xref: /openbmc/qemu/hw/intc/spapr_xive.c (revision 354908ce)
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), xive->nr_irqs, "nr-irqs",
312                             &error_fatal);
313     object_property_set_link(OBJECT(xsrc), OBJECT(xive), "xive",
314                              &error_abort);
315     qdev_realize(DEVICE(xsrc), NULL, &local_err);
316     if (local_err) {
317         error_propagate(errp, local_err);
318         return;
319     }
320     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &xsrc->esb_mmio);
321 
322     /*
323      * Initialize the END ESB source
324      */
325     object_property_set_int(OBJECT(end_xsrc), xive->nr_irqs, "nr-ends",
326                             &error_fatal);
327     object_property_set_link(OBJECT(end_xsrc), OBJECT(xive), "xive",
328                              &error_abort);
329     qdev_realize(DEVICE(end_xsrc), NULL, &local_err);
330     if (local_err) {
331         error_propagate(errp, local_err);
332         return;
333     }
334     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &end_xsrc->esb_mmio);
335 
336     /* Set the mapping address of the END ESB pages after the source ESBs */
337     xive->end_base = xive->vc_base + (1ull << xsrc->esb_shift) * xsrc->nr_irqs;
338 
339     /*
340      * Allocate the routing tables
341      */
342     xive->eat = g_new0(XiveEAS, xive->nr_irqs);
343     xive->endt = g_new0(XiveEND, xive->nr_ends);
344 
345     xive->nodename = g_strdup_printf("interrupt-controller@%" PRIx64,
346                            xive->tm_base + XIVE_TM_USER_PAGE * (1 << TM_SHIFT));
347 
348     qemu_register_reset(spapr_xive_reset, dev);
349 
350     /* TIMA initialization */
351     memory_region_init_io(&xive->tm_mmio, OBJECT(xive), &spapr_xive_tm_ops,
352                           xive, "xive.tima", 4ull << TM_SHIFT);
353     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &xive->tm_mmio);
354 
355     /*
356      * Map all regions. These will be enabled or disabled at reset and
357      * can also be overridden by KVM memory regions if active
358      */
359     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 0, xive->vc_base);
360     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 1, xive->end_base);
361     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 2, xive->tm_base);
362 }
363 
364 static int spapr_xive_get_eas(XiveRouter *xrtr, uint8_t eas_blk,
365                               uint32_t eas_idx, XiveEAS *eas)
366 {
367     SpaprXive *xive = SPAPR_XIVE(xrtr);
368 
369     if (eas_idx >= xive->nr_irqs) {
370         return -1;
371     }
372 
373     *eas = xive->eat[eas_idx];
374     return 0;
375 }
376 
377 static int spapr_xive_get_end(XiveRouter *xrtr,
378                               uint8_t end_blk, uint32_t end_idx, XiveEND *end)
379 {
380     SpaprXive *xive = SPAPR_XIVE(xrtr);
381 
382     if (end_idx >= xive->nr_ends) {
383         return -1;
384     }
385 
386     memcpy(end, &xive->endt[end_idx], sizeof(XiveEND));
387     return 0;
388 }
389 
390 static int spapr_xive_write_end(XiveRouter *xrtr, uint8_t end_blk,
391                                 uint32_t end_idx, XiveEND *end,
392                                 uint8_t word_number)
393 {
394     SpaprXive *xive = SPAPR_XIVE(xrtr);
395 
396     if (end_idx >= xive->nr_ends) {
397         return -1;
398     }
399 
400     memcpy(&xive->endt[end_idx], end, sizeof(XiveEND));
401     return 0;
402 }
403 
404 static int spapr_xive_get_nvt(XiveRouter *xrtr,
405                               uint8_t nvt_blk, uint32_t nvt_idx, XiveNVT *nvt)
406 {
407     uint32_t vcpu_id = spapr_xive_nvt_to_target(nvt_blk, nvt_idx);
408     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
409 
410     if (!cpu) {
411         /* TODO: should we assert() if we can find a NVT ? */
412         return -1;
413     }
414 
415     /*
416      * sPAPR does not maintain a NVT table. Return that the NVT is
417      * valid if we have found a matching CPU
418      */
419     nvt->w0 = cpu_to_be32(NVT_W0_VALID);
420     return 0;
421 }
422 
423 static int spapr_xive_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk,
424                                 uint32_t nvt_idx, XiveNVT *nvt,
425                                 uint8_t word_number)
426 {
427     /*
428      * We don't need to write back to the NVTs because the sPAPR
429      * machine should never hit a non-scheduled NVT. It should never
430      * get called.
431      */
432     g_assert_not_reached();
433 }
434 
435 static int spapr_xive_match_nvt(XivePresenter *xptr, uint8_t format,
436                                 uint8_t nvt_blk, uint32_t nvt_idx,
437                                 bool cam_ignore, uint8_t priority,
438                                 uint32_t logic_serv, XiveTCTXMatch *match)
439 {
440     CPUState *cs;
441     int count = 0;
442 
443     CPU_FOREACH(cs) {
444         PowerPCCPU *cpu = POWERPC_CPU(cs);
445         XiveTCTX *tctx = spapr_cpu_state(cpu)->tctx;
446         int ring;
447 
448         /*
449          * Skip partially initialized vCPUs. This can happen when
450          * vCPUs are hotplugged.
451          */
452         if (!tctx) {
453             continue;
454         }
455 
456         /*
457          * Check the thread context CAM lines and record matches.
458          */
459         ring = xive_presenter_tctx_match(xptr, tctx, format, nvt_blk, nvt_idx,
460                                          cam_ignore, logic_serv);
461         /*
462          * Save the matching thread interrupt context and follow on to
463          * check for duplicates which are invalid.
464          */
465         if (ring != -1) {
466             if (match->tctx) {
467                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: already found a thread "
468                               "context NVT %x/%x\n", nvt_blk, nvt_idx);
469                 return -1;
470             }
471 
472             match->ring = ring;
473             match->tctx = tctx;
474             count++;
475         }
476     }
477 
478     return count;
479 }
480 
481 static uint8_t spapr_xive_get_block_id(XiveRouter *xrtr)
482 {
483     return SPAPR_XIVE_BLOCK_ID;
484 }
485 
486 static const VMStateDescription vmstate_spapr_xive_end = {
487     .name = TYPE_SPAPR_XIVE "/end",
488     .version_id = 1,
489     .minimum_version_id = 1,
490     .fields = (VMStateField []) {
491         VMSTATE_UINT32(w0, XiveEND),
492         VMSTATE_UINT32(w1, XiveEND),
493         VMSTATE_UINT32(w2, XiveEND),
494         VMSTATE_UINT32(w3, XiveEND),
495         VMSTATE_UINT32(w4, XiveEND),
496         VMSTATE_UINT32(w5, XiveEND),
497         VMSTATE_UINT32(w6, XiveEND),
498         VMSTATE_UINT32(w7, XiveEND),
499         VMSTATE_END_OF_LIST()
500     },
501 };
502 
503 static const VMStateDescription vmstate_spapr_xive_eas = {
504     .name = TYPE_SPAPR_XIVE "/eas",
505     .version_id = 1,
506     .minimum_version_id = 1,
507     .fields = (VMStateField []) {
508         VMSTATE_UINT64(w, XiveEAS),
509         VMSTATE_END_OF_LIST()
510     },
511 };
512 
513 static int vmstate_spapr_xive_pre_save(void *opaque)
514 {
515     if (kvm_irqchip_in_kernel()) {
516         return kvmppc_xive_pre_save(SPAPR_XIVE(opaque));
517     }
518 
519     return 0;
520 }
521 
522 /*
523  * Called by the sPAPR IRQ backend 'post_load' method at the machine
524  * level.
