Merge tag 'pull-loongarch-20241016' of https://gitlab.com/gaosong/qemu into staging
[qemu/armbru.git] / hw / ppc / spapr_irq.c
blobaebd7eaabbf5f4ea073d52faa597138c4fbea141
1 /*
2 * QEMU PowerPC sPAPR IRQ interface
4 * Copyright (c) 2018, IBM Corporation.
6 * This code is licensed under the GPL version 2 or later. See the
7 * COPYING file in the top-level directory.
8 */
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"
24 #include "trace.h"
26 QEMU_BUILD_BUG_ON(SPAPR_IRQ_NR_IPIS > SPAPR_XIRQ_BASE);
28 static const TypeInfo spapr_intc_info = {
29 .name = TYPE_SPAPR_INTC,
30 .parent = TYPE_INTERFACE,
31 .class_size = sizeof(SpaprInterruptControllerClass),
34 static void spapr_irq_msi_init(SpaprMachineState *spapr)
36 if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
37 /* Legacy mode doesn't use this allocator */
38 return;
41 spapr->irq_map_nr = spapr_irq_nr_msis(spapr);
42 spapr->irq_map = bitmap_new(spapr->irq_map_nr);
45 int spapr_irq_msi_alloc(SpaprMachineState *spapr, uint32_t num, bool align,
46 Error **errp)
48 int irq;
51 * The 'align_mask' parameter of bitmap_find_next_zero_area()
52 * should be one less than a power of 2; 0 means no
53 * alignment. Adapt the 'align' value of the former allocator
54 * to fit the requirements of bitmap_find_next_zero_area()
56 align -= 1;
58 irq = bitmap_find_next_zero_area(spapr->irq_map, spapr->irq_map_nr, 0, num,
59 align);
60 if (irq == spapr->irq_map_nr) {
61 error_setg(errp, "can't find a free %d-IRQ block", num);
62 return -1;
65 bitmap_set(spapr->irq_map, irq, num);
67 return irq + SPAPR_IRQ_MSI;
70 void spapr_irq_msi_free(SpaprMachineState *spapr, int irq, uint32_t num)
72 bitmap_clear(spapr->irq_map, irq - SPAPR_IRQ_MSI, num);
75 int spapr_irq_init_kvm(SpaprInterruptControllerInitKvm fn,
76 SpaprInterruptController *intc,
77 uint32_t nr_servers,
78 Error **errp)
80 Error *local_err = NULL;
82 if (kvm_enabled() && kvm_kernel_irqchip_allowed()) {
83 if (fn(intc, nr_servers, &local_err) < 0) {
84 if (kvm_kernel_irqchip_required()) {
85 error_prepend(&local_err,
86 "kernel_irqchip requested but unavailable: ");
87 error_propagate(errp, local_err);
88 return -1;
92 * We failed to initialize the KVM device, fallback to
93 * emulated mode
95 error_prepend(&local_err,
96 "kernel_irqchip allowed but unavailable: ");
97 error_append_hint(&local_err,
98 "Falling back to kernel-irqchip=off\n");
99 warn_report_err(local_err);
103 return 0;
107 * XICS IRQ backend.
110 SpaprIrq spapr_irq_xics = {
111 .xics = true,
112 .xive = false,
116 * XIVE IRQ backend.
119 SpaprIrq spapr_irq_xive = {
120 .xics = false,
121 .xive = true,
125 * Dual XIVE and XICS IRQ backend.
127 * Both interrupt mode, XIVE and XICS, objects are created but the
128 * machine starts in legacy interrupt mode (XICS). It can be changed
129 * by the CAS negotiation process and, in that case, the new mode is
130 * activated after an extra machine reset.
134 * Define values in sync with the XIVE and XICS backend
136 SpaprIrq spapr_irq_dual = {
137 .xics = true,
138 .xive = true,
142 static int spapr_irq_check(SpaprMachineState *spapr, Error **errp)
144 ERRP_GUARD();
145 MachineState *machine = MACHINE(spapr);
148 * Sanity checks on non-P9 machines. On these, XIVE is not
149 * advertised, see spapr_dt_ov5_platform_support()
151 if (!ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00,
152 0, spapr->max_compat_pvr)) {
154 * If the 'dual' interrupt mode is selected, force XICS as CAS
155 * negotiation is useless.
157 if (spapr->irq == &spapr_irq_dual) {
158 spapr->irq = &spapr_irq_xics;
159 return 0;
163 * Non-P9 machines using only XIVE is a bogus setup. We have two
164 * scenarios to take into account because of the compat mode:
166 * 1. POWER7/8 machines should fail to init later on when creating
167 * the XIVE interrupt presenters because a POWER9 exception
168 * model is required.
