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 * do_hypercall- Invoke the specified hypercall 93 */ 94 static u64 do_hypercall(u64 control, void *input, void *output) 95 { 96 #ifdef CONFIG_X86_64 97 u64 hv_status = 0; 98 u64 input_address = (input) ? virt_to_phys(input) : 0; 99 u64 output_address = (output) ? virt_to_phys(output) : 0; 100 void *hypercall_page = hv_context.hypercall_page; 101 102 __asm__ __volatile__("mov %0, %%r8" : : "r" (output_address) : "r8"); 103 __asm__ __volatile__("call *%3" : "=a" (hv_status) : 104 "c" (control), "d" (input_address), 105 "m" (hypercall_page)); 106 107 return hv_status; 108 109 #else 110 111 u32 control_hi = control >> 32; 112 u32 control_lo = control & 0xFFFFFFFF; 113 u32 hv_status_hi = 1; 114 u32 hv_status_lo = 1; 115 u64 input_address = (input) ? virt_to_phys(input) : 0; 116 u32 input_address_hi = input_address >> 32; 117 u32 input_address_lo = input_address & 0xFFFFFFFF; 118 u64 output_address = (output) ? virt_to_phys(output) : 0; 119 u32 output_address_hi = output_address >> 32; 120 u32 output_address_lo = output_address & 0xFFFFFFFF; 121 void *hypercall_page = hv_context.hypercall_page; 122 123 __asm__ __volatile__ ("call *%8" : "=d"(hv_status_hi), 124 "=a"(hv_status_lo) : "d" (control_hi), 125 "a" (control_lo), "b" (input_address_hi), 126 "c" (input_address_lo), "D"(output_address_hi), 127 "S"(output_address_lo), "m" (hypercall_page)); 128 129 return hv_status_lo | ((u64)hv_status_hi << 32); 130 #endif /* !x86_64 */ 131 } 132 133 /* 134 * hv_init - Main initialization routine. 135 * 136 * This routine must be called before any other routines in here are called 137 */ 138 int hv_init(void) 139 { 140 int max_leaf; 141 union hv_x64_msr_hypercall_contents hypercall_msr; 142 void *virtaddr = NULL; 143 144 memset(hv_context.synic_event_page, 0, sizeof(void *) * NR_CPUS); 145 memset(hv_context.synic_message_page, 0, 146 sizeof(void *) * NR_CPUS); 147 memset(hv_context.post_msg_page, 0, 148 sizeof(void *) * NR_CPUS); 149 memset(hv_context.vp_index, 0, 150 sizeof(int) * NR_CPUS); 151 memset(hv_context.event_dpc, 0, 152 sizeof(void *) * NR_CPUS); 153 memset(hv_context.clk_evt, 0, 154 sizeof(void *) * NR_CPUS); 155 156 max_leaf = query_hypervisor_info(); 157 158 /* 159 * Write our OS ID. 160 */ 161 hv_context.guestid = generate_guest_id(0, LINUX_VERSION_CODE, 0); 162 wrmsrl(HV_X64_MSR_GUEST_OS_ID, hv_context.guestid); 163 164 /* See if the hypercall page is already set */ 165 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 166 167 virtaddr = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_EXEC); 168 169 if (!virtaddr) 170 goto cleanup; 171 172 hypercall_msr.enable = 1; 173 174 hypercall_msr.guest_physical_address = vmalloc_to_pfn(virtaddr); 175 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 176 177 /* Confirm that hypercall page did get setup. */ 178 hypercall_msr.as_uint64 = 0; 179 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 180 181 if (!hypercall_msr.enable) 182 goto cleanup; 183 184 hv_context.hypercall_page = virtaddr; 185 186 return 0; 187 188 cleanup: 189 if (virtaddr) { 190 if (hypercall_msr.enable) { 191 hypercall_msr.as_uint64 = 0; 192 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 193 } 194 195 vfree(virtaddr); 196 } 197 198 return -ENOTSUPP; 199 } 200 201 /* 202 * hv_cleanup - Cleanup routine. 203 * 204 * This routine is called normally during driver unloading or exiting. 205 */ 206 void hv_cleanup(void) 207 { 208 union hv_x64_msr_hypercall_contents hypercall_msr; 209 210 /* Reset our OS id */ 211 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0); 212 213 if (hv_context.hypercall_page) { 214 hypercall_msr.as_uint64 = 0; 215 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 216 vfree(hv_context.hypercall_page); 217 hv_context.hypercall_page = NULL; 218 } 219 } 220 221 /* 222 * hv_post_message - Post a message using the hypervisor message IPC. 223 * 224 * This involves a hypercall. 