xref: /openbmc/qemu/hw/intc/spapr_xive.c (revision 328d8eb24db8ec415260ee7243adf2e3d7e81bad)
1 /*
2  * QEMU PowerPC sPAPR XIVE interrupt controller model
3  *
4  * Copyright (c) 2017-2018, IBM Corporation.
5  *
6  * This code is licensed under the GPL version 2 or later. See the
7  * COPYING file in the top-level directory.
8  */
9 
10 #include "qemu/osdep.h"
11 #include "qemu/log.h"
12 #include "qemu/module.h"
13 #include "qapi/error.h"
14 #include "qemu/error-report.h"
15 #include "target/ppc/cpu.h"
16 #include "sysemu/cpus.h"
17 #include "sysemu/reset.h"
18 #include "migration/vmstate.h"
19 #include "monitor/monitor.h"
20 #include "hw/ppc/fdt.h"
21 #include "hw/ppc/spapr.h"
22 #include "hw/ppc/spapr_cpu_core.h"
23 #include "hw/ppc/spapr_xive.h"
24 #include "hw/ppc/xive.h"
25 #include "hw/ppc/xive_regs.h"
26 #include "hw/qdev-properties.h"
27 
28 /*
29  * XIVE Virtualization Controller BAR and Thread Managment BAR that we
30  * use for the ESB pages and the TIMA pages
31  */
32 #define SPAPR_XIVE_VC_BASE   0x0006010000000000ull
33 #define SPAPR_XIVE_TM_BASE   0x0006030203180000ull
34 
35 /*
36  * The allocation of VP blocks is a complex operation in OPAL and the
37  * VP identifiers have a relation with the number of HW chips, the
38  * size of the VP blocks, VP grouping, etc. The QEMU sPAPR XIVE
39  * controller model does not have the same constraints and can use a
40  * simple mapping scheme of the CPU vcpu_id
41  *
42  * These identifiers are never returned to the OS.
43  */
44 
45 #define SPAPR_XIVE_NVT_BASE 0x400
46 
47 /*
48  * sPAPR NVT and END indexing helpers
49  */
50 static uint32_t spapr_xive_nvt_to_target(uint8_t nvt_blk, uint32_t nvt_idx)
51 {
52     return nvt_idx - SPAPR_XIVE_NVT_BASE;
53 }
54 
55 static void spapr_xive_cpu_to_nvt(PowerPCCPU *cpu,
56                                   uint8_t *out_nvt_blk, uint32_t *out_nvt_idx)
57 {
58     assert(cpu);
59 
60     if (out_nvt_blk) {
61         *out_nvt_blk = SPAPR_XIVE_BLOCK_ID;
62     }
63 
64     if (out_nvt_blk) {
65         *out_nvt_idx = SPAPR_XIVE_NVT_BASE + cpu->vcpu_id;
66     }
67 }
68 
69 static int spapr_xive_target_to_nvt(uint32_t target,
70                                     uint8_t *out_nvt_blk, uint32_t *out_nvt_idx)
71 {
72     PowerPCCPU *cpu = spapr_find_cpu(target);
73 
74     if (!cpu) {
75         return -1;
76     }
77 
78     spapr_xive_cpu_to_nvt(cpu, out_nvt_blk, out_nvt_idx);
79     return 0;
80 }
81 
82 /*
83  * sPAPR END indexing uses a simple mapping of the CPU vcpu_id, 8
84  * priorities per CPU
85  */
86 int spapr_xive_end_to_target(uint8_t end_blk, uint32_t end_idx,
87                              uint32_t *out_server, uint8_t *out_prio)
88 {
89 
90     assert(end_blk == SPAPR_XIVE_BLOCK_ID);
91 
92     if (out_server) {
93         *out_server = end_idx >> 3;
94     }
95 
96     if (out_prio) {
97         *out_prio = end_idx & 0x7;
98     }
99     return 0;
100 }
101 
102 static void spapr_xive_cpu_to_end(PowerPCCPU *cpu, uint8_t prio,
103                                   uint8_t *out_end_blk, uint32_t *out_end_idx)
104 {
105     assert(cpu);
106 
107     if (out_end_blk) {
108         *out_end_blk = SPAPR_XIVE_BLOCK_ID;
109     }
110 
111     if (out_end_idx) {
112         *out_end_idx = (cpu->vcpu_id << 3) + prio;
113     }
114 }
115 
116 static int spapr_xive_target_to_end(uint32_t target, uint8_t prio,
117                                     uint8_t *out_end_blk, uint32_t *out_end_idx)
118 {
119     PowerPCCPU *cpu = spapr_find_cpu(target);
120 
121     if (!cpu) {
122         return -1;
123     }
124 
125     spapr_xive_cpu_to_end(cpu, prio, out_end_blk, out_end_idx);
126     return 0;
127 }
128 
129 /*
130  * On sPAPR machines, use a simplified output for the XIVE END
131  * structure dumping only the information related to the OS EQ.