170 * 2. POWER9 machines using the POWER8 compat mode won't fail and
171 * will let the OS boot with a partial XIVE setup : DT
172 * properties but no hcalls.
174 * To cover both and not confuse the OS, add an early failure in
175 * QEMU.
177 if (!spapr->irq->xics) {
178 error_setg(errp, "XIVE-only machines require a POWER9 CPU");
179 return -1;
184 * On a POWER9 host, some older KVM XICS devices cannot be destroyed and
185 * re-created. Same happens with KVM nested guests. Detect that early to
186 * avoid QEMU to exit later when the guest reboots.
188 if (kvm_enabled() &&
189 spapr->irq == &spapr_irq_dual &&
190 kvm_kernel_irqchip_required() &&
191 xics_kvm_has_broken_disconnect()) {
192 error_setg(errp,
193 "KVM is incompatible with ic-mode=dual,kernel-irqchip=on");
194 error_append_hint(errp,
195 "This can happen with an old KVM or in a KVM nested guest.\n");
196 error_append_hint(errp,
197 "Try without kernel-irqchip or with kernel-irqchip=off.\n");
198 return -1;
201 return 0;
205 * sPAPR IRQ frontend routines for devices
207 #define ALL_INTCS(spapr_) \
208 { SPAPR_INTC((spapr_)->ics), SPAPR_INTC((spapr_)->xive), }
210 int spapr_irq_cpu_intc_create(SpaprMachineState *spapr,
211 PowerPCCPU *cpu, Error **errp)
213 SpaprInterruptController *intcs[] = ALL_INTCS(spapr);
214 int i;
215 int rc;
217 for (i = 0; i < ARRAY_SIZE(intcs); i++) {
218 SpaprInterruptController *intc = intcs[i];
219 if (intc) {
220 SpaprInterruptControllerClass *sicc = SPAPR_INTC_GET_CLASS(intc);
221 rc = sicc->cpu_intc_create(intc, cpu, errp);
222 if (rc < 0) {
223 return rc;
228 return 0;
231 void spapr_irq_cpu_intc_reset(SpaprMachineState *spapr, PowerPCCPU *cpu)
233 SpaprInterruptController *intcs[] = ALL_INTCS(spapr);
234 int i;
236 for (i = 0; i < ARRAY_SIZE(intcs); i++) {
237 SpaprInterruptController *intc = intcs[i];
238 if (intc) {
239 SpaprInterruptControllerClass *sicc = SPAPR_INTC_GET_CLASS(intc);
240 sicc->cpu_intc_reset(intc, cpu);
245 void spapr_irq_cpu_intc_destroy(SpaprMachineState *spapr, PowerPCCPU *cpu)
247 SpaprInterruptController *intcs[] = ALL_INTCS(spapr);
248 int i;
250 for (i = 0; i < ARRAY_SIZE(intcs); i++) {
251 SpaprInterruptController *intc = intcs[i];
252 if (intc) {
253 SpaprInterruptControllerClass *sicc = SPAPR_INTC_GET_CLASS(intc);
254 sicc->cpu_intc_destroy(intc, cpu);
259 static void spapr_set_irq(void *opaque, int irq, int level)
261 SpaprMachineState *spapr = SPAPR_MACHINE(opaque);
262 SpaprInterruptControllerClass *sicc
263 = SPAPR_INTC_GET_CLASS(spapr->active_intc);
265 sicc->set_irq(spapr->active_intc, irq, level);
268 void spapr_irq_print_info(SpaprMachineState *spapr, GString *buf)
270 SpaprInterruptControllerClass *sicc
271 = SPAPR_INTC_GET_CLASS(spapr->active_intc);
273 sicc->print_info(spapr->active_intc, buf);
276 void spapr_irq_dt(SpaprMachineState *spapr, uint32_t nr_servers,
277 void *fdt, uint32_t phandle)
279 SpaprInterruptControllerClass *sicc
280 = SPAPR_INTC_GET_CLASS(spapr->active_intc);
282 sicc->dt(spapr->active_intc, nr_servers, fdt, phandle);
285 uint32_t spapr_irq_nr_msis(SpaprMachineState *spapr)
287 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
289 if (smc->legacy_irq_allocation) {
290 return smc->nr_xirqs;
291 } else {
292 return SPAPR_XIRQ_BASE + smc->nr_xirqs - SPAPR_IRQ_MSI;
296 void spapr_irq_init(SpaprMachineState *spapr, Error **errp)
298 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
300 if (kvm_enabled() && kvm_kernel_irqchip_split()) {
