xref: /openbmc/qemu/hw/intc/spapr_xive.c (revision f0984d40)
1 /*
2  * QEMU PowerPC sPAPR XIVE interrupt controller model
3  *
4  * Copyright (c) 2017-2018, IBM Corporation.
5  *
6  * This code is licensed under the GPL version 2 or later. See the
7  * COPYING file in the top-level directory.
8  */
9 
10 #include "qemu/osdep.h"
11 #include "qemu/log.h"
12 #include "qemu/module.h"
13 #include "qapi/error.h"
14 #include "qemu/error-report.h"
15 #include "target/ppc/cpu.h"
16 #include "sysemu/cpus.h"
17 #include "sysemu/reset.h"
18 #include "migration/vmstate.h"
19 #include "monitor/monitor.h"
20 #include "hw/ppc/fdt.h"
21 #include "hw/ppc/spapr.h"
22 #include "hw/ppc/spapr_cpu_core.h"
23 #include "hw/ppc/spapr_xive.h"
24 #include "hw/ppc/xive.h"
25 #include "hw/ppc/xive_regs.h"
26 #include "hw/qdev-properties.h"
27 #include "trace.h"
28 
29 /*
30  * XIVE Virtualization Controller BAR and Thread Managment BAR that we
31  * use for the ESB pages and the TIMA pages
32  */
33 #define SPAPR_XIVE_VC_BASE   0x0006010000000000ull
34 #define SPAPR_XIVE_TM_BASE   0x0006030203180000ull
35 
36 /*
37  * The allocation of VP blocks is a complex operation in OPAL and the
38  * VP identifiers have a relation with the number of HW chips, the
39  * size of the VP blocks, VP grouping, etc. The QEMU sPAPR XIVE
40  * controller model does not have the same constraints and can use a
41  * simple mapping scheme of the CPU vcpu_id
42  *
43  * These identifiers are never returned to the OS.
44  */
45 
46 #define SPAPR_XIVE_NVT_BASE 0x400
47 
48 /*
49  * sPAPR NVT and END indexing helpers
50  */
51 static uint32_t spapr_xive_nvt_to_target(uint8_t nvt_blk, uint32_t nvt_idx)
52 {
53     return nvt_idx - SPAPR_XIVE_NVT_BASE;
54 }
55 
56 static void spapr_xive_cpu_to_nvt(PowerPCCPU *cpu,
57                                   uint8_t *out_nvt_blk, uint32_t *out_nvt_idx)
58 {
59     assert(cpu);
60 
61     if (out_nvt_blk) {
62         *out_nvt_blk = SPAPR_XIVE_BLOCK_ID;
63     }
64 
65     if (out_nvt_blk) {
66         *out_nvt_idx = SPAPR_XIVE_NVT_BASE + cpu->vcpu_id;
67     }
68 }
69 
70 static int spapr_xive_target_to_nvt(uint32_t target,
71                                     uint8_t *out_nvt_blk, uint32_t *out_nvt_idx)
72 {
73     PowerPCCPU *cpu = spapr_find_cpu(target);
74 
75     if (!cpu) {
76         return -1;
77     }
78 
79     spapr_xive_cpu_to_nvt(cpu, out_nvt_blk, out_nvt_idx);
80     return 0;
81 }
82 
83 /*
84  * sPAPR END indexing uses a simple mapping of the CPU vcpu_id, 8
85  * priorities per CPU
86  */
87 int spapr_xive_end_to_target(uint8_t end_blk, uint32_t end_idx,
88                              uint32_t *out_server, uint8_t *out_prio)
89 {
90 
91     assert(end_blk == SPAPR_XIVE_BLOCK_ID);
92 
93     if (out_server) {
94         *out_server = end_idx >> 3;
95     }
96 
97     if (out_prio) {
98         *out_prio = end_idx & 0x7;
99     }
100     return 0;
101 }
102 
103 static void spapr_xive_cpu_to_end(PowerPCCPU *cpu, uint8_t prio,
104                                   uint8_t *out_end_blk, uint32_t *out_end_idx)
105 {
106     assert(cpu);
107 
108     if (out_end_blk) {
109         *out_end_blk = SPAPR_XIVE_BLOCK_ID;
110     }
111 
112     if (out_end_idx) {
113         *out_end_idx = (cpu->vcpu_id << 3) + prio;
114     }
115 }
116 
117 static int spapr_xive_target_to_end(uint32_t target, uint8_t prio,
118                                     uint8_t *out_end_blk, uint32_t *out_end_idx)
119 {
120     PowerPCCPU *cpu = spapr_find_cpu(target);
121 
122     if (!cpu) {
123         return -1;
124     }
125 
126     spapr_xive_cpu_to_end(cpu, prio, out_end_blk, out_end_idx);
127     return 0;
128 }
129 
130 /*
131  * On sPAPR machines, use a simplified output for the XIVE END
132  * structure dumping only the information related to the OS EQ.
133  */
134 static void spapr_xive_end_pic_print_info(SpaprXive *xive, XiveEND *end,
135                                           Monitor *mon)
136 {
137     uint64_t qaddr_base = xive_end_qaddr(end);
138     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
139     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
140     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
141     uint32_t qentries = 1 << (qsize + 10);
142     uint32_t nvt = xive_get_field32(END_W6_NVT_INDEX, end->w6);
143     uint8_t priority = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
144 
145     monitor_printf(mon, "%3d/%d % 6d/%5d @%"PRIx64" ^%d",
146                    spapr_xive_nvt_to_target(0, nvt),
147                    priority, qindex, qentries, qaddr_base, qgen);
148 
149     xive_end_queue_pic_print_info(end, 6, mon);
150 }
151 
152 /*
153  * kvm_irqchip_in_kernel() will cause the compiler to turn this
154  * info a nop if CONFIG_KVM isn't defined.
155  */
156 #define spapr_xive_in_kernel(xive) \
157     (kvm_irqchip_in_kernel() && (xive)->fd != -1)
158 
159 static void spapr_xive_pic_print_info(SpaprXive *xive, Monitor *mon)
160 {
161     XiveSource *xsrc = &xive->source;
162     int i;
163 
164     if (spapr_xive_in_kernel(xive)) {
165         Error *local_err = NULL;
166 
167         kvmppc_xive_synchronize_state(xive, &local_err);
168         if (local_err) {
169             error_report_err(local_err);
170             return;
171         }
172     }
173 
174     monitor_printf(mon, "  LISN         PQ    EISN     CPU/PRIO EQ\n");
175 
176     for (i = 0; i < xive->nr_irqs; i++) {
177         uint8_t pq = xive_source_esb_get(xsrc, i);
178         XiveEAS *eas = &xive->eat[i];
179 
180         if (!xive_eas_is_valid(eas)) {
181             continue;
182         }
183 
184         monitor_printf(mon, "  %08x %s %c%c%c %s %08x ", i,
185                        xive_source_irq_is_lsi(xsrc, i) ? "LSI" : "MSI",
186                        pq & XIVE_ESB_VAL_P ? 'P' : '-',
187                        pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
188                        xive_source_is_asserted(xsrc, i) ? 'A' : ' ',
189                        xive_eas_is_masked(eas) ? "M" : " ",
190                        (int) xive_get_field64(EAS_END_DATA, eas->w));
191 
192         if (!xive_eas_is_masked(eas)) {
193             uint32_t end_idx = xive_get_field64(EAS_END_INDEX, eas->w);
194             XiveEND *end;
195 
196             assert(end_idx < xive->nr_ends);
197             end = &xive->endt[end_idx];
198 
199             if (xive_end_is_valid(end)) {
200                 spapr_xive_end_pic_print_info(xive, end, mon);
201             }
202         }
203         monitor_printf(mon, "\n");
204     }
205 }
206 
207 void spapr_xive_mmio_set_enabled(SpaprXive *xive, bool enable)
208 {
209     memory_region_set_enabled(&xive->source.esb_mmio, enable);
210     memory_region_set_enabled(&xive->tm_mmio, enable);
211 
212     /* Disable the END ESBs until a guest OS makes use of them */
213     memory_region_set_enabled(&xive->end_source.esb_mmio, false);
214 }
215 
216 static void spapr_xive_tm_write(void *opaque, hwaddr offset,
217                           uint64_t value, unsigned size)
218 {
219     XiveTCTX *tctx = spapr_cpu_state(POWERPC_CPU(current_cpu))->tctx;
220 
221     xive_tctx_tm_write(XIVE_PRESENTER(opaque), tctx, offset, value, size);
222 }
223 
224 static uint64_t spapr_xive_tm_read(void *opaque, hwaddr offset, unsigned size)
225 {
226     XiveTCTX *tctx = spapr_cpu_state(POWERPC_CPU(current_cpu))->tctx;
227 
228     return xive_tctx_tm_read(XIVE_PRESENTER(opaque), tctx, offset, size);
229 }
230 
231 const MemoryRegionOps spapr_xive_tm_ops = {
232     .read = spapr_xive_tm_read,
233     .write = spapr_xive_tm_write,
