xref: /openbmc/qemu/hw/intc/spapr_xive.c (revision d024a2c1114fadd9b0692be4e594a9b8b31197af)
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 XiveTCTX *spapr_xive_get_tctx(XiveRouter *xrtr, CPUState *cs)
431 {
432     PowerPCCPU *cpu = POWERPC_CPU(cs);
433 
434     return spapr_cpu_state(cpu)->tctx;
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 const VMStateDescription vmstate_spapr_xive_end = {
484     .name = TYPE_SPAPR_XIVE "/end",
485     .version_id = 1,
486     .minimum_version_id = 1,
487     .fields = (VMStateField []) {
488         VMSTATE_UINT32(w0, XiveEND),
489         VMSTATE_UINT32(w1, XiveEND),
490         VMSTATE_UINT32(w2, XiveEND),
491         VMSTATE_UINT32(w3, XiveEND),
492         VMSTATE_UINT32(w4, XiveEND),
493         VMSTATE_UINT32(w5, XiveEND),
494         VMSTATE_UINT32(w6, XiveEND),
495         VMSTATE_UINT32(w7, XiveEND),
496         VMSTATE_END_OF_LIST()
497     },
498 };
499 
500 static const VMStateDescription vmstate_spapr_xive_eas = {
501     .name = TYPE_SPAPR_XIVE "/eas",
502     .version_id = 1,
503     .minimum_version_id = 1,
504     .fields = (VMStateField []) {
505         VMSTATE_UINT64(w, XiveEAS),
506         VMSTATE_END_OF_LIST()
507     },
508 };
509 
510 static int vmstate_spapr_xive_pre_save(void *opaque)
511 {
512     if (kvm_irqchip_in_kernel()) {
513         return kvmppc_xive_pre_save(SPAPR_XIVE(opaque));
514     }
515 
516     return 0;
517 }
518 
519 /*
520  * Called by the sPAPR IRQ backend 'post_load' method at the machine
521  * level.
522  */
523 static int spapr_xive_post_load(SpaprInterruptController *intc, int version_id)
524 {
525     if (kvm_irqchip_in_kernel()) {
526         return kvmppc_xive_post_load(SPAPR_XIVE(intc), version_id);
527     }
528 
529     return 0;
530 }
531 
532 static const VMStateDescription vmstate_spapr_xive = {
533     .name = TYPE_SPAPR_XIVE,
534     .version_id = 1,
535     .minimum_version_id = 1,
536     .pre_save = vmstate_spapr_xive_pre_save,
537     .post_load = NULL, /* handled at the machine level */
538     .fields = (VMStateField[]) {
539         VMSTATE_UINT32_EQUAL(nr_irqs, SpaprXive, NULL),
540         VMSTATE_STRUCT_VARRAY_POINTER_UINT32(eat, SpaprXive, nr_irqs,
541                                      vmstate_spapr_xive_eas, XiveEAS),
542         VMSTATE_STRUCT_VARRAY_POINTER_UINT32(endt, SpaprXive, nr_ends,
543                                              vmstate_spapr_xive_end, XiveEND),
544         VMSTATE_END_OF_LIST()
545     },
546 };
547 
548 static int spapr_xive_claim_irq(SpaprInterruptController *intc, int lisn,
549                                 bool lsi, Error **errp)
550 {
551     SpaprXive *xive = SPAPR_XIVE(intc);
552     XiveSource *xsrc = &xive->source;
553 
554     assert(lisn < xive->nr_irqs);
555 
556     if (xive_eas_is_valid(&xive->eat[lisn])) {
557         error_setg(errp, "IRQ %d is not free", lisn);
558         return -EBUSY;
559     }
560 
561     /*
562      * Set default values when allocating an IRQ number
563      */
564     xive->eat[lisn].w |= cpu_to_be64(EAS_VALID | EAS_MASKED);
565     if (lsi) {
566         xive_source_irq_set_lsi(xsrc, lisn);
567     }
568 
569     if (kvm_irqchip_in_kernel()) {
570         return kvmppc_xive_source_reset_one(xsrc, lisn, errp);
571     }
572 
573     return 0;
574 }
575 
576 static void spapr_xive_free_irq(SpaprInterruptController *intc, int lisn)
577 {
578     SpaprXive *xive = SPAPR_XIVE(intc);
579     assert(lisn < xive->nr_irqs);
580 
581     xive->eat[lisn].w &= cpu_to_be64(~EAS_VALID);
582 }
583 
584 static Property spapr_xive_properties[] = {
585     DEFINE_PROP_UINT32("nr-irqs", SpaprXive, nr_irqs, 0),
586     DEFINE_PROP_UINT32("nr-ends", SpaprXive, nr_ends, 0),
587     DEFINE_PROP_UINT64("vc-base", SpaprXive, vc_base, SPAPR_XIVE_VC_BASE),
588     DEFINE_PROP_UINT64("tm-base", SpaprXive, tm_base, SPAPR_XIVE_TM_BASE),
589     DEFINE_PROP_END_OF_LIST(),
590 };
591 
592 static int spapr_xive_cpu_intc_create(SpaprInterruptController *intc,
593                                       PowerPCCPU *cpu, Error **errp)
594 {
595     SpaprXive *xive = SPAPR_XIVE(intc);
596     Object *obj;
597     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
598 
599     obj = xive_tctx_create(OBJECT(cpu), XIVE_ROUTER(xive), errp);
600     if (!obj) {
601         return -1;
602     }
603 
604     spapr_cpu->tctx = XIVE_TCTX(obj);
605     return 0;
606 }
607 
608 static void xive_tctx_set_os_cam(XiveTCTX *tctx, uint32_t os_cam)
609 {
610     uint32_t qw1w2 = cpu_to_be32(TM_QW1W2_VO | os_cam);
611     memcpy(&tctx->regs[TM_QW1_OS + TM_WORD2], &qw1w2, 4);
612 }
613 
614 static void spapr_xive_cpu_intc_reset(SpaprInterruptController *intc,
615                                      PowerPCCPU *cpu)
616 {
617     XiveTCTX *tctx = spapr_cpu_state(cpu)->tctx;
618     uint8_t  nvt_blk;
619     uint32_t nvt_idx;
620 
621     xive_tctx_reset(tctx);
622 
623     /*
624      * When a Virtual Processor is scheduled to run on a HW thread,
625      * the hypervisor pushes its identifier in the OS CAM line.
