2 * ioapic.c IOAPIC emulation logic
4 * Copyright (c) 2004-2005 Fabrice Bellard
6 * Split the ioapic logic from apic.c
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.
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.
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/>.
23 #include "qemu/osdep.h"
24 #include "qemu/error-report.h"
25 #include "monitor/monitor.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"
38 #define APIC_DELIVERY_MODE_SHIFT 8
39 #define APIC_POLARITY_SHIFT 14
40 #define APIC_TRIG_MODE_SHIFT 15
42 static IOAPICCommonState *ioapics[MAX_IOAPICS];
44 /* global variable from ioapic_common.c */
47 struct ioapic_entry_info {
48 /* fields parsed from IOAPIC entries */
53 uint8_t delivery_mode;
56 /* MSI message generated from above parsed fields */
61 static void ioapic_entry_parse(uint64_t entry, struct ioapic_entry_info *info)
63 memset(info, 0, sizeof(*info));
64 info->masked = (entry >> IOAPIC_LVT_MASKED_SHIFT) & 1;
65 info->trig_mode = (entry >> IOAPIC_LVT_TRIGGER_MODE_SHIFT) & 1;
67 * By default, this would be dest_id[8] + reserved[8]. When IR
68 * is enabled, this would be interrupt_index[15] +
69 * interrupt_format[1]. This field never means anything, but
70 * only used to generate corresponding MSI.
72 info->dest_idx = (entry >> IOAPIC_LVT_DEST_IDX_SHIFT) & 0xffff;
73 info->dest_mode = (entry >> IOAPIC_LVT_DEST_MODE_SHIFT) & 1;
74 info->delivery_mode = (entry >> IOAPIC_LVT_DELIV_MODE_SHIFT) \
76 if (info->delivery_mode == IOAPIC_DM_EXTINT) {
77 info->vector = pic_read_irq(isa_pic);
79 info->vector = entry & IOAPIC_VECTOR_MASK;
82 info->addr = APIC_DEFAULT_ADDRESS | \
83 (info->dest_idx << MSI_ADDR_DEST_IDX_SHIFT) | \
84 (info->dest_mode << MSI_ADDR_DEST_MODE_SHIFT);
85 info->data = (info->vector << MSI_DATA_VECTOR_SHIFT) | \
86 (info->trig_mode << MSI_DATA_TRIGGER_SHIFT) | \
87 (info->delivery_mode << MSI_DATA_DELIVERY_MODE_SHIFT);
90 static void ioapic_service(IOAPICCommonState *s)
92 AddressSpace *ioapic_as = PC_MACHINE(qdev_get_machine())->ioapic_as;
93 struct ioapic_entry_info info;
98 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
103 entry = s->ioredtbl[i];
104 ioapic_entry_parse(entry, &info);
106 if (info.trig_mode == IOAPIC_TRIGGER_EDGE) {
109 coalesce = s->ioredtbl[i] & IOAPIC_LVT_REMOTE_IRR;
110 trace_ioapic_set_remote_irr(i);
111 s->ioredtbl[i] |= IOAPIC_LVT_REMOTE_IRR;
115 /* We are level triggered interrupts, and the
116 * guest should be still working on previous one,
122 if (kvm_irqchip_is_split()) {
123 if (info.trig_mode == IOAPIC_TRIGGER_EDGE) {
124 kvm_set_irq(kvm_state, i, 1);
125 kvm_set_irq(kvm_state, i, 0);
127 kvm_set_irq(kvm_state, i, 1);
133 /* No matter whether IR is enabled, we translate
134 * the IOAPIC message into a MSI one, and its
135 * address space will decide whether we need a
137 stl_le_phys(ioapic_as, info.addr, info.data);
143 static void ioapic_set_irq(void *opaque, int vector, int level)
145 IOAPICCommonState *s = opaque;
147 /* ISA IRQs map to GSI 1-1 except for IRQ0 which maps
148 * to GSI 2. GSI maps to ioapic 1-1. This is not
149 * the cleanest way of doing it but it should work. */
151 trace_ioapic_set_irq(vector, level);
155 if (vector >= 0 && vector < IOAPIC_NUM_PINS) {
156 uint32_t mask = 1 << vector;
157 uint64_t entry = s->ioredtbl[vector];
159 if (((entry >> IOAPIC_LVT_TRIGGER_MODE_SHIFT) & 1) ==
160 IOAPIC_TRIGGER_LEVEL) {
161 /* level triggered */
164 if (!(entry & IOAPIC_LVT_REMOTE_IRR)) {
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)) {
181 static void ioapic_update_kvm_routes(IOAPICCommonState *s)
186 if (kvm_irqchip_is_split()) {
187 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
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);
195 kvm_irqchip_commit_routes(kvm_state);
201 static void ioapic_iec_notifier(void *private, bool global,
202 uint32_t index, uint32_t mask)
204 IOAPICCommonState *s = (IOAPICCommonState *)private;
205 /* For simplicity, we just update all the routes */
206 ioapic_update_kvm_routes(s);
210 void ioapic_eoi_broadcast(int vector)
212 IOAPICCommonState *s;
216 trace_ioapic_eoi_broadcast(vector);
218 for (i = 0; i < MAX_IOAPICS; i++) {
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))) {
