xref: /openbmc/qemu/hw/intc/spapr_xive.c (revision d49e8a9b46e4594223806ae622af462ff7bfa158)
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 /*
209  * When a Virtual Processor is scheduled to run on a HW thread, the
210  * hypervisor pushes its identifier in the OS CAM line. Emulate the
211  * same behavior under QEMU.
212  */
213 void spapr_xive_set_tctx_os_cam(XiveTCTX *tctx)
214 {
215     uint8_t  nvt_blk;
216     uint32_t nvt_idx;
217     uint32_t nvt_cam;
218 
219     spapr_xive_cpu_to_nvt(POWERPC_CPU(tctx->cs), &nvt_blk, &nvt_idx);
220 
221     nvt_cam = cpu_to_be32(TM_QW1W2_VO | xive_nvt_cam_line(nvt_blk, nvt_idx));
222     memcpy(&tctx->regs[TM_QW1_OS + TM_WORD2], &nvt_cam, 4);
223 }
224 
225 static void spapr_xive_end_reset(XiveEND *end)
226 {
227     memset(end, 0, sizeof(*end));
228 
229     /* switch off the escalation and notification ESBs */
230     end->w1 = cpu_to_be32(END_W1_ESe_Q | END_W1_ESn_Q);
231 }
232 
233 static void spapr_xive_reset(void *dev)
234 {
235     SpaprXive *xive = SPAPR_XIVE(dev);
236     int i;
237 
238     /*
239      * The XiveSource has its own reset handler, which mask off all
240      * IRQs (!P|Q)
241      */
242 
243     /* Mask all valid EASs in the IRQ number space. */
244     for (i = 0; i < xive->nr_irqs; i++) {
245         XiveEAS *eas = &xive->eat[i];
246         if (xive_eas_is_valid(eas)) {
247             eas->w = cpu_to_be64(EAS_VALID | EAS_MASKED);
248         } else {
249             eas->w = 0;
250         }
251     }
252 
253     /* Clear all ENDs */
254     for (i = 0; i < xive->nr_ends; i++) {
255         spapr_xive_end_reset(&xive->endt[i]);
256     }
257 }
258 
259 static void spapr_xive_instance_init(Object *obj)
260 {
261     SpaprXive *xive = SPAPR_XIVE(obj);
262 
263     object_initialize_child(obj, "source", &xive->source, sizeof(xive->source),
264                             TYPE_XIVE_SOURCE, &error_abort, NULL);
265 
266     object_initialize_child(obj, "end_source", &xive->end_source,
267                             sizeof(xive->end_source), TYPE_XIVE_END_SOURCE,
268                             &error_abort, NULL);
269 
270     /* Not connected to the KVM XIVE device */
271     xive->fd = -1;
272 }
273 
274 static void spapr_xive_realize(DeviceState *dev, Error **errp)
275 {
276     SpaprXive *xive = SPAPR_XIVE(dev);
277     XiveSource *xsrc = &xive->source;
278     XiveENDSource *end_xsrc = &xive->end_source;
279     Error *local_err = NULL;
280 
281     if (!xive->nr_irqs) {
282         error_setg(errp, "Number of interrupt needs to be greater 0");
283         return;
284     }
285 
286     if (!xive->nr_ends) {
287         error_setg(errp, "Number of interrupt needs to be greater 0");
288         return;
289     }
290 
291     /*
292      * Initialize the internal sources, for IPIs and virtual devices.
293      */
294     object_property_set_int(OBJECT(xsrc), xive->nr_irqs, "nr-irqs",
295                             &error_fatal);
296     object_property_add_const_link(OBJECT(xsrc), "xive", OBJECT(xive),
297                                    &error_fatal);
298     object_property_set_bool(OBJECT(xsrc), true, "realized", &local_err);
299     if (local_err) {
300         error_propagate(errp, local_err);
301         return;
302     }
303     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &xsrc->esb_mmio);
304 
305     /*
306      * Initialize the END ESB source
307      */
308     object_property_set_int(OBJECT(end_xsrc), xive->nr_irqs, "nr-ends",
309                             &error_fatal);
310     object_property_add_const_link(OBJECT(end_xsrc), "xive", OBJECT(xive),
311                                    &error_fatal);
312     object_property_set_bool(OBJECT(end_xsrc), true, "realized", &local_err);
313     if (local_err) {
314         error_propagate(errp, local_err);
315         return;
316     }
317     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &end_xsrc->esb_mmio);
318 
319     /* Set the mapping address of the END ESB pages after the source ESBs */
320     xive->end_base = xive->vc_base + (1ull << xsrc->esb_shift) * xsrc->nr_irqs;
321 
322     /*
323      * Allocate the routing tables
324      */
325     xive->eat = g_new0(XiveEAS, xive->nr_irqs);
326     xive->endt = g_new0(XiveEND, xive->nr_ends);
327 
328     xive->nodename = g_strdup_printf("interrupt-controller@%" PRIx64,
329                            xive->tm_base + XIVE_TM_USER_PAGE * (1 << TM_SHIFT));
330 
331     qemu_register_reset(spapr_xive_reset, dev);
332 
333     /* TIMA initialization */
334     memory_region_init_io(&xive->tm_mmio, OBJECT(xive), &xive_tm_ops, xive,
335                           "xive.tima", 4ull << TM_SHIFT);
336     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &xive->tm_mmio);
337 
338     /*
339      * Map all regions. These will be enabled or disabled at reset and
340      * can also be overridden by KVM memory regions if active
341      */
342     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 0, xive->vc_base);
343     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 1, xive->end_base);
344     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 2, xive->tm_base);
345 }
346 
347 static int spapr_xive_get_eas(XiveRouter *xrtr, uint8_t eas_blk,
348                               uint32_t eas_idx, XiveEAS *eas)
349 {
350     SpaprXive *xive = SPAPR_XIVE(xrtr);
351 
352     if (eas_idx >= xive->nr_irqs) {
353         return -1;
354     }
355 
356     *eas = xive->eat[eas_idx];
357     return 0;
358 }
359 
360 static int spapr_xive_get_end(XiveRouter *xrtr,
361                               uint8_t end_blk, uint32_t end_idx, XiveEND *end)
362 {
363     SpaprXive *xive = SPAPR_XIVE(xrtr);
364 
365     if (end_idx >= xive->nr_ends) {
366         return -1;
367     }
368 
369     memcpy(end, &xive->endt[end_idx], sizeof(XiveEND));
370     return 0;
371 }
372 
373 static int spapr_xive_write_end(XiveRouter *xrtr, uint8_t end_blk,
374                                 uint32_t end_idx, XiveEND *end,
375                                 uint8_t word_number)
376 {
377     SpaprXive *xive = SPAPR_XIVE(xrtr);
378 
379     if (end_idx >= xive->nr_ends) {
380         return -1;
381     }
382 
383     memcpy(&xive->endt[end_idx], end, sizeof(XiveEND));
384     return 0;
385 }
386 
387 static int spapr_xive_get_nvt(XiveRouter *xrtr,
388                               uint8_t nvt_blk, uint32_t nvt_idx, XiveNVT *nvt)
389 {
390     uint32_t vcpu_id = spapr_xive_nvt_to_target(nvt_blk, nvt_idx);
391     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
392 
393     if (!cpu) {
394         /* TODO: should we assert() if we can find a NVT ? */
395         return -1;
396     }
397 
398     /*
399      * sPAPR does not maintain a NVT table. Return that the NVT is
400      * valid if we have found a matching CPU
401      */
402     nvt->w0 = cpu_to_be32(NVT_W0_VALID);
403     return 0;
404 }
405 
406 static int spapr_xive_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk,
407                                 uint32_t nvt_idx, XiveNVT *nvt,
408                                 uint8_t word_number)
409 {
410     /*
411      * We don't need to write back to the NVTs because the sPAPR
412      * machine should never hit a non-scheduled NVT. It should never
413      * get called.
