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 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 "qemu/error-report.h"
  25 #include "monitor/monitor.h"
  26 #include "hw/hw.h"
  27 #include "hw/i386/pc.h"
  28 #include "hw/i386/apic.h"
  29 #include "hw/i386/ioapic.h"
  30 #include "hw/i386/ioapic_internal.h"
  31 #include "include/hw/pci/msi.h"
  32 #include "sysemu/kvm.h"
  33 #include "target-i386/cpu.h"
  34 #include "hw/i386/apic-msidef.h"
  35 #include "hw/i386/x86-iommu.h"
  36
  37 //#define DEBUG_IOAPIC
  38
  39 #ifdef DEBUG_IOAPIC
  40 #define DPRINTF(fmt, ...)                                       \
  41     do { printf("ioapic: " fmt , ## __VA_ARGS__); } while (0)
  42 #else
  43 #define DPRINTF(fmt, ...)
  44 #endif
  45
  46 #define APIC_DELIVERY_MODE_SHIFT 8
  47 #define APIC_POLARITY_SHIFT 14
  48 #define APIC_TRIG_MODE_SHIFT 15
  49
  50 static IOAPICCommonState *ioapics[MAX_IOAPICS];
  51
  52 /* global variable from ioapic_common.c */
  53 extern int ioapic_no;
  54
  55 struct ioapic_entry_info {
  56     /* fields parsed from IOAPIC entries */
  57     uint8_t masked;
  58     uint8_t trig_mode;
  59     uint16_t dest_idx;
  60     uint8_t dest_mode;
  61     uint8_t delivery_mode;
  62     uint8_t vector;
  63
  64     /* MSI message generated from above parsed fields */
  65     uint32_t addr;
  66     uint32_t data;
  67 };
  68
  69 static void ioapic_entry_parse(uint64_t entry, struct ioapic_entry_info *info)
  70 {
  71     memset(info, 0, sizeof(*info));
  72     info->masked = (entry >> IOAPIC_LVT_MASKED_SHIFT) & 1;
  73     info->trig_mode = (entry >> IOAPIC_LVT_TRIGGER_MODE_SHIFT) & 1;
  74     /*
  75      * By default, this would be dest_id[8] + reserved[8]. When IR
  76      * is enabled, this would be interrupt_index[15] +
  77      * interrupt_format[1]. This field never means anything, but
  78      * only used to generate corresponding MSI.
  79      */
  80     info->dest_idx = (entry >> IOAPIC_LVT_DEST_IDX_SHIFT) & 0xffff;
  81     info->dest_mode = (entry >> IOAPIC_LVT_DEST_MODE_SHIFT) & 1;
  82     info->delivery_mode = (entry >> IOAPIC_LVT_DELIV_MODE_SHIFT) \
  83         & IOAPIC_DM_MASK;
  84     if (info->delivery_mode == IOAPIC_DM_EXTINT) {
  85         info->vector = pic_read_irq(isa_pic);
  86     } else {
  87         info->vector = entry & IOAPIC_VECTOR_MASK;
  88     }
  89
  90     info->addr = APIC_DEFAULT_ADDRESS | \
  91         (info->dest_idx << MSI_ADDR_DEST_IDX_SHIFT) | \
  92         (info->dest_mode << MSI_ADDR_DEST_MODE_SHIFT);
  93     info->data = (info->vector << MSI_DATA_VECTOR_SHIFT) | \
  94         (info->trig_mode << MSI_DATA_TRIGGER_SHIFT) | \
  95         (info->delivery_mode << MSI_DATA_DELIVERY_MODE_SHIFT);
  96 }
  97
  98 static void ioapic_service(IOAPICCommonState *s)
  99 {
 100     AddressSpace *ioapic_as = PC_MACHINE(qdev_get_machine())->ioapic_as;
 101     struct ioapic_entry_info info;
 102     uint8_t i;
 103     uint32_t mask;
 104     uint64_t entry;
 105
 106     for (i = 0; i < IOAPIC_NUM_PINS; i++) {
 107         mask = 1 << i;
 108         if (s->irr & mask) {
 109             int coalesce = 0;
 110
 111             entry = s->ioredtbl[i];
 112             ioapic_entry_parse(entry, &info);
 113             if (!info.masked) {
 114                 if (info.trig_mode == IOAPIC_TRIGGER_EDGE) {
