xref: /qemu/hw/intc/spapr_xive.c (revision 6f9ff551)
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 /*
152  * kvm_irqchip_in_kernel() will cause the compiler to turn this
153  * info a nop if CONFIG_KVM isn't defined.
154  */
155 #define spapr_xive_in_kernel(xive) \
156     (kvm_irqchip_in_kernel() && (xive)->fd != -1)
157 
158 void spapr_xive_pic_print_info(SpaprXive *xive, Monitor *mon)
159 {
160     XiveSource *xsrc = &xive->source;
161     int i;
162 
163     if (spapr_xive_in_kernel(xive)) {
164         Error *local_err = NULL;
165 
166         kvmppc_xive_synchronize_state(xive, &local_err);
167         if (local_err) {
168             error_report_err(local_err);
169             return;
170         }
171     }
172 
173     monitor_printf(mon, "  LISN         PQ    EISN     CPU/PRIO EQ\n");
174 
175     for (i = 0; i < xive->nr_irqs; i++) {
176         uint8_t pq = xive_source_esb_get(xsrc, i);
177         XiveEAS *eas = &xive->eat[i];
178 
179         if (!xive_eas_is_valid(eas)) {
180             continue;
181         }
182 
183         monitor_printf(mon, "  %08x %s %c%c%c %s %08x ", i,
184                        xive_source_irq_is_lsi(xsrc, i) ? "LSI" : "MSI",
185                        pq & XIVE_ESB_VAL_P ? 'P' : '-',
186                        pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
187                        xsrc->status[i] & XIVE_STATUS_ASSERTED ? 'A' : ' ',
188                        xive_eas_is_masked(eas) ? "M" : " ",
189                        (int) xive_get_field64(EAS_END_DATA, eas->w));
190 
191         if (!xive_eas_is_masked(eas)) {
192             uint32_t end_idx = xive_get_field64(EAS_END_INDEX, eas->w);
193             XiveEND *end;
194 
195             assert(end_idx < xive->nr_ends);
196             end = &xive->endt[end_idx];
197 
198             if (xive_end_is_valid(end)) {
199                 spapr_xive_end_pic_print_info(xive, end, mon);
200             }
201         }
202         monitor_printf(mon, "\n");
203     }
204 }
205 
206 void spapr_xive_mmio_set_enabled(SpaprXive *xive, bool enable)
207 {
208     memory_region_set_enabled(&xive->source.esb_mmio, enable);
209     memory_region_set_enabled(&xive->tm_mmio, enable);
210 
211     /* Disable the END ESBs until a guest OS makes use of them */
212     memory_region_set_enabled(&xive->end_source.esb_mmio, false);
213 }
214 
215 static void spapr_xive_tm_write(void *opaque, hwaddr offset,
216                           uint64_t value, unsigned size)
217 {
218     XiveTCTX *tctx = spapr_cpu_state(POWERPC_CPU(current_cpu))->tctx;
219 
220     xive_tctx_tm_write(XIVE_PRESENTER(opaque), tctx, offset, value, size);
221 }
222 
223 static uint64_t spapr_xive_tm_read(void *opaque, hwaddr offset, unsigned size)
224 {
225     XiveTCTX *tctx = spapr_cpu_state(POWERPC_CPU(current_cpu))->tctx;
226 
227     return xive_tctx_tm_read(XIVE_PRESENTER(opaque), tctx, offset, size);
228 }
229 
230 const MemoryRegionOps spapr_xive_tm_ops = {
231     .read = spapr_xive_tm_read,
232     .write = spapr_xive_tm_write,
233     .endianness = DEVICE_BIG_ENDIAN,
234     .valid = {
235         .min_access_size = 1,
236         .max_access_size = 8,
237     },
238     .impl = {
239         .min_access_size = 1,
240         .max_access_size = 8,
241     },
242 };
243 
244 static void spapr_xive_end_reset(XiveEND *end)
245 {
246     memset(end, 0, sizeof(*end));
247 
248     /* switch off the escalation and notification ESBs */
249     end->w1 = cpu_to_be32(END_W1_ESe_Q | END_W1_ESn_Q);
250 }
251 
252 static void spapr_xive_reset(void *dev)
253 {
254     SpaprXive *xive = SPAPR_XIVE(dev);
255     int i;
256 
257     /*
258      * The XiveSource has its own reset handler, which mask off all
259      * IRQs (!P|Q)
260      */
261 
262     /* Mask all valid EASs in the IRQ number space. */
263     for (i = 0; i < xive->nr_irqs; i++) {
264         XiveEAS *eas = &xive->eat[i];
265         if (xive_eas_is_valid(eas)) {
266             eas->w = cpu_to_be64(EAS_VALID | EAS_MASKED);
267         } else {
268             eas->w = 0;
269         }
270     }
271 
272     /* Clear all ENDs */
273     for (i = 0; i < xive->nr_ends; i++) {
274         spapr_xive_end_reset(&xive->endt[i]);
275     }
276 }
277 
278 static void spapr_xive_instance_init(Object *obj)
279 {
280     SpaprXive *xive = SPAPR_XIVE(obj);
281 
282     object_initialize_child(obj, "source", &xive->source, TYPE_XIVE_SOURCE);
283 
284     object_initialize_child(obj, "end_source", &xive->end_source,
285                             TYPE_XIVE_END_SOURCE);
286 
287     /* Not connected to the KVM XIVE device */
288     xive->fd = -1;
289 }
290 
291 static void spapr_xive_realize(DeviceState *dev, Error **errp)
292 {
293     SpaprXive *xive = SPAPR_XIVE(dev);
294     SpaprXiveClass *sxc = SPAPR_XIVE_GET_CLASS(xive);
295     XiveSource *xsrc = &xive->source;
296     XiveENDSource *end_xsrc = &xive->end_source;
297     Error *local_err = NULL;
298 
299     sxc->parent_realize(dev, &local_err);
300     if (local_err) {
301         error_propagate(errp, local_err);
302         return;
303     }
304 
305     if (!xive->nr_irqs) {
306         error_setg(errp, "Number of interrupt needs to be greater 0");
307         return;
308     }
309 
310     if (!xive->nr_ends) {
311         error_setg(errp, "Number of interrupt needs to be greater 0");
312         return;
313     }
314 
315     /*
316      * Initialize the internal sources, for IPIs and virtual devices.