626      * Emulate the same behavior under QEMU.
627      */
628     spapr_xive_cpu_to_nvt(cpu, &nvt_blk, &nvt_idx);
629 
630     xive_tctx_set_os_cam(tctx, xive_nvt_cam_line(nvt_blk, nvt_idx));
631 }
632 
633 static void spapr_xive_cpu_intc_destroy(SpaprInterruptController *intc,
634                                         PowerPCCPU *cpu)
635 {
636     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
637 
638     xive_tctx_destroy(spapr_cpu->tctx);
639     spapr_cpu->tctx = NULL;
640 }
641 
642 static void spapr_xive_set_irq(SpaprInterruptController *intc, int irq, int val)
643 {
644     SpaprXive *xive = SPAPR_XIVE(intc);
645 
646     if (kvm_irqchip_in_kernel()) {
647         kvmppc_xive_source_set_irq(&xive->source, irq, val);
648     } else {
649         xive_source_set_irq(&xive->source, irq, val);
650     }
651 }
652 
653 static void spapr_xive_print_info(SpaprInterruptController *intc, Monitor *mon)
654 {
655     SpaprXive *xive = SPAPR_XIVE(intc);
656     CPUState *cs;
657 
658     CPU_FOREACH(cs) {
659         PowerPCCPU *cpu = POWERPC_CPU(cs);
660 
661         xive_tctx_pic_print_info(spapr_cpu_state(cpu)->tctx, mon);
662     }
663 
664     spapr_xive_pic_print_info(xive, mon);
665 }
666 
667 static void spapr_xive_dt(SpaprInterruptController *intc, uint32_t nr_servers,
668                           void *fdt, uint32_t phandle)
669 {
670     SpaprXive *xive = SPAPR_XIVE(intc);
671     int node;
672     uint64_t timas[2 * 2];
673     /* Interrupt number ranges for the IPIs */
674     uint32_t lisn_ranges[] = {
675         cpu_to_be32(0),
676         cpu_to_be32(nr_servers),
677     };
678     /*
679      * EQ size - the sizes of pages supported by the system 4K, 64K,
680      * 2M, 16M. We only advertise 64K for the moment.
681      */
682     uint32_t eq_sizes[] = {
683         cpu_to_be32(16), /* 64K */
684     };
685     /*
686      * The following array is in sync with the reserved priorities
687      * defined by the 'spapr_xive_priority_is_reserved' routine.
688      */
689     uint32_t plat_res_int_priorities[] = {
690         cpu_to_be32(7),    /* start */
691         cpu_to_be32(0xf8), /* count */
692     };
693 
694     /* Thread Interrupt Management Area : User (ring 3) and OS (ring 2) */
695     timas[0] = cpu_to_be64(xive->tm_base +
696                            XIVE_TM_USER_PAGE * (1ull << TM_SHIFT));
697     timas[1] = cpu_to_be64(1ull << TM_SHIFT);
698     timas[2] = cpu_to_be64(xive->tm_base +
699                            XIVE_TM_OS_PAGE * (1ull << TM_SHIFT));
700     timas[3] = cpu_to_be64(1ull << TM_SHIFT);
701 
702     _FDT(node = fdt_add_subnode(fdt, 0, xive->nodename));
703 
704     _FDT(fdt_setprop_string(fdt, node, "device_type", "power-ivpe"));
705     _FDT(fdt_setprop(fdt, node, "reg", timas, sizeof(timas)));
706 
707     _FDT(fdt_setprop_string(fdt, node, "compatible", "ibm,power-ivpe"));
708     _FDT(fdt_setprop(fdt, node, "ibm,xive-eq-sizes", eq_sizes,
709                      sizeof(eq_sizes)));
710     _FDT(fdt_setprop(fdt, node, "ibm,xive-lisn-ranges", lisn_ranges,
711                      sizeof(lisn_ranges)));
712 
713     /* For Linux to link the LSIs to the interrupt controller. */
714     _FDT(fdt_setprop(fdt, node, "interrupt-controller", NULL, 0));
715     _FDT(fdt_setprop_cell(fdt, node, "#interrupt-cells", 2));
716 
717     /* For SLOF */
718     _FDT(fdt_setprop_cell(fdt, node, "linux,phandle", phandle));
719     _FDT(fdt_setprop_cell(fdt, node, "phandle", phandle));
720 
721     /*
722      * The "ibm,plat-res-int-priorities" property defines the priority
723      * ranges reserved by the hypervisor
724      */
725     _FDT(fdt_setprop(fdt, 0, "ibm,plat-res-int-priorities",
726                      plat_res_int_priorities, sizeof(plat_res_int_priorities)));
727 }
728 
729 static int spapr_xive_activate(SpaprInterruptController *intc,
730                                uint32_t nr_servers, Error **errp)
731 {
732     SpaprXive *xive = SPAPR_XIVE(intc);
733 
734     if (kvm_enabled()) {
735         int rc = spapr_irq_init_kvm(kvmppc_xive_connect, intc, nr_servers,
736                                     errp);
737         if (rc < 0) {
738             return rc;
739         }
740     }
741 
742     /* Activate the XIVE MMIOs */
743     spapr_xive_mmio_set_enabled(xive, true);
744 
745     return 0;
746 }
747 
748 static void spapr_xive_deactivate(SpaprInterruptController *intc)
749 {
750     SpaprXive *xive = SPAPR_XIVE(intc);
751 
752     spapr_xive_mmio_set_enabled(xive, false);
753 
754     if (kvm_irqchip_in_kernel()) {
755         kvmppc_xive_disconnect(intc);
756     }
757 }
758 
759 static void spapr_xive_class_init(ObjectClass *klass, void *data)
760 {
761     DeviceClass *dc = DEVICE_CLASS(klass);
762     XiveRouterClass *xrc = XIVE_ROUTER_CLASS(klass);
763     SpaprInterruptControllerClass *sicc = SPAPR_INTC_CLASS(klass);
764     XivePresenterClass *xpc = XIVE_PRESENTER_CLASS(klass);
765 
766     dc->desc    = "sPAPR XIVE Interrupt Controller";
767     dc->props   = spapr_xive_properties;
768     dc->realize = spapr_xive_realize;
769     dc->vmsd    = &vmstate_spapr_xive;
770 