132  */
133 static void spapr_xive_end_pic_print_info(SpaprXive *xive, XiveEND *end,
134                                           Monitor *mon)
135 {
136     uint64_t qaddr_base = xive_end_qaddr(end);
137     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
138     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
139     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
140     uint32_t qentries = 1 << (qsize + 10);
141     uint32_t nvt = xive_get_field32(END_W6_NVT_INDEX, end->w6);
142     uint8_t priority = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
143 
144     monitor_printf(mon, "%3d/%d % 6d/%5d @%"PRIx64" ^%d",
145                    spapr_xive_nvt_to_target(0, nvt),
146                    priority, qindex, qentries, qaddr_base, qgen);
147 
148     xive_end_queue_pic_print_info(end, 6, mon);
149 }
150 
151 void spapr_xive_pic_print_info(SpaprXive *xive, Monitor *mon)
152 {
153     XiveSource *xsrc = &xive->source;
154     int i;
155 
156     if (kvm_irqchip_in_kernel()) {
157         Error *local_err = NULL;
158 
159         kvmppc_xive_synchronize_state(xive, &local_err);
160         if (local_err) {
161             error_report_err(local_err);
162             return;
163         }
164     }
165 
166     monitor_printf(mon, "  LISN         PQ    EISN     CPU/PRIO EQ\n");
167 
168     for (i = 0; i < xive->nr_irqs; i++) {
169         uint8_t pq = xive_source_esb_get(xsrc, i);
170         XiveEAS *eas = &xive->eat[i];
171 
172         if (!xive_eas_is_valid(eas)) {
173             continue;
174         }
175 
176         monitor_printf(mon, "  %08x %s %c%c%c %s %08x ", i,
177                        xive_source_irq_is_lsi(xsrc, i) ? "LSI" : "MSI",
178                        pq & XIVE_ESB_VAL_P ? 'P' : '-',
179                        pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
180                        xsrc->status[i] & XIVE_STATUS_ASSERTED ? 'A' : ' ',
181                        xive_eas_is_masked(eas) ? "M" : " ",
182                        (int) xive_get_field64(EAS_END_DATA, eas->w));
183 
184         if (!xive_eas_is_masked(eas)) {
185             uint32_t end_idx = xive_get_field64(EAS_END_INDEX, eas->w);
186             XiveEND *end;
187 
188             assert(end_idx < xive->nr_ends);
189             end = &xive->endt[end_idx];
190 
191             if (xive_end_is_valid(end)) {
192                 spapr_xive_end_pic_print_info(xive, end, mon);
193             }
194         }
195         monitor_printf(mon, "\n");
196     }
197 }
198 
199 void spapr_xive_mmio_set_enabled(SpaprXive *xive, bool enable)
200 {
201     memory_region_set_enabled(&xive->source.esb_mmio, enable);
202     memory_region_set_enabled(&xive->tm_mmio, enable);
203 
204     /* Disable the END ESBs until a guest OS makes use of them */
205     memory_region_set_enabled(&xive->end_source.esb_mmio, false);
206 }
207 
208 /*
209  * When a Virtual Processor is scheduled to run on a HW thread, the
210  * hypervisor pushes its identifier in the OS CAM line. Emulate the
211  * same behavior under QEMU.
