xref: /openbmc/qemu/hw/intc/spapr_xive.c (revision f6476697)
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 #include "trace.h"
28 
29 /*
30  * XIVE Virtualization Controller BAR and Thread Managment BAR that we
31  * use for the ESB pages and the TIMA pages
32  */
33 #define SPAPR_XIVE_VC_BASE   0x0006010000000000ull
34 #define SPAPR_XIVE_TM_BASE   0x0006030203180000ull
35 
36 /*
37  * The allocation of VP blocks is a complex operation in OPAL and the
38  * VP identifiers have a relation with the number of HW chips, the
39  * size of the VP blocks, VP grouping, etc. The QEMU sPAPR XIVE
40  * controller model does not have the same constraints and can use a
41  * simple mapping scheme of the CPU vcpu_id
42  *
43  * These identifiers are never returned to the OS.
44  */
45 
46 #define SPAPR_XIVE_NVT_BASE 0x400
47 
48 /*
49  * sPAPR NVT and END indexing helpers
50  */
51 static uint32_t spapr_xive_nvt_to_target(uint8_t nvt_blk, uint32_t nvt_idx)
52 {
53     return nvt_idx - SPAPR_XIVE_NVT_BASE;
54 }
55 
56 static void spapr_xive_cpu_to_nvt(PowerPCCPU *cpu,
57                                   uint8_t *out_nvt_blk, uint32_t *out_nvt_idx)
58 {
59     assert(cpu);
60 
61     if (out_nvt_blk) {
62         *out_nvt_blk = SPAPR_XIVE_BLOCK_ID;
63     }
64 
65     if (out_nvt_blk) {
66         *out_nvt_idx = SPAPR_XIVE_NVT_BASE + cpu->vcpu_id;
67     }
68 }
69 
70 static int spapr_xive_target_to_nvt(uint32_t target,
71                                     uint8_t *out_nvt_blk, uint32_t *out_nvt_idx)
72 {
73     PowerPCCPU *cpu = spapr_find_cpu(target);
74 
75     if (!cpu) {
76         return -1;
77     }
78 
79     spapr_xive_cpu_to_nvt(cpu, out_nvt_blk, out_nvt_idx);
80     return 0;
81 }
82 
83 /*
84  * sPAPR END indexing uses a simple mapping of the CPU vcpu_id, 8
85  * priorities per CPU
86  */
87 int spapr_xive_end_to_target(uint8_t end_blk, uint32_t end_idx,
88                              uint32_t *out_server, uint8_t *out_prio)
89 {
90 
91     assert(end_blk == SPAPR_XIVE_BLOCK_ID);
92 
93     if (out_server) {
94         *out_server = end_idx >> 3;
95     }
96 
97     if (out_prio) {
98         *out_prio = end_idx & 0x7;
99     }
100     return 0;
101 }
102 
103 static void spapr_xive_cpu_to_end(PowerPCCPU *cpu, uint8_t prio,
104                                   uint8_t *out_end_blk, uint32_t *out_end_idx)
105 {
106     assert(cpu);
107 
108     if (out_end_blk) {
109         *out_end_blk = SPAPR_XIVE_BLOCK_ID;
110     }
111 
112     if (out_end_idx) {
113         *out_end_idx = (cpu->vcpu_id << 3) + prio;
114     }
115 }
116 
117 static int spapr_xive_target_to_end(uint32_t target, uint8_t prio,
118                                     uint8_t *out_end_blk, uint32_t *out_end_idx)
119 {
120     PowerPCCPU *cpu = spapr_find_cpu(target);
121 
122     if (!cpu) {
123         return -1;
124     }
125 
126     spapr_xive_cpu_to_end(cpu, prio, out_end_blk, out_end_idx);
127     return 0;
128 }
129 
130 /*
131  * On sPAPR machines, use a simplified output for the XIVE END
132  * structure dumping only the information related to the OS EQ.
133  */
134 static void spapr_xive_end_pic_print_info(SpaprXive *xive, XiveEND *end,
135                                           Monitor *mon)
136 {
137     uint64_t qaddr_base = xive_end_qaddr(end);
138     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
139     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
140     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
141     uint32_t qentries = 1 << (qsize + 10);
142     uint32_t nvt = xive_get_field32(END_W6_NVT_INDEX, end->w6);
143     uint8_t priority = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
144 
145     monitor_printf(mon, "%3d/%d % 6d/%5d @%"PRIx64" ^%d",
146                    spapr_xive_nvt_to_target(0, nvt),
147                    priority, qindex, qentries, qaddr_base, qgen);
148 
149     xive_end_queue_pic_print_info(end, 6, mon);
150 }
151 
152 /*
153  * kvm_irqchip_in_kernel() will cause the compiler to turn this
154  * info a nop if CONFIG_KVM isn't defined.
155  */
156 #define spapr_xive_in_kernel(xive) \
157     (kvm_irqchip_in_kernel() && (xive)->fd != -1)
158 
159 static void spapr_xive_pic_print_info(SpaprXive *xive, Monitor *mon)
160 {
161     XiveSource *xsrc = &xive->source;
162     int i;
163 
164     if (spapr_xive_in_kernel(xive)) {
165         Error *local_err = NULL;
166 
167         kvmppc_xive_synchronize_state(xive, &local_err);
168         if (local_err) {
169             error_report_err(local_err);
170             return;
171         }
172     }
173 
174     monitor_printf(mon, "  LISN         PQ    EISN     CPU/PRIO EQ\n");
175 
176     for (i = 0; i < xive->nr_irqs; i++) {
177         uint8_t pq = xive_source_esb_get(xsrc, i);
178         XiveEAS *eas = &xive->eat[i];
179 
180         if (!xive_eas_is_valid(eas)) {
181             continue;
182         }
183 
184         monitor_printf(mon, "  %08x %s %c%c%c %s %08x ", i,
185                        xive_source_irq_is_lsi(xsrc, i) ? "LSI" : "MSI",
186                        pq & XIVE_ESB_VAL_P ? 'P' : '-',
187                        pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
188                        xsrc->status[i] & XIVE_STATUS_ASSERTED ? 'A' : ' ',
189                        xive_eas_is_masked(eas) ? "M" : " ",
190                        (int) xive_get_field64(EAS_END_DATA, eas->w));
191 
192         if (!xive_eas_is_masked(eas)) {
193             uint32_t end_idx = xive_get_field64(EAS_END_INDEX, eas->w);
194             XiveEND *end;
195 
196             assert(end_idx < xive->nr_ends);
197             end = &xive->endt[end_idx];
198 
199             if (xive_end_is_valid(end)) {
200                 spapr_xive_end_pic_print_info(xive, end, mon);
201             }
202         }
203         monitor_printf(mon, "\n");
204     }
205 }
206 
207 void spapr_xive_mmio_set_enabled(SpaprXive *xive, bool enable)
208 {
209     memory_region_set_enabled(&xive->source.esb_mmio, enable);
210     memory_region_set_enabled(&xive->tm_mmio, enable);
211 
212     /* Disable the END ESBs until a guest OS makes use of them */
213     memory_region_set_enabled(&xive->end_source.esb_mmio, false);
214 }
215 
216 static void spapr_xive_tm_write(void *opaque, hwaddr offset,
217                           uint64_t value, unsigned size)
218 {
219     XiveTCTX *tctx = spapr_cpu_state(POWERPC_CPU(current_cpu))->tctx;
220 
221     xive_tctx_tm_write(XIVE_PRESENTER(opaque), tctx, offset, value, size);
222 }
223 
224 static uint64_t spapr_xive_tm_read(void *opaque, hwaddr offset, unsigned size)
225 {
226     XiveTCTX *tctx = spapr_cpu_state(POWERPC_CPU(current_cpu))->tctx;
227 
228     return xive_tctx_tm_read(XIVE_PRESENTER(opaque), tctx, offset, size);
229 }
230 
231 const MemoryRegionOps spapr_xive_tm_ops = {
232     .read = spapr_xive_tm_read,
233     .write = spapr_xive_tm_write,
234     .endianness = DEVICE_BIG_ENDIAN,
235     .valid = {
236         .min_access_size = 1,
237         .max_access_size = 8,
238     },
239     .impl = {
240         .min_access_size = 1,
241         .max_access_size = 8,
242     },
243 };
244 
245 static void spapr_xive_end_reset(XiveEND *end)
246 {
247     memset(end, 0, sizeof(*end));
248 
249     /* switch off the escalation and notification ESBs */
250     end->w1 = cpu_to_be32(END_W1_ESe_Q | END_W1_ESn_Q);
251 }
252 
253 static void spapr_xive_reset(void *dev)
254 {
255     SpaprXive *xive = SPAPR_XIVE(dev);
256     int i;
257 
258     /*
259      * The XiveSource has its own reset handler, which mask off all
260      * IRQs (!P|Q)
261      */
262 
263     /* Mask all valid EASs in the IRQ number space. */
264     for (i = 0; i < xive->nr_irqs; i++) {
265         XiveEAS *eas = &xive->eat[i];
266         if (xive_eas_is_valid(eas)) {
267             eas->w = cpu_to_be64(EAS_VALID | EAS_MASKED);
268         } else {
269             eas->w = 0;
270         }
271     }
272 
273     /* Clear all ENDs */
274     for (i = 0; i < xive->nr_ends; i++) {
275         spapr_xive_end_reset(&xive->endt[i]);
276     }
277 }
278 
279 static void spapr_xive_instance_init(Object *obj)
280 {
281     SpaprXive *xive = SPAPR_XIVE(obj);
282 
283     object_initialize_child(obj, "source", &xive->source, TYPE_XIVE_SOURCE);
284 
285     object_initialize_child(obj, "end_source", &xive->end_source,
286                             TYPE_XIVE_END_SOURCE);
287 
288     /* Not connected to the KVM XIVE device */
289     xive->fd = -1;
290 }
291 
292 static void spapr_xive_realize(DeviceState *dev, Error **errp)
293 {
294     SpaprXive *xive = SPAPR_XIVE(dev);
295     SpaprXiveClass *sxc = SPAPR_XIVE_GET_CLASS(xive);
296     XiveSource *xsrc = &xive->source;
297     XiveENDSource *end_xsrc = &xive->end_source;
298     Error *local_err = NULL;
299 
300     /* Set by spapr_irq_init() */
301     g_assert(xive->nr_irqs);
302     g_assert(xive->nr_ends);
303 
304     sxc->parent_realize(dev, &local_err);
305     if (local_err) {
306         error_propagate(errp, local_err);
307         return;
308     }
309 
310     /*
311      * Initialize the internal sources, for IPIs and virtual devices.
