virt/kvm/ioapic.c

   1 /*
   2  *  Copyright (C) 2001  MandrakeSoft S.A.
   3  *  Copyright 2010 Red Hat, Inc. and/or its affiliates.
   4  *
   5  *    MandrakeSoft S.A.
   6  *    43, rue d'Aboukir
   7  *    75002 Paris - France
   8  *    http://www.linux-mandrake.com/
   9  *    http://www.mandrakesoft.com/
  10  *
  11  *  This library is free software; you can redistribute it and/or
  12  *  modify it under the terms of the GNU Lesser General Public
  13  *  License as published by the Free Software Foundation; either
  14  *  version 2 of the License, or (at your option) any later version.
  15  *
  16  *  This library is distributed in the hope that it will be useful,
  17  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  *  Lesser General Public License for more details.
  20  *
  21  *  You should have received a copy of the GNU Lesser General Public
  22  *  License along with this library; if not, write to the Free Software
  23  *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
  24  *
  25  *  Yunhong Jiang <yunhong.jiang@intel.com>
  26  *  Yaozu (Eddie) Dong <eddie.dong@intel.com>
  27  *  Based on Xen 3.1 code.
  28  */
  29
  30 #include <linux/kvm_host.h>
  31 #include <linux/kvm.h>
  32 #include <linux/mm.h>
  33 #include <linux/highmem.h>
  34 #include <linux/smp.h>
  35 #include <linux/hrtimer.h>
  36 #include <linux/io.h>
  37 #include <linux/slab.h>
  38 #include <linux/export.h>
  39 #include <asm/processor.h>
  40 #include <asm/page.h>
  41 #include <asm/current.h>
  42 #include <trace/events/kvm.h>
  43
  44 #include "ioapic.h"
  45 #include "lapic.h"
  46 #include "irq.h"
  47
  48 #if 0
  49 #define ioapic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg)
  50 #else
  51 #define ioapic_debug(fmt, arg...)
  52 #endif
  53 static int ioapic_deliver(struct kvm_ioapic *vioapic, int irq,
  54                 bool line_status);
  55
  56 static unsigned long ioapic_read_indirect(struct kvm_ioapic *ioapic,
  57                                           unsigned long addr,
  58                                           unsigned long length)
  59 {
  60         unsigned long result = 0;
  61
  62         switch (ioapic->ioregsel) {
  63         case IOAPIC_REG_VERSION:
  64                 result = ((((IOAPIC_NUM_PINS - 1) & 0xff) << 16)
  65                           | (IOAPIC_VERSION_ID & 0xff));
  66                 break;
  67
  68         case IOAPIC_REG_APIC_ID:
  69         case IOAPIC_REG_ARB_ID:
  70                 result = ((ioapic->id & 0xf) << 24);
  71                 break;
  72
  73         default:
  74                 {
  75                         u32 redir_index = (ioapic->ioregsel - 0x10) >> 1;
  76                         u64 redir_content;
  77
  78                         if (redir_index < IOAPIC_NUM_PINS)
  79                                 redir_content =
  80                                         ioapic->redirtbl[redir_index].bits;
  81                         else
  82                                 redir_content = ~0ULL;
  83
  84                         result = (ioapic->ioregsel & 0x1) ?