317      */
318     object_property_set_int(OBJECT(xsrc), "nr-irqs", xive->nr_irqs,
319                             &error_fatal);
320     object_property_set_link(OBJECT(xsrc), "xive", OBJECT(xive), &error_abort);
321     if (!qdev_realize(DEVICE(xsrc), NULL, errp)) {
322         return;
323     }
324     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &xsrc->esb_mmio);
325 
326     /*
327      * Initialize the END ESB source
328      */
329     object_property_set_int(OBJECT(end_xsrc), "nr-ends", xive->nr_irqs,
330                             &error_fatal);
331     object_property_set_link(OBJECT(end_xsrc), "xive", OBJECT(xive),
332                              &error_abort);
333     if (!qdev_realize(DEVICE(end_xsrc), NULL, errp)) {
334         return;
335     }
336     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &end_xsrc->esb_mmio);
337 
338     /* Set the mapping address of the END ESB pages after the source ESBs */
339     xive->end_base = xive->vc_base + xive_source_esb_len(xsrc);
340 
341     /*
342      * Allocate the routing tables
343      */
344     xive->eat = g_new0(XiveEAS, xive->nr_irqs);
345     xive->endt = g_new0(XiveEND, xive->nr_ends);
346 
347     xive->nodename = g_strdup_printf("interrupt-controller@%" PRIx64,
348                            xive->tm_base + XIVE_TM_USER_PAGE * (1 << TM_SHIFT));
349 
350     qemu_register_reset(spapr_xive_reset, dev);
351 
352     /* TIMA initialization */
353     memory_region_init_io(&xive->tm_mmio, OBJECT(xive), &spapr_xive_tm_ops,
354                           xive, "xive.tima", 4ull << TM_SHIFT);
355     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &xive->tm_mmio);
356 
357     /*
358      * Map all regions. These will be enabled or disabled at reset and
359      * can also be overridden by KVM memory regions if active
360      */
361     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 0, xive->vc_base);
362     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 1, xive->end_base);
363     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 2, xive->tm_base);
364 }
365 
366 static int spapr_xive_get_eas(XiveRouter *xrtr, uint8_t eas_blk,
367                               uint32_t eas_idx, XiveEAS *eas)
368 {
369     SpaprXive *xive = SPAPR_XIVE(xrtr);
370 
371     if (eas_idx >= xive->nr_irqs) {
372         return -1;
373     }
374 
375     *eas = xive->eat[eas_idx];
376     return 0;
377 }
378 
379 static int spapr_xive_get_end(XiveRouter *xrtr,
380                               uint8_t end_blk, uint32_t end_idx, XiveEND *end)
381 {
382     SpaprXive *xive = SPAPR_XIVE(xrtr);
383 
384     if (end_idx >= xive->nr_ends) {
385         return -1;
386     }
387 
388     memcpy(end, &xive->endt[end_idx], sizeof(XiveEND));
389     return 0;
390 }
391 
392 static int spapr_xive_write_end(XiveRouter *xrtr, uint8_t end_blk,
393                                 uint32_t end_idx, XiveEND *end,
394                                 uint8_t word_number)
395 {
396     SpaprXive *xive = SPAPR_XIVE(xrtr);
397 
398     if (end_idx >= xive->nr_ends) {
399         return -1;
400     }
401 
402     memcpy(&xive->endt[end_idx], end, sizeof(XiveEND));
403     return 0;
404 }
405 
406 static int spapr_xive_get_nvt(XiveRouter *xrtr,
407                               uint8_t nvt_blk, uint32_t nvt_idx, XiveNVT *nvt)
408 {
409     uint32_t vcpu_id = spapr_xive_nvt_to_target(nvt_blk, nvt_idx);
410     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
411 
412     if (!cpu) {
413         /* TODO: should we assert() if we can find a NVT ? */
414         return -1;
415     }
416 
417     /*
418      * sPAPR does not maintain a NVT table. Return that the NVT is
419      * valid if we have found a matching CPU
420      */
421     nvt->w0 = cpu_to_be32(NVT_W0_VALID);
422     return 0;
423 }
424 
425 static int spapr_xive_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk,
426                                 uint32_t nvt_idx, XiveNVT *nvt,
427                                 uint8_t word_number)
428 {
429     /*
430      * We don't need to write back to the NVTs because the sPAPR
431      * machine should never hit a non-scheduled NVT. It should never
432      * get called.
433      */
434     g_assert_not_reached();
435 }
436 
437 static int spapr_xive_match_nvt(XivePresenter *xptr, uint8_t format,
438                                 uint8_t nvt_blk, uint32_t nvt_idx,
439                                 bool cam_ignore, uint8_t priority,
440                                 uint32_t logic_serv, XiveTCTXMatch *match)
441 {
442     CPUState *cs;
443     int count = 0;
444 
445     CPU_FOREACH(cs) {
446         PowerPCCPU *cpu = POWERPC_CPU(cs);
447         XiveTCTX *tctx = spapr_cpu_state(cpu)->tctx;
448         int ring;
449 
450         /*
451          * Skip partially initialized vCPUs. This can happen when
452          * vCPUs are hotplugged.
453          */
454         if (!tctx) {
455             continue;
456         }
457 
458         /*
459          * Check the thread context CAM lines and record matches.
460          */
461         ring = xive_presenter_tctx_match(xptr, tctx, format, nvt_blk, nvt_idx,
462                                          cam_ignore, logic_serv);
463         /*
464          * Save the matching thread interrupt context and follow on to
465          * check for duplicates which are invalid.
466          */
467         if (ring != -1) {
468             if (match->tctx) {
469                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: already found a thread "
470                               "context NVT %x/%x\n", nvt_blk, nvt_idx);
471                 return -1;
472             }
473 
474             match->ring = ring;
475             match->tctx = tctx;
476             count++;
477         }
478     }
479 
480     return count;
481 }
482 
483 static uint8_t spapr_xive_get_block_id(XiveRouter *xrtr)
484 {
485     return SPAPR_XIVE_BLOCK_ID;
486 }
487 
488 static const VMStateDescription vmstate_spapr_xive_end = {
489     .name = TYPE_SPAPR_XIVE "/end",
490     .version_id = 1,
491     .minimum_version_id = 1,
492     .fields = (VMStateField []) {
493         VMSTATE_UINT32(w0, XiveEND),
494         VMSTATE_UINT32(w1, XiveEND),
495         VMSTATE_UINT32(w2, XiveEND),
496         VMSTATE_UINT32(w3, XiveEND),
497         VMSTATE_UINT32(w4, XiveEND),
498         VMSTATE_UINT32(w5, XiveEND),
499         VMSTATE_UINT32(w6, XiveEND),
500         VMSTATE_UINT32(w7, XiveEND),
501         VMSTATE_END_OF_LIST()
502     },
503 };
504 
505 static const VMStateDescription vmstate_spapr_xive_eas = {
506     .name = TYPE_SPAPR_XIVE "/eas",
507     .version_id = 1,
508     .minimum_version_id = 1,
509     .fields = (VMStateField []) {
510         VMSTATE_UINT64(w, XiveEAS),
511         VMSTATE_END_OF_LIST()
512     },
513 };
514 
515 static int vmstate_spapr_xive_pre_save(void *opaque)
516 {
517     SpaprXive *xive = SPAPR_XIVE(opaque);
518 
519     if (spapr_xive_in_kernel(xive)) {
520         return kvmppc_xive_pre_save(xive);
521     }
522 
523     return 0;
524 }
525 
526 /*
527  * Called by the sPAPR IRQ backend 'post_load' method at the machine
528  * level.