525  */
526 static int spapr_xive_post_load(SpaprInterruptController *intc, int version_id)
527 {
528     if (kvm_irqchip_in_kernel()) {
529         return kvmppc_xive_post_load(SPAPR_XIVE(intc), version_id);
530     }
531 
532     return 0;
533 }
534 
535 static const VMStateDescription vmstate_spapr_xive = {
536     .name = TYPE_SPAPR_XIVE,
537     .version_id = 1,
538     .minimum_version_id = 1,
539     .pre_save = vmstate_spapr_xive_pre_save,
540     .post_load = NULL, /* handled at the machine level */
541     .fields = (VMStateField[]) {
542         VMSTATE_UINT32_EQUAL(nr_irqs, SpaprXive, NULL),
543         VMSTATE_STRUCT_VARRAY_POINTER_UINT32(eat, SpaprXive, nr_irqs,
544                                      vmstate_spapr_xive_eas, XiveEAS),
545         VMSTATE_STRUCT_VARRAY_POINTER_UINT32(endt, SpaprXive, nr_ends,
546                                              vmstate_spapr_xive_end, XiveEND),
547         VMSTATE_END_OF_LIST()
548     },
549 };
550 
551 static int spapr_xive_claim_irq(SpaprInterruptController *intc, int lisn,
552                                 bool lsi, Error **errp)
553 {
554     SpaprXive *xive = SPAPR_XIVE(intc);
555     XiveSource *xsrc = &xive->source;
556 
557     assert(lisn < xive->nr_irqs);
558 
559     if (xive_eas_is_valid(&xive->eat[lisn])) {
560         error_setg(errp, "IRQ %d is not free", lisn);
561         return -EBUSY;
562     }
563 
564     /*
565      * Set default values when allocating an IRQ number
566      */
567     xive->eat[lisn].w |= cpu_to_be64(EAS_VALID | EAS_MASKED);
568     if (lsi) {
569         xive_source_irq_set_lsi(xsrc, lisn);
570     }
571 
572     if (kvm_irqchip_in_kernel()) {
573         return kvmppc_xive_source_reset_one(xsrc, lisn, errp);
574     }
575 
576     return 0;
577 }
578 
579 static void spapr_xive_free_irq(SpaprInterruptController *intc, int lisn)
580 {
581     SpaprXive *xive = SPAPR_XIVE(intc);
582     assert(lisn < xive->nr_irqs);
583 
584     xive->eat[lisn].w &= cpu_to_be64(~EAS_VALID);
585 }
586 
587 static Property spapr_xive_properties[] = {
588     DEFINE_PROP_UINT32("nr-irqs", SpaprXive, nr_irqs, 0),
589     DEFINE_PROP_UINT32("nr-ends", SpaprXive, nr_ends, 0),
590     DEFINE_PROP_UINT64("vc-base", SpaprXive, vc_base, SPAPR_XIVE_VC_BASE),
591     DEFINE_PROP_UINT64("tm-base", SpaprXive, tm_base, SPAPR_XIVE_TM_BASE),
592     DEFINE_PROP_END_OF_LIST(),
593 };
594 
595 static int spapr_xive_cpu_intc_create(SpaprInterruptController *intc,
596                                       PowerPCCPU *cpu, Error **errp)
597 {
598     SpaprXive *xive = SPAPR_XIVE(intc);
599     Object *obj;
600     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
601 
602     obj = xive_tctx_create(OBJECT(cpu), XIVE_PRESENTER(xive), errp);
603     if (!obj) {
604         return -1;
605     }
606 
607     spapr_cpu->tctx = XIVE_TCTX(obj);
608     return 0;
609 }
610 
611 static void xive_tctx_set_os_cam(XiveTCTX *tctx, uint32_t os_cam)
612 {
613     uint32_t qw1w2 = cpu_to_be32(TM_QW1W2_VO | os_cam);
614     memcpy(&tctx->regs[TM_QW1_OS + TM_WORD2], &qw1w2, 4);
615 }
616 
617 static void spapr_xive_cpu_intc_reset(SpaprInterruptController *intc,
618                                      PowerPCCPU *cpu)
619 {
620     XiveTCTX *tctx = spapr_cpu_state(cpu)->tctx;
621     uint8_t  nvt_blk;
622     uint32_t nvt_idx;
623 
624     xive_tctx_reset(tctx);
625 
626     /*
627      * When a Virtual Processor is scheduled to run on a HW thread,
628      * the hypervisor pushes its identifier in the OS CAM line.
629      * Emulate the same behavior under QEMU.
630      */
631     spapr_xive_cpu_to_nvt(cpu, &nvt_blk, &nvt_idx);
632 
633     xive_tctx_set_os_cam(tctx, xive_nvt_cam_line(nvt_blk, nvt_idx));
634 }
635 
636 static void spapr_xive_cpu_intc_destroy(SpaprInterruptController *intc,
637                                         PowerPCCPU *cpu)
638 {
639     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
640 
641     xive_tctx_destroy(spapr_cpu->tctx);
642     spapr_cpu->tctx = NULL;
643 }
644 
645 static void spapr_xive_set_irq(SpaprInterruptController *intc, int irq, int val)
646 {
647     SpaprXive *xive = SPAPR_XIVE(intc);
648 
649     if (kvm_irqchip_in_kernel()) {
650         kvmppc_xive_source_set_irq(&xive->source, irq, val);
651     } else {
652         xive_source_set_irq(&xive->source, irq, val);
653     }
654 }
655 
656 static void spapr_xive_print_info(SpaprInterruptController *intc, Monitor *mon)
657 {
658     SpaprXive *xive = SPAPR_XIVE(intc);
659     CPUState *cs;
660 
661     CPU_FOREACH(cs) {
662         PowerPCCPU *cpu = POWERPC_CPU(cs);
663 
664         xive_tctx_pic_print_info(spapr_cpu_state(cpu)->tctx, mon);
665     }
666 
667     spapr_xive_pic_print_info(xive, mon);
668 }
669 
670 static void spapr_xive_dt(SpaprInterruptController *intc, uint32_t nr_servers,
671                           void *fdt, uint32_t phandle)
672 {
673     SpaprXive *xive = SPAPR_XIVE(intc);
674     int node;
675     uint64_t timas[2 * 2];
676     /* Interrupt number ranges for the IPIs */
677     uint32_t lisn_ranges[] = {
678         cpu_to_be32(SPAPR_IRQ_IPI),
679         cpu_to_be32(SPAPR_IRQ_IPI + nr_servers),
680     };
681     /*
682      * EQ size - the sizes of pages supported by the system 4K, 64K,
683      * 2M, 16M. We only advertise 64K for the moment.