234     .endianness = DEVICE_BIG_ENDIAN,
235     .valid = {
236         .min_access_size = 1,
237         .max_access_size = 8,
238     },
239     .impl = {
240         .min_access_size = 1,
241         .max_access_size = 8,
242     },
243 };
244 
245 static void spapr_xive_end_reset(XiveEND *end)
246 {
247     memset(end, 0, sizeof(*end));
248 
249     /* switch off the escalation and notification ESBs */
250     end->w1 = cpu_to_be32(END_W1_ESe_Q | END_W1_ESn_Q);
251 }
252 
253 static void spapr_xive_reset(void *dev)
254 {
255     SpaprXive *xive = SPAPR_XIVE(dev);
256     int i;
257 
258     /*
259      * The XiveSource has its own reset handler, which mask off all
260      * IRQs (!P|Q)
261      */
262 
263     /* Mask all valid EASs in the IRQ number space. */
264     for (i = 0; i < xive->nr_irqs; i++) {
265         XiveEAS *eas = &xive->eat[i];
266         if (xive_eas_is_valid(eas)) {
267             eas->w = cpu_to_be64(EAS_VALID | EAS_MASKED);
268         } else {
269             eas->w = 0;
270         }
271     }
272 
273     /* Clear all ENDs */
274     for (i = 0; i < xive->nr_ends; i++) {
275         spapr_xive_end_reset(&xive->endt[i]);
276     }
277 }
278 
279 static void spapr_xive_instance_init(Object *obj)
280 {
281     SpaprXive *xive = SPAPR_XIVE(obj);
282 
283     object_initialize_child(obj, "source", &xive->source, TYPE_XIVE_SOURCE);
284 
285     object_initialize_child(obj, "end_source", &xive->end_source,
286                             TYPE_XIVE_END_SOURCE);
287 
288     /* Not connected to the KVM XIVE device */
289     xive->fd = -1;
290 }
291 
292 static void spapr_xive_realize(DeviceState *dev, Error **errp)
293 {
294     SpaprXive *xive = SPAPR_XIVE(dev);
295     SpaprXiveClass *sxc = SPAPR_XIVE_GET_CLASS(xive);
296     XiveSource *xsrc = &xive->source;
297     XiveENDSource *end_xsrc = &xive->end_source;
298     Error *local_err = NULL;
299 
300     /* Set by spapr_irq_init() */
301     g_assert(xive->nr_irqs);
302     g_assert(xive->nr_ends);
303 
304     sxc->parent_realize(dev, &local_err);
305     if (local_err) {
306         error_propagate(errp, local_err);
307         return;
308     }
309 
310     /*
311      * Initialize the internal sources, for IPIs and virtual devices.
312      */
313     object_property_set_int(OBJECT(xsrc), "nr-irqs", xive->nr_irqs,
314                             &error_fatal);
315     object_property_set_link(OBJECT(xsrc), "xive", OBJECT(xive), &error_abort);
316     if (!qdev_realize(DEVICE(xsrc), NULL, errp)) {
317         return;
318     }
319     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &xsrc->esb_mmio);
320 
321     /*
322      * Initialize the END ESB source
323      */
324     object_property_set_int(OBJECT(end_xsrc), "nr-ends", xive->nr_irqs,
325                             &error_fatal);
326     object_property_set_link(OBJECT(end_xsrc), "xive", OBJECT(xive),
327                              &error_abort);
328     if (!qdev_realize(DEVICE(end_xsrc), NULL, errp)) {
329         return;
330     }
331     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &end_xsrc->esb_mmio);
332 
333     /* Set the mapping address of the END ESB pages after the source ESBs */
334     xive->end_base = xive->vc_base + xive_source_esb_len(xsrc);
335 
336     /*
337      * Allocate the routing tables
338      */
339     xive->eat = g_new0(XiveEAS, xive->nr_irqs);
340     xive->endt = g_new0(XiveEND, xive->nr_ends);
341 
342     xive->nodename = g_strdup_printf("interrupt-controller@%" PRIx64,
343                            xive->tm_base + XIVE_TM_USER_PAGE * (1 << TM_SHIFT));
344 
345     qemu_register_reset(spapr_xive_reset, dev);
346 
347     /* TIMA initialization */
348     memory_region_init_io(&xive->tm_mmio, OBJECT(xive), &spapr_xive_tm_ops,
349                           xive, "xive.tima", 4ull << TM_SHIFT);
350     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &xive->tm_mmio);
351 
352     /*
353      * Map all regions. These will be enabled or disabled at reset and
354      * can also be overridden by KVM memory regions if active
355      */
356     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 0, xive->vc_base);
357     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 1, xive->end_base);
358     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 2, xive->tm_base);
359 }
360 
361 static int spapr_xive_get_eas(XiveRouter *xrtr, uint8_t eas_blk,
362                               uint32_t eas_idx, XiveEAS *eas)
363 {
364     SpaprXive *xive = SPAPR_XIVE(xrtr);
365 
366     if (eas_idx >= xive->nr_irqs) {
367         return -1;
368     }
369 
370     *eas = xive->eat[eas_idx];
371     return 0;
372 }
373 
374 static int spapr_xive_get_end(XiveRouter *xrtr,
375                               uint8_t end_blk, uint32_t end_idx, XiveEND *end)
376 {
377     SpaprXive *xive = SPAPR_XIVE(xrtr);
378 
379     if (end_idx >= xive->nr_ends) {
380         return -1;
381     }
382 
383     memcpy(end, &xive->endt[end_idx], sizeof(XiveEND));
384     return 0;
385 }
386 
387 static int spapr_xive_write_end(XiveRouter *xrtr, uint8_t end_blk,
388                                 uint32_t end_idx, XiveEND *end,
389                                 uint8_t word_number)
390 {
391     SpaprXive *xive = SPAPR_XIVE(xrtr);
392 
393     if (end_idx >= xive->nr_ends) {
394         return -1;
395     }
396 
397     memcpy(&xive->endt[end_idx], end, sizeof(XiveEND));
398     return 0;
399 }
400 
401 static int spapr_xive_get_nvt(XiveRouter *xrtr,
402                               uint8_t nvt_blk, uint32_t nvt_idx, XiveNVT *nvt)
403 {
404     uint32_t vcpu_id = spapr_xive_nvt_to_target(nvt_blk, nvt_idx);
405     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
406 
407     if (!cpu) {
408         /* TODO: should we assert() if we can find a NVT ? */
409         return -1;
410     }
411 
412     /*
413      * sPAPR does not maintain a NVT table. Return that the NVT is
414      * valid if we have found a matching CPU
415      */
416     nvt->w0 = cpu_to_be32(NVT_W0_VALID);
417     return 0;
418 }
419 
420 static int spapr_xive_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk,
421                                 uint32_t nvt_idx, XiveNVT *nvt,
422                                 uint8_t word_number)
423 {
424     /*
425      * We don't need to write back to the NVTs because the sPAPR
426      * machine should never hit a non-scheduled NVT. It should never
427      * get called.
428      */
429     g_assert_not_reached();
430 }
431 
432 static int spapr_xive_match_nvt(XivePresenter *xptr, uint8_t format,
433                                 uint8_t nvt_blk, uint32_t nvt_idx,
434                                 bool cam_ignore, uint8_t priority,
435                                 uint32_t logic_serv, XiveTCTXMatch *match)
436 {
437     CPUState *cs;
438     int count = 0;
439 
440     CPU_FOREACH(cs) {
441         PowerPCCPU *cpu = POWERPC_CPU(cs);
442         XiveTCTX *tctx = spapr_cpu_state(cpu)->tctx;
443         int ring;
444 
445         /*
446          * Skip partially initialized vCPUs. This can happen when
447          * vCPUs are hotplugged.
448          */
449         if (!tctx) {
450             continue;
451         }
452 
453         /*
454          * Check the thread context CAM lines and record matches.
455          */
456         ring = xive_presenter_tctx_match(xptr, tctx, format, nvt_blk, nvt_idx,
457                                          cam_ignore, logic_serv);
458         /*
459          * Save the matching thread interrupt context and follow on to
460          * check for duplicates which are invalid.