312      */
313     object_property_set_int(OBJECT(xsrc), "nr-irqs", xive->nr_irqs,
314                             &error_fatal);
315     object_property_set_link(OBJECT(xsrc), "xive", OBJECT(xive), &error_abort);
316     if (!qdev_realize(DEVICE(xsrc), NULL, errp)) {
317         return;
318     }
319     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &xsrc->esb_mmio);
320 
321     /*
322      * Initialize the END ESB source
323      */
324     object_property_set_int(OBJECT(end_xsrc), "nr-ends", xive->nr_irqs,
325                             &error_fatal);
326     object_property_set_link(OBJECT(end_xsrc), "xive", OBJECT(xive),
327                              &error_abort);
328     if (!qdev_realize(DEVICE(end_xsrc), NULL, errp)) {
329         return;
330     }
331     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &end_xsrc->esb_mmio);
332 
333     /* Set the mapping address of the END ESB pages after the source ESBs */
334     xive->end_base = xive->vc_base + xive_source_esb_len(xsrc);
335 
336     /*
337      * Allocate the routing tables
338      */
339     xive->eat = g_new0(XiveEAS, xive->nr_irqs);
340     xive->endt = g_new0(XiveEND, xive->nr_ends);
341 
342     xive->nodename = g_strdup_printf("interrupt-controller@%" PRIx64,
343                            xive->tm_base + XIVE_TM_USER_PAGE * (1 << TM_SHIFT));
344 
345     qemu_register_reset(spapr_xive_reset, dev);
346 
347     /* TIMA initialization */
348     memory_region_init_io(&xive->tm_mmio, OBJECT(xive), &spapr_xive_tm_ops,
349                           xive, "xive.tima", 4ull << TM_SHIFT);
350     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &xive->tm_mmio);
351 
352     /*
353      * Map all regions. These will be enabled or disabled at reset and
354      * can also be overridden by KVM memory regions if active
355      */
356     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 0, xive->vc_base);
357     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 1, xive->end_base);
358     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 2, xive->tm_base);
359 }
360 
361 static int spapr_xive_get_eas(XiveRouter *xrtr, uint8_t eas_blk,
362                               uint32_t eas_idx, XiveEAS *eas)
363 {
364     SpaprXive *xive = SPAPR_XIVE(xrtr);
365 
366     if (eas_idx >= xive->nr_irqs) {
367         return -1;
368     }
369 
370     *eas = xive->eat[eas_idx];
371     return 0;
372 }
373 
374 static int spapr_xive_get_end(XiveRouter *xrtr,
375                               uint8_t end_blk, uint32_t end_idx, XiveEND *end)
376 {
377     SpaprXive *xive = SPAPR_XIVE(xrtr);
378 
379     if (end_idx >= xive->nr_ends) {
380         return -1;
381     }
382 
383     memcpy(end, &xive->endt[end_idx], sizeof(XiveEND));
384     return 0;
385 }
386 
387 static int spapr_xive_write_end(XiveRouter *xrtr, uint8_t end_blk,
388                                 uint32_t end_idx, XiveEND *end,
389                                 uint8_t word_number)
390 {
391     SpaprXive *xive = SPAPR_XIVE(xrtr);
392 
393     if (end_idx >= xive->nr_ends) {
394         return -1;
395     }
396 
397     memcpy(&xive->endt[end_idx], end, sizeof(XiveEND));
398     return 0;
399 }
400 
401 static int spapr_xive_get_nvt(XiveRouter *xrtr,
402                               uint8_t nvt_blk, uint32_t nvt_idx, XiveNVT *nvt)
403 {
404     uint32_t vcpu_id = spapr_xive_nvt_to_target(nvt_blk, nvt_idx);
405     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
406 
407     if (!cpu) {
408         /* TODO: should we assert() if we can find a NVT ? */
409         return -1;
410     }
411 
412     /*
413      * sPAPR does not maintain a NVT table. Return that the NVT is
414      * valid if we have found a matching CPU
415      */
416     nvt->w0 = cpu_to_be32(NVT_W0_VALID);
417     return 0;
418 }
419 
420 static int spapr_xive_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk,
421                                 uint32_t nvt_idx, XiveNVT *nvt,
422                                 uint8_t word_number)
423 {
424     /*
425      * We don't need to write back to the NVTs because the sPAPR
426      * machine should never hit a non-scheduled NVT. It should never
427      * get called.
428      */
429     g_assert_not_reached();
430 }
431 
432 static int spapr_xive_match_nvt(XivePresenter *xptr, uint8_t format,
433                                 uint8_t nvt_blk, uint32_t nvt_idx,
434                                 bool cam_ignore, uint8_t priority,
435                                 uint32_t logic_serv, XiveTCTXMatch *match)
436 {
437     CPUState *cs;
438     int count = 0;
439 
440     CPU_FOREACH(cs) {
441         PowerPCCPU *cpu = POWERPC_CPU(cs);
442         XiveTCTX *tctx = spapr_cpu_state(cpu)->tctx;
443         int ring;
444 
445         /*
446          * Skip partially initialized vCPUs. This can happen when
447          * vCPUs are hotplugged.
448          */
449         if (!tctx) {
450             continue;
451         }
452 
453         /*
454          * Check the thread context CAM lines and record matches.
455          */
456         ring = xive_presenter_tctx_match(xptr, tctx, format, nvt_blk, nvt_idx,
457                                          cam_ignore, logic_serv);
458         /*
459          * Save the matching thread interrupt context and follow on to
460          * check for duplicates which are invalid.