684      */
685     uint32_t eq_sizes[] = {
686         cpu_to_be32(16), /* 64K */
687     };
688     /*
689      * The following array is in sync with the reserved priorities
690      * defined by the 'spapr_xive_priority_is_reserved' routine.
691      */
692     uint32_t plat_res_int_priorities[] = {
693         cpu_to_be32(7),    /* start */
694         cpu_to_be32(0xf8), /* count */
695     };
696 
697     /* Thread Interrupt Management Area : User (ring 3) and OS (ring 2) */
698     timas[0] = cpu_to_be64(xive->tm_base +
699                            XIVE_TM_USER_PAGE * (1ull << TM_SHIFT));
700     timas[1] = cpu_to_be64(1ull << TM_SHIFT);
701     timas[2] = cpu_to_be64(xive->tm_base +
702                            XIVE_TM_OS_PAGE * (1ull << TM_SHIFT));
703     timas[3] = cpu_to_be64(1ull << TM_SHIFT);
704 
705     _FDT(node = fdt_add_subnode(fdt, 0, xive->nodename));
706 
707     _FDT(fdt_setprop_string(fdt, node, "device_type", "power-ivpe"));
708     _FDT(fdt_setprop(fdt, node, "reg", timas, sizeof(timas)));
709 
710     _FDT(fdt_setprop_string(fdt, node, "compatible", "ibm,power-ivpe"));
711     _FDT(fdt_setprop(fdt, node, "ibm,xive-eq-sizes", eq_sizes,
712                      sizeof(eq_sizes)));
713     _FDT(fdt_setprop(fdt, node, "ibm,xive-lisn-ranges", lisn_ranges,
714                      sizeof(lisn_ranges)));
715 
716     /* For Linux to link the LSIs to the interrupt controller. */
717     _FDT(fdt_setprop(fdt, node, "interrupt-controller", NULL, 0));
718     _FDT(fdt_setprop_cell(fdt, node, "#interrupt-cells", 2));
719 
720     /* For SLOF */
721     _FDT(fdt_setprop_cell(fdt, node, "linux,phandle", phandle));
722     _FDT(fdt_setprop_cell(fdt, node, "phandle", phandle));
723 
724     /*
725      * The "ibm,plat-res-int-priorities" property defines the priority
726      * ranges reserved by the hypervisor
727      */
728     _FDT(fdt_setprop(fdt, 0, "ibm,plat-res-int-priorities",
729                      plat_res_int_priorities, sizeof(plat_res_int_priorities)));
730 }
731 
732 static int spapr_xive_activate(SpaprInterruptController *intc,
733                                uint32_t nr_servers, Error **errp)
734 {
735     SpaprXive *xive = SPAPR_XIVE(intc);
736 
737     if (kvm_enabled()) {
738         int rc = spapr_irq_init_kvm(kvmppc_xive_connect, intc, nr_servers,
739                                     errp);
740         if (rc < 0) {
741             return rc;
742         }
743     }
744 
745     /* Activate the XIVE MMIOs */
746     spapr_xive_mmio_set_enabled(xive, true);
747 
748     return 0;
749 }
750 
751 static void spapr_xive_deactivate(SpaprInterruptController *intc)
752 {
753     SpaprXive *xive = SPAPR_XIVE(intc);
754 
755     spapr_xive_mmio_set_enabled(xive, false);
756 
757     if (kvm_irqchip_in_kernel()) {
758         kvmppc_xive_disconnect(intc);
759     }
760 }
761 
762 static void spapr_xive_class_init(ObjectClass *klass, void *data)
763 {
764     DeviceClass *dc = DEVICE_CLASS(klass);
765     XiveRouterClass *xrc = XIVE_ROUTER_CLASS(klass);
766     SpaprInterruptControllerClass *sicc = SPAPR_INTC_CLASS(klass);
767     XivePresenterClass *xpc = XIVE_PRESENTER_CLASS(klass);
768     SpaprXiveClass *sxc = SPAPR_XIVE_CLASS(klass);
769 
770     dc->desc    = "sPAPR XIVE Interrupt Controller";
771     device_class_set_props(dc, spapr_xive_properties);
772     device_class_set_parent_realize(dc, spapr_xive_realize,
773                                     &sxc->parent_realize);
774     dc->vmsd    = &vmstate_spapr_xive;
775 
776     xrc->get_eas = spapr_xive_get_eas;
777     xrc->get_end = spapr_xive_get_end;
778     xrc->write_end = spapr_xive_write_end;
779     xrc->get_nvt = spapr_xive_get_nvt;
780     xrc->write_nvt = spapr_xive_write_nvt;
781     xrc->get_block_id = spapr_xive_get_block_id;
782 
783     sicc->activate = spapr_xive_activate;
784     sicc->deactivate = spapr_xive_deactivate;
785     sicc->cpu_intc_create = spapr_xive_cpu_intc_create;
786     sicc->cpu_intc_reset = spapr_xive_cpu_intc_reset;
787     sicc->cpu_intc_destroy = spapr_xive_cpu_intc_destroy;
788     sicc->claim_irq = spapr_xive_claim_irq;
789     sicc->free_irq = spapr_xive_free_irq;
790     sicc->set_irq = spapr_xive_set_irq;
791     sicc->print_info = spapr_xive_print_info;
792     sicc->dt = spapr_xive_dt;
793     sicc->post_load = spapr_xive_post_load;
794 
795     xpc->match_nvt  = spapr_xive_match_nvt;
796 }
797 
798 static const TypeInfo spapr_xive_info = {
799     .name = TYPE_SPAPR_XIVE,
800     .parent = TYPE_XIVE_ROUTER,
801     .instance_init = spapr_xive_instance_init,
802     .instance_size = sizeof(SpaprXive),
803     .class_init = spapr_xive_class_init,
804     .class_size = sizeof(SpaprXiveClass),
805     .interfaces = (InterfaceInfo[]) {
806         { TYPE_SPAPR_INTC },
807         { }
808     },
809 };
810 
811 static void spapr_xive_register_types(void)
812 {
813     type_register_static(&spapr_xive_info);
814 }
815 
816 type_init(spapr_xive_register_types)
817 
818 /*
819  * XIVE hcalls
820  *
821  * The terminology used by the XIVE hcalls is the following :
822  *
823  *   TARGET vCPU number
824  *   EQ     Event Queue assigned by OS to receive event data
825  *   ESB    page for source interrupt management
826  *   LISN   Logical Interrupt Source Number identifying a source in the
827  *          machine
828  *   EISN   Effective Interrupt Source Number used by guest OS to
829  *          identify source in the guest
830  *
831  * The EAS, END, NVT structures are not exposed.