  85                             (redir_content >> 32) & 0xffffffff :
  86                             redir_content & 0xffffffff;
  87                         break;
  88                 }
  89         }
  90
  91         return result;
  92 }
  93
  94 static void rtc_irq_eoi_tracking_reset(struct kvm_ioapic *ioapic)
  95 {
  96         ioapic->rtc_status.pending_eoi = 0;
  97         bitmap_zero(ioapic->rtc_status.dest_map, KVM_MAX_VCPUS);
  98 }
  99
 100 static void __rtc_irq_eoi_tracking_restore_one(struct kvm_vcpu *vcpu)
 101 {
 102         bool new_val, old_val;
 103         struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
 104         union kvm_ioapic_redirect_entry *e;
 105
 106         e = &ioapic->redirtbl[RTC_GSI];
 107         if (!kvm_apic_match_dest(vcpu, NULL, 0, e->fields.dest_id,
 108                                 e->fields.dest_mode))
 109                 return;
 110
 111         new_val = kvm_apic_pending_eoi(vcpu, e->fields.vector);
 112         old_val = test_bit(vcpu->vcpu_id, ioapic->rtc_status.dest_map);
 113
 114         if (new_val == old_val)
 115                 return;
 116
 117         if (new_val) {
 118                 __set_bit(vcpu->vcpu_id, ioapic->rtc_status.dest_map);
 119                 ioapic->rtc_status.pending_eoi++;
 120         } else {
 121                 __clear_bit(vcpu->vcpu_id, ioapic->rtc_status.dest_map);
 122                 ioapic->rtc_status.pending_eoi--;
 123         }
 124
 125         WARN_ON(ioapic->rtc_status.pending_eoi < 0);
 126 }
 127
 128 void kvm_rtc_eoi_tracking_restore_one(struct kvm_vcpu *vcpu)
 129 {
 130         struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
 131
 132         spin_lock(&ioapic->lock);
 133         __rtc_irq_eoi_tracking_restore_one(vcpu);
 134         spin_unlock(&ioapic->lock);
 135 }
 136
 137 static void kvm_rtc_eoi_tracking_restore_all(struct kvm_ioapic *ioapic)
 138 {
 139         struct kvm_vcpu *vcpu;
 140         int i;
 141
 142         if (RTC_GSI >= IOAPIC_NUM_PINS)
 143                 return;
 144
 145         rtc_irq_eoi_tracking_reset(ioapic);
 146         kvm_for_each_vcpu(i, vcpu, ioapic->kvm)
 147             __rtc_irq_eoi_tracking_restore_one(vcpu);
 148 }
 149
 150 static void rtc_irq_eoi(struct kvm_ioapic *ioapic, struct kvm_vcpu *vcpu)
 151 {
 152         if (test_and_clear_bit(vcpu->vcpu_id, ioapic->rtc_status.dest_map))
 153                 --ioapic->rtc_status.pending_eoi;
 154
 155         WARN_ON(ioapic->rtc_status.pending_eoi < 0);
 156 }
 157
 158 static bool rtc_irq_check_coalesced(struct kvm_ioapic *ioapic)
 159 {
 160         if (ioapic->rtc_status.pending_eoi > 0)
 161                 return true; /* coalesced */
 162
 163         return false;
 164 }
 165
 166 static int ioapic_service(struct kvm_ioapic *ioapic, unsigned int idx,
 167                 bool line_status)
 168 {
 169         union kvm_ioapic_redirect_entry *pent;
 170         int injected = -1;
 171
 172         pent = &ioapic->redirtbl[idx];
 173
 174         if (!pent->fields.mask) {
 175                 injected = ioapic_deliver(ioapic, idx, line_status);
 176                 if (injected && pent->fields.trig_mode == IOAPIC_LEVEL_TRIG)
 177                         pent->fields.remote_irr = 1;
 178         }
 179
 180         return injected;
 181 }
 182
 183 static void update_handled_vectors(struct kvm_ioapic *ioapic)
 184 {
 185         DECLARE_BITMAP(handled_vectors, 256);
 186         int i;
 187
 188         memset(handled_vectors, 0, sizeof(handled_vectors));
 189         for (i = 0; i < IOAPIC_NUM_PINS; ++i)
 190                 __set_bit(ioapic->redirtbl[i].fields.vector, handled_vectors);
 191         memcpy(ioapic->handled_vectors, handled_vectors,
 192                sizeof(handled_vectors));
 193         smp_wmb();
 194 }
 195
 196 void kvm_ioapic_scan_entry(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap,