461          */
462         if (ring != -1) {
463             if (match->tctx) {
464                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: already found a thread "
465                               "context NVT %x/%x\n", nvt_blk, nvt_idx);
466                 return -1;
467             }
468 
469             match->ring = ring;
470             match->tctx = tctx;
471             count++;
472         }
473     }
474 
475     return count;
476 }
477 
478 static uint8_t spapr_xive_get_block_id(XiveRouter *xrtr)
479 {
480     return SPAPR_XIVE_BLOCK_ID;
481 }
482 
483 static int spapr_xive_get_pq(XiveRouter *xrtr, uint8_t blk, uint32_t idx,
484                              uint8_t *pq)
485 {
486     SpaprXive *xive = SPAPR_XIVE(xrtr);
487 
488     assert(SPAPR_XIVE_BLOCK_ID == blk);
489 
490     *pq = xive_source_esb_get(&xive->source, idx);
491     return 0;
492 }
493 
494 static int spapr_xive_set_pq(XiveRouter *xrtr, uint8_t blk, uint32_t idx,
495                              uint8_t *pq)
496 {
497     SpaprXive *xive = SPAPR_XIVE(xrtr);
498 
499     assert(SPAPR_XIVE_BLOCK_ID == blk);
500 
501     *pq = xive_source_esb_set(&xive->source, idx, *pq);
502     return 0;
503 }
504 
505 
506 static const VMStateDescription vmstate_spapr_xive_end = {
507     .name = TYPE_SPAPR_XIVE "/end",
508     .version_id = 1,
509     .minimum_version_id = 1,
510     .fields = (VMStateField []) {
511         VMSTATE_UINT32(w0, XiveEND),
512         VMSTATE_UINT32(w1, XiveEND),
513         VMSTATE_UINT32(w2, XiveEND),
514         VMSTATE_UINT32(w3, XiveEND),
515         VMSTATE_UINT32(w4, XiveEND),
516         VMSTATE_UINT32(w5, XiveEND),
517         VMSTATE_UINT32(w6, XiveEND),
518         VMSTATE_UINT32(w7, XiveEND),
519         VMSTATE_END_OF_LIST()
520     },
521 };
522 
523 static const VMStateDescription vmstate_spapr_xive_eas = {
524     .name = TYPE_SPAPR_XIVE "/eas",
525     .version_id = 1,
526     .minimum_version_id = 1,
527     .fields = (VMStateField []) {
528         VMSTATE_UINT64(w, XiveEAS),
529         VMSTATE_END_OF_LIST()
530     },
531 };
532 
533 static int vmstate_spapr_xive_pre_save(void *opaque)
534 {
535     SpaprXive *xive = SPAPR_XIVE(opaque);
536 
537     if (spapr_xive_in_kernel(xive)) {
538         return kvmppc_xive_pre_save(xive);
539     }
540 
541     return 0;
542 }
543 
544 /*
545  * Called by the sPAPR IRQ backend 'post_load' method at the machine
546  * level.
547  */
548 static int spapr_xive_post_load(SpaprInterruptController *intc, int version_id)
549 {
550     SpaprXive *xive = SPAPR_XIVE(intc);
551 
552     if (spapr_xive_in_kernel(xive)) {
553         return kvmppc_xive_post_load(xive, version_id);
554     }
555 
556     return 0;
557 }
558 
559 static const VMStateDescription vmstate_spapr_xive = {
560     .name = TYPE_SPAPR_XIVE,
561     .version_id = 1,
562     .minimum_version_id = 1,
563     .pre_save = vmstate_spapr_xive_pre_save,
564     .post_load = NULL, /* handled at the machine level */
565     .fields = (VMStateField[]) {
566         VMSTATE_UINT32_EQUAL(nr_irqs, SpaprXive, NULL),
567         VMSTATE_STRUCT_VARRAY_POINTER_UINT32(eat, SpaprXive, nr_irqs,
568                                      vmstate_spapr_xive_eas, XiveEAS),
569         VMSTATE_STRUCT_VARRAY_POINTER_UINT32(endt, SpaprXive, nr_ends,
570                                              vmstate_spapr_xive_end, XiveEND),
571         VMSTATE_END_OF_LIST()
572     },
573 };
574 
575 static int spapr_xive_claim_irq(SpaprInterruptController *intc, int lisn,
576                                 bool lsi, Error **errp)
577 {
578     SpaprXive *xive = SPAPR_XIVE(intc);
579     XiveSource *xsrc = &xive->source;
580 
581     assert(lisn < xive->nr_irqs);
582 
583     trace_spapr_xive_claim_irq(lisn, lsi);
584 
585     if (xive_eas_is_valid(&xive->eat[lisn])) {
586         error_setg(errp, "IRQ %d is not free", lisn);
587         return -EBUSY;
588     }
589 
590     /*
591      * Set default values when allocating an IRQ number
592      */
593     xive->eat[lisn].w |= cpu_to_be64(EAS_VALID | EAS_MASKED);
594     if (lsi) {
595         xive_source_irq_set_lsi(xsrc, lisn);
596     }
597 
598     if (spapr_xive_in_kernel(xive)) {
599         return kvmppc_xive_source_reset_one(xsrc, lisn, errp);
600     }
601 
602     return 0;
603 }
604 
605 static void spapr_xive_free_irq(SpaprInterruptController *intc, int lisn)
606 {
607     SpaprXive *xive = SPAPR_XIVE(intc);
608     assert(lisn < xive->nr_irqs);
609 
610     trace_spapr_xive_free_irq(lisn);
611 
612     xive->eat[lisn].w &= cpu_to_be64(~EAS_VALID);
613 }
614 
615 static Property spapr_xive_properties[] = {
616     DEFINE_PROP_UINT32("nr-irqs", SpaprXive, nr_irqs, 0),
617     DEFINE_PROP_UINT32("nr-ends", SpaprXive, nr_ends, 0),
618     DEFINE_PROP_UINT64("vc-base", SpaprXive, vc_base, SPAPR_XIVE_VC_BASE),
619     DEFINE_PROP_UINT64("tm-base", SpaprXive, tm_base, SPAPR_XIVE_TM_BASE),
620     DEFINE_PROP_UINT8("hv-prio", SpaprXive, hv_prio, 7),
621     DEFINE_PROP_END_OF_LIST(),
622 };
623 
624 static int spapr_xive_cpu_intc_create(SpaprInterruptController *intc,
625                                       PowerPCCPU *cpu, Error **errp)
626 {
627     SpaprXive *xive = SPAPR_XIVE(intc);
628     Object *obj;
629     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
630 
631     obj = xive_tctx_create(OBJECT(cpu), XIVE_PRESENTER(xive), errp);
632     if (!obj) {
633         return -1;
634     }
635 
636     spapr_cpu->tctx = XIVE_TCTX(obj);
637     return 0;
638 }
639 
640 static void xive_tctx_set_os_cam(XiveTCTX *tctx, uint32_t os_cam)
641 {
642     uint32_t qw1w2 = cpu_to_be32(TM_QW1W2_VO | os_cam);
643     memcpy(&tctx->regs[TM_QW1_OS + TM_WORD2], &qw1w2, 4);
644 }
645 
646 static void spapr_xive_cpu_intc_reset(SpaprInterruptController *intc,
647                                      PowerPCCPU *cpu)
648 {
649     XiveTCTX *tctx = spapr_cpu_state(cpu)->tctx;
650     uint8_t  nvt_blk;
651     uint32_t nvt_idx;
652 
653     xive_tctx_reset(tctx);
654 
655     /*
656      * When a Virtual Processor is scheduled to run on a HW thread,
657      * the hypervisor pushes its identifier in the OS CAM line.
658      * Emulate the same behavior under QEMU.
659      */
660     spapr_xive_cpu_to_nvt(cpu, &nvt_blk, &nvt_idx);
661 
662     xive_tctx_set_os_cam(tctx, xive_nvt_cam_line(nvt_blk, nvt_idx));
663 }
664 
665 static void spapr_xive_cpu_intc_destroy(SpaprInterruptController *intc,
666                                         PowerPCCPU *cpu)
667 {
668     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
669 
670     xive_tctx_destroy(spapr_cpu->tctx);
671     spapr_cpu->tctx = NULL;
672 }
673 
674 static void spapr_xive_set_irq(SpaprInterruptController *intc, int irq, int val)
675 {
676     SpaprXive *xive = SPAPR_XIVE(intc);
677 
678     trace_spapr_xive_set_irq(irq, val);
679 
680     if (spapr_xive_in_kernel(xive)) {
681         kvmppc_xive_source_set_irq(&xive->source, irq, val);
682     } else {
683         xive_source_set_irq(&xive->source, irq, val);
684     }
685 }
686 
687 static void spapr_xive_print_info(SpaprInterruptController *intc, Monitor *mon)
688 {
689     SpaprXive *xive = SPAPR_XIVE(intc);
690     CPUState *cs;
691 
692     CPU_FOREACH(cs) {
693         PowerPCCPU *cpu = POWERPC_CPU(cs);
694 
695         xive_tctx_pic_print_info(spapr_cpu_state(cpu)->tctx, mon);
696     }
697 
698     spapr_xive_pic_print_info(xive, mon);
699 }
700 
701 static void spapr_xive_dt(SpaprInterruptController *intc, uint32_t nr_servers,
702                           void *fdt, uint32_t phandle)
703 {
704     SpaprXive *xive = SPAPR_XIVE(intc);
705     int node;
706     uint64_t timas[2 * 2];
707     /* Interrupt number ranges for the IPIs */
708     uint32_t lisn_ranges[] = {
709         cpu_to_be32(SPAPR_IRQ_IPI),
710         cpu_to_be32(SPAPR_IRQ_IPI + nr_servers),
711     };
712     /*
713      * EQ size - the sizes of pages supported by the system 4K, 64K,
714      * 2M, 16M. We only advertise 64K for the moment.