461          */
462         if (ring != -1) {
463             if (match->tctx) {
464                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: already found a thread "
465                               "context NVT %x/%x\n", nvt_blk, nvt_idx);
466                 return -1;
467             }
468 
469             match->ring = ring;
470             match->tctx = tctx;
471             count++;
472         }
473     }
474 
475     return count;
476 }
477 
478 static uint8_t spapr_xive_get_block_id(XiveRouter *xrtr)
479 {
480     return SPAPR_XIVE_BLOCK_ID;
481 }
482 
483 static const VMStateDescription vmstate_spapr_xive_end = {
484     .name = TYPE_SPAPR_XIVE "/end",
485     .version_id = 1,
486     .minimum_version_id = 1,
487     .fields = (VMStateField []) {
488         VMSTATE_UINT32(w0, XiveEND),
489         VMSTATE_UINT32(w1, XiveEND),
490         VMSTATE_UINT32(w2, XiveEND),
491         VMSTATE_UINT32(w3, XiveEND),
492         VMSTATE_UINT32(w4, XiveEND),
493         VMSTATE_UINT32(w5, XiveEND),
494         VMSTATE_UINT32(w6, XiveEND),
495         VMSTATE_UINT32(w7, XiveEND),
496         VMSTATE_END_OF_LIST()
497     },
498 };
499 
500 static const VMStateDescription vmstate_spapr_xive_eas = {
501     .name = TYPE_SPAPR_XIVE "/eas",
502     .version_id = 1,
503     .minimum_version_id = 1,
504     .fields = (VMStateField []) {
505         VMSTATE_UINT64(w, XiveEAS),
506         VMSTATE_END_OF_LIST()
507     },
508 };
509 
510 static int vmstate_spapr_xive_pre_save(void *opaque)
511 {
512     SpaprXive *xive = SPAPR_XIVE(opaque);
513 
514     if (spapr_xive_in_kernel(xive)) {
515         return kvmppc_xive_pre_save(xive);
516     }
517 
518     return 0;
519 }
520 
521 /*
522  * Called by the sPAPR IRQ backend 'post_load' method at the machine
523  * level.
524  */
525 static int spapr_xive_post_load(SpaprInterruptController *intc, int version_id)
526 {
527     SpaprXive *xive = SPAPR_XIVE(intc);
528 
529     if (spapr_xive_in_kernel(xive)) {
530         return kvmppc_xive_post_load(xive, version_id);
531     }
532 
533     return 0;
534 }
535 
536 static const VMStateDescription vmstate_spapr_xive = {
537     .name = TYPE_SPAPR_XIVE,
538     .version_id = 1,
539     .minimum_version_id = 1,
540     .pre_save = vmstate_spapr_xive_pre_save,
541     .post_load = NULL, /* handled at the machine level */
542     .fields = (VMStateField[]) {
543         VMSTATE_UINT32_EQUAL(nr_irqs, SpaprXive, NULL),
544         VMSTATE_STRUCT_VARRAY_POINTER_UINT32(eat, SpaprXive, nr_irqs,
545                                      vmstate_spapr_xive_eas, XiveEAS),
546         VMSTATE_STRUCT_VARRAY_POINTER_UINT32(endt, SpaprXive, nr_ends,
547                                              vmstate_spapr_xive_end, XiveEND),
548         VMSTATE_END_OF_LIST()
549     },
550 };
551 
552 static int spapr_xive_claim_irq(SpaprInterruptController *intc, int lisn,
553                                 bool lsi, Error **errp)
554 {
555     SpaprXive *xive = SPAPR_XIVE(intc);
556     XiveSource *xsrc = &xive->source;
557 
558     assert(lisn < xive->nr_irqs);
559 
560     trace_spapr_xive_claim_irq(lisn, lsi);
561 
562     if (xive_eas_is_valid(&xive->eat[lisn])) {
563         error_setg(errp, "IRQ %d is not free", lisn);
564         return -EBUSY;
565     }
566 
567     /*
568      * Set default values when allocating an IRQ number
569      */
570     xive->eat[lisn].w |= cpu_to_be64(EAS_VALID | EAS_MASKED);
571     if (lsi) {
572         xive_source_irq_set_lsi(xsrc, lisn);
573     }
574 
575     if (spapr_xive_in_kernel(xive)) {
576         return kvmppc_xive_source_reset_one(xsrc, lisn, errp);
577     }
578 
579     return 0;
580 }
581 
582 static void spapr_xive_free_irq(SpaprInterruptController *intc, int lisn)
583 {
584     SpaprXive *xive = SPAPR_XIVE(intc);
585     assert(lisn < xive->nr_irqs);
586 
587     trace_spapr_xive_free_irq(lisn);
588 
589     xive->eat[lisn].w &= cpu_to_be64(~EAS_VALID);
590 }
591 
592 static Property spapr_xive_properties[] = {
593     DEFINE_PROP_UINT32("nr-irqs", SpaprXive, nr_irqs, 0),
594     DEFINE_PROP_UINT32("nr-ends", SpaprXive, nr_ends, 0),
595     DEFINE_PROP_UINT64("vc-base", SpaprXive, vc_base, SPAPR_XIVE_VC_BASE),
596     DEFINE_PROP_UINT64("tm-base", SpaprXive, tm_base, SPAPR_XIVE_TM_BASE),
597     DEFINE_PROP_UINT8("hv-prio", SpaprXive, hv_prio, 7),
598     DEFINE_PROP_END_OF_LIST(),
599 };
600 
601 static int spapr_xive_cpu_intc_create(SpaprInterruptController *intc,
602                                       PowerPCCPU *cpu, Error **errp)
603 {
604     SpaprXive *xive = SPAPR_XIVE(intc);
605     Object *obj;
606     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
607 
608     obj = xive_tctx_create(OBJECT(cpu), XIVE_PRESENTER(xive), errp);
609     if (!obj) {
610         return -1;
611     }
612 
613     spapr_cpu->tctx = XIVE_TCTX(obj);
614     return 0;
615 }
616 
617 static void xive_tctx_set_os_cam(XiveTCTX *tctx, uint32_t os_cam)
618 {
619     uint32_t qw1w2 = cpu_to_be32(TM_QW1W2_VO | os_cam);
620     memcpy(&tctx->regs[TM_QW1_OS + TM_WORD2], &qw1w2, 4);
621 }
622 
623 static void spapr_xive_cpu_intc_reset(SpaprInterruptController *intc,
624                                      PowerPCCPU *cpu)
625 {
626     XiveTCTX *tctx = spapr_cpu_state(cpu)->tctx;
627     uint8_t  nvt_blk;
628     uint32_t nvt_idx;
629 
630     xive_tctx_reset(tctx);
631 
632     /*
633      * When a Virtual Processor is scheduled to run on a HW thread,
634      * the hypervisor pushes its identifier in the OS CAM line.
635      * Emulate the same behavior under QEMU.
636      */
637     spapr_xive_cpu_to_nvt(cpu, &nvt_blk, &nvt_idx);
638 
639     xive_tctx_set_os_cam(tctx, xive_nvt_cam_line(nvt_blk, nvt_idx));
640 }
641 
642 static void spapr_xive_cpu_intc_destroy(SpaprInterruptController *intc,
643                                         PowerPCCPU *cpu)
644 {
645     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
646 
647     xive_tctx_destroy(spapr_cpu->tctx);
648     spapr_cpu->tctx = NULL;
649 }
650 
651 static void spapr_xive_set_irq(SpaprInterruptController *intc, int irq, int val)
652 {
653     SpaprXive *xive = SPAPR_XIVE(intc);
654 
655     trace_spapr_xive_set_irq(irq, val);
656 
657     if (spapr_xive_in_kernel(xive)) {
658         kvmppc_xive_source_set_irq(&xive->source, irq, val);
659     } else {
660         xive_source_set_irq(&xive->source, irq, val);
661     }
662 }
663 
664 static void spapr_xive_print_info(SpaprInterruptController *intc, Monitor *mon)
665 {
666     SpaprXive *xive = SPAPR_XIVE(intc);
667     CPUState *cs;
668 
669     CPU_FOREACH(cs) {
670         PowerPCCPU *cpu = POWERPC_CPU(cs);
671 
672         xive_tctx_pic_print_info(spapr_cpu_state(cpu)->tctx, mon);
673     }
674 
675     spapr_xive_pic_print_info(xive, mon);
676 }
677 
678 static void spapr_xive_dt(SpaprInterruptController *intc, uint32_t nr_servers,
679                           void *fdt, uint32_t phandle)
680 {
681     SpaprXive *xive = SPAPR_XIVE(intc);
682     int node;
683     uint64_t timas[2 * 2];
684     /* Interrupt number ranges for the IPIs */
685     uint32_t lisn_ranges[] = {
686         cpu_to_be32(SPAPR_IRQ_IPI),
687         cpu_to_be32(SPAPR_IRQ_IPI + nr_servers),
688     };
689     /*
690      * EQ size - the sizes of pages supported by the system 4K, 64K,
691      * 2M, 16M. We only advertise 64K for the moment.
692      */
693     uint32_t eq_sizes[] = {
694         cpu_to_be32(16), /* 64K */
695     };
696     /*
697      * QEMU/KVM only needs to define a single range to reserve the
698      * escalation priority. A priority bitmask would have been more
699      * appropriate.