529  */
530 static int spapr_xive_post_load(SpaprInterruptController *intc, int version_id)
531 {
532     SpaprXive *xive = SPAPR_XIVE(intc);
533 
534     if (spapr_xive_in_kernel(xive)) {
535         return kvmppc_xive_post_load(xive, version_id);
536     }
537 
538     return 0;
539 }
540 
541 static const VMStateDescription vmstate_spapr_xive = {
542     .name = TYPE_SPAPR_XIVE,
543     .version_id = 1,
544     .minimum_version_id = 1,
545     .pre_save = vmstate_spapr_xive_pre_save,
546     .post_load = NULL, /* handled at the machine level */
547     .fields = (VMStateField[]) {
548         VMSTATE_UINT32_EQUAL(nr_irqs, SpaprXive, NULL),
549         VMSTATE_STRUCT_VARRAY_POINTER_UINT32(eat, SpaprXive, nr_irqs,
550                                      vmstate_spapr_xive_eas, XiveEAS),
551         VMSTATE_STRUCT_VARRAY_POINTER_UINT32(endt, SpaprXive, nr_ends,
552                                              vmstate_spapr_xive_end, XiveEND),
553         VMSTATE_END_OF_LIST()
554     },
555 };
556 
557 static int spapr_xive_claim_irq(SpaprInterruptController *intc, int lisn,
558                                 bool lsi, Error **errp)
559 {
560     SpaprXive *xive = SPAPR_XIVE(intc);
561     XiveSource *xsrc = &xive->source;
562 
563     assert(lisn < xive->nr_irqs);
564 
565     if (xive_eas_is_valid(&xive->eat[lisn])) {
566         error_setg(errp, "IRQ %d is not free", lisn);
567         return -EBUSY;
568     }
569 
570     /*
571      * Set default values when allocating an IRQ number
572      */
573     xive->eat[lisn].w |= cpu_to_be64(EAS_VALID | EAS_MASKED);
574     if (lsi) {
575         xive_source_irq_set_lsi(xsrc, lisn);
576     }
577 
578     if (spapr_xive_in_kernel(xive)) {
579         return kvmppc_xive_source_reset_one(xsrc, lisn, errp);
580     }
581 
582     return 0;
583 }
584 
585 static void spapr_xive_free_irq(SpaprInterruptController *intc, int lisn)
586 {
587     SpaprXive *xive = SPAPR_XIVE(intc);
588     assert(lisn < xive->nr_irqs);
589 
590     xive->eat[lisn].w &= cpu_to_be64(~EAS_VALID);
591 }
592 
593 static Property spapr_xive_properties[] = {
594     DEFINE_PROP_UINT32("nr-irqs", SpaprXive, nr_irqs, 0),
595     DEFINE_PROP_UINT32("nr-ends", SpaprXive, nr_ends, 0),
596     DEFINE_PROP_UINT64("vc-base", SpaprXive, vc_base, SPAPR_XIVE_VC_BASE),
597     DEFINE_PROP_UINT64("tm-base", SpaprXive, tm_base, SPAPR_XIVE_TM_BASE),
598     DEFINE_PROP_UINT8("hv-prio", SpaprXive, hv_prio, 7),
599     DEFINE_PROP_END_OF_LIST(),
600 };
601 
602 static int spapr_xive_cpu_intc_create(SpaprInterruptController *intc,
603                                       PowerPCCPU *cpu, Error **errp)
604 {
605     SpaprXive *xive = SPAPR_XIVE(intc);
606     Object *obj;
607     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
608 
609     obj = xive_tctx_create(OBJECT(cpu), XIVE_PRESENTER(xive), errp);
610     if (!obj) {
611         return -1;
612     }
613 
614     spapr_cpu->tctx = XIVE_TCTX(obj);
615     return 0;
616 }
617 
618 static void xive_tctx_set_os_cam(XiveTCTX *tctx, uint32_t os_cam)
619 {
620     uint32_t qw1w2 = cpu_to_be32(TM_QW1W2_VO | os_cam);
621     memcpy(&tctx->regs[TM_QW1_OS + TM_WORD2], &qw1w2, 4);
622 }
623 
624 static void spapr_xive_cpu_intc_reset(SpaprInterruptController *intc,
625                                      PowerPCCPU *cpu)
626 {
627     XiveTCTX *tctx = spapr_cpu_state(cpu)->tctx;
628     uint8_t  nvt_blk;
629     uint32_t nvt_idx;
630 
631     xive_tctx_reset(tctx);
632 
633     /*
634      * When a Virtual Processor is scheduled to run on a HW thread,
635      * the hypervisor pushes its identifier in the OS CAM line.
636      * Emulate the same behavior under QEMU.
637      */
638     spapr_xive_cpu_to_nvt(cpu, &nvt_blk, &nvt_idx);
639 
640     xive_tctx_set_os_cam(tctx, xive_nvt_cam_line(nvt_blk, nvt_idx));
641 }
642 
643 static void spapr_xive_cpu_intc_destroy(SpaprInterruptController *intc,
644                                         PowerPCCPU *cpu)
645 {
646     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
647 
648     xive_tctx_destroy(spapr_cpu->tctx);
649     spapr_cpu->tctx = NULL;
650 }
651 
652 static void spapr_xive_set_irq(SpaprInterruptController *intc, int irq, int val)
653 {
654     SpaprXive *xive = SPAPR_XIVE(intc);
655 
656     if (spapr_xive_in_kernel(xive)) {
657         kvmppc_xive_source_set_irq(&xive->source, irq, val);
658     } else {
659         xive_source_set_irq(&xive->source, irq, val);
660     }
661 }
662 
663 static void spapr_xive_print_info(SpaprInterruptController *intc, Monitor *mon)
664 {
665     SpaprXive *xive = SPAPR_XIVE(intc);
666     CPUState *cs;
667 
668     CPU_FOREACH(cs) {
669         PowerPCCPU *cpu = POWERPC_CPU(cs);
670 
671         xive_tctx_pic_print_info(spapr_cpu_state(cpu)->tctx, mon);
672     }
673 
674     spapr_xive_pic_print_info(xive, mon);
675 }
676 
677 static void spapr_xive_dt(SpaprInterruptController *intc, uint32_t nr_servers,
678                           void *fdt, uint32_t phandle)
679 {
680     SpaprXive *xive = SPAPR_XIVE(intc);
681     int node;
682     uint64_t timas[2 * 2];
683     /* Interrupt number ranges for the IPIs */
684     uint32_t lisn_ranges[] = {
685         cpu_to_be32(SPAPR_IRQ_IPI),
686         cpu_to_be32(SPAPR_IRQ_IPI + nr_servers),
687     };
688     /*
689      * EQ size - the sizes of pages supported by the system 4K, 64K,
690      * 2M, 16M. We only advertise 64K for the moment.
691      */
692     uint32_t eq_sizes[] = {
693         cpu_to_be32(16), /* 64K */
694     };
695     /*
696      * QEMU/KVM only needs to define a single range to reserve the
697      * escalation priority. A priority bitmask would have been more
698      * appropriate.
699      */
700     uint32_t plat_res_int_priorities[] = {
701         cpu_to_be32(xive->hv_prio),    /* start */
702         cpu_to_be32(0xff - xive->hv_prio), /* count */
703     };
704 
705     /* Thread Interrupt Management Area : User (ring 3) and OS (ring 2) */
706     timas[0] = cpu_to_be64(xive->tm_base +
707                            XIVE_TM_USER_PAGE * (1ull << TM_SHIFT));
708     timas[1] = cpu_to_be64(1ull << TM_SHIFT);
709     timas[2] = cpu_to_be64(xive->tm_base +
710                            XIVE_TM_OS_PAGE * (1ull << TM_SHIFT));
711     timas[3] = cpu_to_be64(1ull << TM_SHIFT);
712 
713     _FDT(node = fdt_add_subnode(fdt, 0, xive->nodename));
714 
715     _FDT(fdt_setprop_string(fdt, node, "device_type", "power-ivpe"));
716     _FDT(fdt_setprop(fdt, node, "reg", timas, sizeof(timas)));
717 
718     _FDT(fdt_setprop_string(fdt, node, "compatible", "ibm,power-ivpe"));
719     _FDT(fdt_setprop(fdt, node, "ibm,xive-eq-sizes", eq_sizes,
720                      sizeof(eq_sizes)));
721     _FDT(fdt_setprop(fdt, node, "ibm,xive-lisn-ranges", lisn_ranges,
722                      sizeof(lisn_ranges)));
723 
724     /* For Linux to link the LSIs to the interrupt controller. */
725     _FDT(fdt_setprop(fdt, node, "interrupt-controller", NULL, 0));
726     _FDT(fdt_setprop_cell(fdt, node, "#interrupt-cells", 2));
727 
728     /* For SLOF */
729     _FDT(fdt_setprop_cell(fdt, node, "linux,phandle", phandle));
