xref: /openbmc/qemu/hw/intc/spapr_xive.c (revision 8b3aaaa1a9c2182ed9e3c406c90bb4257b43e753)
1 /*
2  * QEMU PowerPC sPAPR XIVE interrupt controller model
3  *
4  * Copyright (c) 2017-2018, IBM Corporation.
5  *
6  * This code is licensed under the GPL version 2 or later. See the
7  * COPYING file in the top-level directory.
8  */
9 
10 #include "qemu/osdep.h"
11 #include "qemu/log.h"
12 #include "qemu/module.h"
13 #include "qapi/error.h"
14 #include "qemu/error-report.h"
15 #include "target/ppc/cpu.h"
16 #include "sysemu/cpus.h"
17 #include "sysemu/reset.h"
18 #include "migration/vmstate.h"
19 #include "monitor/monitor.h"
20 #include "hw/ppc/fdt.h"
21 #include "hw/ppc/spapr.h"
22 #include "hw/ppc/spapr_cpu_core.h"
23 #include "hw/ppc/spapr_xive.h"
24 #include "hw/ppc/xive.h"
25 #include "hw/ppc/xive_regs.h"
26 #include "hw/qdev-properties.h"
27 
28 /*
29  * XIVE Virtualization Controller BAR and Thread Managment BAR that we
30  * use for the ESB pages and the TIMA pages
31  */
32 #define SPAPR_XIVE_VC_BASE   0x0006010000000000ull
33 #define SPAPR_XIVE_TM_BASE   0x0006030203180000ull
34 
35 /*
36  * The allocation of VP blocks is a complex operation in OPAL and the
37  * VP identifiers have a relation with the number of HW chips, the
38  * size of the VP blocks, VP grouping, etc. The QEMU sPAPR XIVE
39  * controller model does not have the same constraints and can use a
40  * simple mapping scheme of the CPU vcpu_id
41  *
42  * These identifiers are never returned to the OS.
43  */
44 
45 #define SPAPR_XIVE_NVT_BASE 0x400
46 
47 /*
48  * sPAPR NVT and END indexing helpers
49  */
50 static uint32_t spapr_xive_nvt_to_target(uint8_t nvt_blk, uint32_t nvt_idx)
51 {
52     return nvt_idx - SPAPR_XIVE_NVT_BASE;
53 }
54 
55 static void spapr_xive_cpu_to_nvt(PowerPCCPU *cpu,
56                                   uint8_t *out_nvt_blk, uint32_t *out_nvt_idx)
57 {
58     assert(cpu);
59 
60     if (out_nvt_blk) {
61         *out_nvt_blk = SPAPR_XIVE_BLOCK_ID;
62     }
63 
64     if (out_nvt_blk) {
65         *out_nvt_idx = SPAPR_XIVE_NVT_BASE + cpu->vcpu_id;
66     }
67 }
68 
69 static int spapr_xive_target_to_nvt(uint32_t target,
70                                     uint8_t *out_nvt_blk, uint32_t *out_nvt_idx)
71 {
72     PowerPCCPU *cpu = spapr_find_cpu(target);
73 
74     if (!cpu) {
75         return -1;
76     }
77 
78     spapr_xive_cpu_to_nvt(cpu, out_nvt_blk, out_nvt_idx);
79     return 0;
80 }
81 
82 /*
83  * sPAPR END indexing uses a simple mapping of the CPU vcpu_id, 8
84  * priorities per CPU
85  */
86 int spapr_xive_end_to_target(uint8_t end_blk, uint32_t end_idx,
87                              uint32_t *out_server, uint8_t *out_prio)
88 {
89 
90     assert(end_blk == SPAPR_XIVE_BLOCK_ID);
91 
92     if (out_server) {
93         *out_server = end_idx >> 3;
94     }
95 
96     if (out_prio) {
97         *out_prio = end_idx & 0x7;
98     }
99     return 0;
100 }
101 
102 static void spapr_xive_cpu_to_end(PowerPCCPU *cpu, uint8_t prio,
103                                   uint8_t *out_end_blk, uint32_t *out_end_idx)
104 {
105     assert(cpu);
106 
107     if (out_end_blk) {
108         *out_end_blk = SPAPR_XIVE_BLOCK_ID;
109     }
110 
111     if (out_end_idx) {
112         *out_end_idx = (cpu->vcpu_id << 3) + prio;
113     }
114 }
115 
116 static int spapr_xive_target_to_end(uint32_t target, uint8_t prio,
117                                     uint8_t *out_end_blk, uint32_t *out_end_idx)
118 {
119     PowerPCCPU *cpu = spapr_find_cpu(target);
120 
121     if (!cpu) {
122         return -1;
123     }
124 
125     spapr_xive_cpu_to_end(cpu, prio, out_end_blk, out_end_idx);
126     return 0;
127 }
128 
129 /*
130  * On sPAPR machines, use a simplified output for the XIVE END
131  * structure dumping only the information related to the OS EQ.
132  */
133 static void spapr_xive_end_pic_print_info(SpaprXive *xive, XiveEND *end,
134                                           Monitor *mon)
135 {
136     uint64_t qaddr_base = xive_end_qaddr(end);
137     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
138     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
139     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
140     uint32_t qentries = 1 << (qsize + 10);
141     uint32_t nvt = xive_get_field32(END_W6_NVT_INDEX, end->w6);
142     uint8_t priority = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
143 
144     monitor_printf(mon, "%3d/%d % 6d/%5d @%"PRIx64" ^%d",
145                    spapr_xive_nvt_to_target(0, nvt),
146                    priority, qindex, qentries, qaddr_base, qgen);
147 
148     xive_end_queue_pic_print_info(end, 6, mon);
149 }
150 
151 void spapr_xive_pic_print_info(SpaprXive *xive, Monitor *mon)
152 {
153     XiveSource *xsrc = &xive->source;
154     int i;
155 
156     if (kvm_irqchip_in_kernel()) {
157         Error *local_err = NULL;
158 
159         kvmppc_xive_synchronize_state(xive, &local_err);
160         if (local_err) {
161             error_report_err(local_err);
162             return;
163         }
164     }
165 
166     monitor_printf(mon, "  LISN         PQ    EISN     CPU/PRIO EQ\n");
167 
168     for (i = 0; i < xive->nr_irqs; i++) {
169         uint8_t pq = xive_source_esb_get(xsrc, i);
170         XiveEAS *eas = &xive->eat[i];
171 
172         if (!xive_eas_is_valid(eas)) {
173             continue;
174         }
175 
176         monitor_printf(mon, "  %08x %s %c%c%c %s %08x ", i,
177                        xive_source_irq_is_lsi(xsrc, i) ? "LSI" : "MSI",
178                        pq & XIVE_ESB_VAL_P ? 'P' : '-',
179                        pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
180                        xsrc->status[i] & XIVE_STATUS_ASSERTED ? 'A' : ' ',
181                        xive_eas_is_masked(eas) ? "M" : " ",
182                        (int) xive_get_field64(EAS_END_DATA, eas->w));
183 
184         if (!xive_eas_is_masked(eas)) {
185             uint32_t end_idx = xive_get_field64(EAS_END_INDEX, eas->w);
186             XiveEND *end;
187 
188             assert(end_idx < xive->nr_ends);
189             end = &xive->endt[end_idx];
190 
191             if (xive_end_is_valid(end)) {
192                 spapr_xive_end_pic_print_info(xive, end, mon);
193             }
194         }
195         monitor_printf(mon, "\n");
196     }
197 }
198 
199 void spapr_xive_mmio_set_enabled(SpaprXive *xive, bool enable)
200 {
201     memory_region_set_enabled(&xive->source.esb_mmio, enable);
202     memory_region_set_enabled(&xive->tm_mmio, enable);
203 
204     /* Disable the END ESBs until a guest OS makes use of them */
205     memory_region_set_enabled(&xive->end_source.esb_mmio, false);
206 }
207 
208 static void spapr_xive_tm_write(void *opaque, hwaddr offset,
209                           uint64_t value, unsigned size)
210 {
211     XiveTCTX *tctx = spapr_cpu_state(POWERPC_CPU(current_cpu))->tctx;
212 
213     xive_tctx_tm_write(XIVE_PRESENTER(opaque), tctx, offset, value, size);
214 }
215 
216 static uint64_t spapr_xive_tm_read(void *opaque, hwaddr offset, unsigned size)
217 {
218     XiveTCTX *tctx = spapr_cpu_state(POWERPC_CPU(current_cpu))->tctx;
219 
220     return xive_tctx_tm_read(XIVE_PRESENTER(opaque), tctx, offset, size);
221 }
222 
223 const MemoryRegionOps spapr_xive_tm_ops = {
224     .read = spapr_xive_tm_read,
225     .write = spapr_xive_tm_write,
226     .endianness = DEVICE_BIG_ENDIAN,
227     .valid = {
