arch/x86/kvm/i8259.c

   1 /*
   2  * 8259 interrupt controller emulation
   3  *
   4  * Copyright (c) 2003-2004 Fabrice Bellard
   5  * Copyright (c) 2007 Intel Corporation
   6  * Copyright 2009 Red Hat, Inc. and/or its affiliates.
   7  *
   8  * Permission is hereby granted, free of charge, to any person obtaining a copy
   9  * of this software and associated documentation files (the "Software"), to deal
  10  * in the Software without restriction, including without limitation the rights
  11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  12  * copies of the Software, and to permit persons to whom the Software is
  13  * furnished to do so, subject to the following conditions:
  14  *
  15  * The above copyright notice and this permission notice shall be included in
  16  * all copies or substantial portions of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  24  * THE SOFTWARE.
  25  * Authors:
  26  *   Yaozu (Eddie) Dong <Eddie.dong@intel.com>
  27  *   Port from Qemu.
  28  */
  29 #include <linux/mm.h>
  30 #include <linux/slab.h>
  31 #include <linux/bitops.h>
  32 #include "irq.h"
  33
  34 #include <linux/kvm_host.h>
  35 #include "trace.h"
  36
  37 #define pr_pic_unimpl(fmt, ...) \
  38         pr_err_ratelimited("kvm: pic: " fmt, ## __VA_ARGS__)
  39
  40 static void pic_irq_request(struct kvm *kvm, int level);
  41
  42 static void pic_lock(struct kvm_pic *s)
  43         __acquires(&s->lock)
  44 {
  45         spin_lock(&s->lock);
  46 }
  47
  48 static void pic_unlock(struct kvm_pic *s)
  49         __releases(&s->lock)
  50 {
  51         bool wakeup = s->wakeup_needed;
  52         struct kvm_vcpu *vcpu, *found = NULL;
  53         int i;
  54
  55         s->wakeup_needed = false;
  56
  57         spin_unlock(&s->lock);
  58
  59         if (wakeup) {
  60                 kvm_for_each_vcpu(i, vcpu, s->kvm) {
  61                         if (kvm_apic_accept_pic_intr(vcpu)) {
  62                                 found = vcpu;
  63                                 break;
  64                         }
  65                 }
  66
  67                 if (!found)
  68                         return;
  69
  70                 kvm_make_request(KVM_REQ_EVENT, found);
  71                 kvm_vcpu_kick(found);
  72         }
  73 }
  74
  75 static void pic_clear_isr(struct kvm_kpic_state *s, int irq)
  76 {
  77         s->isr &= ~(1 << irq);
  78         if (s != &s->pics_state->pics[0])
  79                 irq += 8;
  80         /*
  81          * We are dropping lock while calling ack notifiers since ack
  82          * notifier callbacks for assigned devices call into PIC recursively.
  83          * Other interrupt may be delivered to PIC while lock is dropped but
  84          * it should be safe since PIC state is already updated at this stage.
  85          */
  86         pic_unlock(s->pics_state);
  87         kvm_notify_acked_irq(s->pics_state->kvm, SELECT_PIC(irq), irq);
  88         pic_lock(s->pics_state);
  89 }
  90
  91 /*
  92  * set irq level. If an edge is detected, then the IRR is set to 1
  93  */
  94 static inline int pic_set_irq1(struct kvm_kpic_state *s, int irq, int level)
  95 {
  96         int mask, ret = 1;
  97         mask = 1 << irq;
  98         if (s->elcr & mask)     /* level triggered */
  99                 if (level) {
 100                         ret = !(s->irr & mask);
 101                         s->irr |= mask;
 102                         s->last_irr |= mask;
 103                 } else {
 104                         s->irr &= ~mask;
 105                         s->last_irr &= ~mask;
 106                 }
 107         else    /* edge triggered */
 108                 if (level) {
 109                         if ((s->last_irr & mask) == 0) {
 110                                 ret = !(s->irr & mask);
 111                                 s->irr |= mask;
 112                         }
 113                         s->last_irr |= mask;
 114                 } else
 115                         s->last_irr &= ~mask;
 116
 117         return (s->imr & mask) ? -1 : ret;
 118 }
 119
 120 /*
 121  * return the highest priority found in mask (highest = smallest
 122  * number). Return 8 if no irq
 123  */
 124 static inline int get_priority(struct kvm_kpic_state *s, int mask)
 125 {
 126         int priority;
 127         if (mask == 0)
 128                 return 8;
 129         priority = 0;
 130         while ((mask & (1 << ((priority + s->priority_add) & 7))) == 0)
 131                 priority++;
 132         return priority;
 133 }
 134
 135 /*
 136  * return the pic wanted interrupt. return -1 if none
 137  */
 138 static int pic_get_irq(struct kvm_kpic_state *s)
 139 {
 140         int mask, cur_priority, priority;
 141
 142         mask = s->irr & ~s->imr;
 143         priority = get_priority(s, mask);
 144         if (priority == 8)
 145                 return -1;
 146         /*
 147          * compute current priority. If special fully nested mode on the
 148          * master, the IRQ coming from the slave is not taken into account
 149          * for the priority computation.
