drivers/hv/hv.c

   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 <asm/hyperv.h>
  30 #include <asm/nospec-branch.h>
  31 #include "hyperv_vmbus.h"
  32
  33 /* The one and only */
  34 struct hv_context hv_context = {
  35         .synic_initialized      = false,
  36         .hypercall_page         = NULL,
  37         .signal_event_param     = NULL,
  38         .signal_event_buffer    = NULL,
  39 };
  40
  41 /*
  42  * query_hypervisor_presence
  43  * - Query the cpuid for presence of windows hypervisor
  44  */
  45 static int query_hypervisor_presence(void)
  46 {
  47         unsigned int eax;
  48         unsigned int ebx;
  49         unsigned int ecx;
  50         unsigned int edx;
  51         unsigned int op;
  52
  53         eax = 0;
  54         ebx = 0;
  55         ecx = 0;
  56         edx = 0;
  57         op = HVCPUID_VERSION_FEATURES;
  58         cpuid(op, &eax, &ebx, &ecx, &edx);
  59
  60         return ecx & HV_PRESENT_BIT;
  61 }
  62
  63 /*
  64  * query_hypervisor_info - Get version info of the windows hypervisor
  65  */
  66 static int query_hypervisor_info(void)
  67 {
  68         unsigned int eax;
  69         unsigned int ebx;
  70         unsigned int ecx;
  71         unsigned int edx;
  72         unsigned int max_leaf;
  73         unsigned int op;
  74
  75         /*
  76         * Its assumed that this is called after confirming that Viridian
  77         * is present. Query id and revision.
  78         */
  79         eax = 0;
  80         ebx = 0;
  81         ecx = 0;
  82         edx = 0;
  83         op = HVCPUID_VENDOR_MAXFUNCTION;
  84         cpuid(op, &eax, &ebx, &ecx, &edx);
  85
  86         max_leaf = eax;
  87
  88         if (max_leaf >= HVCPUID_VERSION) {
  89                 eax = 0;
  90                 ebx = 0;
  91                 ecx = 0;
  92                 edx = 0;
  93                 op = HVCPUID_VERSION;
  94                 cpuid(op, &eax, &ebx, &ecx, &edx);
  95                 pr_info("Hyper-V Host OS Build:%d-%d.%d-%d-%d.%d\n",
  96                             eax,
  97                             ebx >> 16,
  98                             ebx & 0xFFFF,
  99                             ecx,
 100                             edx >> 24,
 101                             edx & 0xFFFFFF);
 102         }
 103         return max_leaf;
 104 }
 105
 106 /*
 107  * do_hypercall- Invoke the specified hypercall
 108  */
 109 static u64 do_hypercall(u64 control, void *input, void *output)
 110 {
 111 #ifdef CONFIG_X86_64
 112         u64 hv_status = 0;
 113         u64 input_address = (input) ? virt_to_phys(input) : 0;
 114         u64 output_address = (output) ? virt_to_phys(output) : 0;
 115         void *hypercall_page = hv_context.hypercall_page;
 116
 117         __asm__ __volatile__("mov %4, %%r8\n"
 118                              CALL_NOSPEC
 119                              : "=a" (hv_status), ASM_CALL_CONSTRAINT,
 120                                "+c" (control), "+d" (input_address)
 121                              :  "r" (output_address),
 122                                 THUNK_TARGET(hypercall_page)
 123                              : "cc", "memory", "r8", "r9", "r10", "r11");
 124
 125         return hv_status;
 126
 127 #else
 128
 129         u32 control_hi = control >> 32;
 130         u32 control_lo = control & 0xFFFFFFFF;
 131         u32 hv_status_hi = 1;
 132         u32 hv_status_lo = 1;
 133         u64 input_address = (input) ? virt_to_phys(input) : 0;
 134         u32 input_address_hi = input_address >> 32;
 135         u32 input_address_lo = input_address & 0xFFFFFFFF;
 136         u64 output_address = (output) ? virt_to_phys(output) : 0;
 137         u32 output_address_hi = output_address >> 32;
 138         u32 output_address_lo = output_address & 0xFFFFFFFF;
 139         void *hypercall_page = hv_context.hypercall_page;
 140
 141         __asm__ __volatile__(CALL_NOSPEC
 142                              : "=d" (hv_status_hi), "=a" (hv_status_lo),
 143                                "+c" (input_address_lo), ASM_CALL_CONSTRAINT
 144                              : "d" (control_hi), "a" (control_lo),
 145                                "b" (input_address_hi),
 146                                "D"(output_address_hi), "S"(output_address_lo),
 147                                THUNK_TARGET(hypercall_page)
 148                              : "cc", "memory");
 149
 150         return hv_status_lo | ((u64)hv_status_hi << 32);
 151 #endif /* !x86_64 */
 152 }
 153
 154 /*
 155  * hv_init - Main initialization routine.