700      */
701     uint32_t plat_res_int_priorities[] = {
702         cpu_to_be32(xive->hv_prio),    /* start */
703         cpu_to_be32(0xff - xive->hv_prio), /* count */
704     };
705 
706     /* Thread Interrupt Management Area : User (ring 3) and OS (ring 2) */
707     timas[0] = cpu_to_be64(xive->tm_base +
708                            XIVE_TM_USER_PAGE * (1ull << TM_SHIFT));
709     timas[1] = cpu_to_be64(1ull << TM_SHIFT);
710     timas[2] = cpu_to_be64(xive->tm_base +
711                            XIVE_TM_OS_PAGE * (1ull << TM_SHIFT));
712     timas[3] = cpu_to_be64(1ull << TM_SHIFT);
713 
714     _FDT(node = fdt_add_subnode(fdt, 0, xive->nodename));
715 
716     _FDT(fdt_setprop_string(fdt, node, "device_type", "power-ivpe"));
717     _FDT(fdt_setprop(fdt, node, "reg", timas, sizeof(timas)));
718 
719     _FDT(fdt_setprop_string(fdt, node, "compatible", "ibm,power-ivpe"));
720     _FDT(fdt_setprop(fdt, node, "ibm,xive-eq-sizes", eq_sizes,
721                      sizeof(eq_sizes)));
722     _FDT(fdt_setprop(fdt, node, "ibm,xive-lisn-ranges", lisn_ranges,
723                      sizeof(lisn_ranges)));
724 
725     /* For Linux to link the LSIs to the interrupt controller. */
726     _FDT(fdt_setprop(fdt, node, "interrupt-controller", NULL, 0));
727     _FDT(fdt_setprop_cell(fdt, node, "#interrupt-cells", 2));
728 
729     /* For SLOF */
730     _FDT(fdt_setprop_cell(fdt, node, "linux,phandle", phandle));
731     _FDT(fdt_setprop_cell(fdt, node, "phandle", phandle));
732 
733     /*
734      * The "ibm,plat-res-int-priorities" property defines the priority
735      * ranges reserved by the hypervisor
736      */
737     _FDT(fdt_setprop(fdt, 0, "ibm,plat-res-int-priorities",
738                      plat_res_int_priorities, sizeof(plat_res_int_priorities)));
739 }
740 
741 static int spapr_xive_activate(SpaprInterruptController *intc,
742                                uint32_t nr_servers, Error **errp)
743 {
744     SpaprXive *xive = SPAPR_XIVE(intc);
745 
746     if (kvm_enabled()) {
747         int rc = spapr_irq_init_kvm(kvmppc_xive_connect, intc, nr_servers,
748                                     errp);
749         if (rc < 0) {
750             return rc;
751         }
752     }
753 
754     /* Activate the XIVE MMIOs */
755     spapr_xive_mmio_set_enabled(xive, true);
756 
757     return 0;
758 }
759 
760 static void spapr_xive_deactivate(SpaprInterruptController *intc)
761 {
762     SpaprXive *xive = SPAPR_XIVE(intc);
763 
764     spapr_xive_mmio_set_enabled(xive, false);
765 
766     if (spapr_xive_in_kernel(xive)) {
767         kvmppc_xive_disconnect(intc);
768     }
769 }
770 
771 static bool spapr_xive_in_kernel_xptr(const XivePresenter *xptr)
772 {
773     return spapr_xive_in_kernel(SPAPR_XIVE(xptr));
774 }
775 
776 static void spapr_xive_class_init(ObjectClass *klass, void *data)
777 {
778     DeviceClass *dc = DEVICE_CLASS(klass);
779     XiveRouterClass *xrc = XIVE_ROUTER_CLASS(klass);
780     SpaprInterruptControllerClass *sicc = SPAPR_INTC_CLASS(klass);
781     XivePresenterClass *xpc = XIVE_PRESENTER_CLASS(klass);
782     SpaprXiveClass *sxc = SPAPR_XIVE_CLASS(klass);
783 
784     dc->desc    = "sPAPR XIVE Interrupt Controller";
785     device_class_set_props(dc, spapr_xive_properties);
786     device_class_set_parent_realize(dc, spapr_xive_realize,
787                                     &sxc->parent_realize);
788     dc->vmsd    = &vmstate_spapr_xive;
789 
790     xrc->get_eas = spapr_xive_get_eas;
791     xrc->get_end = spapr_xive_get_end;
792     xrc->write_end = spapr_xive_write_end;
793     xrc->get_nvt = spapr_xive_get_nvt;
794     xrc->write_nvt = spapr_xive_write_nvt;
795     xrc->get_block_id = spapr_xive_get_block_id;
796 
797     sicc->activate = spapr_xive_activate;
798     sicc->deactivate = spapr_xive_deactivate;
799     sicc->cpu_intc_create = spapr_xive_cpu_intc_create;
800     sicc->cpu_intc_reset = spapr_xive_cpu_intc_reset;
801     sicc->cpu_intc_destroy = spapr_xive_cpu_intc_destroy;
802     sicc->claim_irq = spapr_xive_claim_irq;
803     sicc->free_irq = spapr_xive_free_irq;
804     sicc->set_irq = spapr_xive_set_irq;
805     sicc->print_info = spapr_xive_print_info;
806     sicc->dt = spapr_xive_dt;
807     sicc->post_load = spapr_xive_post_load;
808 
809     xpc->match_nvt  = spapr_xive_match_nvt;
810     xpc->in_kernel  = spapr_xive_in_kernel_xptr;
811 }
812 
813 static const TypeInfo spapr_xive_info = {
814     .name = TYPE_SPAPR_XIVE,
815     .parent = TYPE_XIVE_ROUTER,
816     .instance_init = spapr_xive_instance_init,
817     .instance_size = sizeof(SpaprXive),
818     .class_init = spapr_xive_class_init,
819     .class_size = sizeof(SpaprXiveClass),
820     .interfaces = (InterfaceInfo[]) {
821         { TYPE_SPAPR_INTC },
822         { }
823     },
824 };
825 
826 static void spapr_xive_register_types(void)
827 {
828     type_register_static(&spapr_xive_info);
829 }
830 
831 type_init(spapr_xive_register_types)
832 
833 /*
834  * XIVE hcalls
835  *
836  * The terminology used by the XIVE hcalls is the following :
837  *
838  *   TARGET vCPU number
839  *   EQ     Event Queue assigned by OS to receive event data
840  *   ESB    page for source interrupt management
841  *   LISN   Logical Interrupt Source Number identifying a source in the
842  *          machine
843  *   EISN   Effective Interrupt Source Number used by guest OS to
844  *          identify source in the guest
845  *
846  * The EAS, END, NVT structures are not exposed.
847  */
848 
849 /*
850  * On POWER9, the KVM XIVE device uses priority 7 for the escalation
851  * interrupts. So we only allow the guest to use priorities [0..6].
852  */
853 static bool spapr_xive_priority_is_reserved(SpaprXive *xive, uint8_t priority)
854 {
855     return priority >= xive->hv_prio;
856 }
857 
858 /*
859  * The H_INT_GET_SOURCE_INFO hcall() is used to obtain the logical
860  * real address of the MMIO page through which the Event State Buffer
861  * entry associated with the value of the "lisn" parameter is managed.
862  *
863  * Parameters:
864  * Input
865  * - R4: "flags"
866  *         Bits 0-63 reserved
867  * - R5: "lisn" is per "interrupts", "interrupt-map", or
868  *       "ibm,xive-lisn-ranges" properties, or as returned by the
869  *       ibm,query-interrupt-source-number RTAS call, or as returned
870  *       by the H_ALLOCATE_VAS_WINDOW hcall
871  *
872  * Output
873  * - R4: "flags"
874  *         Bits 0-59: Reserved
875  *         Bit 60: H_INT_ESB must be used for Event State Buffer
876  *                 management
877  *         Bit 61: 1 == LSI  0 == MSI
878  *         Bit 62: the full function page supports trigger
879  *         Bit 63: Store EOI Supported
880  * - R5: Logical Real address of full function Event State Buffer
881  *       management page, -1 if H_INT_ESB hcall flag is set to 1.
882  * - R6: Logical Real Address of trigger only Event State Buffer
883  *       management page or -1.
884  * - R7: Power of 2 page size for the ESB management pages returned in
885  *       R5 and R6.
