called before any other routines in here are called 191 */ 192 int hv_init(void) 193 { 194 int max_leaf; 195 union hv_x64_msr_hypercall_contents hypercall_msr; 196 void *virtaddr = NULL; 197 198 memset(hv_context.synic_event_page, 0, sizeof(void *) * NR_CPUS); 199 memset(hv_context.synic_message_page, 0, 200 sizeof(void *) * NR_CPUS); 201 memset(hv_context.post_msg_page, 0, 202 sizeof(void *) * NR_CPUS); 203 memset(hv_context.vp_index, 0, 204 sizeof(int) * NR_CPUS); 205 memset(hv_context.event_dpc, 0, 206 sizeof(void *) * NR_CPUS); 207 memset(hv_context.msg_dpc, 0, 208 sizeof(void *) * NR_CPUS); 209 memset(hv_context.clk_evt, 0, 210 sizeof(void *) * NR_CPUS); 211 212 max_leaf = query_hypervisor_info(); 213 214 /* 215 * Write our OS ID. 216 */ 217 hv_context.guestid = generate_guest_id(0, LINUX_VERSION_CODE, 0); 218 wrmsrl(HV_X64_MSR_GUEST_OS_ID, hv_context.guestid); 219 220 /* See if the hypercall page is already set */ 221 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 222 223 virtaddr = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_EXEC); 224 225 if (!virtaddr) 226 goto cleanup; 227 228 hypercall_msr.enable = 1; 229 230 hypercall_msr.guest_physical_address = vmalloc_to_pfn(virtaddr); 231 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 232 233 /* Confirm that hypercall page did get setup. */ 234 hypercall_msr.as_uint64 = 0; 235 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 236 237 if (!hypercall_msr.enable) 238 goto cleanup; 239 240 hv_context.hypercall_page = virtaddr; 241 242 #ifdef CONFIG_X86_64 243 if (ms_hyperv.features & HV_X64_MSR_REFERENCE_TSC_AVAILABLE) { 244 union hv_x64_msr_hypercall_contents tsc_msr; 245 void *va_tsc; 246 247 va_tsc = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL); 248 if (!va_tsc) 249 goto cleanup; 250 hv_context.tsc_page = va_tsc; 251 252 rdmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64); 253 254 tsc_msr.enable = 1; 255 tsc_msr.guest_physical_address = vmalloc_to_pfn(va_tsc); 256 257 wrmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64); 258 clocksource_register_hz(&hyperv_cs_tsc, NSEC_PER_SEC/100); 259 } 260 #endif 261 return 0; 262 263 cleanup: 264 if (virtaddr) { 265 if (hypercall_msr.enable) { 266 hypercall_msr.as_uint64 = 0; 267 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 268 } 269 270 vfree(virtaddr); 271 } 272 273 return -ENOTSUPP; 274 } 275 276 /* 277 * hv_cleanup - Cleanup routine. 278 * 279 * This routine is called normally during driver unloading or exiting. 280 */ 281 void hv_cleanup(void) 282 { 283 union hv_x64_msr_hypercall_contents hypercall_msr; 284 285 /* Reset our OS id */ 286 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0); 287 288 if (hv_context.hypercall_page) { 289 hypercall_msr.as_uint64 = 0; 290 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 291 vfree(hv_context.hypercall_page); 292 hv_context.hypercall_page = NULL; 293 } 294 295 #ifdef CONFIG_X86_64 296 /* 297 * Cleanup the TSC page based CS. 298 */ 299 if (ms_hyperv.features & HV_X64_MSR_REFERENCE_TSC_AVAILABLE) { 300 /* 301 * Crash can happen in an interrupt context and unregistering 302 * a clocksource is impossible and redundant in this case. 303 */ 304 if (!oops_in_progress) { 305 clocksource_change_rating(&hyperv_cs_tsc, 10); 306 clocksource_unregister(&hyperv_cs_tsc); 307 } 308 309 hypercall_msr.as_uint64 = 0; 310 wrmsrl(HV_X64_MSR_REFERENCE_TSC, hypercall_msr.as_uint64); 311 vfree(hv_context.tsc_page); 312 hv_context.tsc_page = NULL; 313 } 314 #endif 315 } 316 317 /* 318 * hv_post_message - Post a message using the hypervisor message IPC. 319 * 320 * This involves a hypercall. 