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 <asm/hyperv.h> 32#include "hyperv_vmbus.h" 33 34/* The one and only */ 35struct hv_context hv_context = { 36 .synic_initialized = false, 37 .hypercall_page = NULL, 38}; 39 40/* 41 * query_hypervisor_info - Get version info of the windows hypervisor 42 */ 43unsigned int host_info_eax; 44unsigned int host_info_ebx; 45unsigned int host_info_ecx; 46unsigned int host_info_edx; 47 48static int query_hypervisor_info(void) 49{ 50 unsigned int eax; 51 unsigned int ebx; 52 unsigned int ecx; 53 unsigned int edx; 54 unsigned int max_leaf; 55 unsigned int op; 56 57 /* 58 * Its assumed that this is called after confirming that Viridian 59 * is present. Query id and revision. 60 */ 61 eax = 0; 62 ebx = 0; 63 ecx = 0; 64 edx = 0; 65 op = HVCPUID_VENDOR_MAXFUNCTION; 66 cpuid(op, &eax, &ebx, &ecx, &edx); 67 68 max_leaf = eax; 69 70 if (max_leaf >= HVCPUID_VERSION) { 71 eax = 0; 72 ebx = 0; 73 ecx = 0; 74 edx = 0; 75 op = HVCPUID_VERSION; 76 cpuid(op, &eax, &ebx, &ecx, &edx); 77 host_info_eax = eax; 78 host_info_ebx = ebx; 79 host_info_ecx = ecx; 80 host_info_edx = edx; 81 } 82 return max_leaf; 83} 84 85/* 86 * do_hypercall- Invoke the specified hypercall 87 */ 88static u64 do_hypercall(u64 control, void *input, void *output) 89{ 90#ifdef CONFIG_X86_64 91 u64 hv_status = 0; 92 u64 input_address = (input) ? virt_to_phys(input) : 0; 93 u64 output_address = (output) ? virt_to_phys(output) : 0; 94 void *hypercall_page = hv_context.hypercall_page; 95 96 __asm__ __volatile__("mov %0, %%r8" : : "r" (output_address) : "r8"); 97 __asm__ __volatile__("call *%3" : "=a" (hv_status) : 98 "c" (control), "d" (input_address), 99 "m" (hypercall_page)); 100 101 return hv_status; 102 103#else 104 105 u32 control_hi = control >> 32; 106 u32 control_lo = control & 0xFFFFFFFF; 107 u32 hv_status_hi = 1; 108 u32 hv_status_lo = 1; 109 u64 input_address = (input) ? virt_to_phys(input) : 0; 110 u32 input_address_hi = input_address >> 32; 111 u32 input_address_lo = input_address & 0xFFFFFFFF; 112 u64 output_address = (output) ? virt_to_phys(output) : 0; 113 u32 output_address_hi = output_address >> 32; 114 u32 output_address_lo = output_address & 0xFFFFFFFF; 115 void *hypercall_page = hv_context.hypercall_page; 116 117 __asm__ __volatile__ ("call *%8" : "=d"(hv_status_hi), 118 "=a"(hv_status_lo) : "d" (control_hi), 119 "a" (control_lo), "b" (input_address_hi), 120 "c" (input_address_lo), "D"(output_address_hi), 121 "S"(output_address_lo), "m" (hypercall_page)); 122 123 return hv_status_lo | ((u64)hv_status_hi << 32); 124#endif /* !x86_64 */ 125} 126 127/* 128 * hv_init - Main initialization routine. 129 * 130 * This routine must be called before any other routines in here are called 131 */ 132int hv_init(void) 133{ 134 int max_leaf; 135 union hv_x64_msr_hypercall_contents hypercall_msr; 136 void *virtaddr = NULL; 137 138 memset(hv_context.synic_event_page, 0, sizeof(void *) * NR_CPUS); 139 memset(hv_context.synic_message_page, 0, 140 sizeof(void *) * NR_CPUS); 141 memset(hv_context.post_msg_page, 0, 142 sizeof(void *) * NR_CPUS); 143 memset(hv_context.vp_index, 0, 144 sizeof(int) * NR_CPUS); 145 memset(hv_context.event_dpc, 0, 146 sizeof(void *) * NR_CPUS); 147 148 max_leaf = query_hypervisor_info(); 149 150 /* 151 * Write our OS ID. 152 */ 153 hv_context.guestid = generate_guest_id(0, LINUX_VERSION_CODE, 0); 154 wrmsrl(HV_X64_MSR_GUEST_OS_ID, hv_context.guestid); 155 156 /* See if the hypercall page is already set */ 157 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 158 159 virtaddr = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_EXEC); 160 161 