hw/intc/ioapic.c

   1 /*
   2  *  ioapic.c IOAPIC emulation logic
   3  *
   4  *  Copyright (c) 2004-2005 Fabrice Bellard
   5  *
   6  *  Split the ioapic logic from apic.c
   7  *  Xiantao Zhang <xiantao.zhang@intel.com>
   8  *
   9  * This library is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public
  11  * License as published by the Free Software Foundation; either
  12  * version 2.1 of the License, or (at your option) any later version.
  13  *
  14  * This library 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  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
  21  */
  22
  23 #include "qemu/osdep.h"
  24 #include "qapi/error.h"
  25 #include "monitor/monitor.h"
  26 #include "hw/i386/apic.h"
  27 #include "hw/i386/x86.h"
  28 #include "hw/intc/i8259.h"
  29 #include "hw/intc/ioapic.h"
  30 #include "hw/intc/ioapic_internal.h"
  31 #include "hw/pci/msi.h"
  32 #include "hw/qdev-properties.h"
  33 #include "sysemu/kvm.h"
  34 #include "sysemu/sysemu.h"
  35 #include "hw/i386/apic-msidef.h"
  36 #include "hw/i386/x86-iommu.h"
  37 #include "trace.h"
  38
  39 #define APIC_DELIVERY_MODE_SHIFT 8
  40 #define APIC_POLARITY_SHIFT 14
  41 #define APIC_TRIG_MODE_SHIFT 15
  42
  43 static IOAPICCommonState *ioapics[MAX_IOAPICS];
  44
  45 /* global variable from ioapic_common.c */
  46 extern int ioapic_no;
  47
  48 struct ioapic_entry_info {
  49     /* fields parsed from IOAPIC entries */
  50     uint8_t masked;
  51     uint8_t trig_mode;
  52     uint16_t dest_idx;
  53     uint8_t dest_mode;
  54     uint8_t delivery_mode;
  55     uint8_t vector;
  56
  57     /* MSI message generated from above parsed fields */
  58     uint32_t addr;
  59     uint32_t data;
  60 };
  61
  62 static void ioapic_entry_parse(uint64_t entry, struct ioapic_entry_info *info)
  63 {
  64     memset(info, 0, sizeof(*info));
  65     info->masked = (entry >> IOAPIC_LVT_MASKED_SHIFT) & 1;
  66     info->trig_mode = (entry >> IOAPIC_LVT_TRIGGER_MODE_SHIFT) & 1;
  67     /*
  68      * By default, this would be dest_id[8] + reserved[8]. When IR
  69      * is enabled, this would be interrupt_index[15] +
  70      * interrupt_format[1]. This field never means anything, but
  71      * only used to generate corresponding MSI.
  72      */
  73     info->dest_idx = (entry >> IOAPIC_LVT_DEST_IDX_SHIFT) & 0xffff;
  74     info->dest_mode = (entry >> IOAPIC_LVT_DEST_MODE_SHIFT) & 1;
  75     info->delivery_mode = (entry >> IOAPIC_LVT_DELIV_MODE_SHIFT) \
  76         & IOAPIC_DM_MASK;
  77     if (info->delivery_mode == IOAPIC_DM_EXTINT) {
  78         info->vector = pic_read_irq(isa_pic);
  79     } else {
  80         info->vector = entry & IOAPIC_VECTOR_MASK;
  81     }
  82
  83     info->addr = APIC_DEFAULT_ADDRESS | \
  84         (info->dest_idx << MSI_ADDR_DEST_IDX_SHIFT) | \
  85         (info->dest_mode << MSI_ADDR_DEST_MODE_SHIFT);
  86     info->data = (info->vector << MSI_DATA_VECTOR_SHIFT) | \
  87         (info->trig_mode << MSI_DATA_TRIGGER_SHIFT) | \
  88         (info->delivery_mode << MSI_DATA_DELIVERY_MODE_SHIFT);
  89 }
  90
  91 static void ioapic_service(IOAPICCommonState *s)
  92 {
  93     AddressSpace *ioapic_as = X86_MACHINE(qdev_get_machine())->ioapic_as;
  94     struct ioapic_entry_info info;
  95     uint8_t i;
  96     uint32_t mask;
  97     uint64_t entry;
  98
  99     for (i = 0; i < IOAPIC_NUM_PINS; i++) {