771     xrc->get_eas = spapr_xive_get_eas;
772     xrc->get_end = spapr_xive_get_end;
773     xrc->write_end = spapr_xive_write_end;
774     xrc->get_nvt = spapr_xive_get_nvt;
775     xrc->write_nvt = spapr_xive_write_nvt;
776     xrc->get_tctx = spapr_xive_get_tctx;
777 
778     sicc->activate = spapr_xive_activate;
779     sicc->deactivate = spapr_xive_deactivate;
780     sicc->cpu_intc_create = spapr_xive_cpu_intc_create;
781     sicc->cpu_intc_reset = spapr_xive_cpu_intc_reset;
782     sicc->cpu_intc_destroy = spapr_xive_cpu_intc_destroy;
783     sicc->claim_irq = spapr_xive_claim_irq;
784     sicc->free_irq = spapr_xive_free_irq;
785     sicc->set_irq = spapr_xive_set_irq;
786     sicc->print_info = spapr_xive_print_info;
787     sicc->dt = spapr_xive_dt;
788     sicc->post_load = spapr_xive_post_load;
789 
790     xpc->match_nvt  = spapr_xive_match_nvt;
791 }
792 
793 static const TypeInfo spapr_xive_info = {
794     .name = TYPE_SPAPR_XIVE,
795     .parent = TYPE_XIVE_ROUTER,
796     .instance_init = spapr_xive_instance_init,
797     .instance_size = sizeof(SpaprXive),
798     .class_init = spapr_xive_class_init,
799     .interfaces = (InterfaceInfo[]) {
800         { TYPE_SPAPR_INTC },
801         { }
802     },
803 };
804 
805 static void spapr_xive_register_types(void)
806 {
807     type_register_static(&spapr_xive_info);
808 }
809 
810 type_init(spapr_xive_register_types)
811 
812 /*
813  * XIVE hcalls
814  *
815  * The terminology used by the XIVE hcalls is the following :
816  *
817  *   TARGET vCPU number
818  *   EQ     Event Queue assigned by OS to receive event data
819  *   ESB    page for source interrupt management
820  *   LISN   Logical Interrupt Source Number identifying a source in the
821  *          machine
822  *   EISN   Effective Interrupt Source Number used by guest OS to
823  *          identify source in the guest
824  *
825  * The EAS, END, NVT structures are not exposed.
826  */
827 
828 /*
829  * Linux hosts under OPAL reserve priority 7 for their own escalation
830  * interrupts (DD2.X POWER9). So we only allow the guest to use
831  * priorities [0..6].
832  */
833 static bool spapr_xive_priority_is_reserved(uint8_t priority)
834 {
835     switch (priority) {
836     case 0 ... 6:
837         return false;
838     case 7: /* OPAL escalation queue */
839     default:
840         return true;
841     }
842 }
843 
844 /*
845  * The H_INT_GET_SOURCE_INFO hcall() is used to obtain the logical
846  * real address of the MMIO page through which the Event State Buffer
847  * entry associated with the value of the "lisn" parameter is managed.
848  *
849  * Parameters:
850  * Input
851  * - R4: "flags"
852  *         Bits 0-63 reserved
853  * - R5: "lisn" is per "interrupts", "interrupt-map", or
854  *       "ibm,xive-lisn-ranges" properties, or as returned by the
855  *       ibm,query-interrupt-source-number RTAS call, or as returned
856  *       by the H_ALLOCATE_VAS_WINDOW hcall
857  *
858  * Output
859  * - R4: "flags"
860  *         Bits 0-59: Reserved
861  *         Bit 60: H_INT_ESB must be used for Event State Buffer
862  *                 management
863  *         Bit 61: 1 == LSI  0 == MSI
864  *         Bit 62: the full function page supports trigger
865  *         Bit 63: Store EOI Supported
866  * - R5: Logical Real address of full function Event State Buffer
867  *       management page, -1 if H_INT_ESB hcall flag is set to 1.
868  * - R6: Logical Real Address of trigger only Event State Buffer
869  *       management page or -1.
870  * - R7: Power of 2 page size for the ESB management pages returned in
871  *       R5 and R6.
872  */
873 
874 #define SPAPR_XIVE_SRC_H_INT_ESB     PPC_BIT(60) /* ESB manage with H_INT_ESB */
875 #define SPAPR_XIVE_SRC_LSI           PPC_BIT(61) /* Virtual LSI type */
876 #define SPAPR_XIVE_SRC_TRIGGER       PPC_BIT(62) /* Trigger and management
877                                                     on same page */
878 #define SPAPR_XIVE_SRC_STORE_EOI     PPC_BIT(63) /* Store EOI support */
879 
880 static target_ulong h_int_get_source_info(PowerPCCPU *cpu,
881                                           SpaprMachineState *spapr,
882                                           target_ulong opcode,
883                                           target_ulong *args)
884 {
885     SpaprXive *xive = spapr->xive;
886     XiveSource *xsrc = &xive->source;
887     target_ulong flags  = args[0];
888     target_ulong lisn   = args[1];
889 
890     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
891         return H_FUNCTION;
892     }
893 
894     if (flags) {
895         return H_PARAMETER;
896     }
897 
898     if (lisn >= xive->nr_irqs) {
899         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
900                       lisn);
901         return H_P2;
902     }
903 
904     if (!xive_eas_is_valid(&xive->eat[lisn])) {
905         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
906                       lisn);
907         return H_P2;
908     }
909 
910     /*
911      * All sources are emulated under the main XIVE object and share
912      * the same characteristics.
