hv.c revision b0209501dc7586cbfbf6d023f2dd3ce4621aff2c
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 */ 43static int query_hypervisor_info(void) 44{ 45 unsigned int eax; 46 unsigned int ebx; 47 unsigned int ecx; 48 unsigned int edx; 49 unsigned int max_leaf; 50 unsigned int op; 51 52 /* 53 * Its assumed that this is called after confirming that Viridian 54 * is present. Query id and revision. 55 */ 56 eax = 0; 57 ebx = 0; 58 ecx = 0; 59 edx = 0; 60 op = HVCPUID_VENDOR_MAXFUNCTION; 61 cpuid(op, &eax, &ebx, &ecx, &edx); 62 63 max_leaf = eax; 64 65 if (max_leaf >= HVCPUID_VERSION) { 66 eax = 0; 67 ebx = 0; 68 ecx = 0; 69 edx = 0; 70 op = HVCPUID_VERSION; 71 cpuid(op, &eax, &ebx, &ecx, &edx); 72 pr_info("Hyper-V Host OS Build:%d-%d.%d-%d-%d.%d\n", 73 eax, 74 ebx >> 16, 75 ebx & 0xFFFF, 76 ecx, 77 edx >> 24, 78 edx & 0xFFFFFF); 79 } 80 return max_leaf; 81} 82 83/* 84 * do_hypercall- Invoke the specified hypercall 85 */ 86static u64 do_hypercall(u64 control, void *input, void *output) 87{ 88#ifdef CONFIG_X86_64 89 u64 hv_status = 0; 90 u64 input_address = (input) ? virt_to_phys(input) : 0; 91 u64 output_address = (output) ? virt_to_phys(output) : 0; 92 void *hypercall_page = hv_context.hypercall_page; 93 94 __asm__ __volatile__("mov %0, %%r8" : : "r" (output_address) : "r8"); 95 __asm__ __volatile__("call *%3" : "=a" (hv_status) : 96 "c" (control), "d" (input_address), 97 "m" (hypercall_page)); 98 99 return hv_status; 100 101#else 102 103 u32 control_hi = control >> 32; 104 u32 control_lo = control & 0xFFFFFFFF; 105 u32 hv_status_hi = 1; 106 u32 hv_status_lo = 1; 107 u64 input_address = (input) ? virt_to_phys(input) : 0; 108 u32 input_address_hi = input_address >> 32; 109 u32 input_address_lo = input_address & 0xFFFFFFFF; 110 u64 output_address = (output) ? virt_to_phys(output) : 0; 111 u32 output_address_hi = output_address >> 32; 112 u32 output_address_lo = output_address & 0xFFFFFFFF; 113 void *hypercall_page = hv_context.hypercall_page; 114 115 __asm__ __volatile__ ("call *%8" : "=d"(hv_status_hi), 116 "=a"(hv_status_lo) : "d" (control_hi), 117 "a" (control_lo), "b" (input_address_hi), 118 "c" (input_address_lo), "D"(output_address_hi), 119 "S"(output_address_lo), "m" (hypercall_page)); 120 121 return hv_status_lo | ((u64)hv_status_hi << 32); 122#endif /* !x86_64 */ 123} 124 125/* 126 * hv_init - Main initialization routine. 127 * 128 * This routine must be called before any other routines in here are called 129 */ 130int hv_init(void) 131{ 132 int max_leaf; 133 union hv_x64_msr_hypercall_contents hypercall_msr; 134 void *virtaddr = NULL; 135 136 memset(hv_context.synic_event_page, 0, sizeof(void *) * NR_CPUS); 137 memset(hv_context.synic_message_page, 0, 138 sizeof(void *) * NR_CPUS); 139 memset(hv_context.vp_index, 0, 140 sizeof(int) * NR_CPUS); 141 memset(hv_context.event_dpc, 0, 142 sizeof(void *) * NR_CPUS); 143 144 max_leaf = query_hypervisor_info(); 145 146 /* 147 * Write our OS ID. 148 */ 149 hv_context.guestid = generate_guest_id(0, LINUX_VERSION_CODE, 0); 150 wrmsrl(HV_X64_MSR_GUEST_OS_ID, hv_context.guestid); 151 152 /* See if the hypercall page is already set */ 153 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 154 155 virtaddr = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_EXEC); 156 157 if (!virtaddr) 158 goto cleanup; 159 160 hypercall_msr.enable = 1; 161 162 hypercall_msr.guest_physical_address = vmalloc_to_pfn(virtaddr); 163 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 164 165 /* Confirm that hypercall page did get setup. */ 166 hypercall_msr.as_uint64 = 0; 167 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 168 169 if (!hypercall_msr.enable) 170 goto cleanup; 171 172 hv_context.hypercall_page = virtaddr; 173 174 return 0; 175 176cleanup: 177 if (virtaddr) { 178 if (hypercall_msr.enable) { 179 hypercall_msr.as_uint64 = 0; 180 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 181 } 182 183 vfree(virtaddr); 184 } 185 186 return -ENOTSUPP; 187} 188 189/* 190 * hv_cleanup - Cleanup routine. 191 * 192 * This routine is called normally during driver unloading or exiting. 193 */ 194void hv_cleanup(void) 195{ 196 union hv_x64_msr_hypercall_contents hypercall_msr; 197 198 /* Reset our OS id */ 199 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0); 200 201 if (hv_context.hypercall_page) { 202 hypercall_msr.as_uint64 = 0; 203 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64); 204 vfree(hv_context.hypercall_page); 205 hv_context.hypercall_page = NULL; 206 } 207} 208 209/* 210 * hv_post_message - Post a message using the hypervisor message IPC. 211 * 212 * This involves a hypercall. 213 */ 214int hv_post_message(union hv_connection_id connection_id, 215 enum hv_message_type message_type, 216 void *payload, size_t payload_size) 217{ 218 struct aligned_input { 219 u64 alignment8; 220 struct hv_input_post_message msg; 221 }; 222 223 struct hv_input_post_message *aligned_msg; 224 u16 status; 225 unsigned long addr; 226 227 if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT) 228 return -EMSGSIZE; 229 230 addr = (unsigned long)kmalloc(sizeof(struct aligned_input), GFP_ATOMIC); 231 if (!addr) 232 return -ENOMEM; 233 234 aligned_msg = (struct hv_input_post_message *) 235 (ALIGN(addr, HV_HYPERCALL_PARAM_ALIGN)); 236 237 aligned_msg->connectionid = connection_id; 238 aligned_msg->message_type = message_type; 239 aligned_msg->payload_size = payload_size; 240 memcpy((void *)aligned_msg->payload, payload, payload_size); 241 242 status = do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL) 243 & 0xFFFF; 244 245 kfree((void *)addr); 246 247 return status; 248} 249 250 251/* 252 * hv_signal_event - 253 * Signal an event on the specified connection using the hypervisor event IPC. 254 * 255 * This involves a hypercall. 256 */ 257u16 hv_signal_event(void *con_id) 258{ 259 u16 status; 260 261 status = (do_hypercall(HVCALL_SIGNAL_EVENT, con_id, NULL) & 0xFFFF); 262 263 return status; 264} 265 266/* 267 * hv_synic_init - Initialize the Synthethic Interrupt Controller. 268 * 269 * If it is already initialized by another entity (ie x2v shim), we need to 270 * retrieve the initialized message and event pages. Otherwise, we create and 271 * initialize the message and event pages. 272 */ 273void hv_synic_init(void *irqarg) 274{ 275 u64 version; 276 union hv_synic_simp simp; 277 union hv_synic_siefp siefp; 278 union hv_synic_sint shared_sint; 279 union hv_synic_scontrol sctrl; 280 u64 vp_index; 281 282 u32 irq_vector = *((u32 *)(irqarg)); 283 int cpu = smp_processor_id(); 284 285 if (!hv_context.hypercall_page) 286 return; 287 288 /* Check the version */ 289 rdmsrl(HV_X64_MSR_SVERSION, version); 290 291 hv_context.event_dpc[cpu] = (struct tasklet_struct *) 292 kmalloc(sizeof(struct tasklet_struct), 293 GFP_ATOMIC); 294 