xref: /openbmc/qemu/hw/intc/spapr_xive.c (revision f7160f32)
1 /*
2  * QEMU PowerPC sPAPR XIVE interrupt controller model
3  *
4  * Copyright (c) 2017-2018, 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/module.h"
13 #include "qapi/error.h"
14 #include "qemu/error-report.h"
15 #include "target/ppc/cpu.h"
16 #include "sysemu/cpus.h"
17 #include "sysemu/reset.h"
18 #include "migration/vmstate.h"
19 #include "monitor/monitor.h"
20 #include "hw/ppc/fdt.h"
21 #include "hw/ppc/spapr.h"
22 #include "hw/ppc/spapr_cpu_core.h"
23 #include "hw/ppc/spapr_xive.h"
24 #include "hw/ppc/xive.h"
25 #include "hw/ppc/xive_regs.h"
26 #include "hw/qdev-properties.h"
27 
28 /*
29  * XIVE Virtualization Controller BAR and Thread Managment BAR that we
30  * use for the ESB pages and the TIMA pages
31  */
32 #define SPAPR_XIVE_VC_BASE   0x0006010000000000ull
33 #define SPAPR_XIVE_TM_BASE   0x0006030203180000ull
34 
35 /*
36  * The allocation of VP blocks is a complex operation in OPAL and the
37  * VP identifiers have a relation with the number of HW chips, the
38  * size of the VP blocks, VP grouping, etc. The QEMU sPAPR XIVE
39  * controller model does not have the same constraints and can use a
40  * simple mapping scheme of the CPU vcpu_id
41  *
42  * These identifiers are never returned to the OS.
43  */
44 
45 #define SPAPR_XIVE_NVT_BASE 0x400
46 
47 /*
48  * sPAPR NVT and END indexing helpers
49  */
50 static uint32_t spapr_xive_nvt_to_target(uint8_t nvt_blk, uint32_t nvt_idx)
51 {
52     return nvt_idx - SPAPR_XIVE_NVT_BASE;
53 }
54 
55 static void spapr_xive_cpu_to_nvt(PowerPCCPU *cpu,
56                                   uint8_t *out_nvt_blk, uint32_t *out_nvt_idx)
57 {
58     assert(cpu);
59 
60     if (out_nvt_blk) {
61         *out_nvt_blk = SPAPR_XIVE_BLOCK_ID;
62     }
63 
64     if (out_nvt_blk) {
65         *out_nvt_idx = SPAPR_XIVE_NVT_BASE + cpu->vcpu_id;
66     }
67 }
68 
69 static int spapr_xive_target_to_nvt(uint32_t target,
70                                     uint8_t *out_nvt_blk, uint32_t *out_nvt_idx)
71 {
72     PowerPCCPU *cpu = spapr_find_cpu(target);
73 
74     if (!cpu) {
75         return -1;
76     }
77 
78     spapr_xive_cpu_to_nvt(cpu, out_nvt_blk, out_nvt_idx);
79     return 0;
80 }
81 
82 /*
83  * sPAPR END indexing uses a simple mapping of the CPU vcpu_id, 8
84  * priorities per CPU
85  */
86 int spapr_xive_end_to_target(uint8_t end_blk, uint32_t end_idx,
87                              uint32_t *out_server, uint8_t *out_prio)
88 {
89 
90     assert(end_blk == SPAPR_XIVE_BLOCK_ID);
91 
92     if (out_server) {
93         *out_server = end_idx >> 3;
94     }
95 
96     if (out_prio) {
97         *out_prio = end_idx & 0x7;
98     }
99     return 0;
100 }
101 
102 static void spapr_xive_cpu_to_end(PowerPCCPU *cpu, uint8_t prio,
103                                   uint8_t *out_end_blk, uint32_t *out_end_idx)
104 {
105     assert(cpu);
106 
107     if (out_end_blk) {
108         *out_end_blk = SPAPR_XIVE_BLOCK_ID;
109     }
110 
111     if (out_end_idx) {
112         *out_end_idx = (cpu->vcpu_id << 3) + prio;
113     }
114 }
115 
116 static int spapr_xive_target_to_end(uint32_t target, uint8_t prio,
117                                     uint8_t *out_end_blk, uint32_t *out_end_idx)
118 {
119     PowerPCCPU *cpu = spapr_find_cpu(target);
120 
121     if (!cpu) {
122         return -1;
123     }
124 
125     spapr_xive_cpu_to_end(cpu, prio, out_end_blk, out_end_idx);
126     return 0;
127 }
128 
129 /*
130  * On sPAPR machines, use a simplified output for the XIVE END
131  * structure dumping only the information related to the OS EQ.
132  */
133 static void spapr_xive_end_pic_print_info(SpaprXive *xive, XiveEND *end,
134                                           Monitor *mon)
135 {
136     uint64_t qaddr_base = xive_end_qaddr(end);
137     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
138     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
139     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
140     uint32_t qentries = 1 << (qsize + 10);
141     uint32_t nvt = xive_get_field32(END_W6_NVT_INDEX, end->w6);
142     uint8_t priority = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
143 
144     monitor_printf(mon, "%3d/%d % 6d/%5d @%"PRIx64" ^%d",
145                    spapr_xive_nvt_to_target(0, nvt),
146                    priority, qindex, qentries, qaddr_base, qgen);
147 
148     xive_end_queue_pic_print_info(end, 6, mon);
149 }
150 
151 void spapr_xive_pic_print_info(SpaprXive *xive, Monitor *mon)
152 {
153     XiveSource *xsrc = &xive->source;
154     int i;
155 
156     if (kvm_irqchip_in_kernel()) {
157         Error *local_err = NULL;
158 
159         kvmppc_xive_synchronize_state(xive, &local_err);
160         if (local_err) {
161             error_report_err(local_err);
162             return;
163         }
164     }
165 
166     monitor_printf(mon, "  LISN         PQ    EISN     CPU/PRIO EQ\n");
167 
168     for (i = 0; i < xive->nr_irqs; i++) {
169         uint8_t pq = xive_source_esb_get(xsrc, i);
170         XiveEAS *eas = &xive->eat[i];
171 
172         if (!xive_eas_is_valid(eas)) {
173             continue;
174         }
175 
176         monitor_printf(mon, "  %08x %s %c%c%c %s %08x ", i,
177                        xive_source_irq_is_lsi(xsrc, i) ? "LSI" : "MSI",
178                        pq & XIVE_ESB_VAL_P ? 'P' : '-',
179                        pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
180                        xsrc->status[i] & XIVE_STATUS_ASSERTED ? 'A' : ' ',
181                        xive_eas_is_masked(eas) ? "M" : " ",
182                        (int) xive_get_field64(EAS_END_DATA, eas->w));
183 
184         if (!xive_eas_is_masked(eas)) {
185             uint32_t end_idx = xive_get_field64(EAS_END_INDEX, eas->w);
186             XiveEND *end;
187 
188             assert(end_idx < xive->nr_ends);
189             end = &xive->endt[end_idx];
190 
191             if (xive_end_is_valid(end)) {
192                 spapr_xive_end_pic_print_info(xive, end, mon);
193             }
194         }
195         monitor_printf(mon, "\n");
196     }
197 }
198 
199 void spapr_xive_mmio_set_enabled(SpaprXive *xive, bool enable)
200 {
201     memory_region_set_enabled(&xive->source.esb_mmio, enable);
202     memory_region_set_enabled(&xive->tm_mmio, enable);
203 
204     /* Disable the END ESBs until a guest OS makes use of them */
205     memory_region_set_enabled(&xive->end_source.esb_mmio, false);
206 }
207 
208 static void spapr_xive_tm_write(void *opaque, hwaddr offset,
209                           uint64_t value, unsigned size)
210 {
211     XiveTCTX *tctx = spapr_cpu_state(POWERPC_CPU(current_cpu))->tctx;
212 
213     xive_tctx_tm_write(XIVE_PRESENTER(opaque), tctx, offset, value, size);
214 }
215 
216 static uint64_t spapr_xive_tm_read(void *opaque, hwaddr offset, unsigned size)
217 {
218     XiveTCTX *tctx = spapr_cpu_state(POWERPC_CPU(current_cpu))->tctx;
219 
220     return xive_tctx_tm_read(XIVE_PRESENTER(opaque), tctx, offset, size);
221 }
222 
223 const MemoryRegionOps spapr_xive_tm_ops = {
224     .read = spapr_xive_tm_read,
225     .write = spapr_xive_tm_write,
226     .endianness = DEVICE_BIG_ENDIAN,
227     .valid = {
228         .min_access_size = 1,
229         .max_access_size = 8,
230     },
231     .impl = {
232         .min_access_size = 1,
233         .max_access_size = 8,
234     },
235 };
236 
237 static void spapr_xive_end_reset(XiveEND *end)
238 {