 197                         u32 *tmr)
 198 {
 199         struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
 200         union kvm_ioapic_redirect_entry *e;
 201         int index;
 202
 203         spin_lock(&ioapic->lock);
 204         for (index = 0; index < IOAPIC_NUM_PINS; index++) {
 205                 e = &ioapic->redirtbl[index];
 206                 if (!e->fields.mask &&
 207                         (e->fields.trig_mode == IOAPIC_LEVEL_TRIG ||
 208                          kvm_irq_has_notifier(ioapic->kvm, KVM_IRQCHIP_IOAPIC,
 209                                  index) || index == RTC_GSI)) {
 210                         if (kvm_apic_match_dest(vcpu, NULL, 0,
 211                                 e->fields.dest_id, e->fields.dest_mode)) {
 212                                 __set_bit(e->fields.vector,
 213                                         (unsigned long *)eoi_exit_bitmap);
 214                                 if (e->fields.trig_mode == IOAPIC_LEVEL_TRIG)
 215                                         __set_bit(e->fields.vector,
 216                                                 (unsigned long *)tmr);
 217                         }
 218                 }
 219         }
 220         spin_unlock(&ioapic->lock);
 221 }
 222
 223 #ifdef CONFIG_X86
 224 void kvm_vcpu_request_scan_ioapic(struct kvm *kvm)
 225 {
 226         struct kvm_ioapic *ioapic = kvm->arch.vioapic;
 227
 228         if (!ioapic)
 229                 return;
 230         kvm_make_scan_ioapic_request(kvm);
 231 }
 232 #else
 233 void kvm_vcpu_request_scan_ioapic(struct kvm *kvm)
 234 {
 235         return;
 236 }
 237 #endif
 238
 239 static void ioapic_write_indirect(struct kvm_ioapic *ioapic, u32 val)
 240 {
 241         unsigned index;
 242         bool mask_before, mask_after;
 243         union kvm_ioapic_redirect_entry *e;
 244
 245         switch (ioapic->ioregsel) {
 246         case IOAPIC_REG_VERSION:
 247                 /* Writes are ignored. */
 248                 break;
 249
 250         case IOAPIC_REG_APIC_ID:
 251                 ioapic->id = (val >> 24) & 0xf;
 252                 break;
 253
 254         case IOAPIC_REG_ARB_ID:
 255                 break;
 256
 257         default:
 258                 index = (ioapic->ioregsel - 0x10) >> 1;
 259
 260                 ioapic_debug("change redir index %x val %x\n", index, val);
 261                 if (index >= IOAPIC_NUM_PINS)
 262                         return;
 263                 e = &ioapic->redirtbl[index];
 264                 mask_before = e->fields.mask;
 265                 if (ioapic->ioregsel & 1) {
 266                         e->bits &= 0xffffffff;
 267                         e->bits |= (u64) val << 32;
 268                 } else {
 269                         e->bits &= ~0xffffffffULL;
 270                         e->bits |= (u32) val;
 271                         e->fields.remote_irr = 0;
 272                 }
 273                 update_handled_vectors(ioapic);
 274                 mask_after = e->fields.mask;
 275                 if (mask_before != mask_after)
 276                         kvm_fire_mask_notifiers(ioapic->kvm, KVM_IRQCHIP_IOAPIC, index, mask_after);
 277                 if (e->fields.trig_mode == IOAPIC_LEVEL_TRIG
 278                     && ioapic->irr & (1 << index))
 279                         ioapic_service(ioapic, index, false);
 280                 kvm_vcpu_request_scan_ioapic(ioapic->kvm);
 281                 break;
 282         }
 283 }
 284
 285 static int ioapic_deliver(struct kvm_ioapic *ioapic, int irq, bool line_status)
 286 {
 287         union kvm_ioapic_redirect_entry *entry = &ioapic->redirtbl[irq];
 288         struct kvm_lapic_irq irqe;
 289         int ret;
 290
 291         ioapic_debug("dest=%x dest_mode=%x delivery_mode=%x "
 292                      "vector=%x trig_mode=%x\n",
 293                      entry->fields.dest_id, entry->fields.dest_mode,