913      */
914     args[0] = 0;
915     if (!xive_source_esb_has_2page(xsrc)) {
916         args[0] |= SPAPR_XIVE_SRC_TRIGGER;
917     }
918     if (xsrc->esb_flags & XIVE_SRC_STORE_EOI) {
919         args[0] |= SPAPR_XIVE_SRC_STORE_EOI;
920     }
921 
922     /*
923      * Force the use of the H_INT_ESB hcall in case of an LSI
924      * interrupt. This is necessary under KVM to re-trigger the
925      * interrupt if the level is still asserted
926      */
927     if (xive_source_irq_is_lsi(xsrc, lisn)) {
928         args[0] |= SPAPR_XIVE_SRC_H_INT_ESB | SPAPR_XIVE_SRC_LSI;
929     }
930 
931     if (!(args[0] & SPAPR_XIVE_SRC_H_INT_ESB)) {
932         args[1] = xive->vc_base + xive_source_esb_mgmt(xsrc, lisn);
933     } else {
934         args[1] = -1;
935     }
936 
937     if (xive_source_esb_has_2page(xsrc) &&
938         !(args[0] & SPAPR_XIVE_SRC_H_INT_ESB)) {
939         args[2] = xive->vc_base + xive_source_esb_page(xsrc, lisn);
940     } else {
941         args[2] = -1;
942     }
943 
944     if (xive_source_esb_has_2page(xsrc)) {
945         args[3] = xsrc->esb_shift - 1;
946     } else {
947         args[3] = xsrc->esb_shift;
948     }
949 
950     return H_SUCCESS;
951 }
952 
953 /*
954  * The H_INT_SET_SOURCE_CONFIG hcall() is used to assign a Logical
955  * Interrupt Source to a target. The Logical Interrupt Source is
956  * designated with the "lisn" parameter and the target is designated
957  * with the "target" and "priority" parameters.  Upon return from the
958  * hcall(), no additional interrupts will be directed to the old EQ.
959  *
960  * Parameters:
961  * Input:
962  * - R4: "flags"
963  *         Bits 0-61: Reserved
964  *         Bit 62: set the "eisn" in the EAS
965  *         Bit 63: masks the interrupt source in the hardware interrupt
966  *       control structure. An interrupt masked by this mechanism will
967  *       be dropped, but it's source state bits will still be
968  *       set. There is no race-free way of unmasking and restoring the
969  *       source. Thus this should only be used in interrupts that are
970  *       also masked at the source, and only in cases where the
971  *       interrupt is not meant to be used for a large amount of time
972  *       because no valid target exists for it for example
973  * - R5: "lisn" is per "interrupts", "interrupt-map", or
974  *       "ibm,xive-lisn-ranges" properties, or as returned by the
975  *       ibm,query-interrupt-source-number RTAS call, or as returned by
976  *       the H_ALLOCATE_VAS_WINDOW hcall
977  * - R6: "target" is per "ibm,ppc-interrupt-server#s" or
978  *       "ibm,ppc-interrupt-gserver#s"
979  * - R7: "priority" is a valid priority not in
980  *       "ibm,plat-res-int-priorities"
981  * - R8: "eisn" is the guest EISN associated with the "lisn"
982  *
983  * Output:
984  * - None
985  */
986 
987 #define SPAPR_XIVE_SRC_SET_EISN PPC_BIT(62)
988 #define SPAPR_XIVE_SRC_MASK     PPC_BIT(63)
989 
990 static target_ulong h_int_set_source_config(PowerPCCPU *cpu,
991                                             SpaprMachineState *spapr,
992                                             target_ulong opcode,
993                                             target_ulong *args)
994 {
995     SpaprXive *xive = spapr->xive;
996     XiveEAS eas, new_eas;
997     target_ulong flags    = args[0];
998     target_ulong lisn     = args[1];
999     target_ulong target   = args[2];
1000     target_ulong priority = args[3];
1001     target_ulong eisn     = args[4];
1002     uint8_t end_blk;
1003     uint32_t end_idx;
1004 
1005     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1006         return H_FUNCTION;
1007     }
1008 
1009     if (flags & ~(SPAPR_XIVE_SRC_SET_EISN | SPAPR_XIVE_SRC_MASK)) {
1010         return H_PARAMETER;
1011     }
1012 
1013     if (lisn >= xive->nr_irqs) {
1014         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1015                       lisn);
1016         return H_P2;
1017     }
1018 
1019     eas = xive->eat[lisn];
1020     if (!xive_eas_is_valid(&eas)) {
1021         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1022                       lisn);
1023         return H_P2;
1024     }
1025 
1026     /* priority 0xff is used to reset the EAS */
1027     if (priority == 0xff) {
1028         new_eas.w = cpu_to_be64(EAS_VALID | EAS_MASKED);
1029         goto out;
1030     }
1031 
1032     if (flags & SPAPR_XIVE_SRC_MASK) {
1033         new_eas.w = eas.w | cpu_to_be64(EAS_MASKED);
1034     } else {
1035         new_eas.w = eas.w & cpu_to_be64(~EAS_MASKED);
1036     }
1037 
1038     if (spapr_xive_priority_is_reserved(priority)) {
1039         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1040                       " is reserved\n", priority);
1041         return H_P4;
1042     }
1043 
1044     /*
1045      * Validate that "target" is part of the list of threads allocated
1046      * to the partition. For that, find the END corresponding to the
1047      * target.
1048      */
1049     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1050         return H_P3;
1051     }
1052 
1053     new_eas.w = xive_set_field64(EAS_END_BLOCK, new_eas.w, end_blk);
1054     new_eas.w = xive_set_field64(EAS_END_INDEX, new_eas.w, end_idx);
1055 
1056     if (flags & SPAPR_XIVE_SRC_SET_EISN) {
1057         new_eas.w = xive_set_field64(EAS_END_DATA, new_eas.w, eisn);
1058     }
1059 
1060     if (kvm_irqchip_in_kernel()) {
1061         Error *local_err = NULL;
1062 
1063         kvmppc_xive_set_source_config(xive, lisn, &new_eas, &local_err);
1064         if (local_err) {
1065             error_report_err(local_err);
1066             return H_HARDWARE;
1067         }
1068     }
1069 
1070 out:
1071     xive->eat[lisn] = new_eas;
1072     return H_SUCCESS;
1073 }
1074 
1075 /*
1076  * The H_INT_GET_SOURCE_CONFIG hcall() is used to determine to which
1077  * target/priority pair is assigned to the specified Logical Interrupt
1078  * Source.
