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 "qapi/error.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 "hw/pci/msi.h"
  32 #include "sysemu/kvm.h"
  33 #include "hw/i386/apic-msidef.h"
  34 #include "hw/i386/x86-iommu.h"
  35 #include "trace.h"
  36
  37 #define APIC_DELIVERY_MODE_SHIFT 8
  38 #define APIC_POLARITY_SHIFT 14
  39 #define APIC_TRIG_MODE_SHIFT 15
  40
  41 static IOAPICCommonState *ioapics[MAX_IOAPICS];
  42
  43 /* global variable from ioapic_common.c */
  44 extern int ioapic_no;
  45
  46 struct ioapic_entry_info {
  47     /* fields parsed from IOAPIC entries */
  48     uint8_t masked;
  49     uint8_t trig_mode;
  50     uint16_t dest_idx;
  51     uint8_t dest_mode;
  52     uint8_t delivery_mode;
  53     uint8_t vector;
  54
  55     /* MSI message generated from above parsed fields */
  56     uint32_t addr;
  57     uint32_t data;
  58 };
  59
  60 static void ioapic_entry_parse(uint64_t entry, struct ioapic_entry_info *info)
  61 {
  62     memset(info, 0, sizeof(*info));
  63     info->masked = (entry >> IOAPIC_LVT_MASKED_SHIFT) & 1;
  64     info->trig_mode = (entry >> IOAPIC_LVT_TRIGGER_MODE_SHIFT) & 1;
  65     /*
  66      * By default, this would be dest_id[8] + reserved[8]. When IR
  67      * is enabled, this would be interrupt_index[15] +
  68      * interrupt_format[1]. This field never means anything, but
  69      * only used to generate corresponding MSI.
  70      */
  71     info->dest_idx = (entry >> IOAPIC_LVT_DEST_IDX_SHIFT) & 0xffff;
  72     info->dest_mode = (entry >> IOAPIC_LVT_DEST_MODE_SHIFT) & 1;
  73     info->delivery_mode = (entry >> IOAPIC_LVT_DELIV_MODE_SHIFT) \
  74         & IOAPIC_DM_MASK;
  75     if (info->delivery_mode == IOAPIC_DM_EXTINT) {
  76         info->vector = pic_read_irq(isa_pic);
  77     } else {
  78         info->vector = entry & IOAPIC_VECTOR_MASK;
  79     }
  80
  81     info->addr = APIC_DEFAULT_ADDRESS | \
  82         (info->dest_idx << MSI_ADDR_DEST_IDX_SHIFT) | \
  83         (info->dest_mode << MSI_ADDR_DEST_MODE_SHIFT);
  84     info->data = (info->vector << MSI_DATA_VECTOR_SHIFT) | \
  85         (info->trig_mode << MSI_DATA_TRIGGER_SHIFT) | \
  86         (info->delivery_mode << MSI_DATA_DELIVERY_MODE_SHIFT);
  87 }
  88
  89 static void ioapic_service(IOAPICCommonState *s)
  90 {
  91     AddressSpace *ioapic_as = PC_MACHINE(qdev_get_machine())->ioapic_as;
  92     struct ioapic_entry_info info;
  93     uint8_t i;
  94     uint32_t mask;
  95     uint64_t entry;
  96
  97     for (i = 0; i < IOAPIC_NUM_PINS; i++) {
  98         mask = 1 << i;
  99         if (s->irr & mask) {
 100             int coalesce = 0;
 101
 102             entry = s->ioredtbl[i];
 103             ioapic_entry_parse(entry, &info);
 104             if (!info.masked) {
 105                 if (info.trig_mode == IOAPIC_TRIGGER_EDGE) {
 106                     s->irr &= ~mask;
 107                 } else {
 108                     coalesce = s->ioredtbl[i] & IOAPIC_LVT_REMOTE_IRR;
 109                     trace_ioapic_set_remote_irr(i);
 110                     s->ioredtbl[i] |= IOAPIC_LVT_REMOTE_IRR;
 111                 }
 112
 113                 if (coalesce) {
 114                     /* We are level triggered interrupts, and the
 115                      * guest should be still working on previous one,
 116                      * so skip it. */
 117                     continue;
 118                 }
 119
 120 #ifdef CONFIG_KVM
 121                 if (kvm_irqchip_is_split()) {
 122                     if (info.trig_mode == IOAPIC_TRIGGER_EDGE) {
 123                         kvm_set_irq(kvm_state, i, 1);
