]> git.proxmox.com Git - mirror_qemu.git/blob - hw/ppc/spapr_irq.c
Merge remote-tracking branch 'remotes/kevin/tags/for-upstream' into staging
[mirror_qemu.git] / hw / ppc / spapr_irq.c
1 /*
2 * QEMU PowerPC sPAPR IRQ interface
3 *
4 * Copyright (c) 2018, IBM Corporation.
5 *
6 * This code is licensed under the GPL version 2 or later. See the
7 * COPYING file in the top-level directory.
8 */
9
10 #include "qemu/osdep.h"
11 #include "qemu/log.h"
12 #include "qemu/error-report.h"
13 #include "qapi/error.h"
14 #include "hw/irq.h"
15 #include "hw/ppc/spapr.h"
16 #include "hw/ppc/spapr_cpu_core.h"
17 #include "hw/ppc/spapr_xive.h"
18 #include "hw/ppc/xics.h"
19 #include "hw/ppc/xics_spapr.h"
20 #include "hw/qdev-properties.h"
21 #include "cpu-models.h"
22 #include "sysemu/kvm.h"
23
24 #include "trace.h"
25
26 static const TypeInfo spapr_intc_info = {
27 .name = TYPE_SPAPR_INTC,
28 .parent = TYPE_INTERFACE,
29 .class_size = sizeof(SpaprInterruptControllerClass),
30 };
31
32 static void spapr_irq_msi_init(SpaprMachineState *spapr)
33 {
34 if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
35 /* Legacy mode doesn't use this allocator */
36 return;
37 }
38
39 spapr->irq_map_nr = spapr_irq_nr_msis(spapr);
40 spapr->irq_map = bitmap_new(spapr->irq_map_nr);
41 }
42
43 int spapr_irq_msi_alloc(SpaprMachineState *spapr, uint32_t num, bool align,
44 Error **errp)
45 {
46 int irq;
47
48 /*
49 * The 'align_mask' parameter of bitmap_find_next_zero_area()
50 * should be one less than a power of 2; 0 means no
51 * alignment. Adapt the 'align' value of the former allocator
52 * to fit the requirements of bitmap_find_next_zero_area()
53 */
54 align -= 1;
55
56 irq = bitmap_find_next_zero_area(spapr->irq_map, spapr->irq_map_nr, 0, num,
57 align);
58 if (irq == spapr->irq_map_nr) {
59 error_setg(errp, "can't find a free %d-IRQ block", num);
60 return -1;
61 }
62
63 bitmap_set(spapr->irq_map, irq, num);
64
65 return irq + SPAPR_IRQ_MSI;
66 }
67
68 void spapr_irq_msi_free(SpaprMachineState *spapr, int irq, uint32_t num)
69 {
70 bitmap_clear(spapr->irq_map, irq - SPAPR_IRQ_MSI, num);
71 }
72
73 int spapr_irq_init_kvm(SpaprInterruptControllerInitKvm fn,
74 SpaprInterruptController *intc,
75 uint32_t nr_servers,
76 Error **errp)
77 {
78 Error *local_err = NULL;
79
80 if (kvm_enabled() && kvm_kernel_irqchip_allowed()) {
81 if (fn(intc, nr_servers, &local_err) < 0) {
82 if (kvm_kernel_irqchip_required()) {
83 error_prepend(&local_err,
84 "kernel_irqchip requested but unavailable: ");
85 error_propagate(errp, local_err);
86 return -1;
87 }
88
89 /*
90 * We failed to initialize the KVM device, fallback to
91 * emulated mode
92 */
93 error_prepend(&local_err,
94 "kernel_irqchip allowed but unavailable: ");
95 error_append_hint(&local_err,
96 "Falling back to kernel-irqchip=off\n");
97 warn_report_err(local_err);
98 }
99 }
100
101 return 0;
102 }
103
104 /*
105 * XICS IRQ backend.
106 */
107
108 SpaprIrq spapr_irq_xics = {
109 .xics = true,
110 .xive = false,
111 };
112
113 /*
114 * XIVE IRQ backend.