1079  *
1080  * Parameters:
1081  * Input:
1082  * - R4: "flags"
1083  *         Bits 0-63 Reserved
1084  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1085  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1086  *       ibm,query-interrupt-source-number RTAS call, or as
1087  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1088  *
1089  * Output:
1090  * - R4: Target to which the specified Logical Interrupt Source is
1091  *       assigned
1092  * - R5: Priority to which the specified Logical Interrupt Source is
1093  *       assigned
1094  * - R6: EISN for the specified Logical Interrupt Source (this will be
1095  *       equivalent to the LISN if not changed by H_INT_SET_SOURCE_CONFIG)
1096  */
1097 static target_ulong h_int_get_source_config(PowerPCCPU *cpu,
1098                                             SpaprMachineState *spapr,
1099                                             target_ulong opcode,
1100                                             target_ulong *args)
1101 {
1102     SpaprXive *xive = spapr->xive;
1103     target_ulong flags = args[0];
1104     target_ulong lisn = args[1];
1105     XiveEAS eas;
1106     XiveEND *end;
1107     uint8_t nvt_blk;
1108     uint32_t end_idx, nvt_idx;
1109 
1110     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1111         return H_FUNCTION;
1112     }
1113 
1114     if (flags) {
1115         return H_PARAMETER;
1116     }
1117 
1118     if (lisn >= xive->nr_irqs) {
1119         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1120                       lisn);
1121         return H_P2;
1122     }
1123 
1124     eas = xive->eat[lisn];
1125     if (!xive_eas_is_valid(&eas)) {
1126         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1127                       lisn);
1128         return H_P2;
1129     }
1130 
1131     /* EAS_END_BLOCK is unused on sPAPR */
1132     end_idx = xive_get_field64(EAS_END_INDEX, eas.w);
1133 
1134     assert(end_idx < xive->nr_ends);
1135     end = &xive->endt[end_idx];
1136 
1137     nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end->w6);
1138     nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end->w6);
1139     args[0] = spapr_xive_nvt_to_target(nvt_blk, nvt_idx);
1140 
1141     if (xive_eas_is_masked(&eas)) {
1142         args[1] = 0xff;
1143     } else {
1144         args[1] = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
1145     }
1146 
1147     args[2] = xive_get_field64(EAS_END_DATA, eas.w);
1148 
1149     return H_SUCCESS;
1150 }
1151 
1152 /*
1153  * The H_INT_GET_QUEUE_INFO hcall() is used to get the logical real
1154  * address of the notification management page associated with the
1155  * specified target and priority.
1156  *
1157  * Parameters:
1158  * Input:
1159  * - R4: "flags"
1160  *         Bits 0-63 Reserved
1161  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1162  *       "ibm,ppc-interrupt-gserver#s"
1163  * - R6: "priority" is a valid priority not in
1164  *       "ibm,plat-res-int-priorities"
1165  *
1166  * Output:
1167  * - R4: Logical real address of notification page
1168  * - R5: Power of 2 page size of the notification page
1169  */
1170 static target_ulong h_int_get_queue_info(PowerPCCPU *cpu,
1171                                          SpaprMachineState *spapr,
1172                                          target_ulong opcode,
1173                                          target_ulong *args)
1174 {
1175     SpaprXive *xive = spapr->xive;
1176     XiveENDSource *end_xsrc = &xive->end_source;
1177     target_ulong flags = args[0];
1178     target_ulong target = args[1];
1179     target_ulong priority = args[2];
1180     XiveEND *end;
1181     uint8_t end_blk;
1182     uint32_t end_idx;
1183 
1184     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1185         return H_FUNCTION;
1186     }
1187 
1188     if (flags) {
1189         return H_PARAMETER;
1190     }
1191 
1192     /*
1193      * H_STATE should be returned if a H_INT_RESET is in progress.
1194      * This is not needed when running the emulation under QEMU
1195      */
1196 
1197     if (spapr_xive_priority_is_reserved(priority)) {
1198         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1199                       " is reserved\n", priority);
1200         return H_P3;
1201     }
1202 
1203     /*
1204      * Validate that "target" is part of the list of threads allocated
1205      * to the partition. For that, find the END corresponding to the
1206      * target.
1207      */
1208     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1209         return H_P2;
1210     }
1211 
1212     assert(end_idx < xive->nr_ends);
1213     end = &xive->endt[end_idx];
1214 
1215     args[0] = xive->end_base + (1ull << (end_xsrc->esb_shift + 1)) * end_idx;
1216     if (xive_end_is_enqueue(end)) {
1217         args[1] = xive_get_field32(END_W0_QSIZE, end->w0) + 12;
1218     } else {
1219         args[1] = 0;
1220     }
1221 
1222     return H_SUCCESS;
1223 }
1224 
1225 /*
1226  * The H_INT_SET_QUEUE_CONFIG hcall() is used to set or reset a EQ for
1227  * a given "target" and "priority".  It is also used to set the
1228  * notification config associated with the EQ.  An EQ size of 0 is
1229  * used to reset the EQ config for a given target and priority. If
1230  * resetting the EQ config, the END associated with the given "target"
1231  * and "priority" will be changed to disable queueing.
1232  *
1233  * Upon return from the hcall(), no additional interrupts will be
1234  * directed to the old EQ (if one was set). The old EQ (if one was
1235  * set) should be investigated for interrupts that occurred prior to
1236  * or during the hcall().
1237  *
1238  * Parameters:
1239  * Input:
1240  * - R4: "flags"
1241  *         Bits 0-62: Reserved
1242  *         Bit 63: Unconditional Notify (n) per the XIVE spec
1243  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1244  *       "ibm,ppc-interrupt-gserver#s"
1245  * - R6: "priority" is a valid priority not in
1246  *       "ibm,plat-res-int-priorities"
1247  * - R7: "eventQueue": The logical real address of the start of the EQ
1248  * - R8: "eventQueueSize": The power of 2 EQ size per "ibm,xive-eq-sizes"
1249  *
1250  * Output:
1251  * - None
1252  */
1253 
1254 #define SPAPR_XIVE_END_ALWAYS_NOTIFY PPC_BIT(63)
1255 
1256 static target_ulong h_int_set_queue_config(PowerPCCPU *cpu,
1257                                            SpaprMachineState *spapr,
1258                                            target_ulong opcode,
1259                                            target_ulong *args)
1260 {
1261     SpaprXive *xive = spapr->xive;
1262     target_ulong flags = args[0];
1263     target_ulong target = args[1];
1264     target_ulong priority = args[2];
1265     target_ulong qpage = args[3];
1266     target_ulong qsize = args[4];
1267     XiveEND end;
1268     uint8_t end_blk, nvt_blk;
1269     uint32_t end_idx, nvt_idx;
1270 
1271     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1272         return H_FUNCTION;
1273     }
1274 
1275     if (flags & ~SPAPR_XIVE_END_ALWAYS_NOTIFY) {
1276         return H_PARAMETER;
1277     }
1278 
1279     /*
1280      * H_STATE should be returned if a H_INT_RESET is in progress.