 294                      entry->fields.delivery_mode, entry->fields.vector,
 295                      entry->fields.trig_mode);
 296
 297         irqe.dest_id = entry->fields.dest_id;
 298         irqe.vector = entry->fields.vector;
 299         irqe.dest_mode = entry->fields.dest_mode;
 300         irqe.trig_mode = entry->fields.trig_mode;
 301         irqe.delivery_mode = entry->fields.delivery_mode << 8;
 302         irqe.level = 1;
 303         irqe.shorthand = 0;
 304
 305         if (irq == RTC_GSI && line_status) {
 306                 BUG_ON(ioapic->rtc_status.pending_eoi != 0);
 307                 ret = kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe,
 308                                 ioapic->rtc_status.dest_map);
 309                 ioapic->rtc_status.pending_eoi = ret;
 310         } else
 311                 ret = kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe, NULL);
 312
 313         return ret;
 314 }
 315
 316 int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int irq_source_id,
 317                        int level, bool line_status)
 318 {
 319         u32 old_irr;
 320         u32 mask = 1 << irq;
 321         union kvm_ioapic_redirect_entry entry;
 322         int ret, irq_level;
 323
 324         BUG_ON(irq < 0 || irq >= IOAPIC_NUM_PINS);
 325
 326         spin_lock(&ioapic->lock);
 327         old_irr = ioapic->irr;
 328         irq_level = __kvm_irq_line_state(&ioapic->irq_states[irq],
 329                                          irq_source_id, level);
 330         entry = ioapic->redirtbl[irq];
 331         irq_level ^= entry.fields.polarity;
 332         if (!irq_level) {
 333                 ioapic->irr &= ~mask;
 334                 ret = 1;
 335         } else {
 336                 int edge = (entry.fields.trig_mode == IOAPIC_EDGE_TRIG);
 337
 338                 if (irq == RTC_GSI && line_status &&
 339                         rtc_irq_check_coalesced(ioapic)) {
 340                         ret = 0; /* coalesced */
 341                         goto out;
 342                 }
 343                 ioapic->irr |= mask;
 344                 if ((edge && old_irr != ioapic->irr) ||
 345                     (!edge && !entry.fields.remote_irr))
 346                         ret = ioapic_service(ioapic, irq, line_status);
 347                 else
 348                         ret = 0; /* report coalesced interrupt */
 349         }
 350 out:
 351         trace_kvm_ioapic_set_irq(entry.bits, irq, ret == 0);
 352         spin_unlock(&ioapic->lock);
 353
 354         return ret;
 355 }
 356
 357 void kvm_ioapic_clear_all(struct kvm_ioapic *ioapic, int irq_source_id)
 358 {
 359         int i;
 360
 361         spin_lock(&ioapic->lock);
 362         for (i = 0; i < KVM_IOAPIC_NUM_PINS; i++)
 363                 __clear_bit(irq_source_id, &ioapic->irq_states[i]);
 364         spin_unlock(&ioapic->lock);
 365 }
 366
 367 static void __kvm_ioapic_update_eoi(struct kvm_vcpu *vcpu,
 368                         struct kvm_ioapic *ioapic, int vector, int trigger_mode)
 369 {
 370         int i;
 371
 372         for (i = 0; i < IOAPIC_NUM_PINS; i++) {
 373                 union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
 374
 375                 if (ent->fields.vector != vector)
 376                         continue;
 377
 378                 if (i == RTC_GSI)
 379                         rtc_irq_eoi(ioapic, vcpu);
 380                 /*
 381                  * We are dropping lock while calling ack notifiers because ack
 382                  * notifier callbacks for assigned devices call into IOAPIC
 383                  * recursively. Since remote_irr is cleared only after call
 384                  * to notifiers if the same vector will be delivered while lock
 385                  * is dropped it will be put into irr and will be delivered
 386                  * after ack notifier returns.