 115                     s->irr &= ~mask;
 116                 } else {
 117                     coalesce = s->ioredtbl[i] & IOAPIC_LVT_REMOTE_IRR;
 118                     s->ioredtbl[i] |= IOAPIC_LVT_REMOTE_IRR;
 119                 }
 120
 121                 if (coalesce) {
 122                     /* We are level triggered interrupts, and the
 123                      * guest should be still working on previous one,
 124                      * so skip it. */
 125                     continue;
 126                 }
 127
 128 #ifdef CONFIG_KVM
 129                 if (kvm_irqchip_is_split()) {
 130                     if (info.trig_mode == IOAPIC_TRIGGER_EDGE) {
 131                         kvm_set_irq(kvm_state, i, 1);
 132                         kvm_set_irq(kvm_state, i, 0);
 133                     } else {
 134                         kvm_set_irq(kvm_state, i, 1);
 135                     }
 136                     continue;
 137                 }
 138 #endif
 139
 140                 /* No matter whether IR is enabled, we translate
 141                  * the IOAPIC message into a MSI one, and its
 142                  * address space will decide whether we need a
 143                  * translation. */
 144                 stl_le_phys(ioapic_as, info.addr, info.data);
 145             }
 146         }
 147     }
 148 }
 149
 150 static void ioapic_set_irq(void *opaque, int vector, int level)
 151 {
 152     IOAPICCommonState *s = opaque;
 153
 154     /* ISA IRQs map to GSI 1-1 except for IRQ0 which maps
 155      * to GSI 2.  GSI maps to ioapic 1-1.  This is not
 156      * the cleanest way of doing it but it should work. */
 157
 158     DPRINTF("%s: %s vec %x\n", __func__, level ? "raise" : "lower", vector);
 159     if (vector == 0) {
 160         vector = 2;
 161     }
 162     if (vector >= 0 && vector < IOAPIC_NUM_PINS) {
 163         uint32_t mask = 1 << vector;
 164         uint64_t entry = s->ioredtbl[vector];
 165
 166         if (((entry >> IOAPIC_LVT_TRIGGER_MODE_SHIFT) & 1) ==
 167             IOAPIC_TRIGGER_LEVEL) {
 168             /* level triggered */
 169             if (level) {
 170                 s->irr |= mask;
 171                 if (!(entry & IOAPIC_LVT_REMOTE_IRR)) {
 172                     ioapic_service(s);
 173                 }
 174             } else {
 175                 s->irr &= ~mask;
 176             }
 177         } else {
 178             /* According to the 82093AA manual, we must ignore edge requests
 179              * if the input pin is masked. */
 180             if (level && !(entry & IOAPIC_LVT_MASKED)) {
 181                 s->irr |= mask;
 182                 ioapic_service(s);
 183             }
 184         }
 185     }
 186 }
 187
 188 static void ioapic_update_kvm_routes(IOAPICCommonState *s)
 189 {
 190 #ifdef CONFIG_KVM
 191     int i;
 192
 193     if (kvm_irqchip_is_split()) {
 194         for (i = 0; i < IOAPIC_NUM_PINS; i++) {
 195             MSIMessage msg;
 196             struct ioapic_entry_info info;
 197             ioapic_entry_parse(s->ioredtbl[i], &info);
 198             msg.address = info.addr;
 199             msg.data = info.data;
 200             kvm_irqchip_update_msi_route(kvm_state, i, msg, NULL);
 201         }
 202         kvm_irqchip_commit_routes(kvm_state);
 203     }
 204 #endif
 205 }
 206
 207 #ifdef CONFIG_KVM