239     memset(end, 0, sizeof(*end));
240 
241     /* switch off the escalation and notification ESBs */
242     end->w1 = cpu_to_be32(END_W1_ESe_Q | END_W1_ESn_Q);
243 }
244 
245 static void spapr_xive_reset(void *dev)
246 {
247     SpaprXive *xive = SPAPR_XIVE(dev);
248     int i;
249 
250     /*
251      * The XiveSource has its own reset handler, which mask off all
252      * IRQs (!P|Q)
253      */
254 
255     /* Mask all valid EASs in the IRQ number space. */
256     for (i = 0; i < xive->nr_irqs; i++) {
257         XiveEAS *eas = &xive->eat[i];
258         if (xive_eas_is_valid(eas)) {
259             eas->w = cpu_to_be64(EAS_VALID | EAS_MASKED);
260         } else {
261             eas->w = 0;
262         }
263     }
264 
265     /* Clear all ENDs */
266     for (i = 0; i < xive->nr_ends; i++) {
267         spapr_xive_end_reset(&xive->endt[i]);
268     }
269 }
270 
271 static void spapr_xive_instance_init(Object *obj)
272 {
273     SpaprXive *xive = SPAPR_XIVE(obj);
274 
275     object_initialize_child(obj, "source", &xive->source, TYPE_XIVE_SOURCE);
276 
277     object_initialize_child(obj, "end_source", &xive->end_source,
278                             TYPE_XIVE_END_SOURCE);
279 
280     /* Not connected to the KVM XIVE device */
281     xive->fd = -1;
282 }
283 
284 static void spapr_xive_realize(DeviceState *dev, Error **errp)
285 {
286     SpaprXive *xive = SPAPR_XIVE(dev);
287     SpaprXiveClass *sxc = SPAPR_XIVE_GET_CLASS(xive);
288     XiveSource *xsrc = &xive->source;
289     XiveENDSource *end_xsrc = &xive->end_source;
290     Error *local_err = NULL;
291 
292     sxc->parent_realize(dev, &local_err);
293     if (local_err) {
294         error_propagate(errp, local_err);
295         return;
296     }
297 
298     if (!xive->nr_irqs) {
299         error_setg(errp, "Number of interrupt needs to be greater 0");
300         return;
301     }
302 
303     if (!xive->nr_ends) {
304         error_setg(errp, "Number of interrupt needs to be greater 0");
305         return;
306     }
307 
308     /*
309      * Initialize the internal sources, for IPIs and virtual devices.
310      */
311     object_property_set_int(OBJECT(xsrc), "nr-irqs", xive->nr_irqs,
312                             &error_fatal);
313     object_property_set_link(OBJECT(xsrc), "xive", OBJECT(xive), &error_abort);
314     if (!qdev_realize(DEVICE(xsrc), NULL, errp)) {
315         return;
316     }
317     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &xsrc->esb_mmio);
318 
319     /*
320      * Initialize the END ESB source
321      */
322     object_property_set_int(OBJECT(end_xsrc), "nr-ends", xive->nr_irqs,
323                             &error_fatal);
324     object_property_set_link(OBJECT(end_xsrc), "xive", OBJECT(xive),
325                              &error_abort);
326     if (!qdev_realize(DEVICE(end_xsrc), NULL, errp)) {
327         return;
328     }
329     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &end_xsrc->esb_mmio);
330 
331     /* Set the mapping address of the END ESB pages after the source ESBs */
332     xive->end_base = xive->vc_base + (1ull << xsrc->esb_shift) * xsrc->nr_irqs;
333 
334     /*
335      * Allocate the routing tables
336      */
337     xive->eat = g_new0(XiveEAS, xive->nr_irqs);
338     xive->endt = g_new0(XiveEND, xive->nr_ends);
339 
340     xive->nodename = g_strdup_printf("interrupt-controller@%" PRIx64,
341                            xive->tm_base + XIVE_TM_USER_PAGE * (1 << TM_SHIFT));
342 
343     qemu_register_reset(spapr_xive_reset, dev);
344 
345     /* TIMA initialization */
346     memory_region_init_io(&xive->tm_mmio, OBJECT(xive), &spapr_xive_tm_ops,
347                           xive, "xive.tima", 4ull << TM_SHIFT);
348     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &xive->tm_mmio);
349 
350     /*
351      * Map all regions. These will be enabled or disabled at reset and
352      * can also be overridden by KVM memory regions if active
353      */
354     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 0, xive->vc_base);
355     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 1, xive->end_base);
356     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 2, xive->tm_base);
357 }
358 
359 static int spapr_xive_get_eas(XiveRouter *xrtr, uint8_t eas_blk,
360                               uint32_t eas_idx, XiveEAS *eas)
361 {
362     SpaprXive *xive = SPAPR_XIVE(xrtr);
363 
364     if (eas_idx >= xive->nr_irqs) {
365         return -1;
366     }
367 
368     *eas = xive->eat[eas_idx];
369     return 0;
370 }
371 
372 static int spapr_xive_get_end(XiveRouter *xrtr,
373                               uint8_t end_blk, uint32_t end_idx, XiveEND *end)
374 {
375     SpaprXive *xive = SPAPR_XIVE(xrtr);
376 
377     if (end_idx >= xive->nr_ends) {
378         return -1;
379     }
380 
381     memcpy(end, &xive->endt[end_idx], sizeof(XiveEND));
382     return 0;
383 }
384 
385 static int spapr_xive_write_end(XiveRouter *xrtr, uint8_t end_blk,
386                                 uint32_t end_idx, XiveEND *end,
387                                 uint8_t word_number)
388 {
389     SpaprXive *xive = SPAPR_XIVE(xrtr);
390 
391     if (end_idx >= xive->nr_ends) {
392         return -1;
393     }
394 
395     memcpy(&xive->endt[end_idx], end, sizeof(XiveEND));
396     return 0;
397 }
398 
399 static int spapr_xive_get_nvt(XiveRouter *xrtr,
400                               uint8_t nvt_blk, uint32_t nvt_idx, XiveNVT *nvt)
401 {
402     uint32_t vcpu_id = spapr_xive_nvt_to_target(nvt_blk, nvt_idx);
403     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
404 
405     if (!cpu) {
406         /* TODO: should we assert() if we can find a NVT ? */
407         return -1;
408     }
409 
410     /*
411      * sPAPR does not maintain a NVT table. Return that the NVT is
412      * valid if we have found a matching CPU
413      */
414     nvt->w0 = cpu_to_be32(NVT_W0_VALID);
415     return 0;
416 }
417 
418 static int spapr_xive_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk,
419                                 uint32_t nvt_idx, XiveNVT *nvt,
420                                 uint8_t word_number)
421 {
422     /*
423      * We don't need to write back to the NVTs because the sPAPR
424      * machine should never hit a non-scheduled NVT. It should never
425      * get called.
426      */
427     g_assert_not_reached();
428 }
429 
430 static int spapr_xive_match_nvt(XivePresenter *xptr, uint8_t format,
431                                 uint8_t nvt_blk, uint32_t nvt_idx,
432                                 bool cam_ignore, uint8_t priority,
433                                 uint32_t logic_serv, XiveTCTXMatch *match)
434 {
435     CPUState *cs;
436     int count = 0;
437 
438     CPU_FOREACH(cs) {
439         PowerPCCPU *cpu = POWERPC_CPU(cs);
440         XiveTCTX *tctx = spapr_cpu_state(cpu)->tctx;
441         int ring;
442 
443         /*
444          * Skip partially initialized vCPUs. This can happen when
445          * vCPUs are hotplugged.
446          */
447         if (!tctx) {
448             continue;
449         }
450 
451         /*
452          * Check the thread context CAM lines and record matches.
453          */
454         ring = xive_presenter_tctx_match(xptr, tctx, format, nvt_blk, nvt_idx,
455                                          cam_ignore, logic_serv);
456         /*
457          * Save the matching thread interrupt context and follow on to
458          * check for duplicates which are invalid.