228         .min_access_size = 1,
229         .max_access_size = 8,
230     },
231     .impl = {
232         .min_access_size = 1,
233         .max_access_size = 8,
234     },
235 };
236 
237 static void spapr_xive_end_reset(XiveEND *end)
238 {
239     memset(end, 0, sizeof(*end));
240 
241     /* switch off the escalation and notification ESBs */
242     end->w1 = cpu_to_be32(END_W1_ESe_Q | END_W1_ESn_Q);
243 }
244 
245 static void spapr_xive_reset(void *dev)
246 {
247     SpaprXive *xive = SPAPR_XIVE(dev);
248     int i;
249 
250     /*
251      * The XiveSource has its own reset handler, which mask off all
252      * IRQs (!P|Q)
253      */
254 
255     /* Mask all valid EASs in the IRQ number space. */
256     for (i = 0; i < xive->nr_irqs; i++) {
257         XiveEAS *eas = &xive->eat[i];
258         if (xive_eas_is_valid(eas)) {
259             eas->w = cpu_to_be64(EAS_VALID | EAS_MASKED);
260         } else {
261             eas->w = 0;
262         }
263     }
264 
265     /* Clear all ENDs */
266     for (i = 0; i < xive->nr_ends; i++) {
267         spapr_xive_end_reset(&xive->endt[i]);
268     }
269 }
270 
271 static void spapr_xive_instance_init(Object *obj)
272 {
273     SpaprXive *xive = SPAPR_XIVE(obj);
274 
275     object_initialize_child(obj, "source", &xive->source, sizeof(xive->source),
276                             TYPE_XIVE_SOURCE, &error_abort, NULL);
277 
278     object_initialize_child(obj, "end_source", &xive->end_source,
279                             sizeof(xive->end_source), TYPE_XIVE_END_SOURCE,
280                             &error_abort, NULL);
281 
282     /* Not connected to the KVM XIVE device */
283     xive->fd = -1;
284 }
285 
286 static void spapr_xive_realize(DeviceState *dev, Error **errp)
287 {
288     SpaprXive *xive = SPAPR_XIVE(dev);
289     XiveSource *xsrc = &xive->source;
290     XiveENDSource *end_xsrc = &xive->end_source;
291     Error *local_err = NULL;
292 
293     if (!xive->nr_irqs) {
294         error_setg(errp, "Number of interrupt needs to be greater 0");
295         return;
296     }
297 
298     if (!xive->nr_ends) {
299         error_setg(errp, "Number of interrupt needs to be greater 0");
300         return;
301     }
302 
303     /*
304      * Initialize the internal sources, for IPIs and virtual devices.
305      */
306     object_property_set_int(OBJECT(xsrc), xive->nr_irqs, "nr-irqs",
307                             &error_fatal);
308     object_property_set_link(OBJECT(xsrc), OBJECT(xive), "xive",
309                              &error_abort);
310     object_property_set_bool(OBJECT(xsrc), true, "realized", &local_err);
311     if (local_err) {
312         error_propagate(errp, local_err);
313         return;
314     }
315     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &xsrc->esb_mmio);
316 
317     /*
318      * Initialize the END ESB source
319      */
320     object_property_set_int(OBJECT(end_xsrc), xive->nr_irqs, "nr-ends",
321                             &error_fatal);
322     object_property_set_link(OBJECT(end_xsrc), OBJECT(xive), "xive",
323                              &error_abort);
324     object_property_set_bool(OBJECT(end_xsrc), true, "realized", &local_err);
325     if (local_err) {
326         error_propagate(errp, local_err);
327         return;
328     }
329     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &end_xsrc->esb_mmio);
330 
331     /* Set the mapping address of the END ESB pages after the source ESBs */
332     xive->end_base = xive->vc_base + (1ull << xsrc->esb_shift) * xsrc->nr_irqs;
333 
334     /*
335      * Allocate the routing tables
336      */
337     xive->eat = g_new0(XiveEAS, xive->nr_irqs);
338     xive->endt = g_new0(XiveEND, xive->nr_ends);
339 
340     xive->nodename = g_strdup_printf("interrupt-controller@%" PRIx64,
341                            xive->tm_base + XIVE_TM_USER_PAGE * (1 << TM_SHIFT));
342 
343     qemu_register_reset(spapr_xive_reset, dev);
344 
345     /* TIMA initialization */
346     memory_region_init_io(&xive->tm_mmio, OBJECT(xive), &spapr_xive_tm_ops,
347                           xive, "xive.tima", 4ull << TM_SHIFT);
348     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &xive->tm_mmio);
349 
350     /*
351      * Map all regions. These will be enabled or disabled at reset and
352      * can also be overridden by KVM memory regions if active
353      */
354     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 0, xive->vc_base);
355     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 1, xive->end_base);
356     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 2, xive->tm_base);
357 }
358 
359 static int spapr_xive_get_eas(XiveRouter *xrtr, uint8_t eas_blk,
360                               uint32_t eas_idx, XiveEAS *eas)
361 {
362     SpaprXive *xive = SPAPR_XIVE(xrtr);
363 
364     if (eas_idx >= xive->nr_irqs) {
365         return -1;
366     }
367 
368     *eas = xive->eat[eas_idx];
369     return 0;
370 }
371 
372 static int spapr_xive_get_end(XiveRouter *xrtr,
373                               uint8_t end_blk, uint32_t end_idx, XiveEND *end)
374 {
375     SpaprXive *xive = SPAPR_XIVE(xrtr);
376 
377     if (end_idx >= xive->nr_ends) {
378         return -1;
379     }
380 
381     memcpy(end, &xive->endt[end_idx], sizeof(XiveEND));
382     return 0;
383 }
384 
385 static int spapr_xive_write_end(XiveRouter *xrtr, uint8_t end_blk,
386                                 uint32_t end_idx, XiveEND *end,
387                                 uint8_t word_number)
388 {
389     SpaprXive *xive = SPAPR_XIVE(xrtr);
390 
391     if (end_idx >= xive->nr_ends) {
392         return -1;
393     }
394 
395     memcpy(&xive->endt[end_idx], end, sizeof(XiveEND));
396     return 0;
397 }
398 
399 static int spapr_xive_get_nvt(XiveRouter *xrtr,
400                               uint8_t nvt_blk, uint32_t nvt_idx, XiveNVT *nvt)
401 {
402     uint32_t vcpu_id = spapr_xive_nvt_to_target(nvt_blk, nvt_idx);
403     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
404 
405     if (!cpu) {
406         /* TODO: should we assert() if we can find a NVT ? */
407         return -1;
408     }
409 
410     /*
411      * sPAPR does not maintain a NVT table. Return that the NVT is
412      * valid if we have found a matching CPU
413      */
414     nvt->w0 = cpu_to_be32(NVT_W0_VALID);
415     return 0;
416 }
417 
418 static int spapr_xive_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk,
419                                 uint32_t nvt_idx, XiveNVT *nvt,
420                                 uint8_t word_number)
421 {
422     /*
423      * We don't need to write back to the NVTs because the sPAPR
424      * machine should never hit a non-scheduled NVT. It should never
425      * get called.
426      */
427     g_assert_not_reached();
428 }
429 
430 static int spapr_xive_match_nvt(XivePresenter *xptr, uint8_t format,
431                                 uint8_t nvt_blk, uint32_t nvt_idx,
432                                 bool cam_ignore, uint8_t priority,
433                                 uint32_t logic_serv, XiveTCTXMatch *match)
434 {
435     CPUState *cs;
436     int count = 0;
437 
438     CPU_FOREACH(cs) {
439         PowerPCCPU *cpu = POWERPC_CPU(cs);
440         XiveTCTX *tctx = spapr_cpu_state(cpu)->tctx;
441         int ring;
442 
443         /*
444          * Skip partially initialized vCPUs. This can happen when
445          * vCPUs are hotplugged.
