accel/hvf/hvf-accel-ops.c

   1 /*
   2  * Copyright 2008 IBM Corporation
   3  *           2008 Red Hat, Inc.
   4  * Copyright 2011 Intel Corporation
   5  * Copyright 2016 Veertu, Inc.
   6  * Copyright 2017 The Android Open Source Project
   7  *
   8  * QEMU Hypervisor.framework support
   9  *
  10  * This program is free software; you can redistribute it and/or
  11  * modify it under the terms of version 2 of the GNU General Public
  12  * License as published by the Free Software Foundation.
  13  *
  14  * This program is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU General Public License
  20  * along with this program; if not, see <http://www.gnu.org/licenses/>.
  21  *
  22  * This file contain code under public domain from the hvdos project:
  23  * https://github.com/mist64/hvdos
  24  *
  25  * Parts Copyright (c) 2011 NetApp, Inc.
  26  * All rights reserved.
  27  *
  28  * Redistribution and use in source and binary forms, with or without
  29  * modification, are permitted provided that the following conditions
  30  * are met:
  31  * 1. Redistributions of source code must retain the above copyright
  32  *    notice, this list of conditions and the following disclaimer.
  33  * 2. Redistributions in binary form must reproduce the above copyright
  34  *    notice, this list of conditions and the following disclaimer in the
  35  *    documentation and/or other materials provided with the distribution.
  36  *
  37  * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
  38  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  39  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  40  * ARE DISCLAIMED.  IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
  41  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  42  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  43  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  44  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  45  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  46  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  47  * SUCH DAMAGE.
  48  */
  49
  50 #include "qemu/osdep.h"
  51 #include "qemu/error-report.h"
  52 #include "qemu/main-loop.h"
  53 #include "exec/address-spaces.h"
  54 #include "exec/exec-all.h"
  55 #include "sysemu/cpus.h"
  56 #include "sysemu/hvf.h"
  57 #include "sysemu/hvf_int.h"
  58 #include "sysemu/runstate.h"
  59 #include "qemu/guest-random.h"
  60
  61 HVFState *hvf_state;
  62
  63 #ifdef __aarch64__
  64 #define HV_VM_DEFAULT NULL
  65 #endif
  66
  67 /* Memory slots */
  68
  69 hvf_slot *hvf_find_overlap_slot(uint64_t start, uint64_t size)
  70 {
  71     hvf_slot *slot;
  72     int x;
  73     for (x = 0; x < hvf_state->num_slots; ++x) {
  74         slot = &hvf_state->slots[x];
  75         if (slot->size && start < (slot->start + slot->size) &&
  76             (start + size) > slot->start) {
  77             return slot;
  78         }
  79     }
  80     return NULL;
  81 }
  82
  83 struct mac_slot {
  84     int present;
  85     uint64_t size;
  86     uint64_t gpa_start;
  87     uint64_t gva;
  88 };
  89
  90 struct mac_slot mac_slots[32];
  91
  92 static int do_hvf_set_memory(hvf_slot *slot, hv_memory_flags_t flags)
  93 {
  94     struct mac_slot *macslot;
  95     hv_return_t ret;
  96
  97     macslot = &mac_slots[slot->slot_id];
  98
  99     if (macslot->present) {
 100         if (macslot->size != slot->size) {
 101             macslot->present = 0;
 102             ret = hv_vm_unmap(macslot->gpa_start, macslot->size);
 103             assert_hvf_ok(ret);
 104         }
 105     }
 106
 107     if (!slot->size) {
 108         return 0;
 109     }
 110
 111     macslot->present = 1;
 112     macslot->gpa_start = slot->start;
 113     macslot->size = slot->size;
 114     ret = hv_vm_map(slot->mem, slot->start, slot->size, flags);
 115     assert_hvf_ok(ret);
 116     return 0;
 117 }
 118
 119 static void hvf_set_phys_mem(MemoryRegionSection *section, bool add)
 120 {
 121     hvf_slot *mem;
 122     MemoryRegion *area = section->mr;
 123     bool writeable = !area->readonly && !area->rom_device;
 124     hv_memory_flags_t flags;
 125
 126     if (!memory_region_is_ram(area)) {
 127         if (writeable) {
 128             return;
 129         } else if (!memory_region_is_romd(area)) {
 130             /*
 131              * If the memory device is not in romd_mode, then we actually want
 132              * to remove the hvf memory slot so all accesses will trap.
