hv.c revision dec317fd6accfdba4daff13aa4d9f5abde02b347
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 29#include "hyperv.h" 30#include "hyperv_vmbus.h" 31 32/* The one and only */ 33struct hv_context hv_context = { 34 .synic_initialized = false, 35 .hypercall_page = NULL, 36 .signal_event_param = NULL, 37 .signal_event_buffer = NULL, 38}; 39 40/* 41 * query_hypervisor_presence 42 * - Query the cpuid for presence of windows hypervisor 43 */ 44static int query_hypervisor_presence(void) 45{ 46 unsigned int eax; 47 unsigned int ebx; 48 unsigned int ecx; 49 unsigned int edx; 50 unsigned int op; 51 52 eax = 0; 53 ebx = 0; 54 ecx = 0; 55 edx = 0; 56 op = HVCPUID_VERSION_FEATURES; 57 cpuid(op, &eax, &ebx, &ecx, &edx); 58 59 return ecx & HV_PRESENT_BIT; 60} 61 62/* 63 * query_hypervisor_info - Get version info of the windows hypervisor 64 */ 65static int query_hypervisor_info(void) 66{ 67 unsigned int eax; 68 unsigned int ebx; 69 unsigned int ecx; 70 unsigned int edx; 71 unsigned int max_leaf; 72 unsigned int op; 73 74 /* 75 * Its assumed that this is called after confirming that Viridian 76 * is present. Query id and revision. 77 */ 78 eax = 0; 79 ebx = 0; 80 ecx = 0; 81 edx = 0; 82 op = HVCPUID_VENDOR_MAXFUNCTION; 83 cpuid(op, &eax, &ebx, &ecx, &edx); 84 85 max_leaf = eax; 86 87 if (max_leaf >= HVCPUID_VERSION) { 88 eax = 0; 89 ebx = 0; 90 ecx = 0; 91 edx = 0; 92 op = HVCPUID_VERSION; 93 cpuid(op, &eax, &ebx, &ecx, &edx); 94 pr_info("Hyper-V Host OS Build:%d-%d.%d-%d-%d.%d\n", 95 eax, 96 ebx >> 16, 97 ebx & 0xFFFF, 98 ecx, 99 edx >> 24, 100 edx & 0xFFFFFF); 101 } 102 return max_leaf; 103} 104 105/* 106 * do_hypercall- Invoke the specified hypercall 107 */ 108static u64 do_hypercall(u64 control, void *input, void *output) 109{ 110#ifdef CONFIG_X86_64 111 u64 hv_status = 0; 112 u64 input_address = (input) ? virt_to_phys(input) : 0; 113 u64 output_address = (output) ? virt_to_phys(output) : 0; 114 void *hypercall_page = hv_context.hypercall_page; 115 116 __asm__ __volatile__("mov %0, %%r8" : : "r" (output_address) : "r8"); 117 __asm__ __volatile__("call *%3" : "=a" (hv_status) : 118 "c" (control), "d" (input_address), 119 "m" (hypercall_page)); 120 121 return hv_status; 122 123#else 124 125 u32 control_hi = control >> 32; 126 u32 control_lo = control & 0xFFFFFFFF; 127 u32 hv_status_hi = 1; 128 u32 hv_status_lo = 1; 129 u64 input_address = (input) ? virt_to_phys(input) : 0; 130 u32 input_address_hi = input_address >> 32; 131 u32 input_address_lo = input_address & 0xFFFFFFFF; 132 u64 output_address = (output) ? virt_to_phys(output) : 0; 133 u32 output_address_hi = output_address >> 32; 134 u32 output_address_lo = output_address & 0xFFFFFFFF; 135 void *hypercall_page = hv_context.hypercall_page; 136 137 __asm__ __volatile__ ("call *%8" : "=d"(hv_status_hi), 138 "=a"(hv_status_lo) : "d" (control_hi), 139 "a" (control_lo), "b" (input_address_hi), 140 "c" (input_address_lo), "D"(output_address_hi), 141 "S"(output_address_lo), "m" (hypercall_page)); 142 143 return hv_status_lo | ((u64)hv_status_hi << 32); 144#endif /* !x86_64 */ 145} 146 147/* 148 * hv_init - Main initialization routine. 