 124                         kvm_set_irq(kvm_state, i, 0);
 125                     } else {
 126                         kvm_set_irq(kvm_state, i, 1);
 127                     }
 128                     continue;
 129                 }
 130 #endif
 131
 132                 /* No matter whether IR is enabled, we translate
 133                  * the IOAPIC message into a MSI one, and its
 134                  * address space will decide whether we need a
 135                  * translation. */
 136                 stl_le_phys(ioapic_as, info.addr, info.data);
 137             }
 138         }
 139     }
 140 }
 141
 142 static void ioapic_set_irq(void *opaque, int vector, int level)
 143 {
 144     IOAPICCommonState *s = opaque;
 145
 146     /* ISA IRQs map to GSI 1-1 except for IRQ0 which maps
 147      * to GSI 2.  GSI maps to ioapic 1-1.  This is not
 148      * the cleanest way of doing it but it should work. */
 149
 150     trace_ioapic_set_irq(vector, level);
 151     ioapic_stat_update_irq(s, vector, level);
 152     if (vector == 0) {
 153         vector = 2;
 154     }
 155     if (vector < IOAPIC_NUM_PINS) {
 156         uint32_t mask = 1 << vector;
 157         uint64_t entry = s->ioredtbl[vector];
 158
 159         if (((entry >> IOAPIC_LVT_TRIGGER_MODE_SHIFT) & 1) ==
 160             IOAPIC_TRIGGER_LEVEL) {
 161             /* level triggered */
 162             if (level) {
 163                 s->irr |= mask;
 164                 if (!(entry & IOAPIC_LVT_REMOTE_IRR)) {
 165                     ioapic_service(s);
 166                 }
 167             } else {
 168                 s->irr &= ~mask;
 169             }
 170         } else {
 171             /* According to the 82093AA manual, we must ignore edge requests
 172              * if the input pin is masked. */
 173             if (level && !(entry & IOAPIC_LVT_MASKED)) {
 174                 s->irr |= mask;
 175                 ioapic_service(s);
 176             }
 177         }
 178     }
 179 }
 180
 181 static void ioapic_update_kvm_routes(IOAPICCommonState *s)
 182 {
 183 #ifdef CONFIG_KVM
 184     int i;
 185
 186     if (kvm_irqchip_is_split()) {
 187         for (i = 0; i < IOAPIC_NUM_PINS; i++) {
 188             MSIMessage msg;
 189             struct ioapic_entry_info info;
 190             ioapic_entry_parse(s->ioredtbl[i], &info);
 191             msg.address = info.addr;
 192             msg.data = info.data;
 193             kvm_irqchip_update_msi_route(kvm_state, i, msg, NULL);
 194         }
 195         kvm_irqchip_commit_routes(kvm_state);
 196     }
 197 #endif
 198 }
 199
 200 #ifdef CONFIG_KVM
 201 static void ioapic_iec_notifier(void *private, bool global,
 202                                 uint32_t index, uint32_t mask)
 203 {
 204     IOAPICCommonState *s = (IOAPICCommonState *)private;
 205     /* For simplicity, we just update all the routes */
 206     ioapic_update_kvm_routes(s);
 207 }
 208 #endif
 209
 210 void ioapic_eoi_broadcast(int vector)
 211 {
 212     IOAPICCommonState *s;
 213     uint64_t entry;
 214     int i, n;
 215
 216     trace_ioapic_eoi_broadcast(vector);
 217
 218     for (i = 0; i < MAX_IOAPICS; i++) {
 219         s = ioapics[i];
 220         if (!s) {
 221             continue;
 222         }
 223         for (n = 0; n < IOAPIC_NUM_PINS; n++) {
 224             entry = s->ioredtbl[n];
 225             if ((entry & IOAPIC_LVT_REMOTE_IRR)
 226                 && (entry & IOAPIC_VECTOR_MASK) == vector) {
 227                 trace_ioapic_clear_remote_irr(n, vector);
 228                 s->ioredtbl[n] = entry & ~IOAPIC_LVT_REMOTE_IRR;