832  */
833 
834 /*
835  * Linux hosts under OPAL reserve priority 7 for their own escalation
836  * interrupts (DD2.X POWER9). So we only allow the guest to use
837  * priorities [0..6].
838  */
839 static bool spapr_xive_priority_is_reserved(uint8_t priority)
840 {
841     switch (priority) {
842     case 0 ... 6:
843         return false;
844     case 7: /* OPAL escalation queue */
845     default:
846         return true;
847     }
848 }
849 
850 /*
851  * The H_INT_GET_SOURCE_INFO hcall() is used to obtain the logical
852  * real address of the MMIO page through which the Event State Buffer
853  * entry associated with the value of the "lisn" parameter is managed.
854  *
855  * Parameters:
856  * Input
857  * - R4: "flags"
858  *         Bits 0-63 reserved
859  * - R5: "lisn" is per "interrupts", "interrupt-map", or
860  *       "ibm,xive-lisn-ranges" properties, or as returned by the
861  *       ibm,query-interrupt-source-number RTAS call, or as returned
862  *       by the H_ALLOCATE_VAS_WINDOW hcall
863  *
864  * Output
865  * - R4: "flags"
866  *         Bits 0-59: Reserved
867  *         Bit 60: H_INT_ESB must be used for Event State Buffer
868  *                 management
869  *         Bit 61: 1 == LSI  0 == MSI
870  *         Bit 62: the full function page supports trigger
871  *         Bit 63: Store EOI Supported
872  * - R5: Logical Real address of full function Event State Buffer
873  *       management page, -1 if H_INT_ESB hcall flag is set to 1.
874  * - R6: Logical Real Address of trigger only Event State Buffer
875  *       management page or -1.
876  * - R7: Power of 2 page size for the ESB management pages returned in
877  *       R5 and R6.
878  */
879 
880 #define SPAPR_XIVE_SRC_H_INT_ESB     PPC_BIT(60) /* ESB manage with H_INT_ESB */
881 #define SPAPR_XIVE_SRC_LSI           PPC_BIT(61) /* Virtual LSI type */
882 #define SPAPR_XIVE_SRC_TRIGGER       PPC_BIT(62) /* Trigger and management
883                                                     on same page */
884 #define SPAPR_XIVE_SRC_STORE_EOI     PPC_BIT(63) /* Store EOI support */
885 
886 static target_ulong h_int_get_source_info(PowerPCCPU *cpu,
887                                           SpaprMachineState *spapr,
888                                           target_ulong opcode,
889                                           target_ulong *args)
890 {
891     SpaprXive *xive = spapr->xive;
892     XiveSource *xsrc = &xive->source;
893     target_ulong flags  = args[0];
894     target_ulong lisn   = args[1];
895 
896     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
897         return H_FUNCTION;
898     }
899 
900     if (flags) {
901         return H_PARAMETER;
902     }
903 
904     if (lisn >= xive->nr_irqs) {
905         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
906                       lisn);
907         return H_P2;
908     }
909 
910     if (!xive_eas_is_valid(&xive->eat[lisn])) {
911         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
912                       lisn);
913         return H_P2;
914     }
915 
916     /*
917      * All sources are emulated under the main XIVE object and share
918      * the same characteristics.
919      */
920     args[0] = 0;
921     if (!xive_source_esb_has_2page(xsrc)) {
922         args[0] |= SPAPR_XIVE_SRC_TRIGGER;
923     }
924     if (xsrc->esb_flags & XIVE_SRC_STORE_EOI) {
925         args[0] |= SPAPR_XIVE_SRC_STORE_EOI;
926     }
927 
928     /*
929      * Force the use of the H_INT_ESB hcall in case of an LSI
930      * interrupt. This is necessary under KVM to re-trigger the
931      * interrupt if the level is still asserted
932      */
933     if (xive_source_irq_is_lsi(xsrc, lisn)) {
934         args[0] |= SPAPR_XIVE_SRC_H_INT_ESB | SPAPR_XIVE_SRC_LSI;
935     }
936 
937     if (!(args[0] & SPAPR_XIVE_SRC_H_INT_ESB)) {
938         args[1] = xive->vc_base + xive_source_esb_mgmt(xsrc, lisn);
939     } else {
940         args[1] = -1;
941     }
942 
943     if (xive_source_esb_has_2page(xsrc) &&
944         !(args[0] & SPAPR_XIVE_SRC_H_INT_ESB)) {
945         args[2] = xive->vc_base + xive_source_esb_page(xsrc, lisn);
946     } else {
947         args[2] = -1;
948     }
949 
950     if (xive_source_esb_has_2page(xsrc)) {
951         args[3] = xsrc->esb_shift - 1;
952     } else {
953         args[3] = xsrc->esb_shift;
954     }
955 
956     return H_SUCCESS;
957 }
958 
959 /*
960  * The H_INT_SET_SOURCE_CONFIG hcall() is used to assign a Logical
961  * Interrupt Source to a target. The Logical Interrupt Source is
962  * designated with the "lisn" parameter and the target is designated
963  * with the "target" and "priority" parameters.  Upon return from the
964  * hcall(), no additional interrupts will be directed to the old EQ.
965  *
966  * Parameters:
967  * Input:
968  * - R4: "flags"
969  *         Bits 0-61: Reserved
970  *         Bit 62: set the "eisn" in the EAS
971  *         Bit 63: masks the interrupt source in the hardware interrupt
972  *       control structure. An interrupt masked by this mechanism will
973  *       be dropped, but it's source state bits will still be
974  *       set. There is no race-free way of unmasking and restoring the
975  *       source. Thus this should only be used in interrupts that are
976  *       also masked at the source, and only in cases where the
977  *       interrupt is not meant to be used for a large amount of time
978  *       because no valid target exists for it for example
979  * - R5: "lisn" is per "interrupts", "interrupt-map", or
980  *       "ibm,xive-lisn-ranges" properties, or as returned by the
981  *       ibm,query-interrupt-source-number RTAS call, or as returned by
982  *       the H_ALLOCATE_VAS_WINDOW hcall
983  * - R6: "target" is per "ibm,ppc-interrupt-server#s" or
984  *       "ibm,ppc-interrupt-gserver#s"
985  * - R7: "priority" is a valid priority not in
986  *       "ibm,plat-res-int-priorities"
987  * - R8: "eisn" is the guest EISN associated with the "lisn"
988  *
989  * Output:
990  * - None
991  */
992 
993 #define SPAPR_XIVE_SRC_SET_EISN PPC_BIT(62)
994 #define SPAPR_XIVE_SRC_MASK     PPC_BIT(63)
995 
996 static target_ulong h_int_set_source_config(PowerPCCPU *cpu,
997                                             SpaprMachineState *spapr,
998                                             target_ulong opcode,
999                                             target_ulong *args)
1000 {
1001     SpaprXive *xive = spapr->xive;
1002     XiveEAS eas, new_eas;
1003     target_ulong flags    = args[0];
1004     target_ulong lisn     = args[1];
1005     target_ulong target   = args[2];
1006     target_ulong priority = args[3];
1007     target_ulong eisn     = args[4];
1008     uint8_t end_blk;
1009     uint32_t end_idx;
1010 
1011     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1012         return H_FUNCTION;
1013     }
1014 
1015     if (flags & ~(SPAPR_XIVE_SRC_SET_EISN | SPAPR_XIVE_SRC_MASK)) {
1016         return H_PARAMETER;
1017     }
1018 
1019     if (lisn >= xive->nr_irqs) {
1020         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1021                       lisn);
1022         return H_P2;
1023     }
1024 
1025     eas = xive->eat[lisn];
1026     if (!xive_eas_is_valid(&eas)) {
1027         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1028                       lisn);
1029         return H_P2;
1030     }
1031 
1032     /* priority 0xff is used to reset the EAS */
1033     if (priority == 0xff) {
1034         new_eas.w = cpu_to_be64(EAS_VALID | EAS_MASKED);
1035         goto out;
1036     }
1037 
1038     if (flags & SPAPR_XIVE_SRC_MASK) {
1039         new_eas.w = eas.w | cpu_to_be64(EAS_MASKED);
1040     } else {
1041         new_eas.w = eas.w & cpu_to_be64(~EAS_MASKED);
1042     }
1043 
1044     if (spapr_xive_priority_is_reserved(priority)) {
1045         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1046                       " is reserved\n", priority);
1047         return H_P4;
1048     }
1049 
1050     /*
1051      * Validate that "target" is part of the list of threads allocated
1052      * to the partition. For that, find the END corresponding to the
1053      * target.