212  */
213 void spapr_xive_set_tctx_os_cam(XiveTCTX *tctx)
214 {
215     uint8_t  nvt_blk;
216     uint32_t nvt_idx;
217     uint32_t nvt_cam;
218 
219     spapr_xive_cpu_to_nvt(POWERPC_CPU(tctx->cs), &nvt_blk, &nvt_idx);
220 
221     nvt_cam = cpu_to_be32(TM_QW1W2_VO | xive_nvt_cam_line(nvt_blk, nvt_idx));
222     memcpy(&tctx->regs[TM_QW1_OS + TM_WORD2], &nvt_cam, 4);
223 }
224 
225 static void spapr_xive_end_reset(XiveEND *end)
226 {
227     memset(end, 0, sizeof(*end));
228 
229     /* switch off the escalation and notification ESBs */
230     end->w1 = cpu_to_be32(END_W1_ESe_Q | END_W1_ESn_Q);
231 }
232 
233 static void spapr_xive_reset(void *dev)
234 {
235     SpaprXive *xive = SPAPR_XIVE(dev);
236     int i;
237 
238     /*
239      * The XiveSource has its own reset handler, which mask off all
240      * IRQs (!P|Q)
241      */
242 
243     /* Mask all valid EASs in the IRQ number space. */
244     for (i = 0; i < xive->nr_irqs; i++) {
245         XiveEAS *eas = &xive->eat[i];
246         if (xive_eas_is_valid(eas)) {
247             eas->w = cpu_to_be64(EAS_VALID | EAS_MASKED);
248         } else {
249             eas->w = 0;
250         }
251     }
252 
253     /* Clear all ENDs */
254     for (i = 0; i < xive->nr_ends; i++) {
255         spapr_xive_end_reset(&xive->endt[i]);
256     }
257 }
258 
259 static void spapr_xive_instance_init(Object *obj)
260 {
261     SpaprXive *xive = SPAPR_XIVE(obj);
262 
263     object_initialize_child(obj, "source", &xive->source, sizeof(xive->source),
264                             TYPE_XIVE_SOURCE, &error_abort, NULL);
265 
266     object_initialize_child(obj, "end_source", &xive->end_source,
267                             sizeof(xive->end_source), TYPE_XIVE_END_SOURCE,
268                             &error_abort, NULL);
269 
270     /* Not connected to the KVM XIVE device */
271     xive->fd = -1;
272 }
273 
274 static void spapr_xive_realize(DeviceState *dev, Error **errp)
275 {
276     SpaprXive *xive = SPAPR_XIVE(dev);
277     XiveSource *xsrc = &xive->source;
278     XiveENDSource *end_xsrc = &xive->end_source;
279     Error *local_err = NULL;
280 
281     if (!xive->nr_irqs) {
282         error_setg(errp, "Number of interrupt needs to be greater 0");
283         return;
284     }
285 
286     if (!xive->nr_ends) {
287         error_setg(errp, "Number of interrupt needs to be greater 0");
288         return;
289     }
290 
291     /*
292      * Initialize the internal sources, for IPIs and virtual devices.