 156  *
 157  * This routine must be called before any other routines in here are called
 158  */
 159 int hv_init(void)
 160 {
 161         int max_leaf;
 162         union hv_x64_msr_hypercall_contents hypercall_msr;
 163         void *virtaddr = NULL;
 164
 165         memset(hv_context.synic_event_page, 0, sizeof(void *) * MAX_NUM_CPUS);
 166         memset(hv_context.synic_message_page, 0,
 167                sizeof(void *) * MAX_NUM_CPUS);
 168         memset(hv_context.post_msg_page, 0,
 169                sizeof(void *) * MAX_NUM_CPUS);
 170
 171         if (!query_hypervisor_presence())
 172                 goto cleanup;
 173
 174         max_leaf = query_hypervisor_info();
 175
 176         rdmsrl(HV_X64_MSR_GUEST_OS_ID, hv_context.guestid);
 177
 178         if (hv_context.guestid != 0)
 179                 goto cleanup;
 180
 181         /* Write our OS info */
 182         wrmsrl(HV_X64_MSR_GUEST_OS_ID, HV_LINUX_GUEST_ID);
 183         hv_context.guestid = HV_LINUX_GUEST_ID;
 184
 185         /* See if the hypercall page is already set */
 186         rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 187
 188         virtaddr = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_RX);
 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 0;
 226
 227 cleanup:
 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
 237         return -ENOTSUPP;
 238 }
 239
 240 /*
 241  * hv_cleanup - Cleanup routine.
 242  *
 243  * This routine is called normally during driver unloading or exiting.
 244  */
 245 void 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  */
 266 u16 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
 271         struct hv_input_post_message *aligned_msg;
 272         u16 status;
 273
 274         if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT)
 275                 return -EMSGSIZE;
 276
 277         aligned_msg = (struct hv_input_post_message *)
 278                         hv_context.post_msg_page[get_cpu()];
 279
 280         aligned_msg->connectionid = connection_id;
 281         aligned_msg->reserved = 0;
 282         aligned_msg->message_type = message_type;
 283         aligned_msg->payload_size = payload_size;
 284         memcpy((void *)aligned_msg->payload, payload, payload_size);
 285
 286         status = do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL)
 287                 & 0xFFFF;
 288
 289         put_cpu();
 290         return status;
 291 }
 292
 293
 294 /*
 295  * hv_signal_event -
 296  * Signal an event on the specified connection using the hypervisor event IPC.
 297  *
 298  * This involves a hypercall.
 299  */
 300 u16 hv_signal_event(void)
 301 {
 302         u16 status;
 303
 304         status = do_hypercall(HVCALL_SIGNAL_EVENT,
 305                                hv_context.signal_event_param,
 306                                NULL) & 0xFFFF;
 307         return status;
 308 }
 309
 310 /*
 311  * hv_synic_init - Initialize the Synthethic Interrupt Controller.
 312  *
 313  * If it is already initialized by another entity (ie x2v shim), we need to
 314  * retrieve the initialized message and event pages.  Otherwise, we create and
 315  * initialize the message and event pages.