115 */
116
117 SpaprIrq spapr_irq_xive = {
118 .xics = false,
119 .xive = true,
120 };
121
122 /*
123 * Dual XIVE and XICS IRQ backend.
124 *
125 * Both interrupt mode, XIVE and XICS, objects are created but the
126 * machine starts in legacy interrupt mode (XICS). It can be changed
127 * by the CAS negotiation process and, in that case, the new mode is
128 * activated after an extra machine reset.
129 */
130
131 /*
132 * Define values in sync with the XIVE and XICS backend
133 */
134 SpaprIrq spapr_irq_dual = {
135 .xics = true,
136 .xive = true,
137 };
138
139
140 static int spapr_irq_check(SpaprMachineState *spapr, Error **errp)
141 {
142 MachineState *machine = MACHINE(spapr);
143
144 /*
145 * Sanity checks on non-P9 machines. On these, XIVE is not
146 * advertised, see spapr_dt_ov5_platform_support()
147 */
148 if (!ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00,
149 0, spapr->max_compat_pvr)) {
150 /*
151 * If the 'dual' interrupt mode is selected, force XICS as CAS
152 * negotiation is useless.
153 */
154 if (spapr->irq == &spapr_irq_dual) {
155 spapr->irq = &spapr_irq_xics;
156 return 0;
157 }
158
159 /*
160 * Non-P9 machines using only XIVE is a bogus setup. We have two
161 * scenarios to take into account because of the compat mode:
162 *
163 * 1. POWER7/8 machines should fail to init later on when creating
164 * the XIVE interrupt presenters because a POWER9 exception
165 * model is required.
166
167 * 2. POWER9 machines using the POWER8 compat mode won't fail and
168 * will let the OS boot with a partial XIVE setup : DT
169 * properties but no hcalls.
170 *
171 * To cover both and not confuse the OS, add an early failure in
172 * QEMU.
173 */
174 if (spapr->irq == &spapr_irq_xive) {
175 error_setg(errp, "XIVE-only machines require a POWER9 CPU");
176 return -1;
177 }
178 }
179
180 /*
181 * On a POWER9 host, some older KVM XICS devices cannot be destroyed and
182 * re-created. Detect that early to avoid QEMU to exit later when the
183 * guest reboots.
184 */
185 if (kvm_enabled() &&
186 spapr->irq == &spapr_irq_dual &&
187 kvm_kernel_irqchip_required() &&
188 xics_kvm_has_broken_disconnect(spapr)) {
189 error_setg(errp, "KVM is too old to support ic-mode=dual,kernel-irqchip=on");
190 return -1;
191 }
192
193 return 0;
194 }
195
196 /*
197 * sPAPR IRQ frontend routines for devices
198 */
199 #define ALL_INTCS(spapr_) \
200 { SPAPR_INTC((spapr_)->ics), SPAPR_INTC((spapr_)->xive), }
201
202 int spapr_irq_cpu_intc_create(SpaprMachineState *spapr,
203 PowerPCCPU *cpu, Error **errp)
204 {
205 SpaprInterruptController *intcs[] = ALL_INTCS(spapr);
206 int i;
207 int rc;
208
209 for (i = 0; i < ARRAY_SIZE(intcs); i++) {
210 SpaprInterruptController *intc = intcs[i];
211 if (intc) {
212 SpaprInterruptControllerClass *sicc = SPAPR_INTC_GET_CLASS(intc);
213 rc = sicc->cpu_intc_create(intc, cpu, errp);
214 if (rc < 0) {
215 return rc;
216 }
217 }
218 }
219
220 return 0;
221 }
222
223 void spapr_irq_cpu_intc_reset(SpaprMachineState *spapr, PowerPCCPU *cpu)
224 {
225 SpaprInterruptController *intcs[] = ALL_INTCS(spapr);
226 int i;
227
228 for (i = 0; i < ARRAY_SIZE(intcs); i++) {
229 SpaprInterruptController *intc = intcs[i];
230 if (intc) {
231 SpaprInterruptControllerClass *sicc = SPAPR_INTC_GET_CLASS(intc);
232 sicc->cpu_intc_reset(intc, cpu);
233 }
234 }
235 }
236
237 void spapr_irq_cpu_intc_destroy(SpaprMachineState *spapr, PowerPCCPU *cpu)
238 {