459          */
460         if (ring != -1) {
461             if (match->tctx) {
462                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: already found a thread "
463                               "context NVT %x/%x\n", nvt_blk, nvt_idx);
464                 return -1;
465             }
466 
467             match->ring = ring;
468             match->tctx = tctx;
469             count++;
470         }
471     }
472 
473     return count;
474 }
475 
476 static uint8_t spapr_xive_get_block_id(XiveRouter *xrtr)
477 {
478     return SPAPR_XIVE_BLOCK_ID;
479 }
480 
481 static const VMStateDescription vmstate_spapr_xive_end = {
482     .name = TYPE_SPAPR_XIVE "/end",
483     .version_id = 1,
484     .minimum_version_id = 1,
485     .fields = (VMStateField []) {
486         VMSTATE_UINT32(w0, XiveEND),
487         VMSTATE_UINT32(w1, XiveEND),
488         VMSTATE_UINT32(w2, XiveEND),
489         VMSTATE_UINT32(w3, XiveEND),
490         VMSTATE_UINT32(w4, XiveEND),
491         VMSTATE_UINT32(w5, XiveEND),
492         VMSTATE_UINT32(w6, XiveEND),
493         VMSTATE_UINT32(w7, XiveEND),
494         VMSTATE_END_OF_LIST()
495     },
496 };
497 
498 static const VMStateDescription vmstate_spapr_xive_eas = {
499     .name = TYPE_SPAPR_XIVE "/eas",
500     .version_id = 1,
501     .minimum_version_id = 1,
502     .fields = (VMStateField []) {
503         VMSTATE_UINT64(w, XiveEAS),
504         VMSTATE_END_OF_LIST()
505     },
506 };
507 
508 static int vmstate_spapr_xive_pre_save(void *opaque)
509 {
510     if (kvm_irqchip_in_kernel()) {
511         return kvmppc_xive_pre_save(SPAPR_XIVE(opaque));
512     }
513 
514     return 0;
515 }
516 
517 /*
518  * Called by the sPAPR IRQ backend 'post_load' method at the machine
519  * level.
520  */
521 static int spapr_xive_post_load(SpaprInterruptController *intc, int version_id)
522 {
523     if (kvm_irqchip_in_kernel()) {
524         return kvmppc_xive_post_load(SPAPR_XIVE(intc), version_id);
525     }
526 
527     return 0;
528 }
529 
530 static const VMStateDescription vmstate_spapr_xive = {
531     .name = TYPE_SPAPR_XIVE,
532     .version_id = 1,
533     .minimum_version_id = 1,
534     .pre_save = vmstate_spapr_xive_pre_save,
535     .post_load = NULL, /* handled at the machine level */
536     .fields = (VMStateField[]) {
537         VMSTATE_UINT32_EQUAL(nr_irqs, SpaprXive, NULL),
538         VMSTATE_STRUCT_VARRAY_POINTER_UINT32(eat, SpaprXive, nr_irqs,
539                                      vmstate_spapr_xive_eas, XiveEAS),
540         VMSTATE_STRUCT_VARRAY_POINTER_UINT32(endt, SpaprXive, nr_ends,
541                                              vmstate_spapr_xive_end, XiveEND),
542         VMSTATE_END_OF_LIST()
543     },
544 };
545 
546 static int spapr_xive_claim_irq(SpaprInterruptController *intc, int lisn,
547                                 bool lsi, Error **errp)
548 {
549     SpaprXive *xive = SPAPR_XIVE(intc);
550     XiveSource *xsrc = &xive->source;
551 
552     assert(lisn < xive->nr_irqs);
553 
554     if (xive_eas_is_valid(&xive->eat[lisn])) {
555         error_setg(errp, "IRQ %d is not free", lisn);
556         return -EBUSY;
557     }
558 
559     /*
560      * Set default values when allocating an IRQ number
561      */
562     xive->eat[lisn].w |= cpu_to_be64(EAS_VALID | EAS_MASKED);
563     if (lsi) {
564         xive_source_irq_set_lsi(xsrc, lisn);
565     }
566 
567     if (kvm_irqchip_in_kernel()) {
568         return kvmppc_xive_source_reset_one(xsrc, lisn, errp);
569     }
570 
571     return 0;
572 }
573 
574 static void spapr_xive_free_irq(SpaprInterruptController *intc, int lisn)
575 {
576     SpaprXive *xive = SPAPR_XIVE(intc);
577     assert(lisn < xive->nr_irqs);
578 
579     xive->eat[lisn].w &= cpu_to_be64(~EAS_VALID);
580 }
581 
582 static Property spapr_xive_properties[] = {
583     DEFINE_PROP_UINT32("nr-irqs", SpaprXive, nr_irqs, 0),
584     DEFINE_PROP_UINT32("nr-ends", SpaprXive, nr_ends, 0),
585     DEFINE_PROP_UINT64("vc-base", SpaprXive, vc_base, SPAPR_XIVE_VC_BASE),
586     DEFINE_PROP_UINT64("tm-base", SpaprXive, tm_base, SPAPR_XIVE_TM_BASE),
587     DEFINE_PROP_END_OF_LIST(),
588 };
589 
590 static int spapr_xive_cpu_intc_create(SpaprInterruptController *intc,
591                                       PowerPCCPU *cpu, Error **errp)
592 {
593     SpaprXive *xive = SPAPR_XIVE(intc);
594     Object *obj;
595     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
596 
597     obj = xive_tctx_create(OBJECT(cpu), XIVE_PRESENTER(xive), errp);
598     if (!obj) {
599         return -1;
600     }
601 
602     spapr_cpu->tctx = XIVE_TCTX(obj);
603     return 0;
604 }
605 
606 static void xive_tctx_set_os_cam(XiveTCTX *tctx, uint32_t os_cam)
607 {
608     uint32_t qw1w2 = cpu_to_be32(TM_QW1W2_VO | os_cam);
609     memcpy(&tctx->regs[TM_QW1_OS + TM_WORD2], &qw1w2, 4);
610 }
611 
612 static void spapr_xive_cpu_intc_reset(SpaprInterruptController *intc,
613                                      PowerPCCPU *cpu)
614 {
615     XiveTCTX *tctx = spapr_cpu_state(cpu)->tctx;
616     uint8_t  nvt_blk;
617     uint32_t nvt_idx;
618 
619     xive_tctx_reset(tctx);
620 
621     /*
622      * When a Virtual Processor is scheduled to run on a HW thread,
623      * the hypervisor pushes its identifier in the OS CAM line.
624      * Emulate the same behavior under QEMU.
625      */
626     spapr_xive_cpu_to_nvt(cpu, &nvt_blk, &nvt_idx);
627 
628     xive_tctx_set_os_cam(tctx, xive_nvt_cam_line(nvt_blk, nvt_idx));
629 }
630 
631 static void spapr_xive_cpu_intc_destroy(SpaprInterruptController *intc,
632                                         PowerPCCPU *cpu)
633 {
634     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
635 
636     xive_tctx_destroy(spapr_cpu->tctx);
637     spapr_cpu->tctx = NULL;
638 }
639 
640 static void spapr_xive_set_irq(SpaprInterruptController *intc, int irq, int val)
641 {
642     SpaprXive *xive = SPAPR_XIVE(intc);
643 
644     if (kvm_irqchip_in_kernel()) {
645         kvmppc_xive_source_set_irq(&xive->source, irq, val);
646     } else {
647         xive_source_set_irq(&xive->source, irq, val);
648     }
649 }
650 
651 static void spapr_xive_print_info(SpaprInterruptController *intc, Monitor *mon)
652 {
653     SpaprXive *xive = SPAPR_XIVE(intc);
654     CPUState *cs;
655 
656     CPU_FOREACH(cs) {
657         PowerPCCPU *cpu = POWERPC_CPU(cs);
658 
659         xive_tctx_pic_print_info(spapr_cpu_state(cpu)->tctx, mon);
660     }
661 
662     spapr_xive_pic_print_info(xive, mon);
663 }
664 
665 static void spapr_xive_dt(SpaprInterruptController *intc, uint32_t nr_servers,
666                           void *fdt, uint32_t phandle)
667 {
668     SpaprXive *xive = SPAPR_XIVE(intc);
669     int node;
670     uint64_t timas[2 * 2];
671     /* Interrupt number ranges for the IPIs */
672     uint32_t lisn_ranges[] = {
673         cpu_to_be32(SPAPR_IRQ_IPI),
674         cpu_to_be32(SPAPR_IRQ_IPI + nr_servers),
675     };
676     /*
677      * EQ size - the sizes of pages supported by the system 4K, 64K,
678      * 2M, 16M. We only advertise 64K for the moment.
