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