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 <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         u64 input_address = (input) ? virt_to_phys(input) : 0;
  97         u64 output_address = (output) ? virt_to_phys(output) : 0;
  98         void *hypercall_page = hv_context.hypercall_page;
  99 #ifdef CONFIG_X86_64
 100         u64 hv_status = 0;
 101
 102         if (!hypercall_page)
 103                 return (u64)ULLONG_MAX;
 104
 105         __asm__ __volatile__("mov %0, %%r8" : : "r" (output_address) : "r8");
 106         __asm__ __volatile__("call *%3" : "=a" (hv_status) :
 107                              "c" (control), "d" (input_address),
 108                              "m" (hypercall_page));
 109
 110         return hv_status;
 111
 112 #else
 113
 114         u32 control_hi = control >> 32;
 115         u32 control_lo = control & 0xFFFFFFFF;
 116         u32 hv_status_hi = 1;
 117         u32 hv_status_lo = 1;
 118         u32 input_address_hi = input_address >> 32;
 119         u32 input_address_lo = input_address & 0xFFFFFFFF;
 120         u32 output_address_hi = output_address >> 32;
 121         u32 output_address_lo = output_address & 0xFFFFFFFF;
 122
 123         if (!hypercall_page)
 124                 return (u64)ULLONG_MAX;
 125
 126         __asm__ __volatile__ ("call *%8" : "=d"(hv_status_hi),
 127                               "=a"(hv_status_lo) : "d" (control_hi),
 128                               "a" (control_lo), "b" (input_address_hi),
 129                               "c" (input_address_lo), "D"(output_address_hi),
 130                               "S"(output_address_lo), "m" (hypercall_page));
 131
 132         return hv_status_lo | ((u64)hv_status_hi << 32);
 133 #endif /* !x86_64 */
 134 }
 135
 136 /*
 137  * hv_init - Main initialization routine.
 138  *
 139  * This routine must be called before any other routines in here are called
 140  */
 141 int hv_init(void)
 142 {
 143         int max_leaf;
 144         union hv_x64_msr_hypercall_contents hypercall_msr;
 145         void *virtaddr = NULL;
 146
 147         memset(hv_context.synic_event_page, 0, sizeof(void *) * NR_CPUS);
 148         memset(hv_context.synic_message_page, 0,
 149                sizeof(void *) * NR_CPUS);
 150         memset(hv_context.post_msg_page, 0,
 151                sizeof(void *) * NR_CPUS);
 152         memset(hv_context.vp_index, 0,
 153                sizeof(int) * NR_CPUS);
 154         memset(hv_context.event_dpc, 0,
 155                sizeof(void *) * NR_CPUS);
 156         memset(hv_context.clk_evt, 0,
 157                sizeof(void *) * NR_CPUS);
 158
 159         max_leaf = query_hypervisor_info();
 160
 161         /*
 162          * Write our OS ID.
 163          */
 164         hv_context.guestid = generate_guest_id(0, LINUX_VERSION_CODE, 0);
 165         wrmsrl(HV_X64_MSR_GUEST_OS_ID, hv_context.guestid);
 166
 167         /* See if the hypercall page is already set */
 168         rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 169
 170         virtaddr = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_EXEC);
 171
 172         if (!virtaddr)
 173                 goto cleanup;
 174
 175         hypercall_msr.enable = 1;
 176
 177         hypercall_msr.guest_physical_address = vmalloc_to_pfn(virtaddr);
 178         wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 179
 180         /* Confirm that hypercall page did get setup. */
 181         hypercall_msr.as_uint64 = 0;
 182         rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 183
 184         if (!hypercall_msr.enable)
 185                 goto cleanup;
 186
 187         hv_context.hypercall_page = virtaddr;
 188
 189         return 0;
 190
 191 cleanup:
 192         if (virtaddr) {
 193                 if (hypercall_msr.enable) {
 194                         hypercall_msr.as_uint64 = 0;
 195                         wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 196                 }
 197
 198                 vfree(virtaddr);
 199         }
 200
 201         return -ENOTSUPP;
 202 }
 203
 204 /*
 205  * hv_cleanup - Cleanup routine.