237 void ioapic_dump_state(Monitor *mon, const QDict *qdict)
241 for (i = 0; i < MAX_IOAPICS; i++) {
242 if (ioapics[i] != 0) {
243 ioapic_print_redtbl(mon, ioapics[i]);
249 ioapic_mem_read(void *opaque, hwaddr addr, unsigned int size)
251 IOAPICCommonState *s = opaque;
258 case IOAPIC_IOREGSEL:
265 switch (s->ioregsel) {
268 val = s->id << IOAPIC_ID_SHIFT;
272 ((IOAPIC_NUM_PINS - 1) << IOAPIC_VER_ENTRIES_SHIFT);
275 index = (s->ioregsel - IOAPIC_REG_REDTBL_BASE) >> 1;
276 if (index >= 0 && index < IOAPIC_NUM_PINS) {
277 if (s->ioregsel & 1) {
278 val = s->ioredtbl[index] >> 32;
280 val = s->ioredtbl[index] & 0xffffffff;
287 trace_ioapic_mem_read(addr, s->ioregsel, size, val);
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
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
302 * (See linux kernel __eoi_ioapic_pin() comment in commit c0205701)
304 * This is based on the assumption that, Remote IRR bit will be
305 * cleared by IOAPIC hardware when configured as edge-triggered
308 * Without this, level-triggered interrupts in IR mode might fail to
312 ioapic_fix_edge_remote_irr(uint64_t *entry)
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);
321 ioapic_mem_write(void *opaque, hwaddr addr, uint64_t val,
324 IOAPICCommonState *s = opaque;
328 trace_ioapic_mem_write(addr, s->ioregsel, size, val);
331 case IOAPIC_IOREGSEL:
338 switch (s->ioregsel) {
340 s->id = (val >> IOAPIC_ID_SHIFT) & IOAPIC_ID_MASK;
346 index = (s->ioregsel - IOAPIC_REG_REDTBL_BASE) >> 1;
347 if (index >= 0 && index < IOAPIC_NUM_PINS) {
348 uint64_t ro_bits = s->ioredtbl[index] & IOAPIC_RO_BITS;
349 if (s->ioregsel & 1) {
350 s->ioredtbl[index] &= 0xffffffff;
351 s->ioredtbl[index] |= (uint64_t)val << 32;
353 s->ioredtbl[index] &= ~0xffffffffULL;
354 s->ioredtbl[index] |= val;
356 /* restore RO bits */
357 s->ioredtbl[index] &= IOAPIC_RW_BITS;
358 s->ioredtbl[index] |= ro_bits;
359 ioapic_fix_edge_remote_irr(&s->ioredtbl[index]);
365 /* Explicit EOI is only supported for IOAPIC version 0x20 */
366 if (size != 4 || s->version != 0x20) {
369 ioapic_eoi_broadcast(val);
373 ioapic_update_kvm_routes(s);
376 static const MemoryRegionOps ioapic_io_ops = {
377 .read = ioapic_mem_read,
378 .write = ioapic_mem_write,
379 .endianness = DEVICE_NATIVE_ENDIAN,
382 static void ioapic_machine_done_notify(Notifier *notifier, void *data)
385 IOAPICCommonState *s = container_of(notifier, IOAPICCommonState,
388 if (kvm_irqchip_is_split()) {
389 X86IOMMUState *iommu = x86_iommu_get_default();
391 /* Register this IOAPIC with IOMMU IEC notifier, so that
392 * when there are IR invalidates, we can be notified to
393 * update kernel IR cache. */
394 x86_iommu_iec_register_notifier(iommu, ioapic_iec_notifier, s);
400 #define IOAPIC_VER_DEF 0x20
402 static void ioapic_realize(DeviceState *dev, Error **errp)
404 IOAPICCommonState *s = IOAPIC_COMMON(dev);
406 if (s->version != 0x11 && s->version != 0x20) {
407 error_report("IOAPIC only supports version 0x11 or 0x20 "
408 "(default: 0x%x).", IOAPIC_VER_DEF);
412 memory_region_init_io(&s->io_memory, OBJECT(s), &ioapic_io_ops, s,
415 qdev_init_gpio_in(dev, ioapic_set_irq, IOAPIC_NUM_PINS);
417 ioapics[ioapic_no] = s;
418 s->machine_done.notify = ioapic_machine_done_notify;
419 qemu_add_machine_init_done_notifier(&s->machine_done);
422 static Property ioapic_properties[] = {
423 DEFINE_PROP_UINT8("version", IOAPICCommonState, version, IOAPIC_VER_DEF),
424 DEFINE_PROP_END_OF_LIST(),
427 static void ioapic_class_init(ObjectClass *klass, void *data)
429 IOAPICCommonClass *k = IOAPIC_COMMON_CLASS(klass);
430 DeviceClass *dc = DEVICE_CLASS(klass);
432 k->realize = ioapic_realize;
434 * If APIC is in kernel, we need to update the kernel cache after
435 * migration, otherwise first 24 gsi routes will be invalid.
437 k->post_load = ioapic_update_kvm_routes;
438 dc->reset = ioapic_reset_common;
439 dc->props = ioapic_properties;
442 static const TypeInfo ioapic_info = {
444 .parent = TYPE_IOAPIC_COMMON,
445 .instance_size = sizeof(IOAPICCommonState),
446 .class_init = ioapic_class_init,
449 static void ioapic_register_types(void)
451 type_register_static(&ioapic_info);
454 type_init(ioapic_register_types)