715      */
716     uint32_t eq_sizes[] = {
717         cpu_to_be32(16), /* 64K */
718     };
719     /*
720      * QEMU/KVM only needs to define a single range to reserve the
721      * escalation priority. A priority bitmask would have been more
722      * appropriate.
723      */
724     uint32_t plat_res_int_priorities[] = {
725         cpu_to_be32(xive->hv_prio),    /* start */
726         cpu_to_be32(0xff - xive->hv_prio), /* count */
727     };
728 
729     /* Thread Interrupt Management Area : User (ring 3) and OS (ring 2) */
730     timas[0] = cpu_to_be64(xive->tm_base +
731                            XIVE_TM_USER_PAGE * (1ull << TM_SHIFT));
732     timas[1] = cpu_to_be64(1ull << TM_SHIFT);
733     timas[2] = cpu_to_be64(xive->tm_base +
734                            XIVE_TM_OS_PAGE * (1ull << TM_SHIFT));
735     timas[3] = cpu_to_be64(1ull << TM_SHIFT);
736 
737     _FDT(node = fdt_add_subnode(fdt, 0, xive->nodename));
738 
739     _FDT(fdt_setprop_string(fdt, node, "device_type", "power-ivpe"));
740     _FDT(fdt_setprop(fdt, node, "reg", timas, sizeof(timas)));
741 
742     _FDT(fdt_setprop_string(fdt, node, "compatible", "ibm,power-ivpe"));
743     _FDT(fdt_setprop(fdt, node, "ibm,xive-eq-sizes", eq_sizes,
744                      sizeof(eq_sizes)));
745     _FDT(fdt_setprop(fdt, node, "ibm,xive-lisn-ranges", lisn_ranges,
746                      sizeof(lisn_ranges)));
747 
748     /* For Linux to link the LSIs to the interrupt controller. */
749     _FDT(fdt_setprop(fdt, node, "interrupt-controller", NULL, 0));
750     _FDT(fdt_setprop_cell(fdt, node, "#interrupt-cells", 2));
751 
752     /* For SLOF */
753     _FDT(fdt_setprop_cell(fdt, node, "linux,phandle", phandle));
754     _FDT(fdt_setprop_cell(fdt, node, "phandle", phandle));
755 
756     /*
757      * The "ibm,plat-res-int-priorities" property defines the priority
758      * ranges reserved by the hypervisor
759      */
760     _FDT(fdt_setprop(fdt, 0, "ibm,plat-res-int-priorities",
761                      plat_res_int_priorities, sizeof(plat_res_int_priorities)));
762 }
763 
764 static int spapr_xive_activate(SpaprInterruptController *intc,
765                                uint32_t nr_servers, Error **errp)
766 {
767     SpaprXive *xive = SPAPR_XIVE(intc);
768 
769     if (kvm_enabled()) {
770         int rc = spapr_irq_init_kvm(kvmppc_xive_connect, intc, nr_servers,
771                                     errp);
772         if (rc < 0) {
773             return rc;
774         }
775     }
776 
777     /* Activate the XIVE MMIOs */
778     spapr_xive_mmio_set_enabled(xive, true);
779 
780     return 0;
781 }
782 
783 static void spapr_xive_deactivate(SpaprInterruptController *intc)
784 {
785     SpaprXive *xive = SPAPR_XIVE(intc);
786 
787     spapr_xive_mmio_set_enabled(xive, false);
788 
789     if (spapr_xive_in_kernel(xive)) {
790         kvmppc_xive_disconnect(intc);
791     }
792 }
793 
794 static bool spapr_xive_in_kernel_xptr(const XivePresenter *xptr)
795 {
796     return spapr_xive_in_kernel(SPAPR_XIVE(xptr));
797 }
798 
799 static void spapr_xive_class_init(ObjectClass *klass, void *data)
800 {
801     DeviceClass *dc = DEVICE_CLASS(klass);
802     XiveRouterClass *xrc = XIVE_ROUTER_CLASS(klass);
803     SpaprInterruptControllerClass *sicc = SPAPR_INTC_CLASS(klass);
804     XivePresenterClass *xpc = XIVE_PRESENTER_CLASS(klass);
805     SpaprXiveClass *sxc = SPAPR_XIVE_CLASS(klass);
806 
807     dc->desc    = "sPAPR XIVE Interrupt Controller";
808     device_class_set_props(dc, spapr_xive_properties);
809     device_class_set_parent_realize(dc, spapr_xive_realize,
810                                     &sxc->parent_realize);
811     dc->vmsd    = &vmstate_spapr_xive;
812 
813     xrc->get_eas = spapr_xive_get_eas;
814     xrc->get_pq  = spapr_xive_get_pq;
815     xrc->set_pq  = spapr_xive_set_pq;
816     xrc->get_end = spapr_xive_get_end;
817     xrc->write_end = spapr_xive_write_end;
818     xrc->get_nvt = spapr_xive_get_nvt;
819     xrc->write_nvt = spapr_xive_write_nvt;
820     xrc->get_block_id = spapr_xive_get_block_id;
821 
822     sicc->activate = spapr_xive_activate;
823     sicc->deactivate = spapr_xive_deactivate;
824     sicc->cpu_intc_create = spapr_xive_cpu_intc_create;
825     sicc->cpu_intc_reset = spapr_xive_cpu_intc_reset;
826     sicc->cpu_intc_destroy = spapr_xive_cpu_intc_destroy;
827     sicc->claim_irq = spapr_xive_claim_irq;
828     sicc->free_irq = spapr_xive_free_irq;
829     sicc->set_irq = spapr_xive_set_irq;
830     sicc->print_info = spapr_xive_print_info;
831     sicc->dt = spapr_xive_dt;
832     sicc->post_load = spapr_xive_post_load;
833 
834     xpc->match_nvt  = spapr_xive_match_nvt;
835     xpc->in_kernel  = spapr_xive_in_kernel_xptr;
836 }
837 
838 static const TypeInfo spapr_xive_info = {
839     .name = TYPE_SPAPR_XIVE,
840     .parent = TYPE_XIVE_ROUTER,
841     .instance_init = spapr_xive_instance_init,
842     .instance_size = sizeof(SpaprXive),
843     .class_init = spapr_xive_class_init,
844     .class_size = sizeof(SpaprXiveClass),
845     .interfaces = (InterfaceInfo[]) {
846         { TYPE_SPAPR_INTC },
847         { }
848     },
849 };
850 
851 static void spapr_xive_register_types(void)
852 {
853     type_register_static(&spapr_xive_info);
854 }
855 
856 type_init(spapr_xive_register_types)
857 
858 /*
859  * XIVE hcalls
860  *
861  * The terminology used by the XIVE hcalls is the following :
862  *
863  *   TARGET vCPU number
864  *   EQ     Event Queue assigned by OS to receive event data
865  *   ESB    page for source interrupt management
866  *   LISN   Logical Interrupt Source Number identifying a source in the
867  *          machine
868  *   EISN   Effective Interrupt Source Number used by guest OS to
869  *          identify source in the guest
870  *
871  * The EAS, END, NVT structures are not exposed.
872  */
873 
874 /*
875  * On POWER9, the KVM XIVE device uses priority 7 for the escalation
876  * interrupts. So we only allow the guest to use priorities [0..6].
877  */
878 static bool spapr_xive_priority_is_reserved(SpaprXive *xive, uint8_t priority)
879 {
880     return priority >= xive->hv_prio;
881 }
882 
883 /*
884  * The H_INT_GET_SOURCE_INFO hcall() is used to obtain the logical
885  * real address of the MMIO page through which the Event State Buffer
886  * entry associated with the value of the "lisn" parameter is managed.