301 error_setg(errp, "kernel_irqchip split mode not supported on pseries");
302 return;
305 if (spapr_irq_check(spapr, errp) < 0) {
306 return;
309 /* Initialize the MSI IRQ allocator. */
310 spapr_irq_msi_init(spapr);
312 if (spapr->irq->xics) {
313 Object *obj;
315 obj = object_new(TYPE_ICS_SPAPR);
317 object_property_add_child(OBJECT(spapr), "ics", obj);
318 object_property_set_link(obj, ICS_PROP_XICS, OBJECT(spapr),
319 &error_abort);
320 object_property_set_int(obj, "nr-irqs", smc->nr_xirqs, &error_abort);
321 if (!qdev_realize(DEVICE(obj), NULL, errp)) {
322 return;
325 spapr->ics = ICS_SPAPR(obj);
328 if (spapr->irq->xive) {
329 uint32_t nr_servers = spapr_max_server_number(spapr);
330 DeviceState *dev;
331 int i;
333 dev = qdev_new(TYPE_SPAPR_XIVE);
334 qdev_prop_set_uint32(dev, "nr-irqs", smc->nr_xirqs + SPAPR_IRQ_NR_IPIS);
336 * 8 XIVE END structures per CPU. One for each available
337 * priority
339 qdev_prop_set_uint32(dev, "nr-ends", nr_servers << 3);
340 object_property_set_link(OBJECT(dev), "xive-fabric", OBJECT(spapr),
341 &error_abort);
342 sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
344 spapr->xive = SPAPR_XIVE(dev);
346 /* Enable the CPU IPIs */
347 for (i = 0; i < nr_servers; ++i) {
348 SpaprInterruptControllerClass *sicc
349 = SPAPR_INTC_GET_CLASS(spapr->xive);
351 if (sicc->claim_irq(SPAPR_INTC(spapr->xive), SPAPR_IRQ_IPI + i,
352 false, errp) < 0) {
353 return;
357 spapr_xive_hcall_init(spapr);
360 spapr->qirqs = qemu_allocate_irqs(spapr_set_irq, spapr,
361 smc->nr_xirqs + SPAPR_IRQ_NR_IPIS);
364 * Mostly we don't actually need this until reset, except that not
365 * having this set up can cause VFIO devices to issue a
366 * false-positive warning during realize(), because they don't yet
367 * have an in-kernel irq chip.
369 spapr_irq_update_active_intc(spapr);
372 int spapr_irq_claim(SpaprMachineState *spapr, int irq, bool lsi, Error **errp)
374 SpaprInterruptController *intcs[] = ALL_INTCS(spapr);
375 int i;
376 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
377 int rc;
379 assert(irq >= SPAPR_XIRQ_BASE);
380 assert(irq < (smc->nr_xirqs + SPAPR_XIRQ_BASE));
382 for (i = 0; i < ARRAY_SIZE(intcs); i++) {
383 SpaprInterruptController *intc = intcs[i];
384 if (intc) {
385 SpaprInterruptControllerClass *sicc = SPAPR_INTC_GET_CLASS(intc);
386 rc = sicc->claim_irq(intc, irq, lsi, errp);
387 if (rc < 0) {
388 return rc;
393 return 0;
396 void spapr_irq_free(SpaprMachineState *spapr, int irq, int num)
398 SpaprInterruptController *intcs[] = ALL_INTCS(spapr);
399 int i, j;
400 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
402 assert(irq >= SPAPR_XIRQ_BASE);
403 assert((irq + num) <= (smc->nr_xirqs + SPAPR_XIRQ_BASE));
405 for (i = irq; i < (irq + num); i++) {
406 for (j = 0; j < ARRAY_SIZE(intcs); j++) {
407 SpaprInterruptController *intc = intcs[j];
409 if (intc) {
410 SpaprInterruptControllerClass *sicc
411 = SPAPR_INTC_GET_CLASS(intc);
412 sicc->free_irq(intc, i);
418 qemu_irq spapr_qirq(SpaprMachineState *spapr, int irq)
420 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
423 * This interface is basically for VIO and PHB devices to find the
424 * right qemu_irq to manipulate, so we only allow access to the
425 * external irqs for now. Currently anything which needs to
426 * access the IPIs most naturally gets there via the guest side
427 * interfaces, we can change this if we need to in future.