446          */
447         if (!tctx) {
448             continue;
449         }
450 
451         /*
452          * Check the thread context CAM lines and record matches.
453          */
454         ring = xive_presenter_tctx_match(xptr, tctx, format, nvt_blk, nvt_idx,
455                                          cam_ignore, logic_serv);
456         /*
457          * Save the matching thread interrupt context and follow on to
458          * check for duplicates which are invalid.
459          */
460         if (ring != -1) {
461             if (match->tctx) {
462                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: already found a thread "
463                               "context NVT %x/%x\n", nvt_blk, nvt_idx);
464                 return -1;
465             }
466 
467             match->ring = ring;
468             match->tctx = tctx;
469             count++;
470         }
471     }
472 
473     return count;
474 }
475 
476 static const VMStateDescription vmstate_spapr_xive_end = {
477     .name = TYPE_SPAPR_XIVE "/end",
478     .version_id = 1,
479     .minimum_version_id = 1,
480     .fields = (VMStateField []) {
481         VMSTATE_UINT32(w0, XiveEND),
482         VMSTATE_UINT32(w1, XiveEND),
483         VMSTATE_UINT32(w2, XiveEND),
484         VMSTATE_UINT32(w3, XiveEND),
485         VMSTATE_UINT32(w4, XiveEND),
486         VMSTATE_UINT32(w5, XiveEND),
487         VMSTATE_UINT32(w6, XiveEND),
488         VMSTATE_UINT32(w7, XiveEND),
489         VMSTATE_END_OF_LIST()
490     },
491 };
492 
493 static const VMStateDescription vmstate_spapr_xive_eas = {
494     .name = TYPE_SPAPR_XIVE "/eas",
495     .version_id = 1,
496     .minimum_version_id = 1,
497     .fields = (VMStateField []) {
498         VMSTATE_UINT64(w, XiveEAS),
499         VMSTATE_END_OF_LIST()
500     },
501 };
502 
503 static int vmstate_spapr_xive_pre_save(void *opaque)
504 {
505     if (kvm_irqchip_in_kernel()) {
506         return kvmppc_xive_pre_save(SPAPR_XIVE(opaque));
507     }
508 
509     return 0;
510 }
511 
512 /*
513  * Called by the sPAPR IRQ backend 'post_load' method at the machine
514  * level.
515  */
516 static int spapr_xive_post_load(SpaprInterruptController *intc, int version_id)
517 {
518     if (kvm_irqchip_in_kernel()) {
519         return kvmppc_xive_post_load(SPAPR_XIVE(intc), version_id);
520     }
521 
522     return 0;
523 }
524 
525 static const VMStateDescription vmstate_spapr_xive = {
526     .name = TYPE_SPAPR_XIVE,
527     .version_id = 1,
528     .minimum_version_id = 1,
529     .pre_save = vmstate_spapr_xive_pre_save,
530     .post_load = NULL, /* handled at the machine level */
531     .fields = (VMStateField[]) {
532         VMSTATE_UINT32_EQUAL(nr_irqs, SpaprXive, NULL),
533         VMSTATE_STRUCT_VARRAY_POINTER_UINT32(eat, SpaprXive, nr_irqs,
534                                      vmstate_spapr_xive_eas, XiveEAS),
535         VMSTATE_STRUCT_VARRAY_POINTER_UINT32(endt, SpaprXive, nr_ends,
536                                              vmstate_spapr_xive_end, XiveEND),
537         VMSTATE_END_OF_LIST()
538     },
539 };
540 
541 static int spapr_xive_claim_irq(SpaprInterruptController *intc, int lisn,
542                                 bool lsi, Error **errp)
543 {
544     SpaprXive *xive = SPAPR_XIVE(intc);
545     XiveSource *xsrc = &xive->source;
546 
547     assert(lisn < xive->nr_irqs);
548 
549     if (xive_eas_is_valid(&xive->eat[lisn])) {
550         error_setg(errp, "IRQ %d is not free", lisn);
551         return -EBUSY;
552     }
553 
554     /*
555      * Set default values when allocating an IRQ number
556      */
557     xive->eat[lisn].w |= cpu_to_be64(EAS_VALID | EAS_MASKED);
558     if (lsi) {
559         xive_source_irq_set_lsi(xsrc, lisn);
560     }
561 
562     if (kvm_irqchip_in_kernel()) {
563         return kvmppc_xive_source_reset_one(xsrc, lisn, errp);
564     }
565 
566     return 0;
567 }
568 
569 static void spapr_xive_free_irq(SpaprInterruptController *intc, int lisn)
570 {
571     SpaprXive *xive = SPAPR_XIVE(intc);
572     assert(lisn < xive->nr_irqs);
573 
574     xive->eat[lisn].w &= cpu_to_be64(~EAS_VALID);
575 }
576 
577 static Property spapr_xive_properties[] = {
578     DEFINE_PROP_UINT32("nr-irqs", SpaprXive, nr_irqs, 0),
579     DEFINE_PROP_UINT32("nr-ends", SpaprXive, nr_ends, 0),
580     DEFINE_PROP_UINT64("vc-base", SpaprXive, vc_base, SPAPR_XIVE_VC_BASE),
581     DEFINE_PROP_UINT64("tm-base", SpaprXive, tm_base, SPAPR_XIVE_TM_BASE),
582     DEFINE_PROP_END_OF_LIST(),
583 };
584 
585 static int spapr_xive_cpu_intc_create(SpaprInterruptController *intc,
586                                       PowerPCCPU *cpu, Error **errp)
587 {
588     SpaprXive *xive = SPAPR_XIVE(intc);
589     Object *obj;
590     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
591 
592     obj = xive_tctx_create(OBJECT(cpu), XIVE_ROUTER(xive), errp);
593     if (!obj) {
594         return -1;
595     }
596 
597     spapr_cpu->tctx = XIVE_TCTX(obj);
598     return 0;
599 }
600 
601 static void xive_tctx_set_os_cam(XiveTCTX *tctx, uint32_t os_cam)
602 {
603     uint32_t qw1w2 = cpu_to_be32(TM_QW1W2_VO | os_cam);
604     memcpy(&tctx->regs[TM_QW1_OS + TM_WORD2], &qw1w2, 4);
605 }
606 
607 static void spapr_xive_cpu_intc_reset(SpaprInterruptController *intc,
608                                      PowerPCCPU *cpu)
609 {
610     XiveTCTX *tctx = spapr_cpu_state(cpu)->tctx;
611     uint8_t  nvt_blk;
612     uint32_t nvt_idx;
613 
614     xive_tctx_reset(tctx);
615 
616     /*
617      * When a Virtual Processor is scheduled to run on a HW thread,
618      * the hypervisor pushes its identifier in the OS CAM line.
619      * Emulate the same behavior under QEMU.
620      */
621     spapr_xive_cpu_to_nvt(cpu, &nvt_blk, &nvt_idx);
622 
623     xive_tctx_set_os_cam(tctx, xive_nvt_cam_line(nvt_blk, nvt_idx));
624 }
625 
626 static void spapr_xive_cpu_intc_destroy(SpaprInterruptController *intc,
627                                         PowerPCCPU *cpu)
628 {
629     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
630 
631     xive_tctx_destroy(spapr_cpu->tctx);
632     spapr_cpu->tctx = NULL;
633 }
634 
635 static void spapr_xive_set_irq(SpaprInterruptController *intc, int irq, int val)
636 {
637     SpaprXive *xive = SPAPR_XIVE(intc);
638 
639     if (kvm_irqchip_in_kernel()) {
640         kvmppc_xive_source_set_irq(&xive->source, irq, val);
641     } else {
642         xive_source_set_irq(&xive->source, irq, val);
643     }
644 }
645 
646 static void spapr_xive_print_info(SpaprInterruptController *intc, Monitor *mon)
647 {
648     SpaprXive *xive = SPAPR_XIVE(intc);
649     CPUState *cs;
650 
651     CPU_FOREACH(cs) {
652         PowerPCCPU *cpu = POWERPC_CPU(cs);
653 
654         xive_tctx_pic_print_info(spapr_cpu_state(cpu)->tctx, mon);
655     }
656 
657     spapr_xive_pic_print_info(xive, mon);
658 }
659 
660 static void spapr_xive_dt(SpaprInterruptController *intc, uint32_t nr_servers,
661                           void *fdt, uint32_t phandle)
662 {
663     SpaprXive *xive = SPAPR_XIVE(intc);
664     int node;
665     uint64_t timas[2 * 2];
666     /* Interrupt number ranges for the IPIs */
667     uint32_t lisn_ranges[] = {
668         cpu_to_be32(0),
669         cpu_to_be32(nr_servers),
670     };
671     /*
672      * EQ size - the sizes of pages supported by the system 4K, 64K,
673      * 2M, 16M. We only advertise 64K for the moment.