730     _FDT(fdt_setprop_cell(fdt, node, "phandle", phandle));
731 
732     /*
733      * The "ibm,plat-res-int-priorities" property defines the priority
734      * ranges reserved by the hypervisor
735      */
736     _FDT(fdt_setprop(fdt, 0, "ibm,plat-res-int-priorities",
737                      plat_res_int_priorities, sizeof(plat_res_int_priorities)));
738 }
739 
740 static int spapr_xive_activate(SpaprInterruptController *intc,
741                                uint32_t nr_servers, Error **errp)
742 {
743     SpaprXive *xive = SPAPR_XIVE(intc);
744 
745     if (kvm_enabled()) {
746         int rc = spapr_irq_init_kvm(kvmppc_xive_connect, intc, nr_servers,
747                                     errp);
748         if (rc < 0) {
749             return rc;
750         }
751     }
752 
753     /* Activate the XIVE MMIOs */
754     spapr_xive_mmio_set_enabled(xive, true);
755 
756     return 0;
757 }
758 
759 static void spapr_xive_deactivate(SpaprInterruptController *intc)
760 {
761     SpaprXive *xive = SPAPR_XIVE(intc);
762 
763     spapr_xive_mmio_set_enabled(xive, false);
764 
765     if (spapr_xive_in_kernel(xive)) {
766         kvmppc_xive_disconnect(intc);
767     }
768 }
769 
770 static bool spapr_xive_in_kernel_xptr(const XivePresenter *xptr)
771 {
772     return spapr_xive_in_kernel(SPAPR_XIVE(xptr));
773 }
774 
775 static void spapr_xive_class_init(ObjectClass *klass, void *data)
776 {
777     DeviceClass *dc = DEVICE_CLASS(klass);
778     XiveRouterClass *xrc = XIVE_ROUTER_CLASS(klass);
779     SpaprInterruptControllerClass *sicc = SPAPR_INTC_CLASS(klass);
780     XivePresenterClass *xpc = XIVE_PRESENTER_CLASS(klass);
781     SpaprXiveClass *sxc = SPAPR_XIVE_CLASS(klass);
782 
783     dc->desc    = "sPAPR XIVE Interrupt Controller";
784     device_class_set_props(dc, spapr_xive_properties);
785     device_class_set_parent_realize(dc, spapr_xive_realize,
786                                     &sxc->parent_realize);
787     dc->vmsd    = &vmstate_spapr_xive;
788 
789     xrc->get_eas = spapr_xive_get_eas;
790     xrc->get_end = spapr_xive_get_end;
791     xrc->write_end = spapr_xive_write_end;
792     xrc->get_nvt = spapr_xive_get_nvt;
793     xrc->write_nvt = spapr_xive_write_nvt;
794     xrc->get_block_id = spapr_xive_get_block_id;
795 
796     sicc->activate = spapr_xive_activate;
797     sicc->deactivate = spapr_xive_deactivate;
798     sicc->cpu_intc_create = spapr_xive_cpu_intc_create;
799     sicc->cpu_intc_reset = spapr_xive_cpu_intc_reset;
800     sicc->cpu_intc_destroy = spapr_xive_cpu_intc_destroy;
801     sicc->claim_irq = spapr_xive_claim_irq;
802     sicc->free_irq = spapr_xive_free_irq;
803     sicc->set_irq = spapr_xive_set_irq;
804     sicc->print_info = spapr_xive_print_info;
805     sicc->dt = spapr_xive_dt;
806     sicc->post_load = spapr_xive_post_load;
807 
808     xpc->match_nvt  = spapr_xive_match_nvt;
809     xpc->in_kernel  = spapr_xive_in_kernel_xptr;
810 }
811 
812 static const TypeInfo spapr_xive_info = {
813     .name = TYPE_SPAPR_XIVE,
814     .parent = TYPE_XIVE_ROUTER,
815     .instance_init = spapr_xive_instance_init,
816     .instance_size = sizeof(SpaprXive),
817     .class_init = spapr_xive_class_init,
818     .class_size = sizeof(SpaprXiveClass),
819     .interfaces = (InterfaceInfo[]) {
820         { TYPE_SPAPR_INTC },
821         { }
822     },
823 };
824 
825 static void spapr_xive_register_types(void)
826 {
827     type_register_static(&spapr_xive_info);
828 }
829 
830 type_init(spapr_xive_register_types)
831 
832 /*
833  * XIVE hcalls
834  *
835  * The terminology used by the XIVE hcalls is the following :
836  *
837  *   TARGET vCPU number
838  *   EQ     Event Queue assigned by OS to receive event data
839  *   ESB    page for source interrupt management
840  *   LISN   Logical Interrupt Source Number identifying a source in the
841  *          machine
842  *   EISN   Effective Interrupt Source Number used by guest OS to
843  *          identify source in the guest
844  *
845  * The EAS, END, NVT structures are not exposed.
846  */
847 
848 /*
849  * On POWER9, the KVM XIVE device uses priority 7 for the escalation
850  * interrupts. So we only allow the guest to use priorities [0..6].
851  */
852 static bool spapr_xive_priority_is_reserved(SpaprXive *xive, uint8_t priority)
853 {
854     return priority >= xive->hv_prio;
855 }
856 
857 /*
858  * The H_INT_GET_SOURCE_INFO hcall() is used to obtain the logical
859  * real address of the MMIO page through which the Event State Buffer
860  * entry associated with the value of the "lisn" parameter is managed.
861  *
862  * Parameters:
863  * Input
864  * - R4: "flags"
865  *         Bits 0-63 reserved
866  * - R5: "lisn" is per "interrupts", "interrupt-map", or
867  *       "ibm,xive-lisn-ranges" properties, or as returned by the
868  *       ibm,query-interrupt-source-number RTAS call, or as returned
869  *       by the H_ALLOCATE_VAS_WINDOW hcall
870  *
871  * Output
872  * - R4: "flags"
873  *         Bits 0-59: Reserved
874  *         Bit 60: H_INT_ESB must be used for Event State Buffer
875  *                 management
876  *         Bit 61: 1 == LSI  0 == MSI
877  *         Bit 62: the full function page supports trigger
878  *         Bit 63: Store EOI Supported
879  * - R5: Logical Real address of full function Event State Buffer
880  *       management page, -1 if H_INT_ESB hcall flag is set to 1.
881  * - R6: Logical Real Address of trigger only Event State Buffer
882  *       management page or -1.
883  * - R7: Power of 2 page size for the ESB management pages returned in
884  *       R5 and R6.
885  */
886 
887 #define SPAPR_XIVE_SRC_H_INT_ESB     PPC_BIT(60) /* ESB manage with H_INT_ESB */
888 #define SPAPR_XIVE_SRC_LSI           PPC_BIT(61) /* Virtual LSI type */
889 #define SPAPR_XIVE_SRC_TRIGGER       PPC_BIT(62) /* Trigger and management
890                                                     on same page */
891 #define SPAPR_XIVE_SRC_STORE_EOI     PPC_BIT(63) /* Store EOI support */
892 
893 static target_ulong h_int_get_source_info(PowerPCCPU *cpu,
894                                           SpaprMachineState *spapr,
895                                           target_ulong opcode,
896                                           target_ulong *args)
897 {
898     SpaprXive *xive = spapr->xive;
899     XiveSource *xsrc = &xive->source;
900     target_ulong flags  = args[0];
901     target_ulong lisn   = args[1];
902 
903     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
904         return H_FUNCTION;
905     }
906 
907     if (flags) {
908         return H_PARAMETER;
909     }
910 
911     if (lisn >= xive->nr_irqs) {
912         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
913                       lisn);
914         return H_P2;
915     }
916 
917     if (!xive_eas_is_valid(&xive->eat[lisn])) {
918         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
919                       lisn);
920         return H_P2;
921     }
922 
923     /*
924      * All sources are emulated under the main XIVE object and share
925      * the same characteristics.