 208 static void ioapic_iec_notifier(void *private, bool global,
 209                                 uint32_t index, uint32_t mask)
 210 {
 211     IOAPICCommonState *s = (IOAPICCommonState *)private;
 212     /* For simplicity, we just update all the routes */
 213     ioapic_update_kvm_routes(s);
 214 }
 215 #endif
 216
 217 void ioapic_eoi_broadcast(int vector)
 218 {
 219     IOAPICCommonState *s;
 220     uint64_t entry;
 221     int i, n;
 222
 223     for (i = 0; i < MAX_IOAPICS; i++) {
 224         s = ioapics[i];
 225         if (!s) {
 226             continue;
 227         }
 228         for (n = 0; n < IOAPIC_NUM_PINS; n++) {
 229             entry = s->ioredtbl[n];
 230             if ((entry & IOAPIC_LVT_REMOTE_IRR)
 231                 && (entry & IOAPIC_VECTOR_MASK) == vector) {
 232                 s->ioredtbl[n] = entry & ~IOAPIC_LVT_REMOTE_IRR;
 233                 if (!(entry & IOAPIC_LVT_MASKED) && (s->irr & (1 << n))) {
 234                     ioapic_service(s);
 235                 }
 236             }
 237         }
 238     }
 239 }
 240
 241 void ioapic_dump_state(Monitor *mon, const QDict *qdict)
 242 {
 243     int i;
 244
 245     for (i = 0; i < MAX_IOAPICS; i++) {
 246         if (ioapics[i] != 0) {
 247             ioapic_print_redtbl(mon, ioapics[i]);
 248         }
 249     }
 250 }
 251
 252 static uint64_t
 253 ioapic_mem_read(void *opaque, hwaddr addr, unsigned int size)
 254 {
 255     IOAPICCommonState *s = opaque;
 256     int index;
 257     uint32_t val = 0;
 258
 259     switch (addr & 0xff) {
 260     case IOAPIC_IOREGSEL:
 261         val = s->ioregsel;
 262         break;
 263     case IOAPIC_IOWIN:
 264         if (size != 4) {
 265             break;
 266         }
 267         switch (s->ioregsel) {
 268         case IOAPIC_REG_ID:
 269         case IOAPIC_REG_ARB:
 270             val = s->id << IOAPIC_ID_SHIFT;
 271             break;
 272         case IOAPIC_REG_VER:
 273             val = s->version |
 274                 ((IOAPIC_NUM_PINS - 1) << IOAPIC_VER_ENTRIES_SHIFT);
 275             break;
 276         default:
 277             index = (s->ioregsel - IOAPIC_REG_REDTBL_BASE) >> 1;
 278             if (index >= 0 && index < IOAPIC_NUM_PINS) {
 279                 if (s->ioregsel & 1) {
 280                     val = s->ioredtbl[index] >> 32;
 281                 } else {
 282                     val = s->ioredtbl[index] & 0xffffffff;
 283                 }
 284             }
 285         }
 286         DPRINTF("read: %08x = %08x\n", s->ioregsel, val);
 287         break;
 288     }
 289     return val;
 290 }
 291
 292 /*
 293  * This is to satisfy the hack in Linux kernel. One hack of it is to
 294  * simulate clearing the Remote IRR bit of IOAPIC entry using the
 295  * following:
 296  *
 297  * "For IO-APIC's with EOI register, we use that to do an explicit EOI.
 298  * Otherwise, we simulate the EOI message manually by changing the trigger
 299  * mode to edge and then back to level, with RTE being masked during
 300  * this."
 301  *
 302  * (See linux kernel __eoi_ioapic_pin() comment in commit c0205701)
 303  *
 304  * This is based on the assumption that, Remote IRR bit will be
 305  * cleared by IOAPIC hardware when configured as edge-triggered
 306  * interrupts.
 307  *
 308  * Without this, level-triggered interrupts in IR mode might fail to
 309  * work correctly.