1054      */
1055     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1056         return H_P3;
1057     }
1058 
1059     new_eas.w = xive_set_field64(EAS_END_BLOCK, new_eas.w, end_blk);
1060     new_eas.w = xive_set_field64(EAS_END_INDEX, new_eas.w, end_idx);
1061 
1062     if (flags & SPAPR_XIVE_SRC_SET_EISN) {
1063         new_eas.w = xive_set_field64(EAS_END_DATA, new_eas.w, eisn);
1064     }
1065 
1066     if (kvm_irqchip_in_kernel()) {
1067         Error *local_err = NULL;
1068 
1069         kvmppc_xive_set_source_config(xive, lisn, &new_eas, &local_err);
1070         if (local_err) {
1071             error_report_err(local_err);
1072             return H_HARDWARE;
1073         }
1074     }
1075 
1076 out:
1077     xive->eat[lisn] = new_eas;
1078     return H_SUCCESS;
1079 }
1080 
1081 /*
1082  * The H_INT_GET_SOURCE_CONFIG hcall() is used to determine to which
1083  * target/priority pair is assigned to the specified Logical Interrupt
1084  * Source.
1085  *
1086  * Parameters:
1087  * Input:
1088  * - R4: "flags"
1089  *         Bits 0-63 Reserved
1090  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1091  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1092  *       ibm,query-interrupt-source-number RTAS call, or as
1093  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1094  *
1095  * Output:
1096  * - R4: Target to which the specified Logical Interrupt Source is
1097  *       assigned
1098  * - R5: Priority to which the specified Logical Interrupt Source is
1099  *       assigned
1100  * - R6: EISN for the specified Logical Interrupt Source (this will be
1101  *       equivalent to the LISN if not changed by H_INT_SET_SOURCE_CONFIG)
1102  */
1103 static target_ulong h_int_get_source_config(PowerPCCPU *cpu,
1104                                             SpaprMachineState *spapr,
1105                                             target_ulong opcode,
1106                                             target_ulong *args)
1107 {
1108     SpaprXive *xive = spapr->xive;
1109     target_ulong flags = args[0];
1110     target_ulong lisn = args[1];
1111     XiveEAS eas;
1112     XiveEND *end;
1113     uint8_t nvt_blk;
1114     uint32_t end_idx, nvt_idx;
1115 
1116     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1117         return H_FUNCTION;
1118     }
1119 
1120     if (flags) {
1121         return H_PARAMETER;
1122     }
1123 
1124     if (lisn >= xive->nr_irqs) {
1125         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1126                       lisn);
1127         return H_P2;
1128     }
1129 
1130     eas = xive->eat[lisn];
1131     if (!xive_eas_is_valid(&eas)) {
1132         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1133                       lisn);
1134         return H_P2;
1135     }
1136 
1137     /* EAS_END_BLOCK is unused on sPAPR */
1138     end_idx = xive_get_field64(EAS_END_INDEX, eas.w);
1139 
1140     assert(end_idx < xive->nr_ends);
1141     end = &xive->endt[end_idx];
1142 
1143     nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end->w6);
1144     nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end->w6);
1145     args[0] = spapr_xive_nvt_to_target(nvt_blk, nvt_idx);
1146 
1147     if (xive_eas_is_masked(&eas)) {
1148         args[1] = 0xff;
1149     } else {
1150         args[1] = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
1151     }
1152 
1153     args[2] = xive_get_field64(EAS_END_DATA, eas.w);
1154 
1155     return H_SUCCESS;
1156 }
1157 
1158 /*
1159  * The H_INT_GET_QUEUE_INFO hcall() is used to get the logical real
1160  * address of the notification management page associated with the
1161  * specified target and priority.
1162  *
1163  * Parameters:
1164  * Input:
1165  * - R4: "flags"
1166  *         Bits 0-63 Reserved
1167  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1168  *       "ibm,ppc-interrupt-gserver#s"
1169  * - R6: "priority" is a valid priority not in
1170  *       "ibm,plat-res-int-priorities"
1171  *
1172  * Output:
1173  * - R4: Logical real address of notification page
1174  * - R5: Power of 2 page size of the notification page
1175  */
1176 static target_ulong h_int_get_queue_info(PowerPCCPU *cpu,
1177                                          SpaprMachineState *spapr,
1178                                          target_ulong opcode,
1179                                          target_ulong *args)
1180 {
1181     SpaprXive *xive = spapr->xive;
1182     XiveENDSource *end_xsrc = &xive->end_source;
1183     target_ulong flags = args[0];
1184     target_ulong target = args[1];
1185     target_ulong priority = args[2];
1186     XiveEND *end;
1187     uint8_t end_blk;
1188     uint32_t end_idx;
1189 
1190     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1191         return H_FUNCTION;
1192     }
1193 
1194     if (flags) {
1195         return H_PARAMETER;
1196     }
1197 
1198     /*
1199      * H_STATE should be returned if a H_INT_RESET is in progress.
1200      * This is not needed when running the emulation under QEMU
1201      */
1202 
1203     if (spapr_xive_priority_is_reserved(priority)) {
1204         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1205                       " is reserved\n", priority);
1206         return H_P3;
1207     }
1208 
1209     /*
1210      * Validate that "target" is part of the list of threads allocated
1211      * to the partition. For that, find the END corresponding to the
1212      * target.