679      */
680     uint32_t eq_sizes[] = {
681         cpu_to_be32(16), /* 64K */
682     };
683     /*
684      * The following array is in sync with the reserved priorities
685      * defined by the 'spapr_xive_priority_is_reserved' routine.
686      */
687     uint32_t plat_res_int_priorities[] = {
688         cpu_to_be32(7),    /* start */
689         cpu_to_be32(0xf8), /* count */
690     };
691 
692     /* Thread Interrupt Management Area : User (ring 3) and OS (ring 2) */
693     timas[0] = cpu_to_be64(xive->tm_base +
694                            XIVE_TM_USER_PAGE * (1ull << TM_SHIFT));
695     timas[1] = cpu_to_be64(1ull << TM_SHIFT);
696     timas[2] = cpu_to_be64(xive->tm_base +
697                            XIVE_TM_OS_PAGE * (1ull << TM_SHIFT));
698     timas[3] = cpu_to_be64(1ull << TM_SHIFT);
699 
700     _FDT(node = fdt_add_subnode(fdt, 0, xive->nodename));
701 
702     _FDT(fdt_setprop_string(fdt, node, "device_type", "power-ivpe"));
703     _FDT(fdt_setprop(fdt, node, "reg", timas, sizeof(timas)));
704 
705     _FDT(fdt_setprop_string(fdt, node, "compatible", "ibm,power-ivpe"));
706     _FDT(fdt_setprop(fdt, node, "ibm,xive-eq-sizes", eq_sizes,
707                      sizeof(eq_sizes)));
708     _FDT(fdt_setprop(fdt, node, "ibm,xive-lisn-ranges", lisn_ranges,
709                      sizeof(lisn_ranges)));
710 
711     /* For Linux to link the LSIs to the interrupt controller. */
712     _FDT(fdt_setprop(fdt, node, "interrupt-controller", NULL, 0));
713     _FDT(fdt_setprop_cell(fdt, node, "#interrupt-cells", 2));
714 
715     /* For SLOF */
716     _FDT(fdt_setprop_cell(fdt, node, "linux,phandle", phandle));
717     _FDT(fdt_setprop_cell(fdt, node, "phandle", phandle));
718 
719     /*
720      * The "ibm,plat-res-int-priorities" property defines the priority
721      * ranges reserved by the hypervisor
722      */
723     _FDT(fdt_setprop(fdt, 0, "ibm,plat-res-int-priorities",
724                      plat_res_int_priorities, sizeof(plat_res_int_priorities)));
725 }
726 
727 static int spapr_xive_activate(SpaprInterruptController *intc,
728                                uint32_t nr_servers, Error **errp)
729 {
730     SpaprXive *xive = SPAPR_XIVE(intc);
731 
732     if (kvm_enabled()) {
733         int rc = spapr_irq_init_kvm(kvmppc_xive_connect, intc, nr_servers,
734                                     errp);
735         if (rc < 0) {
736             return rc;
737         }
738     }
739 
740     /* Activate the XIVE MMIOs */
741     spapr_xive_mmio_set_enabled(xive, true);
742 
743     return 0;
744 }
745 
746 static void spapr_xive_deactivate(SpaprInterruptController *intc)
747 {
748     SpaprXive *xive = SPAPR_XIVE(intc);
749 
750     spapr_xive_mmio_set_enabled(xive, false);
751 
752     if (kvm_irqchip_in_kernel()) {
753         kvmppc_xive_disconnect(intc);
754     }
755 }
756 
757 static void spapr_xive_class_init(ObjectClass *klass, void *data)
758 {
759     DeviceClass *dc = DEVICE_CLASS(klass);
760     XiveRouterClass *xrc = XIVE_ROUTER_CLASS(klass);
761     SpaprInterruptControllerClass *sicc = SPAPR_INTC_CLASS(klass);
762     XivePresenterClass *xpc = XIVE_PRESENTER_CLASS(klass);
763     SpaprXiveClass *sxc = SPAPR_XIVE_CLASS(klass);
764 
765     dc->desc    = "sPAPR XIVE Interrupt Controller";
766     device_class_set_props(dc, spapr_xive_properties);
767     device_class_set_parent_realize(dc, spapr_xive_realize,
768                                     &sxc->parent_realize);
769     dc->vmsd    = &vmstate_spapr_xive;
770 
771     xrc->get_eas = spapr_xive_get_eas;
772     xrc->get_end = spapr_xive_get_end;
773     xrc->write_end = spapr_xive_write_end;
774     xrc->get_nvt = spapr_xive_get_nvt;
775     xrc->write_nvt = spapr_xive_write_nvt;
776     xrc->get_block_id = spapr_xive_get_block_id;
777 
778     sicc->activate = spapr_xive_activate;
779     sicc->deactivate = spapr_xive_deactivate;
780     sicc->cpu_intc_create = spapr_xive_cpu_intc_create;
781     sicc->cpu_intc_reset = spapr_xive_cpu_intc_reset;
782     sicc->cpu_intc_destroy = spapr_xive_cpu_intc_destroy;
783     sicc->claim_irq = spapr_xive_claim_irq;
784     sicc->free_irq = spapr_xive_free_irq;
785     sicc->set_irq = spapr_xive_set_irq;
786     sicc->print_info = spapr_xive_print_info;
787     sicc->dt = spapr_xive_dt;
788     sicc->post_load = spapr_xive_post_load;
789 
790     xpc->match_nvt  = spapr_xive_match_nvt;
791 }
792 
793 static const TypeInfo spapr_xive_info = {
794     .name = TYPE_SPAPR_XIVE,
795     .parent = TYPE_XIVE_ROUTER,
796     .instance_init = spapr_xive_instance_init,
797     .instance_size = sizeof(SpaprXive),
798     .class_init = spapr_xive_class_init,
799     .class_size = sizeof(SpaprXiveClass),
800     .interfaces = (InterfaceInfo[]) {
801         { TYPE_SPAPR_INTC },
802         { }
803     },
804 };
805 
806 static void spapr_xive_register_types(void)
807 {
808     type_register_static(&spapr_xive_info);
809 }
810 
811 type_init(spapr_xive_register_types)
812 
813 /*
814  * XIVE hcalls
815  *
816  * The terminology used by the XIVE hcalls is the following :
817  *
818  *   TARGET vCPU number
819  *   EQ     Event Queue assigned by OS to receive event data
820  *   ESB    page for source interrupt management
821  *   LISN   Logical Interrupt Source Number identifying a source in the
822  *          machine
823  *   EISN   Effective Interrupt Source Number used by guest OS to
824  *          identify source in the guest
825  *
826  * The EAS, END, NVT structures are not exposed.
827  */
828 
829 /*
830  * Linux hosts under OPAL reserve priority 7 for their own escalation
831  * interrupts (DD2.X POWER9). So we only allow the guest to use
832  * priorities [0..6].
833  */
834 static bool spapr_xive_priority_is_reserved(uint8_t priority)
835 {
836     switch (priority) {
837     case 0 ... 6:
838         return false;
839     case 7: /* OPAL escalation queue */
840     default:
841         return true;
842     }
843 }
844 
845 /*
846  * The H_INT_GET_SOURCE_INFO hcall() is used to obtain the logical
847  * real address of the MMIO page through which the Event State Buffer
848  * entry associated with the value of the "lisn" parameter is managed.
849  *
850  * Parameters:
851  * Input
852  * - R4: "flags"
853  *         Bits 0-63 reserved
854  * - R5: "lisn" is per "interrupts", "interrupt-map", or
855  *       "ibm,xive-lisn-ranges" properties, or as returned by the
856  *       ibm,query-interrupt-source-number RTAS call, or as returned
857  *       by the H_ALLOCATE_VAS_WINDOW hcall
858  *
859  * Output
860  * - R4: "flags"
861  *         Bits 0-59: Reserved
862  *         Bit 60: H_INT_ESB must be used for Event State Buffer
863  *                 management
864  *         Bit 61: 1 == LSI  0 == MSI
865  *         Bit 62: the full function page supports trigger
866  *         Bit 63: Store EOI Supported
867  * - R5: Logical Real address of full function Event State Buffer
868  *       management page, -1 if H_INT_ESB hcall flag is set to 1.