293      */
294     object_property_set_int(OBJECT(xsrc), xive->nr_irqs, "nr-irqs",
295                             &error_fatal);
296     object_property_add_const_link(OBJECT(xsrc), "xive", OBJECT(xive),
297                                    &error_fatal);
298     object_property_set_bool(OBJECT(xsrc), true, "realized", &local_err);
299     if (local_err) {
300         error_propagate(errp, local_err);
301         return;
302     }
303     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &xsrc->esb_mmio);
304 
305     /*
306      * Initialize the END ESB source
307      */
308     object_property_set_int(OBJECT(end_xsrc), xive->nr_irqs, "nr-ends",
309                             &error_fatal);
310     object_property_add_const_link(OBJECT(end_xsrc), "xive", OBJECT(xive),
311                                    &error_fatal);
312     object_property_set_bool(OBJECT(end_xsrc), true, "realized", &local_err);
313     if (local_err) {
314         error_propagate(errp, local_err);
315         return;
316     }
317     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &end_xsrc->esb_mmio);
318 
319     /* Set the mapping address of the END ESB pages after the source ESBs */
320     xive->end_base = xive->vc_base + (1ull << xsrc->esb_shift) * xsrc->nr_irqs;
321 
322     /*
323      * Allocate the routing tables
324      */
325     xive->eat = g_new0(XiveEAS, xive->nr_irqs);
326     xive->endt = g_new0(XiveEND, xive->nr_ends);
327 
328     xive->nodename = g_strdup_printf("interrupt-controller@%" PRIx64,
329                            xive->tm_base + XIVE_TM_USER_PAGE * (1 << TM_SHIFT));
330 
331     qemu_register_reset(spapr_xive_reset, dev);
332 
333     /* TIMA initialization */
334     memory_region_init_io(&xive->tm_mmio, OBJECT(xive), &xive_tm_ops, xive,
335                           "xive.tima", 4ull << TM_SHIFT);
336     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &xive->tm_mmio);
337 
338     /*
339      * Map all regions. These will be enabled or disabled at reset and
340      * can also be overridden by KVM memory regions if active
341      */
342     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 0, xive->vc_base);
343     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 1, xive->end_base);
344     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 2, xive->tm_base);
345 }
346 
347 static int spapr_xive_get_eas(XiveRouter *xrtr, uint8_t eas_blk,
348                               uint32_t eas_idx, XiveEAS *eas)
349 {
350     SpaprXive *xive = SPAPR_XIVE(xrtr);
351 
352     if (eas_idx >= xive->nr_irqs) {
353         return -1;
354     }
355 
356     *eas = xive->eat[eas_idx];
357     return 0;
358 }
359 
360 static int spapr_xive_get_end(XiveRouter *xrtr,
361                               uint8_t end_blk, uint32_t end_idx, XiveEND *end)
362 {
363     SpaprXive *xive = SPAPR_XIVE(xrtr);
364 
365     if (end_idx >= xive->nr_ends) {
366         return -1;
367     }
368 
369     memcpy(end, &xive->endt[end_idx], sizeof(XiveEND));
370     return 0;
371 }
372 
373 static int spapr_xive_write_end(XiveRouter *xrtr, uint8_t end_blk,
374                                 uint32_t end_idx, XiveEND *end,
375                                 uint8_t word_number)
376 {
377     SpaprXive *xive = SPAPR_XIVE(xrtr);
378 
379     if (end_idx >= xive->nr_ends) {
380         return -1;
381     }
382 
383     memcpy(&xive->endt[end_idx], end, sizeof(XiveEND));
384     return 0;
385 }
386 
387 static int spapr_xive_get_nvt(XiveRouter *xrtr,
388                               uint8_t nvt_blk, uint32_t nvt_idx, XiveNVT *nvt)
389 {
390     uint32_t vcpu_id = spapr_xive_nvt_to_target(nvt_blk, nvt_idx);
391     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
392 
393     if (!cpu) {
394         /* TODO: should we assert() if we can find a NVT ? */
395         return -1;
396     }
397 
398     /*
399      * sPAPR does not maintain a NVT table. Return that the NVT is
400      * valid if we have found a matching CPU
401      */
402     nvt->w0 = cpu_to_be32(NVT_W0_VALID);
403     return 0;
404 }
405 
406 static int spapr_xive_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk,
407                                 uint32_t nvt_idx, XiveNVT *nvt,
408                                 uint8_t word_number)
409 {
410     /*
411      * We don't need to write back to the NVTs because the sPAPR
412      * machine should never hit a non-scheduled NVT. It should never
413      * get called.