886  */
887 
888 #define SPAPR_XIVE_SRC_H_INT_ESB     PPC_BIT(60) /* ESB manage with H_INT_ESB */
889 #define SPAPR_XIVE_SRC_LSI           PPC_BIT(61) /* Virtual LSI type */
890 #define SPAPR_XIVE_SRC_TRIGGER       PPC_BIT(62) /* Trigger and management
891                                                     on same page */
892 #define SPAPR_XIVE_SRC_STORE_EOI     PPC_BIT(63) /* Store EOI support */
893 
894 static target_ulong h_int_get_source_info(PowerPCCPU *cpu,
895                                           SpaprMachineState *spapr,
896                                           target_ulong opcode,
897                                           target_ulong *args)
898 {
899     SpaprXive *xive = spapr->xive;
900     XiveSource *xsrc = &xive->source;
901     target_ulong flags  = args[0];
902     target_ulong lisn   = args[1];
903 
904     trace_spapr_xive_get_source_info(flags, lisn);
905 
906     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
907         return H_FUNCTION;
908     }
909 
910     if (flags) {
911         return H_PARAMETER;
912     }
913 
914     if (lisn >= xive->nr_irqs) {
915         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
916                       lisn);
917         return H_P2;
918     }
919 
920     if (!xive_eas_is_valid(&xive->eat[lisn])) {
921         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
922                       lisn);
923         return H_P2;
924     }
925 
926     /*
927      * All sources are emulated under the main XIVE object and share
928      * the same characteristics.
929      */
930     args[0] = 0;
931     if (!xive_source_esb_has_2page(xsrc)) {
932         args[0] |= SPAPR_XIVE_SRC_TRIGGER;
933     }
934     if (xsrc->esb_flags & XIVE_SRC_STORE_EOI) {
935         args[0] |= SPAPR_XIVE_SRC_STORE_EOI;
936     }
937 
938     /*
939      * Force the use of the H_INT_ESB hcall in case of an LSI
940      * interrupt. This is necessary under KVM to re-trigger the
941      * interrupt if the level is still asserted
942      */
943     if (xive_source_irq_is_lsi(xsrc, lisn)) {
944         args[0] |= SPAPR_XIVE_SRC_H_INT_ESB | SPAPR_XIVE_SRC_LSI;
945     }
946 
947     if (!(args[0] & SPAPR_XIVE_SRC_H_INT_ESB)) {
948         args[1] = xive->vc_base + xive_source_esb_mgmt(xsrc, lisn);
949     } else {
950         args[1] = -1;
951     }
952 
953     if (xive_source_esb_has_2page(xsrc) &&
954         !(args[0] & SPAPR_XIVE_SRC_H_INT_ESB)) {
955         args[2] = xive->vc_base + xive_source_esb_page(xsrc, lisn);
956     } else {
957         args[2] = -1;
958     }
959 
960     if (xive_source_esb_has_2page(xsrc)) {
961         args[3] = xsrc->esb_shift - 1;
962     } else {
963         args[3] = xsrc->esb_shift;
964     }
965 
966     return H_SUCCESS;
967 }
968 
969 /*
970  * The H_INT_SET_SOURCE_CONFIG hcall() is used to assign a Logical
971  * Interrupt Source to a target. The Logical Interrupt Source is
972  * designated with the "lisn" parameter and the target is designated
973  * with the "target" and "priority" parameters.  Upon return from the
974  * hcall(), no additional interrupts will be directed to the old EQ.
975  *
976  * Parameters:
977  * Input:
978  * - R4: "flags"
979  *         Bits 0-61: Reserved
980  *         Bit 62: set the "eisn" in the EAS
981  *         Bit 63: masks the interrupt source in the hardware interrupt
982  *       control structure. An interrupt masked by this mechanism will
983  *       be dropped, but it's source state bits will still be
984  *       set. There is no race-free way of unmasking and restoring the
985  *       source. Thus this should only be used in interrupts that are
986  *       also masked at the source, and only in cases where the
987  *       interrupt is not meant to be used for a large amount of time
988  *       because no valid target exists for it for example
989  * - R5: "lisn" is per "interrupts", "interrupt-map", or
990  *       "ibm,xive-lisn-ranges" properties, or as returned by the
991  *       ibm,query-interrupt-source-number RTAS call, or as returned by
992  *       the H_ALLOCATE_VAS_WINDOW hcall
993  * - R6: "target" is per "ibm,ppc-interrupt-server#s" or
994  *       "ibm,ppc-interrupt-gserver#s"
995  * - R7: "priority" is a valid priority not in
996  *       "ibm,plat-res-int-priorities"
997  * - R8: "eisn" is the guest EISN associated with the "lisn"
998  *
999  * Output:
1000  * - None
1001  */
1002 
1003 #define SPAPR_XIVE_SRC_SET_EISN PPC_BIT(62)
1004 #define SPAPR_XIVE_SRC_MASK     PPC_BIT(63)
1005 
1006 static target_ulong h_int_set_source_config(PowerPCCPU *cpu,
1007                                             SpaprMachineState *spapr,
1008                                             target_ulong opcode,
1009                                             target_ulong *args)
1010 {
1011     SpaprXive *xive = spapr->xive;
1012     XiveEAS eas, new_eas;
1013     target_ulong flags    = args[0];
1014     target_ulong lisn     = args[1];
1015     target_ulong target   = args[2];
1016     target_ulong priority = args[3];
1017     target_ulong eisn     = args[4];
1018     uint8_t end_blk;
1019     uint32_t end_idx;
1020 
1021     trace_spapr_xive_set_source_config(flags, lisn, target, priority, eisn);
1022 
1023     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1024         return H_FUNCTION;
1025     }
1026 
1027     if (flags & ~(SPAPR_XIVE_SRC_SET_EISN | SPAPR_XIVE_SRC_MASK)) {
1028         return H_PARAMETER;
1029     }
1030 
1031     if (lisn >= xive->nr_irqs) {
1032         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1033                       lisn);
1034         return H_P2;
1035     }
1036 
1037     eas = xive->eat[lisn];
1038     if (!xive_eas_is_valid(&eas)) {
1039         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1040                       lisn);
1041         return H_P2;
1042     }
1043 
1044     /* priority 0xff is used to reset the EAS */
1045     if (priority == 0xff) {
1046         new_eas.w = cpu_to_be64(EAS_VALID | EAS_MASKED);
1047         goto out;
1048     }
1049 
1050     if (flags & SPAPR_XIVE_SRC_MASK) {
1051         new_eas.w = eas.w | cpu_to_be64(EAS_MASKED);
1052     } else {
1053         new_eas.w = eas.w & cpu_to_be64(~EAS_MASKED);
1054     }
1055 
1056     if (spapr_xive_priority_is_reserved(xive, priority)) {
1057         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1058                       " is reserved\n", priority);
1059         return H_P4;
1060     }
1061 
1062     /*
1063      * Validate that "target" is part of the list of threads allocated
1064      * to the partition. For that, find the END corresponding to the
1065      * target.
1066      */
1067     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1068         return H_P3;
1069     }
1070 
1071     new_eas.w = xive_set_field64(EAS_END_BLOCK, new_eas.w, end_blk);
1072     new_eas.w = xive_set_field64(EAS_END_INDEX, new_eas.w, end_idx);
1073 
1074     if (flags & SPAPR_XIVE_SRC_SET_EISN) {
1075         new_eas.w = xive_set_field64(EAS_END_DATA, new_eas.w, eisn);
1076     }
1077 
1078     if (spapr_xive_in_kernel(xive)) {
1079         Error *local_err = NULL;
1080 
1081         kvmppc_xive_set_source_config(xive, lisn, &new_eas, &local_err);
1082         if (local_err) {
1083             error_report_err(local_err);
1084             return H_HARDWARE;
1085         }
1086     }
1087 
1088 out:
1089     xive->eat[lisn] = new_eas;
1090     return H_SUCCESS;
1091 }
1092 
1093 /*
1094  * The H_INT_GET_SOURCE_CONFIG hcall() is used to determine to which
1095  * target/priority pair is assigned to the specified Logical Interrupt
1096  * Source.