414      */
415     g_assert_not_reached();
416 }
417 
418 static XiveTCTX *spapr_xive_get_tctx(XiveRouter *xrtr, CPUState *cs)
419 {
420     PowerPCCPU *cpu = POWERPC_CPU(cs);
421 
422     return spapr_cpu_state(cpu)->tctx;
423 }
424 
425 static const VMStateDescription vmstate_spapr_xive_end = {
426     .name = TYPE_SPAPR_XIVE "/end",
427     .version_id = 1,
428     .minimum_version_id = 1,
429     .fields = (VMStateField []) {
430         VMSTATE_UINT32(w0, XiveEND),
431         VMSTATE_UINT32(w1, XiveEND),
432         VMSTATE_UINT32(w2, XiveEND),
433         VMSTATE_UINT32(w3, XiveEND),
434         VMSTATE_UINT32(w4, XiveEND),
435         VMSTATE_UINT32(w5, XiveEND),
436         VMSTATE_UINT32(w6, XiveEND),
437         VMSTATE_UINT32(w7, XiveEND),
438         VMSTATE_END_OF_LIST()
439     },
440 };
441 
442 static const VMStateDescription vmstate_spapr_xive_eas = {
443     .name = TYPE_SPAPR_XIVE "/eas",
444     .version_id = 1,
445     .minimum_version_id = 1,
446     .fields = (VMStateField []) {
447         VMSTATE_UINT64(w, XiveEAS),
448         VMSTATE_END_OF_LIST()
449     },
450 };
451 
452 static int vmstate_spapr_xive_pre_save(void *opaque)
453 {
454     if (kvm_irqchip_in_kernel()) {
455         return kvmppc_xive_pre_save(SPAPR_XIVE(opaque));
456     }
457 
458     return 0;
459 }
460 
461 /*
462  * Called by the sPAPR IRQ backend 'post_load' method at the machine
463  * level.
464  */
465 static int spapr_xive_post_load(SpaprInterruptController *intc, int version_id)
466 {
467     if (kvm_irqchip_in_kernel()) {
468         return kvmppc_xive_post_load(SPAPR_XIVE(intc), version_id);
469     }
470 
471     return 0;
472 }
473 
474 static const VMStateDescription vmstate_spapr_xive = {
475     .name = TYPE_SPAPR_XIVE,
476     .version_id = 1,
477     .minimum_version_id = 1,
478     .pre_save = vmstate_spapr_xive_pre_save,
479     .post_load = NULL, /* handled at the machine level */
480     .fields = (VMStateField[]) {
481         VMSTATE_UINT32_EQUAL(nr_irqs, SpaprXive, NULL),
482         VMSTATE_STRUCT_VARRAY_POINTER_UINT32(eat, SpaprXive, nr_irqs,
483                                      vmstate_spapr_xive_eas, XiveEAS),
484         VMSTATE_STRUCT_VARRAY_POINTER_UINT32(endt, SpaprXive, nr_ends,
485                                              vmstate_spapr_xive_end, XiveEND),
486         VMSTATE_END_OF_LIST()
487     },
488 };
489 
490 static int spapr_xive_claim_irq(SpaprInterruptController *intc, int lisn,
491                                 bool lsi, Error **errp)
492 {
493     SpaprXive *xive = SPAPR_XIVE(intc);
494     XiveSource *xsrc = &xive->source;
495 
496     assert(lisn < xive->nr_irqs);
497 
498     if (xive_eas_is_valid(&xive->eat[lisn])) {
499         error_setg(errp, "IRQ %d is not free", lisn);
500         return -EBUSY;
501     }
502 
503     /*
504      * Set default values when allocating an IRQ number
505      */
506     xive->eat[lisn].w |= cpu_to_be64(EAS_VALID | EAS_MASKED);
507     if (lsi) {
508         xive_source_irq_set_lsi(xsrc, lisn);
509     }
510 
511     if (kvm_irqchip_in_kernel()) {
512         return kvmppc_xive_source_reset_one(xsrc, lisn, errp);
513     }
514 
515     return 0;
516 }
517 
518 static void spapr_xive_free_irq(SpaprInterruptController *intc, int lisn)
519 {
520     SpaprXive *xive = SPAPR_XIVE(intc);
521     assert(lisn < xive->nr_irqs);
522 
523     xive->eat[lisn].w &= cpu_to_be64(~EAS_VALID);
524 }
525 
526 static Property spapr_xive_properties[] = {
527     DEFINE_PROP_UINT32("nr-irqs", SpaprXive, nr_irqs, 0),
528     DEFINE_PROP_UINT32("nr-ends", SpaprXive, nr_ends, 0),
529     DEFINE_PROP_UINT64("vc-base", SpaprXive, vc_base, SPAPR_XIVE_VC_BASE),
530     DEFINE_PROP_UINT64("tm-base", SpaprXive, tm_base, SPAPR_XIVE_TM_BASE),
531     DEFINE_PROP_END_OF_LIST(),
532 };
533 
534 static int spapr_xive_cpu_intc_create(SpaprInterruptController *intc,
535                                       PowerPCCPU *cpu, Error **errp)
536 {
537     SpaprXive *xive = SPAPR_XIVE(intc);
538     Object *obj;
539     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
540 
541     obj = xive_tctx_create(OBJECT(cpu), XIVE_ROUTER(xive), errp);
542     if (!obj) {
543         return -1;
544     }
545 
546     spapr_cpu->tctx = XIVE_TCTX(obj);
547 
548     /*
549      * (TCG) Early setting the OS CAM line for hotplugged CPUs as they
550      * don't beneficiate from the reset of the XIVE IRQ backend
551      */
552     spapr_xive_set_tctx_os_cam(spapr_cpu->tctx);
553     return 0;
554 }
555 
556 static void spapr_xive_cpu_intc_reset(SpaprInterruptController *intc,
557                                      PowerPCCPU *cpu)
558 {
559     XiveTCTX *tctx = spapr_cpu_state(cpu)->tctx;
560 
561     xive_tctx_reset(tctx);
562 }
563 
564 static void spapr_xive_set_irq(SpaprInterruptController *intc, int irq, int val)
565 {
566     SpaprXive *xive = SPAPR_XIVE(intc);
567 
568     if (kvm_irqchip_in_kernel()) {
569         kvmppc_xive_source_set_irq(&xive->source, irq, val);
570     } else {
571         xive_source_set_irq(&xive->source, irq, val);
572     }
573 }
574 
575 static void spapr_xive_print_info(SpaprInterruptController *intc, Monitor *mon)
576 {
577     SpaprXive *xive = SPAPR_XIVE(intc);
578     CPUState *cs;
579 
580     CPU_FOREACH(cs) {
581         PowerPCCPU *cpu = POWERPC_CPU(cs);
582 
583         xive_tctx_pic_print_info(spapr_cpu_state(cpu)->tctx, mon);
584     }
585 
586     spapr_xive_pic_print_info(xive, mon);
587 }
588 
589 static void spapr_xive_dt(SpaprInterruptController *intc, uint32_t nr_servers,
590                           void *fdt, uint32_t phandle)
591 {
592     SpaprXive *xive = SPAPR_XIVE(intc);
593     int node;
594     uint64_t timas[2 * 2];
595     /* Interrupt number ranges for the IPIs */
596     uint32_t lisn_ranges[] = {
597         cpu_to_be32(0),
598         cpu_to_be32(nr_servers),
599     };
600     /*
601      * EQ size - the sizes of pages supported by the system 4K, 64K,
602      * 2M, 16M. We only advertise 64K for the moment.