 206  *
 207  * This routine is called normally during driver unloading or exiting.
 208  */
 209 void hv_cleanup(void)
 210 {
 211         union hv_x64_msr_hypercall_contents hypercall_msr;
 212
 213         /* Reset our OS id */
 214         wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
 215
 216         if (hv_context.hypercall_page) {
 217                 hypercall_msr.as_uint64 = 0;
 218                 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 219                 vfree(hv_context.hypercall_page);
 220                 hv_context.hypercall_page = NULL;
 221         }
 222 }
 223
 224 /*
 225  * hv_post_message - Post a message using the hypervisor message IPC.
 226  *
 227  * This involves a hypercall.
 228  */
 229 int hv_post_message(union hv_connection_id connection_id,
 230                   enum hv_message_type message_type,
 231                   void *payload, size_t payload_size)
 232 {
 233
 234         struct hv_input_post_message *aligned_msg;
 235         u16 status;
 236
 237         if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT)
 238                 return -EMSGSIZE;
 239
 240         aligned_msg = (struct hv_input_post_message *)
 241                         hv_context.post_msg_page[get_cpu()];
 242
 243         aligned_msg->connectionid = connection_id;
 244         aligned_msg->reserved = 0;
 245         aligned_msg->message_type = message_type;
 246         aligned_msg->payload_size = payload_size;
 247         memcpy((void *)aligned_msg->payload, payload, payload_size);
 248
 249         status = do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL)
 250                 & 0xFFFF;
 251
 252         put_cpu();
 253         return status;
 254 }
 255
 256
 257 /*
 258  * hv_signal_event -
 259  * Signal an event on the specified connection using the hypervisor event IPC.
 260  *
 261  * This involves a hypercall.
 262  */
 263 u16 hv_signal_event(void *con_id)
 264 {
 265         u16 status;
 266
 267         status = (do_hypercall(HVCALL_SIGNAL_EVENT, con_id, NULL) & 0xFFFF);
 268
 269         return status;
 270 }
 271
 272 static int hv_ce_set_next_event(unsigned long delta,
 273                                 struct clock_event_device *evt)
 274 {
 275         cycle_t current_tick;
 276
 277         WARN_ON(!clockevent_state_oneshot(evt));
 278
 279         rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
 280         current_tick += delta;
 281         wrmsrl(HV_X64_MSR_STIMER0_COUNT, current_tick);
 282         return 0;
 283 }
 284
 285 static int hv_ce_shutdown(struct clock_event_device *evt)
 286 {
 287         wrmsrl(HV_X64_MSR_STIMER0_COUNT, 0);
 288         wrmsrl(HV_X64_MSR_STIMER0_CONFIG, 0);
 289
 290         return 0;
 291 }
 292
 293 static int hv_ce_set_oneshot(struct clock_event_device *evt)
 294 {
 295         union hv_timer_config timer_cfg;
 296
 297         timer_cfg.enable = 1;
 298         timer_cfg.auto_enable = 1;
 299         timer_cfg.sintx = VMBUS_MESSAGE_SINT;
 300         wrmsrl(HV_X64_MSR_STIMER0_CONFIG, timer_cfg.as_uint64);
 301
 302         return 0;
 303 }
 304
 305 static void hv_init_clockevent_device(struct clock_event_device *dev, int cpu)
 306 {
 307         dev->name = "Hyper-V clockevent";
 308         dev->features = CLOCK_EVT_FEAT_ONESHOT;
 309         dev->cpumask = cpumask_of(cpu);
 310         dev->rating = 1000;
 311         /*
 312          * Avoid settint dev->owner = THIS_MODULE deliberately as doing so will
 313          * result in clockevents_config_and_register() taking additional
 314          * references to the hv_vmbus module making it impossible to unload.