1097  *
1098  * Parameters:
1099  * Input:
1100  * - R4: "flags"
1101  *         Bits 0-63 Reserved
1102  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1103  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1104  *       ibm,query-interrupt-source-number RTAS call, or as
1105  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1106  *
1107  * Output:
1108  * - R4: Target to which the specified Logical Interrupt Source is
1109  *       assigned
1110  * - R5: Priority to which the specified Logical Interrupt Source is
1111  *       assigned
1112  * - R6: EISN for the specified Logical Interrupt Source (this will be
1113  *       equivalent to the LISN if not changed by H_INT_SET_SOURCE_CONFIG)
1114  */
1115 static target_ulong h_int_get_source_config(PowerPCCPU *cpu,
1116                                             SpaprMachineState *spapr,
1117                                             target_ulong opcode,
1118                                             target_ulong *args)
1119 {
1120     SpaprXive *xive = spapr->xive;
1121     target_ulong flags = args[0];
1122     target_ulong lisn = args[1];
1123     XiveEAS eas;
1124     XiveEND *end;
1125     uint8_t nvt_blk;
1126     uint32_t end_idx, nvt_idx;
1127 
1128     trace_spapr_xive_get_source_config(flags, lisn);
1129 
1130     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1131         return H_FUNCTION;
1132     }
1133 
1134     if (flags) {
1135         return H_PARAMETER;
1136     }
1137 
1138     if (lisn >= xive->nr_irqs) {
1139         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1140                       lisn);
1141         return H_P2;
1142     }
1143 
1144     eas = xive->eat[lisn];
1145     if (!xive_eas_is_valid(&eas)) {
1146         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1147                       lisn);
1148         return H_P2;
1149     }
1150 
1151     /* EAS_END_BLOCK is unused on sPAPR */
1152     end_idx = xive_get_field64(EAS_END_INDEX, eas.w);
1153 
1154     assert(end_idx < xive->nr_ends);
1155     end = &xive->endt[end_idx];
1156 
1157     nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end->w6);
1158     nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end->w6);
1159     args[0] = spapr_xive_nvt_to_target(nvt_blk, nvt_idx);
1160 
1161     if (xive_eas_is_masked(&eas)) {
1162         args[1] = 0xff;
1163     } else {
1164         args[1] = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
1165     }
1166 
1167     args[2] = xive_get_field64(EAS_END_DATA, eas.w);
1168 
1169     return H_SUCCESS;
1170 }
1171 
1172 /*
1173  * The H_INT_GET_QUEUE_INFO hcall() is used to get the logical real
1174  * address of the notification management page associated with the
1175  * specified target and priority.
1176  *
1177  * Parameters:
1178  * Input:
1179  * - R4: "flags"
1180  *         Bits 0-63 Reserved
1181  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1182  *       "ibm,ppc-interrupt-gserver#s"
1183  * - R6: "priority" is a valid priority not in
1184  *       "ibm,plat-res-int-priorities"
1185  *
1186  * Output:
1187  * - R4: Logical real address of notification page
1188  * - R5: Power of 2 page size of the notification page
1189  */
1190 static target_ulong h_int_get_queue_info(PowerPCCPU *cpu,
1191                                          SpaprMachineState *spapr,
1192                                          target_ulong opcode,
1193                                          target_ulong *args)
1194 {
1195     SpaprXive *xive = spapr->xive;
1196     XiveENDSource *end_xsrc = &xive->end_source;
1197     target_ulong flags = args[0];
1198     target_ulong target = args[1];
1199     target_ulong priority = args[2];
1200     XiveEND *end;
1201     uint8_t end_blk;
1202     uint32_t end_idx;
1203 
1204     trace_spapr_xive_get_queue_info(flags, target, priority);
1205 
1206     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1207         return H_FUNCTION;
1208     }
1209 
1210     if (flags) {
1211         return H_PARAMETER;
1212     }
1213 
1214     /*
1215      * H_STATE should be returned if a H_INT_RESET is in progress.
1216      * This is not needed when running the emulation under QEMU
1217      */
1218 
1219     if (spapr_xive_priority_is_reserved(xive, priority)) {
1220         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1221                       " is reserved\n", priority);
1222         return H_P3;
1223     }
1224 
1225     /*
1226      * Validate that "target" is part of the list of threads allocated
1227      * to the partition. For that, find the END corresponding to the
1228      * target.
1229      */
1230     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1231         return H_P2;
1232     }
1233 
1234     assert(end_idx < xive->nr_ends);
1235     end = &xive->endt[end_idx];
1236 
1237     args[0] = xive->end_base + (1ull << (end_xsrc->esb_shift + 1)) * end_idx;
1238     if (xive_end_is_enqueue(end)) {
1239         args[1] = xive_get_field32(END_W0_QSIZE, end->w0) + 12;
1240     } else {
1241         args[1] = 0;
1242     }
1243 
1244     return H_SUCCESS;
1245 }
1246 
1247 /*
1248  * The H_INT_SET_QUEUE_CONFIG hcall() is used to set or reset a EQ for
1249  * a given "target" and "priority".  It is also used to set the
1250  * notification config associated with the EQ.  An EQ size of 0 is
1251  * used to reset the EQ config for a given target and priority. If
1252  * resetting the EQ config, the END associated with the given "target"
1253  * and "priority" will be changed to disable queueing.
1254  *
1255  * Upon return from the hcall(), no additional interrupts will be
1256  * directed to the old EQ (if one was set). The old EQ (if one was
1257  * set) should be investigated for interrupts that occurred prior to
1258  * or during the hcall().
1259  *
1260  * Parameters:
1261  * Input:
1262  * - R4: "flags"
1263  *         Bits 0-62: Reserved
1264  *         Bit 63: Unconditional Notify (n) per the XIVE spec
1265  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1266  *       "ibm,ppc-interrupt-gserver#s"
1267  * - R6: "priority" is a valid priority not in
1268  *       "ibm,plat-res-int-priorities"
1269  * - R7: "eventQueue": The logical real address of the start of the EQ
1270  * - R8: "eventQueueSize": The power of 2 EQ size per "ibm,xive-eq-sizes"
1271  *
1272  * Output:
1273  * - None
1274  */
1275 
1276 #define SPAPR_XIVE_END_ALWAYS_NOTIFY PPC_BIT(63)
1277 
1278 static target_ulong h_int_set_queue_config(PowerPCCPU *cpu,
1279                                            SpaprMachineState *spapr,
1280                                            target_ulong opcode,
1281                                            target_ulong *args)
1282 {
1283     SpaprXive *xive = spapr->xive;
1284     target_ulong flags = args[0];
1285     target_ulong target = args[1];
1286     target_ulong priority = args[2];
1287     target_ulong qpage = args[3];
1288     target_ulong qsize = args[4];
1289     XiveEND end;
1290     uint8_t end_blk, nvt_blk;
1291     uint32_t end_idx, nvt_idx;
1292 
1293     trace_spapr_xive_set_queue_config(flags, target, priority, qpage, qsize);
1294 
1295     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1296         return H_FUNCTION;
1297     }
1298 
1299     if (flags & ~SPAPR_XIVE_END_ALWAYS_NOTIFY) {
1300         return H_PARAMETER;
1301     }
1302 
1303     /*
1304      * H_STATE should be returned if a H_INT_RESET is in progress.
1305      * This is not needed when running the emulation under QEMU
1306      */
1307 
1308     if (spapr_xive_priority_is_reserved(xive, priority)) {
1309         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1310                       " is reserved\n", priority);
1311         return H_P3;
1312     }
1313 
1314     /*
1315      * Validate that "target" is part of the list of threads allocated
1316      * to the partition. For that, find the END corresponding to the
1317      * target.
1318      */
1319 
1320     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1321         return H_P2;
1322     }
1323 
1324     assert(end_idx < xive->nr_ends);
1325     memcpy(&end, &xive->endt[end_idx], sizeof(XiveEND));
1326 
1327     switch (qsize) {
1328     case 12:
1329     case 16:
1330     case 21:
1331     case 24:
1332         if (!QEMU_IS_ALIGNED(qpage, 1ul << qsize)) {
1333             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: EQ @0x%" HWADDR_PRIx
1334                           " is not naturally aligned with %" HWADDR_PRIx "\n",
1335                           qpage, (hwaddr)1 << qsize);
1336             return H_P4;
1337         }
1338         end.w2 = cpu_to_be32((qpage >> 32) & 0x0fffffff);
1339         end.w3 = cpu_to_be32(qpage & 0xffffffff);
1340         end.w0 |= cpu_to_be32(END_W0_ENQUEUE);
1341         end.w0 = xive_set_field32(END_W0_QSIZE, end.w0, qsize - 12);
1342         break;
1343     case 0:
1344         /* reset queue and disable queueing */
1345         spapr_xive_end_reset(&end);
1346         goto out;
1347 
1348     default:
1349         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid EQ size %"PRIx64"\n",
1350                       qsize);
1351         return H_P5;
1352     }
1353 
1354     if (qsize) {
1355         hwaddr plen = 1 << qsize;
1356         void *eq;
1357 
1358         /*
1359          * Validate the guest EQ. We should also check that the queue
1360          * has been zeroed by the OS.