869  * - R6: Logical Real Address of trigger only Event State Buffer
870  *       management page or -1.
871  * - R7: Power of 2 page size for the ESB management pages returned in
872  *       R5 and R6.
873  */
874 
875 #define SPAPR_XIVE_SRC_H_INT_ESB     PPC_BIT(60) /* ESB manage with H_INT_ESB */
876 #define SPAPR_XIVE_SRC_LSI           PPC_BIT(61) /* Virtual LSI type */
877 #define SPAPR_XIVE_SRC_TRIGGER       PPC_BIT(62) /* Trigger and management
878                                                     on same page */
879 #define SPAPR_XIVE_SRC_STORE_EOI     PPC_BIT(63) /* Store EOI support */
880 
881 static target_ulong h_int_get_source_info(PowerPCCPU *cpu,
882                                           SpaprMachineState *spapr,
883                                           target_ulong opcode,
884                                           target_ulong *args)
885 {
886     SpaprXive *xive = spapr->xive;
887     XiveSource *xsrc = &xive->source;
888     target_ulong flags  = args[0];
889     target_ulong lisn   = args[1];
890 
891     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
892         return H_FUNCTION;
893     }
894 
895     if (flags) {
896         return H_PARAMETER;
897     }
898 
899     if (lisn >= xive->nr_irqs) {
900         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
901                       lisn);
902         return H_P2;
903     }
904 
905     if (!xive_eas_is_valid(&xive->eat[lisn])) {
906         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
907                       lisn);
908         return H_P2;
909     }
910 
911     /*
912      * All sources are emulated under the main XIVE object and share
913      * the same characteristics.
914      */
915     args[0] = 0;
916     if (!xive_source_esb_has_2page(xsrc)) {
917         args[0] |= SPAPR_XIVE_SRC_TRIGGER;
918     }
919     if (xsrc->esb_flags & XIVE_SRC_STORE_EOI) {
920         args[0] |= SPAPR_XIVE_SRC_STORE_EOI;
921     }
922 
923     /*
924      * Force the use of the H_INT_ESB hcall in case of an LSI
925      * interrupt. This is necessary under KVM to re-trigger the
926      * interrupt if the level is still asserted
927      */
928     if (xive_source_irq_is_lsi(xsrc, lisn)) {
929         args[0] |= SPAPR_XIVE_SRC_H_INT_ESB | SPAPR_XIVE_SRC_LSI;
930     }
931 
932     if (!(args[0] & SPAPR_XIVE_SRC_H_INT_ESB)) {
933         args[1] = xive->vc_base + xive_source_esb_mgmt(xsrc, lisn);
934     } else {
935         args[1] = -1;
936     }
937 
938     if (xive_source_esb_has_2page(xsrc) &&
939         !(args[0] & SPAPR_XIVE_SRC_H_INT_ESB)) {
940         args[2] = xive->vc_base + xive_source_esb_page(xsrc, lisn);
941     } else {
942         args[2] = -1;
943     }
944 
945     if (xive_source_esb_has_2page(xsrc)) {
946         args[3] = xsrc->esb_shift - 1;
947     } else {
948         args[3] = xsrc->esb_shift;
949     }
950 
951     return H_SUCCESS;
952 }
953 
954 /*
955  * The H_INT_SET_SOURCE_CONFIG hcall() is used to assign a Logical
956  * Interrupt Source to a target. The Logical Interrupt Source is
957  * designated with the "lisn" parameter and the target is designated
958  * with the "target" and "priority" parameters.  Upon return from the
959  * hcall(), no additional interrupts will be directed to the old EQ.
960  *
961  * Parameters:
962  * Input:
963  * - R4: "flags"
964  *         Bits 0-61: Reserved
965  *         Bit 62: set the "eisn" in the EAS
966  *         Bit 63: masks the interrupt source in the hardware interrupt
967  *       control structure. An interrupt masked by this mechanism will
968  *       be dropped, but it's source state bits will still be
969  *       set. There is no race-free way of unmasking and restoring the
970  *       source. Thus this should only be used in interrupts that are
971  *       also masked at the source, and only in cases where the
972  *       interrupt is not meant to be used for a large amount of time
973  *       because no valid target exists for it for example
974  * - R5: "lisn" is per "interrupts", "interrupt-map", or
975  *       "ibm,xive-lisn-ranges" properties, or as returned by the
976  *       ibm,query-interrupt-source-number RTAS call, or as returned by
977  *       the H_ALLOCATE_VAS_WINDOW hcall
978  * - R6: "target" is per "ibm,ppc-interrupt-server#s" or
979  *       "ibm,ppc-interrupt-gserver#s"
980  * - R7: "priority" is a valid priority not in
981  *       "ibm,plat-res-int-priorities"
982  * - R8: "eisn" is the guest EISN associated with the "lisn"
983  *
984  * Output:
985  * - None
986  */
987 
988 #define SPAPR_XIVE_SRC_SET_EISN PPC_BIT(62)
989 #define SPAPR_XIVE_SRC_MASK     PPC_BIT(63)
990 
991 static target_ulong h_int_set_source_config(PowerPCCPU *cpu,
992                                             SpaprMachineState *spapr,
993                                             target_ulong opcode,
994                                             target_ulong *args)
995 {
996     SpaprXive *xive = spapr->xive;
997     XiveEAS eas, new_eas;
998     target_ulong flags    = args[0];
999     target_ulong lisn     = args[1];
1000     target_ulong target   = args[2];
1001     target_ulong priority = args[3];
1002     target_ulong eisn     = args[4];
1003     uint8_t end_blk;
1004     uint32_t end_idx;
1005 
1006     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1007         return H_FUNCTION;
1008     }
1009 
1010     if (flags & ~(SPAPR_XIVE_SRC_SET_EISN | SPAPR_XIVE_SRC_MASK)) {
1011         return H_PARAMETER;
1012     }
1013 
1014     if (lisn >= xive->nr_irqs) {
1015         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1016                       lisn);
1017         return H_P2;
1018     }
1019 
1020     eas = xive->eat[lisn];
1021     if (!xive_eas_is_valid(&eas)) {
1022         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1023                       lisn);
1024         return H_P2;
1025     }
1026 
1027     /* priority 0xff is used to reset the EAS */
1028     if (priority == 0xff) {
1029         new_eas.w = cpu_to_be64(EAS_VALID | EAS_MASKED);
1030         goto out;
1031     }
1032 
1033     if (flags & SPAPR_XIVE_SRC_MASK) {
1034         new_eas.w = eas.w | cpu_to_be64(EAS_MASKED);
1035     } else {
1036         new_eas.w = eas.w & cpu_to_be64(~EAS_MASKED);
1037     }
1038 
1039     if (spapr_xive_priority_is_reserved(priority)) {
1040         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1041                       " is reserved\n", priority);
1042         return H_P4;
1043     }
1044 
1045     /*
1046      * Validate that "target" is part of the list of threads allocated
1047      * to the partition. For that, find the END corresponding to the
1048      * target.
1049      */
1050     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1051         return H_P3;
1052     }
1053 
1054     new_eas.w = xive_set_field64(EAS_END_BLOCK, new_eas.w, end_blk);
1055     new_eas.w = xive_set_field64(EAS_END_INDEX, new_eas.w, end_idx);
1056 
1057     if (flags & SPAPR_XIVE_SRC_SET_EISN) {
1058         new_eas.w = xive_set_field64(EAS_END_DATA, new_eas.w, eisn);
1059     }
1060 
1061     if (kvm_irqchip_in_kernel()) {
1062         Error *local_err = NULL;
1063 
1064         kvmppc_xive_set_source_config(xive, lisn, &new_eas, &local_err);
1065         if (local_err) {
1066             error_report_err(local_err);
1067             return H_HARDWARE;
1068         }
1069     }
1070 
1071 out:
1072     xive->eat[lisn] = new_eas;
1073     return H_SUCCESS;
1074 }
1075 
1076 /*
1077  * The H_INT_GET_SOURCE_CONFIG hcall() is used to determine to which
1078  * target/priority pair is assigned to the specified Logical Interrupt
1079  * Source.