 100         mask = 1 << i;
 101         if (s->irr & mask) {
 102             int coalesce = 0;
 103
 104             entry = s->ioredtbl[i];
 105             ioapic_entry_parse(entry, &info);
 106             if (!info.masked) {
 107                 if (info.trig_mode == IOAPIC_TRIGGER_EDGE) {
 108                     s->irr &= ~mask;
 109                 } else {
 110                     coalesce = s->ioredtbl[i] & IOAPIC_LVT_REMOTE_IRR;
 111                     trace_ioapic_set_remote_irr(i);
 112                     s->ioredtbl[i] |= IOAPIC_LVT_REMOTE_IRR;
 113                 }
 114
 115                 if (coalesce) {
 116                     /* We are level triggered interrupts, and the
 117                      * guest should be still working on previous one,
 118                      * so skip it. */
 119                     continue;
 120                 }
 121
 122 #ifdef CONFIG_KVM
 123                 if (kvm_irqchip_is_split()) {
 124                     if (info.trig_mode == IOAPIC_TRIGGER_EDGE) {
 125                         kvm_set_irq(kvm_state, i, 1);
 126                         kvm_set_irq(kvm_state, i, 0);
 127                     } else {
 128                         kvm_set_irq(kvm_state, i, 1);
 129                     }
 130                     continue;
 131                 }
 132 #endif
 133
 134                 /* No matter whether IR is enabled, we translate
 135                  * the IOAPIC message into a MSI one, and its
 136                  * address space will decide whether we need a
 137                  * translation. */
 138                 stl_le_phys(ioapic_as, info.addr, info.data);
 139             }
 140         }
 141     }
 142 }
 143
 144 #define SUCCESSIVE_IRQ_MAX_COUNT 10000
 145
 146 static void delayed_ioapic_service_cb(void *opaque)
 147 {
 148     IOAPICCommonState *s = opaque;
 149
 150     ioapic_service(s);
 151 }
 152
 153 static void ioapic_set_irq(void *opaque, int vector, int level)
 154 {
 155     IOAPICCommonState *s = opaque;
 156
 157     /* ISA IRQs map to GSI 1-1 except for IRQ0 which maps
 158      * to GSI 2.  GSI maps to ioapic 1-1.  This is not
 159      * the cleanest way of doing it but it should work. */
 160
 161     trace_ioapic_set_irq(vector, level);
 162     ioapic_stat_update_irq(s, vector, level);
 163     if (vector == 0) {
 164         vector = 2;
 165     }
 166     if (vector < IOAPIC_NUM_PINS) {
 167         uint32_t mask = 1 << vector;
 168         uint64_t entry = s->ioredtbl[vector];
 169
 170         if (((entry >> IOAPIC_LVT_TRIGGER_MODE_SHIFT) & 1) ==
 171             IOAPIC_TRIGGER_LEVEL) {
 172             /* level triggered */
 173             if (level) {
 174                 s->irr |= mask;
 175                 if (!(entry & IOAPIC_LVT_REMOTE_IRR)) {
 176                     ioapic_service(s);
 177                 }
 178             } else {
 179                 s->irr &= ~mask;
 180             }
 181         } else {
 182             /* According to the 82093AA manual, we must ignore edge requests
 183              * if the input pin is masked. */
 184             if (level && !(entry & IOAPIC_LVT_MASKED)) {
 185                 s->irr |= mask;
 186                 ioapic_service(s);
 187             }
 188         }
 189     }
 190 }
 191
 192 static void ioapic_update_kvm_routes(IOAPICCommonState *s)
 193 {
 194 #ifdef CONFIG_KVM
 195     int i;