if (!virtaddr) 162 goto cleanup; 163 164 hypercall_msr.enable = 1; 165 166 hypercall_msr.guest_physical_address = vmalloc_to_pfn(virtaddr); 167 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 168 169 /* Confirm that hypercall page did get setup. */ 170 hypercall_msr.as_uint64 = 0; 171 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 172 173 if (!hypercall_msr.enable) 174 goto cleanup; 175 176 hv_context.hypercall_page = virtaddr; 177 178 return 0; 179 180cleanup: 181 if (virtaddr) { 182 if (hypercall_msr.enable) { 183 hypercall_msr.as_uint64 = 0; 184 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 185 } 186 187 vfree(virtaddr); 188 } 189 190 return -ENOTSUPP; 191} 192 193/* 194 * hv_cleanup - Cleanup routine. 195 * 196 * This routine is called normally during driver unloading or exiting. 197 */ 198void hv_cleanup(void) 199{ 200 union hv_x64_msr_hypercall_contents hypercall_msr; 201 202 /* Reset our OS id */ 203 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0); 204 205 if (hv_context.hypercall_page) { 206 hypercall_msr.as_uint64 = 0; 207 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 208 vfree(hv_context.hypercall_page); 209 hv_context.hypercall_page = NULL; 210 } 211} 212 213/* 214 * hv_post_message - Post a message using the hypervisor message IPC. 215 * 216 * This involves a hypercall. 217 */ 218int hv_post_message(union hv_connection_id connection_id, 219 enum hv_message_type message_type, 220 void *payload, size_t payload_size) 221{ 222 223 struct hv_input_post_message *aligned_msg; 224 u16 status; 225 226 if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT) 227 return -EMSGSIZE; 228 229 aligned_msg = (struct hv_input_post_message *) 230 hv_context.post_msg_page[get_cpu()]; 231 232 aligned_msg->connectionid = connection_id; 233 aligned_msg->reserved = 0; 234 aligned_msg->message_type = message_type; 235 aligned_msg->payload_size = payload_size; 236 memcpy((void *)aligned_msg->payload, payload, payload_size); 237 238 status = do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL) 239 & 0xFFFF; 240 241 put_cpu(); 242 return status; 243} 244 245 246/* 247 * hv_signal_event - 248 * Signal an event on the specified connection using the hypervisor event IPC. 249 * 250 * This involves a hypercall. 251 */ 252u16 hv_signal_event(void *con_id) 253{ 254 u16 status; 255 256 status = (do_hypercall(HVCALL_SIGNAL_EVENT, con_id, NULL) & 0xFFFF); 257 258 return status; 259} 260 261 262int hv_synic_alloc(void) 263{ 264 size_t size = sizeof(struct tasklet_struct); 265 int cpu; 266 267 for_each_online_cpu(cpu) { 268 hv_context.event_dpc[cpu] = kmalloc(size, GFP_ATOMIC); 269 if (hv_context.event_dpc[cpu] == NULL) { 270 pr_err("Unable to allocate event dpc\n"); 271 goto err; 272 } 273 tasklet_init(hv_context.event_dpc[cpu], vmbus_on_event, cpu); 274 275 hv_context.synic_message_page[cpu] = 276 (void *)get_zeroed_page(GFP_ATOMIC); 277 278 if (hv_context.synic_message_page[cpu] == NULL) { 279 pr_err("Unable to allocate SYNIC message page\n"); 280 goto err; 281 } 282 283 hv_context.synic_event_page[cpu] = 284 (void *)get_zeroed_page(GFP_ATOMIC); 285 286 if (hv_context.synic_event_page[cpu] == NULL) { 287 pr_err("Unable to allocate SYNIC event page\n"); 288 goto err; 289 } 290 291 hv_context.post_msg_page[cpu] = 292 (void *)get_zeroed_page(GFP_ATOMIC); 293 294 if (hv_context.post_msg_page[cpu] == NULL) { 295 pr_err("Unable to allocate post msg page\n"); 296 goto err; 297 } 298 } 299 300 return 0; 301err: 302 return -ENOMEM; 303} 304 305static void hv_synic_free_cpu(int cpu) 306{ 307 kfree(hv_context.event_dpc[cpu]); 308 if (hv_context.synic_event_page[cpu]) 309 free_page((unsigned long)hv_context.synic_event_page[cpu]); 310 if (hv_context.synic_message_page[cpu]) 311 free_page((unsigned long)hv_context.synic_message_page[cpu]); 312 if (hv_context.post_msg_page[cpu]) 313 free_page((unsigned long)hv_context.post_msg_page[cpu]); 314} 315 316void hv_synic_free(void) 317{ 318 int cpu; 319 320 for_each_online_cpu(cpu) 321 hv_synic_free_cpu(cpu); 322} 323 324/* 325 * hv_synic_init - Initialize the Synthethic Interrupt Controller. 