239 SpaprInterruptController *intcs[] = ALL_INTCS(spapr);
240 int i;
241
242 for (i = 0; i < ARRAY_SIZE(intcs); i++) {
243 SpaprInterruptController *intc = intcs[i];
244 if (intc) {
245 SpaprInterruptControllerClass *sicc = SPAPR_INTC_GET_CLASS(intc);
246 sicc->cpu_intc_destroy(intc, cpu);
247 }
248 }
249 }
250
251 static void spapr_set_irq(void *opaque, int irq, int level)
252 {
253 SpaprMachineState *spapr = SPAPR_MACHINE(opaque);
254 SpaprInterruptControllerClass *sicc
255 = SPAPR_INTC_GET_CLASS(spapr->active_intc);
256
257 sicc->set_irq(spapr->active_intc, irq, level);
258 }
259
260 void spapr_irq_print_info(SpaprMachineState *spapr, Monitor *mon)
261 {
262 SpaprInterruptControllerClass *sicc
263 = SPAPR_INTC_GET_CLASS(spapr->active_intc);
264
265 sicc->print_info(spapr->active_intc, mon);
266 }
267
268 void spapr_irq_dt(SpaprMachineState *spapr, uint32_t nr_servers,
269 void *fdt, uint32_t phandle)
270 {
271 SpaprInterruptControllerClass *sicc
272 = SPAPR_INTC_GET_CLASS(spapr->active_intc);
273
274 sicc->dt(spapr->active_intc, nr_servers, fdt, phandle);
275 }
276
277 uint32_t spapr_irq_nr_msis(SpaprMachineState *spapr)
278 {
279 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
280
281 if (smc->legacy_irq_allocation) {
282 return smc->nr_xirqs;
283 } else {
284 return SPAPR_XIRQ_BASE + smc->nr_xirqs - SPAPR_IRQ_MSI;
285 }
286 }
287
288 void spapr_irq_init(SpaprMachineState *spapr, Error **errp)
289 {
290 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
291
292 if (kvm_enabled() && kvm_kernel_irqchip_split()) {
293 error_setg(errp, "kernel_irqchip split mode not supported on pseries");
294 return;
295 }
296
297 if (spapr_irq_check(spapr, errp) < 0) {
298 return;
299 }
300
301 /* Initialize the MSI IRQ allocator. */
302 spapr_irq_msi_init(spapr);
303
304 if (spapr->irq->xics) {
305 Error *local_err = NULL;
306 Object *obj;
307
308 obj = object_new(TYPE_ICS_SPAPR);
309
310 object_property_add_child(OBJECT(spapr), "ics", obj, &error_abort);
311 object_property_set_link(obj, OBJECT(spapr), ICS_PROP_XICS,
312 &error_abort);
313 object_property_set_int(obj, smc->nr_xirqs, "nr-irqs", &error_abort);
314 object_property_set_bool(obj, true, "realized", &local_err);
315 if (local_err) {
316 error_propagate(errp, local_err);
317 return;
318 }
319
320 spapr->ics = ICS_SPAPR(obj);
321 }
322
323 if (spapr->irq->xive) {
324 uint32_t nr_servers = spapr_max_server_number(spapr);
325 DeviceState *dev;
326 int i;
327
328 dev = qdev_create(NULL, TYPE_SPAPR_XIVE);
329 qdev_prop_set_uint32(dev, "nr-irqs", smc->nr_xirqs + SPAPR_XIRQ_BASE);
330 /*
331 * 8 XIVE END structures per CPU. One for each available
332 * priority
333 */
334 qdev_prop_set_uint32(dev, "nr-ends", nr_servers << 3);
335 qdev_init_nofail(dev);
336
337 spapr->xive = SPAPR_XIVE(dev);
338
339 /* Enable the CPU IPIs */
340 for (i = 0; i < nr_servers; ++i) {
341 SpaprInterruptControllerClass *sicc
342 = SPAPR_INTC_GET_CLASS(spapr->xive);
343
344 if (sicc->claim_irq(SPAPR_INTC(spapr->xive), SPAPR_IRQ_IPI + i,
345 false, errp) < 0) {
346 return;
347 }
348 }
349
350 spapr_xive_hcall_init(spapr);
351 }
352
353 spapr->qirqs = qemu_allocate_irqs(spapr_set_irq, spapr,
354 smc->nr_xirqs + SPAPR_XIRQ_BASE);
355
356 /*
357 * Mostly we don't actually need this until reset, except that not
358 * having this set up can cause VFIO devices to issue a
359 * false-positive warning during realize(), because they don't yet
360 * have an in-kernel irq chip.