887  *
888  * Parameters:
889  * Input
890  * - R4: "flags"
891  *         Bits 0-63 reserved
892  * - R5: "lisn" is per "interrupts", "interrupt-map", or
893  *       "ibm,xive-lisn-ranges" properties, or as returned by the
894  *       ibm,query-interrupt-source-number RTAS call, or as returned
895  *       by the H_ALLOCATE_VAS_WINDOW hcall
896  *
897  * Output
898  * - R4: "flags"
899  *         Bits 0-59: Reserved
900  *         Bit 60: H_INT_ESB must be used for Event State Buffer
901  *                 management
902  *         Bit 61: 1 == LSI  0 == MSI
903  *         Bit 62: the full function page supports trigger
904  *         Bit 63: Store EOI Supported
905  * - R5: Logical Real address of full function Event State Buffer
906  *       management page, -1 if H_INT_ESB hcall flag is set to 1.
907  * - R6: Logical Real Address of trigger only Event State Buffer
908  *       management page or -1.
909  * - R7: Power of 2 page size for the ESB management pages returned in
910  *       R5 and R6.
911  */
912 
913 #define SPAPR_XIVE_SRC_H_INT_ESB     PPC_BIT(60) /* ESB manage with H_INT_ESB */
914 #define SPAPR_XIVE_SRC_LSI           PPC_BIT(61) /* Virtual LSI type */
915 #define SPAPR_XIVE_SRC_TRIGGER       PPC_BIT(62) /* Trigger and management
916                                                     on same page */
917 #define SPAPR_XIVE_SRC_STORE_EOI     PPC_BIT(63) /* Store EOI support */
918 
919 static target_ulong h_int_get_source_info(PowerPCCPU *cpu,
920                                           SpaprMachineState *spapr,
921                                           target_ulong opcode,
922                                           target_ulong *args)
923 {
924     SpaprXive *xive = spapr->xive;
925     XiveSource *xsrc = &xive->source;
926     target_ulong flags  = args[0];
927     target_ulong lisn   = args[1];
928 
929     trace_spapr_xive_get_source_info(flags, lisn);
930 
931     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
932         return H_FUNCTION;
933     }
934 
935     if (flags) {
936         return H_PARAMETER;
937     }
938 
939     if (lisn >= xive->nr_irqs) {
940         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
941                       lisn);
942         return H_P2;
943     }
944 
945     if (!xive_eas_is_valid(&xive->eat[lisn])) {
946         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
947                       lisn);
948         return H_P2;
949     }
950 
951     /*
952      * All sources are emulated under the main XIVE object and share
953      * the same characteristics.
954      */
955     args[0] = 0;
956     if (!xive_source_esb_has_2page(xsrc)) {
957         args[0] |= SPAPR_XIVE_SRC_TRIGGER;
958     }
959     if (xsrc->esb_flags & XIVE_SRC_STORE_EOI) {
960         args[0] |= SPAPR_XIVE_SRC_STORE_EOI;
961     }
962 
963     /*
964      * Force the use of the H_INT_ESB hcall in case of an LSI
965      * interrupt. This is necessary under KVM to re-trigger the
966      * interrupt if the level is still asserted
967      */
968     if (xive_source_irq_is_lsi(xsrc, lisn)) {
969         args[0] |= SPAPR_XIVE_SRC_H_INT_ESB | SPAPR_XIVE_SRC_LSI;
970     }
971 
972     if (!(args[0] & SPAPR_XIVE_SRC_H_INT_ESB)) {
973         args[1] = xive->vc_base + xive_source_esb_mgmt(xsrc, lisn);
974     } else {
975         args[1] = -1;
976     }
977 
978     if (xive_source_esb_has_2page(xsrc) &&
979         !(args[0] & SPAPR_XIVE_SRC_H_INT_ESB)) {
980         args[2] = xive->vc_base + xive_source_esb_page(xsrc, lisn);
981     } else {
982         args[2] = -1;
983     }
984 
985     if (xive_source_esb_has_2page(xsrc)) {
986         args[3] = xsrc->esb_shift - 1;
987     } else {
988         args[3] = xsrc->esb_shift;
989     }
990 
991     return H_SUCCESS;
992 }
993 
994 /*
995  * The H_INT_SET_SOURCE_CONFIG hcall() is used to assign a Logical
996  * Interrupt Source to a target. The Logical Interrupt Source is
997  * designated with the "lisn" parameter and the target is designated
998  * with the "target" and "priority" parameters.  Upon return from the
999  * hcall(), no additional interrupts will be directed to the old EQ.
1000  *
1001  * Parameters:
1002  * Input:
1003  * - R4: "flags"
1004  *         Bits 0-61: Reserved
1005  *         Bit 62: set the "eisn" in the EAS
1006  *         Bit 63: masks the interrupt source in the hardware interrupt
1007  *       control structure. An interrupt masked by this mechanism will
1008  *       be dropped, but it's source state bits will still be
1009  *       set. There is no race-free way of unmasking and restoring the
1010  *       source. Thus this should only be used in interrupts that are
1011  *       also masked at the source, and only in cases where the
1012  *       interrupt is not meant to be used for a large amount of time
1013  *       because no valid target exists for it for example
1014  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1015  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1016  *       ibm,query-interrupt-source-number RTAS call, or as returned by
1017  *       the H_ALLOCATE_VAS_WINDOW hcall
1018  * - R6: "target" is per "ibm,ppc-interrupt-server#s" or
1019  *       "ibm,ppc-interrupt-gserver#s"
1020  * - R7: "priority" is a valid priority not in
1021  *       "ibm,plat-res-int-priorities"
1022  * - R8: "eisn" is the guest EISN associated with the "lisn"
1023  *
1024  * Output:
1025  * - None
1026  */
1027 
1028 #define SPAPR_XIVE_SRC_SET_EISN PPC_BIT(62)
1029 #define SPAPR_XIVE_SRC_MASK     PPC_BIT(63)
1030 
1031 static target_ulong h_int_set_source_config(PowerPCCPU *cpu,
1032                                             SpaprMachineState *spapr,
1033                                             target_ulong opcode,
1034                                             target_ulong *args)
1035 {
1036     SpaprXive *xive = spapr->xive;
1037     XiveEAS eas, new_eas;
1038     target_ulong flags    = args[0];
1039     target_ulong lisn     = args[1];
1040     target_ulong target   = args[2];
1041     target_ulong priority = args[3];
1042     target_ulong eisn     = args[4];
1043     uint8_t end_blk;
1044     uint32_t end_idx;
1045 
1046     trace_spapr_xive_set_source_config(flags, lisn, target, priority, eisn);
1047 
1048     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1049         return H_FUNCTION;
1050     }
1051 
1052     if (flags & ~(SPAPR_XIVE_SRC_SET_EISN | SPAPR_XIVE_SRC_MASK)) {
1053         return H_PARAMETER;
1054     }
1055 
1056     if (lisn >= xive->nr_irqs) {
1057         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1058                       lisn);
1059         return H_P2;
1060     }
1061 
1062     eas = xive->eat[lisn];
1063     if (!xive_eas_is_valid(&eas)) {
1064         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1065                       lisn);
1066         return H_P2;
1067     }
1068 
1069     /* priority 0xff is used to reset the EAS */
1070     if (priority == 0xff) {
1071         new_eas.w = cpu_to_be64(EAS_VALID | EAS_MASKED);
1072         goto out;
1073     }
1074 
1075     if (flags & SPAPR_XIVE_SRC_MASK) {
1076         new_eas.w = eas.w | cpu_to_be64(EAS_MASKED);
1077     } else {
1078         new_eas.w = eas.w & cpu_to_be64(~EAS_MASKED);
1079     }
1080 
1081     if (spapr_xive_priority_is_reserved(xive, priority)) {
1082         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1083                       " is reserved\n", priority);
1084         return H_P4;
1085     }
1086 
1087     /*
1088      * Validate that "target" is part of the list of threads allocated
1089      * to the partition. For that, find the END corresponding to the
1090      * target.
1091      */
1092     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1093         return H_P3;
1094     }
1095 
1096     new_eas.w = xive_set_field64(EAS_END_BLOCK, new_eas.w, end_blk);
1097     new_eas.w = xive_set_field64(EAS_END_INDEX, new_eas.w, end_idx);
1098 
1099     if (flags & SPAPR_XIVE_SRC_SET_EISN) {
1100         new_eas.w = xive_set_field64(EAS_END_DATA, new_eas.w, eisn);
1101     }
1102 
1103     if (spapr_xive_in_kernel(xive)) {
1104         Error *local_err = NULL;
1105 
1106         kvmppc_xive_set_source_config(xive, lisn, &new_eas, &local_err);
1107         if (local_err) {
1108             error_report_err(local_err);
1109             return H_HARDWARE;
1110         }
1111     }
1112 
1113 out:
1114     xive->eat[lisn] = new_eas;
1115     return H_SUCCESS;
1116 }
1117 
1118 /*
1119  * The H_INT_GET_SOURCE_CONFIG hcall() is used to determine to which
1120  * target/priority pair is assigned to the specified Logical Interrupt
1121  * Source.