1213      */
1214     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1215         return H_P2;
1216     }
1217 
1218     assert(end_idx < xive->nr_ends);
1219     end = &xive->endt[end_idx];
1220 
1221     args[0] = xive->end_base + (1ull << (end_xsrc->esb_shift + 1)) * end_idx;
1222     if (xive_end_is_enqueue(end)) {
1223         args[1] = xive_get_field32(END_W0_QSIZE, end->w0) + 12;
1224     } else {
1225         args[1] = 0;
1226     }
1227 
1228     return H_SUCCESS;
1229 }
1230 
1231 /*
1232  * The H_INT_SET_QUEUE_CONFIG hcall() is used to set or reset a EQ for
1233  * a given "target" and "priority".  It is also used to set the
1234  * notification config associated with the EQ.  An EQ size of 0 is
1235  * used to reset the EQ config for a given target and priority. If
1236  * resetting the EQ config, the END associated with the given "target"
1237  * and "priority" will be changed to disable queueing.
1238  *
1239  * Upon return from the hcall(), no additional interrupts will be
1240  * directed to the old EQ (if one was set). The old EQ (if one was
1241  * set) should be investigated for interrupts that occurred prior to
1242  * or during the hcall().
1243  *
1244  * Parameters:
1245  * Input:
1246  * - R4: "flags"
1247  *         Bits 0-62: Reserved
1248  *         Bit 63: Unconditional Notify (n) per the XIVE spec
1249  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1250  *       "ibm,ppc-interrupt-gserver#s"
1251  * - R6: "priority" is a valid priority not in
1252  *       "ibm,plat-res-int-priorities"
1253  * - R7: "eventQueue": The logical real address of the start of the EQ
1254  * - R8: "eventQueueSize": The power of 2 EQ size per "ibm,xive-eq-sizes"
1255  *
1256  * Output:
1257  * - None
1258  */
1259 
1260 #define SPAPR_XIVE_END_ALWAYS_NOTIFY PPC_BIT(63)
1261 
1262 static target_ulong h_int_set_queue_config(PowerPCCPU *cpu,
1263                                            SpaprMachineState *spapr,
1264                                            target_ulong opcode,
1265                                            target_ulong *args)
1266 {
1267     SpaprXive *xive = spapr->xive;
1268     target_ulong flags = args[0];
1269     target_ulong target = args[1];
1270     target_ulong priority = args[2];
1271     target_ulong qpage = args[3];
1272     target_ulong qsize = args[4];
1273     XiveEND end;
1274     uint8_t end_blk, nvt_blk;
1275     uint32_t end_idx, nvt_idx;
1276 
1277     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1278         return H_FUNCTION;
1279     }
1280 
1281     if (flags & ~SPAPR_XIVE_END_ALWAYS_NOTIFY) {
1282         return H_PARAMETER;
1283     }
1284 
1285     /*
1286      * H_STATE should be returned if a H_INT_RESET is in progress.
1287      * This is not needed when running the emulation under QEMU
1288      */
1289 
1290     if (spapr_xive_priority_is_reserved(priority)) {
1291         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1292                       " is reserved\n", priority);
1293         return H_P3;
1294     }
1295 
1296     /*
1297      * Validate that "target" is part of the list of threads allocated
1298      * to the partition. For that, find the END corresponding to the
1299      * target.
1300      */
1301 
1302     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1303         return H_P2;
1304     }
1305 
1306     assert(end_idx < xive->nr_ends);
1307     memcpy(&end, &xive->endt[end_idx], sizeof(XiveEND));
1308 
1309     switch (qsize) {
1310     case 12:
1311     case 16:
1312     case 21:
1313     case 24:
1314         if (!QEMU_IS_ALIGNED(qpage, 1ul << qsize)) {
1315             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: EQ @0x%" HWADDR_PRIx
1316                           " is not naturally aligned with %" HWADDR_PRIx "\n",
1317                           qpage, (hwaddr)1 << qsize);
1318             return H_P4;
1319         }
1320         end.w2 = cpu_to_be32((qpage >> 32) & 0x0fffffff);
1321         end.w3 = cpu_to_be32(qpage & 0xffffffff);
1322         end.w0 |= cpu_to_be32(END_W0_ENQUEUE);
1323         end.w0 = xive_set_field32(END_W0_QSIZE, end.w0, qsize - 12);
1324         break;
1325     case 0:
1326         /* reset queue and disable queueing */
1327         spapr_xive_end_reset(&end);
1328         goto out;
1329 
1330     default:
1331         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid EQ size %"PRIx64"\n",
1332                       qsize);
1333         return H_P5;
1334     }
1335 
1336     if (qsize) {
1337         hwaddr plen = 1 << qsize;
1338         void *eq;
1339 
1340         /*
1341          * Validate the guest EQ. We should also check that the queue
1342          * has been zeroed by the OS.
1343          */
1344         eq = address_space_map(CPU(cpu)->as, qpage, &plen, true,
1345                                MEMTXATTRS_UNSPECIFIED);
1346         if (plen != 1 << qsize) {
1347             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to map EQ @0x%"
1348                           HWADDR_PRIx "\n", qpage);
1349             return H_P4;
1350         }
1351         address_space_unmap(CPU(cpu)->as, eq, plen, true, plen);
1352     }
1353 
1354     /* "target" should have been validated above */
1355     if (spapr_xive_target_to_nvt(target, &nvt_blk, &nvt_idx)) {
1356         g_assert_not_reached();
1357     }
1358 
1359     /*
1360      * Ensure the priority and target are correctly set (they will not
1361      * be right after allocation)
1362      */
1363     end.w6 = xive_set_field32(END_W6_NVT_BLOCK, 0ul, nvt_blk) |
1364         xive_set_field32(END_W6_NVT_INDEX, 0ul, nvt_idx);
1365     end.w7 = xive_set_field32(END_W7_F0_PRIORITY, 0ul, priority);
1366 
1367     if (flags & SPAPR_XIVE_END_ALWAYS_NOTIFY) {
1368         end.w0 |= cpu_to_be32(END_W0_UCOND_NOTIFY);
1369     } else {
1370         end.w0 &= cpu_to_be32((uint32_t)~END_W0_UCOND_NOTIFY);
1371     }
1372 
1373     /*
1374      * The generation bit for the END starts at 1 and The END page
1375      * offset counter starts at 0.
1376      */
1377     end.w1 = cpu_to_be32(END_W1_GENERATION) |
1378         xive_set_field32(END_W1_PAGE_OFF, 0ul, 0ul);
1379     end.w0 |= cpu_to_be32(END_W0_VALID);
1380 
1381     /*
1382      * TODO: issue syncs required to ensure all in-flight interrupts
1383      * are complete on the old END
1384      */
1385 
1386 out:
1387     if (kvm_irqchip_in_kernel()) {
1388         Error *local_err = NULL;
1389 
1390         kvmppc_xive_set_queue_config(xive, end_blk, end_idx, &end, &local_err);
1391         if (local_err) {
1392             error_report_err(local_err);
1393             return H_HARDWARE;
1394         }
1395     }
1396 
1397     /* Update END */
1398     memcpy(&xive->endt[end_idx], &end, sizeof(XiveEND));
1399     return H_SUCCESS;
1400 }
1401 
1402 /*
1403  * The H_INT_GET_QUEUE_CONFIG hcall() is used to get a EQ for a given
1404  * target and priority.