603      */
604     uint32_t eq_sizes[] = {
605         cpu_to_be32(16), /* 64K */
606     };
607     /*
608      * The following array is in sync with the reserved priorities
609      * defined by the 'spapr_xive_priority_is_reserved' routine.
610      */
611     uint32_t plat_res_int_priorities[] = {
612         cpu_to_be32(7),    /* start */
613         cpu_to_be32(0xf8), /* count */
614     };
615 
616     /* Thread Interrupt Management Area : User (ring 3) and OS (ring 2) */
617     timas[0] = cpu_to_be64(xive->tm_base +
618                            XIVE_TM_USER_PAGE * (1ull << TM_SHIFT));
619     timas[1] = cpu_to_be64(1ull << TM_SHIFT);
620     timas[2] = cpu_to_be64(xive->tm_base +
621                            XIVE_TM_OS_PAGE * (1ull << TM_SHIFT));
622     timas[3] = cpu_to_be64(1ull << TM_SHIFT);
623 
624     _FDT(node = fdt_add_subnode(fdt, 0, xive->nodename));
625 
626     _FDT(fdt_setprop_string(fdt, node, "device_type", "power-ivpe"));
627     _FDT(fdt_setprop(fdt, node, "reg", timas, sizeof(timas)));
628 
629     _FDT(fdt_setprop_string(fdt, node, "compatible", "ibm,power-ivpe"));
630     _FDT(fdt_setprop(fdt, node, "ibm,xive-eq-sizes", eq_sizes,
631                      sizeof(eq_sizes)));
632     _FDT(fdt_setprop(fdt, node, "ibm,xive-lisn-ranges", lisn_ranges,
633                      sizeof(lisn_ranges)));
634 
635     /* For Linux to link the LSIs to the interrupt controller. */
636     _FDT(fdt_setprop(fdt, node, "interrupt-controller", NULL, 0));
637     _FDT(fdt_setprop_cell(fdt, node, "#interrupt-cells", 2));
638 
639     /* For SLOF */
640     _FDT(fdt_setprop_cell(fdt, node, "linux,phandle", phandle));
641     _FDT(fdt_setprop_cell(fdt, node, "phandle", phandle));
642 
643     /*
644      * The "ibm,plat-res-int-priorities" property defines the priority
645      * ranges reserved by the hypervisor
646      */
647     _FDT(fdt_setprop(fdt, 0, "ibm,plat-res-int-priorities",
648                      plat_res_int_priorities, sizeof(plat_res_int_priorities)));
649 }
650 
651 static int spapr_xive_activate(SpaprInterruptController *intc, Error **errp)
652 {
653     SpaprXive *xive = SPAPR_XIVE(intc);
654     CPUState *cs;
655 
656     CPU_FOREACH(cs) {
657         PowerPCCPU *cpu = POWERPC_CPU(cs);
658 
659         /* (TCG) Set the OS CAM line of the thread interrupt context. */
660         spapr_xive_set_tctx_os_cam(spapr_cpu_state(cpu)->tctx);
661     }
662 
663     if (kvm_enabled()) {
664         int rc = spapr_irq_init_kvm(kvmppc_xive_connect, intc, errp);
665         if (rc < 0) {
666             return rc;
667         }
668     }
669 
670     /* Activate the XIVE MMIOs */
671     spapr_xive_mmio_set_enabled(xive, true);
672 
673     return 0;
674 }
675 
676 static void spapr_xive_deactivate(SpaprInterruptController *intc)
677 {
678     SpaprXive *xive = SPAPR_XIVE(intc);
679 
680     spapr_xive_mmio_set_enabled(xive, false);
681 
682     if (kvm_irqchip_in_kernel()) {
683         kvmppc_xive_disconnect(intc);
684     }
685 }
686 
687 static void spapr_xive_class_init(ObjectClass *klass, void *data)
688 {
689     DeviceClass *dc = DEVICE_CLASS(klass);
690     XiveRouterClass *xrc = XIVE_ROUTER_CLASS(klass);
691     SpaprInterruptControllerClass *sicc = SPAPR_INTC_CLASS(klass);
692 
693     dc->desc    = "sPAPR XIVE Interrupt Controller";
694     dc->props   = spapr_xive_properties;
695     dc->realize = spapr_xive_realize;
696     dc->vmsd    = &vmstate_spapr_xive;
697 
698     xrc->get_eas = spapr_xive_get_eas;
699     xrc->get_end = spapr_xive_get_end;
700     xrc->write_end = spapr_xive_write_end;
701     xrc->get_nvt = spapr_xive_get_nvt;
702     xrc->write_nvt = spapr_xive_write_nvt;
703     xrc->get_tctx = spapr_xive_get_tctx;
704 
705     sicc->activate = spapr_xive_activate;
706     sicc->deactivate = spapr_xive_deactivate;
707     sicc->cpu_intc_create = spapr_xive_cpu_intc_create;
708     sicc->cpu_intc_reset = spapr_xive_cpu_intc_reset;
709     sicc->claim_irq = spapr_xive_claim_irq;
710     sicc->free_irq = spapr_xive_free_irq;
711     sicc->set_irq = spapr_xive_set_irq;
712     sicc->print_info = spapr_xive_print_info;
713     sicc->dt = spapr_xive_dt;
714     sicc->post_load = spapr_xive_post_load;
715 }
716 
717 static const TypeInfo spapr_xive_info = {
718     .name = TYPE_SPAPR_XIVE,
719     .parent = TYPE_XIVE_ROUTER,
720     .instance_init = spapr_xive_instance_init,
721     .instance_size = sizeof(SpaprXive),
722     .class_init = spapr_xive_class_init,
723     .interfaces = (InterfaceInfo[]) {
724         { TYPE_SPAPR_INTC },
725         { }
726     },
727 };
728 
729 static void spapr_xive_register_types(void)
730 {
731     type_register_static(&spapr_xive_info);
732 }
733 
734 type_init(spapr_xive_register_types)
735 
736 /*
737  * XIVE hcalls
738  *
739  * The terminology used by the XIVE hcalls is the following :
740  *
741  *   TARGET vCPU number
742  *   EQ     Event Queue assigned by OS to receive event data
743  *   ESB    page for source interrupt management
744  *   LISN   Logical Interrupt Source Number identifying a source in the
745  *          machine
746  *   EISN   Effective Interrupt Source Number used by guest OS to
747  *          identify source in the guest
748  *
749  * The EAS, END, NVT structures are not exposed.