674      */
675     uint32_t eq_sizes[] = {
676         cpu_to_be32(16), /* 64K */
677     };
678     /*
679      * The following array is in sync with the reserved priorities
680      * defined by the 'spapr_xive_priority_is_reserved' routine.
681      */
682     uint32_t plat_res_int_priorities[] = {
683         cpu_to_be32(7),    /* start */
684         cpu_to_be32(0xf8), /* count */
685     };
686 
687     /* Thread Interrupt Management Area : User (ring 3) and OS (ring 2) */
688     timas[0] = cpu_to_be64(xive->tm_base +
689                            XIVE_TM_USER_PAGE * (1ull << TM_SHIFT));
690     timas[1] = cpu_to_be64(1ull << TM_SHIFT);
691     timas[2] = cpu_to_be64(xive->tm_base +
692                            XIVE_TM_OS_PAGE * (1ull << TM_SHIFT));
693     timas[3] = cpu_to_be64(1ull << TM_SHIFT);
694 
695     _FDT(node = fdt_add_subnode(fdt, 0, xive->nodename));
696 
697     _FDT(fdt_setprop_string(fdt, node, "device_type", "power-ivpe"));
698     _FDT(fdt_setprop(fdt, node, "reg", timas, sizeof(timas)));
699 
700     _FDT(fdt_setprop_string(fdt, node, "compatible", "ibm,power-ivpe"));
701     _FDT(fdt_setprop(fdt, node, "ibm,xive-eq-sizes", eq_sizes,
702                      sizeof(eq_sizes)));
703     _FDT(fdt_setprop(fdt, node, "ibm,xive-lisn-ranges", lisn_ranges,
704                      sizeof(lisn_ranges)));
705 
706     /* For Linux to link the LSIs to the interrupt controller. */
707     _FDT(fdt_setprop(fdt, node, "interrupt-controller", NULL, 0));
708     _FDT(fdt_setprop_cell(fdt, node, "#interrupt-cells", 2));
709 
710     /* For SLOF */
711     _FDT(fdt_setprop_cell(fdt, node, "linux,phandle", phandle));
712     _FDT(fdt_setprop_cell(fdt, node, "phandle", phandle));
713 
714     /*
715      * The "ibm,plat-res-int-priorities" property defines the priority
716      * ranges reserved by the hypervisor
717      */
718     _FDT(fdt_setprop(fdt, 0, "ibm,plat-res-int-priorities",
719                      plat_res_int_priorities, sizeof(plat_res_int_priorities)));
720 }
721 
722 static int spapr_xive_activate(SpaprInterruptController *intc,
723                                uint32_t nr_servers, Error **errp)
724 {
725     SpaprXive *xive = SPAPR_XIVE(intc);
726 
727     if (kvm_enabled()) {
728         int rc = spapr_irq_init_kvm(kvmppc_xive_connect, intc, nr_servers,
729                                     errp);
730         if (rc < 0) {
731             return rc;
732         }
733     }
734 
735     /* Activate the XIVE MMIOs */
736     spapr_xive_mmio_set_enabled(xive, true);
737 
738     return 0;
739 }
740 
741 static void spapr_xive_deactivate(SpaprInterruptController *intc)
742 {
743     SpaprXive *xive = SPAPR_XIVE(intc);
744 
745     spapr_xive_mmio_set_enabled(xive, false);
746 
747     if (kvm_irqchip_in_kernel()) {
748         kvmppc_xive_disconnect(intc);
749     }
750 }
751 
752 static void spapr_xive_class_init(ObjectClass *klass, void *data)
753 {
754     DeviceClass *dc = DEVICE_CLASS(klass);
755     XiveRouterClass *xrc = XIVE_ROUTER_CLASS(klass);
756     SpaprInterruptControllerClass *sicc = SPAPR_INTC_CLASS(klass);
757     XivePresenterClass *xpc = XIVE_PRESENTER_CLASS(klass);
758 
759     dc->desc    = "sPAPR XIVE Interrupt Controller";
760     dc->props   = spapr_xive_properties;
761     dc->realize = spapr_xive_realize;
762     dc->vmsd    = &vmstate_spapr_xive;
763 
764     xrc->get_eas = spapr_xive_get_eas;
765     xrc->get_end = spapr_xive_get_end;
766     xrc->write_end = spapr_xive_write_end;
767     xrc->get_nvt = spapr_xive_get_nvt;
768     xrc->write_nvt = spapr_xive_write_nvt;
769 
770     sicc->activate = spapr_xive_activate;
771     sicc->deactivate = spapr_xive_deactivate;
772     sicc->cpu_intc_create = spapr_xive_cpu_intc_create;
773     sicc->cpu_intc_reset = spapr_xive_cpu_intc_reset;
774     sicc->cpu_intc_destroy = spapr_xive_cpu_intc_destroy;
775     sicc->claim_irq = spapr_xive_claim_irq;
776     sicc->free_irq = spapr_xive_free_irq;
777     sicc->set_irq = spapr_xive_set_irq;
778     sicc->print_info = spapr_xive_print_info;
779     sicc->dt = spapr_xive_dt;
780     sicc->post_load = spapr_xive_post_load;
781 
782     xpc->match_nvt  = spapr_xive_match_nvt;
783 }
784 
785 static const TypeInfo spapr_xive_info = {
786     .name = TYPE_SPAPR_XIVE,
787     .parent = TYPE_XIVE_ROUTER,
788     .instance_init = spapr_xive_instance_init,
789     .instance_size = sizeof(SpaprXive),
790     .class_init = spapr_xive_class_init,
791     .interfaces = (InterfaceInfo[]) {
792         { TYPE_SPAPR_INTC },
793         { }
794     },
795 };
796 
797 static void spapr_xive_register_types(void)
798 {
799     type_register_static(&spapr_xive_info);
800 }
801 
802 type_init(spapr_xive_register_types)
803 
804 /*
805  * XIVE hcalls
806  *
807  * The terminology used by the XIVE hcalls is the following :
808  *
809  *   TARGET vCPU number
810  *   EQ     Event Queue assigned by OS to receive event data
811  *   ESB    page for source interrupt management
812  *   LISN   Logical Interrupt Source Number identifying a source in the
813  *          machine
814  *   EISN   Effective Interrupt Source Number used by guest OS to
815  *          identify source in the guest
816  *
817  * The EAS, END, NVT structures are not exposed.
818  */
819 
820 /*
821  * Linux hosts under OPAL reserve priority 7 for their own escalation
822  * interrupts (DD2.X POWER9). So we only allow the guest to use
823  * priorities [0..6].
824  */
825 static bool spapr_xive_priority_is_reserved(uint8_t priority)
826 {
827     switch (priority) {
828     case 0 ... 6:
829         return false;
830     case 7: /* OPAL escalation queue */
831     default:
832         return true;
833     }
834 }
835 
836 /*
837  * The H_INT_GET_SOURCE_INFO hcall() is used to obtain the logical
838  * real address of the MMIO page through which the Event State Buffer
839  * entry associated with the value of the "lisn" parameter is managed.
840  *
841  * Parameters:
842  * Input
843  * - R4: "flags"
844  *         Bits 0-63 reserved
845  * - R5: "lisn" is per "interrupts", "interrupt-map", or
846  *       "ibm,xive-lisn-ranges" properties, or as returned by the
847  *       ibm,query-interrupt-source-number RTAS call, or as returned
848  *       by the H_ALLOCATE_VAS_WINDOW hcall
849  *
850  * Output
851  * - R4: "flags"
852  *         Bits 0-59: Reserved
853  *         Bit 60: H_INT_ESB must be used for Event State Buffer
854  *                 management
855  *         Bit 61: 1 == LSI  0 == MSI
856  *         Bit 62: the full function page supports trigger
857  *         Bit 63: Store EOI Supported
858  * - R5: Logical Real address of full function Event State Buffer
859  *       management page, -1 if H_INT_ESB hcall flag is set to 1.