1361          */
1362         eq = address_space_map(CPU(cpu)->as, qpage, &plen, true,
1363                                MEMTXATTRS_UNSPECIFIED);
1364         if (plen != 1 << qsize) {
1365             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to map EQ @0x%"
1366                           HWADDR_PRIx "\n", qpage);
1367             return H_P4;
1368         }
1369         address_space_unmap(CPU(cpu)->as, eq, plen, true, plen);
1370     }
1371 
1372     /* "target" should have been validated above */
1373     if (spapr_xive_target_to_nvt(target, &nvt_blk, &nvt_idx)) {
1374         g_assert_not_reached();
1375     }
1376 
1377     /*
1378      * Ensure the priority and target are correctly set (they will not
1379      * be right after allocation)
1380      */
1381     end.w6 = xive_set_field32(END_W6_NVT_BLOCK, 0ul, nvt_blk) |
1382         xive_set_field32(END_W6_NVT_INDEX, 0ul, nvt_idx);
1383     end.w7 = xive_set_field32(END_W7_F0_PRIORITY, 0ul, priority);
1384 
1385     if (flags & SPAPR_XIVE_END_ALWAYS_NOTIFY) {
1386         end.w0 |= cpu_to_be32(END_W0_UCOND_NOTIFY);
1387     } else {
1388         end.w0 &= cpu_to_be32((uint32_t)~END_W0_UCOND_NOTIFY);
1389     }
1390 
1391     /*
1392      * The generation bit for the END starts at 1 and The END page
1393      * offset counter starts at 0.
1394      */
1395     end.w1 = cpu_to_be32(END_W1_GENERATION) |
1396         xive_set_field32(END_W1_PAGE_OFF, 0ul, 0ul);
1397     end.w0 |= cpu_to_be32(END_W0_VALID);
1398 
1399     /*
1400      * TODO: issue syncs required to ensure all in-flight interrupts
1401      * are complete on the old END
1402      */
1403 
1404 out:
1405     if (spapr_xive_in_kernel(xive)) {
1406         Error *local_err = NULL;
1407 
1408         kvmppc_xive_set_queue_config(xive, end_blk, end_idx, &end, &local_err);
1409         if (local_err) {
1410             error_report_err(local_err);
1411             return H_HARDWARE;
1412         }
1413     }
1414 
1415     /* Update END */
1416     memcpy(&xive->endt[end_idx], &end, sizeof(XiveEND));
1417     return H_SUCCESS;
1418 }
1419 
1420 /*
1421  * The H_INT_GET_QUEUE_CONFIG hcall() is used to get a EQ for a given
1422  * target and priority.
1423  *
1424  * Parameters:
1425  * Input:
1426  * - R4: "flags"
1427  *         Bits 0-62: Reserved
1428  *         Bit 63: Debug: Return debug data
1429  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1430  *       "ibm,ppc-interrupt-gserver#s"
1431  * - R6: "priority" is a valid priority not in
1432  *       "ibm,plat-res-int-priorities"
1433  *
1434  * Output:
1435  * - R4: "flags":
1436  *       Bits 0-61: Reserved
1437  *       Bit 62: The value of Event Queue Generation Number (g) per
1438  *              the XIVE spec if "Debug" = 1
1439  *       Bit 63: The value of Unconditional Notify (n) per the XIVE spec
1440  * - R5: The logical real address of the start of the EQ
1441  * - R6: The power of 2 EQ size per "ibm,xive-eq-sizes"
1442  * - R7: The value of Event Queue Offset Counter per XIVE spec
1443  *       if "Debug" = 1, else 0
1444  *
1445  */
1446 
1447 #define SPAPR_XIVE_END_DEBUG     PPC_BIT(63)
1448 
1449 static target_ulong h_int_get_queue_config(PowerPCCPU *cpu,
1450                                            SpaprMachineState *spapr,
1451                                            target_ulong opcode,
1452                                            target_ulong *args)
1453 {
1454     SpaprXive *xive = spapr->xive;
1455     target_ulong flags = args[0];
1456     target_ulong target = args[1];
1457     target_ulong priority = args[2];
1458     XiveEND *end;
1459     uint8_t end_blk;
1460     uint32_t end_idx;
1461 
1462     trace_spapr_xive_get_queue_config(flags, target, priority);
1463 
1464     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1465         return H_FUNCTION;
1466     }
1467 
1468     if (flags & ~SPAPR_XIVE_END_DEBUG) {
1469         return H_PARAMETER;
1470     }
1471 
1472     /*
1473      * H_STATE should be returned if a H_INT_RESET is in progress.
1474      * This is not needed when running the emulation under QEMU
1475      */
1476 
1477     if (spapr_xive_priority_is_reserved(xive, priority)) {
1478         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1479                       " is reserved\n", priority);
1480         return H_P3;
1481     }
1482 
1483     /*
1484      * Validate that "target" is part of the list of threads allocated
1485      * to the partition. For that, find the END corresponding to the
1486      * target.
1487      */
1488     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1489         return H_P2;
1490     }
1491 
1492     assert(end_idx < xive->nr_ends);
1493     end = &xive->endt[end_idx];
1494 
1495     args[0] = 0;
1496     if (xive_end_is_notify(end)) {
1497         args[0] |= SPAPR_XIVE_END_ALWAYS_NOTIFY;
1498     }
1499 
1500     if (xive_end_is_enqueue(end)) {
1501         args[1] = xive_end_qaddr(end);
1502         args[2] = xive_get_field32(END_W0_QSIZE, end->w0) + 12;
1503     } else {
1504         args[1] = 0;
1505         args[2] = 0;
1506     }
1507 
1508     if (spapr_xive_in_kernel(xive)) {
1509         Error *local_err = NULL;
1510 
1511         kvmppc_xive_get_queue_config(xive, end_blk, end_idx, end, &local_err);
1512         if (local_err) {
1513             error_report_err(local_err);
1514             return H_HARDWARE;
1515         }
1516     }
1517 
1518     /* TODO: do we need any locking on the END ? */
1519     if (flags & SPAPR_XIVE_END_DEBUG) {
1520         /* Load the event queue generation number into the return flags */
1521         args[0] |= (uint64_t)xive_get_field32(END_W1_GENERATION, end->w1) << 62;
1522 
1523         /* Load R7 with the event queue offset counter */
1524         args[3] = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1525     } else {
1526         args[3] = 0;
1527     }
1528 
1529     return H_SUCCESS;
1530 }
1531 
1532 /*
1533  * The H_INT_SET_OS_REPORTING_LINE hcall() is used to set the
1534  * reporting cache line pair for the calling thread.  The reporting
1535  * cache lines will contain the OS interrupt context when the OS
1536  * issues a CI store byte to @TIMA+0xC10 to acknowledge the OS
1537  * interrupt. The reporting cache lines can be reset by inputting -1
1538  * in "reportingLine".  Issuing the CI store byte without reporting
1539  * cache lines registered will result in the data not being accessible
1540  * to the OS.
1541  *
1542  * Parameters:
1543  * Input:
1544  * - R4: "flags"
1545  *         Bits 0-63: Reserved
1546  * - R5: "reportingLine": The logical real address of the reporting cache
1547  *       line pair
1548  *
1549  * Output:
1550  * - None
1551  */
1552 static target_ulong h_int_set_os_reporting_line(PowerPCCPU *cpu,
1553                                                 SpaprMachineState *spapr,
1554                                                 target_ulong opcode,
1555                                                 target_ulong *args)
1556 {
1557     target_ulong flags   = args[0];
1558 
1559     trace_spapr_xive_set_os_reporting_line(flags);
1560 
1561     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1562         return H_FUNCTION;
1563     }
1564 
1565     /*
1566      * H_STATE should be returned if a H_INT_RESET is in progress.
1567      * This is not needed when running the emulation under QEMU
1568      */
1569 
1570     /* TODO: H_INT_SET_OS_REPORTING_LINE */
1571     return H_FUNCTION;
1572 }
1573 
1574 /*
1575  * The H_INT_GET_OS_REPORTING_LINE hcall() is used to get the logical
1576  * real address of the reporting cache line pair set for the input
1577  * "target".  If no reporting cache line pair has been set, -1 is
1578  * returned.