149 * 150 * This routine must be called before any other routines in here are called 151 */ 152int hv_init(void) 153{ 154 int ret = 0; 155 int max_leaf; 156 union hv_x64_msr_hypercall_contents hypercall_msr; 157 void *virtaddr = NULL; 158 159 memset(hv_context.synic_event_page, 0, sizeof(void *) * MAX_NUM_CPUS); 160 memset(hv_context.synic_message_page, 0, 161 sizeof(void *) * MAX_NUM_CPUS); 162 163 if (!query_hypervisor_presence()) 164 goto cleanup; 165 166 max_leaf = query_hypervisor_info(); 167 /* HvQueryHypervisorFeatures(maxLeaf); */ 168 169 /* 170 * We only support running on top of Hyper-V 171 */ 172 rdmsrl(HV_X64_MSR_GUEST_OS_ID, hv_context.guestid); 173 174 if (hv_context.guestid != 0) 175 goto cleanup; 176 177 /* Write our OS info */ 178 wrmsrl(HV_X64_MSR_GUEST_OS_ID, HV_LINUX_GUEST_ID); 179 hv_context.guestid = HV_LINUX_GUEST_ID; 180 181 /* See if the hypercall page is already set */ 182 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 183 184 /* 185 * Allocate the hypercall page memory 186 * virtaddr = osd_page_alloc(1); 187 */ 188 virtaddr = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_EXEC); 189 190 if (!virtaddr) 191 goto cleanup; 192 193 hypercall_msr.enable = 1; 194 195 hypercall_msr.guest_physical_address = vmalloc_to_pfn(virtaddr); 196 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 197 198 /* Confirm that hypercall page did get setup. */ 199 hypercall_msr.as_uint64 = 0; 200 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 201 202 if (!hypercall_msr.enable) 203 goto cleanup; 204 205 hv_context.hypercall_page = virtaddr; 206 207 /* Setup the global signal event param for the signal event hypercall */ 208 hv_context.signal_event_buffer = 209 kmalloc(sizeof(struct hv_input_signal_event_buffer), 210 GFP_KERNEL); 211 if (!hv_context.signal_event_buffer) 212 goto cleanup; 213 214 hv_context.signal_event_param = 215 (struct hv_input_signal_event *) 216 (ALIGN((unsigned long) 217 hv_context.signal_event_buffer, 218 HV_HYPERCALL_PARAM_ALIGN)); 219 hv_context.signal_event_param->connectionid.asu32 = 0; 220 hv_context.signal_event_param->connectionid.u.id = 221 VMBUS_EVENT_CONNECTION_ID; 222 hv_context.signal_event_param->flag_number = 0; 223 hv_context.signal_event_param->rsvdz = 0; 224 225 return ret; 226 227cleanup: 228 if (virtaddr) { 229 if (hypercall_msr.enable) { 230 hypercall_msr.as_uint64 = 0; 231 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 232 } 233 234 vfree(virtaddr); 235 } 236 ret = -1; 237 return ret; 238} 239 240/* 241 * hv_cleanup - Cleanup routine. 242 * 243 * This routine is called normally during driver unloading or exiting. 244 */ 245void hv_cleanup(void) 246{ 247 union hv_x64_msr_hypercall_contents hypercall_msr; 248 249 kfree(hv_context.signal_event_buffer); 250 hv_context.signal_event_buffer = NULL; 251 hv_context.signal_event_param = NULL; 252 253 if (hv_context.hypercall_page) { 254 hypercall_msr.as_uint64 = 0; 255 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 256 vfree(hv_context.hypercall_page); 257 hv_context.hypercall_page = NULL; 258 } 259} 260 261/* 262 * hv_post_message - Post a message using the hypervisor message IPC. 263 * 264 * This involves a hypercall. 265 */ 266u16 hv_post_message(union hv_connection_id connection_id, 267 enum hv_message_type message_type, 268 void *payload, size_t payload_size) 269{ 270 struct aligned_input { 271 u64 alignment8; 272 struct hv_input_post_message msg; 273 }; 274 275 struct hv_input_post_message *aligned_msg; 276 u16 status; 277 unsigned long addr; 278 279 if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT) 280 return -EMSGSIZE; 281 282 addr = (unsigned long)kmalloc(sizeof(struct aligned_input), GFP_ATOMIC); 283 if (!addr) 284 return -ENOMEM; 285 286 aligned_msg = (struct hv_input_post_message *) 287 (ALIGN(addr, HV_HYPERCALL_PARAM_ALIGN)); 288 289 aligned_msg->connectionid = connection_id; 290 aligned_msg->message_type = message_type; 291 aligned_msg->payload_size = payload_size; 292 memcpy((void *)aligned_msg->payload, payload, payload_size); 293 294 status = do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL) 295 & 0xFFFF; 296 297 kfree((void *)addr); 298 299 return status; 300} 301 302 303/* 304 * hv_signal_event - 305 * Signal an event on the specified connection using the hypervisor event IPC. 306 * 307 * This involves a hypercall. 