 229                 if (!(entry & IOAPIC_LVT_MASKED) && (s->irr & (1 << n))) {
 230                     ioapic_service(s);
 231                 }
 232             }
 233         }
 234     }
 235 }
 236
 237 static uint64_t
 238 ioapic_mem_read(void *opaque, hwaddr addr, unsigned int size)
 239 {
 240     IOAPICCommonState *s = opaque;
 241     int index;
 242     uint32_t val = 0;
 243
 244     addr &= 0xff;
 245
 246     switch (addr) {
 247     case IOAPIC_IOREGSEL:
 248         val = s->ioregsel;
 249         break;
 250     case IOAPIC_IOWIN:
 251         if (size != 4) {
 252             break;
 253         }
 254         switch (s->ioregsel) {
 255         case IOAPIC_REG_ID:
 256         case IOAPIC_REG_ARB:
 257             val = s->id << IOAPIC_ID_SHIFT;
 258             break;
 259         case IOAPIC_REG_VER:
 260             val = s->version |
 261                 ((IOAPIC_NUM_PINS - 1) << IOAPIC_VER_ENTRIES_SHIFT);
 262             break;
 263         default:
 264             index = (s->ioregsel - IOAPIC_REG_REDTBL_BASE) >> 1;
 265             if (index >= 0 && index < IOAPIC_NUM_PINS) {
 266                 if (s->ioregsel & 1) {
 267                     val = s->ioredtbl[index] >> 32;
 268                 } else {
 269                     val = s->ioredtbl[index] & 0xffffffff;
 270                 }
 271             }
 272         }
 273         break;
 274     }
 275
 276     trace_ioapic_mem_read(addr, s->ioregsel, size, val);
 277
 278     return val;
 279 }
 280
 281 /*
 282  * This is to satisfy the hack in Linux kernel. One hack of it is to
 283  * simulate clearing the Remote IRR bit of IOAPIC entry using the
 284  * following:
 285  *
 286  * "For IO-APIC's with EOI register, we use that to do an explicit EOI.
 287  * Otherwise, we simulate the EOI message manually by changing the trigger
 288  * mode to edge and then back to level, with RTE being masked during
 289  * this."
 290  *
 291  * (See linux kernel __eoi_ioapic_pin() comment in commit c0205701)
 292  *
 293  * This is based on the assumption that, Remote IRR bit will be
 294  * cleared by IOAPIC hardware when configured as edge-triggered
 295  * interrupts.
 296  *
 297  * Without this, level-triggered interrupts in IR mode might fail to
 298  * work correctly.
 299  */
 300 static inline void
 301 ioapic_fix_edge_remote_irr(uint64_t *entry)
 302 {
 303     if (!(*entry & IOAPIC_LVT_TRIGGER_MODE)) {
 304         /* Edge-triggered interrupts, make sure remote IRR is zero */
 305         *entry &= ~((uint64_t)IOAPIC_LVT_REMOTE_IRR);
 306     }
 307 }
 308
 309 static void
 310 ioapic_mem_write(void *opaque, hwaddr addr, uint64_t val,
 311                  unsigned int size)
 312 {
 313     IOAPICCommonState *s = opaque;
 314     int index;
 315
 316     addr &= 0xff;
 317     trace_ioapic_mem_write(addr, s->ioregsel, size, val);
 318
 319     switch (addr) {
 320     case IOAPIC_IOREGSEL:
 321         s->ioregsel = val;
 322         break;
 323     case IOAPIC_IOWIN:
 324         if (size != 4) {
 325             break;
 326         }
 327         switch (s->ioregsel) {
 328         case IOAPIC_REG_ID:
 329             s->id = (val >> IOAPIC_ID_SHIFT) & IOAPIC_ID_MASK;
 330             break;
 331         case IOAPIC_REG_VER:
 332         case IOAPIC_REG_ARB:
 333             break;
 334         default:
 335             index = (s->ioregsel - IOAPIC_REG_REDTBL_BASE) >> 1;
 336             if (index >= 0 && index < IOAPIC_NUM_PINS) {
 337                 uint64_t ro_bits = s->ioredtbl[index] & IOAPIC_RO_BITS;
 338                 if (s->ioregsel & 1) {
 339                     s->ioredtbl[index] &= 0xffffffff;