 133              */
 134              add = false;
 135         }
 136     }
 137
 138     mem = hvf_find_overlap_slot(
 139             section->offset_within_address_space,
 140             int128_get64(section->size));
 141
 142     if (mem && add) {
 143         if (mem->size == int128_get64(section->size) &&
 144             mem->start == section->offset_within_address_space &&
 145             mem->mem == (memory_region_get_ram_ptr(area) +
 146             section->offset_within_region)) {
 147             return; /* Same region was attempted to register, go away. */
 148         }
 149     }
 150
 151     /* Region needs to be reset. set the size to 0 and remap it. */
 152     if (mem) {
 153         mem->size = 0;
 154         if (do_hvf_set_memory(mem, 0)) {
 155             error_report("Failed to reset overlapping slot");
 156             abort();
 157         }
 158     }
 159
 160     if (!add) {
 161         return;
 162     }
 163
 164     if (area->readonly ||
 165         (!memory_region_is_ram(area) && memory_region_is_romd(area))) {
 166         flags = HV_MEMORY_READ | HV_MEMORY_EXEC;
 167     } else {
 168         flags = HV_MEMORY_READ | HV_MEMORY_WRITE | HV_MEMORY_EXEC;
 169     }
 170
 171     /* Now make a new slot. */
 172     int x;
 173
 174     for (x = 0; x < hvf_state->num_slots; ++x) {
 175         mem = &hvf_state->slots[x];
 176         if (!mem->size) {
 177             break;
 178         }
 179     }
 180
 181     if (x == hvf_state->num_slots) {
 182         error_report("No free slots");
 183         abort();
 184     }
 185
 186     mem->size = int128_get64(section->size);
 187     mem->mem = memory_region_get_ram_ptr(area) + section->offset_within_region;
 188     mem->start = section->offset_within_address_space;
 189     mem->region = area;
 190
 191     if (do_hvf_set_memory(mem, flags)) {
 192         error_report("Error registering new memory slot");
 193         abort();
 194     }
 195 }
 196
 197 static void do_hvf_cpu_synchronize_state(CPUState *cpu, run_on_cpu_data arg)
 198 {
 199     if (!cpu->vcpu_dirty) {
 200         hvf_get_registers(cpu);
 201         cpu->vcpu_dirty = true;
 202     }
 203 }
 204
 205 static void hvf_cpu_synchronize_state(CPUState *cpu)
 206 {
 207     if (!cpu->vcpu_dirty) {
 208         run_on_cpu(cpu, do_hvf_cpu_synchronize_state, RUN_ON_CPU_NULL);
 209     }
 210 }
 211
 212 static void do_hvf_cpu_synchronize_set_dirty(CPUState *cpu,
 213                                              run_on_cpu_data arg)
 214 {
 215     /* QEMU state is the reference, push it to HVF now and on next entry */
 216     cpu->vcpu_dirty = true;
 217 }
 218
 219 static void hvf_cpu_synchronize_post_reset(CPUState *cpu)
 220 {
 221     run_on_cpu(cpu, do_hvf_cpu_synchronize_set_dirty, RUN_ON_CPU_NULL);
 222 }
 223
 224 static void hvf_cpu_synchronize_post_init(CPUState *cpu)
 225 {
 226     run_on_cpu(cpu, do_hvf_cpu_synchronize_set_dirty, RUN_ON_CPU_NULL);
 227 }
 228
 229 static void hvf_cpu_synchronize_pre_loadvm(CPUState *cpu)
 230 {
 231     run_on_cpu(cpu, do_hvf_cpu_synchronize_set_dirty, RUN_ON_CPU_NULL);
 232 }
 233
 234 static void hvf_set_dirty_tracking(MemoryRegionSection *section, bool on)
 235 {
 236     hvf_slot *slot;
 237
 238     slot = hvf_find_overlap_slot(
 239             section->offset_within_address_space,
 240             int128_get64(section->size));
 241
 242     /* protect region against writes; begin tracking it */
 243     if (on) {
 244         slot->flags |= HVF_SLOT_LOG;
 245         hv_vm_protect((uintptr_t)slot->start, (size_t)slot->size,
 246                       HV_MEMORY_READ | HV_MEMORY_EXEC);
 247     /* stop tracking region*/
 248     } else {
 249         slot->flags &= ~HVF_SLOT_LOG;
 250         hv_vm_protect((uintptr_t)slot->start, (size_t)slot->size,
 251                       HV_MEMORY_READ | HV_MEMORY_WRITE | HV_MEMORY_EXEC);
 252     }
 253 }
 254
 255 static void hvf_log_start(MemoryListener *listener,
 256                           MemoryRegionSection *section, int old, int new)
 257 {
 258     if (old != 0) {
 259         return;
 260     }
 261
 262     hvf_set_dirty_tracking(section, 1);
 263 }
 264
 265 static void hvf_log_stop(MemoryListener *listener,
 266                          MemoryRegionSection *section, int old, int new)
 267 {
 268     if (new != 0) {
 269         return;
 270     }
 271
 272     hvf_set_dirty_tracking(section, 0);
 273 }
 274
 275 static void hvf_log_sync(MemoryListener *listener,
 276                          MemoryRegionSection *section)
 277 {
 278     /*
 279      * sync of dirty pages is handled elsewhere; just make sure we keep
 280      * tracking the region.