308 */ 309u16 hv_signal_event(void) 310{ 311 u16 status; 312 313 status = do_hypercall(HVCALL_SIGNAL_EVENT, 314 hv_context.signal_event_param, 315 NULL) & 0xFFFF; 316 return status; 317} 318 319/* 320 * hv_synic_init - Initialize the Synthethic Interrupt Controller. 321 * 322 * If it is already initialized by another entity (ie x2v shim), we need to 323 * retrieve the initialized message and event pages. Otherwise, we create and 324 * initialize the message and event pages. 325 */ 326void hv_synic_init(void *irqarg) 327{ 328 u64 version; 329 union hv_synic_simp simp; 330 union hv_synic_siefp siefp; 331 union hv_synic_sint shared_sint; 332 union hv_synic_scontrol sctrl; 333 334 u32 irq_vector = *((u32 *)(irqarg)); 335 int cpu = smp_processor_id(); 336 337 if (!hv_context.hypercall_page) 338 return; 339 340 /* Check the version */ 341 rdmsrl(HV_X64_MSR_SVERSION, version); 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 cleanup; 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 cleanup; 357 } 358 359 /* Setup the Synic's message page */ 360 rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64); 361 simp.simp_enabled = 1; 362 simp.base_simp_gpa = virt_to_phys(hv_context.synic_message_page[cpu]) 363 >> PAGE_SHIFT; 364 365 wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64); 366 367 /* Setup the Synic's event page */ 368 rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64); 369 siefp.siefp_enabled = 1; 370 siefp.base_siefp_gpa = virt_to_phys(hv_context.synic_event_page[cpu]) 371 >> PAGE_SHIFT; 372 373 wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64); 374 375 /* Setup the shared SINT. */ 376 rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64); 377 378 shared_sint.as_uint64 = 0; 379 shared_sint.vector = irq_vector; /* HV_SHARED_SINT_IDT_VECTOR + 0x20; */ 380 shared_sint.masked = false; 381 shared_sint.auto_eoi = true; 382 383 wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64); 384 385 /* Enable the global synic bit */ 386 rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64); 387 sctrl.enable = 1; 388 389 wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64); 390 391 hv_context.synic_initialized = true; 392 return; 393 394cleanup: 395 if (hv_context.synic_event_page[cpu]) 396 free_page((unsigned long)hv_context.synic_event_page[cpu]); 397 398 if (hv_context.synic_message_page[cpu]) 399 free_page((unsigned long)hv_context.synic_message_page[cpu]); 400 return; 401} 402 403/* 404 * hv_synic_cleanup - Cleanup routine for hv_synic_init(). 405 */ 406void hv_synic_cleanup(void *arg) 407{ 408 union hv_synic_sint shared_sint; 409 union hv_synic_simp simp; 410 union hv_synic_siefp siefp; 411 int cpu = smp_processor_id(); 412 413 if (!hv_context.synic_initialized) 414 return; 415 416 rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64); 417 418 shared_sint.masked = 1; 419 420 /* Need to correctly cleanup in the case of SMP!!! */ 421 /* Disable the interrupt */ 422 wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64); 423 424 rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64); 425 simp.simp_enabled = 0; 426 simp.base_simp_gpa = 0; 427 428 wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64); 429 430 rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64); 431 siefp.siefp_enabled = 0; 432 siefp.base_siefp_gpa = 0; 433 434 wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64); 435 436 free_page((unsigned long)hv_context.synic_message_page[cpu]); 437 free_page((unsigned long)hv_context.synic_event_page[cpu]); 438} 439