1122  *
1123  * Parameters:
1124  * Input:
1125  * - R4: "flags"
1126  *         Bits 0-63 Reserved
1127  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1128  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1129  *       ibm,query-interrupt-source-number RTAS call, or as
1130  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1131  *
1132  * Output:
1133  * - R4: Target to which the specified Logical Interrupt Source is
1134  *       assigned
1135  * - R5: Priority to which the specified Logical Interrupt Source is
1136  *       assigned
1137  * - R6: EISN for the specified Logical Interrupt Source (this will be
1138  *       equivalent to the LISN if not changed by H_INT_SET_SOURCE_CONFIG)
1139  */
1140 static target_ulong h_int_get_source_config(PowerPCCPU *cpu,
1141                                             SpaprMachineState *spapr,
1142                                             target_ulong opcode,
1143                                             target_ulong *args)
1144 {
1145     SpaprXive *xive = spapr->xive;
1146     target_ulong flags = args[0];
1147     target_ulong lisn = args[1];
1148     XiveEAS eas;
1149     XiveEND *end;
1150     uint8_t nvt_blk;
1151     uint32_t end_idx, nvt_idx;
1152 
1153     trace_spapr_xive_get_source_config(flags, lisn);
1154 
1155     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1156         return H_FUNCTION;
1157     }
1158 
1159     if (flags) {
1160         return H_PARAMETER;
1161     }
1162 
1163     if (lisn >= xive->nr_irqs) {
1164         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1165                       lisn);
1166         return H_P2;
1167     }
1168 
1169     eas = xive->eat[lisn];
1170     if (!xive_eas_is_valid(&eas)) {
1171         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1172                       lisn);
1173         return H_P2;
1174     }
1175 
1176     /* EAS_END_BLOCK is unused on sPAPR */
1177     end_idx = xive_get_field64(EAS_END_INDEX, eas.w);
1178 
1179     assert(end_idx < xive->nr_ends);
1180     end = &xive->endt[end_idx];
1181 
1182     nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end->w6);
1183     nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end->w6);
1184     args[0] = spapr_xive_nvt_to_target(nvt_blk, nvt_idx);
1185 
1186     if (xive_eas_is_masked(&eas)) {
1187         args[1] = 0xff;
1188     } else {
1189         args[1] = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
1190     }
1191 
1192     args[2] = xive_get_field64(EAS_END_DATA, eas.w);
1193 
1194     return H_SUCCESS;
1195 }
1196 
1197 /*
1198  * The H_INT_GET_QUEUE_INFO hcall() is used to get the logical real
1199  * address of the notification management page associated with the
1200  * specified target and priority.
1201  *
1202  * Parameters:
1203  * Input:
1204  * - R4: "flags"
1205  *         Bits 0-63 Reserved
1206  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1207  *       "ibm,ppc-interrupt-gserver#s"
1208  * - R6: "priority" is a valid priority not in
1209  *       "ibm,plat-res-int-priorities"
1210  *
1211  * Output:
1212  * - R4: Logical real address of notification page
1213  * - R5: Power of 2 page size of the notification page
1214  */
1215 static target_ulong h_int_get_queue_info(PowerPCCPU *cpu,
1216                                          SpaprMachineState *spapr,
1217                                          target_ulong opcode,
1218                                          target_ulong *args)
1219 {
1220     SpaprXive *xive = spapr->xive;
1221     XiveENDSource *end_xsrc = &xive->end_source;
1222     target_ulong flags = args[0];
1223     target_ulong target = args[1];
1224     target_ulong priority = args[2];
1225     XiveEND *end;
1226     uint8_t end_blk;
1227     uint32_t end_idx;
1228 
1229     trace_spapr_xive_get_queue_info(flags, target, priority);
1230 
1231     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1232         return H_FUNCTION;
1233     }
1234 
1235     if (flags) {
1236         return H_PARAMETER;
1237     }
1238 
1239     /*
1240      * H_STATE should be returned if a H_INT_RESET is in progress.
1241      * This is not needed when running the emulation under QEMU
1242      */
1243 
1244     if (spapr_xive_priority_is_reserved(xive, priority)) {
1245         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1246                       " is reserved\n", priority);
1247         return H_P3;
1248     }
1249 
1250     /*
1251      * Validate that "target" is part of the list of threads allocated
1252      * to the partition. For that, find the END corresponding to the
1253      * target.
1254      */
1255     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1256         return H_P2;
1257     }
1258 
1259     assert(end_idx < xive->nr_ends);
1260     end = &xive->endt[end_idx];
1261 
1262     args[0] = xive->end_base + (1ull << (end_xsrc->esb_shift + 1)) * end_idx;
1263     if (xive_end_is_enqueue(end)) {
1264         args[1] = xive_get_field32(END_W0_QSIZE, end->w0) + 12;
1265     } else {
1266         args[1] = 0;
1267     }
1268 
1269     return H_SUCCESS;
1270 }
1271 
1272 /*
1273  * The H_INT_SET_QUEUE_CONFIG hcall() is used to set or reset a EQ for
1274  * a given "target" and "priority".  It is also used to set the
1275  * notification config associated with the EQ.  An EQ size of 0 is
1276  * used to reset the EQ config for a given target and priority. If
1277  * resetting the EQ config, the END associated with the given "target"
1278  * and "priority" will be changed to disable queueing.
1279  *
1280  * Upon return from the hcall(), no additional interrupts will be
1281  * directed to the old EQ (if one was set). The old EQ (if one was
1282  * set) should be investigated for interrupts that occurred prior to
1283  * or during the hcall().
1284  *
1285  * Parameters:
1286  * Input:
1287  * - R4: "flags"
1288  *         Bits 0-62: Reserved
1289  *         Bit 63: Unconditional Notify (n) per the XIVE spec
1290  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1291  *       "ibm,ppc-interrupt-gserver#s"
1292  * - R6: "priority" is a valid priority not in
1293  *       "ibm,plat-res-int-priorities"
1294  * - R7: "eventQueue": The logical real address of the start of the EQ
1295  * - R8: "eventQueueSize": The power of 2 EQ size per "ibm,xive-eq-sizes"
1296  *
1297  * Output:
1298  * - None
1299  */
1300 
1301 #define SPAPR_XIVE_END_ALWAYS_NOTIFY PPC_BIT(63)
1302 
1303 static target_ulong h_int_set_queue_config(PowerPCCPU *cpu,
1304                                            SpaprMachineState *spapr,
1305                                            target_ulong opcode,
1306                                            target_ulong *args)
1307 {
1308     SpaprXive *xive = spapr->xive;
1309     target_ulong flags = args[0];
1310     target_ulong target = args[1];
1311     target_ulong priority = args[2];
1312     target_ulong qpage = args[3];
1313     target_ulong qsize = args[4];
1314     XiveEND end;
1315     uint8_t end_blk, nvt_blk;
1316     uint32_t end_idx, nvt_idx;
1317 
1318     trace_spapr_xive_set_queue_config(flags, target, priority, qpage, qsize);
1319 
1320     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1321         return H_FUNCTION;
1322     }
1323 
1324     if (flags & ~SPAPR_XIVE_END_ALWAYS_NOTIFY) {
1325         return H_PARAMETER;
1326     }
1327 
1328     /*
1329      * H_STATE should be returned if a H_INT_RESET is in progress.
1330      * This is not needed when running the emulation under QEMU
1331      */
1332 
1333     if (spapr_xive_priority_is_reserved(xive, priority)) {
1334         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1335                       " is reserved\n", priority);
1336         return H_P3;
1337     }
1338 
1339     /*
1340      * Validate that "target" is part of the list of threads allocated
1341      * to the partition. For that, find the END corresponding to the
1342      * target.
1343      */
1344 
1345     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1346         return H_P2;
1347     }
1348 
1349     assert(end_idx < xive->nr_ends);
1350     memcpy(&end, &xive->endt[end_idx], sizeof(XiveEND));
1351 
1352     switch (qsize) {
1353     case 12:
1354     case 16:
1355     case 21:
1356     case 24:
1357         if (!QEMU_IS_ALIGNED(qpage, 1ul << qsize)) {
1358             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: EQ @0x%" HWADDR_PRIx
1359                           " is not naturally aligned with %" HWADDR_PRIx "\n",
1360                           qpage, (hwaddr)1 << qsize);
1361             return H_P4;
1362         }
1363         end.w2 = cpu_to_be32((qpage >> 32) & 0x0fffffff);
1364         end.w3 = cpu_to_be32(qpage & 0xffffffff);
1365         end.w0 |= cpu_to_be32(END_W0_ENQUEUE);
1366         end.w0 = xive_set_field32(END_W0_QSIZE, end.w0, qsize - 12);
1367         break;
1368     case 0:
1369         /* reset queue and disable queueing */
1370         spapr_xive_end_reset(&end);
1371         goto out;
1372 
1373     default:
1374         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid EQ size %"PRIx64"\n",
1375                       qsize);
1376         return H_P5;
1377     }
1378 
1379     if (qsize) {
1380         hwaddr plen = 1 << qsize;
1381         void *eq;
1382 
1383         /*
1384          * Validate the guest EQ. We should also check that the queue
1385          * has been zeroed by the OS.