 196
 197     if (kvm_irqchip_is_split()) {
 198         for (i = 0; i < IOAPIC_NUM_PINS; i++) {
 199             MSIMessage msg;
 200             struct ioapic_entry_info info;
 201             ioapic_entry_parse(s->ioredtbl[i], &info);
 202             if (!info.masked) {
 203                 msg.address = info.addr;
 204                 msg.data = info.data;
 205                 kvm_irqchip_update_msi_route(kvm_state, i, msg, NULL);
 206             }
 207         }
 208         kvm_irqchip_commit_routes(kvm_state);
 209     }
 210 #endif
 211 }
 212
 213 #ifdef CONFIG_KVM
 214 static void ioapic_iec_notifier(void *private, bool global,
 215                                 uint32_t index, uint32_t mask)
 216 {
 217     IOAPICCommonState *s = (IOAPICCommonState *)private;
 218     /* For simplicity, we just update all the routes */
 219     ioapic_update_kvm_routes(s);
 220 }
 221 #endif
 222
 223 void ioapic_eoi_broadcast(int vector)
 224 {
 225     IOAPICCommonState *s;
 226     uint64_t entry;
 227     int i, n;
 228
 229     trace_ioapic_eoi_broadcast(vector);
 230
 231     for (i = 0; i < MAX_IOAPICS; i++) {
 232         s = ioapics[i];
 233         if (!s) {
 234             continue;
 235         }
 236         for (n = 0; n < IOAPIC_NUM_PINS; n++) {
 237             entry = s->ioredtbl[n];
 238
 239             if ((entry & IOAPIC_VECTOR_MASK) != vector ||
 240                 ((entry >> IOAPIC_LVT_TRIGGER_MODE_SHIFT) & 1) != IOAPIC_TRIGGER_LEVEL) {
 241                 continue;
 242             }
 243
 244 #ifdef CONFIG_KVM
 245             /*
 246              * When IOAPIC is in the userspace while APIC is still in
 247              * the kernel (i.e., split irqchip), we have a trick to
 248              * kick the resamplefd logic for registered irqfds from
 249              * userspace to deactivate the IRQ.  When that happens, it
 250              * means the irq bypassed userspace IOAPIC (so the irr and
 251              * remote-irr of the table entry should be bypassed too
 252              * even if interrupt come).  Still kick the resamplefds if
 253              * they're bound to the IRQ, to make sure to EOI the
 254              * interrupt for the hardware correctly.
 255              *
 256              * Note: We still need to go through the irr & remote-irr
 257              * operations below because we don't know whether there're
 258              * emulated devices that are using/sharing the same IRQ.
 259              */
 260             kvm_resample_fd_notify(n);
 261 #endif
 262
 263             if (!(entry & IOAPIC_LVT_REMOTE_IRR)) {
 264                 continue;
 265             }
 266
 267             trace_ioapic_clear_remote_irr(n, vector);
 268             s->ioredtbl[n] = entry & ~IOAPIC_LVT_REMOTE_IRR;
 269
 270             if (!(entry & IOAPIC_LVT_MASKED) && (s->irr & (1 << n))) {
 271                 ++s->irq_eoi[n];
 272                 if (s->irq_eoi[n] >= SUCCESSIVE_IRQ_MAX_COUNT) {
 273                     /*
 274                      * Real hardware does not deliver the interrupt immediately
 275                      * during eoi broadcast, and this lets a buggy guest make
 276                      * slow progress even if it does not correctly handle a
 277                      * level-triggered interrupt. Emulate this behavior if we
 278                      * detect an interrupt storm.