 150          */
 151         mask = s->isr;
 152         if (s->special_fully_nested_mode && s == &s->pics_state->pics[0])
 153                 mask &= ~(1 << 2);
 154         cur_priority = get_priority(s, mask);
 155         if (priority < cur_priority)
 156                 /*
 157                  * higher priority found: an irq should be generated
 158                  */
 159                 return (priority + s->priority_add) & 7;
 160         else
 161                 return -1;
 162 }
 163
 164 /*
 165  * raise irq to CPU if necessary. must be called every time the active
 166  * irq may change
 167  */
 168 static void pic_update_irq(struct kvm_pic *s)
 169 {
 170         int irq2, irq;
 171
 172         irq2 = pic_get_irq(&s->pics[1]);
 173         if (irq2 >= 0) {
 174                 /*
 175                  * if irq request by slave pic, signal master PIC
 176                  */
 177                 pic_set_irq1(&s->pics[0], 2, 1);
 178                 pic_set_irq1(&s->pics[0], 2, 0);
 179         }
 180         irq = pic_get_irq(&s->pics[0]);
 181         pic_irq_request(s->kvm, irq >= 0);
 182 }
 183
 184 void kvm_pic_update_irq(struct kvm_pic *s)
 185 {
 186         pic_lock(s);
 187         pic_update_irq(s);
 188         pic_unlock(s);
 189 }
 190
 191 int kvm_pic_set_irq(void *opaque, int irq, int level)
 192 {
 193         struct kvm_pic *s = opaque;
 194         int ret = -1;
 195
 196         pic_lock(s);
 197         if (irq >= 0 && irq < PIC_NUM_PINS) {
 198                 ret = pic_set_irq1(&s->pics[irq >> 3], irq & 7, level);
 199                 pic_update_irq(s);
 200                 trace_kvm_pic_set_irq(irq >> 3, irq & 7, s->pics[irq >> 3].elcr,
 201                                       s->pics[irq >> 3].imr, ret == 0);
 202         }
 203         pic_unlock(s);
 204
 205         return ret;
 206 }
 207
 208 /*
 209  * acknowledge interrupt 'irq'
 210  */
 211 static inline void pic_intack(struct kvm_kpic_state *s, int irq)
 212 {
 213         s->isr |= 1 << irq;
 214         /*
 215          * We don't clear a level sensitive interrupt here
 216          */
 217         if (!(s->elcr & (1 << irq)))
 218                 s->irr &= ~(1 << irq);
 219
 220         if (s->auto_eoi) {
 221                 if (s->rotate_on_auto_eoi)
 222                         s->priority_add = (irq + 1) & 7;
 223                 pic_clear_isr(s, irq);
 224         }
 225
 226 }
 227
 228 int kvm_pic_read_irq(struct kvm *kvm)
 229 {
 230         int irq, irq2, intno;
 231         struct kvm_pic *s = pic_irqchip(kvm);
 232
 233         pic_lock(s);
 234         irq = pic_get_irq(&s->pics[0]);
 235         if (irq >= 0) {
 236                 pic_intack(&s->pics[0], irq);
 237                 if (irq == 2) {
 238                         irq2 = pic_get_irq(&s->pics[1]);
 239                         if (irq2 >= 0)
 240                                 pic_intack(&s->pics[1], irq2);
 241                         else
 242                                 /*
 243                                  * spurious IRQ on slave controller
 244                                  */
 245                                 irq2 = 7;
 246                         intno = s->pics[1].irq_base + irq2;
 247                         irq = irq2 + 8;
 248                 } else
 249                         intno = s->pics[0].irq_base + irq;
 250         } else {
 251                 /*
 252                  * spurious IRQ on host controller
 253                  */
 254                 irq = 7;
 255                 intno = s->pics[0].irq_base + irq;
 256         }
 257         pic_update_irq(s);
 258         pic_unlock(s);
 259
 260         return intno;
 261 }
 262
 263 void kvm_pic_reset(struct kvm_kpic_state *s)
 264 {
 265         int irq, i;
 266         struct kvm_vcpu *vcpu;
 267         u8 irr = s->irr, isr = s->imr;
 268         bool found = false;
 269
 270         s->last_irr = 0;
 271         s->irr = 0;
 272         s->imr = 0;
 273         s->isr = 0;
 274         s->priority_add = 0;
 275         s->irq_base = 0;
 276         s->read_reg_select = 0;
 277         s->poll = 0;
 278         s->special_mask = 0;
 279         s->init_state = 0;
 280         s->auto_eoi = 0;
 281         s->rotate_on_auto_eoi = 0;
 282         s->special_fully_nested_mode = 0;
 283         s->init4 = 0;
 284
 285         kvm_for_each_vcpu(i, vcpu, s->pics_state->kvm)
 286                 if (kvm_apic_accept_pic_intr(vcpu)) {
 287                         found = true;
 288                         break;
 289                 }
 290
 291
 292         if (!found)