1386          */
1387         eq = address_space_map(CPU(cpu)->as, qpage, &plen, true,
1388                                MEMTXATTRS_UNSPECIFIED);
1389         if (plen != 1 << qsize) {
1390             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to map EQ @0x%"
1391                           HWADDR_PRIx "\n", qpage);
1392             return H_P4;
1393         }
1394         address_space_unmap(CPU(cpu)->as, eq, plen, true, plen);
1395     }
1396 
1397     /* "target" should have been validated above */
1398     if (spapr_xive_target_to_nvt(target, &nvt_blk, &nvt_idx)) {
1399         g_assert_not_reached();
1400     }
1401 
1402     /*
1403      * Ensure the priority and target are correctly set (they will not
1404      * be right after allocation)
1405      */
1406     end.w6 = xive_set_field32(END_W6_NVT_BLOCK, 0ul, nvt_blk) |
1407         xive_set_field32(END_W6_NVT_INDEX, 0ul, nvt_idx);
1408     end.w7 = xive_set_field32(END_W7_F0_PRIORITY, 0ul, priority);
1409 
1410     if (flags & SPAPR_XIVE_END_ALWAYS_NOTIFY) {
1411         end.w0 |= cpu_to_be32(END_W0_UCOND_NOTIFY);
1412     } else {
1413         end.w0 &= cpu_to_be32((uint32_t)~END_W0_UCOND_NOTIFY);
1414     }
1415 
1416     /*
1417      * The generation bit for the END starts at 1 and The END page
1418      * offset counter starts at 0.
1419      */
1420     end.w1 = cpu_to_be32(END_W1_GENERATION) |
1421         xive_set_field32(END_W1_PAGE_OFF, 0ul, 0ul);
1422     end.w0 |= cpu_to_be32(END_W0_VALID);
1423 
1424     /*
1425      * TODO: issue syncs required to ensure all in-flight interrupts
1426      * are complete on the old END
1427      */
1428 
1429 out:
1430     if (spapr_xive_in_kernel(xive)) {
1431         Error *local_err = NULL;
1432 
1433         kvmppc_xive_set_queue_config(xive, end_blk, end_idx, &end, &local_err);
1434         if (local_err) {
1435             error_report_err(local_err);
1436             return H_HARDWARE;
1437         }
1438     }
1439 
1440     /* Update END */
1441     memcpy(&xive->endt[end_idx], &end, sizeof(XiveEND));
1442     return H_SUCCESS;
1443 }
1444 
1445 /*
1446  * The H_INT_GET_QUEUE_CONFIG hcall() is used to get a EQ for a given
1447  * target and priority.
1448  *
1449  * Parameters:
1450  * Input:
1451  * - R4: "flags"
1452  *         Bits 0-62: Reserved
1453  *         Bit 63: Debug: Return debug data
1454  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1455  *       "ibm,ppc-interrupt-gserver#s"
1456  * - R6: "priority" is a valid priority not in
1457  *       "ibm,plat-res-int-priorities"
1458  *
1459  * Output:
1460  * - R4: "flags":
1461  *       Bits 0-61: Reserved
1462  *       Bit 62: The value of Event Queue Generation Number (g) per
1463  *              the XIVE spec if "Debug" = 1
1464  *       Bit 63: The value of Unconditional Notify (n) per the XIVE spec
1465  * - R5: The logical real address of the start of the EQ
1466  * - R6: The power of 2 EQ size per "ibm,xive-eq-sizes"
1467  * - R7: The value of Event Queue Offset Counter per XIVE spec
1468  *       if "Debug" = 1, else 0
1469  *
1470  */
1471 
1472 #define SPAPR_XIVE_END_DEBUG     PPC_BIT(63)
1473 
1474 static target_ulong h_int_get_queue_config(PowerPCCPU *cpu,
1475                                            SpaprMachineState *spapr,
1476                                            target_ulong opcode,
1477                                            target_ulong *args)
1478 {
1479     SpaprXive *xive = spapr->xive;
1480     target_ulong flags = args[0];
1481     target_ulong target = args[1];
1482     target_ulong priority = args[2];
1483     XiveEND *end;
1484     uint8_t end_blk;
1485     uint32_t end_idx;
1486 
1487     trace_spapr_xive_get_queue_config(flags, target, priority);
1488 
1489     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1490         return H_FUNCTION;
1491     }
1492 
1493     if (flags & ~SPAPR_XIVE_END_DEBUG) {
1494         return H_PARAMETER;
1495     }
1496 
1497     /*
1498      * H_STATE should be returned if a H_INT_RESET is in progress.
1499      * This is not needed when running the emulation under QEMU
1500      */
1501 
1502     if (spapr_xive_priority_is_reserved(xive, priority)) {
1503         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1504                       " is reserved\n", priority);
1505         return H_P3;
1506     }
1507 
1508     /*
1509      * Validate that "target" is part of the list of threads allocated
1510      * to the partition. For that, find the END corresponding to the
1511      * target.
1512      */
1513     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1514         return H_P2;
1515     }
1516 
1517     assert(end_idx < xive->nr_ends);
1518     end = &xive->endt[end_idx];
1519 
1520     args[0] = 0;
1521     if (xive_end_is_notify(end)) {
1522         args[0] |= SPAPR_XIVE_END_ALWAYS_NOTIFY;
1523     }
1524 
1525     if (xive_end_is_enqueue(end)) {
1526         args[1] = xive_end_qaddr(end);
1527         args[2] = xive_get_field32(END_W0_QSIZE, end->w0) + 12;
1528     } else {
1529         args[1] = 0;
1530         args[2] = 0;
1531     }
1532 
1533     if (spapr_xive_in_kernel(xive)) {
1534         Error *local_err = NULL;
1535 
1536         kvmppc_xive_get_queue_config(xive, end_blk, end_idx, end, &local_err);
1537         if (local_err) {
1538             error_report_err(local_err);
1539             return H_HARDWARE;
1540         }
1541     }
1542 
1543     /* TODO: do we need any locking on the END ? */
1544     if (flags & SPAPR_XIVE_END_DEBUG) {
1545         /* Load the event queue generation number into the return flags */
1546         args[0] |= (uint64_t)xive_get_field32(END_W1_GENERATION, end->w1) << 62;
1547 
1548         /* Load R7 with the event queue offset counter */
1549         args[3] = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1550     } else {
1551         args[3] = 0;
1552     }
1553 
1554     return H_SUCCESS;
1555 }
1556 
1557 /*
1558  * The H_INT_SET_OS_REPORTING_LINE hcall() is used to set the
1559  * reporting cache line pair for the calling thread.  The reporting
1560  * cache lines will contain the OS interrupt context when the OS
1561  * issues a CI store byte to @TIMA+0xC10 to acknowledge the OS
1562  * interrupt. The reporting cache lines can be reset by inputting -1
1563  * in "reportingLine".  Issuing the CI store byte without reporting
1564  * cache lines registered will result in the data not being accessible
1565  * to the OS.
1566  *
1567  * Parameters:
1568  * Input:
1569  * - R4: "flags"
1570  *         Bits 0-63: Reserved
1571  * - R5: "reportingLine": The logical real address of the reporting cache
1572  *       line pair
1573  *
1574  * Output:
1575  * - None
1576  */
1577 static target_ulong h_int_set_os_reporting_line(PowerPCCPU *cpu,
1578                                                 SpaprMachineState *spapr,
1579                                                 target_ulong opcode,
1580                                                 target_ulong *args)
1581 {
1582     target_ulong flags   = args[0];
1583 
1584     trace_spapr_xive_set_os_reporting_line(flags);
1585 
1586     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1587         return H_FUNCTION;
1588     }
1589 
1590     /*
1591      * H_STATE should be returned if a H_INT_RESET is in progress.
1592      * This is not needed when running the emulation under QEMU
1593      */
1594 
1595     /* TODO: H_INT_SET_OS_REPORTING_LINE */
1596     return H_FUNCTION;
1597 }
1598 
1599 /*
1600  * The H_INT_GET_OS_REPORTING_LINE hcall() is used to get the logical
1601  * real address of the reporting cache line pair set for the input
1602  * "target".  If no reporting cache line pair has been set, -1 is
1603  * returned.