361 */
362 spapr_irq_update_active_intc(spapr);
363 }
364
365 int spapr_irq_claim(SpaprMachineState *spapr, int irq, bool lsi, Error **errp)
366 {
367 SpaprInterruptController *intcs[] = ALL_INTCS(spapr);
368 int i;
369 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
370 int rc;
371
372 assert(irq >= SPAPR_XIRQ_BASE);
373 assert(irq < (smc->nr_xirqs + SPAPR_XIRQ_BASE));
374
375 for (i = 0; i < ARRAY_SIZE(intcs); i++) {
376 SpaprInterruptController *intc = intcs[i];
377 if (intc) {
378 SpaprInterruptControllerClass *sicc = SPAPR_INTC_GET_CLASS(intc);
379 rc = sicc->claim_irq(intc, irq, lsi, errp);
380 if (rc < 0) {
381 return rc;
382 }
383 }
384 }
385
386 return 0;
387 }
388
389 void spapr_irq_free(SpaprMachineState *spapr, int irq, int num)
390 {
391 SpaprInterruptController *intcs[] = ALL_INTCS(spapr);
392 int i, j;
393 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
394
395 assert(irq >= SPAPR_XIRQ_BASE);
396 assert((irq + num) <= (smc->nr_xirqs + SPAPR_XIRQ_BASE));
397
398 for (i = irq; i < (irq + num); i++) {
399 for (j = 0; j < ARRAY_SIZE(intcs); j++) {
400 SpaprInterruptController *intc = intcs[j];
401
402 if (intc) {
403 SpaprInterruptControllerClass *sicc
404 = SPAPR_INTC_GET_CLASS(intc);
405 sicc->free_irq(intc, i);
406 }
407 }
408 }
409 }
410
411 qemu_irq spapr_qirq(SpaprMachineState *spapr, int irq)
412 {
413 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
414
415 /*
416 * This interface is basically for VIO and PHB devices to find the
417 * right qemu_irq to manipulate, so we only allow access to the
418 * external irqs for now. Currently anything which needs to
419 * access the IPIs most naturally gets there via the guest side
420 * interfaces, we can change this if we need to in future.
421 */
422 assert(irq >= SPAPR_XIRQ_BASE);
423 assert(irq < (smc->nr_xirqs + SPAPR_XIRQ_BASE));
424
425 if (spapr->ics) {
426 assert(ics_valid_irq(spapr->ics, irq));
427 }
428 if (spapr->xive) {
429 assert(irq < spapr->xive->nr_irqs);
430 assert(xive_eas_is_valid(&spapr->xive->eat[irq]));
431 }
432
433 return spapr->qirqs[irq];
434 }
435
436 int spapr_irq_post_load(SpaprMachineState *spapr, int version_id)
437 {
438 SpaprInterruptControllerClass *sicc;
439
440 spapr_irq_update_active_intc(spapr);
441 sicc = SPAPR_INTC_GET_CLASS(spapr->active_intc);
442 return sicc->post_load(spapr->active_intc, version_id);
443 }
444
445 void spapr_irq_reset(SpaprMachineState *spapr, Error **errp)
446 {
447 assert(!spapr->irq_map || bitmap_empty(spapr->irq_map, spapr->irq_map_nr));
448
449 spapr_irq_update_active_intc(spapr);
450 }
451
452 int spapr_irq_get_phandle(SpaprMachineState *spapr, void *fdt, Error **errp)
453 {
454 const char *nodename = "interrupt-controller";
455 int offset, phandle;
456
457 offset = fdt_subnode_offset(fdt, 0, nodename);
458 if (offset < 0) {
459 error_setg(errp, "Can't find node \"%s\": %s",
460 nodename, fdt_strerror(offset));
461 return -1;
462 }
463
464 phandle = fdt_get_phandle(fdt, offset);
465 if (!phandle) {
466 error_setg(errp, "Can't get phandle of node \"%s\"", nodename);