 279                      */
 280                     s->irq_eoi[n] = 0;
 281                     timer_mod_anticipate(s->delayed_ioapic_service_timer,
 282                                          qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) +
 283                                          NANOSECONDS_PER_SECOND / 100);
 284                     trace_ioapic_eoi_delayed_reassert(n);
 285                 } else {
 286                     ioapic_service(s);
 287                 }
 288             } else {
 289                 s->irq_eoi[n] = 0;
 290             }
 291         }
 292     }
 293 }
 294
 295 static uint64_t
 296 ioapic_mem_read(void *opaque, hwaddr addr, unsigned int size)
 297 {
 298     IOAPICCommonState *s = opaque;
 299     int index;
 300     uint32_t val = 0;
 301
 302     addr &= 0xff;
 303
 304     switch (addr) {
 305     case IOAPIC_IOREGSEL:
 306         val = s->ioregsel;
 307         break;
 308     case IOAPIC_IOWIN:
 309         if (size != 4) {
 310             break;
 311         }
 312         switch (s->ioregsel) {
 313         case IOAPIC_REG_ID:
 314         case IOAPIC_REG_ARB:
 315             val = s->id << IOAPIC_ID_SHIFT;
 316             break;
 317         case IOAPIC_REG_VER:
 318             val = s->version |
 319                 ((IOAPIC_NUM_PINS - 1) << IOAPIC_VER_ENTRIES_SHIFT);
 320             break;
 321         default:
 322             index = (s->ioregsel - IOAPIC_REG_REDTBL_BASE) >> 1;
 323             if (index >= 0 && index < IOAPIC_NUM_PINS) {
 324                 if (s->ioregsel & 1) {
 325                     val = s->ioredtbl[index] >> 32;
 326                 } else {
 327                     val = s->ioredtbl[index] & 0xffffffff;
 328                 }
 329             }
 330         }
 331         break;
 332     }
 333
 334     trace_ioapic_mem_read(addr, s->ioregsel, size, val);
 335
 336     return val;
 337 }
 338
 339 /*
 340  * This is to satisfy the hack in Linux kernel. One hack of it is to
 341  * simulate clearing the Remote IRR bit of IOAPIC entry using the
 342  * following:
 343  *
 344  * "For IO-APIC's with EOI register, we use that to do an explicit EOI.
 345  * Otherwise, we simulate the EOI message manually by changing the trigger
 346  * mode to edge and then back to level, with RTE being masked during
 347  * this."
 348  *
 349  * (See linux kernel __eoi_ioapic_pin() comment in commit c0205701)
 350  *
 351  * This is based on the assumption that, Remote IRR bit will be
 352  * cleared by IOAPIC hardware when configured as edge-triggered
 353  * interrupts.
 354  *
 355  * Without this, level-triggered interrupts in IR mode might fail to
 356  * work correctly.
 357  */
 358 static inline void
 359 ioapic_fix_edge_remote_irr(uint64_t *entry)
 360 {
 361     if (!(*entry & IOAPIC_LVT_TRIGGER_MODE)) {
 362         /* Edge-triggered interrupts, make sure remote IRR is zero */
 363         *entry &= ~((uint64_t)IOAPIC_LVT_REMOTE_IRR);
 364     }
 365 }
 366
 367 static void
 368 ioapic_mem_write(void *opaque, hwaddr addr, uint64_t val,
 369                  unsigned int size)
 370 {
 371     IOAPICCommonState *s = opaque;
 372     int index;
 373
 374     addr &= 0xff;
 375     trace_ioapic_mem_write(addr, s->ioregsel, size, val);
 376
 377     switch (addr) {
 378     case IOAPIC_IOREGSEL:
 379         s->ioregsel = val;
 380         break;
 381     case IOAPIC_IOWIN:
 382         if (size != 4) {
 383             break;
 384         }
 385         switch (s->ioregsel) {
 386         case IOAPIC_REG_ID:
 387             s->id = (val >> IOAPIC_ID_SHIFT) & IOAPIC_ID_MASK;
 388             break;
 389         case IOAPIC_REG_VER:
 390         case IOAPIC_REG_ARB:
 391             break;
 392         default:
 393             index = (s->ioregsel - IOAPIC_REG_REDTBL_BASE) >> 1;
 394             if (index >= 0 && index < IOAPIC_NUM_PINS) {
 395                 uint64_t ro_bits = s->ioredtbl[index] & IOAPIC_RO_BITS;
 396                 if (s->ioregsel & 1) {
 397                     s->ioredtbl[index] &= 0xffffffff;