860  * - R6: Logical Real Address of trigger only Event State Buffer
861  *       management page or -1.
862  * - R7: Power of 2 page size for the ESB management pages returned in
863  *       R5 and R6.
864  */
865 
866 #define SPAPR_XIVE_SRC_H_INT_ESB     PPC_BIT(60) /* ESB manage with H_INT_ESB */
867 #define SPAPR_XIVE_SRC_LSI           PPC_BIT(61) /* Virtual LSI type */
868 #define SPAPR_XIVE_SRC_TRIGGER       PPC_BIT(62) /* Trigger and management
869                                                     on same page */
870 #define SPAPR_XIVE_SRC_STORE_EOI     PPC_BIT(63) /* Store EOI support */
871 
872 static target_ulong h_int_get_source_info(PowerPCCPU *cpu,
873                                           SpaprMachineState *spapr,
874                                           target_ulong opcode,
875                                           target_ulong *args)
876 {
877     SpaprXive *xive = spapr->xive;
878     XiveSource *xsrc = &xive->source;
879     target_ulong flags  = args[0];
880     target_ulong lisn   = args[1];
881 
882     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
883         return H_FUNCTION;
884     }
885 
886     if (flags) {
887         return H_PARAMETER;
888     }
889 
890     if (lisn >= xive->nr_irqs) {
891         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
892                       lisn);
893         return H_P2;
894     }
895 
896     if (!xive_eas_is_valid(&xive->eat[lisn])) {
897         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
898                       lisn);
899         return H_P2;
900     }
901 
902     /*
903      * All sources are emulated under the main XIVE object and share
904      * the same characteristics.
905      */
906     args[0] = 0;
907     if (!xive_source_esb_has_2page(xsrc)) {
908         args[0] |= SPAPR_XIVE_SRC_TRIGGER;
909     }
910     if (xsrc->esb_flags & XIVE_SRC_STORE_EOI) {
911         args[0] |= SPAPR_XIVE_SRC_STORE_EOI;
912     }
913 
914     /*
915      * Force the use of the H_INT_ESB hcall in case of an LSI
916      * interrupt. This is necessary under KVM to re-trigger the
917      * interrupt if the level is still asserted
918      */
919     if (xive_source_irq_is_lsi(xsrc, lisn)) {
920         args[0] |= SPAPR_XIVE_SRC_H_INT_ESB | SPAPR_XIVE_SRC_LSI;
921     }
922 
923     if (!(args[0] & SPAPR_XIVE_SRC_H_INT_ESB)) {
924         args[1] = xive->vc_base + xive_source_esb_mgmt(xsrc, lisn);
925     } else {
926         args[1] = -1;
927     }
928 
929     if (xive_source_esb_has_2page(xsrc) &&
930         !(args[0] & SPAPR_XIVE_SRC_H_INT_ESB)) {
931         args[2] = xive->vc_base + xive_source_esb_page(xsrc, lisn);
932     } else {
933         args[2] = -1;
934     }
935 
936     if (xive_source_esb_has_2page(xsrc)) {
937         args[3] = xsrc->esb_shift - 1;
938     } else {
939         args[3] = xsrc->esb_shift;
940     }
941 
942     return H_SUCCESS;
943 }
944 
945 /*
946  * The H_INT_SET_SOURCE_CONFIG hcall() is used to assign a Logical
947  * Interrupt Source to a target. The Logical Interrupt Source is
948  * designated with the "lisn" parameter and the target is designated
949  * with the "target" and "priority" parameters.  Upon return from the
950  * hcall(), no additional interrupts will be directed to the old EQ.
951  *
952  * Parameters:
953  * Input:
954  * - R4: "flags"
955  *         Bits 0-61: Reserved
956  *         Bit 62: set the "eisn" in the EAS
957  *         Bit 63: masks the interrupt source in the hardware interrupt
958  *       control structure. An interrupt masked by this mechanism will
959  *       be dropped, but it's source state bits will still be
960  *       set. There is no race-free way of unmasking and restoring the
961  *       source. Thus this should only be used in interrupts that are
962  *       also masked at the source, and only in cases where the
963  *       interrupt is not meant to be used for a large amount of time
964  *       because no valid target exists for it for example
965  * - R5: "lisn" is per "interrupts", "interrupt-map", or
966  *       "ibm,xive-lisn-ranges" properties, or as returned by the
967  *       ibm,query-interrupt-source-number RTAS call, or as returned by
968  *       the H_ALLOCATE_VAS_WINDOW hcall
969  * - R6: "target" is per "ibm,ppc-interrupt-server#s" or
970  *       "ibm,ppc-interrupt-gserver#s"
971  * - R7: "priority" is a valid priority not in
972  *       "ibm,plat-res-int-priorities"
973  * - R8: "eisn" is the guest EISN associated with the "lisn"
974  *
975  * Output:
976  * - None
977  */
978 
979 #define SPAPR_XIVE_SRC_SET_EISN PPC_BIT(62)
980 #define SPAPR_XIVE_SRC_MASK     PPC_BIT(63)
981 
982 static target_ulong h_int_set_source_config(PowerPCCPU *cpu,
983                                             SpaprMachineState *spapr,
984                                             target_ulong opcode,
985                                             target_ulong *args)
986 {
987     SpaprXive *xive = spapr->xive;
988     XiveEAS eas, new_eas;
989     target_ulong flags    = args[0];
990     target_ulong lisn     = args[1];
991     target_ulong target   = args[2];
992     target_ulong priority = args[3];
993     target_ulong eisn     = args[4];
994     uint8_t end_blk;
995     uint32_t end_idx;
996 
997     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
998         return H_FUNCTION;
999     }
1000 
1001     if (flags & ~(SPAPR_XIVE_SRC_SET_EISN | SPAPR_XIVE_SRC_MASK)) {
1002         return H_PARAMETER;
1003     }
1004 
1005     if (lisn >= xive->nr_irqs) {
1006         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1007                       lisn);
1008         return H_P2;
1009     }
1010 
1011     eas = xive->eat[lisn];
1012     if (!xive_eas_is_valid(&eas)) {
1013         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1014                       lisn);
1015         return H_P2;
1016     }
1017 
1018     /* priority 0xff is used to reset the EAS */
1019     if (priority == 0xff) {
1020         new_eas.w = cpu_to_be64(EAS_VALID | EAS_MASKED);
1021         goto out;
1022     }
1023 
1024     if (flags & SPAPR_XIVE_SRC_MASK) {
1025         new_eas.w = eas.w | cpu_to_be64(EAS_MASKED);
1026     } else {
1027         new_eas.w = eas.w & cpu_to_be64(~EAS_MASKED);
1028     }
1029 
1030     if (spapr_xive_priority_is_reserved(priority)) {
1031         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1032                       " is reserved\n", priority);
1033         return H_P4;
1034     }
1035 
1036     /*
1037      * Validate that "target" is part of the list of threads allocated
1038      * to the partition. For that, find the END corresponding to the
1039      * target.
1040      */
1041     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1042         return H_P3;
1043     }
1044 
1045     new_eas.w = xive_set_field64(EAS_END_BLOCK, new_eas.w, end_blk);
1046     new_eas.w = xive_set_field64(EAS_END_INDEX, new_eas.w, end_idx);
1047 
1048     if (flags & SPAPR_XIVE_SRC_SET_EISN) {
1049         new_eas.w = xive_set_field64(EAS_END_DATA, new_eas.w, eisn);
1050     }
1051 
1052     if (kvm_irqchip_in_kernel()) {
1053         Error *local_err = NULL;
1054 
1055         kvmppc_xive_set_source_config(xive, lisn, &new_eas, &local_err);
1056         if (local_err) {
1057             error_report_err(local_err);
1058             return H_HARDWARE;
1059         }
1060     }
1061 
1062 out:
1063     xive->eat[lisn] = new_eas;
1064     return H_SUCCESS;
1065 }
1066 
1067 /*
1068  * The H_INT_GET_SOURCE_CONFIG hcall() is used to determine to which
1069  * target/priority pair is assigned to the specified Logical Interrupt
1070  * Source.