 315          */
 316
 317         dev->set_state_shutdown = hv_ce_shutdown;
 318         dev->set_state_oneshot = hv_ce_set_oneshot;
 319         dev->set_next_event = hv_ce_set_next_event;
 320 }
 321
 322
 323 int hv_synic_alloc(void)
 324 {
 325         size_t size = sizeof(struct tasklet_struct);
 326         size_t ced_size = sizeof(struct clock_event_device);
 327         int cpu;
 328
 329         hv_context.hv_numa_map = kzalloc(sizeof(struct cpumask) * nr_node_ids,
 330                                          GFP_ATOMIC);
 331         if (hv_context.hv_numa_map == NULL) {
 332                 pr_err("Unable to allocate NUMA map\n");
 333                 goto err;
 334         }
 335
 336         for_each_online_cpu(cpu) {
 337                 hv_context.event_dpc[cpu] = kmalloc(size, GFP_ATOMIC);
 338                 if (hv_context.event_dpc[cpu] == NULL) {
 339                         pr_err("Unable to allocate event dpc\n");
 340                         goto err;
 341                 }
 342                 tasklet_init(hv_context.event_dpc[cpu], vmbus_on_event, cpu);
 343
 344                 hv_context.clk_evt[cpu] = kzalloc(ced_size, GFP_ATOMIC);
 345                 if (hv_context.clk_evt[cpu] == NULL) {
 346                         pr_err("Unable to allocate clock event device\n");
 347                         goto err;
 348                 }
 349
 350                 hv_init_clockevent_device(hv_context.clk_evt[cpu], cpu);
 351
 352                 hv_context.synic_message_page[cpu] =
 353                         (void *)get_zeroed_page(GFP_ATOMIC);
 354
 355                 if (hv_context.synic_message_page[cpu] == NULL) {
 356                         pr_err("Unable to allocate SYNIC message page\n");
 357                         goto err;
 358                 }
 359
 360                 hv_context.synic_event_page[cpu] =
 361                         (void *)get_zeroed_page(GFP_ATOMIC);
 362
 363                 if (hv_context.synic_event_page[cpu] == NULL) {
 364                         pr_err("Unable to allocate SYNIC event page\n");
 365                         goto err;
 366                 }
 367
 368                 hv_context.post_msg_page[cpu] =
 369                         (void *)get_zeroed_page(GFP_ATOMIC);
 370
 371                 if (hv_context.post_msg_page[cpu] == NULL) {
 372                         pr_err("Unable to allocate post msg page\n");
 373                         goto err;
 374                 }
 375         }
 376
 377         return 0;
 378 err:
 379         return -ENOMEM;
 380 }
 381
 382 static void hv_synic_free_cpu(int cpu)
 383 {
 384         kfree(hv_context.event_dpc[cpu]);
 385         kfree(hv_context.clk_evt[cpu]);
 386         if (hv_context.synic_event_page[cpu])
 387                 free_page((unsigned long)hv_context.synic_event_page[cpu]);
 388         if (hv_context.synic_message_page[cpu])
 389                 free_page((unsigned long)hv_context.synic_message_page[cpu]);
 390         if (hv_context.post_msg_page[cpu])
 391                 free_page((unsigned long)hv_context.post_msg_page[cpu]);
 392 }
 393
 394 void hv_synic_free(void)
 395 {
 396         int cpu;
 397
 398         kfree(hv_context.hv_numa_map);
 399         for_each_online_cpu(cpu)
 400                 hv_synic_free_cpu(cpu);
 401 }
 402
 403 /*
 404  * hv_synic_init - Initialize the Synthethic Interrupt Controller.
 405  *
 406  * If it is already initialized by another entity (ie x2v shim), we need to
 407  * retrieve the initialized message and event pages.  Otherwise, we create and
 408  * initialize the message and event pages.