1405  *
1406  * Parameters:
1407  * Input:
1408  * - R4: "flags"
1409  *         Bits 0-62: Reserved
1410  *         Bit 63: Debug: Return debug data
1411  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1412  *       "ibm,ppc-interrupt-gserver#s"
1413  * - R6: "priority" is a valid priority not in
1414  *       "ibm,plat-res-int-priorities"
1415  *
1416  * Output:
1417  * - R4: "flags":
1418  *       Bits 0-61: Reserved
1419  *       Bit 62: The value of Event Queue Generation Number (g) per
1420  *              the XIVE spec if "Debug" = 1
1421  *       Bit 63: The value of Unconditional Notify (n) per the XIVE spec
1422  * - R5: The logical real address of the start of the EQ
1423  * - R6: The power of 2 EQ size per "ibm,xive-eq-sizes"
1424  * - R7: The value of Event Queue Offset Counter per XIVE spec
1425  *       if "Debug" = 1, else 0
1426  *
1427  */
1428 
1429 #define SPAPR_XIVE_END_DEBUG     PPC_BIT(63)
1430 
1431 static target_ulong h_int_get_queue_config(PowerPCCPU *cpu,
1432                                            SpaprMachineState *spapr,
1433                                            target_ulong opcode,
1434                                            target_ulong *args)
1435 {
1436     SpaprXive *xive = spapr->xive;
1437     target_ulong flags = args[0];
1438     target_ulong target = args[1];
1439     target_ulong priority = args[2];
1440     XiveEND *end;
1441     uint8_t end_blk;
1442     uint32_t end_idx;
1443 
1444     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1445         return H_FUNCTION;
1446     }
1447 
1448     if (flags & ~SPAPR_XIVE_END_DEBUG) {
1449         return H_PARAMETER;
1450     }
1451 
1452     /*
1453      * H_STATE should be returned if a H_INT_RESET is in progress.
1454      * This is not needed when running the emulation under QEMU
1455      */
1456 
1457     if (spapr_xive_priority_is_reserved(priority)) {
1458         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1459                       " is reserved\n", priority);
1460         return H_P3;
1461     }
1462 
1463     /*
1464      * Validate that "target" is part of the list of threads allocated
1465      * to the partition. For that, find the END corresponding to the
1466      * target.
1467      */
1468     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1469         return H_P2;
1470     }
1471 
1472     assert(end_idx < xive->nr_ends);
1473     end = &xive->endt[end_idx];
1474 
1475     args[0] = 0;
1476     if (xive_end_is_notify(end)) {
1477         args[0] |= SPAPR_XIVE_END_ALWAYS_NOTIFY;
1478     }
1479 
1480     if (xive_end_is_enqueue(end)) {
1481         args[1] = xive_end_qaddr(end);
1482         args[2] = xive_get_field32(END_W0_QSIZE, end->w0) + 12;
1483     } else {
1484         args[1] = 0;
1485         args[2] = 0;
1486     }
1487 
1488     if (kvm_irqchip_in_kernel()) {
1489         Error *local_err = NULL;
1490 
1491         kvmppc_xive_get_queue_config(xive, end_blk, end_idx, end, &local_err);
1492         if (local_err) {
1493             error_report_err(local_err);
1494             return H_HARDWARE;
1495         }
1496     }
1497 
1498     /* TODO: do we need any locking on the END ? */
1499     if (flags & SPAPR_XIVE_END_DEBUG) {
1500         /* Load the event queue generation number into the return flags */
1501         args[0] |= (uint64_t)xive_get_field32(END_W1_GENERATION, end->w1) << 62;
1502 
1503         /* Load R7 with the event queue offset counter */
1504         args[3] = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1505     } else {
1506         args[3] = 0;
1507     }
1508 
1509     return H_SUCCESS;
1510 }
1511 
1512 /*
1513  * The H_INT_SET_OS_REPORTING_LINE hcall() is used to set the
1514  * reporting cache line pair for the calling thread.  The reporting
1515  * cache lines will contain the OS interrupt context when the OS
1516  * issues a CI store byte to @TIMA+0xC10 to acknowledge the OS
1517  * interrupt. The reporting cache lines can be reset by inputting -1
1518  * in "reportingLine".  Issuing the CI store byte without reporting
1519  * cache lines registered will result in the data not being accessible
1520  * to the OS.
1521  *
1522  * Parameters:
1523  * Input:
1524  * - R4: "flags"
1525  *         Bits 0-63: Reserved
1526  * - R5: "reportingLine": The logical real address of the reporting cache
1527  *       line pair
1528  *
1529  * Output:
1530  * - None
1531  */
1532 static target_ulong h_int_set_os_reporting_line(PowerPCCPU *cpu,
1533                                                 SpaprMachineState *spapr,
1534                                                 target_ulong opcode,
1535                                                 target_ulong *args)
1536 {
1537     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1538         return H_FUNCTION;
1539     }
1540 
1541     /*
1542      * H_STATE should be returned if a H_INT_RESET is in progress.
1543      * This is not needed when running the emulation under QEMU
1544      */
1545 
1546     /* TODO: H_INT_SET_OS_REPORTING_LINE */
1547     return H_FUNCTION;
1548 }
1549 
1550 /*
1551  * The H_INT_GET_OS_REPORTING_LINE hcall() is used to get the logical
1552  * real address of the reporting cache line pair set for the input
1553  * "target".  If no reporting cache line pair has been set, -1 is
1554  * returned.
1555  *
1556  * Parameters:
1557  * Input:
1558  * - R4: "flags"
1559  *         Bits 0-63: Reserved
1560  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1561  *       "ibm,ppc-interrupt-gserver#s"
1562  * - R6: "reportingLine": The logical real address of the reporting
1563  *        cache line pair
1564  *
1565  * Output:
1566  * - R4: The logical real address of the reporting line if set, else -1
1567  */
1568 static target_ulong h_int_get_os_reporting_line(PowerPCCPU *cpu,
1569                                                 SpaprMachineState *spapr,
1570                                                 target_ulong opcode,
1571                                                 target_ulong *args)
1572 {
1573     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1574         return H_FUNCTION;
1575     }
1576 
1577     /*
1578      * H_STATE should be returned if a H_INT_RESET is in progress.
1579      * This is not needed when running the emulation under QEMU
1580      */
1581 
1582     /* TODO: H_INT_GET_OS_REPORTING_LINE */
1583     return H_FUNCTION;
1584 }
1585 
1586 /*
1587  * The H_INT_ESB hcall() is used to issue a load or store to the ESB
1588  * page for the input "lisn".  This hcall is only supported for LISNs
1589  * that have the ESB hcall flag set to 1 when returned from hcall()
1590  * H_INT_GET_SOURCE_INFO.