 293                 return;
 294
 295         for (irq = 0; irq < PIC_NUM_PINS/2; irq++)
 296                 if (irr & (1 << irq) || isr & (1 << irq))
 297                         pic_clear_isr(s, irq);
 298 }
 299
 300 static void pic_ioport_write(void *opaque, u32 addr, u32 val)
 301 {
 302         struct kvm_kpic_state *s = opaque;
 303         int priority, cmd, irq;
 304
 305         addr &= 1;
 306         if (addr == 0) {
 307                 if (val & 0x10) {
 308                         s->init4 = val & 1;
 309                         s->last_irr = 0;
 310                         s->imr = 0;
 311                         s->priority_add = 0;
 312                         s->special_mask = 0;
 313                         s->read_reg_select = 0;
 314                         if (!s->init4) {
 315                                 s->special_fully_nested_mode = 0;
 316                                 s->auto_eoi = 0;
 317                         }
 318                         s->init_state = 1;
 319                         if (val & 0x02)
 320                                 pr_pic_unimpl("single mode not supported");
 321                         if (val & 0x08)
 322                                 pr_pic_unimpl(
 323                                         "level sensitive irq not supported");
 324                 } else if (val & 0x08) {
 325                         if (val & 0x04)
 326                                 s->poll = 1;
 327                         if (val & 0x02)
 328                                 s->read_reg_select = val & 1;
 329                         if (val & 0x40)
 330                                 s->special_mask = (val >> 5) & 1;
 331                 } else {
 332                         cmd = val >> 5;
 333                         switch (cmd) {
 334                         case 0:
 335                         case 4:
 336                                 s->rotate_on_auto_eoi = cmd >> 2;
 337                                 break;
 338                         case 1: /* end of interrupt */
 339                         case 5:
 340                                 priority = get_priority(s, s->isr);
 341                                 if (priority != 8) {
 342                                         irq = (priority + s->priority_add) & 7;
 343                                         if (cmd == 5)
 344                                                 s->priority_add = (irq + 1) & 7;
 345                                         pic_clear_isr(s, irq);
 346                                         pic_update_irq(s->pics_state);
 347                                 }
 348                                 break;
 349                         case 3:
 350                                 irq = val & 7;
 351                                 pic_clear_isr(s, irq);
 352                                 pic_update_irq(s->pics_state);
 353                                 break;
 354                         case 6:
 355                                 s->priority_add = (val + 1) & 7;
 356                                 pic_update_irq(s->pics_state);
 357                                 break;
 358                         case 7:
 359                                 irq = val & 7;
 360                                 s->priority_add = (irq + 1) & 7;
 361                                 pic_clear_isr(s, irq);
 362                                 pic_update_irq(s->pics_state);
 363                                 break;
 364                         default:
 365                                 break;  /* no operation */
 366                         }
 367                 }
 368         } else
 369                 switch (s->init_state) {
 370                 case 0: { /* normal mode */
 371                         u8 imr_diff = s->imr ^ val,
 372                                 off = (s == &s->pics_state->pics[0]) ? 0 : 8;
 373                         s->imr = val;
 374                         for (irq = 0; irq < PIC_NUM_PINS/2; irq++)
 375                                 if (imr_diff & (1 << irq))
 376                                         kvm_fire_mask_notifiers(
 377                                                 s->pics_state->kvm,
 378                                                 SELECT_PIC(irq + off),
 379                                                 irq + off,
 380                                                 !!(s->imr & (1 << irq)));
 381                         pic_update_irq(s->pics_state);
 382                         break;
 383                 }
 384                 case 1:
 385                         s->irq_base = val & 0xf8;
 386                         s->init_state = 2;
 387                         break;
 388                 case 2:
 389                         if (s->init4)
 390                                 s->init_state = 3;
 391                         else
 392                                 s->init_state = 0;
 393                         break;
 394                 case 3:
 395                         s->special_fully_nested_mode = (val >> 4) & 1;
 396                         s->auto_eoi = (val >> 1) & 1;
 397                         s->init_state = 0;
 398                         break;
 399                 }
 400 }
 401
 402 static u32 pic_poll_read(struct kvm_kpic_state *s, u32 addr1)
 403 {
 404         int ret;
 405
 406         ret = pic_get_irq(s);
 407         if (ret >= 0) {
 408                 if (addr1 >> 7) {
 409                         s->pics_state->pics[0].isr &= ~(1 << 2);
 410                         s->pics_state->pics[0].irr &= ~(1 << 2);
 411                 }
 412                 s->irr &= ~(1 << ret);
 413                 pic_clear_isr(s, ret);
 414                 if (addr1 >> 7 || ret != 2)
 415                         pic_update_irq(s->pics_state);
 416         } else {
 417                 ret = 0x07;
 418                 pic_update_irq(s->pics_state);
 419         }
 420
 421         return ret;
 422 }
 423
 424 static u32 pic_ioport_read(void *opaque, u32 addr1)
 425 {
 426         struct kvm_kpic_state *s = opaque;
 427         unsigned int addr;
 428         int ret;
 429
 430         addr = addr1;
 431         addr &= 1;
 432         if (s->poll) {
 433                 ret = pic_poll_read(s, addr1);
 434                 s->poll = 0;
 435         } else
 436                 if (addr == 0)
 437                         if (s->read_reg_select)
 438                                 ret = s->isr;
 439                         else
 440                                 ret = s->irr;
 441                 else
 442                         ret = s->imr;
 443         return ret;
 444 }
 445
 446 static void elcr_ioport_write(void *opaque, u32 addr, u32 val)
 447 {
 448         struct kvm_kpic_state *s = opaque;
 449         s->elcr = val & s->elcr_mask;
 450 }
 451
 452 static u32 elcr_ioport_read(void *opaque, u32 addr1)
 453 {
 454         struct kvm_kpic_state *s = opaque;
 455         return s->elcr;
 456 }
 457
 458 static int picdev_in_range(gpa_t addr)
 459 {
 460         switch (addr) {
 461         case 0x20:
 462         case 0x21:
 463         case 0xa0:
 464         case 0xa1:
 465         case 0x4d0:
 466         case 0x4d1:
 467                 return 1;
 468         default:
 469                 return 0;
 470         }
 471 }
 472
 473 static int picdev_write(struct kvm_pic *s,
 474                          gpa_t addr, int len, const void *val)
 475 {
 476         unsigned char data = *(unsigned char *)val;
 477         if (!picdev_in_range(addr))
 478                 return -EOPNOTSUPP;
 479
 480         if (len != 1) {
 481                 pr_pic_unimpl("non byte write\n");
 482                 return 0;
 483         }
 484         pic_lock(s);
 485         switch (addr) {
 486         case 0x20:
 487         case 0x21:
 488         case 0xa0:
 489         case 0xa1:
 490                 pic_ioport_write(&s->pics[addr >> 7], addr, data);
 491                 break;
 492         case 0x4d0:
 493         case 0x4d1:
 494                 elcr_ioport_write(&s->pics[addr & 1], addr, data);
 495                 break;
 496         }
 497         pic_unlock(s);
 498         return 0;
 499 }
 500
 501 static int picdev_read(struct kvm_pic *s,
 502                        gpa_t addr, int len, void *val)
 503 {
 504         unsigned char data = 0;
 505         if (!picdev_in_range(addr))
 506                 return -EOPNOTSUPP;
 507
 508         if (len != 1) {
 509                 pr_pic_unimpl("non byte read\n");
 510                 return 0;
 511         }
 512         pic_lock(s);
 513         switch (addr) {
 514         case 0x20:
 515         case 0x21:
 516         case 0xa0:
 517         case 0xa1:
 518                 data = pic_ioport_read(&s->pics[addr >> 7], addr);
 519                 break;
 520         case 0x4d0:
 521         case 0x4d1:
 522                 data = elcr_ioport_read(&s->pics[addr & 1], addr);
 523                 break;
 524         }
 525         *(unsigned char *)val = data;
 526         pic_unlock(s);
 527         return 0;
 528 }
 529
 530 static int picdev_master_write(struct kvm_io_device *dev,
 531                                gpa_t addr, int len, const void *val)