 281      */
 282     hvf_set_dirty_tracking(section, 1);
 283 }
 284
 285 static void hvf_region_add(MemoryListener *listener,
 286                            MemoryRegionSection *section)
 287 {
 288     hvf_set_phys_mem(section, true);
 289 }
 290
 291 static void hvf_region_del(MemoryListener *listener,
 292                            MemoryRegionSection *section)
 293 {
 294     hvf_set_phys_mem(section, false);
 295 }
 296
 297 static MemoryListener hvf_memory_listener = {
 298     .name = "hvf",
 299     .priority = 10,
 300     .region_add = hvf_region_add,
 301     .region_del = hvf_region_del,
 302     .log_start = hvf_log_start,
 303     .log_stop = hvf_log_stop,
 304     .log_sync = hvf_log_sync,
 305 };
 306
 307 static void dummy_signal(int sig)
 308 {
 309 }
 310
 311 bool hvf_allowed;
 312
 313 static int hvf_accel_init(MachineState *ms)
 314 {
 315     int x;
 316     hv_return_t ret;
 317     HVFState *s;
 318
 319     ret = hv_vm_create(HV_VM_DEFAULT);
 320     assert_hvf_ok(ret);
 321
 322     s = g_new0(HVFState, 1);
 323
 324     s->num_slots = ARRAY_SIZE(s->slots);
 325     for (x = 0; x < s->num_slots; ++x) {
 326         s->slots[x].size = 0;
 327         s->slots[x].slot_id = x;
 328     }
 329
 330     hvf_state = s;
 331     memory_listener_register(&hvf_memory_listener, &address_space_memory);
 332
 333     return hvf_arch_init();
 334 }
 335
 336 static void hvf_accel_class_init(ObjectClass *oc, void *data)
 337 {
 338     AccelClass *ac = ACCEL_CLASS(oc);
 339     ac->name = "HVF";
 340     ac->init_machine = hvf_accel_init;
 341     ac->allowed = &hvf_allowed;
 342 }
 343
 344 static const TypeInfo hvf_accel_type = {
 345     .name = TYPE_HVF_ACCEL,
 346     .parent = TYPE_ACCEL,
 347     .class_init = hvf_accel_class_init,
 348 };
 349
 350 static void hvf_type_init(void)
 351 {
 352     type_register_static(&hvf_accel_type);
 353 }
 354
 355 type_init(hvf_type_init);
 356
 357 static void hvf_vcpu_destroy(CPUState *cpu)
 358 {
 359     hv_return_t ret = hv_vcpu_destroy(cpu->hvf->fd);
 360     assert_hvf_ok(ret);
 361
 362     hvf_arch_vcpu_destroy(cpu);
 363     g_free(cpu->hvf);
 364     cpu->hvf = NULL;
 365 }
 366
 367 static int hvf_init_vcpu(CPUState *cpu)
 368 {
 369     int r;
 370
 371     cpu->hvf = g_malloc0(sizeof(*cpu->hvf));
 372
 373     /* init cpu signals */
 374     struct sigaction sigact;
 375
 376     memset(&sigact, 0, sizeof(sigact));
 377     sigact.sa_handler = dummy_signal;
 378     sigaction(SIG_IPI, &sigact, NULL);
 379
 380     pthread_sigmask(SIG_BLOCK, NULL, &cpu->hvf->unblock_ipi_mask);
 381     sigdelset(&cpu->hvf->unblock_ipi_mask, SIG_IPI);
 382
 383 #ifdef __aarch64__
 384     r = hv_vcpu_create(&cpu->hvf->fd, (hv_vcpu_exit_t **)&cpu->hvf->exit, NULL);
 385 #else
 386     r = hv_vcpu_create((hv_vcpuid_t *)&cpu->hvf->fd, HV_VCPU_DEFAULT);
 387 #endif
 388     cpu->vcpu_dirty = 1;
 389     assert_hvf_ok(r);
 390
 391     return hvf_arch_init_vcpu(cpu);
 392 }
 393
 394 /*
 395  * The HVF-specific vCPU thread function. This one should only run when the host
 396  * CPU supports the VMX "unrestricted guest" feature.