 387                  */
 388                 spin_unlock(&ioapic->lock);
 389                 kvm_notify_acked_irq(ioapic->kvm, KVM_IRQCHIP_IOAPIC, i);
 390                 spin_lock(&ioapic->lock);
 391
 392                 if (trigger_mode != IOAPIC_LEVEL_TRIG)
 393                         continue;
 394
 395                 ASSERT(ent->fields.trig_mode == IOAPIC_LEVEL_TRIG);
 396                 ent->fields.remote_irr = 0;
 397                 if (!ent->fields.mask && (ioapic->irr & (1 << i)))
 398                         ioapic_service(ioapic, i, false);
 399         }
 400 }
 401
 402 bool kvm_ioapic_handles_vector(struct kvm *kvm, int vector)
 403 {
 404         struct kvm_ioapic *ioapic = kvm->arch.vioapic;
 405         smp_rmb();
 406         return test_bit(vector, ioapic->handled_vectors);
 407 }
 408
 409 void kvm_ioapic_update_eoi(struct kvm_vcpu *vcpu, int vector, int trigger_mode)
 410 {
 411         struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
 412
 413         spin_lock(&ioapic->lock);
 414         __kvm_ioapic_update_eoi(vcpu, ioapic, vector, trigger_mode);
 415         spin_unlock(&ioapic->lock);
 416 }
 417
 418 static inline struct kvm_ioapic *to_ioapic(struct kvm_io_device *dev)
 419 {
 420         return container_of(dev, struct kvm_ioapic, dev);
 421 }
 422
 423 static inline int ioapic_in_range(struct kvm_ioapic *ioapic, gpa_t addr)
 424 {
 425         return ((addr >= ioapic->base_address &&
 426                  (addr < ioapic->base_address + IOAPIC_MEM_LENGTH)));
 427 }
 428
 429 static int ioapic_mmio_read(struct kvm_io_device *this, gpa_t addr, int len,
 430                             void *val)
 431 {
 432         struct kvm_ioapic *ioapic = to_ioapic(this);
 433         u32 result;
 434         if (!ioapic_in_range(ioapic, addr))
 435                 return -EOPNOTSUPP;
 436
 437         ioapic_debug("addr %lx\n", (unsigned long)addr);
 438         ASSERT(!(addr & 0xf));  /* check alignment */
 439
 440         addr &= 0xff;
 441         spin_lock(&ioapic->lock);
 442         switch (addr) {
 443         case IOAPIC_REG_SELECT:
 444                 result = ioapic->ioregsel;
 445                 break;
 446
 447         case IOAPIC_REG_WINDOW:
 448                 result = ioapic_read_indirect(ioapic, addr, len);
 449                 break;
 450
 451         default:
 452                 result = 0;
 453                 break;
 454         }
 455         spin_unlock(&ioapic->lock);
 456
 457         switch (len) {
 458         case 8:
 459                 *(u64 *) val = result;
 460                 break;
 461         case 1:
 462         case 2:
 463         case 4:
 464                 memcpy(val, (char *)&result, len);
 465                 break;
 466         default:
 467                 printk(KERN_WARNING "ioapic: wrong length %d\n", len);
 468         }
 469         return 0;
 470 }
 471
 472 static int ioapic_mmio_write(struct kvm_io_device *this, gpa_t addr, int len,
 473                              const void *val)
 474 {
 475         struct kvm_ioapic *ioapic = to_ioapic(this);
 476         u32 data;
 477         if (!ioapic_in_range(ioapic, addr))
 478                 return -EOPNOTSUPP;
 479
 480         ioapic_debug("ioapic_mmio_write addr=%p len=%d val=%p\n",
 481                      (void*)addr, len, val);
 482         ASSERT(!(addr & 0xf));  /* check alignment */
 483
 484         switch (len) {
 485         case 8:
 486         case 4:
 487                 data = *(u32 *) val;
 488                 break;
 489         case 2:
 490                 data = *(u16 *) val;
 491                 break;
 492         case 1:
 493                 data = *(u8  *) val;