1281      * This is not needed when running the emulation under QEMU
1282      */
1283 
1284     if (spapr_xive_priority_is_reserved(priority)) {
1285         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1286                       " is reserved\n", priority);
1287         return H_P3;
1288     }
1289 
1290     /*
1291      * Validate that "target" is part of the list of threads allocated
1292      * to the partition. For that, find the END corresponding to the
1293      * target.
1294      */
1295 
1296     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1297         return H_P2;
1298     }
1299 
1300     assert(end_idx < xive->nr_ends);
1301     memcpy(&end, &xive->endt[end_idx], sizeof(XiveEND));
1302 
1303     switch (qsize) {
1304     case 12:
1305     case 16:
1306     case 21:
1307     case 24:
1308         if (!QEMU_IS_ALIGNED(qpage, 1ul << qsize)) {
1309             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: EQ @0x%" HWADDR_PRIx
1310                           " is not naturally aligned with %" HWADDR_PRIx "\n",
1311                           qpage, (hwaddr)1 << qsize);
1312             return H_P4;
1313         }
1314         end.w2 = cpu_to_be32((qpage >> 32) & 0x0fffffff);
1315         end.w3 = cpu_to_be32(qpage & 0xffffffff);
1316         end.w0 |= cpu_to_be32(END_W0_ENQUEUE);
1317         end.w0 = xive_set_field32(END_W0_QSIZE, end.w0, qsize - 12);
1318         break;
1319     case 0:
1320         /* reset queue and disable queueing */
1321         spapr_xive_end_reset(&end);
1322         goto out;
1323 
1324     default:
1325         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid EQ size %"PRIx64"\n",
1326                       qsize);
1327         return H_P5;
1328     }
1329 
1330     if (qsize) {
1331         hwaddr plen = 1 << qsize;
1332         void *eq;
1333 
1334         /*
1335          * Validate the guest EQ. We should also check that the queue
1336          * has been zeroed by the OS.
1337          */
1338         eq = address_space_map(CPU(cpu)->as, qpage, &plen, true,
1339                                MEMTXATTRS_UNSPECIFIED);
1340         if (plen != 1 << qsize) {
1341             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to map EQ @0x%"
1342                           HWADDR_PRIx "\n", qpage);
1343             return H_P4;
1344         }
1345         address_space_unmap(CPU(cpu)->as, eq, plen, true, plen);
1346     }
1347 
1348     /* "target" should have been validated above */
1349     if (spapr_xive_target_to_nvt(target, &nvt_blk, &nvt_idx)) {
1350         g_assert_not_reached();
1351     }
1352 
1353     /*
1354      * Ensure the priority and target are correctly set (they will not
1355      * be right after allocation)
1356      */
1357     end.w6 = xive_set_field32(END_W6_NVT_BLOCK, 0ul, nvt_blk) |
1358         xive_set_field32(END_W6_NVT_INDEX, 0ul, nvt_idx);
1359     end.w7 = xive_set_field32(END_W7_F0_PRIORITY, 0ul, priority);
1360 
1361     if (flags & SPAPR_XIVE_END_ALWAYS_NOTIFY) {
1362         end.w0 |= cpu_to_be32(END_W0_UCOND_NOTIFY);
1363     } else {
1364         end.w0 &= cpu_to_be32((uint32_t)~END_W0_UCOND_NOTIFY);
1365     }
1366 
1367     /*
1368      * The generation bit for the END starts at 1 and The END page
1369      * offset counter starts at 0.
1370      */
1371     end.w1 = cpu_to_be32(END_W1_GENERATION) |
1372         xive_set_field32(END_W1_PAGE_OFF, 0ul, 0ul);
1373     end.w0 |= cpu_to_be32(END_W0_VALID);
1374 
1375     /*
1376      * TODO: issue syncs required to ensure all in-flight interrupts
1377      * are complete on the old END
1378      */
1379 
1380 out:
1381     if (kvm_irqchip_in_kernel()) {
1382         Error *local_err = NULL;
1383 
1384         kvmppc_xive_set_queue_config(xive, end_blk, end_idx, &end, &local_err);
1385         if (local_err) {
1386             error_report_err(local_err);
1387             return H_HARDWARE;
1388         }
1389     }
1390 
1391     /* Update END */
1392     memcpy(&xive->endt[end_idx], &end, sizeof(XiveEND));
1393     return H_SUCCESS;
1394 }
1395 
1396 /*
1397  * The H_INT_GET_QUEUE_CONFIG hcall() is used to get a EQ for a given
1398  * target and priority.
1399  *
1400  * Parameters:
1401  * Input:
1402  * - R4: "flags"
1403  *         Bits 0-62: Reserved
1404  *         Bit 63: Debug: Return debug data
1405  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1406  *       "ibm,ppc-interrupt-gserver#s"
1407  * - R6: "priority" is a valid priority not in
1408  *       "ibm,plat-res-int-priorities"
1409  *
1410  * Output:
1411  * - R4: "flags":
1412  *       Bits 0-61: Reserved
1413  *       Bit 62: The value of Event Queue Generation Number (g) per
1414  *              the XIVE spec if "Debug" = 1
1415  *       Bit 63: The value of Unconditional Notify (n) per the XIVE spec
1416  * - R5: The logical real address of the start of the EQ
1417  * - R6: The power of 2 EQ size per "ibm,xive-eq-sizes"
1418  * - R7: The value of Event Queue Offset Counter per XIVE spec
1419  *       if "Debug" = 1, else 0
1420  *
1421  */
1422 
1423 #define SPAPR_XIVE_END_DEBUG     PPC_BIT(63)
1424 
1425 static target_ulong h_int_get_queue_config(PowerPCCPU *cpu,
1426                                            SpaprMachineState *spapr,
1427                                            target_ulong opcode,
1428                                            target_ulong *args)
1429 {
1430     SpaprXive *xive = spapr->xive;
1431     target_ulong flags = args[0];
1432     target_ulong target = args[1];
1433     target_ulong priority = args[2];
1434     XiveEND *end;
1435     uint8_t end_blk;
1436     uint32_t end_idx;
1437 
1438     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1439         return H_FUNCTION;
1440     }
1441 
1442     if (flags & ~SPAPR_XIVE_END_DEBUG) {
1443         return H_PARAMETER;
1444     }
1445 
1446     /*
1447      * H_STATE should be returned if a H_INT_RESET is in progress.