414      */
415     g_assert_not_reached();
416 }
417 
418 static XiveTCTX *spapr_xive_get_tctx(XiveRouter *xrtr, CPUState *cs)
419 {
420     PowerPCCPU *cpu = POWERPC_CPU(cs);
421 
422     return spapr_cpu_state(cpu)->tctx;
423 }
424 
425 static const VMStateDescription vmstate_spapr_xive_end = {
426     .name = TYPE_SPAPR_XIVE "/end",
427     .version_id = 1,
428     .minimum_version_id = 1,
429     .fields = (VMStateField []) {
430         VMSTATE_UINT32(w0, XiveEND),
431         VMSTATE_UINT32(w1, XiveEND),
432         VMSTATE_UINT32(w2, XiveEND),
433         VMSTATE_UINT32(w3, XiveEND),
434         VMSTATE_UINT32(w4, XiveEND),
435         VMSTATE_UINT32(w5, XiveEND),
436         VMSTATE_UINT32(w6, XiveEND),
437         VMSTATE_UINT32(w7, XiveEND),
438         VMSTATE_END_OF_LIST()
439     },
440 };
441 
442 static const VMStateDescription vmstate_spapr_xive_eas = {
443     .name = TYPE_SPAPR_XIVE "/eas",
444     .version_id = 1,
445     .minimum_version_id = 1,
446     .fields = (VMStateField []) {
447         VMSTATE_UINT64(w, XiveEAS),
448         VMSTATE_END_OF_LIST()
449     },
450 };
451 
452 static int vmstate_spapr_xive_pre_save(void *opaque)
453 {
454     if (kvm_irqchip_in_kernel()) {
455         return kvmppc_xive_pre_save(SPAPR_XIVE(opaque));
456     }
457 
458     return 0;
459 }
460 
461 /*
462  * Called by the sPAPR IRQ backend 'post_load' method at the machine
463  * level.
464  */
465 int spapr_xive_post_load(SpaprXive *xive, int version_id)
466 {
467     if (kvm_irqchip_in_kernel()) {
468         return kvmppc_xive_post_load(xive, version_id);
469     }
470 
471     return 0;
472 }
473 
474 static const VMStateDescription vmstate_spapr_xive = {
475     .name = TYPE_SPAPR_XIVE,
476     .version_id = 1,
477     .minimum_version_id = 1,
478     .pre_save = vmstate_spapr_xive_pre_save,
479     .post_load = NULL, /* handled at the machine level */
480     .fields = (VMStateField[]) {
481         VMSTATE_UINT32_EQUAL(nr_irqs, SpaprXive, NULL),
482         VMSTATE_STRUCT_VARRAY_POINTER_UINT32(eat, SpaprXive, nr_irqs,
483                                      vmstate_spapr_xive_eas, XiveEAS),
484         VMSTATE_STRUCT_VARRAY_POINTER_UINT32(endt, SpaprXive, nr_ends,
485                                              vmstate_spapr_xive_end, XiveEND),
486         VMSTATE_END_OF_LIST()
487     },
488 };
489 
490 static int spapr_xive_claim_irq(SpaprInterruptController *intc, int lisn,
491                                 bool lsi, Error **errp)
492 {
493     SpaprXive *xive = SPAPR_XIVE(intc);
494     XiveSource *xsrc = &xive->source;
495 
496     assert(lisn < xive->nr_irqs);
497 
498     if (xive_eas_is_valid(&xive->eat[lisn])) {
499         error_setg(errp, "IRQ %d is not free", lisn);
500         return -EBUSY;
501     }
502 
503     /*
504      * Set default values when allocating an IRQ number
505      */
506     xive->eat[lisn].w |= cpu_to_be64(EAS_VALID | EAS_MASKED);
507     if (lsi) {
508         xive_source_irq_set_lsi(xsrc, lisn);
509     }
510 
511     if (kvm_irqchip_in_kernel()) {
512         return kvmppc_xive_source_reset_one(xsrc, lisn, errp);
513     }
514 
515     return 0;
516 }
517 
518 static void spapr_xive_free_irq(SpaprInterruptController *intc, int lisn)
519 {