if (hv_context.event_dpc[cpu] == NULL) { 295 pr_err("Unable to allocate event dpc\n"); 296 goto cleanup; 297 } 298 tasklet_init(hv_context.event_dpc[cpu], vmbus_on_event, cpu); 299 300 hv_context.synic_message_page[cpu] = 301 (void *)get_zeroed_page(GFP_ATOMIC); 302 303 if (hv_context.synic_message_page[cpu] == NULL) { 304 pr_err("Unable to allocate SYNIC message page\n"); 305 goto cleanup; 306 } 307 308 hv_context.synic_event_page[cpu] = 309 (void *)get_zeroed_page(GFP_ATOMIC); 310 311 if (hv_context.synic_event_page[cpu] == NULL) { 312 pr_err("Unable to allocate SYNIC event page\n"); 313 goto cleanup; 314 } 315 316 /* Setup the Synic's message page */ 317 rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64); 318 simp.simp_enabled = 1; 319 simp.base_simp_gpa = virt_to_phys(hv_context.synic_message_page[cpu]) 320 >> PAGE_SHIFT; 321 322 wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64); 323 324 /* Setup the Synic's event page */ 325 rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64); 326 siefp.siefp_enabled = 1; 327 siefp.base_siefp_gpa = virt_to_phys(hv_context.synic_event_page[cpu]) 328 >> PAGE_SHIFT; 329 330 wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64); 331 332 /* Setup the shared SINT. */ 333 rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64); 334 335 shared_sint.as_uint64 = 0; 336 shared_sint.vector = irq_vector; /* HV_SHARED_SINT_IDT_VECTOR + 0x20; */ 337 shared_sint.masked = false; 338 shared_sint.auto_eoi = true; 339 340 wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64); 341 342 /* Enable the global synic bit */ 343 rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64); 344 sctrl.enable = 1; 345 346 wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64); 347 348 hv_context.synic_initialized = true; 349 350 /* 351 * Setup the mapping between Hyper-V's notion 352 * of cpuid and Linux' notion of cpuid. 353 * This array will be indexed using Linux cpuid. 354 */ 355 rdmsrl(HV_X64_MSR_VP_INDEX, vp_index); 356 hv_context.vp_index[cpu] = (u32)vp_index; 357 return; 358 359cleanup: 360 if (hv_context.synic_event_page[cpu]) 361 free_page((unsigned long)hv_context.synic_event_page[cpu]); 362 363 if (hv_context.synic_message_page[cpu]) 364 free_page((unsigned long)hv_context.synic_message_page[cpu]); 365 return; 366} 367 368/* 369 * hv_synic_cleanup - Cleanup routine for hv_synic_init(). 370 */ 371void hv_synic_cleanup(void *arg) 372{ 373 union hv_synic_sint shared_sint; 374 union hv_synic_simp simp; 375 union hv_synic_siefp siefp; 376 int cpu = smp_processor_id(); 377 378 if (!hv_context.synic_initialized) 379 return; 380 381 rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64); 382 383 shared_sint.masked = 1; 384 385 /* Need to correctly cleanup in the case of SMP!!! */ 386 /* Disable the interrupt */ 387 wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64); 388 389 rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64); 390 simp.simp_enabled = 0; 391 simp.base_simp_gpa = 0; 392 393 wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64); 394 395 rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64); 396 siefp.siefp_enabled = 0; 397 siefp.base_siefp_gpa = 0; 398 399 wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64); 400 401 free_page((unsigned long)hv_context.synic_message_page[cpu]); 402 free_page((unsigned long)hv_context.synic_event_page[cpu]); 403} 404