1448      * This is not needed when running the emulation under QEMU
1449      */
1450 
1451     if (spapr_xive_priority_is_reserved(priority)) {
1452         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1453                       " is reserved\n", priority);
1454         return H_P3;
1455     }
1456 
1457     /*
1458      * Validate that "target" is part of the list of threads allocated
1459      * to the partition. For that, find the END corresponding to the
1460      * target.
1461      */
1462     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1463         return H_P2;
1464     }
1465 
1466     assert(end_idx < xive->nr_ends);
1467     end = &xive->endt[end_idx];
1468 
1469     args[0] = 0;
1470     if (xive_end_is_notify(end)) {
1471         args[0] |= SPAPR_XIVE_END_ALWAYS_NOTIFY;
1472     }
1473 
1474     if (xive_end_is_enqueue(end)) {
1475         args[1] = xive_end_qaddr(end);
1476         args[2] = xive_get_field32(END_W0_QSIZE, end->w0) + 12;
1477     } else {
1478         args[1] = 0;
1479         args[2] = 0;
1480     }
1481 
1482     if (kvm_irqchip_in_kernel()) {
1483         Error *local_err = NULL;
1484 
1485         kvmppc_xive_get_queue_config(xive, end_blk, end_idx, end, &local_err);
1486         if (local_err) {
1487             error_report_err(local_err);
1488             return H_HARDWARE;
1489         }
1490     }
1491 
1492     /* TODO: do we need any locking on the END ? */
1493     if (flags & SPAPR_XIVE_END_DEBUG) {
1494         /* Load the event queue generation number into the return flags */
1495         args[0] |= (uint64_t)xive_get_field32(END_W1_GENERATION, end->w1) << 62;
1496 
1497         /* Load R7 with the event queue offset counter */
1498         args[3] = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1499     } else {
1500         args[3] = 0;
1501     }
1502 
1503     return H_SUCCESS;
1504 }
1505 
1506 /*
1507  * The H_INT_SET_OS_REPORTING_LINE hcall() is used to set the
1508  * reporting cache line pair for the calling thread.  The reporting
1509  * cache lines will contain the OS interrupt context when the OS
1510  * issues a CI store byte to @TIMA+0xC10 to acknowledge the OS
1511  * interrupt. The reporting cache lines can be reset by inputting -1
1512  * in "reportingLine".  Issuing the CI store byte without reporting
1513  * cache lines registered will result in the data not being accessible
1514  * to the OS.
1515  *
1516  * Parameters:
1517  * Input:
1518  * - R4: "flags"
1519  *         Bits 0-63: Reserved
1520  * - R5: "reportingLine": The logical real address of the reporting cache
1521  *       line pair
1522  *
1523  * Output:
1524  * - None
1525  */
1526 static target_ulong h_int_set_os_reporting_line(PowerPCCPU *cpu,
1527                                                 SpaprMachineState *spapr,
1528                                                 target_ulong opcode,
1529                                                 target_ulong *args)
1530 {
1531     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1532         return H_FUNCTION;
1533     }
1534 
1535     /*
1536      * H_STATE should be returned if a H_INT_RESET is in progress.
1537      * This is not needed when running the emulation under QEMU
1538      */
1539 
1540     /* TODO: H_INT_SET_OS_REPORTING_LINE */
1541     return H_FUNCTION;
1542 }
1543 
1544 /*
1545  * The H_INT_GET_OS_REPORTING_LINE hcall() is used to get the logical
1546  * real address of the reporting cache line pair set for the input
1547  * "target".  If no reporting cache line pair has been set, -1 is
1548  * returned.
1549  *
1550  * Parameters:
1551  * Input:
1552  * - R4: "flags"
1553  *         Bits 0-63: Reserved
1554  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1555  *       "ibm,ppc-interrupt-gserver#s"
1556  * - R6: "reportingLine": The logical real address of the reporting
1557  *        cache line pair
1558  *
1559  * Output:
1560  * - R4: The logical real address of the reporting line if set, else -1
1561  */
1562 static target_ulong h_int_get_os_reporting_line(PowerPCCPU *cpu,
1563                                                 SpaprMachineState *spapr,
1564                                                 target_ulong opcode,
1565                                                 target_ulong *args)
1566 {
1567     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1568         return H_FUNCTION;
1569     }
1570 
1571     /*
1572      * H_STATE should be returned if a H_INT_RESET is in progress.
1573      * This is not needed when running the emulation under QEMU
1574      */
1575 
1576     /* TODO: H_INT_GET_OS_REPORTING_LINE */
1577     return H_FUNCTION;
1578 }
1579 
1580 /*
1581  * The H_INT_ESB hcall() is used to issue a load or store to the ESB
1582  * page for the input "lisn".  This hcall is only supported for LISNs
1583  * that have the ESB hcall flag set to 1 when returned from hcall()
1584  * H_INT_GET_SOURCE_INFO.