926      */
927     args[0] = 0;
928     if (!xive_source_esb_has_2page(xsrc)) {
929         args[0] |= SPAPR_XIVE_SRC_TRIGGER;
930     }
931     if (xsrc->esb_flags & XIVE_SRC_STORE_EOI) {
932         args[0] |= SPAPR_XIVE_SRC_STORE_EOI;
933     }
934 
935     /*
936      * Force the use of the H_INT_ESB hcall in case of an LSI
937      * interrupt. This is necessary under KVM to re-trigger the
938      * interrupt if the level is still asserted
939      */
940     if (xive_source_irq_is_lsi(xsrc, lisn)) {
941         args[0] |= SPAPR_XIVE_SRC_H_INT_ESB | SPAPR_XIVE_SRC_LSI;
942     }
943 
944     if (!(args[0] & SPAPR_XIVE_SRC_H_INT_ESB)) {
945         args[1] = xive->vc_base + xive_source_esb_mgmt(xsrc, lisn);
946     } else {
947         args[1] = -1;
948     }
949 
950     if (xive_source_esb_has_2page(xsrc) &&
951         !(args[0] & SPAPR_XIVE_SRC_H_INT_ESB)) {
952         args[2] = xive->vc_base + xive_source_esb_page(xsrc, lisn);
953     } else {
954         args[2] = -1;
955     }
956 
957     if (xive_source_esb_has_2page(xsrc)) {
958         args[3] = xsrc->esb_shift - 1;
959     } else {
960         args[3] = xsrc->esb_shift;
961     }
962 
963     return H_SUCCESS;
964 }
965 
966 /*
967  * The H_INT_SET_SOURCE_CONFIG hcall() is used to assign a Logical
968  * Interrupt Source to a target. The Logical Interrupt Source is
969  * designated with the "lisn" parameter and the target is designated
970  * with the "target" and "priority" parameters.  Upon return from the
971  * hcall(), no additional interrupts will be directed to the old EQ.
972  *
973  * Parameters:
974  * Input:
975  * - R4: "flags"
976  *         Bits 0-61: Reserved
977  *         Bit 62: set the "eisn" in the EAS
978  *         Bit 63: masks the interrupt source in the hardware interrupt
979  *       control structure. An interrupt masked by this mechanism will
980  *       be dropped, but it's source state bits will still be
981  *       set. There is no race-free way of unmasking and restoring the
982  *       source. Thus this should only be used in interrupts that are
983  *       also masked at the source, and only in cases where the
984  *       interrupt is not meant to be used for a large amount of time
985  *       because no valid target exists for it for example
986  * - R5: "lisn" is per "interrupts", "interrupt-map", or
987  *       "ibm,xive-lisn-ranges" properties, or as returned by the
988  *       ibm,query-interrupt-source-number RTAS call, or as returned by
989  *       the H_ALLOCATE_VAS_WINDOW hcall
990  * - R6: "target" is per "ibm,ppc-interrupt-server#s" or
991  *       "ibm,ppc-interrupt-gserver#s"
992  * - R7: "priority" is a valid priority not in
993  *       "ibm,plat-res-int-priorities"
994  * - R8: "eisn" is the guest EISN associated with the "lisn"
995  *
996  * Output:
997  * - None
998  */
999 
1000 #define SPAPR_XIVE_SRC_SET_EISN PPC_BIT(62)
1001 #define SPAPR_XIVE_SRC_MASK     PPC_BIT(63)
1002 
1003 static target_ulong h_int_set_source_config(PowerPCCPU *cpu,
1004                                             SpaprMachineState *spapr,
1005                                             target_ulong opcode,
1006                                             target_ulong *args)
1007 {
1008     SpaprXive *xive = spapr->xive;
1009     XiveEAS eas, new_eas;
1010     target_ulong flags    = args[0];
1011     target_ulong lisn     = args[1];
1012     target_ulong target   = args[2];
1013     target_ulong priority = args[3];
1014     target_ulong eisn     = args[4];
1015     uint8_t end_blk;
1016     uint32_t end_idx;
1017 
1018     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1019         return H_FUNCTION;
1020     }
1021 
1022     if (flags & ~(SPAPR_XIVE_SRC_SET_EISN | SPAPR_XIVE_SRC_MASK)) {
1023         return H_PARAMETER;
1024     }
1025 
1026     if (lisn >= xive->nr_irqs) {
1027         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1028                       lisn);
1029         return H_P2;
1030     }
1031 
1032     eas = xive->eat[lisn];
1033     if (!xive_eas_is_valid(&eas)) {
1034         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1035                       lisn);
1036         return H_P2;
1037     }
1038 
1039     /* priority 0xff is used to reset the EAS */
1040     if (priority == 0xff) {
1041         new_eas.w = cpu_to_be64(EAS_VALID | EAS_MASKED);
1042         goto out;
1043     }
1044 
1045     if (flags & SPAPR_XIVE_SRC_MASK) {
1046         new_eas.w = eas.w | cpu_to_be64(EAS_MASKED);
1047     } else {
1048         new_eas.w = eas.w & cpu_to_be64(~EAS_MASKED);
1049     }
1050 
1051     if (spapr_xive_priority_is_reserved(xive, priority)) {
1052         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1053                       " is reserved\n", priority);
1054         return H_P4;
1055     }
1056 
1057     /*
1058      * Validate that "target" is part of the list of threads allocated
1059      * to the partition. For that, find the END corresponding to the
1060      * target.
1061      */
1062     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1063         return H_P3;
1064     }
1065 
1066     new_eas.w = xive_set_field64(EAS_END_BLOCK, new_eas.w, end_blk);
1067     new_eas.w = xive_set_field64(EAS_END_INDEX, new_eas.w, end_idx);
1068 
1069     if (flags & SPAPR_XIVE_SRC_SET_EISN) {
1070         new_eas.w = xive_set_field64(EAS_END_DATA, new_eas.w, eisn);
1071     }
1072 
1073     if (spapr_xive_in_kernel(xive)) {
1074         Error *local_err = NULL;
1075 
1076         kvmppc_xive_set_source_config(xive, lisn, &new_eas, &local_err);
1077         if (local_err) {
1078             error_report_err(local_err);
1079             return H_HARDWARE;
1080         }
1081     }
1082 
1083 out:
1084     xive->eat[lisn] = new_eas;
1085     return H_SUCCESS;
1086 }
1087 
1088 /*
1089  * The H_INT_GET_SOURCE_CONFIG hcall() is used to determine to which
1090  * target/priority pair is assigned to the specified Logical Interrupt
1091  * Source.
1092  *
1093  * Parameters:
1094  * Input:
1095  * - R4: "flags"
1096  *         Bits 0-63 Reserved
1097  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1098  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1099  *       ibm,query-interrupt-source-number RTAS call, or as
1100  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1101  *
1102  * Output:
1103  * - R4: Target to which the specified Logical Interrupt Source is
1104  *       assigned
1105  * - R5: Priority to which the specified Logical Interrupt Source is
1106  *       assigned
1107  * - R6: EISN for the specified Logical Interrupt Source (this will be
1108  *       equivalent to the LISN if not changed by H_INT_SET_SOURCE_CONFIG)
1109  */
1110 static target_ulong h_int_get_source_config(PowerPCCPU *cpu,
1111                                             SpaprMachineState *spapr,
1112                                             target_ulong opcode,
1113                                             target_ulong *args)
1114 {
1115     SpaprXive *xive = spapr->xive;
1116     target_ulong flags = args[0];
1117     target_ulong lisn = args[1];
1118     XiveEAS eas;
1119     XiveEND *end;
1120     uint8_t nvt_blk;
1121     uint32_t end_idx, nvt_idx;
1122 
1123     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1124         return H_FUNCTION;
1125     }
1126 
1127     if (flags) {
1128         return H_PARAMETER;
1129     }
1130 
1131     if (lisn >= xive->nr_irqs) {
1132         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1133                       lisn);
1134         return H_P2;
1135     }
1136 
1137     eas = xive->eat[lisn];
1138     if (!xive_eas_is_valid(&eas)) {
1139         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1140                       lisn);
1141         return H_P2;
1142     }
1143 
1144     /* EAS_END_BLOCK is unused on sPAPR */
1145     end_idx = xive_get_field64(EAS_END_INDEX, eas.w);
1146 
1147     assert(end_idx < xive->nr_ends);
1148     end = &xive->endt[end_idx];
1149 
1150     nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end->w6);
1151     nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end->w6);
1152     args[0] = spapr_xive_nvt_to_target(nvt_blk, nvt_idx);
1153 
1154     if (xive_eas_is_masked(&eas)) {
1155         args[1] = 0xff;
1156     } else {
1157         args[1] = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
1158     }
1159 
1160     args[2] = xive_get_field64(EAS_END_DATA, eas.w);
1161 
1162     return H_SUCCESS;
1163 }
1164 
1165 /*
1166  * The H_INT_GET_QUEUE_INFO hcall() is used to get the logical real
1167  * address of the notification management page associated with the
1168  * specified target and priority.