750  */
751 
752 /*
753  * Linux hosts under OPAL reserve priority 7 for their own escalation
754  * interrupts (DD2.X POWER9). So we only allow the guest to use
755  * priorities [0..6].
756  */
757 static bool spapr_xive_priority_is_reserved(uint8_t priority)
758 {
759     switch (priority) {
760     case 0 ... 6:
761         return false;
762     case 7: /* OPAL escalation queue */
763     default:
764         return true;
765     }
766 }
767 
768 /*
769  * The H_INT_GET_SOURCE_INFO hcall() is used to obtain the logical
770  * real address of the MMIO page through which the Event State Buffer
771  * entry associated with the value of the "lisn" parameter is managed.
772  *
773  * Parameters:
774  * Input
775  * - R4: "flags"
776  *         Bits 0-63 reserved
777  * - R5: "lisn" is per "interrupts", "interrupt-map", or
778  *       "ibm,xive-lisn-ranges" properties, or as returned by the
779  *       ibm,query-interrupt-source-number RTAS call, or as returned
780  *       by the H_ALLOCATE_VAS_WINDOW hcall
781  *
782  * Output
783  * - R4: "flags"
784  *         Bits 0-59: Reserved
785  *         Bit 60: H_INT_ESB must be used for Event State Buffer
786  *                 management
787  *         Bit 61: 1 == LSI  0 == MSI
788  *         Bit 62: the full function page supports trigger
789  *         Bit 63: Store EOI Supported
790  * - R5: Logical Real address of full function Event State Buffer
791  *       management page, -1 if H_INT_ESB hcall flag is set to 1.
792  * - R6: Logical Real Address of trigger only Event State Buffer
793  *       management page or -1.
794  * - R7: Power of 2 page size for the ESB management pages returned in
795  *       R5 and R6.
796  */
797 
798 #define SPAPR_XIVE_SRC_H_INT_ESB     PPC_BIT(60) /* ESB manage with H_INT_ESB */
799 #define SPAPR_XIVE_SRC_LSI           PPC_BIT(61) /* Virtual LSI type */
800 #define SPAPR_XIVE_SRC_TRIGGER       PPC_BIT(62) /* Trigger and management
801                                                     on same page */
802 #define SPAPR_XIVE_SRC_STORE_EOI     PPC_BIT(63) /* Store EOI support */
803 
804 static target_ulong h_int_get_source_info(PowerPCCPU *cpu,
805                                           SpaprMachineState *spapr,
806                                           target_ulong opcode,
807                                           target_ulong *args)
808 {
809     SpaprXive *xive = spapr->xive;
810     XiveSource *xsrc = &xive->source;
811     target_ulong flags  = args[0];
812     target_ulong lisn   = args[1];
813 
814     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
815         return H_FUNCTION;
816     }
817 
818     if (flags) {
819         return H_PARAMETER;
820     }
821 
822     if (lisn >= xive->nr_irqs) {
823         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
824                       lisn);
825         return H_P2;
826     }
827 
828     if (!xive_eas_is_valid(&xive->eat[lisn])) {
829         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
830                       lisn);
831         return H_P2;
832     }
833 
834     /*
835      * All sources are emulated under the main XIVE object and share
836      * the same characteristics.
837      */
838     args[0] = 0;
839     if (!xive_source_esb_has_2page(xsrc)) {
840         args[0] |= SPAPR_XIVE_SRC_TRIGGER;
841     }
842     if (xsrc->esb_flags & XIVE_SRC_STORE_EOI) {
843         args[0] |= SPAPR_XIVE_SRC_STORE_EOI;
844     }
845 
846     /*
847      * Force the use of the H_INT_ESB hcall in case of an LSI
848      * interrupt. This is necessary under KVM to re-trigger the
849      * interrupt if the level is still asserted
850      */
851     if (xive_source_irq_is_lsi(xsrc, lisn)) {
852         args[0] |= SPAPR_XIVE_SRC_H_INT_ESB | SPAPR_XIVE_SRC_LSI;
853     }
854 
855     if (!(args[0] & SPAPR_XIVE_SRC_H_INT_ESB)) {
856         args[1] = xive->vc_base + xive_source_esb_mgmt(xsrc, lisn);
857     } else {
858         args[1] = -1;
859     }
860 
861     if (xive_source_esb_has_2page(xsrc) &&
862         !(args[0] & SPAPR_XIVE_SRC_H_INT_ESB)) {
863         args[2] = xive->vc_base + xive_source_esb_page(xsrc, lisn);
864     } else {
865         args[2] = -1;
866     }
867 
868     if (xive_source_esb_has_2page(xsrc)) {
869         args[3] = xsrc->esb_shift - 1;
870     } else {
871         args[3] = xsrc->esb_shift;
872     }
873 
874     return H_SUCCESS;
875 }
876 
877 /*
878  * The H_INT_SET_SOURCE_CONFIG hcall() is used to assign a Logical
879  * Interrupt Source to a target. The Logical Interrupt Source is
880  * designated with the "lisn" parameter and the target is designated
881  * with the "target" and "priority" parameters.  Upon return from the
882  * hcall(), no additional interrupts will be directed to the old EQ.