 397  */
 398 static void *hvf_cpu_thread_fn(void *arg)
 399 {
 400     CPUState *cpu = arg;
 401
 402     int r;
 403
 404     assert(hvf_enabled());
 405
 406     rcu_register_thread();
 407
 408     qemu_mutex_lock_iothread();
 409     qemu_thread_get_self(cpu->thread);
 410
 411     cpu->thread_id = qemu_get_thread_id();
 412     cpu->can_do_io = 1;
 413     current_cpu = cpu;
 414
 415     hvf_init_vcpu(cpu);
 416
 417     /* signal CPU creation */
 418     cpu_thread_signal_created(cpu);
 419     qemu_guest_random_seed_thread_part2(cpu->random_seed);
 420
 421     do {
 422         if (cpu_can_run(cpu)) {
 423             r = hvf_vcpu_exec(cpu);
 424             if (r == EXCP_DEBUG) {
 425                 cpu_handle_guest_debug(cpu);
 426             }
 427         }
 428         qemu_wait_io_event(cpu);
 429     } while (!cpu->unplug || cpu_can_run(cpu));
 430
 431     hvf_vcpu_destroy(cpu);
 432     cpu_thread_signal_destroyed(cpu);
 433     qemu_mutex_unlock_iothread();
 434     rcu_unregister_thread();
 435     return NULL;
 436 }
 437
 438 static void hvf_start_vcpu_thread(CPUState *cpu)
 439 {
 440     char thread_name[VCPU_THREAD_NAME_SIZE];
 441
 442     /*
 443      * HVF currently does not support TCG, and only runs in
 444      * unrestricted-guest mode.
 445      */
 446     assert(hvf_enabled());
 447
 448     cpu->thread = g_malloc0(sizeof(QemuThread));
 449     cpu->halt_cond = g_malloc0(sizeof(QemuCond));
 450     qemu_cond_init(cpu->halt_cond);
 451
 452     snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/HVF",
 453              cpu->cpu_index);
 454     qemu_thread_create(cpu->thread, thread_name, hvf_cpu_thread_fn,
 455                        cpu, QEMU_THREAD_JOINABLE);
 456 }
 457
 458 static void hvf_accel_ops_class_init(ObjectClass *oc, void *data)
 459 {
 460     AccelOpsClass *ops = ACCEL_OPS_CLASS(oc);
 461
 462     ops->create_vcpu_thread = hvf_start_vcpu_thread;
 463     ops->kick_vcpu_thread = hvf_kick_vcpu_thread;
 464
 465     ops->synchronize_post_reset = hvf_cpu_synchronize_post_reset;
 466     ops->synchronize_post_init = hvf_cpu_synchronize_post_init;
 467     ops->synchronize_state = hvf_cpu_synchronize_state;
 468     ops->synchronize_pre_loadvm = hvf_cpu_synchronize_pre_loadvm;
 469 };
 470 static const TypeInfo hvf_accel_ops_type = {
 471     .name = ACCEL_OPS_NAME("hvf"),
 472
 473     .parent = TYPE_ACCEL_OPS,
 474     .class_init = hvf_accel_ops_class_init,
 475     .abstract = true,
 476 };
 477 static void hvf_accel_ops_register_types(void)
 478 {
 479     type_register_static(&hvf_accel_ops_type);
 480 }
 481 type_init(hvf_accel_ops_register_types);