1071  *
1072  * Parameters:
1073  * Input:
1074  * - R4: "flags"
1075  *         Bits 0-63 Reserved
1076  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1077  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1078  *       ibm,query-interrupt-source-number RTAS call, or as
1079  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1080  *
1081  * Output:
1082  * - R4: Target to which the specified Logical Interrupt Source is
1083  *       assigned
1084  * - R5: Priority to which the specified Logical Interrupt Source is
1085  *       assigned
1086  * - R6: EISN for the specified Logical Interrupt Source (this will be
1087  *       equivalent to the LISN if not changed by H_INT_SET_SOURCE_CONFIG)
1088  */
1089 static target_ulong h_int_get_source_config(PowerPCCPU *cpu,
1090                                             SpaprMachineState *spapr,
1091                                             target_ulong opcode,
1092                                             target_ulong *args)
1093 {
1094     SpaprXive *xive = spapr->xive;
1095     target_ulong flags = args[0];
1096     target_ulong lisn = args[1];
1097     XiveEAS eas;
1098     XiveEND *end;
1099     uint8_t nvt_blk;
1100     uint32_t end_idx, nvt_idx;
1101 
1102     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1103         return H_FUNCTION;
1104     }
1105 
1106     if (flags) {
1107         return H_PARAMETER;
1108     }
1109 
1110     if (lisn >= xive->nr_irqs) {
1111         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1112                       lisn);
1113         return H_P2;
1114     }
1115 
1116     eas = xive->eat[lisn];
1117     if (!xive_eas_is_valid(&eas)) {
1118         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1119                       lisn);
1120         return H_P2;
1121     }
1122 
1123     /* EAS_END_BLOCK is unused on sPAPR */
1124     end_idx = xive_get_field64(EAS_END_INDEX, eas.w);
1125 
1126     assert(end_idx < xive->nr_ends);
1127     end = &xive->endt[end_idx];
1128 
1129     nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end->w6);
1130     nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end->w6);
1131     args[0] = spapr_xive_nvt_to_target(nvt_blk, nvt_idx);
1132 
1133     if (xive_eas_is_masked(&eas)) {
1134         args[1] = 0xff;
1135     } else {
1136         args[1] = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
1137     }
1138 
1139     args[2] = xive_get_field64(EAS_END_DATA, eas.w);
1140 
1141     return H_SUCCESS;
1142 }
1143 
1144 /*
1145  * The H_INT_GET_QUEUE_INFO hcall() is used to get the logical real
1146  * address of the notification management page associated with the
1147  * specified target and priority.
1148  *
1149  * Parameters:
1150  * Input:
1151  * - R4: "flags"
1152  *         Bits 0-63 Reserved
1153  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1154  *       "ibm,ppc-interrupt-gserver#s"
1155  * - R6: "priority" is a valid priority not in
1156  *       "ibm,plat-res-int-priorities"
1157  *
1158  * Output:
1159  * - R4: Logical real address of notification page
1160  * - R5: Power of 2 page size of the notification page
1161  */
1162 static target_ulong h_int_get_queue_info(PowerPCCPU *cpu,
1163                                          SpaprMachineState *spapr,
1164                                          target_ulong opcode,
1165                                          target_ulong *args)
1166 {
1167     SpaprXive *xive = spapr->xive;
1168     XiveENDSource *end_xsrc = &xive->end_source;
1169     target_ulong flags = args[0];
1170     target_ulong target = args[1];
1171     target_ulong priority = args[2];
1172     XiveEND *end;
1173     uint8_t end_blk;
1174     uint32_t end_idx;
1175 
1176     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1177         return H_FUNCTION;
1178     }
1179 
1180     if (flags) {
1181         return H_PARAMETER;
1182     }
1183 
1184     /*
1185      * H_STATE should be returned if a H_INT_RESET is in progress.
1186      * This is not needed when running the emulation under QEMU
1187      */
1188 
1189     if (spapr_xive_priority_is_reserved(priority)) {
1190         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1191                       " is reserved\n", priority);
1192         return H_P3;
1193     }
1194 
1195     /*
1196      * Validate that "target" is part of the list of threads allocated
1197      * to the partition. For that, find the END corresponding to the
1198      * target.
1199      */
1200     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1201         return H_P2;
1202     }
1203 
1204     assert(end_idx < xive->nr_ends);
1205     end = &xive->endt[end_idx];
1206 
1207     args[0] = xive->end_base + (1ull << (end_xsrc->esb_shift + 1)) * end_idx;
1208     if (xive_end_is_enqueue(end)) {
1209         args[1] = xive_get_field32(END_W0_QSIZE, end->w0) + 12;
1210     } else {
1211         args[1] = 0;
1212     }
1213 
1214     return H_SUCCESS;
1215 }
1216 
1217 /*
1218  * The H_INT_SET_QUEUE_CONFIG hcall() is used to set or reset a EQ for
1219  * a given "target" and "priority".  It is also used to set the
1220  * notification config associated with the EQ.  An EQ size of 0 is
1221  * used to reset the EQ config for a given target and priority. If
1222  * resetting the EQ config, the END associated with the given "target"
1223  * and "priority" will be changed to disable queueing.
1224  *
1225  * Upon return from the hcall(), no additional interrupts will be
1226  * directed to the old EQ (if one was set). The old EQ (if one was
1227  * set) should be investigated for interrupts that occurred prior to
1228  * or during the hcall().
1229  *
1230  * Parameters:
1231  * Input:
1232  * - R4: "flags"
1233  *         Bits 0-62: Reserved
1234  *         Bit 63: Unconditional Notify (n) per the XIVE spec
1235  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1236  *       "ibm,ppc-interrupt-gserver#s"
1237  * - R6: "priority" is a valid priority not in
1238  *       "ibm,plat-res-int-priorities"
1239  * - R7: "eventQueue": The logical real address of the start of the EQ
1240  * - R8: "eventQueueSize": The power of 2 EQ size per "ibm,xive-eq-sizes"
1241  *
1242  * Output:
1243  * - None
1244  */
1245 
1246 #define SPAPR_XIVE_END_ALWAYS_NOTIFY PPC_BIT(63)
1247 
1248 static target_ulong h_int_set_queue_config(PowerPCCPU *cpu,
1249                                            SpaprMachineState *spapr,
1250                                            target_ulong opcode,
1251                                            target_ulong *args)
1252 {
1253     SpaprXive *xive = spapr->xive;
1254     target_ulong flags = args[0];
1255     target_ulong target = args[1];
1256     target_ulong priority = args[2];
1257     target_ulong qpage = args[3];
1258     target_ulong qsize = args[4];
1259     XiveEND end;
1260     uint8_t end_blk, nvt_blk;
1261     uint32_t end_idx, nvt_idx;
1262 
1263     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1264         return H_FUNCTION;
1265     }
1266 
1267     if (flags & ~SPAPR_XIVE_END_ALWAYS_NOTIFY) {
1268         return H_PARAMETER;
1269     }
1270 
1271     /*
1272      * H_STATE should be returned if a H_INT_RESET is in progress.
1273      * This is not needed when running the emulation under QEMU
1274      */
1275 
1276     if (spapr_xive_priority_is_reserved(priority)) {
1277         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1278                       " is reserved\n", priority);
1279         return H_P3;
1280     }
1281 
1282     /*
1283      * Validate that "target" is part of the list of threads allocated
1284      * to the partition. For that, find the END corresponding to the
1285      * target.
1286      */
1287 
1288     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1289         return H_P2;
1290     }
1291 
1292     assert(end_idx < xive->nr_ends);
1293     memcpy(&end, &xive->endt[end_idx], sizeof(XiveEND));
1294 
1295     switch (qsize) {
1296     case 12:
1297     case 16:
1298     case 21:
1299     case 24:
1300         if (!QEMU_IS_ALIGNED(qpage, 1ul << qsize)) {
1301             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: EQ @0x%" HWADDR_PRIx
1302                           " is not naturally aligned with %" HWADDR_PRIx "\n",
1303                           qpage, (hwaddr)1 << qsize);
1304             return H_P4;
1305         }
1306         end.w2 = cpu_to_be32((qpage >> 32) & 0x0fffffff);
1307         end.w3 = cpu_to_be32(qpage & 0xffffffff);
1308         end.w0 |= cpu_to_be32(END_W0_ENQUEUE);
1309         end.w0 = xive_set_field32(END_W0_QSIZE, end.w0, qsize - 12);
1310         break;
1311     case 0:
1312         /* reset queue and disable queueing */
1313         spapr_xive_end_reset(&end);
1314         goto out;
1315 
1316     default:
1317         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid EQ size %"PRIx64"\n",
1318                       qsize);
1319         return H_P5;
1320     }
1321 
1322     if (qsize) {
1323         hwaddr plen = 1 << qsize;
1324         void *eq;
1325 
1326         /*
1327          * Validate the guest EQ. We should also check that the queue
1328          * has been zeroed by the OS.