883  *
884  * Parameters:
885  * Input:
886  * - R4: "flags"
887  *         Bits 0-61: Reserved
888  *         Bit 62: set the "eisn" in the EAS
889  *         Bit 63: masks the interrupt source in the hardware interrupt
890  *       control structure. An interrupt masked by this mechanism will
891  *       be dropped, but it's source state bits will still be
892  *       set. There is no race-free way of unmasking and restoring the
893  *       source. Thus this should only be used in interrupts that are
894  *       also masked at the source, and only in cases where the
895  *       interrupt is not meant to be used for a large amount of time
896  *       because no valid target exists for it for example
897  * - R5: "lisn" is per "interrupts", "interrupt-map", or
898  *       "ibm,xive-lisn-ranges" properties, or as returned by the
899  *       ibm,query-interrupt-source-number RTAS call, or as returned by
900  *       the H_ALLOCATE_VAS_WINDOW hcall
901  * - R6: "target" is per "ibm,ppc-interrupt-server#s" or
902  *       "ibm,ppc-interrupt-gserver#s"
903  * - R7: "priority" is a valid priority not in
904  *       "ibm,plat-res-int-priorities"
905  * - R8: "eisn" is the guest EISN associated with the "lisn"
906  *
907  * Output:
908  * - None
909  */
910 
911 #define SPAPR_XIVE_SRC_SET_EISN PPC_BIT(62)
912 #define SPAPR_XIVE_SRC_MASK     PPC_BIT(63)
913 
914 static target_ulong h_int_set_source_config(PowerPCCPU *cpu,
915                                             SpaprMachineState *spapr,
916                                             target_ulong opcode,
917                                             target_ulong *args)
918 {
919     SpaprXive *xive = spapr->xive;
920     XiveEAS eas, new_eas;
921     target_ulong flags    = args[0];
922     target_ulong lisn     = args[1];
923     target_ulong target   = args[2];
924     target_ulong priority = args[3];
925     target_ulong eisn     = args[4];
926     uint8_t end_blk;
927     uint32_t end_idx;
928 
929     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
930         return H_FUNCTION;
931     }
932 
933     if (flags & ~(SPAPR_XIVE_SRC_SET_EISN | SPAPR_XIVE_SRC_MASK)) {
934         return H_PARAMETER;
935     }
936 
937     if (lisn >= xive->nr_irqs) {
938         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
939                       lisn);
940         return H_P2;
941     }
942 
943     eas = xive->eat[lisn];
944     if (!xive_eas_is_valid(&eas)) {
945         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
946                       lisn);
947         return H_P2;
948     }
949 
950     /* priority 0xff is used to reset the EAS */
951     if (priority == 0xff) {
952         new_eas.w = cpu_to_be64(EAS_VALID | EAS_MASKED);
953         goto out;
954     }
955 
956     if (flags & SPAPR_XIVE_SRC_MASK) {
957         new_eas.w = eas.w | cpu_to_be64(EAS_MASKED);
958     } else {
959         new_eas.w = eas.w & cpu_to_be64(~EAS_MASKED);
960     }
961 
962     if (spapr_xive_priority_is_reserved(priority)) {
963         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
964                       " is reserved\n", priority);
965         return H_P4;
966     }
967 
968     /*
969      * Validate that "target" is part of the list of threads allocated
970      * to the partition. For that, find the END corresponding to the
971      * target.
972      */
973     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
974         return H_P3;
975     }
976 
977     new_eas.w = xive_set_field64(EAS_END_BLOCK, new_eas.w, end_blk);
978     new_eas.w = xive_set_field64(EAS_END_INDEX, new_eas.w, end_idx);
979 
980     if (flags & SPAPR_XIVE_SRC_SET_EISN) {
981         new_eas.w = xive_set_field64(EAS_END_DATA, new_eas.w, eisn);
982     }
983 
984     if (kvm_irqchip_in_kernel()) {
985         Error *local_err = NULL;
986 
987         kvmppc_xive_set_source_config(xive, lisn, &new_eas, &local_err);
988         if (local_err) {
989             error_report_err(local_err);
990             return H_HARDWARE;
991         }
992     }
993 
994 out:
995     xive->eat[lisn] = new_eas;
996     return H_SUCCESS;
997 }
998 
999 /*
1000  * The H_INT_GET_SOURCE_CONFIG hcall() is used to determine to which
1001  * target/priority pair is assigned to the specified Logical Interrupt
1002  * Source.
1003  *
1004  * Parameters:
1005  * Input:
1006  * - R4: "flags"
1007  *         Bits 0-63 Reserved
1008  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1009  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1010  *       ibm,query-interrupt-source-number RTAS call, or as
1011  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1012  *
1013  * Output:
1014  * - R4: Target to which the specified Logical Interrupt Source is
1015  *       assigned
1016  * - R5: Priority to which the specified Logical Interrupt Source is
1017  *       assigned
1018  * - R6: EISN for the specified Logical Interrupt Source (this will be
1019  *       equivalent to the LISN if not changed by H_INT_SET_SOURCE_CONFIG)
1020  */
1021 static target_ulong h_int_get_source_config(PowerPCCPU *cpu,
1022                                             SpaprMachineState *spapr,
1023                                             target_ulong opcode,
1024                                             target_ulong *args)
1025 {
1026     SpaprXive *xive = spapr->xive;
1027     target_ulong flags = args[0];
1028     target_ulong lisn = args[1];
1029     XiveEAS eas;
1030     XiveEND *end;
1031     uint8_t nvt_blk;
1032     uint32_t end_idx, nvt_idx;
1033 
1034     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1035         return H_FUNCTION;
1036     }
1037 
1038     if (flags) {
1039         return H_PARAMETER;
1040     }
1041 
1042     if (lisn >= xive->nr_irqs) {
1043         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1044                       lisn);
1045         return H_P2;
1046     }
1047 
1048     eas = xive->eat[lisn];
1049     if (!xive_eas_is_valid(&eas)) {
1050         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1051                       lisn);
1052         return H_P2;
1053     }
1054 
1055     /* EAS_END_BLOCK is unused on sPAPR */
1056     end_idx = xive_get_field64(EAS_END_INDEX, eas.w);
1057 
1058     assert(end_idx < xive->nr_ends);
1059     end = &xive->endt[end_idx];
1060 
1061     nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end->w6);
1062     nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end->w6);
1063     args[0] = spapr_xive_nvt_to_target(nvt_blk, nvt_idx);
1064 
1065     if (xive_eas_is_masked(&eas)) {
1066         args[1] = 0xff;
1067     } else {
1068         args[1] = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
1069     }
1070 
1071     args[2] = xive_get_field64(EAS_END_DATA, eas.w);
1072 
1073     return H_SUCCESS;
1074 }
1075 
1076 /*
1077  * The H_INT_GET_QUEUE_INFO hcall() is used to get the logical real
1078  * address of the notification management page associated with the
1079  * specified target and priority.
1080  *
1081  * Parameters:
1082  * Input:
1083  * - R4: "flags"
1084  *         Bits 0-63 Reserved
1085  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1086  *       "ibm,ppc-interrupt-gserver#s"
1087  * - R6: "priority" is a valid priority not in
1088  *       "ibm,plat-res-int-priorities"
1089  *
1090  * Output:
1091  * - R4: Logical real address of notification page
1092  * - R5: Power of 2 page size of the notification page
1093  */
1094 static target_ulong h_int_get_queue_info(PowerPCCPU *cpu,
1095                                          SpaprMachineState *spapr,
1096                                          target_ulong opcode,
1097                                          target_ulong *args)
1098 {
1099     SpaprXive *xive = spapr->xive;
1100     XiveENDSource *end_xsrc = &xive->end_source;
1101     target_ulong flags = args[0];
1102     target_ulong target = args[1];
1103     target_ulong priority = args[2];
1104     XiveEND *end;
1105     uint8_t end_blk;
1106     uint32_t end_idx;
1107 
1108     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1109         return H_FUNCTION;
1110     }
1111 
1112     if (flags) {
1113         return H_PARAMETER;
1114     }
1115 
1116     /*
1117      * H_STATE should be returned if a H_INT_RESET is in progress.
1118      * This is not needed when running the emulation under QEMU
1119      */
1120 
1121     if (spapr_xive_priority_is_reserved(priority)) {
1122         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1123                       " is reserved\n", priority);
1124         return H_P3;
1125     }
1126 
1127     /*
1128      * Validate that "target" is part of the list of threads allocated
1129      * to the partition. For that, find the END corresponding to the
1130      * target.