 310  */
 311 static inline void
 312 ioapic_fix_edge_remote_irr(uint64_t *entry)
 313 {
 314     if (!(*entry & IOAPIC_LVT_TRIGGER_MODE)) {
 315         /* Edge-triggered interrupts, make sure remote IRR is zero */
 316         *entry &= ~((uint64_t)IOAPIC_LVT_REMOTE_IRR);
 317     }
 318 }
 319
 320 static void
 321 ioapic_mem_write(void *opaque, hwaddr addr, uint64_t val,
 322                  unsigned int size)
 323 {
 324     IOAPICCommonState *s = opaque;
 325     int index;
 326
 327     switch (addr & 0xff) {
 328     case IOAPIC_IOREGSEL:
 329         s->ioregsel = val;
 330         break;
 331     case IOAPIC_IOWIN:
 332         if (size != 4) {
 333             break;
 334         }
 335         DPRINTF("write: %08x = %08" PRIx64 "\n", s->ioregsel, val);
 336         switch (s->ioregsel) {
 337         case IOAPIC_REG_ID:
 338             s->id = (val >> IOAPIC_ID_SHIFT) & IOAPIC_ID_MASK;
 339             break;
 340         case IOAPIC_REG_VER:
 341         case IOAPIC_REG_ARB:
 342             break;
 343         default:
 344             index = (s->ioregsel - IOAPIC_REG_REDTBL_BASE) >> 1;
 345             if (index >= 0 && index < IOAPIC_NUM_PINS) {
 346                 uint64_t ro_bits = s->ioredtbl[index] & IOAPIC_RO_BITS;
 347                 if (s->ioregsel & 1) {
 348                     s->ioredtbl[index] &= 0xffffffff;
 349                     s->ioredtbl[index] |= (uint64_t)val << 32;
 350                 } else {
 351                     s->ioredtbl[index] &= ~0xffffffffULL;
 352                     s->ioredtbl[index] |= val;
 353                 }
 354                 /* restore RO bits */
 355                 s->ioredtbl[index] &= IOAPIC_RW_BITS;
 356                 s->ioredtbl[index] |= ro_bits;
 357                 ioapic_fix_edge_remote_irr(&s->ioredtbl[index]);
 358                 ioapic_service(s);
 359             }
 360         }
 361         break;
 362     case IOAPIC_EOI:
 363         /* Explicit EOI is only supported for IOAPIC version 0x20 */
 364         if (size != 4 || s->version != 0x20) {
 365             break;
 366         }
 367         ioapic_eoi_broadcast(val);
 368         break;
 369     }
 370
 371     ioapic_update_kvm_routes(s);
 372 }
 373
 374 static const MemoryRegionOps ioapic_io_ops = {
 375     .read = ioapic_mem_read,
 376     .write = ioapic_mem_write,
 377     .endianness = DEVICE_NATIVE_ENDIAN,
 378 };
 379
 380 static void ioapic_machine_done_notify(Notifier *notifier, void *data)
 381 {
 382 #ifdef CONFIG_KVM
 383     IOAPICCommonState *s = container_of(notifier, IOAPICCommonState,
 384                                         machine_done);
 385
 386     if (kvm_irqchip_is_split()) {
 387         X86IOMMUState *iommu = x86_iommu_get_default();
 388         if (iommu) {
 389             /* Register this IOAPIC with IOMMU IEC notifier, so that
 390              * when there are IR invalidates, we can be notified to
 391              * update kernel IR cache. */
 392             x86_iommu_iec_register_notifier(iommu, ioapic_iec_notifier, s);
 393         }
 394     }
 395 #endif
 396 }
 397
 398 static void ioapic_realize(DeviceState *dev, Error **errp)
 399 {
 400     IOAPICCommonState *s = IOAPIC_COMMON(dev);
 401
 402     if (s->version != 0x11 && s->version != 0x20) {
 403         error_report("IOAPIC only supports version 0x11 or 0x20 "
 404                      "(default: 0x11).");
 405         exit(1);
 406     }
 407
 408     memory_region_init_io(&s->io_memory, OBJECT(s), &ioapic_io_ops, s,
 409                           "ioapic", 0x1000);
 410
 411     qdev_init_gpio_in(dev, ioapic_set_irq, IOAPIC_NUM_PINS);
 412
 413     ioapics[ioapic_no] = s;
 414     s->machine_done.notify = ioapic_machine_done_notify;
 415     qemu_add_machine_init_done_notifier(&s->machine_done);
 416 }
 417
 418 static Property ioapic_properties[] = {
 419     DEFINE_PROP_UINT8("version", IOAPICCommonState, version, 0x11),
 420     DEFINE_PROP_END_OF_LIST(),
 421 };
 422
 423 static void ioapic_class_init(ObjectClass *klass, void *data)
 424 {
 425     IOAPICCommonClass *k = IOAPIC_COMMON_CLASS(klass);
 426     DeviceClass *dc = DEVICE_CLASS(klass);
 427
 428     k->realize = ioapic_realize;
 429     dc->reset = ioapic_reset_common;
 430     dc->props = ioapic_properties;
 431 }
 432
 433 static const TypeInfo ioapic_info = {
 434     .name          = "ioapic",
 435     .parent        = TYPE_IOAPIC_COMMON,
 436     .instance_size = sizeof(IOAPICCommonState),
 437     .class_init    = ioapic_class_init,
 438 };
 439
 440 static void ioapic_register_types(void)
 441 {
 442     type_register_static(&ioapic_info);
 443 }
 444
 445 type_init(ioapic_register_types)