1604  *
1605  * Parameters:
1606  * Input:
1607  * - R4: "flags"
1608  *         Bits 0-63: Reserved
1609  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1610  *       "ibm,ppc-interrupt-gserver#s"
1611  * - R6: "reportingLine": The logical real address of the reporting
1612  *        cache line pair
1613  *
1614  * Output:
1615  * - R4: The logical real address of the reporting line if set, else -1
1616  */
1617 static target_ulong h_int_get_os_reporting_line(PowerPCCPU *cpu,
1618                                                 SpaprMachineState *spapr,
1619                                                 target_ulong opcode,
1620                                                 target_ulong *args)
1621 {
1622     target_ulong flags   = args[0];
1623 
1624     trace_spapr_xive_get_os_reporting_line(flags);
1625 
1626     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1627         return H_FUNCTION;
1628     }
1629 
1630     /*
1631      * H_STATE should be returned if a H_INT_RESET is in progress.
1632      * This is not needed when running the emulation under QEMU
1633      */
1634 
1635     /* TODO: H_INT_GET_OS_REPORTING_LINE */
1636     return H_FUNCTION;
1637 }
1638 
1639 /*
1640  * The H_INT_ESB hcall() is used to issue a load or store to the ESB
1641  * page for the input "lisn".  This hcall is only supported for LISNs
1642  * that have the ESB hcall flag set to 1 when returned from hcall()
1643  * H_INT_GET_SOURCE_INFO.
1644  *
1645  * Parameters:
1646  * Input:
1647  * - R4: "flags"
1648  *         Bits 0-62: Reserved
1649  *         bit 63: Store: Store=1, store operation, else load operation
1650  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1651  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1652  *       ibm,query-interrupt-source-number RTAS call, or as
1653  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1654  * - R6: "esbOffset" is the offset into the ESB page for the load or
1655  *       store operation
1656  * - R7: "storeData" is the data to write for a store operation
1657  *
1658  * Output:
1659  * - R4: The value of the load if load operation, else -1
1660  */
1661 
1662 #define SPAPR_XIVE_ESB_STORE PPC_BIT(63)
1663 
1664 static target_ulong h_int_esb(PowerPCCPU *cpu,
1665                               SpaprMachineState *spapr,
1666                               target_ulong opcode,
1667                               target_ulong *args)
1668 {
1669     SpaprXive *xive = spapr->xive;
1670     XiveEAS eas;
1671     target_ulong flags  = args[0];
1672     target_ulong lisn   = args[1];
1673     target_ulong offset = args[2];
1674     target_ulong data   = args[3];
1675     hwaddr mmio_addr;
1676     XiveSource *xsrc = &xive->source;
1677 
1678     trace_spapr_xive_esb(flags, lisn, offset, data);
1679 
1680     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1681         return H_FUNCTION;
1682     }
1683 
1684     if (flags & ~SPAPR_XIVE_ESB_STORE) {
1685         return H_PARAMETER;
1686     }
1687 
1688     if (lisn >= xive->nr_irqs) {
1689         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1690                       lisn);
1691         return H_P2;
1692     }
1693 
1694     eas = xive->eat[lisn];
1695     if (!xive_eas_is_valid(&eas)) {
1696         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1697                       lisn);
1698         return H_P2;
1699     }
1700 
1701     if (offset > (1ull << xsrc->esb_shift)) {
1702         return H_P3;
1703     }
1704 
1705     if (spapr_xive_in_kernel(xive)) {
1706         args[0] = kvmppc_xive_esb_rw(xsrc, lisn, offset, data,
1707                                      flags & SPAPR_XIVE_ESB_STORE);
1708     } else {
1709         mmio_addr = xive->vc_base + xive_source_esb_mgmt(xsrc, lisn) + offset;
1710 
1711         if (dma_memory_rw(&address_space_memory, mmio_addr, &data, 8,
1712                           (flags & SPAPR_XIVE_ESB_STORE),
1713                           MEMTXATTRS_UNSPECIFIED)) {
1714             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to access ESB @0x%"
1715                           HWADDR_PRIx "\n", mmio_addr);
1716             return H_HARDWARE;
1717         }
1718         args[0] = (flags & SPAPR_XIVE_ESB_STORE) ? -1 : data;
1719     }
1720     return H_SUCCESS;
1721 }
1722 
1723 /*
1724  * The H_INT_SYNC hcall() is used to issue hardware syncs that will
1725  * ensure any in flight events for the input lisn are in the event
1726  * queue.
1727  *
1728  * Parameters:
1729  * Input:
1730  * - R4: "flags"
1731  *         Bits 0-63: Reserved
1732  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1733  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1734  *       ibm,query-interrupt-source-number RTAS call, or as
1735  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1736  *
1737  * Output:
1738  * - None
1739  */
1740 static target_ulong h_int_sync(PowerPCCPU *cpu,
1741                                SpaprMachineState *spapr,
1742                                target_ulong opcode,
1743                                target_ulong *args)
1744 {
1745     SpaprXive *xive = spapr->xive;
1746     XiveEAS eas;
1747     target_ulong flags = args[0];
1748     target_ulong lisn = args[1];
1749 
1750     trace_spapr_xive_sync(flags, lisn);
1751 
1752     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1753         return H_FUNCTION;
1754     }
1755 
1756     if (flags) {
1757         return H_PARAMETER;
1758     }
1759 
1760     if (lisn >= xive->nr_irqs) {
1761         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1762                       lisn);
1763         return H_P2;
1764     }
1765 
1766     eas = xive->eat[lisn];
1767     if (!xive_eas_is_valid(&eas)) {
1768         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1769                       lisn);
1770         return H_P2;
1771     }
1772 
1773     /*
1774      * H_STATE should be returned if a H_INT_RESET is in progress.
1775      * This is not needed when running the emulation under QEMU
1776      */
1777 
1778     /*
1779      * This is not real hardware. Nothing to be done unless when
1780      * under KVM
1781      */
1782 
1783     if (spapr_xive_in_kernel(xive)) {
1784         Error *local_err = NULL;
1785 
1786         kvmppc_xive_sync_source(xive, lisn, &local_err);
1787         if (local_err) {
1788             error_report_err(local_err);
1789             return H_HARDWARE;
1790         }
1791     }
1792     return H_SUCCESS;
1793 }
1794 
1795 /*
1796  * The H_INT_RESET hcall() is used to reset all of the partition's
1797  * interrupt exploitation structures to their initial state.  This
1798  * means losing all previously set interrupt state set via
1799  * H_INT_SET_SOURCE_CONFIG and H_INT_SET_QUEUE_CONFIG.
1800  *
1801  * Parameters:
1802  * Input:
1803  * - R4: "flags"
1804  *         Bits 0-63: Reserved
1805  *
1806  * Output:
1807  * - None
1808  */
1809 static target_ulong h_int_reset(PowerPCCPU *cpu,
1810                                 SpaprMachineState *spapr,
1811                                 target_ulong opcode,
1812                                 target_ulong *args)
1813 {
1814     SpaprXive *xive = spapr->xive;
1815     target_ulong flags   = args[0];
1816 
1817     trace_spapr_xive_reset(flags);
1818 
1819     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1820         return H_FUNCTION;
1821     }
1822 
1823     if (flags) {
1824         return H_PARAMETER;
1825     }
1826 
1827     device_cold_reset(DEVICE(xive));
1828 
1829     if (spapr_xive_in_kernel(xive)) {
1830         Error *local_err = NULL;
1831 
1832         kvmppc_xive_reset(xive, &local_err);
1833         if (local_err) {
1834             error_report_err(local_err);
1835             return H_HARDWARE;
1836         }
1837     }
1838     return H_SUCCESS;
1839 }
1840 
1841 void spapr_xive_hcall_init(SpaprMachineState *spapr)
1842 {
1843     spapr_register_hypercall(H_INT_GET_SOURCE_INFO, h_int_get_source_info);
1844     spapr_register_hypercall(H_INT_SET_SOURCE_CONFIG, h_int_set_source_config);
1845     spapr_register_hypercall(H_INT_GET_SOURCE_CONFIG, h_int_get_source_config);
1846     spapr_register_hypercall(H_INT_GET_QUEUE_INFO, h_int_get_queue_info);
1847     spapr_register_hypercall(H_INT_SET_QUEUE_CONFIG, h_int_set_queue_config);
1848     spapr_register_hypercall(H_INT_GET_QUEUE_CONFIG, h_int_get_queue_config);
1849     spapr_register_hypercall(H_INT_SET_OS_REPORTING_LINE,
1850                              h_int_set_os_reporting_line);
1851     spapr_register_hypercall(H_INT_GET_OS_REPORTING_LINE,
1852                              h_int_get_os_reporting_line);
1853     spapr_register_hypercall(H_INT_ESB, h_int_esb);
1854     spapr_register_hypercall(H_INT_SYNC, h_int_sync);
1855     spapr_register_hypercall(H_INT_RESET, h_int_reset);
1856 }
1857