1169  *
1170  * Parameters:
1171  * Input:
1172  * - R4: "flags"
1173  *         Bits 0-63 Reserved
1174  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1175  *       "ibm,ppc-interrupt-gserver#s"
1176  * - R6: "priority" is a valid priority not in
1177  *       "ibm,plat-res-int-priorities"
1178  *
1179  * Output:
1180  * - R4: Logical real address of notification page
1181  * - R5: Power of 2 page size of the notification page
1182  */
1183 static target_ulong h_int_get_queue_info(PowerPCCPU *cpu,
1184                                          SpaprMachineState *spapr,
1185                                          target_ulong opcode,
1186                                          target_ulong *args)
1187 {
1188     SpaprXive *xive = spapr->xive;
1189     XiveENDSource *end_xsrc = &xive->end_source;
1190     target_ulong flags = args[0];
1191     target_ulong target = args[1];
1192     target_ulong priority = args[2];
1193     XiveEND *end;
1194     uint8_t end_blk;
1195     uint32_t end_idx;
1196 
1197     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1198         return H_FUNCTION;
1199     }
1200 
1201     if (flags) {
1202         return H_PARAMETER;
1203     }
1204 
1205     /*
1206      * H_STATE should be returned if a H_INT_RESET is in progress.
1207      * This is not needed when running the emulation under QEMU
1208      */
1209 
1210     if (spapr_xive_priority_is_reserved(xive, priority)) {
1211         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1212                       " is reserved\n", priority);
1213         return H_P3;
1214     }
1215 
1216     /*
1217      * Validate that "target" is part of the list of threads allocated
1218      * to the partition. For that, find the END corresponding to the
1219      * target.
1220      */
1221     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1222         return H_P2;
1223     }
1224 
1225     assert(end_idx < xive->nr_ends);
1226     end = &xive->endt[end_idx];
1227 
1228     args[0] = xive->end_base + (1ull << (end_xsrc->esb_shift + 1)) * end_idx;
1229     if (xive_end_is_enqueue(end)) {
1230         args[1] = xive_get_field32(END_W0_QSIZE, end->w0) + 12;
1231     } else {
1232         args[1] = 0;
1233     }
1234 
1235     return H_SUCCESS;
1236 }
1237 
1238 /*
1239  * The H_INT_SET_QUEUE_CONFIG hcall() is used to set or reset a EQ for
1240  * a given "target" and "priority".  It is also used to set the
1241  * notification config associated with the EQ.  An EQ size of 0 is
1242  * used to reset the EQ config for a given target and priority. If
1243  * resetting the EQ config, the END associated with the given "target"
1244  * and "priority" will be changed to disable queueing.
1245  *
1246  * Upon return from the hcall(), no additional interrupts will be
1247  * directed to the old EQ (if one was set). The old EQ (if one was
1248  * set) should be investigated for interrupts that occurred prior to
1249  * or during the hcall().
1250  *
1251  * Parameters:
1252  * Input:
1253  * - R4: "flags"
1254  *         Bits 0-62: Reserved
1255  *         Bit 63: Unconditional Notify (n) per the XIVE spec
1256  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1257  *       "ibm,ppc-interrupt-gserver#s"
1258  * - R6: "priority" is a valid priority not in
1259  *       "ibm,plat-res-int-priorities"
1260  * - R7: "eventQueue": The logical real address of the start of the EQ
1261  * - R8: "eventQueueSize": The power of 2 EQ size per "ibm,xive-eq-sizes"
1262  *
1263  * Output:
1264  * - None
1265  */
1266 
1267 #define SPAPR_XIVE_END_ALWAYS_NOTIFY PPC_BIT(63)
1268 
1269 static target_ulong h_int_set_queue_config(PowerPCCPU *cpu,
1270                                            SpaprMachineState *spapr,
1271                                            target_ulong opcode,
1272                                            target_ulong *args)
1273 {
1274     SpaprXive *xive = spapr->xive;
1275     target_ulong flags = args[0];
1276     target_ulong target = args[1];
1277     target_ulong priority = args[2];
1278     target_ulong qpage = args[3];
1279     target_ulong qsize = args[4];
1280     XiveEND end;
1281     uint8_t end_blk, nvt_blk;
1282     uint32_t end_idx, nvt_idx;
1283 
1284     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1285         return H_FUNCTION;
1286     }
1287 
1288     if (flags & ~SPAPR_XIVE_END_ALWAYS_NOTIFY) {
1289         return H_PARAMETER;
1290     }
1291 
1292     /*
1293      * H_STATE should be returned if a H_INT_RESET is in progress.
1294      * This is not needed when running the emulation under QEMU
1295      */
1296 
1297     if (spapr_xive_priority_is_reserved(xive, priority)) {
1298         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1299                       " is reserved\n", priority);
1300         return H_P3;
1301     }
1302 
1303     /*
1304      * Validate that "target" is part of the list of threads allocated
1305      * to the partition. For that, find the END corresponding to the
1306      * target.
1307      */
1308 
1309     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1310         return H_P2;
1311     }
1312 
1313     assert(end_idx < xive->nr_ends);
1314     memcpy(&end, &xive->endt[end_idx], sizeof(XiveEND));
1315 
1316     switch (qsize) {
1317     case 12:
1318     case 16:
1319     case 21:
1320     case 24:
1321         if (!QEMU_IS_ALIGNED(qpage, 1ul << qsize)) {
1322             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: EQ @0x%" HWADDR_PRIx
1323                           " is not naturally aligned with %" HWADDR_PRIx "\n",
1324                           qpage, (hwaddr)1 << qsize);
1325             return H_P4;
1326         }
1327         end.w2 = cpu_to_be32((qpage >> 32) & 0x0fffffff);
1328         end.w3 = cpu_to_be32(qpage & 0xffffffff);
1329         end.w0 |= cpu_to_be32(END_W0_ENQUEUE);
1330         end.w0 = xive_set_field32(END_W0_QSIZE, end.w0, qsize - 12);
1331         break;
1332     case 0:
1333         /* reset queue and disable queueing */
1334         spapr_xive_end_reset(&end);
1335         goto out;
1336 
1337     default:
1338         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid EQ size %"PRIx64"\n",
1339                       qsize);
1340         return H_P5;
1341     }
1342 
1343     if (qsize) {
1344         hwaddr plen = 1 << qsize;
1345         void *eq;
1346 
1347         /*
1348          * Validate the guest EQ. We should also check that the queue
1349          * has been zeroed by the OS.