 532 {
 533         return picdev_write(container_of(dev, struct kvm_pic, dev_master),
 534                             addr, len, val);
 535 }
 536
 537 static int picdev_master_read(struct kvm_io_device *dev,
 538                               gpa_t addr, int len, void *val)
 539 {
 540         return picdev_read(container_of(dev, struct kvm_pic, dev_master),
 541                             addr, len, val);
 542 }
 543
 544 static int picdev_slave_write(struct kvm_io_device *dev,
 545                               gpa_t addr, int len, const void *val)
 546 {
 547         return picdev_write(container_of(dev, struct kvm_pic, dev_slave),
 548                             addr, len, val);
 549 }
 550
 551 static int picdev_slave_read(struct kvm_io_device *dev,
 552                              gpa_t addr, int len, void *val)
 553 {
 554         return picdev_read(container_of(dev, struct kvm_pic, dev_slave),
 555                             addr, len, val);
 556 }
 557
 558 static int picdev_eclr_write(struct kvm_io_device *dev,
 559                              gpa_t addr, int len, const void *val)
 560 {
 561         return picdev_write(container_of(dev, struct kvm_pic, dev_eclr),
 562                             addr, len, val);
 563 }
 564
 565 static int picdev_eclr_read(struct kvm_io_device *dev,
 566                             gpa_t addr, int len, void *val)
 567 {
 568         return picdev_read(container_of(dev, struct kvm_pic, dev_eclr),
 569                             addr, len, val);
 570 }
 571
 572 /*
 573  * callback when PIC0 irq status changed
 574  */
 575 static void pic_irq_request(struct kvm *kvm, int level)
 576 {
 577         struct kvm_pic *s = pic_irqchip(kvm);
 578
 579         if (!s->output)
 580                 s->wakeup_needed = true;
 581         s->output = level;
 582 }
 583
 584 static const struct kvm_io_device_ops picdev_master_ops = {
 585         .read     = picdev_master_read,
 586         .write    = picdev_master_write,
 587 };
 588
 589 static const struct kvm_io_device_ops picdev_slave_ops = {
 590         .read     = picdev_slave_read,
 591         .write    = picdev_slave_write,
 592 };
 593
 594 static const struct kvm_io_device_ops picdev_eclr_ops = {
 595         .read     = picdev_eclr_read,
 596         .write    = picdev_eclr_write,
 597 };
 598
 599 struct kvm_pic *kvm_create_pic(struct kvm *kvm)
 600 {
 601         struct kvm_pic *s;
 602         int ret;
 603
 604         s = kzalloc(sizeof(struct kvm_pic), GFP_KERNEL);
 605         if (!s)
 606                 return NULL;
 607         spin_lock_init(&s->lock);
 608         s->kvm = kvm;
 609         s->pics[0].elcr_mask = 0xf8;
 610         s->pics[1].elcr_mask = 0xde;
 611         s->pics[0].pics_state = s;
 612         s->pics[1].pics_state = s;
 613
 614         /*
 615          * Initialize PIO device
 616          */
 617         kvm_iodevice_init(&s->dev_master, &picdev_master_ops);
 618         kvm_iodevice_init(&s->dev_slave, &picdev_slave_ops);
 619         kvm_iodevice_init(&s->dev_eclr, &picdev_eclr_ops);
 620         mutex_lock(&kvm->slots_lock);
 621         ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS, 0x20, 2,
 622                                       &s->dev_master);
 623         if (ret < 0)
 624                 goto fail_unlock;
 625
 626         ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS, 0xa0, 2, &s->dev_slave);
 627         if (ret < 0)
 628                 goto fail_unreg_2;
 629
 630         ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS, 0x4d0, 2, &s->dev_eclr);
 631         if (ret < 0)
 632                 goto fail_unreg_1;
 633
 634         mutex_unlock(&kvm->slots_lock);
 635
 636         return s;
 637
 638 fail_unreg_1:
 639         kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &s->dev_slave);
 640
 641 fail_unreg_2:
 642         kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &s->dev_master);
 643
 644 fail_unlock:
 645         mutex_unlock(&kvm->slots_lock);
 646
 647         kfree(s);
 648
 649         return NULL;
 650 }
 651
 652 void kvm_destroy_pic(struct kvm *kvm)
 653 {
 654         struct kvm_pic *vpic = kvm->arch.vpic;
 655
 656         if (vpic) {
 657                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &vpic->dev_master);
 658                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &vpic->dev_slave);
 659                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &vpic->dev_eclr);
 660                 kvm->arch.vpic = NULL;
 661                 kfree(vpic);
 662         }
 663 }