1080  *
1081  * Parameters:
1082  * Input:
1083  * - R4: "flags"
1084  *         Bits 0-63 Reserved
1085  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1086  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1087  *       ibm,query-interrupt-source-number RTAS call, or as
1088  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1089  *
1090  * Output:
1091  * - R4: Target to which the specified Logical Interrupt Source is
1092  *       assigned
1093  * - R5: Priority to which the specified Logical Interrupt Source is
1094  *       assigned
1095  * - R6: EISN for the specified Logical Interrupt Source (this will be
1096  *       equivalent to the LISN if not changed by H_INT_SET_SOURCE_CONFIG)
1097  */
1098 static target_ulong h_int_get_source_config(PowerPCCPU *cpu,
1099                                             SpaprMachineState *spapr,
1100                                             target_ulong opcode,
1101                                             target_ulong *args)
1102 {
1103     SpaprXive *xive = spapr->xive;
1104     target_ulong flags = args[0];
1105     target_ulong lisn = args[1];
1106     XiveEAS eas;
1107     XiveEND *end;
1108     uint8_t nvt_blk;
1109     uint32_t end_idx, nvt_idx;
1110 
1111     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1112         return H_FUNCTION;
1113     }
1114 
1115     if (flags) {
1116         return H_PARAMETER;
1117     }
1118 
1119     if (lisn >= xive->nr_irqs) {
1120         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1121                       lisn);
1122         return H_P2;
1123     }
1124 
1125     eas = xive->eat[lisn];
1126     if (!xive_eas_is_valid(&eas)) {
1127         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1128                       lisn);
1129         return H_P2;
1130     }
1131 
1132     /* EAS_END_BLOCK is unused on sPAPR */
1133     end_idx = xive_get_field64(EAS_END_INDEX, eas.w);
1134 
1135     assert(end_idx < xive->nr_ends);
1136     end = &xive->endt[end_idx];
1137 
1138     nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end->w6);
1139     nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end->w6);
1140     args[0] = spapr_xive_nvt_to_target(nvt_blk, nvt_idx);
1141 
1142     if (xive_eas_is_masked(&eas)) {
1143         args[1] = 0xff;
1144     } else {
1145         args[1] = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
1146     }
1147 
1148     args[2] = xive_get_field64(EAS_END_DATA, eas.w);
1149 
1150     return H_SUCCESS;
1151 }
1152 
1153 /*
1154  * The H_INT_GET_QUEUE_INFO hcall() is used to get the logical real
1155  * address of the notification management page associated with the
1156  * specified target and priority.
1157  *
1158  * Parameters:
1159  * Input:
1160  * - R4: "flags"
1161  *         Bits 0-63 Reserved
1162  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1163  *       "ibm,ppc-interrupt-gserver#s"
1164  * - R6: "priority" is a valid priority not in
1165  *       "ibm,plat-res-int-priorities"
1166  *
1167  * Output:
1168  * - R4: Logical real address of notification page
1169  * - R5: Power of 2 page size of the notification page
1170  */
1171 static target_ulong h_int_get_queue_info(PowerPCCPU *cpu,
1172                                          SpaprMachineState *spapr,
1173                                          target_ulong opcode,
1174                                          target_ulong *args)
1175 {
1176     SpaprXive *xive = spapr->xive;
1177     XiveENDSource *end_xsrc = &xive->end_source;
1178     target_ulong flags = args[0];
1179     target_ulong target = args[1];
1180     target_ulong priority = args[2];
1181     XiveEND *end;
1182     uint8_t end_blk;
1183     uint32_t end_idx;
1184 
1185     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1186         return H_FUNCTION;
1187     }
1188 
1189     if (flags) {
1190         return H_PARAMETER;
1191     }
1192 
1193     /*
1194      * H_STATE should be returned if a H_INT_RESET is in progress.
1195      * This is not needed when running the emulation under QEMU
1196      */
1197 
1198     if (spapr_xive_priority_is_reserved(priority)) {
1199         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1200                       " is reserved\n", priority);
1201         return H_P3;
1202     }
1203 
1204     /*
1205      * Validate that "target" is part of the list of threads allocated
1206      * to the partition. For that, find the END corresponding to the
1207      * target.
1208      */
1209     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1210         return H_P2;
1211     }
1212 
1213     assert(end_idx < xive->nr_ends);
1214     end = &xive->endt[end_idx];
1215 
1216     args[0] = xive->end_base + (1ull << (end_xsrc->esb_shift + 1)) * end_idx;
1217     if (xive_end_is_enqueue(end)) {
1218         args[1] = xive_get_field32(END_W0_QSIZE, end->w0) + 12;
1219     } else {
1220         args[1] = 0;
1221     }
1222 
1223     return H_SUCCESS;
1224 }
1225 
1226 /*
1227  * The H_INT_SET_QUEUE_CONFIG hcall() is used to set or reset a EQ for
1228  * a given "target" and "priority".  It is also used to set the
1229  * notification config associated with the EQ.  An EQ size of 0 is
1230  * used to reset the EQ config for a given target and priority. If
1231  * resetting the EQ config, the END associated with the given "target"
1232  * and "priority" will be changed to disable queueing.
1233  *
1234  * Upon return from the hcall(), no additional interrupts will be
1235  * directed to the old EQ (if one was set). The old EQ (if one was
1236  * set) should be investigated for interrupts that occurred prior to
1237  * or during the hcall().
1238  *
1239  * Parameters:
1240  * Input:
1241  * - R4: "flags"
1242  *         Bits 0-62: Reserved
1243  *         Bit 63: Unconditional Notify (n) per the XIVE spec
1244  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1245  *       "ibm,ppc-interrupt-gserver#s"
1246  * - R6: "priority" is a valid priority not in
1247  *       "ibm,plat-res-int-priorities"
1248  * - R7: "eventQueue": The logical real address of the start of the EQ
1249  * - R8: "eventQueueSize": The power of 2 EQ size per "ibm,xive-eq-sizes"
1250  *
1251  * Output:
1252  * - None
1253  */
1254 
1255 #define SPAPR_XIVE_END_ALWAYS_NOTIFY PPC_BIT(63)
1256 
1257 static target_ulong h_int_set_queue_config(PowerPCCPU *cpu,
1258                                            SpaprMachineState *spapr,
1259                                            target_ulong opcode,
1260                                            target_ulong *args)
1261 {
1262     SpaprXive *xive = spapr->xive;
1263     target_ulong flags = args[0];
1264     target_ulong target = args[1];
1265     target_ulong priority = args[2];
1266     target_ulong qpage = args[3];
1267     target_ulong qsize = args[4];
1268     XiveEND end;
1269     uint8_t end_blk, nvt_blk;
1270     uint32_t end_idx, nvt_idx;
1271 
1272     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1273         return H_FUNCTION;
1274     }
1275 
1276     if (flags & ~SPAPR_XIVE_END_ALWAYS_NOTIFY) {
1277         return H_PARAMETER;
1278     }
1279 
1280     /*
1281      * H_STATE should be returned if a H_INT_RESET is in progress.
1282      * This is not needed when running the emulation under QEMU
1283      */
1284 
1285     if (spapr_xive_priority_is_reserved(priority)) {
1286         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1287                       " is reserved\n", priority);
1288         return H_P3;
1289     }
1290 
1291     /*
1292      * Validate that "target" is part of the list of threads allocated
1293      * to the partition. For that, find the END corresponding to the
1294      * target.
1295      */
1296 
1297     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1298         return H_P2;
1299     }
1300 
1301     assert(end_idx < xive->nr_ends);
1302     memcpy(&end, &xive->endt[end_idx], sizeof(XiveEND));
1303 
1304     switch (qsize) {
1305     case 12:
1306     case 16:
1307     case 21:
1308     case 24:
1309         if (!QEMU_IS_ALIGNED(qpage, 1ul << qsize)) {
1310             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: EQ @0x%" HWADDR_PRIx
1311                           " is not naturally aligned with %" HWADDR_PRIx "\n",
1312                           qpage, (hwaddr)1 << qsize);
1313             return H_P4;
1314         }
1315         end.w2 = cpu_to_be32((qpage >> 32) & 0x0fffffff);
1316         end.w3 = cpu_to_be32(qpage & 0xffffffff);
1317         end.w0 |= cpu_to_be32(END_W0_ENQUEUE);
1318         end.w0 = xive_set_field32(END_W0_QSIZE, end.w0, qsize - 12);
1319         break;
1320     case 0:
1321         /* reset queue and disable queueing */
1322         spapr_xive_end_reset(&end);
1323         goto out;
1324 
1325     default:
1326         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid EQ size %"PRIx64"\n",
1327                       qsize);
1328         return H_P5;
1329     }
1330 
1331     if (qsize) {
1332         hwaddr plen = 1 << qsize;
1333         void *eq;
1334 
1335         /*
1336          * Validate the guest EQ. We should also check that the queue
1337          * has been zeroed by the OS.