1131      */
1132     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1133         return H_P2;
1134     }
1135 
1136     assert(end_idx < xive->nr_ends);
1137     end = &xive->endt[end_idx];
1138 
1139     args[0] = xive->end_base + (1ull << (end_xsrc->esb_shift + 1)) * end_idx;
1140     if (xive_end_is_enqueue(end)) {
1141         args[1] = xive_get_field32(END_W0_QSIZE, end->w0) + 12;
1142     } else {
1143         args[1] = 0;
1144     }
1145 
1146     return H_SUCCESS;
1147 }
1148 
1149 /*
1150  * The H_INT_SET_QUEUE_CONFIG hcall() is used to set or reset a EQ for
1151  * a given "target" and "priority".  It is also used to set the
1152  * notification config associated with the EQ.  An EQ size of 0 is
1153  * used to reset the EQ config for a given target and priority. If
1154  * resetting the EQ config, the END associated with the given "target"
1155  * and "priority" will be changed to disable queueing.
1156  *
1157  * Upon return from the hcall(), no additional interrupts will be
1158  * directed to the old EQ (if one was set). The old EQ (if one was
1159  * set) should be investigated for interrupts that occurred prior to
1160  * or during the hcall().
1161  *
1162  * Parameters:
1163  * Input:
1164  * - R4: "flags"
1165  *         Bits 0-62: Reserved
1166  *         Bit 63: Unconditional Notify (n) per the XIVE spec
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  * - R7: "eventQueue": The logical real address of the start of the EQ
1172  * - R8: "eventQueueSize": The power of 2 EQ size per "ibm,xive-eq-sizes"
1173  *
1174  * Output:
1175  * - None
1176  */
1177 
1178 #define SPAPR_XIVE_END_ALWAYS_NOTIFY PPC_BIT(63)
1179 
1180 static target_ulong h_int_set_queue_config(PowerPCCPU *cpu,
1181                                            SpaprMachineState *spapr,
1182                                            target_ulong opcode,
1183                                            target_ulong *args)
1184 {
1185     SpaprXive *xive = spapr->xive;
1186     target_ulong flags = args[0];
1187     target_ulong target = args[1];
1188     target_ulong priority = args[2];
1189     target_ulong qpage = args[3];
1190     target_ulong qsize = args[4];
1191     XiveEND end;
1192     uint8_t end_blk, nvt_blk;
1193     uint32_t end_idx, nvt_idx;
1194 
1195     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1196         return H_FUNCTION;
1197     }
1198 
1199     if (flags & ~SPAPR_XIVE_END_ALWAYS_NOTIFY) {
1200         return H_PARAMETER;
1201     }
1202 
1203     /*
1204      * H_STATE should be returned if a H_INT_RESET is in progress.
1205      * This is not needed when running the emulation under QEMU
1206      */
1207 
1208     if (spapr_xive_priority_is_reserved(priority)) {
1209         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1210                       " is reserved\n", priority);
1211         return H_P3;
1212     }
1213 
1214     /*
1215      * Validate that "target" is part of the list of threads allocated
1216      * to the partition. For that, find the END corresponding to the
1217      * target.
1218      */
1219 
1220     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1221         return H_P2;
1222     }
1223 
1224     assert(end_idx < xive->nr_ends);
1225     memcpy(&end, &xive->endt[end_idx], sizeof(XiveEND));
1226 
1227     switch (qsize) {
1228     case 12:
1229     case 16:
1230     case 21:
1231     case 24:
1232         if (!QEMU_IS_ALIGNED(qpage, 1ul << qsize)) {
1233             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: EQ @0x%" HWADDR_PRIx
1234                           " is not naturally aligned with %" HWADDR_PRIx "\n",
1235                           qpage, (hwaddr)1 << qsize);
1236             return H_P4;
1237         }
1238         end.w2 = cpu_to_be32((qpage >> 32) & 0x0fffffff);
1239         end.w3 = cpu_to_be32(qpage & 0xffffffff);
1240         end.w0 |= cpu_to_be32(END_W0_ENQUEUE);
1241         end.w0 = xive_set_field32(END_W0_QSIZE, end.w0, qsize - 12);
1242         break;
1243     case 0:
1244         /* reset queue and disable queueing */
1245         spapr_xive_end_reset(&end);
1246         goto out;
1247 
1248     default:
1249         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid EQ size %"PRIx64"\n",
1250                       qsize);
1251         return H_P5;
1252     }
1253 
1254     if (qsize) {
1255         hwaddr plen = 1 << qsize;
1256         void *eq;
1257 
1258         /*
1259          * Validate the guest EQ. We should also check that the queue
1260          * has been zeroed by the OS.
1261          */
1262         eq = address_space_map(CPU(cpu)->as, qpage, &plen, true,
1263                                MEMTXATTRS_UNSPECIFIED);
1264         if (plen != 1 << qsize) {
1265             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to map EQ @0x%"
1266                           HWADDR_PRIx "\n", qpage);
1267             return H_P4;
1268         }
1269         address_space_unmap(CPU(cpu)->as, eq, plen, true, plen);
1270     }
1271 
1272     /* "target" should have been validated above */
1273     if (spapr_xive_target_to_nvt(target, &nvt_blk, &nvt_idx)) {
1274         g_assert_not_reached();
1275     }
1276 
1277     /*
1278      * Ensure the priority and target are correctly set (they will not
1279      * be right after allocation)
1280      */
1281     end.w6 = xive_set_field32(END_W6_NVT_BLOCK, 0ul, nvt_blk) |
1282         xive_set_field32(END_W6_NVT_INDEX, 0ul, nvt_idx);
1283     end.w7 = xive_set_field32(END_W7_F0_PRIORITY, 0ul, priority);
1284 
1285     if (flags & SPAPR_XIVE_END_ALWAYS_NOTIFY) {
1286         end.w0 |= cpu_to_be32(END_W0_UCOND_NOTIFY);
1287     } else {
1288         end.w0 &= cpu_to_be32((uint32_t)~END_W0_UCOND_NOTIFY);
1289     }
1290 
1291     /*
1292      * The generation bit for the END starts at 1 and The END page
1293      * offset counter starts at 0.
1294      */
1295     end.w1 = cpu_to_be32(END_W1_GENERATION) |
1296         xive_set_field32(END_W1_PAGE_OFF, 0ul, 0ul);
1297     end.w0 |= cpu_to_be32(END_W0_VALID);
1298 
1299     /*
1300      * TODO: issue syncs required to ensure all in-flight interrupts
1301      * are complete on the old END
1302      */
1303 
1304 out:
1305     if (kvm_irqchip_in_kernel()) {
1306         Error *local_err = NULL;
1307 
1308         kvmppc_xive_set_queue_config(xive, end_blk, end_idx, &end, &local_err);
1309         if (local_err) {
1310             error_report_err(local_err);
1311             return H_HARDWARE;
1312         }
1313     }
1314 
1315     /* Update END */
1316     memcpy(&xive->endt[end_idx], &end, sizeof(XiveEND));
1317     return H_SUCCESS;
1318 }
1319 
1320 /*
1321  * The H_INT_GET_QUEUE_CONFIG hcall() is used to get a EQ for a given
1322  * target and priority.