321 */ 322 int hv_post_message(union hv_connection_id connection_id, 323 enum hv_message_type message_type, 324 void *payload, size_t payload_size) 325 { 326 327 struct hv_input_post_message *aligned_msg; 328 u64 status; 329 330 if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT) 331 return -EMSGSIZE; 332 333 aligned_msg = (struct hv_input_post_message *) 334 hv_context.post_msg_page[get_cpu()]; 335 336 aligned_msg->connectionid = connection_id; 337 aligned_msg->reserved = 0; 338 aligned_msg->message_type = message_type; 339 aligned_msg->payload_size = payload_size; 340 memcpy((void *)aligned_msg->payload, payload, payload_size); 341 342 status = hv_do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL); 343 344 put_cpu(); 345 return status & 0xFFFF; 346 } 347 348 static int hv_ce_set_next_event(unsigned long delta, 349 struct clock_event_device *evt) 350 { 351 cycle_t current_tick; 352 353 WARN_ON(!clockevent_state_oneshot(evt)); 354 355 rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick); 356 current_tick += delta; 357 wrmsrl(HV_X64_MSR_STIMER0_COUNT, current_tick); 358 return 0; 359 } 360 361 static int hv_ce_shutdown(struct clock_event_device *evt) 362 { 363 wrmsrl(HV_X64_MSR_STIMER0_COUNT, 0); 364 wrmsrl(HV_X64_MSR_STIMER0_CONFIG, 0); 365 366 return 0; 367 } 368 369 static int hv_ce_set_oneshot(struct clock_event_device *evt) 370 { 371 union hv_timer_config timer_cfg; 372 373 timer_cfg.enable = 1; 374 timer_cfg.auto_enable = 1; 375 timer_cfg.sintx = VMBUS_MESSAGE_SINT; 376 wrmsrl(HV_X64_MSR_STIMER0_CONFIG, timer_cfg.as_uint64); 377 378 return 0; 379 } 380 381 static void hv_init_clockevent_device(struct clock_event_device *dev, int cpu) 382 { 383 dev->name = "Hyper-V clockevent"; 384 dev->features = CLOCK_EVT_FEAT_ONESHOT; 385 dev->cpumask = cpumask_of(cpu); 386 dev->rating = 1000; 387 /* 388 * Avoid settint dev->owner = THIS_MODULE deliberately as doing so will 389 * result in clockevents_config_and_register() taking additional 390 * references to the hv_vmbus module making it impossible to unload. 391 */ 392 393 dev->set_state_shutdown = hv_ce_shutdown; 394 dev->set_state_oneshot = hv_ce_set_oneshot; 395 dev->set_next_event = hv_ce_set_next_event; 396 } 397 398 399 int hv_synic_alloc(void) 400 { 401 size_t size = sizeof(struct tasklet_struct); 402 size_t ced_size = sizeof(struct clock_event_device); 403 int cpu; 404 405 hv_context.hv_numa_map = kzalloc(sizeof(struct cpumask) * nr_node_ids, 406 GFP_ATOMIC); 407 if (hv_context.hv_numa_map == NULL) { 408 pr_err("Unable to allocate NUMA map\n"); 409 goto err; 410 } 411 412 for_each_online_cpu(cpu) { 413 hv_context.event_dpc[cpu] = kmalloc(size, GFP_ATOMIC); 414 if (hv_context.event_dpc[cpu] == NULL) { 415 pr_err("Unable to allocate event dpc\n"); 416 goto err; 417 } 418 tasklet_init(hv_context.event_dpc[cpu], vmbus_on_event, cpu); 419 420 hv_context.msg_dpc[cpu] = kmalloc(size, GFP_ATOMIC); 421 if (hv_context.msg_dpc[cpu] == NULL) { 422 pr_err("Unable to allocate event dpc\n"); 423 goto err; 424 } 425 tasklet_init(hv_context.msg_dpc[cpu], vmbus_on_msg_dpc, cpu); 426 427 hv_context.clk_evt[cpu] = kzalloc(ced_size, GFP_ATOMIC); 428 if (hv_context.clk_evt[cpu] == NULL) { 429 pr_err("Unable to allocate clock event device\n"); 430 goto err; 431 } 432 433 hv_init_clockevent_device(hv_context.clk_evt[cpu], cpu); 434 435 hv_context.synic_message_page[cpu] = 436 (void *)get_zeroed_page(GFP_ATOMIC); 437 438 if (hv_context.synic_message_page[cpu] == NULL) { 439 pr_err("Unable to allocate SYNIC message page\n"); 440 goto err; 441 } 442 443 hv_context.synic_event_page[cpu] = 444 (void *)get_zeroed_page(GFP_ATOMIC); 445 446 if (hv_context.synic_event_page[cpu] == NULL) { 447 pr_err("Unable to allocate SYNIC event page\n"); 448 goto err; 449 } 450 451 hv_context.post_msg_page[cpu] = 452 (void *)get_zeroed_page(GFP_ATOMIC); 453 454 if (hv_context.post_msg_page[cpu] == NULL) { 455 pr_err("Unable to allocate post msg page\n"); 456 goto err; 457 } 458 } 459 460 return 0; 461 err: 462 return -ENOMEM; 463 } 464 465 static void hv_synic_free_cpu(int cpu) 466 { 467 kfree(hv_context.event_dpc[cpu]); 468 kfree(hv_context.msg_dpc[cpu]); 469 kfree(hv_context.clk_evt[cpu]); 470 if (hv_context.synic_event_page[cpu]) 471 free_page((unsigned long)hv_context.synic_event_page[cpu]); 472 if (hv_context.synic_message_page[cpu]) 473 free_page((unsigned long)hv_context.synic_message_page[cpu]); 474 if (hv_context.post_msg_page[cpu]) 475 free_page((unsigned long)hv_context.post_msg_page[cpu]); 476 } 477 478 void hv_synic_free(void) 479 { 480 int cpu; 481 482 kfree(hv_context.hv_numa_map); 483 for_each_online_cpu(cpu) 484 hv_synic_free_cpu(cpu); 485 } 486 487 /* 488 * hv_synic_init - Initialize the Synthethic Interrupt Controller. 