225 */ 226 int hv_post_message(union hv_connection_id connection_id, 227 enum hv_message_type message_type, 228 void *payload, size_t payload_size) 229 { 230 231 struct hv_input_post_message *aligned_msg; 232 u16 status; 233 234 if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT) 235 return -EMSGSIZE; 236 237 aligned_msg = (struct hv_input_post_message *) 238 hv_context.post_msg_page[get_cpu()]; 239 240 aligned_msg->connectionid = connection_id; 241 aligned_msg->reserved = 0; 242 aligned_msg->message_type = message_type; 243 aligned_msg->payload_size = payload_size; 244 memcpy((void *)aligned_msg->payload, payload, payload_size); 245 246 status = do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL) 247 & 0xFFFF; 248 249 put_cpu(); 250 return status; 251 } 252 253 254 /* 255 * hv_signal_event - 256 * Signal an event on the specified connection using the hypervisor event IPC. 257 * 258 * This involves a hypercall. 259 */ 260 u16 hv_signal_event(void *con_id) 261 { 262 u16 status; 263 264 status = (do_hypercall(HVCALL_SIGNAL_EVENT, con_id, NULL) & 0xFFFF); 265 266 return status; 267 } 268 269 static int hv_ce_set_next_event(unsigned long delta, 270 struct clock_event_device *evt) 271 { 272 cycle_t current_tick; 273 274 WARN_ON(evt->mode != CLOCK_EVT_MODE_ONESHOT); 275 276 rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick); 277 current_tick += delta; 278 wrmsrl(HV_X64_MSR_STIMER0_COUNT, current_tick); 279 return 0; 280 } 281 282 static void hv_ce_setmode(enum clock_event_mode mode, 283 struct clock_event_device *evt) 284 { 285 union hv_timer_config timer_cfg; 286 287 switch (mode) { 288 case CLOCK_EVT_MODE_PERIODIC: 289 /* unsupported */ 290 break; 291 292 case CLOCK_EVT_MODE_ONESHOT: 293 timer_cfg.enable = 1; 294 timer_cfg.auto_enable = 1; 295 timer_cfg.sintx = VMBUS_MESSAGE_SINT; 296 wrmsrl(HV_X64_MSR_STIMER0_CONFIG, timer_cfg.as_uint64); 297 break; 298 299 case CLOCK_EVT_MODE_UNUSED: 300 case CLOCK_EVT_MODE_SHUTDOWN: 301 wrmsrl(HV_X64_MSR_STIMER0_COUNT, 0); 302 wrmsrl(HV_X64_MSR_STIMER0_CONFIG, 0); 303 break; 304 case CLOCK_EVT_MODE_RESUME: 305 break; 306 } 307 } 308 309 static void hv_init_clockevent_device(struct clock_event_device *dev, int cpu) 310 { 311 dev->name = "Hyper-V clockevent"; 312 dev->features = CLOCK_EVT_FEAT_ONESHOT; 313 dev->cpumask = cpumask_of(cpu); 314 dev->rating = 1000; 315 dev->owner = THIS_MODULE; 316 317 dev->set_mode = hv_ce_setmode; 318 dev->set_next_event = hv_ce_set_next_event; 319 } 320 321 322 int hv_synic_alloc(void) 323 { 324 size_t size = sizeof(struct tasklet_struct); 325 size_t ced_size = sizeof(struct clock_event_device); 326 int cpu; 327 328 for_each_online_cpu(cpu) { 329 hv_context.event_dpc[cpu] = kmalloc(size, GFP_ATOMIC); 330 if (hv_context.event_dpc[cpu] == NULL) { 331 pr_err("Unable to allocate event dpc\n"); 332 goto err; 333 } 334 tasklet_init(hv_context.event_dpc[cpu], vmbus_on_event, cpu); 335 336 hv_context.clk_evt[cpu] = kzalloc(ced_size, GFP_ATOMIC); 337 if (hv_context.clk_evt[cpu] == NULL) { 338 pr_err("Unable to allocate clock event device\n"); 339 goto err; 340 } 341 hv_init_clockevent_device(hv_context.clk_evt[cpu], cpu); 342 343 hv_context.synic_message_page[cpu] = 344 (void *)get_zeroed_page(GFP_ATOMIC); 345 346 if (hv_context.synic_message_page[cpu] == NULL) { 347 pr_err("Unable to allocate SYNIC message page\n"); 348 goto err; 349 } 350 351 hv_context.synic_event_page[cpu] = 352 (void *)get_zeroed_page(GFP_ATOMIC); 353 354 if (hv_context.synic_event_page[cpu] == NULL) { 355 pr_err("Unable to allocate SYNIC event page\n"); 356 goto err; 357 } 358 359 hv_context.post_msg_page[cpu] = 360 (void *)get_zeroed_page(GFP_ATOMIC); 361 362 if (hv_context.post_msg_page[cpu] == NULL) { 363 pr_err("Unable to allocate post msg page\n"); 364 goto err; 365 } 366 } 367 368 return 0; 369 err: 370 return -ENOMEM; 371 } 372 373 static void hv_synic_free_cpu(int cpu) 374 { 375 kfree(hv_context.event_dpc[cpu]); 376 kfree(hv_context.clk_evt[cpu]); 377 if (hv_context.synic_event_page[cpu]) 378 free_page((unsigned long)hv_context.synic_event_page[cpu]); 379 if (hv_context.synic_message_page[cpu]) 380 free_page((unsigned long)hv_context.synic_message_page[cpu]); 381 if (hv_context.post_msg_page[cpu]) 382 free_page((unsigned long)hv_context.post_msg_page[cpu]); 383 } 384 385 void hv_synic_free(void) 386 { 387 int cpu; 388 389 for_each_online_cpu(cpu) 390 hv_synic_free_cpu(cpu); 391 } 392 393 /* 394 * hv_synic_init - Initialize the Synthethic Interrupt Controller. 