 316  */
 317 void hv_synic_init(void *irqarg)
 318 {
 319         u64 version;
 320         union hv_synic_simp simp;
 321         union hv_synic_siefp siefp;
 322         union hv_synic_sint shared_sint;
 323         union hv_synic_scontrol sctrl;
 324
 325         u32 irq_vector = *((u32 *)(irqarg));
 326         int cpu = smp_processor_id();
 327
 328         if (!hv_context.hypercall_page)
 329                 return;
 330
 331         /* Check the version */
 332         rdmsrl(HV_X64_MSR_SVERSION, version);
 333
 334         hv_context.synic_message_page[cpu] =
 335                 (void *)get_zeroed_page(GFP_ATOMIC);
 336
 337         if (hv_context.synic_message_page[cpu] == NULL) {
 338                 pr_err("Unable to allocate SYNIC message page\n");
 339                 goto cleanup;
 340         }
 341
 342         hv_context.synic_event_page[cpu] =
 343                 (void *)get_zeroed_page(GFP_ATOMIC);
 344
 345         if (hv_context.synic_event_page[cpu] == NULL) {
 346                 pr_err("Unable to allocate SYNIC event page\n");
 347                 goto cleanup;
 348         }
 349
 350         hv_context.post_msg_page[cpu] =
 351                 (void *)get_zeroed_page(GFP_ATOMIC);
 352
 353         if (hv_context.post_msg_page[cpu] == NULL) {
 354                 pr_err("Unable to allocate post msg page\n");
 355                 goto cleanup;
 356         }
 357
 358         /* Setup the Synic's message page */
 359         rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
 360         simp.simp_enabled = 1;
 361         simp.base_simp_gpa = virt_to_phys(hv_context.synic_message_page[cpu])
 362                 >> PAGE_SHIFT;
 363
 364         wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
 365
 366         /* Setup the Synic's event page */
 367         rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
 368         siefp.siefp_enabled = 1;
 369         siefp.base_siefp_gpa = virt_to_phys(hv_context.synic_event_page[cpu])
 370                 >> PAGE_SHIFT;
 371
 372         wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
 373
 374         /* Setup the shared SINT. */
 375         rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
 376
 377         shared_sint.as_uint64 = 0;
 378         shared_sint.vector = irq_vector; /* HV_SHARED_SINT_IDT_VECTOR + 0x20; */
 379         shared_sint.masked = false;
 380         shared_sint.auto_eoi = false;
 381
 382         wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
 383
 384         /* Enable the global synic bit */
 385         rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
 386         sctrl.enable = 1;
 387
 388         wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
 389
 390         hv_context.synic_initialized = true;
 391         return;
 392
 393 cleanup:
 394         if (hv_context.synic_event_page[cpu])
 395                 free_page((unsigned long)hv_context.synic_event_page[cpu]);
 396
 397         if (hv_context.synic_message_page[cpu])
 398                 free_page((unsigned long)hv_context.synic_message_page[cpu]);
 399         if (hv_context.post_msg_page[cpu])
 400                 free_page((unsigned long)hv_context.post_msg_page[cpu]);
 401         return;
 402 }
 403
 404 /*
 405  * hv_synic_cleanup - Cleanup routine for hv_synic_init().
 406  */
 407 void hv_synic_cleanup(void *arg)
 408 {
 409         union hv_synic_sint shared_sint;
 410         union hv_synic_simp simp;
 411         union hv_synic_siefp siefp;
 412         int cpu = smp_processor_id();
 413
 414         if (!hv_context.synic_initialized)
 415                 return;
 416
 417         rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
 418
 419         shared_sint.masked = 1;
 420
 421         /* Need to correctly cleanup in the case of SMP!!! */
 422         /* Disable the interrupt */
 423         wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
 424
 425         rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
 426         simp.simp_enabled = 0;
 427         simp.base_simp_gpa = 0;
 428
 429         wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
 430
 431         rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
 432         siefp.siefp_enabled = 0;
 433         siefp.base_siefp_gpa = 0;
 434
 435         wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
 436
 437         free_page((unsigned long)hv_context.synic_message_page[cpu]);
 438         free_page((unsigned long)hv_context.synic_event_page[cpu]);
 439         free_page((unsigned long)hv_context.post_msg_page[cpu]);
 440 }