1585  *
1586  * Parameters:
1587  * Input:
1588  * - R4: "flags"
1589  *         Bits 0-62: Reserved
1590  *         bit 63: Store: Store=1, store operation, else load operation
1591  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1592  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1593  *       ibm,query-interrupt-source-number RTAS call, or as
1594  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1595  * - R6: "esbOffset" is the offset into the ESB page for the load or
1596  *       store operation
1597  * - R7: "storeData" is the data to write for a store operation
1598  *
1599  * Output:
1600  * - R4: The value of the load if load operation, else -1
1601  */
1602 
1603 #define SPAPR_XIVE_ESB_STORE PPC_BIT(63)
1604 
1605 static target_ulong h_int_esb(PowerPCCPU *cpu,
1606                               SpaprMachineState *spapr,
1607                               target_ulong opcode,
1608                               target_ulong *args)
1609 {
1610     SpaprXive *xive = spapr->xive;
1611     XiveEAS eas;
1612     target_ulong flags  = args[0];
1613     target_ulong lisn   = args[1];
1614     target_ulong offset = args[2];
1615     target_ulong data   = args[3];
1616     hwaddr mmio_addr;
1617     XiveSource *xsrc = &xive->source;
1618 
1619     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1620         return H_FUNCTION;
1621     }
1622 
1623     if (flags & ~SPAPR_XIVE_ESB_STORE) {
1624         return H_PARAMETER;
1625     }
1626 
1627     if (lisn >= xive->nr_irqs) {
1628         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1629                       lisn);
1630         return H_P2;
1631     }
1632 
1633     eas = xive->eat[lisn];
1634     if (!xive_eas_is_valid(&eas)) {
1635         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1636                       lisn);
1637         return H_P2;
1638     }
1639 
1640     if (offset > (1ull << xsrc->esb_shift)) {
1641         return H_P3;
1642     }
1643 
1644     if (kvm_irqchip_in_kernel()) {
1645         args[0] = kvmppc_xive_esb_rw(xsrc, lisn, offset, data,
1646                                      flags & SPAPR_XIVE_ESB_STORE);
1647     } else {
1648         mmio_addr = xive->vc_base + xive_source_esb_mgmt(xsrc, lisn) + offset;
1649 
1650         if (dma_memory_rw(&address_space_memory, mmio_addr, &data, 8,
1651                           (flags & SPAPR_XIVE_ESB_STORE))) {
1652             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to access ESB @0x%"
1653                           HWADDR_PRIx "\n", mmio_addr);
1654             return H_HARDWARE;
1655         }
1656         args[0] = (flags & SPAPR_XIVE_ESB_STORE) ? -1 : data;
1657     }
1658     return H_SUCCESS;
1659 }
1660 
1661 /*
1662  * The H_INT_SYNC hcall() is used to issue hardware syncs that will
1663  * ensure any in flight events for the input lisn are in the event
1664  * queue.
1665  *
1666  * Parameters:
1667  * Input:
1668  * - R4: "flags"
1669  *         Bits 0-63: Reserved
1670  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1671  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1672  *       ibm,query-interrupt-source-number RTAS call, or as
1673  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1674  *
1675  * Output:
1676  * - None
1677  */
1678 static target_ulong h_int_sync(PowerPCCPU *cpu,
1679                                SpaprMachineState *spapr,
1680                                target_ulong opcode,
1681                                target_ulong *args)
1682 {
1683     SpaprXive *xive = spapr->xive;
1684     XiveEAS eas;
1685     target_ulong flags = args[0];
1686     target_ulong lisn = args[1];
1687 
1688     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1689         return H_FUNCTION;
1690     }
1691 
1692     if (flags) {
1693         return H_PARAMETER;
1694     }
1695 
1696     if (lisn >= xive->nr_irqs) {
1697         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1698                       lisn);
1699         return H_P2;
1700     }
1701 
1702     eas = xive->eat[lisn];
1703     if (!xive_eas_is_valid(&eas)) {
1704         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1705                       lisn);
1706         return H_P2;
1707     }
1708 
1709     /*
1710      * H_STATE should be returned if a H_INT_RESET is in progress.
1711      * This is not needed when running the emulation under QEMU
1712      */
1713 
1714     /*
1715      * This is not real hardware. Nothing to be done unless when
1716      * under KVM
1717      */
1718 
1719     if (kvm_irqchip_in_kernel()) {
1720         Error *local_err = NULL;
1721 
1722         kvmppc_xive_sync_source(xive, lisn, &local_err);
1723         if (local_err) {
1724             error_report_err(local_err);
1725             return H_HARDWARE;
1726         }
1727     }
1728     return H_SUCCESS;
1729 }
1730 
1731 /*
1732  * The H_INT_RESET hcall() is used to reset all of the partition's
1733  * interrupt exploitation structures to their initial state.  This
1734  * means losing all previously set interrupt state set via
1735  * H_INT_SET_SOURCE_CONFIG and H_INT_SET_QUEUE_CONFIG.
1736  *
1737  * Parameters:
1738  * Input:
1739  * - R4: "flags"
1740  *         Bits 0-63: Reserved
1741  *
1742  * Output:
1743  * - None
1744  */
1745 static target_ulong h_int_reset(PowerPCCPU *cpu,
1746                                 SpaprMachineState *spapr,
1747                                 target_ulong opcode,
1748                                 target_ulong *args)
1749 {
1750     SpaprXive *xive = spapr->xive;
1751     target_ulong flags   = args[0];
1752 
1753     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1754         return H_FUNCTION;
1755     }
1756 
1757     if (flags) {
1758         return H_PARAMETER;
1759     }
1760 
1761     device_reset(DEVICE(xive));
1762 
1763     if (kvm_irqchip_in_kernel()) {
1764         Error *local_err = NULL;
1765 
1766         kvmppc_xive_reset(xive, &local_err);
1767         if (local_err) {
1768             error_report_err(local_err);
1769             return H_HARDWARE;
1770         }
1771     }
1772     return H_SUCCESS;
1773 }
1774 
1775 void spapr_xive_hcall_init(SpaprMachineState *spapr)
1776 {
1777     spapr_register_hypercall(H_INT_GET_SOURCE_INFO, h_int_get_source_info);
1778     spapr_register_hypercall(H_INT_SET_SOURCE_CONFIG, h_int_set_source_config);
1779     spapr_register_hypercall(H_INT_GET_SOURCE_CONFIG, h_int_get_source_config);
1780     spapr_register_hypercall(H_INT_GET_QUEUE_INFO, h_int_get_queue_info);
1781     spapr_register_hypercall(H_INT_SET_QUEUE_CONFIG, h_int_set_queue_config);
1782     spapr_register_hypercall(H_INT_GET_QUEUE_CONFIG, h_int_get_queue_config);
1783     spapr_register_hypercall(H_INT_SET_OS_REPORTING_LINE,
1784                              h_int_set_os_reporting_line);
1785     spapr_register_hypercall(H_INT_GET_OS_REPORTING_LINE,
1786                              h_int_get_os_reporting_line);
1787     spapr_register_hypercall(H_INT_ESB, h_int_esb);
1788     spapr_register_hypercall(H_INT_SYNC, h_int_sync);
1789     spapr_register_hypercall(H_INT_RESET, h_int_reset);
1790 }
1791