395 * 396 * If it is already initialized by another entity (ie x2v shim), we need to 397 * retrieve the initialized message and event pages. Otherwise, we create and 398 * initialize the message and event pages. 399 */ 400 void hv_synic_init(void *arg) 401 { 402 u64 version; 403 union hv_synic_simp simp; 404 union hv_synic_siefp siefp; 405 union hv_synic_sint shared_sint; 406 union hv_synic_scontrol sctrl; 407 u64 vp_index; 408 409 int cpu = smp_processor_id(); 410 411 if (!hv_context.hypercall_page) 412 return; 413 414 /* Check the version */ 415 rdmsrl(HV_X64_MSR_SVERSION, version); 416 417 /* Setup the Synic's message page */ 418 rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64); 419 simp.simp_enabled = 1; 420 simp.base_simp_gpa = virt_to_phys(hv_context.synic_message_page[cpu]) 421 >> PAGE_SHIFT; 422 423 wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64); 424 425 /* Setup the Synic's event page */ 426 rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64); 427 siefp.siefp_enabled = 1; 428 siefp.base_siefp_gpa = virt_to_phys(hv_context.synic_event_page[cpu]) 429 >> PAGE_SHIFT; 430 431 wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64); 432 433 /* Setup the shared SINT. */ 434 rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64); 435 436 shared_sint.as_uint64 = 0; 437 shared_sint.vector = HYPERVISOR_CALLBACK_VECTOR; 438 shared_sint.masked = false; 439 shared_sint.auto_eoi = true; 440 441 wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64); 442 443 /* Enable the global synic bit */ 444 rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64); 445 sctrl.enable = 1; 446 447 wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64); 448 449 hv_context.synic_initialized = true; 450 451 /* 452 * Setup the mapping between Hyper-V's notion 453 * of cpuid and Linux' notion of cpuid. 454 * This array will be indexed using Linux cpuid. 455 */ 456 rdmsrl(HV_X64_MSR_VP_INDEX, vp_index); 457 hv_context.vp_index[cpu] = (u32)vp_index; 458 459 INIT_LIST_HEAD(&hv_context.percpu_list[cpu]); 460 461 /* 462 * Register the per-cpu clockevent source. 463 */ 464 if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE) 465 clockevents_config_and_register(hv_context.clk_evt[cpu], 466 HV_TIMER_FREQUENCY, 467 HV_MIN_DELTA_TICKS, 468 HV_MAX_MAX_DELTA_TICKS); 469 return; 470 } 471 472 /* 473 * hv_synic_cleanup - Cleanup routine for hv_synic_init(). 474 */ 475 void hv_synic_cleanup(void *arg) 476 { 477 union hv_synic_sint shared_sint; 478 union hv_synic_simp simp; 479 union hv_synic_siefp siefp; 480 int cpu = smp_processor_id(); 481 482 if (!hv_context.synic_initialized) 483 return; 484 485 rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64); 486 487 shared_sint.masked = 1; 488 489 /* Need to correctly cleanup in the case of SMP!!! */ 490 /* Disable the interrupt */ 491 wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64); 492 493 rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64); 494 simp.simp_enabled = 0; 495 simp.base_simp_gpa = 0; 496 497 wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64); 498 499 rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64); 500 siefp.siefp_enabled = 0; 501 siefp.base_siefp_gpa = 0; 502 503 wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64); 504 505 free_page((unsigned long)hv_context.synic_message_page[cpu]); 506 free_page((unsigned long)hv_context.synic_event_page[cpu]); 507 } 508