 409  */
 410 void hv_synic_init(void *arg)
 411 {
 412         u64 version;
 413         union hv_synic_simp simp;
 414         union hv_synic_siefp siefp;
 415         union hv_synic_sint shared_sint;
 416         union hv_synic_scontrol sctrl;
 417         u64 vp_index;
 418
 419         int cpu = smp_processor_id();
 420
 421         if (!hv_context.hypercall_page)
 422                 return;
 423
 424         /* Check the version */
 425         rdmsrl(HV_X64_MSR_SVERSION, version);
 426
 427         /* Setup the Synic's message page */
 428         rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
 429         simp.simp_enabled = 1;
 430         simp.base_simp_gpa = virt_to_phys(hv_context.synic_message_page[cpu])
 431                 >> PAGE_SHIFT;
 432
 433         wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
 434
 435         /* Setup the Synic's event page */
 436         rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
 437         siefp.siefp_enabled = 1;
 438         siefp.base_siefp_gpa = virt_to_phys(hv_context.synic_event_page[cpu])
 439                 >> PAGE_SHIFT;
 440
 441         wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
 442
 443         /* Setup the shared SINT. */
 444         rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
 445
 446         shared_sint.as_uint64 = 0;
 447         shared_sint.vector = HYPERVISOR_CALLBACK_VECTOR;
 448         shared_sint.masked = false;
 449         shared_sint.auto_eoi = true;
 450
 451         wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
 452
 453         /* Enable the global synic bit */
 454         rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
 455         sctrl.enable = 1;
 456
 457         wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
 458
 459         hv_context.synic_initialized = true;
 460
 461         /*
 462          * Setup the mapping between Hyper-V's notion
 463          * of cpuid and Linux' notion of cpuid.
 464          * This array will be indexed using Linux cpuid.
 465          */
 466         rdmsrl(HV_X64_MSR_VP_INDEX, vp_index);
 467         hv_context.vp_index[cpu] = (u32)vp_index;
 468
 469         INIT_LIST_HEAD(&hv_context.percpu_list[cpu]);
 470
 471         /*
 472          * Register the per-cpu clockevent source.
 473          */
 474         if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE)
 475                 clockevents_config_and_register(hv_context.clk_evt[cpu],
 476                                                 HV_TIMER_FREQUENCY,
 477                                                 HV_MIN_DELTA_TICKS,
 478                                                 HV_MAX_MAX_DELTA_TICKS);
 479         return;
 480 }
 481
 482 /*
 483  * hv_synic_clockevents_cleanup - Cleanup clockevent devices
 484  */
 485 void hv_synic_clockevents_cleanup(void)
 486 {
 487         int cpu;
 488
 489         if (!(ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE))
 490                 return;
 491
 492         for_each_online_cpu(cpu)
 493                 clockevents_unbind_device(hv_context.clk_evt[cpu], cpu);
 494 }
 495
 496 /*
 497  * hv_synic_cleanup - Cleanup routine for hv_synic_init().
 498  */
 499 void hv_synic_cleanup(void *arg)
 500 {
 501         union hv_synic_sint shared_sint;
 502         union hv_synic_simp simp;
 503         union hv_synic_siefp siefp;
 504         union hv_synic_scontrol sctrl;
 505         int cpu = smp_processor_id();
 506
 507         if (!hv_context.synic_initialized)
 508                 return;
 509
 510         /* Turn off clockevent device */
 511         if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE)
 512                 hv_ce_shutdown(hv_context.clk_evt[cpu]);
 513
 514         rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
 515
 516         shared_sint.masked = 1;
 517
 518         /* Need to correctly cleanup in the case of SMP!!! */
 519         /* Disable the interrupt */
 520         wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
 521
 522         rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
 523         simp.simp_enabled = 0;
 524         simp.base_simp_gpa = 0;
 525
 526         wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
 527
 528         rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
 529         siefp.siefp_enabled = 0;
 530         siefp.base_siefp_gpa = 0;
 531
 532         wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
 533
 534         /* Disable the global synic bit */
 535         rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
 536         sctrl.enable = 0;
 537         wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
 538 }