1591  *
1592  * Parameters:
1593  * Input:
1594  * - R4: "flags"
1595  *         Bits 0-62: Reserved
1596  *         bit 63: Store: Store=1, store operation, else load operation
1597  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1598  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1599  *       ibm,query-interrupt-source-number RTAS call, or as
1600  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1601  * - R6: "esbOffset" is the offset into the ESB page for the load or
1602  *       store operation
1603  * - R7: "storeData" is the data to write for a store operation
1604  *
1605  * Output:
1606  * - R4: The value of the load if load operation, else -1
1607  */
1608 
1609 #define SPAPR_XIVE_ESB_STORE PPC_BIT(63)
1610 
1611 static target_ulong h_int_esb(PowerPCCPU *cpu,
1612                               SpaprMachineState *spapr,
1613                               target_ulong opcode,
1614                               target_ulong *args)
1615 {
1616     SpaprXive *xive = spapr->xive;
1617     XiveEAS eas;
1618     target_ulong flags  = args[0];
1619     target_ulong lisn   = args[1];
1620     target_ulong offset = args[2];
1621     target_ulong data   = args[3];
1622     hwaddr mmio_addr;
1623     XiveSource *xsrc = &xive->source;
1624 
1625     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1626         return H_FUNCTION;
1627     }
1628 
1629     if (flags & ~SPAPR_XIVE_ESB_STORE) {
1630         return H_PARAMETER;
1631     }
1632 
1633     if (lisn >= xive->nr_irqs) {
1634         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1635                       lisn);
1636         return H_P2;
1637     }
1638 
1639     eas = xive->eat[lisn];
1640     if (!xive_eas_is_valid(&eas)) {
1641         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1642                       lisn);
1643         return H_P2;
1644     }
1645 
1646     if (offset > (1ull << xsrc->esb_shift)) {
1647         return H_P3;
1648     }
1649 
1650     if (kvm_irqchip_in_kernel()) {
1651         args[0] = kvmppc_xive_esb_rw(xsrc, lisn, offset, data,
1652                                      flags & SPAPR_XIVE_ESB_STORE);
1653     } else {
1654         mmio_addr = xive->vc_base + xive_source_esb_mgmt(xsrc, lisn) + offset;
1655 
1656         if (dma_memory_rw(&address_space_memory, mmio_addr, &data, 8,
1657                           (flags & SPAPR_XIVE_ESB_STORE))) {
1658             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to access ESB @0x%"
1659                           HWADDR_PRIx "\n", mmio_addr);
1660             return H_HARDWARE;
1661         }
1662         args[0] = (flags & SPAPR_XIVE_ESB_STORE) ? -1 : data;
1663     }
1664     return H_SUCCESS;
1665 }
1666 
1667 /*
1668  * The H_INT_SYNC hcall() is used to issue hardware syncs that will
1669  * ensure any in flight events for the input lisn are in the event
1670  * queue.
1671  *
1672  * Parameters:
1673  * Input:
1674  * - R4: "flags"
1675  *         Bits 0-63: Reserved
1676  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1677  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1678  *       ibm,query-interrupt-source-number RTAS call, or as
1679  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1680  *
1681  * Output:
1682  * - None
1683  */
1684 static target_ulong h_int_sync(PowerPCCPU *cpu,
1685                                SpaprMachineState *spapr,
1686                                target_ulong opcode,
1687                                target_ulong *args)
1688 {
1689     SpaprXive *xive = spapr->xive;
1690     XiveEAS eas;
1691     target_ulong flags = args[0];
1692     target_ulong lisn = args[1];
1693 
1694     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1695         return H_FUNCTION;
1696     }
1697 
1698     if (flags) {
1699         return H_PARAMETER;
1700     }
1701 
1702     if (lisn >= xive->nr_irqs) {
1703         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1704                       lisn);
1705         return H_P2;
1706     }
1707 
1708     eas = xive->eat[lisn];
1709     if (!xive_eas_is_valid(&eas)) {
1710         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1711                       lisn);
1712         return H_P2;
1713     }
1714 
1715     /*
1716      * H_STATE should be returned if a H_INT_RESET is in progress.
1717      * This is not needed when running the emulation under QEMU
1718      */
1719 
1720     /*
1721      * This is not real hardware. Nothing to be done unless when
1722      * under KVM
1723      */
1724 
1725     if (kvm_irqchip_in_kernel()) {
1726         Error *local_err = NULL;
1727 
1728         kvmppc_xive_sync_source(xive, lisn, &local_err);
1729         if (local_err) {
1730             error_report_err(local_err);
1731             return H_HARDWARE;
1732         }
1733     }
1734     return H_SUCCESS;
1735 }
1736 
1737 /*
1738  * The H_INT_RESET hcall() is used to reset all of the partition's
1739  * interrupt exploitation structures to their initial state.  This
1740  * means losing all previously set interrupt state set via
1741  * H_INT_SET_SOURCE_CONFIG and H_INT_SET_QUEUE_CONFIG.
1742  *
1743  * Parameters:
1744  * Input:
1745  * - R4: "flags"
1746  *         Bits 0-63: Reserved
1747  *
1748  * Output:
1749  * - None
1750  */
1751 static target_ulong h_int_reset(PowerPCCPU *cpu,
1752                                 SpaprMachineState *spapr,
1753                                 target_ulong opcode,
1754                                 target_ulong *args)
1755 {
1756     SpaprXive *xive = spapr->xive;
1757     target_ulong flags   = args[0];
1758 
1759     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1760         return H_FUNCTION;
1761     }
1762 
1763     if (flags) {
1764         return H_PARAMETER;
1765     }
1766 
1767     device_legacy_reset(DEVICE(xive));
1768 
1769     if (kvm_irqchip_in_kernel()) {
1770         Error *local_err = NULL;
1771 
1772         kvmppc_xive_reset(xive, &local_err);
1773         if (local_err) {
1774             error_report_err(local_err);
1775             return H_HARDWARE;
1776         }
1777     }
1778     return H_SUCCESS;
1779 }
1780 
1781 void spapr_xive_hcall_init(SpaprMachineState *spapr)
1782 {
1783     spapr_register_hypercall(H_INT_GET_SOURCE_INFO, h_int_get_source_info);
1784     spapr_register_hypercall(H_INT_SET_SOURCE_CONFIG, h_int_set_source_config);
1785     spapr_register_hypercall(H_INT_GET_SOURCE_CONFIG, h_int_get_source_config);
1786     spapr_register_hypercall(H_INT_GET_QUEUE_INFO, h_int_get_queue_info);
1787     spapr_register_hypercall(H_INT_SET_QUEUE_CONFIG, h_int_set_queue_config);
1788     spapr_register_hypercall(H_INT_GET_QUEUE_CONFIG, h_int_get_queue_config);
1789     spapr_register_hypercall(H_INT_SET_OS_REPORTING_LINE,
1790                              h_int_set_os_reporting_line);
1791     spapr_register_hypercall(H_INT_GET_OS_REPORTING_LINE,
1792                              h_int_get_os_reporting_line);
1793     spapr_register_hypercall(H_INT_ESB, h_int_esb);
1794     spapr_register_hypercall(H_INT_SYNC, h_int_sync);
1795     spapr_register_hypercall(H_INT_RESET, h_int_reset);
1796 }
1797