326 * 327 * If it is already initialized by another entity (ie x2v shim), we need to 328 * retrieve the initialized message and event pages. Otherwise, we create and 329 * initialize the message and event pages. 330 */ 331void hv_synic_init(void *arg) 332{ 333 u64 version; 334 union hv_synic_simp simp; 335 union hv_synic_siefp siefp; 336 union hv_synic_sint shared_sint; 337 union hv_synic_scontrol sctrl; 338 u64 vp_index; 339 340 int cpu = smp_processor_id(); 341 342 if (!hv_context.hypercall_page) 343 return; 344 345 /* Check the version */ 346 rdmsrl(HV_X64_MSR_SVERSION, version); 347 348 /* Setup the Synic's message page */ 349 rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64); 350 simp.simp_enabled = 1; 351 simp.base_simp_gpa = virt_to_phys(hv_context.synic_message_page[cpu]) 352 >> PAGE_SHIFT; 353 354 wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64); 355 356 /* Setup the Synic's event page */ 357 rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64); 358 siefp.siefp_enabled = 1; 359 siefp.base_siefp_gpa = virt_to_phys(hv_context.synic_event_page[cpu]) 360 >> PAGE_SHIFT; 361 362 wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64); 363 364 /* Setup the shared SINT. */ 365 rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64); 366 367 shared_sint.as_uint64 = 0; 368 shared_sint.vector = HYPERVISOR_CALLBACK_VECTOR; 369 shared_sint.masked = false; 370 shared_sint.auto_eoi = true; 371 372 wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64); 373 374 /* Enable the global synic bit */ 375 rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64); 376 sctrl.enable = 1; 377 378 wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64); 379 380 hv_context.synic_initialized = true; 381 382 /* 383 * Setup the mapping between Hyper-V's notion 384 * of cpuid and Linux' notion of cpuid. 385 * This array will be indexed using Linux cpuid. 386 */ 387 rdmsrl(HV_X64_MSR_VP_INDEX, vp_index); 388 hv_context.vp_index[cpu] = (u32)vp_index; 389 390 INIT_LIST_HEAD(&hv_context.percpu_list[cpu]); 391 return; 392} 393 394/* 395 * hv_synic_cleanup - Cleanup routine for hv_synic_init(). 396 */ 397void hv_synic_cleanup(void *arg) 398{ 399 union hv_synic_sint shared_sint; 400 union hv_synic_simp simp; 401 union hv_synic_siefp siefp; 402 int cpu = smp_processor_id(); 403 404 if (!hv_context.synic_initialized) 405 return; 406 407 rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64); 408 409 shared_sint.masked = 1; 410 411 /* Need to correctly cleanup in the case of SMP!!! */ 412 /* Disable the interrupt */ 413 wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64); 414 415 rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64); 416 simp.simp_enabled = 0; 417 simp.base_simp_gpa = 0; 418 419 wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64); 420 421 rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64); 422 siefp.siefp_enabled = 0; 423 siefp.base_siefp_gpa = 0; 424 425 wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64); 426 427 free_page((unsigned long)hv_context.synic_message_page[cpu]); 428 free_page((unsigned long)hv_context.synic_event_page[cpu]); 429} 430