 494                 break;
 495         default:
 496                 printk(KERN_WARNING "ioapic: Unsupported size %d\n", len);
 497                 return 0;
 498         }
 499
 500         addr &= 0xff;
 501         spin_lock(&ioapic->lock);
 502         switch (addr) {
 503         case IOAPIC_REG_SELECT:
 504                 ioapic->ioregsel = data & 0xFF; /* 8-bit register */
 505                 break;
 506
 507         case IOAPIC_REG_WINDOW:
 508                 ioapic_write_indirect(ioapic, data);
 509                 break;
 510 #ifdef  CONFIG_IA64
 511         case IOAPIC_REG_EOI:
 512                 __kvm_ioapic_update_eoi(NULL, ioapic, data, IOAPIC_LEVEL_TRIG);
 513                 break;
 514 #endif
 515
 516         default:
 517                 break;
 518         }
 519         spin_unlock(&ioapic->lock);
 520         return 0;
 521 }
 522
 523 static void kvm_ioapic_reset(struct kvm_ioapic *ioapic)
 524 {
 525         int i;
 526
 527         for (i = 0; i < IOAPIC_NUM_PINS; i++)
 528                 ioapic->redirtbl[i].fields.mask = 1;
 529         ioapic->base_address = IOAPIC_DEFAULT_BASE_ADDRESS;
 530         ioapic->ioregsel = 0;
 531         ioapic->irr = 0;
 532         ioapic->id = 0;
 533         rtc_irq_eoi_tracking_reset(ioapic);
 534         update_handled_vectors(ioapic);
 535 }
 536
 537 static const struct kvm_io_device_ops ioapic_mmio_ops = {
 538         .read     = ioapic_mmio_read,
 539         .write    = ioapic_mmio_write,
 540 };
 541
 542 int kvm_ioapic_init(struct kvm *kvm)
 543 {
 544         struct kvm_ioapic *ioapic;
 545         int ret;
 546
 547         ioapic = kzalloc(sizeof(struct kvm_ioapic), GFP_KERNEL);
 548         if (!ioapic)
 549                 return -ENOMEM;
 550         spin_lock_init(&ioapic->lock);
 551         kvm->arch.vioapic = ioapic;
 552         kvm_ioapic_reset(ioapic);
 553         kvm_iodevice_init(&ioapic->dev, &ioapic_mmio_ops);
 554         ioapic->kvm = kvm;
 555         mutex_lock(&kvm->slots_lock);
 556         ret = kvm_io_bus_register_dev(kvm, KVM_MMIO_BUS, ioapic->base_address,
 557                                       IOAPIC_MEM_LENGTH, &ioapic->dev);
 558         mutex_unlock(&kvm->slots_lock);
 559         if (ret < 0) {
 560                 kvm->arch.vioapic = NULL;
 561                 kfree(ioapic);
 562         }
 563
 564         return ret;
 565 }
 566
 567 void kvm_ioapic_destroy(struct kvm *kvm)
 568 {
 569         struct kvm_ioapic *ioapic = kvm->arch.vioapic;
 570
 571         if (ioapic) {
 572                 kvm_io_bus_unregister_dev(kvm, KVM_MMIO_BUS, &ioapic->dev);
 573                 kvm->arch.vioapic = NULL;
 574                 kfree(ioapic);
 575         }
 576 }
 577
 578 int kvm_get_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
 579 {
 580         struct kvm_ioapic *ioapic = ioapic_irqchip(kvm);
 581         if (!ioapic)
 582                 return -EINVAL;
 583
 584         spin_lock(&ioapic->lock);
 585         memcpy(state, ioapic, sizeof(struct kvm_ioapic_state));
 586         spin_unlock(&ioapic->lock);
 587         return 0;
 588 }
 589
 590 int kvm_set_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
 591 {
 592         struct kvm_ioapic *ioapic = ioapic_irqchip(kvm);
 593         if (!ioapic)
 594                 return -EINVAL;
 595
 596         spin_lock(&ioapic->lock);
 597         memcpy(ioapic, state, sizeof(struct kvm_ioapic_state));
 598         update_handled_vectors(ioapic);
 599         kvm_vcpu_request_scan_ioapic(kvm);
 600         kvm_rtc_eoi_tracking_restore_all(ioapic);
 601         spin_unlock(&ioapic->lock);
 602         return 0;
 603 }