1350          */
1351         eq = address_space_map(CPU(cpu)->as, qpage, &plen, true,
1352                                MEMTXATTRS_UNSPECIFIED);
1353         if (plen != 1 << qsize) {
1354             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to map EQ @0x%"
1355                           HWADDR_PRIx "\n", qpage);
1356             return H_P4;
1357         }
1358         address_space_unmap(CPU(cpu)->as, eq, plen, true, plen);
1359     }
1360 
1361     /* "target" should have been validated above */
1362     if (spapr_xive_target_to_nvt(target, &nvt_blk, &nvt_idx)) {
1363         g_assert_not_reached();
1364     }
1365 
1366     /*
1367      * Ensure the priority and target are correctly set (they will not
1368      * be right after allocation)
1369      */
1370     end.w6 = xive_set_field32(END_W6_NVT_BLOCK, 0ul, nvt_blk) |
1371         xive_set_field32(END_W6_NVT_INDEX, 0ul, nvt_idx);
1372     end.w7 = xive_set_field32(END_W7_F0_PRIORITY, 0ul, priority);
1373 
1374     if (flags & SPAPR_XIVE_END_ALWAYS_NOTIFY) {
1375         end.w0 |= cpu_to_be32(END_W0_UCOND_NOTIFY);
1376     } else {
1377         end.w0 &= cpu_to_be32((uint32_t)~END_W0_UCOND_NOTIFY);
1378     }
1379 
1380     /*
1381      * The generation bit for the END starts at 1 and The END page
1382      * offset counter starts at 0.
1383      */
1384     end.w1 = cpu_to_be32(END_W1_GENERATION) |
1385         xive_set_field32(END_W1_PAGE_OFF, 0ul, 0ul);
1386     end.w0 |= cpu_to_be32(END_W0_VALID);
1387 
1388     /*
1389      * TODO: issue syncs required to ensure all in-flight interrupts
1390      * are complete on the old END
1391      */
1392 
1393 out:
1394     if (spapr_xive_in_kernel(xive)) {
1395         Error *local_err = NULL;
1396 
1397         kvmppc_xive_set_queue_config(xive, end_blk, end_idx, &end, &local_err);
1398         if (local_err) {
1399             error_report_err(local_err);
1400             return H_HARDWARE;
1401         }
1402     }
1403 
1404     /* Update END */
1405     memcpy(&xive->endt[end_idx], &end, sizeof(XiveEND));
1406     return H_SUCCESS;
1407 }
1408 
1409 /*
1410  * The H_INT_GET_QUEUE_CONFIG hcall() is used to get a EQ for a given
1411  * target and priority.
1412  *
1413  * Parameters:
1414  * Input:
1415  * - R4: "flags"
1416  *         Bits 0-62: Reserved
1417  *         Bit 63: Debug: Return debug data
1418  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1419  *       "ibm,ppc-interrupt-gserver#s"
1420  * - R6: "priority" is a valid priority not in
1421  *       "ibm,plat-res-int-priorities"
1422  *
1423  * Output:
1424  * - R4: "flags":
1425  *       Bits 0-61: Reserved
1426  *       Bit 62: The value of Event Queue Generation Number (g) per
1427  *              the XIVE spec if "Debug" = 1
1428  *       Bit 63: The value of Unconditional Notify (n) per the XIVE spec
1429  * - R5: The logical real address of the start of the EQ
1430  * - R6: The power of 2 EQ size per "ibm,xive-eq-sizes"
1431  * - R7: The value of Event Queue Offset Counter per XIVE spec
1432  *       if "Debug" = 1, else 0
1433  *
1434  */
1435 
1436 #define SPAPR_XIVE_END_DEBUG     PPC_BIT(63)
1437 
1438 static target_ulong h_int_get_queue_config(PowerPCCPU *cpu,
1439                                            SpaprMachineState *spapr,
1440                                            target_ulong opcode,
1441                                            target_ulong *args)
1442 {
1443     SpaprXive *xive = spapr->xive;
1444     target_ulong flags = args[0];
1445     target_ulong target = args[1];
1446     target_ulong priority = args[2];
1447     XiveEND *end;
1448     uint8_t end_blk;
1449     uint32_t end_idx;
1450 
1451     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1452         return H_FUNCTION;
1453     }
1454 
1455     if (flags & ~SPAPR_XIVE_END_DEBUG) {
1456         return H_PARAMETER;
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     if (spapr_xive_priority_is_reserved(xive, priority)) {
1465         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1466                       " is reserved\n", priority);
1467         return H_P3;
1468     }
1469 
1470     /*
1471      * Validate that "target" is part of the list of threads allocated
1472      * to the partition. For that, find the END corresponding to the
1473      * target.
1474      */
1475     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1476         return H_P2;
1477     }
1478 
1479     assert(end_idx < xive->nr_ends);
1480     end = &xive->endt[end_idx];
1481 
1482     args[0] = 0;
1483     if (xive_end_is_notify(end)) {
1484         args[0] |= SPAPR_XIVE_END_ALWAYS_NOTIFY;
1485     }
1486 
1487     if (xive_end_is_enqueue(end)) {
1488         args[1] = xive_end_qaddr(end);
1489         args[2] = xive_get_field32(END_W0_QSIZE, end->w0) + 12;
1490     } else {
1491         args[1] = 0;
1492         args[2] = 0;
1493     }
1494 
1495     if (spapr_xive_in_kernel(xive)) {
1496         Error *local_err = NULL;
1497 
1498         kvmppc_xive_get_queue_config(xive, end_blk, end_idx, end, &local_err);
1499         if (local_err) {
1500             error_report_err(local_err);
1501             return H_HARDWARE;
1502         }
1503     }
1504 
1505     /* TODO: do we need any locking on the END ? */
1506     if (flags & SPAPR_XIVE_END_DEBUG) {
1507         /* Load the event queue generation number into the return flags */
1508         args[0] |= (uint64_t)xive_get_field32(END_W1_GENERATION, end->w1) << 62;
1509 
1510         /* Load R7 with the event queue offset counter */
1511         args[3] = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1512     } else {
1513         args[3] = 0;
1514     }
1515 
1516     return H_SUCCESS;
1517 }
1518 
1519 /*
1520  * The H_INT_SET_OS_REPORTING_LINE hcall() is used to set the
1521  * reporting cache line pair for the calling thread.  The reporting
1522  * cache lines will contain the OS interrupt context when the OS
1523  * issues a CI store byte to @TIMA+0xC10 to acknowledge the OS
1524  * interrupt. The reporting cache lines can be reset by inputting -1
1525  * in "reportingLine".  Issuing the CI store byte without reporting
1526  * cache lines registered will result in the data not being accessible
1527  * to the OS.
1528  *
1529  * Parameters:
1530  * Input:
1531  * - R4: "flags"
1532  *         Bits 0-63: Reserved
1533  * - R5: "reportingLine": The logical real address of the reporting cache
1534  *       line pair
1535  *
1536  * Output:
1537  * - None
1538  */
1539 static target_ulong h_int_set_os_reporting_line(PowerPCCPU *cpu,
1540                                                 SpaprMachineState *spapr,
1541                                                 target_ulong opcode,
1542                                                 target_ulong *args)
1543 {
1544     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1545         return H_FUNCTION;
1546     }
1547 
1548     /*
1549      * H_STATE should be returned if a H_INT_RESET is in progress.
1550      * This is not needed when running the emulation under QEMU
1551      */
1552 
1553     /* TODO: H_INT_SET_OS_REPORTING_LINE */
1554     return H_FUNCTION;
1555 }
1556 
1557 /*
1558  * The H_INT_GET_OS_REPORTING_LINE hcall() is used to get the logical
1559  * real address of the reporting cache line pair set for the input
1560  * "target".  If no reporting cache line pair has been set, -1 is
1561  * returned.
1562  *
1563  * Parameters:
1564  * Input:
1565  * - R4: "flags"
1566  *         Bits 0-63: Reserved
1567  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1568  *       "ibm,ppc-interrupt-gserver#s"
1569  * - R6: "reportingLine": The logical real address of the reporting
1570  *        cache line pair
1571  *
1572  * Output:
1573  * - R4: The logical real address of the reporting line if set, else -1
1574  */
1575 static target_ulong h_int_get_os_reporting_line(PowerPCCPU *cpu,
1576                                                 SpaprMachineState *spapr,
1577                                                 target_ulong opcode,
1578                                                 target_ulong *args)
1579 {
1580     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1581         return H_FUNCTION;
1582     }
1583 
1584     /*
1585      * H_STATE should be returned if a H_INT_RESET is in progress.