1329          */
1330         eq = address_space_map(CPU(cpu)->as, qpage, &plen, true,
1331                                MEMTXATTRS_UNSPECIFIED);
1332         if (plen != 1 << qsize) {
1333             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to map EQ @0x%"
1334                           HWADDR_PRIx "\n", qpage);
1335             return H_P4;
1336         }
1337         address_space_unmap(CPU(cpu)->as, eq, plen, true, plen);
1338     }
1339 
1340     /* "target" should have been validated above */
1341     if (spapr_xive_target_to_nvt(target, &nvt_blk, &nvt_idx)) {
1342         g_assert_not_reached();
1343     }
1344 
1345     /*
1346      * Ensure the priority and target are correctly set (they will not
1347      * be right after allocation)
1348      */
1349     end.w6 = xive_set_field32(END_W6_NVT_BLOCK, 0ul, nvt_blk) |
1350         xive_set_field32(END_W6_NVT_INDEX, 0ul, nvt_idx);
1351     end.w7 = xive_set_field32(END_W7_F0_PRIORITY, 0ul, priority);
1352 
1353     if (flags & SPAPR_XIVE_END_ALWAYS_NOTIFY) {
1354         end.w0 |= cpu_to_be32(END_W0_UCOND_NOTIFY);
1355     } else {
1356         end.w0 &= cpu_to_be32((uint32_t)~END_W0_UCOND_NOTIFY);
1357     }
1358 
1359     /*
1360      * The generation bit for the END starts at 1 and The END page
1361      * offset counter starts at 0.
1362      */
1363     end.w1 = cpu_to_be32(END_W1_GENERATION) |
1364         xive_set_field32(END_W1_PAGE_OFF, 0ul, 0ul);
1365     end.w0 |= cpu_to_be32(END_W0_VALID);
1366 
1367     /*
1368      * TODO: issue syncs required to ensure all in-flight interrupts
1369      * are complete on the old END
1370      */
1371 
1372 out:
1373     if (kvm_irqchip_in_kernel()) {
1374         Error *local_err = NULL;
1375 
1376         kvmppc_xive_set_queue_config(xive, end_blk, end_idx, &end, &local_err);
1377         if (local_err) {
1378             error_report_err(local_err);
1379             return H_HARDWARE;
1380         }
1381     }
1382 
1383     /* Update END */
1384     memcpy(&xive->endt[end_idx], &end, sizeof(XiveEND));
1385     return H_SUCCESS;
1386 }
1387 
1388 /*
1389  * The H_INT_GET_QUEUE_CONFIG hcall() is used to get a EQ for a given
1390  * target and priority.
1391  *
1392  * Parameters:
1393  * Input:
1394  * - R4: "flags"
1395  *         Bits 0-62: Reserved
1396  *         Bit 63: Debug: Return debug data
1397  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1398  *       "ibm,ppc-interrupt-gserver#s"
1399  * - R6: "priority" is a valid priority not in
1400  *       "ibm,plat-res-int-priorities"
1401  *
1402  * Output:
1403  * - R4: "flags":
1404  *       Bits 0-61: Reserved
1405  *       Bit 62: The value of Event Queue Generation Number (g) per
1406  *              the XIVE spec if "Debug" = 1
1407  *       Bit 63: The value of Unconditional Notify (n) per the XIVE spec
1408  * - R5: The logical real address of the start of the EQ
1409  * - R6: The power of 2 EQ size per "ibm,xive-eq-sizes"
1410  * - R7: The value of Event Queue Offset Counter per XIVE spec
1411  *       if "Debug" = 1, else 0
1412  *
1413  */
1414 
1415 #define SPAPR_XIVE_END_DEBUG     PPC_BIT(63)
1416 
1417 static target_ulong h_int_get_queue_config(PowerPCCPU *cpu,
1418                                            SpaprMachineState *spapr,
1419                                            target_ulong opcode,
1420                                            target_ulong *args)
1421 {
1422     SpaprXive *xive = spapr->xive;
1423     target_ulong flags = args[0];
1424     target_ulong target = args[1];
1425     target_ulong priority = args[2];
1426     XiveEND *end;
1427     uint8_t end_blk;
1428     uint32_t end_idx;
1429 
1430     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1431         return H_FUNCTION;
1432     }
1433 
1434     if (flags & ~SPAPR_XIVE_END_DEBUG) {
1435         return H_PARAMETER;
1436     }
1437 
1438     /*
1439      * H_STATE should be returned if a H_INT_RESET is in progress.
1440      * This is not needed when running the emulation under QEMU
1441      */
1442 
1443     if (spapr_xive_priority_is_reserved(priority)) {
1444         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1445                       " is reserved\n", priority);
1446         return H_P3;
1447     }
1448 
1449     /*
1450      * Validate that "target" is part of the list of threads allocated
1451      * to the partition. For that, find the END corresponding to the
1452      * target.
1453      */
1454     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1455         return H_P2;
1456     }
1457 
1458     assert(end_idx < xive->nr_ends);
1459     end = &xive->endt[end_idx];
1460 
1461     args[0] = 0;
1462     if (xive_end_is_notify(end)) {
1463         args[0] |= SPAPR_XIVE_END_ALWAYS_NOTIFY;
1464     }
1465 
1466     if (xive_end_is_enqueue(end)) {
1467         args[1] = xive_end_qaddr(end);
1468         args[2] = xive_get_field32(END_W0_QSIZE, end->w0) + 12;
1469     } else {
1470         args[1] = 0;
1471         args[2] = 0;
1472     }
1473 
1474     if (kvm_irqchip_in_kernel()) {
1475         Error *local_err = NULL;
1476 
1477         kvmppc_xive_get_queue_config(xive, end_blk, end_idx, end, &local_err);
1478         if (local_err) {
1479             error_report_err(local_err);
1480             return H_HARDWARE;
1481         }
1482     }
1483 
1484     /* TODO: do we need any locking on the END ? */
1485     if (flags & SPAPR_XIVE_END_DEBUG) {
1486         /* Load the event queue generation number into the return flags */
1487         args[0] |= (uint64_t)xive_get_field32(END_W1_GENERATION, end->w1) << 62;
1488 
1489         /* Load R7 with the event queue offset counter */
1490         args[3] = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1491     } else {
1492         args[3] = 0;
1493     }
1494 
1495     return H_SUCCESS;
1496 }
1497 
1498 /*
1499  * The H_INT_SET_OS_REPORTING_LINE hcall() is used to set the
1500  * reporting cache line pair for the calling thread.  The reporting
1501  * cache lines will contain the OS interrupt context when the OS
1502  * issues a CI store byte to @TIMA+0xC10 to acknowledge the OS
1503  * interrupt. The reporting cache lines can be reset by inputting -1
1504  * in "reportingLine".  Issuing the CI store byte without reporting
1505  * cache lines registered will result in the data not being accessible
1506  * to the OS.
1507  *
1508  * Parameters:
1509  * Input:
1510  * - R4: "flags"
1511  *         Bits 0-63: Reserved
1512  * - R5: "reportingLine": The logical real address of the reporting cache
1513  *       line pair
1514  *
1515  * Output:
1516  * - None
1517  */
1518 static target_ulong h_int_set_os_reporting_line(PowerPCCPU *cpu,
1519                                                 SpaprMachineState *spapr,
1520                                                 target_ulong opcode,
1521                                                 target_ulong *args)
1522 {
1523     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1524         return H_FUNCTION;
1525     }
1526 
1527     /*
1528      * H_STATE should be returned if a H_INT_RESET is in progress.
1529      * This is not needed when running the emulation under QEMU
1530      */
1531 
1532     /* TODO: H_INT_SET_OS_REPORTING_LINE */
1533     return H_FUNCTION;
1534 }
1535 
1536 /*
1537  * The H_INT_GET_OS_REPORTING_LINE hcall() is used to get the logical
1538  * real address of the reporting cache line pair set for the input
1539  * "target".  If no reporting cache line pair has been set, -1 is
1540  * returned.