 340                     s->ioredtbl[index] |= (uint64_t)val << 32;
 341                 } else {
 342                     s->ioredtbl[index] &= ~0xffffffffULL;
 343                     s->ioredtbl[index] |= val;
 344                 }
 345                 /* restore RO bits */
 346                 s->ioredtbl[index] &= IOAPIC_RW_BITS;
 347                 s->ioredtbl[index] |= ro_bits;
 348                 ioapic_fix_edge_remote_irr(&s->ioredtbl[index]);
 349                 ioapic_service(s);
 350             }
 351         }
 352         break;
 353     case IOAPIC_EOI:
 354         /* Explicit EOI is only supported for IOAPIC version 0x20 */
 355         if (size != 4 || s->version != 0x20) {
 356             break;
 357         }
 358         ioapic_eoi_broadcast(val);
 359         break;
 360     }
 361
 362     ioapic_update_kvm_routes(s);
 363 }
 364
 365 static const MemoryRegionOps ioapic_io_ops = {
 366     .read = ioapic_mem_read,
 367     .write = ioapic_mem_write,
 368     .endianness = DEVICE_NATIVE_ENDIAN,
 369 };
 370
 371 static void ioapic_machine_done_notify(Notifier *notifier, void *data)
 372 {
 373 #ifdef CONFIG_KVM
 374     IOAPICCommonState *s = container_of(notifier, IOAPICCommonState,
 375                                         machine_done);
 376
 377     if (kvm_irqchip_is_split()) {
 378         X86IOMMUState *iommu = x86_iommu_get_default();
 379         if (iommu) {
 380             /* Register this IOAPIC with IOMMU IEC notifier, so that
 381              * when there are IR invalidates, we can be notified to
 382              * update kernel IR cache. */
 383             x86_iommu_iec_register_notifier(iommu, ioapic_iec_notifier, s);
 384         }
 385     }
 386 #endif
 387 }
 388
 389 #define IOAPIC_VER_DEF 0x20
 390
 391 static void ioapic_realize(DeviceState *dev, Error **errp)
 392 {
 393     IOAPICCommonState *s = IOAPIC_COMMON(dev);
 394
 395     if (s->version != 0x11 && s->version != 0x20) {
 396         error_setg(errp, "IOAPIC only supports version 0x11 or 0x20 "
 397                    "(default: 0x%x).", IOAPIC_VER_DEF);
 398         return;
 399     }
 400
 401     memory_region_init_io(&s->io_memory, OBJECT(s), &ioapic_io_ops, s,
 402                           "ioapic", 0x1000);
 403
 404     qdev_init_gpio_in(dev, ioapic_set_irq, IOAPIC_NUM_PINS);
 405
 406     ioapics[ioapic_no] = s;
 407     s->machine_done.notify = ioapic_machine_done_notify;
 408     qemu_add_machine_init_done_notifier(&s->machine_done);
 409 }
 410
 411 static Property ioapic_properties[] = {
 412     DEFINE_PROP_UINT8("version", IOAPICCommonState, version, IOAPIC_VER_DEF),
 413     DEFINE_PROP_END_OF_LIST(),
 414 };
 415
 416 static void ioapic_class_init(ObjectClass *klass, void *data)
 417 {
 418     IOAPICCommonClass *k = IOAPIC_COMMON_CLASS(klass);
 419     DeviceClass *dc = DEVICE_CLASS(klass);
 420
 421     k->realize = ioapic_realize;
 422     /*
 423      * If APIC is in kernel, we need to update the kernel cache after
 424      * migration, otherwise first 24 gsi routes will be invalid.
 425      */
 426     k->post_load = ioapic_update_kvm_routes;
 427     dc->reset = ioapic_reset_common;
 428     dc->props = ioapic_properties;
 429 }
 430
 431 static const TypeInfo ioapic_info = {
 432     .name          = "ioapic",
 433     .parent        = TYPE_IOAPIC_COMMON,
 434     .instance_size = sizeof(IOAPICCommonState),
 435     .class_init    = ioapic_class_init,
 436 };
 437
 438 static void ioapic_register_types(void)
 439 {
 440     type_register_static(&ioapic_info);
 441 }
 442
 443 type_init(ioapic_register_types)