1579  *
1580  * Parameters:
1581  * Input:
1582  * - R4: "flags"
1583  *         Bits 0-63: Reserved
1584  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1585  *       "ibm,ppc-interrupt-gserver#s"
1586  * - R6: "reportingLine": The logical real address of the reporting
1587  *        cache line pair
1588  *
1589  * Output:
1590  * - R4: The logical real address of the reporting line if set, else -1
1591  */
1592 static target_ulong h_int_get_os_reporting_line(PowerPCCPU *cpu,
1593                                                 SpaprMachineState *spapr,
1594                                                 target_ulong opcode,
1595                                                 target_ulong *args)
1596 {
1597     target_ulong flags   = args[0];
1598 
1599     trace_spapr_xive_get_os_reporting_line(flags);
1600 
1601     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1602         return H_FUNCTION;
1603     }
1604 
1605     /*
1606      * H_STATE should be returned if a H_INT_RESET is in progress.
1607      * This is not needed when running the emulation under QEMU
1608      */
1609 
1610     /* TODO: H_INT_GET_OS_REPORTING_LINE */
1611     return H_FUNCTION;
1612 }
1613 
1614 /*
1615  * The H_INT_ESB hcall() is used to issue a load or store to the ESB
1616  * page for the input "lisn".  This hcall is only supported for LISNs
1617  * that have the ESB hcall flag set to 1 when returned from hcall()
1618  * H_INT_GET_SOURCE_INFO.
1619  *
1620  * Parameters:
1621  * Input:
1622  * - R4: "flags"
1623  *         Bits 0-62: Reserved
1624  *         bit 63: Store: Store=1, store operation, else load operation
1625  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1626  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1627  *       ibm,query-interrupt-source-number RTAS call, or as
1628  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1629  * - R6: "esbOffset" is the offset into the ESB page for the load or
1630  *       store operation
1631  * - R7: "storeData" is the data to write for a store operation
1632  *
1633  * Output:
1634  * - R4: The value of the load if load operation, else -1
1635  */
1636 
1637 #define SPAPR_XIVE_ESB_STORE PPC_BIT(63)
1638 
1639 static target_ulong h_int_esb(PowerPCCPU *cpu,
1640                               SpaprMachineState *spapr,
1641                               target_ulong opcode,
1642                               target_ulong *args)
1643 {
1644     SpaprXive *xive = spapr->xive;
1645     XiveEAS eas;
1646     target_ulong flags  = args[0];
1647     target_ulong lisn   = args[1];
1648     target_ulong offset = args[2];
1649     target_ulong data   = args[3];
1650     hwaddr mmio_addr;
1651     XiveSource *xsrc = &xive->source;
1652 
1653     trace_spapr_xive_esb(flags, lisn, offset, data);
1654 
1655     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1656         return H_FUNCTION;
1657     }
1658 
1659     if (flags & ~SPAPR_XIVE_ESB_STORE) {
1660         return H_PARAMETER;
1661     }
1662 
1663     if (lisn >= xive->nr_irqs) {
1664         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1665                       lisn);
1666         return H_P2;
1667     }
1668 
1669     eas = xive->eat[lisn];
1670     if (!xive_eas_is_valid(&eas)) {
1671         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1672                       lisn);
1673         return H_P2;
1674     }
1675 
1676     if (offset > (1ull << xsrc->esb_shift)) {
1677         return H_P3;
1678     }
1679 
1680     if (spapr_xive_in_kernel(xive)) {
1681         args[0] = kvmppc_xive_esb_rw(xsrc, lisn, offset, data,
1682                                      flags & SPAPR_XIVE_ESB_STORE);
1683     } else {
1684         mmio_addr = xive->vc_base + xive_source_esb_mgmt(xsrc, lisn) + offset;
1685 
1686         if (dma_memory_rw(&address_space_memory, mmio_addr, &data, 8,
1687                           (flags & SPAPR_XIVE_ESB_STORE))) {
1688             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to access ESB @0x%"
1689                           HWADDR_PRIx "\n", mmio_addr);
1690             return H_HARDWARE;
1691         }
1692         args[0] = (flags & SPAPR_XIVE_ESB_STORE) ? -1 : data;
1693     }
1694     return H_SUCCESS;
1695 }
1696 
1697 /*
1698  * The H_INT_SYNC hcall() is used to issue hardware syncs that will
1699  * ensure any in flight events for the input lisn are in the event
1700  * queue.
1701  *
1702  * Parameters:
1703  * Input:
1704  * - R4: "flags"
1705  *         Bits 0-63: Reserved
1706  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1707  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1708  *       ibm,query-interrupt-source-number RTAS call, or as
1709  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1710  *
1711  * Output:
1712  * - None
1713  */
1714 static target_ulong h_int_sync(PowerPCCPU *cpu,
1715                                SpaprMachineState *spapr,
1716                                target_ulong opcode,
1717                                target_ulong *args)
1718 {
1719     SpaprXive *xive = spapr->xive;
1720     XiveEAS eas;
1721     target_ulong flags = args[0];
1722     target_ulong lisn = args[1];
1723 
1724     trace_spapr_xive_sync(flags, lisn);
1725 
1726     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1727         return H_FUNCTION;
1728     }
1729 
1730     if (flags) {
1731         return H_PARAMETER;
1732     }
1733 
1734     if (lisn >= xive->nr_irqs) {
1735         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1736                       lisn);
1737         return H_P2;
1738     }
1739 
1740     eas = xive->eat[lisn];
1741     if (!xive_eas_is_valid(&eas)) {
1742         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1743                       lisn);
1744         return H_P2;
1745     }
1746 
1747     /*
1748      * H_STATE should be returned if a H_INT_RESET is in progress.
1749      * This is not needed when running the emulation under QEMU
1750      */
1751 
1752     /*
1753      * This is not real hardware. Nothing to be done unless when
1754      * under KVM
1755      */
1756 
1757     if (spapr_xive_in_kernel(xive)) {
1758         Error *local_err = NULL;
1759 
1760         kvmppc_xive_sync_source(xive, lisn, &local_err);
1761         if (local_err) {
1762             error_report_err(local_err);
1763             return H_HARDWARE;
1764         }
1765     }
1766     return H_SUCCESS;
1767 }
1768 
1769 /*
1770  * The H_INT_RESET hcall() is used to reset all of the partition's
1771  * interrupt exploitation structures to their initial state.  This
1772  * means losing all previously set interrupt state set via
1773  * H_INT_SET_SOURCE_CONFIG and H_INT_SET_QUEUE_CONFIG.
1774  *
1775  * Parameters:
1776  * Input:
1777  * - R4: "flags"
1778  *         Bits 0-63: Reserved
1779  *
1780  * Output:
1781  * - None
1782  */
1783 static target_ulong h_int_reset(PowerPCCPU *cpu,
1784                                 SpaprMachineState *spapr,
1785                                 target_ulong opcode,
1786                                 target_ulong *args)
1787 {
1788     SpaprXive *xive = spapr->xive;
1789     target_ulong flags   = args[0];
1790 
1791     trace_spapr_xive_reset(flags);
1792 
1793     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1794         return H_FUNCTION;
1795     }
1796 
1797     if (flags) {
1798         return H_PARAMETER;
1799     }
1800 
1801     device_cold_reset(DEVICE(xive));
1802 
1803     if (spapr_xive_in_kernel(xive)) {
1804         Error *local_err = NULL;
1805 
1806         kvmppc_xive_reset(xive, &local_err);
1807         if (local_err) {
1808             error_report_err(local_err);
1809             return H_HARDWARE;
1810         }
1811     }
1812     return H_SUCCESS;
1813 }
1814 
1815 void spapr_xive_hcall_init(SpaprMachineState *spapr)
1816 {
1817     spapr_register_hypercall(H_INT_GET_SOURCE_INFO, h_int_get_source_info);
1818     spapr_register_hypercall(H_INT_SET_SOURCE_CONFIG, h_int_set_source_config);
1819     spapr_register_hypercall(H_INT_GET_SOURCE_CONFIG, h_int_get_source_config);
1820     spapr_register_hypercall(H_INT_GET_QUEUE_INFO, h_int_get_queue_info);
1821     spapr_register_hypercall(H_INT_SET_QUEUE_CONFIG, h_int_set_queue_config);
1822     spapr_register_hypercall(H_INT_GET_QUEUE_CONFIG, h_int_get_queue_config);
1823     spapr_register_hypercall(H_INT_SET_OS_REPORTING_LINE,
1824                              h_int_set_os_reporting_line);
1825     spapr_register_hypercall(H_INT_GET_OS_REPORTING_LINE,
1826                              h_int_get_os_reporting_line);
1827     spapr_register_hypercall(H_INT_ESB, h_int_esb);
1828     spapr_register_hypercall(H_INT_SYNC, h_int_sync);
1829     spapr_register_hypercall(H_INT_RESET, h_int_reset);
1830 }
1831