489 * 490 * If it is already initialized by another entity (ie x2v shim), we need to 491 * retrieve the initialized message and event pages. Otherwise, we create and 492 * initialize the message and event pages. 493 */ 494 void hv_synic_init(void *arg) 495 { 496 u64 version; 497 union hv_synic_simp simp; 498 union hv_synic_siefp siefp; 499 union hv_synic_sint shared_sint; 500 union hv_synic_scontrol sctrl; 501 u64 vp_index; 502 503 int cpu = smp_processor_id(); 504 505 if (!hv_context.hypercall_page) 506 return; 507 508 /* Check the version */ 509 rdmsrl(HV_X64_MSR_SVERSION, version); 510 511 /* Setup the Synic's message page */ 512 rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64); 513 simp.simp_enabled = 1; 514 simp.base_simp_gpa = virt_to_phys(hv_context.synic_message_page[cpu]) 515 >> PAGE_SHIFT; 516 517 wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64); 518 519 /* Setup the Synic's event page */ 520 rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64); 521 siefp.siefp_enabled = 1; 522 siefp.base_siefp_gpa = virt_to_phys(hv_context.synic_event_page[cpu]) 523 >> PAGE_SHIFT; 524 525 wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64); 526 527 /* Setup the shared SINT. */ 528 rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64); 529 530 shared_sint.as_uint64 = 0; 531 shared_sint.vector = HYPERVISOR_CALLBACK_VECTOR; 532 shared_sint.masked = false; 533 shared_sint.auto_eoi = true; 534 535 wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64); 536 537 /* Enable the global synic bit */ 538 rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64); 539 sctrl.enable = 1; 540 541 wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64); 542 543 hv_context.synic_initialized = true; 544 545 /* 546 * Setup the mapping between Hyper-V's notion 547 * of cpuid and Linux' notion of cpuid. 548 * This array will be indexed using Linux cpuid. 549 */ 550 rdmsrl(HV_X64_MSR_VP_INDEX, vp_index); 551 hv_context.vp_index[cpu] = (u32)vp_index; 552 553 INIT_LIST_HEAD(&hv_context.percpu_list[cpu]); 554 555 /* 556 * Register the per-cpu clockevent source. 557 */ 558 if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE) 559 clockevents_config_and_register(hv_context.clk_evt[cpu], 560 HV_TIMER_FREQUENCY, 561 HV_MIN_DELTA_TICKS, 562 HV_MAX_MAX_DELTA_TICKS); 563 return; 564 } 565 566 /* 567 * hv_synic_clockevents_cleanup - Cleanup clockevent devices 568 */ 569 void hv_synic_clockevents_cleanup(void) 570 { 571 int cpu; 572 573 if (!(ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE)) 574 return; 575 576 for_each_online_cpu(cpu) 577 clockevents_unbind_device(hv_context.clk_evt[cpu], cpu); 578 } 579 580 /* 581 * hv_synic_cleanup - Cleanup routine for hv_synic_init(). 582 */ 583 void hv_synic_cleanup(void *arg) 584 { 585 union hv_synic_sint shared_sint; 586 union hv_synic_simp simp; 587 union hv_synic_siefp siefp; 588 union hv_synic_scontrol sctrl; 589 int cpu = smp_processor_id(); 590 591 if (!hv_context.synic_initialized) 592 return; 593 594 /* Turn off clockevent device */ 595 if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE) 596 hv_ce_shutdown(hv_context.clk_evt[cpu]); 597 598 rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64); 599 600 shared_sint.masked = 1; 601 602 /* Need to correctly cleanup in the case of SMP!!! */ 603 /* Disable the interrupt */ 604 wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64); 605 606 rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64); 607 simp.simp_enabled = 0; 608 simp.base_simp_gpa = 0; 609 610 wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64); 611 612 rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64); 613 siefp.siefp_enabled = 0; 614 siefp.base_siefp_gpa = 0; 615 616 wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64); 617 618 /* Disable the global synic bit */ 619 rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64); 620 sctrl.enable = 0; 621 wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64); 622 } 623