 398                     s->ioredtbl[index] |= (uint64_t)val << 32;
 399                 } else {
 400                     s->ioredtbl[index] &= ~0xffffffffULL;
 401                     s->ioredtbl[index] |= val;
 402                 }
 403                 /* restore RO bits */
 404                 s->ioredtbl[index] &= IOAPIC_RW_BITS;
 405                 s->ioredtbl[index] |= ro_bits;
 406                 s->irq_eoi[index] = 0;
 407                 ioapic_fix_edge_remote_irr(&s->ioredtbl[index]);
 408                 ioapic_update_kvm_routes(s);
 409                 ioapic_service(s);
 410             }
 411         }
 412         break;
 413     case IOAPIC_EOI:
 414         /* Explicit EOI is only supported for IOAPIC version 0x20 */
 415         if (size != 4 || s->version != 0x20) {
 416             break;
 417         }
 418         ioapic_eoi_broadcast(val);
 419         break;
 420     }
 421 }
 422
 423 static const MemoryRegionOps ioapic_io_ops = {
 424     .read = ioapic_mem_read,
 425     .write = ioapic_mem_write,
 426     .endianness = DEVICE_NATIVE_ENDIAN,
 427 };
 428
 429 static void ioapic_machine_done_notify(Notifier *notifier, void *data)
 430 {
 431 #ifdef CONFIG_KVM
 432     IOAPICCommonState *s = container_of(notifier, IOAPICCommonState,
 433                                         machine_done);
 434
 435     if (kvm_irqchip_is_split()) {
 436         X86IOMMUState *iommu = x86_iommu_get_default();
 437         if (iommu) {
 438             /* Register this IOAPIC with IOMMU IEC notifier, so that
 439              * when there are IR invalidates, we can be notified to
 440              * update kernel IR cache. */
 441             x86_iommu_iec_register_notifier(iommu, ioapic_iec_notifier, s);
 442         }
 443     }
 444 #endif
 445 }
 446
 447 #define IOAPIC_VER_DEF 0x20
 448
 449 static void ioapic_realize(DeviceState *dev, Error **errp)
 450 {
 451     IOAPICCommonState *s = IOAPIC_COMMON(dev);
 452
 453     if (s->version != 0x11 && s->version != 0x20) {
 454         error_setg(errp, "IOAPIC only supports version 0x11 or 0x20 "
 455                    "(default: 0x%x).", IOAPIC_VER_DEF);
 456         return;
 457     }
 458
 459     memory_region_init_io(&s->io_memory, OBJECT(s), &ioapic_io_ops, s,
 460                           "ioapic", 0x1000);
 461
 462     s->delayed_ioapic_service_timer =
 463         timer_new_ns(QEMU_CLOCK_VIRTUAL, delayed_ioapic_service_cb, s);
 464
 465     qdev_init_gpio_in(dev, ioapic_set_irq, IOAPIC_NUM_PINS);
 466
 467     ioapics[ioapic_no] = s;
 468     s->machine_done.notify = ioapic_machine_done_notify;
 469     qemu_add_machine_init_done_notifier(&s->machine_done);
 470 }
 471
 472 static void ioapic_unrealize(DeviceState *dev)
 473 {
 474     IOAPICCommonState *s = IOAPIC_COMMON(dev);
 475
 476     timer_free(s->delayed_ioapic_service_timer);
 477 }
 478
 479 static Property ioapic_properties[] = {
 480     DEFINE_PROP_UINT8("version", IOAPICCommonState, version, IOAPIC_VER_DEF),
 481     DEFINE_PROP_END_OF_LIST(),
 482 };
 483
 484 static void ioapic_class_init(ObjectClass *klass, void *data)
 485 {
 486     IOAPICCommonClass *k = IOAPIC_COMMON_CLASS(klass);
 487     DeviceClass *dc = DEVICE_CLASS(klass);
 488
 489     k->realize = ioapic_realize;
 490     k->unrealize = ioapic_unrealize;
 491     /*
 492      * If APIC is in kernel, we need to update the kernel cache after
 493      * migration, otherwise first 24 gsi routes will be invalid.
 494      */
 495     k->post_load = ioapic_update_kvm_routes;
 496     device_class_set_legacy_reset(dc, ioapic_reset_common);
 497     device_class_set_props(dc, ioapic_properties);
 498 }
 499
 500 static const TypeInfo ioapic_info = {
 501     .name          = TYPE_IOAPIC,
 502     .parent        = TYPE_IOAPIC_COMMON,
 503     .instance_size = sizeof(IOAPICCommonState),
 504     .class_init    = ioapic_class_init,
 505 };
 506
 507 static void ioapic_register_types(void)
 508 {
 509     type_register_static(&ioapic_info);
 510 }
 511
 512 type_init(ioapic_register_types)