1323  *
1324  * Parameters:
1325  * Input:
1326  * - R4: "flags"
1327  *         Bits 0-62: Reserved
1328  *         Bit 63: Debug: Return debug data
1329  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1330  *       "ibm,ppc-interrupt-gserver#s"
1331  * - R6: "priority" is a valid priority not in
1332  *       "ibm,plat-res-int-priorities"
1333  *
1334  * Output:
1335  * - R4: "flags":
1336  *       Bits 0-61: Reserved
1337  *       Bit 62: The value of Event Queue Generation Number (g) per
1338  *              the XIVE spec if "Debug" = 1
1339  *       Bit 63: The value of Unconditional Notify (n) per the XIVE spec
1340  * - R5: The logical real address of the start of the EQ
1341  * - R6: The power of 2 EQ size per "ibm,xive-eq-sizes"
1342  * - R7: The value of Event Queue Offset Counter per XIVE spec
1343  *       if "Debug" = 1, else 0
1344  *
1345  */
1346 
1347 #define SPAPR_XIVE_END_DEBUG     PPC_BIT(63)
1348 
1349 static target_ulong h_int_get_queue_config(PowerPCCPU *cpu,
1350                                            SpaprMachineState *spapr,
1351                                            target_ulong opcode,
1352                                            target_ulong *args)
1353 {
1354     SpaprXive *xive = spapr->xive;
1355     target_ulong flags = args[0];
1356     target_ulong target = args[1];
1357     target_ulong priority = args[2];
1358     XiveEND *end;
1359     uint8_t end_blk;
1360     uint32_t end_idx;
1361 
1362     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1363         return H_FUNCTION;
1364     }
1365 
1366     if (flags & ~SPAPR_XIVE_END_DEBUG) {
1367         return H_PARAMETER;
1368     }
1369 
1370     /*
1371      * H_STATE should be returned if a H_INT_RESET is in progress.
1372      * This is not needed when running the emulation under QEMU
1373      */
1374 
1375     if (spapr_xive_priority_is_reserved(priority)) {
1376         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1377                       " is reserved\n", priority);
1378         return H_P3;
1379     }
1380 
1381     /*
1382      * Validate that "target" is part of the list of threads allocated
1383      * to the partition. For that, find the END corresponding to the
1384      * target.
1385      */
1386     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1387         return H_P2;
1388     }
1389 
1390     assert(end_idx < xive->nr_ends);
1391     end = &xive->endt[end_idx];
1392 
1393     args[0] = 0;
1394     if (xive_end_is_notify(end)) {
1395         args[0] |= SPAPR_XIVE_END_ALWAYS_NOTIFY;
1396     }
1397 
1398     if (xive_end_is_enqueue(end)) {
1399         args[1] = xive_end_qaddr(end);
1400         args[2] = xive_get_field32(END_W0_QSIZE, end->w0) + 12;
1401     } else {
1402         args[1] = 0;
1403         args[2] = 0;
1404     }
1405 
1406     if (kvm_irqchip_in_kernel()) {
1407         Error *local_err = NULL;
1408 
1409         kvmppc_xive_get_queue_config(xive, end_blk, end_idx, end, &local_err);
1410         if (local_err) {
1411             error_report_err(local_err);
1412             return H_HARDWARE;
1413         }
1414     }
1415 
1416     /* TODO: do we need any locking on the END ? */
1417     if (flags & SPAPR_XIVE_END_DEBUG) {
1418         /* Load the event queue generation number into the return flags */
1419         args[0] |= (uint64_t)xive_get_field32(END_W1_GENERATION, end->w1) << 62;
1420 
1421         /* Load R7 with the event queue offset counter */
1422         args[3] = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1423     } else {
1424         args[3] = 0;
1425     }
1426 
1427     return H_SUCCESS;
1428 }
1429 
1430 /*
1431  * The H_INT_SET_OS_REPORTING_LINE hcall() is used to set the
1432  * reporting cache line pair for the calling thread.  The reporting
1433  * cache lines will contain the OS interrupt context when the OS
1434  * issues a CI store byte to @TIMA+0xC10 to acknowledge the OS
1435  * interrupt. The reporting cache lines can be reset by inputting -1
1436  * in "reportingLine".  Issuing the CI store byte without reporting
1437  * cache lines registered will result in the data not being accessible
1438  * to the OS.
1439  *
1440  * Parameters:
1441  * Input:
1442  * - R4: "flags"
1443  *         Bits 0-63: Reserved
1444  * - R5: "reportingLine": The logical real address of the reporting cache
1445  *       line pair
1446  *
1447  * Output:
1448  * - None
1449  */
1450 static target_ulong h_int_set_os_reporting_line(PowerPCCPU *cpu,
1451                                                 SpaprMachineState *spapr,
1452                                                 target_ulong opcode,
1453                                                 target_ulong *args)
1454 {
1455     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1456         return H_FUNCTION;
1457     }
1458 
1459     /*
1460      * H_STATE should be returned if a H_INT_RESET is in progress.
1461      * This is not needed when running the emulation under QEMU
1462      */
1463 
1464     /* TODO: H_INT_SET_OS_REPORTING_LINE */
1465     return H_FUNCTION;
1466 }
1467 
1468 /*
1469  * The H_INT_GET_OS_REPORTING_LINE hcall() is used to get the logical
1470  * real address of the reporting cache line pair set for the input
1471  * "target".  If no reporting cache line pair has been set, -1 is
1472  * returned.
1473  *
1474  * Parameters:
1475  * Input:
1476  * - R4: "flags"
1477  *         Bits 0-63: Reserved
1478  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1479  *       "ibm,ppc-interrupt-gserver#s"
1480  * - R6: "reportingLine": The logical real address of the reporting
1481  *        cache line pair
1482  *
1483  * Output:
1484  * - R4: The logical real address of the reporting line if set, else -1
1485  */
1486 static target_ulong h_int_get_os_reporting_line(PowerPCCPU *cpu,
1487                                                 SpaprMachineState *spapr,
1488                                                 target_ulong opcode,
1489                                                 target_ulong *args)
1490 {
1491     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1492         return H_FUNCTION;
1493     }
1494 
1495     /*
1496      * H_STATE should be returned if a H_INT_RESET is in progress.
1497      * This is not needed when running the emulation under QEMU
1498      */
1499 
1500     /* TODO: H_INT_GET_OS_REPORTING_LINE */
1501     return H_FUNCTION;
1502 }
1503 
1504 /*
1505  * The H_INT_ESB hcall() is used to issue a load or store to the ESB
1506  * page for the input "lisn".  This hcall is only supported for LISNs
1507  * that have the ESB hcall flag set to 1 when returned from hcall()
1508  * H_INT_GET_SOURCE_INFO.