1338          */
1339         eq = address_space_map(CPU(cpu)->as, qpage, &plen, true,
1340                                MEMTXATTRS_UNSPECIFIED);
1341         if (plen != 1 << qsize) {
1342             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to map EQ @0x%"
1343                           HWADDR_PRIx "\n", qpage);
1344             return H_P4;
1345         }
1346         address_space_unmap(CPU(cpu)->as, eq, plen, true, plen);
1347     }
1348 
1349     /* "target" should have been validated above */
1350     if (spapr_xive_target_to_nvt(target, &nvt_blk, &nvt_idx)) {
1351         g_assert_not_reached();
1352     }
1353 
1354     /*
1355      * Ensure the priority and target are correctly set (they will not
1356      * be right after allocation)
1357      */
1358     end.w6 = xive_set_field32(END_W6_NVT_BLOCK, 0ul, nvt_blk) |
1359         xive_set_field32(END_W6_NVT_INDEX, 0ul, nvt_idx);
1360     end.w7 = xive_set_field32(END_W7_F0_PRIORITY, 0ul, priority);
1361 
1362     if (flags & SPAPR_XIVE_END_ALWAYS_NOTIFY) {
1363         end.w0 |= cpu_to_be32(END_W0_UCOND_NOTIFY);
1364     } else {
1365         end.w0 &= cpu_to_be32((uint32_t)~END_W0_UCOND_NOTIFY);
1366     }
1367 
1368     /*
1369      * The generation bit for the END starts at 1 and The END page
1370      * offset counter starts at 0.
1371      */
1372     end.w1 = cpu_to_be32(END_W1_GENERATION) |
1373         xive_set_field32(END_W1_PAGE_OFF, 0ul, 0ul);
1374     end.w0 |= cpu_to_be32(END_W0_VALID);
1375 
1376     /*
1377      * TODO: issue syncs required to ensure all in-flight interrupts
1378      * are complete on the old END
1379      */
1380 
1381 out:
1382     if (kvm_irqchip_in_kernel()) {
1383         Error *local_err = NULL;
1384 
1385         kvmppc_xive_set_queue_config(xive, end_blk, end_idx, &end, &local_err);
1386         if (local_err) {
1387             error_report_err(local_err);
1388             return H_HARDWARE;
1389         }
1390     }
1391 
1392     /* Update END */
1393     memcpy(&xive->endt[end_idx], &end, sizeof(XiveEND));
1394     return H_SUCCESS;
1395 }
1396 
1397 /*
1398  * The H_INT_GET_QUEUE_CONFIG hcall() is used to get a EQ for a given
1399  * target and priority.
1400  *
1401  * Parameters:
1402  * Input:
1403  * - R4: "flags"
1404  *         Bits 0-62: Reserved
1405  *         Bit 63: Debug: Return debug data
1406  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1407  *       "ibm,ppc-interrupt-gserver#s"
1408  * - R6: "priority" is a valid priority not in
1409  *       "ibm,plat-res-int-priorities"
1410  *
1411  * Output:
1412  * - R4: "flags":
1413  *       Bits 0-61: Reserved
1414  *       Bit 62: The value of Event Queue Generation Number (g) per
1415  *              the XIVE spec if "Debug" = 1
1416  *       Bit 63: The value of Unconditional Notify (n) per the XIVE spec
1417  * - R5: The logical real address of the start of the EQ
1418  * - R6: The power of 2 EQ size per "ibm,xive-eq-sizes"
1419  * - R7: The value of Event Queue Offset Counter per XIVE spec
1420  *       if "Debug" = 1, else 0
1421  *
1422  */
1423 
1424 #define SPAPR_XIVE_END_DEBUG     PPC_BIT(63)
1425 
1426 static target_ulong h_int_get_queue_config(PowerPCCPU *cpu,
1427                                            SpaprMachineState *spapr,
1428                                            target_ulong opcode,
1429                                            target_ulong *args)
1430 {
1431     SpaprXive *xive = spapr->xive;
1432     target_ulong flags = args[0];
1433     target_ulong target = args[1];
1434     target_ulong priority = args[2];
1435     XiveEND *end;
1436     uint8_t end_blk;
1437     uint32_t end_idx;
1438 
1439     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1440         return H_FUNCTION;
1441     }
1442 
1443     if (flags & ~SPAPR_XIVE_END_DEBUG) {
1444         return H_PARAMETER;
1445     }
1446 
1447     /*
1448      * H_STATE should be returned if a H_INT_RESET is in progress.
1449      * This is not needed when running the emulation under QEMU
1450      */
1451 
1452     if (spapr_xive_priority_is_reserved(priority)) {
1453         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1454                       " is reserved\n", priority);
1455         return H_P3;
1456     }
1457 
1458     /*
1459      * Validate that "target" is part of the list of threads allocated
1460      * to the partition. For that, find the END corresponding to the
1461      * target.
1462      */
1463     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1464         return H_P2;
1465     }
1466 
1467     assert(end_idx < xive->nr_ends);
1468     end = &xive->endt[end_idx];
1469 
1470     args[0] = 0;
1471     if (xive_end_is_notify(end)) {
1472         args[0] |= SPAPR_XIVE_END_ALWAYS_NOTIFY;
1473     }
1474 
1475     if (xive_end_is_enqueue(end)) {
1476         args[1] = xive_end_qaddr(end);
1477         args[2] = xive_get_field32(END_W0_QSIZE, end->w0) + 12;
1478     } else {
1479         args[1] = 0;
1480         args[2] = 0;
1481     }
1482 
1483     if (kvm_irqchip_in_kernel()) {
1484         Error *local_err = NULL;
1485 
1486         kvmppc_xive_get_queue_config(xive, end_blk, end_idx, end, &local_err);
1487         if (local_err) {
1488             error_report_err(local_err);
1489             return H_HARDWARE;
1490         }
1491     }
1492 
1493     /* TODO: do we need any locking on the END ? */
1494     if (flags & SPAPR_XIVE_END_DEBUG) {
1495         /* Load the event queue generation number into the return flags */
1496         args[0] |= (uint64_t)xive_get_field32(END_W1_GENERATION, end->w1) << 62;
1497 
1498         /* Load R7 with the event queue offset counter */
1499         args[3] = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1500     } else {
1501         args[3] = 0;
1502     }
1503 
1504     return H_SUCCESS;
1505 }
1506 
1507 /*
1508  * The H_INT_SET_OS_REPORTING_LINE hcall() is used to set the
1509  * reporting cache line pair for the calling thread.  The reporting
1510  * cache lines will contain the OS interrupt context when the OS
1511  * issues a CI store byte to @TIMA+0xC10 to acknowledge the OS
1512  * interrupt. The reporting cache lines can be reset by inputting -1
1513  * in "reportingLine".  Issuing the CI store byte without reporting
1514  * cache lines registered will result in the data not being accessible
1515  * to the OS.
1516  *
1517  * Parameters:
1518  * Input:
1519  * - R4: "flags"
1520  *         Bits 0-63: Reserved
1521  * - R5: "reportingLine": The logical real address of the reporting cache
1522  *       line pair
1523  *
1524  * Output:
1525  * - None
1526  */
1527 static target_ulong h_int_set_os_reporting_line(PowerPCCPU *cpu,
1528                                                 SpaprMachineState *spapr,
1529                                                 target_ulong opcode,
1530                                                 target_ulong *args)
1531 {
1532     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1533         return H_FUNCTION;
1534     }
1535 
1536     /*
1537      * H_STATE should be returned if a H_INT_RESET is in progress.
1538      * This is not needed when running the emulation under QEMU
1539      */
1540 
1541     /* TODO: H_INT_SET_OS_REPORTING_LINE */
1542     return H_FUNCTION;
1543 }
1544 
1545 /*
1546  * The H_INT_GET_OS_REPORTING_LINE hcall() is used to get the logical
1547  * real address of the reporting cache line pair set for the input
1548  * "target".  If no reporting cache line pair has been set, -1 is
1549  * returned.