429 assert(irq >= SPAPR_XIRQ_BASE);
430 assert(irq < (smc->nr_xirqs + SPAPR_XIRQ_BASE));
432 if (spapr->ics) {
433 assert(ics_valid_irq(spapr->ics, irq));
435 if (spapr->xive) {
436 assert(irq < spapr->xive->nr_irqs);
437 assert(xive_eas_is_valid(&spapr->xive->eat[irq]));
440 return spapr->qirqs[irq];
443 int spapr_irq_post_load(SpaprMachineState *spapr, int version_id)
445 SpaprInterruptControllerClass *sicc;
447 spapr_irq_update_active_intc(spapr);
448 sicc = SPAPR_INTC_GET_CLASS(spapr->active_intc);
449 return sicc->post_load(spapr->active_intc, version_id);
452 void spapr_irq_reset(SpaprMachineState *spapr, Error **errp)
454 assert(!spapr->irq_map || bitmap_empty(spapr->irq_map, spapr->irq_map_nr));
456 spapr_irq_update_active_intc(spapr);
459 int spapr_irq_get_phandle(SpaprMachineState *spapr, void *fdt, Error **errp)
461 const char *nodename = "interrupt-controller";
462 int offset, phandle;
464 offset = fdt_subnode_offset(fdt, 0, nodename);
465 if (offset < 0) {
466 error_setg(errp, "Can't find node \"%s\": %s",
467 nodename, fdt_strerror(offset));
468 return -1;
471 phandle = fdt_get_phandle(fdt, offset);
472 if (!phandle) {
473 error_setg(errp, "Can't get phandle of node \"%s\"", nodename);
474 return -1;
477 return phandle;
480 static void set_active_intc(SpaprMachineState *spapr,
481 SpaprInterruptController *new_intc)
483 SpaprInterruptControllerClass *sicc;
484 uint32_t nr_servers = spapr_max_server_number(spapr);
486 assert(new_intc);
488 if (new_intc == spapr->active_intc) {
489 /* Nothing to do */
490 return;
493 if (spapr->active_intc) {
494 sicc = SPAPR_INTC_GET_CLASS(spapr->active_intc);
495 if (sicc->deactivate) {
496 sicc->deactivate(spapr->active_intc);
500 sicc = SPAPR_INTC_GET_CLASS(new_intc);
501 if (sicc->activate) {
502 sicc->activate(new_intc, nr_servers, &error_fatal);
505 spapr->active_intc = new_intc;
508 * We've changed the kernel irqchip, let VFIO devices know they
509 * need to readjust.
511 kvm_irqchip_change_notify();
514 void spapr_irq_update_active_intc(SpaprMachineState *spapr)
516 SpaprInterruptController *new_intc;
518 if (!spapr->ics) {
520 * XXX before we run CAS, ov5_cas is initialized empty, which
521 * indicates XICS, even if we have ic-mode=xive. TODO: clean
522 * up the CAS path so that we have a clearer way of handling
523 * this.
525 new_intc = SPAPR_INTC(spapr->xive);
526 } else if (spapr->ov5_cas
527 && spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
528 new_intc = SPAPR_INTC(spapr->xive);
529 } else {
530 new_intc = SPAPR_INTC(spapr->ics);
533 set_active_intc(spapr, new_intc);
537 * XICS legacy routines - to deprecate one day
540 static int ics_find_free_block(ICSState *ics, int num, int alignnum)
542 int first, i;
544 for (first = 0; first < ics->nr_irqs; first += alignnum) {
545 if (num > (ics->nr_irqs - first)) {
546 return -1;
548 for (i = first; i < first + num; ++i) {
549 if (!ics_irq_free(ics, i)) {
550 break;
553 if (i == (first + num)) {
554 return first;
558 return -1;
561 int spapr_irq_find(SpaprMachineState *spapr, int num, bool align, Error **errp)
563 ICSState *ics = spapr->ics;
564 int first = -1;
566 assert(ics);
569 * MSIMesage::data is used for storing VIRQ so
570 * it has to be aligned to num to support multiple
571 * MSI vectors. MSI-X is not affected by this.
572 * The hint is used for the first IRQ, the rest should
573 * be allocated continuously.
575 if (align) {
576 assert((num == 1) || (num == 2) || (num == 4) ||
577 (num == 8) || (num == 16) || (num == 32));
578 first = ics_find_free_block(ics, num, num);
579 } else {
580 first = ics_find_free_block(ics, num, 1);
583 if (first < 0) {
584 error_setg(errp, "can't find a free %d-IRQ block", num);
585 return -1;
588 return first + ics->offset;
591 SpaprIrq spapr_irq_xics_legacy = {
592 .xics = true,
593 .xive = false,
596 static void spapr_irq_register_types(void)
598 type_register_static(&spapr_intc_info);
601 type_init(spapr_irq_register_types)