1509  *
1510  * Parameters:
1511  * Input:
1512  * - R4: "flags"
1513  *         Bits 0-62: Reserved
1514  *         bit 63: Store: Store=1, store operation, else load operation
1515  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1516  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1517  *       ibm,query-interrupt-source-number RTAS call, or as
1518  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1519  * - R6: "esbOffset" is the offset into the ESB page for the load or
1520  *       store operation
1521  * - R7: "storeData" is the data to write for a store operation
1522  *
1523  * Output:
1524  * - R4: The value of the load if load operation, else -1
1525  */
1526 
1527 #define SPAPR_XIVE_ESB_STORE PPC_BIT(63)
1528 
1529 static target_ulong h_int_esb(PowerPCCPU *cpu,
1530                               SpaprMachineState *spapr,
1531                               target_ulong opcode,
1532                               target_ulong *args)
1533 {
1534     SpaprXive *xive = spapr->xive;
1535     XiveEAS eas;
1536     target_ulong flags  = args[0];
1537     target_ulong lisn   = args[1];
1538     target_ulong offset = args[2];
1539     target_ulong data   = args[3];
1540     hwaddr mmio_addr;
1541     XiveSource *xsrc = &xive->source;
1542 
1543     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1544         return H_FUNCTION;
1545     }
1546 
1547     if (flags & ~SPAPR_XIVE_ESB_STORE) {
1548         return H_PARAMETER;
1549     }
1550 
1551     if (lisn >= xive->nr_irqs) {
1552         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1553                       lisn);
1554         return H_P2;
1555     }
1556 
1557     eas = xive->eat[lisn];
1558     if (!xive_eas_is_valid(&eas)) {
1559         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1560                       lisn);
1561         return H_P2;
1562     }
1563 
1564     if (offset > (1ull << xsrc->esb_shift)) {
1565         return H_P3;
1566     }
1567 
1568     if (kvm_irqchip_in_kernel()) {
1569         args[0] = kvmppc_xive_esb_rw(xsrc, lisn, offset, data,
1570                                      flags & SPAPR_XIVE_ESB_STORE);
1571     } else {
1572         mmio_addr = xive->vc_base + xive_source_esb_mgmt(xsrc, lisn) + offset;
1573 
1574         if (dma_memory_rw(&address_space_memory, mmio_addr, &data, 8,
1575                           (flags & SPAPR_XIVE_ESB_STORE))) {
1576             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to access ESB @0x%"
1577                           HWADDR_PRIx "\n", mmio_addr);
1578             return H_HARDWARE;
1579         }
1580         args[0] = (flags & SPAPR_XIVE_ESB_STORE) ? -1 : data;
1581     }
1582     return H_SUCCESS;
1583 }
1584 
1585 /*
1586  * The H_INT_SYNC hcall() is used to issue hardware syncs that will
1587  * ensure any in flight events for the input lisn are in the event
1588  * queue.
1589  *
1590  * Parameters:
1591  * Input:
1592  * - R4: "flags"
1593  *         Bits 0-63: Reserved
1594  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1595  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1596  *       ibm,query-interrupt-source-number RTAS call, or as
1597  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1598  *
1599  * Output:
1600  * - None
1601  */
1602 static target_ulong h_int_sync(PowerPCCPU *cpu,
1603                                SpaprMachineState *spapr,
1604                                target_ulong opcode,
1605                                target_ulong *args)
1606 {
1607     SpaprXive *xive = spapr->xive;
1608     XiveEAS eas;
1609     target_ulong flags = args[0];
1610     target_ulong lisn = args[1];
1611 
1612     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1613         return H_FUNCTION;
1614     }
1615 
1616     if (flags) {
1617         return H_PARAMETER;
1618     }
1619 
1620     if (lisn >= xive->nr_irqs) {
1621         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1622                       lisn);
1623         return H_P2;
1624     }
1625 
1626     eas = xive->eat[lisn];
1627     if (!xive_eas_is_valid(&eas)) {
1628         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1629                       lisn);
1630         return H_P2;
1631     }
1632 
1633     /*
1634      * H_STATE should be returned if a H_INT_RESET is in progress.
1635      * This is not needed when running the emulation under QEMU
1636      */
1637 
1638     /*
1639      * This is not real hardware. Nothing to be done unless when
1640      * under KVM
1641      */
1642 
1643     if (kvm_irqchip_in_kernel()) {
1644         Error *local_err = NULL;
1645 
1646         kvmppc_xive_sync_source(xive, lisn, &local_err);
1647         if (local_err) {
1648             error_report_err(local_err);
1649             return H_HARDWARE;
1650         }
1651     }
1652     return H_SUCCESS;
1653 }
1654 
1655 /*
1656  * The H_INT_RESET hcall() is used to reset all of the partition's
1657  * interrupt exploitation structures to their initial state.  This
1658  * means losing all previously set interrupt state set via
1659  * H_INT_SET_SOURCE_CONFIG and H_INT_SET_QUEUE_CONFIG.
1660  *
1661  * Parameters:
1662  * Input:
1663  * - R4: "flags"
1664  *         Bits 0-63: Reserved
1665  *
1666  * Output:
1667  * - None
1668  */
1669 static target_ulong h_int_reset(PowerPCCPU *cpu,
1670                                 SpaprMachineState *spapr,
1671                                 target_ulong opcode,
1672                                 target_ulong *args)
1673 {
1674     SpaprXive *xive = spapr->xive;
1675     target_ulong flags   = args[0];
1676 
1677     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1678         return H_FUNCTION;
1679     }
1680 
1681     if (flags) {
1682         return H_PARAMETER;
1683     }
1684 
1685     device_reset(DEVICE(xive));
1686 
1687     if (kvm_irqchip_in_kernel()) {
1688         Error *local_err = NULL;
1689 
1690         kvmppc_xive_reset(xive, &local_err);
1691         if (local_err) {
1692             error_report_err(local_err);
1693             return H_HARDWARE;
1694         }
1695     }
1696     return H_SUCCESS;
1697 }
1698 
1699 void spapr_xive_hcall_init(SpaprMachineState *spapr)
1700 {
1701     spapr_register_hypercall(H_INT_GET_SOURCE_INFO, h_int_get_source_info);
1702     spapr_register_hypercall(H_INT_SET_SOURCE_CONFIG, h_int_set_source_config);
1703     spapr_register_hypercall(H_INT_GET_SOURCE_CONFIG, h_int_get_source_config);
1704     spapr_register_hypercall(H_INT_GET_QUEUE_INFO, h_int_get_queue_info);
1705     spapr_register_hypercall(H_INT_SET_QUEUE_CONFIG, h_int_set_queue_config);
1706     spapr_register_hypercall(H_INT_GET_QUEUE_CONFIG, h_int_get_queue_config);
1707     spapr_register_hypercall(H_INT_SET_OS_REPORTING_LINE,
1708                              h_int_set_os_reporting_line);
1709     spapr_register_hypercall(H_INT_GET_OS_REPORTING_LINE,
1710                              h_int_get_os_reporting_line);
1711     spapr_register_hypercall(H_INT_ESB, h_int_esb);
1712     spapr_register_hypercall(H_INT_SYNC, h_int_sync);
1713     spapr_register_hypercall(H_INT_RESET, h_int_reset);
1714 }
1715