1541  *
1542  * Parameters:
1543  * Input:
1544  * - R4: "flags"
1545  *         Bits 0-63: Reserved
1546  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1547  *       "ibm,ppc-interrupt-gserver#s"
1548  * - R6: "reportingLine": The logical real address of the reporting
1549  *        cache line pair
1550  *
1551  * Output:
1552  * - R4: The logical real address of the reporting line if set, else -1
1553  */
1554 static target_ulong h_int_get_os_reporting_line(PowerPCCPU *cpu,
1555                                                 SpaprMachineState *spapr,
1556                                                 target_ulong opcode,
1557                                                 target_ulong *args)
1558 {
1559     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1560         return H_FUNCTION;
1561     }
1562 
1563     /*
1564      * H_STATE should be returned if a H_INT_RESET is in progress.
1565      * This is not needed when running the emulation under QEMU
1566      */
1567 
1568     /* TODO: H_INT_GET_OS_REPORTING_LINE */
1569     return H_FUNCTION;
1570 }
1571 
1572 /*
1573  * The H_INT_ESB hcall() is used to issue a load or store to the ESB
1574  * page for the input "lisn".  This hcall is only supported for LISNs
1575  * that have the ESB hcall flag set to 1 when returned from hcall()
1576  * H_INT_GET_SOURCE_INFO.
1577  *
1578  * Parameters:
1579  * Input:
1580  * - R4: "flags"
1581  *         Bits 0-62: Reserved
1582  *         bit 63: Store: Store=1, store operation, else load operation
1583  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1584  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1585  *       ibm,query-interrupt-source-number RTAS call, or as
1586  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1587  * - R6: "esbOffset" is the offset into the ESB page for the load or
1588  *       store operation
1589  * - R7: "storeData" is the data to write for a store operation
1590  *
1591  * Output:
1592  * - R4: The value of the load if load operation, else -1
1593  */
1594 
1595 #define SPAPR_XIVE_ESB_STORE PPC_BIT(63)
1596 
1597 static target_ulong h_int_esb(PowerPCCPU *cpu,
1598                               SpaprMachineState *spapr,
1599                               target_ulong opcode,
1600                               target_ulong *args)
1601 {
1602     SpaprXive *xive = spapr->xive;
1603     XiveEAS eas;
1604     target_ulong flags  = args[0];
1605     target_ulong lisn   = args[1];
1606     target_ulong offset = args[2];
1607     target_ulong data   = args[3];
1608     hwaddr mmio_addr;
1609     XiveSource *xsrc = &xive->source;
1610 
1611     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1612         return H_FUNCTION;
1613     }
1614 
1615     if (flags & ~SPAPR_XIVE_ESB_STORE) {
1616         return H_PARAMETER;
1617     }
1618 
1619     if (lisn >= xive->nr_irqs) {
1620         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1621                       lisn);
1622         return H_P2;
1623     }
1624 
1625     eas = xive->eat[lisn];
1626     if (!xive_eas_is_valid(&eas)) {
1627         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1628                       lisn);
1629         return H_P2;
1630     }
1631 
1632     if (offset > (1ull << xsrc->esb_shift)) {
1633         return H_P3;
1634     }
1635 
1636     if (kvm_irqchip_in_kernel()) {
1637         args[0] = kvmppc_xive_esb_rw(xsrc, lisn, offset, data,
1638                                      flags & SPAPR_XIVE_ESB_STORE);
1639     } else {
1640         mmio_addr = xive->vc_base + xive_source_esb_mgmt(xsrc, lisn) + offset;
1641 
1642         if (dma_memory_rw(&address_space_memory, mmio_addr, &data, 8,
1643                           (flags & SPAPR_XIVE_ESB_STORE))) {
1644             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to access ESB @0x%"
1645                           HWADDR_PRIx "\n", mmio_addr);
1646             return H_HARDWARE;
1647         }
1648         args[0] = (flags & SPAPR_XIVE_ESB_STORE) ? -1 : data;
1649     }
1650     return H_SUCCESS;
1651 }
1652 
1653 /*
1654  * The H_INT_SYNC hcall() is used to issue hardware syncs that will
1655  * ensure any in flight events for the input lisn are in the event
1656  * queue.
1657  *
1658  * Parameters:
1659  * Input:
1660  * - R4: "flags"
1661  *         Bits 0-63: Reserved
1662  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1663  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1664  *       ibm,query-interrupt-source-number RTAS call, or as
1665  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1666  *
1667  * Output:
1668  * - None
1669  */
1670 static target_ulong h_int_sync(PowerPCCPU *cpu,
1671                                SpaprMachineState *spapr,
1672                                target_ulong opcode,
1673                                target_ulong *args)
1674 {
1675     SpaprXive *xive = spapr->xive;
1676     XiveEAS eas;
1677     target_ulong flags = args[0];
1678     target_ulong lisn = args[1];
1679 
1680     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1681         return H_FUNCTION;
1682     }
1683 
1684     if (flags) {
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     /*
1702      * H_STATE should be returned if a H_INT_RESET is in progress.
1703      * This is not needed when running the emulation under QEMU
1704      */
1705 
1706     /*
1707      * This is not real hardware. Nothing to be done unless when
1708      * under KVM
1709      */
1710 
1711     if (kvm_irqchip_in_kernel()) {
1712         Error *local_err = NULL;
1713 
1714         kvmppc_xive_sync_source(xive, lisn, &local_err);
1715         if (local_err) {
1716             error_report_err(local_err);
1717             return H_HARDWARE;
1718         }
1719     }
1720     return H_SUCCESS;
1721 }
1722 
1723 /*
1724  * The H_INT_RESET hcall() is used to reset all of the partition's
1725  * interrupt exploitation structures to their initial state.  This
1726  * means losing all previously set interrupt state set via
1727  * H_INT_SET_SOURCE_CONFIG and H_INT_SET_QUEUE_CONFIG.
1728  *
1729  * Parameters:
1730  * Input:
1731  * - R4: "flags"
1732  *         Bits 0-63: Reserved
1733  *
1734  * Output:
1735  * - None
1736  */
1737 static target_ulong h_int_reset(PowerPCCPU *cpu,
1738                                 SpaprMachineState *spapr,
1739                                 target_ulong opcode,
1740                                 target_ulong *args)
1741 {
1742     SpaprXive *xive = spapr->xive;
1743     target_ulong flags   = args[0];
1744 
1745     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1746         return H_FUNCTION;
1747     }
1748 
1749     if (flags) {
1750         return H_PARAMETER;
1751     }
1752 
1753     device_reset(DEVICE(xive));
1754 
1755     if (kvm_irqchip_in_kernel()) {
1756         Error *local_err = NULL;
1757 
1758         kvmppc_xive_reset(xive, &local_err);
1759         if (local_err) {
1760             error_report_err(local_err);
1761             return H_HARDWARE;
1762         }
1763     }
1764     return H_SUCCESS;
1765 }
1766 
1767 void spapr_xive_hcall_init(SpaprMachineState *spapr)
1768 {
1769     spapr_register_hypercall(H_INT_GET_SOURCE_INFO, h_int_get_source_info);
1770     spapr_register_hypercall(H_INT_SET_SOURCE_CONFIG, h_int_set_source_config);
1771     spapr_register_hypercall(H_INT_GET_SOURCE_CONFIG, h_int_get_source_config);
1772     spapr_register_hypercall(H_INT_GET_QUEUE_INFO, h_int_get_queue_info);
1773     spapr_register_hypercall(H_INT_SET_QUEUE_CONFIG, h_int_set_queue_config);
1774     spapr_register_hypercall(H_INT_GET_QUEUE_CONFIG, h_int_get_queue_config);
1775     spapr_register_hypercall(H_INT_SET_OS_REPORTING_LINE,
1776                              h_int_set_os_reporting_line);
1777     spapr_register_hypercall(H_INT_GET_OS_REPORTING_LINE,
1778                              h_int_get_os_reporting_line);
1779     spapr_register_hypercall(H_INT_ESB, h_int_esb);
1780     spapr_register_hypercall(H_INT_SYNC, h_int_sync);
1781     spapr_register_hypercall(H_INT_RESET, h_int_reset);
1782 }
1783