1586      * This is not needed when running the emulation under QEMU
1587      */
1588 
1589     /* TODO: H_INT_GET_OS_REPORTING_LINE */
1590     return H_FUNCTION;
1591 }
1592 
1593 /*
1594  * The H_INT_ESB hcall() is used to issue a load or store to the ESB
1595  * page for the input "lisn".  This hcall is only supported for LISNs
1596  * that have the ESB hcall flag set to 1 when returned from hcall()
1597  * H_INT_GET_SOURCE_INFO.
1598  *
1599  * Parameters:
1600  * Input:
1601  * - R4: "flags"
1602  *         Bits 0-62: Reserved
1603  *         bit 63: Store: Store=1, store operation, else load operation
1604  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1605  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1606  *       ibm,query-interrupt-source-number RTAS call, or as
1607  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1608  * - R6: "esbOffset" is the offset into the ESB page for the load or
1609  *       store operation
1610  * - R7: "storeData" is the data to write for a store operation
1611  *
1612  * Output:
1613  * - R4: The value of the load if load operation, else -1
1614  */
1615 
1616 #define SPAPR_XIVE_ESB_STORE PPC_BIT(63)
1617 
1618 static target_ulong h_int_esb(PowerPCCPU *cpu,
1619                               SpaprMachineState *spapr,
1620                               target_ulong opcode,
1621                               target_ulong *args)
1622 {
1623     SpaprXive *xive = spapr->xive;
1624     XiveEAS eas;
1625     target_ulong flags  = args[0];
1626     target_ulong lisn   = args[1];
1627     target_ulong offset = args[2];
1628     target_ulong data   = args[3];
1629     hwaddr mmio_addr;
1630     XiveSource *xsrc = &xive->source;
1631 
1632     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1633         return H_FUNCTION;
1634     }
1635 
1636     if (flags & ~SPAPR_XIVE_ESB_STORE) {
1637         return H_PARAMETER;
1638     }
1639 
1640     if (lisn >= xive->nr_irqs) {
1641         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1642                       lisn);
1643         return H_P2;
1644     }
1645 
1646     eas = xive->eat[lisn];
1647     if (!xive_eas_is_valid(&eas)) {
1648         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1649                       lisn);
1650         return H_P2;
1651     }
1652 
1653     if (offset > (1ull << xsrc->esb_shift)) {
1654         return H_P3;
1655     }
1656 
1657     if (spapr_xive_in_kernel(xive)) {
1658         args[0] = kvmppc_xive_esb_rw(xsrc, lisn, offset, data,
1659                                      flags & SPAPR_XIVE_ESB_STORE);
1660     } else {
1661         mmio_addr = xive->vc_base + xive_source_esb_mgmt(xsrc, lisn) + offset;
1662 
1663         if (dma_memory_rw(&address_space_memory, mmio_addr, &data, 8,
1664                           (flags & SPAPR_XIVE_ESB_STORE))) {
1665             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to access ESB @0x%"
1666                           HWADDR_PRIx "\n", mmio_addr);
1667             return H_HARDWARE;
1668         }
1669         args[0] = (flags & SPAPR_XIVE_ESB_STORE) ? -1 : data;
1670     }
1671     return H_SUCCESS;
1672 }
1673 
1674 /*
1675  * The H_INT_SYNC hcall() is used to issue hardware syncs that will
1676  * ensure any in flight events for the input lisn are in the event
1677  * queue.
1678  *
1679  * Parameters:
1680  * Input:
1681  * - R4: "flags"
1682  *         Bits 0-63: Reserved
1683  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1684  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1685  *       ibm,query-interrupt-source-number RTAS call, or as
1686  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1687  *
1688  * Output:
1689  * - None
1690  */
1691 static target_ulong h_int_sync(PowerPCCPU *cpu,
1692                                SpaprMachineState *spapr,
1693                                target_ulong opcode,
1694                                target_ulong *args)
1695 {
1696     SpaprXive *xive = spapr->xive;
1697     XiveEAS eas;
1698     target_ulong flags = args[0];
1699     target_ulong lisn = args[1];
1700 
1701     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1702         return H_FUNCTION;
1703     }
1704 
1705     if (flags) {
1706         return H_PARAMETER;
1707     }
1708 
1709     if (lisn >= xive->nr_irqs) {
1710         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1711                       lisn);
1712         return H_P2;
1713     }
1714 
1715     eas = xive->eat[lisn];
1716     if (!xive_eas_is_valid(&eas)) {
1717         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1718                       lisn);
1719         return H_P2;
1720     }
1721 
1722     /*
1723      * H_STATE should be returned if a H_INT_RESET is in progress.
1724      * This is not needed when running the emulation under QEMU
1725      */
1726 
1727     /*
1728      * This is not real hardware. Nothing to be done unless when
1729      * under KVM
1730      */
1731 
1732     if (spapr_xive_in_kernel(xive)) {
1733         Error *local_err = NULL;
1734 
1735         kvmppc_xive_sync_source(xive, lisn, &local_err);
1736         if (local_err) {
1737             error_report_err(local_err);
1738             return H_HARDWARE;
1739         }
1740     }
1741     return H_SUCCESS;
1742 }
1743 
1744 /*
1745  * The H_INT_RESET hcall() is used to reset all of the partition's
1746  * interrupt exploitation structures to their initial state.  This
1747  * means losing all previously set interrupt state set via
1748  * H_INT_SET_SOURCE_CONFIG and H_INT_SET_QUEUE_CONFIG.
1749  *
1750  * Parameters:
1751  * Input:
1752  * - R4: "flags"
1753  *         Bits 0-63: Reserved
1754  *
1755  * Output:
1756  * - None
1757  */
1758 static target_ulong h_int_reset(PowerPCCPU *cpu,
1759                                 SpaprMachineState *spapr,
1760                                 target_ulong opcode,
1761                                 target_ulong *args)
1762 {
1763     SpaprXive *xive = spapr->xive;
1764     target_ulong flags   = args[0];
1765 
1766     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1767         return H_FUNCTION;
1768     }
1769 
1770     if (flags) {
1771         return H_PARAMETER;
1772     }
1773 
1774     device_legacy_reset(DEVICE(xive));
1775 
1776     if (spapr_xive_in_kernel(xive)) {
1777         Error *local_err = NULL;
1778 
1779         kvmppc_xive_reset(xive, &local_err);
1780         if (local_err) {
1781             error_report_err(local_err);
1782             return H_HARDWARE;
1783         }
1784     }
1785     return H_SUCCESS;
1786 }
1787 
1788 void spapr_xive_hcall_init(SpaprMachineState *spapr)
1789 {
1790     spapr_register_hypercall(H_INT_GET_SOURCE_INFO, h_int_get_source_info);
1791     spapr_register_hypercall(H_INT_SET_SOURCE_CONFIG, h_int_set_source_config);
1792     spapr_register_hypercall(H_INT_GET_SOURCE_CONFIG, h_int_get_source_config);
1793     spapr_register_hypercall(H_INT_GET_QUEUE_INFO, h_int_get_queue_info);
1794     spapr_register_hypercall(H_INT_SET_QUEUE_CONFIG, h_int_set_queue_config);
1795     spapr_register_hypercall(H_INT_GET_QUEUE_CONFIG, h_int_get_queue_config);
1796     spapr_register_hypercall(H_INT_SET_OS_REPORTING_LINE,
1797                              h_int_set_os_reporting_line);
1798     spapr_register_hypercall(H_INT_GET_OS_REPORTING_LINE,
1799                              h_int_get_os_reporting_line);
1800     spapr_register_hypercall(H_INT_ESB, h_int_esb);
1801     spapr_register_hypercall(H_INT_SYNC, h_int_sync);
1802     spapr_register_hypercall(H_INT_RESET, h_int_reset);
1803 }
1804