1550  *
1551  * Parameters:
1552  * Input:
1553  * - R4: "flags"
1554  *         Bits 0-63: Reserved
1555  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1556  *       "ibm,ppc-interrupt-gserver#s"
1557  * - R6: "reportingLine": The logical real address of the reporting
1558  *        cache line pair
1559  *
1560  * Output:
1561  * - R4: The logical real address of the reporting line if set, else -1
1562  */
1563 static target_ulong h_int_get_os_reporting_line(PowerPCCPU *cpu,
1564                                                 SpaprMachineState *spapr,
1565                                                 target_ulong opcode,
1566                                                 target_ulong *args)
1567 {
1568     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1569         return H_FUNCTION;
1570     }
1571 
1572     /*
1573      * H_STATE should be returned if a H_INT_RESET is in progress.
1574      * This is not needed when running the emulation under QEMU
1575      */
1576 
1577     /* TODO: H_INT_GET_OS_REPORTING_LINE */
1578     return H_FUNCTION;
1579 }
1580 
1581 /*
1582  * The H_INT_ESB hcall() is used to issue a load or store to the ESB
1583  * page for the input "lisn".  This hcall is only supported for LISNs
1584  * that have the ESB hcall flag set to 1 when returned from hcall()
1585  * H_INT_GET_SOURCE_INFO.
1586  *
1587  * Parameters:
1588  * Input:
1589  * - R4: "flags"
1590  *         Bits 0-62: Reserved
1591  *         bit 63: Store: Store=1, store operation, else load operation
1592  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1593  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1594  *       ibm,query-interrupt-source-number RTAS call, or as
1595  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1596  * - R6: "esbOffset" is the offset into the ESB page for the load or
1597  *       store operation
1598  * - R7: "storeData" is the data to write for a store operation
1599  *
1600  * Output:
1601  * - R4: The value of the load if load operation, else -1
1602  */
1603 
1604 #define SPAPR_XIVE_ESB_STORE PPC_BIT(63)
1605 
1606 static target_ulong h_int_esb(PowerPCCPU *cpu,
1607                               SpaprMachineState *spapr,
1608                               target_ulong opcode,
1609                               target_ulong *args)
1610 {
1611     SpaprXive *xive = spapr->xive;
1612     XiveEAS eas;
1613     target_ulong flags  = args[0];
1614     target_ulong lisn   = args[1];
1615     target_ulong offset = args[2];
1616     target_ulong data   = args[3];
1617     hwaddr mmio_addr;
1618     XiveSource *xsrc = &xive->source;
1619 
1620     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1621         return H_FUNCTION;
1622     }
1623 
1624     if (flags & ~SPAPR_XIVE_ESB_STORE) {
1625         return H_PARAMETER;
1626     }
1627 
1628     if (lisn >= xive->nr_irqs) {
1629         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1630                       lisn);
1631         return H_P2;
1632     }
1633 
1634     eas = xive->eat[lisn];
1635     if (!xive_eas_is_valid(&eas)) {
1636         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1637                       lisn);
1638         return H_P2;
1639     }
1640 
1641     if (offset > (1ull << xsrc->esb_shift)) {
1642         return H_P3;
1643     }
1644 
1645     if (kvm_irqchip_in_kernel()) {
1646         args[0] = kvmppc_xive_esb_rw(xsrc, lisn, offset, data,
1647                                      flags & SPAPR_XIVE_ESB_STORE);
1648     } else {
1649         mmio_addr = xive->vc_base + xive_source_esb_mgmt(xsrc, lisn) + offset;
1650 
1651         if (dma_memory_rw(&address_space_memory, mmio_addr, &data, 8,
1652                           (flags & SPAPR_XIVE_ESB_STORE))) {
1653             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to access ESB @0x%"
1654                           HWADDR_PRIx "\n", mmio_addr);
1655             return H_HARDWARE;
1656         }
1657         args[0] = (flags & SPAPR_XIVE_ESB_STORE) ? -1 : data;
1658     }
1659     return H_SUCCESS;
1660 }
1661 
1662 /*
1663  * The H_INT_SYNC hcall() is used to issue hardware syncs that will
1664  * ensure any in flight events for the input lisn are in the event
1665  * queue.
1666  *
1667  * Parameters:
1668  * Input:
1669  * - R4: "flags"
1670  *         Bits 0-63: Reserved
1671  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1672  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1673  *       ibm,query-interrupt-source-number RTAS call, or as
1674  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1675  *
1676  * Output:
1677  * - None
1678  */
1679 static target_ulong h_int_sync(PowerPCCPU *cpu,
1680                                SpaprMachineState *spapr,
1681                                target_ulong opcode,
1682                                target_ulong *args)
1683 {
1684     SpaprXive *xive = spapr->xive;
1685     XiveEAS eas;
1686     target_ulong flags = args[0];
1687     target_ulong lisn = args[1];
1688 
1689     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1690         return H_FUNCTION;
1691     }
1692 
1693     if (flags) {
1694         return H_PARAMETER;
1695     }
1696 
1697     if (lisn >= xive->nr_irqs) {
1698         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1699                       lisn);
1700         return H_P2;
1701     }
1702 
1703     eas = xive->eat[lisn];
1704     if (!xive_eas_is_valid(&eas)) {
1705         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1706                       lisn);
1707         return H_P2;
1708     }
1709 
1710     /*
1711      * H_STATE should be returned if a H_INT_RESET is in progress.
1712      * This is not needed when running the emulation under QEMU
1713      */
1714 
1715     /*
1716      * This is not real hardware. Nothing to be done unless when
1717      * under KVM
1718      */
1719 
1720     if (kvm_irqchip_in_kernel()) {
1721         Error *local_err = NULL;
1722 
1723         kvmppc_xive_sync_source(xive, lisn, &local_err);
1724         if (local_err) {
1725             error_report_err(local_err);
1726             return H_HARDWARE;
1727         }
1728     }
1729     return H_SUCCESS;
1730 }
1731 
1732 /*
1733  * The H_INT_RESET hcall() is used to reset all of the partition's
1734  * interrupt exploitation structures to their initial state.  This
1735  * means losing all previously set interrupt state set via
1736  * H_INT_SET_SOURCE_CONFIG and H_INT_SET_QUEUE_CONFIG.
1737  *
1738  * Parameters:
1739  * Input:
1740  * - R4: "flags"
1741  *         Bits 0-63: Reserved
1742  *
1743  * Output:
1744  * - None
1745  */
1746 static target_ulong h_int_reset(PowerPCCPU *cpu,
1747                                 SpaprMachineState *spapr,
1748                                 target_ulong opcode,
1749                                 target_ulong *args)
1750 {
1751     SpaprXive *xive = spapr->xive;
1752     target_ulong flags   = args[0];
1753 
1754     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1755         return H_FUNCTION;
1756     }
1757 
1758     if (flags) {
1759         return H_PARAMETER;
1760     }
1761 
1762     device_legacy_reset(DEVICE(xive));
1763 
1764     if (kvm_irqchip_in_kernel()) {
1765         Error *local_err = NULL;
1766 
1767         kvmppc_xive_reset(xive, &local_err);
1768         if (local_err) {
1769             error_report_err(local_err);
1770             return H_HARDWARE;
1771         }
1772     }
1773     return H_SUCCESS;
1774 }
1775 
1776 void spapr_xive_hcall_init(SpaprMachineState *spapr)
1777 {
1778     spapr_register_hypercall(H_INT_GET_SOURCE_INFO, h_int_get_source_info);
1779     spapr_register_hypercall(H_INT_SET_SOURCE_CONFIG, h_int_set_source_config);
1780     spapr_register_hypercall(H_INT_GET_SOURCE_CONFIG, h_int_get_source_config);
1781     spapr_register_hypercall(H_INT_GET_QUEUE_INFO, h_int_get_queue_info);
1782     spapr_register_hypercall(H_INT_SET_QUEUE_CONFIG, h_int_set_queue_config);
1783     spapr_register_hypercall(H_INT_GET_QUEUE_CONFIG, h_int_get_queue_config);
1784     spapr_register_hypercall(H_INT_SET_OS_REPORTING_LINE,
1785                              h_int_set_os_reporting_line);
1786     spapr_register_hypercall(H_INT_GET_OS_REPORTING_LINE,
1787                              h_int_get_os_reporting_line);
1788     spapr_register_hypercall(H_INT_ESB, h_int_esb);
1789     spapr_register_hypercall(H_INT_SYNC, h_int_sync);
1790     spapr_register_hypercall(H_INT_RESET, h_int_reset);
1791 }
1792