467 return -1;
468 }
469
470 return phandle;
471 }
472
473 static void set_active_intc(SpaprMachineState *spapr,
474 SpaprInterruptController *new_intc)
475 {
476 SpaprInterruptControllerClass *sicc;
477 uint32_t nr_servers = spapr_max_server_number(spapr);
478
479 assert(new_intc);
480
481 if (new_intc == spapr->active_intc) {
482 /* Nothing to do */
483 return;
484 }
485
486 if (spapr->active_intc) {
487 sicc = SPAPR_INTC_GET_CLASS(spapr->active_intc);
488 if (sicc->deactivate) {
489 sicc->deactivate(spapr->active_intc);
490 }
491 }
492
493 sicc = SPAPR_INTC_GET_CLASS(new_intc);
494 if (sicc->activate) {
495 sicc->activate(new_intc, nr_servers, &error_fatal);
496 }
497
498 spapr->active_intc = new_intc;
499
500 /*
501 * We've changed the kernel irqchip, let VFIO devices know they
502 * need to readjust.
503 */
504 kvm_irqchip_change_notify();
505 }
506
507 void spapr_irq_update_active_intc(SpaprMachineState *spapr)
508 {
509 SpaprInterruptController *new_intc;
510
511 if (!spapr->ics) {
512 /*
513 * XXX before we run CAS, ov5_cas is initialized empty, which
514 * indicates XICS, even if we have ic-mode=xive. TODO: clean
515 * up the CAS path so that we have a clearer way of handling
516 * this.
517 */
518 new_intc = SPAPR_INTC(spapr->xive);
519 } else if (spapr->ov5_cas
520 && spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
521 new_intc = SPAPR_INTC(spapr->xive);
522 } else {
523 new_intc = SPAPR_INTC(spapr->ics);
524 }
525
526 set_active_intc(spapr, new_intc);
527 }
528
529 /*
530 * XICS legacy routines - to deprecate one day
531 */
532
533 static int ics_find_free_block(ICSState *ics, int num, int alignnum)
534 {
535 int first, i;
536
537 for (first = 0; first < ics->nr_irqs; first += alignnum) {
538 if (num > (ics->nr_irqs - first)) {
539 return -1;
540 }
541 for (i = first; i < first + num; ++i) {
542 if (!ics_irq_free(ics, i)) {
543 break;
544 }
545 }
546 if (i == (first + num)) {
547 return first;
548 }
549 }
550
551 return -1;
552 }
553
554 int spapr_irq_find(SpaprMachineState *spapr, int num, bool align, Error **errp)
555 {
556 ICSState *ics = spapr->ics;
557 int first = -1;
558
559 assert(ics);
560
561 /*
562 * MSIMesage::data is used for storing VIRQ so
563 * it has to be aligned to num to support multiple
564 * MSI vectors. MSI-X is not affected by this.
565 * The hint is used for the first IRQ, the rest should
566 * be allocated continuously.
567 */
568 if (align) {
569 assert((num == 1) || (num == 2) || (num == 4) ||
570 (num == 8) || (num == 16) || (num == 32));
571 first = ics_find_free_block(ics, num, num);
572 } else {
573 first = ics_find_free_block(ics, num, 1);
574 }
575
576 if (first < 0) {
577 error_setg(errp, "can't find a free %d-IRQ block", num);
578 return -1;
579 }
580
581 return first + ics->offset;
582 }
583
584 SpaprIrq spapr_irq_xics_legacy = {
585 .xics = true,
586 .xive = false,
587 };
588
589 static void spapr_irq_register_types(void)
590 {
591 type_register_static(&spapr_intc_info);
592 }
593
594 type_init(spapr_irq_register_types)