520     SpaprXive *xive = SPAPR_XIVE(intc);
521     assert(lisn < xive->nr_irqs);
522 
523     xive->eat[lisn].w &= cpu_to_be64(~EAS_VALID);
524 }
525 
526 static Property spapr_xive_properties[] = {
527     DEFINE_PROP_UINT32("nr-irqs", SpaprXive, nr_irqs, 0),
528     DEFINE_PROP_UINT32("nr-ends", SpaprXive, nr_ends, 0),
529     DEFINE_PROP_UINT64("vc-base", SpaprXive, vc_base, SPAPR_XIVE_VC_BASE),
530     DEFINE_PROP_UINT64("tm-base", SpaprXive, tm_base, SPAPR_XIVE_TM_BASE),
531     DEFINE_PROP_END_OF_LIST(),
532 };
533 
534 static int spapr_xive_cpu_intc_create(SpaprInterruptController *intc,
535                                       PowerPCCPU *cpu, Error **errp)
536 {
537     SpaprXive *xive = SPAPR_XIVE(intc);
538     Object *obj;
539     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
540 
541     obj = xive_tctx_create(OBJECT(cpu), XIVE_ROUTER(xive), errp);
542     if (!obj) {
543         return -1;
544     }
545 
546     spapr_cpu->tctx = XIVE_TCTX(obj);
547 
548     /*
549      * (TCG) Early setting the OS CAM line for hotplugged CPUs as they
550      * don't beneficiate from the reset of the XIVE IRQ backend
551      */
552     spapr_xive_set_tctx_os_cam(spapr_cpu->tctx);
553     return 0;
554 }
555 
556 static void spapr_xive_set_irq(SpaprInterruptController *intc, int irq, int val)
557 {
558     SpaprXive *xive = SPAPR_XIVE(intc);
559 
560     if (kvm_irqchip_in_kernel()) {
561         kvmppc_xive_source_set_irq(&xive->source, irq, val);
562     } else {
563         xive_source_set_irq(&xive->source, irq, val);
564     }
565 }
566 
567 static void spapr_xive_print_info(SpaprInterruptController *intc, Monitor *mon)
568 {
569     SpaprXive *xive = SPAPR_XIVE(intc);
570     CPUState *cs;
571 
572     CPU_FOREACH(cs) {
573         PowerPCCPU *cpu = POWERPC_CPU(cs);
574 
575         xive_tctx_pic_print_info(spapr_cpu_state(cpu)->tctx, mon);
576     }
577 
578     spapr_xive_pic_print_info(xive, mon);
579 }
580 
581 static void spapr_xive_class_init(ObjectClass *klass, void *data)
582 {
583     DeviceClass *dc = DEVICE_CLASS(klass);
584     XiveRouterClass *xrc = XIVE_ROUTER_CLASS(klass);
585     SpaprInterruptControllerClass *sicc = SPAPR_INTC_CLASS(klass);
586 
587     dc->desc    = "sPAPR XIVE Interrupt Controller";
588     dc->props   = spapr_xive_properties;
589     dc->realize = spapr_xive_realize;
590     dc->vmsd    = &vmstate_spapr_xive;
591 
592     xrc->get_eas = spapr_xive_get_eas;
593     xrc->get_end = spapr_xive_get_end;
594     xrc->write_end = spapr_xive_write_end;
595     xrc->get_nvt = spapr_xive_get_nvt;
596     xrc->write_nvt = spapr_xive_write_nvt;
597     xrc->get_tctx = spapr_xive_get_tctx;
598 
599     sicc->cpu_intc_create = spapr_xive_cpu_intc_create;
600     sicc->claim_irq = spapr_xive_claim_irq;
601     sicc->free_irq = spapr_xive_free_irq;
602     sicc->set_irq = spapr_xive_set_irq;
603     sicc->print_info = spapr_xive_print_info;
604 }
605 
606 static const TypeInfo spapr_xive_info = {
607     .name = TYPE_SPAPR_XIVE,
608     .parent = TYPE_XIVE_ROUTER,
609     .instance_init = spapr_xive_instance_init,
610     .instance_size = sizeof(SpaprXive),
611     .class_init = spapr_xive_class_init,
612     .interfaces = (InterfaceInfo[]) {
613         { TYPE_SPAPR_INTC },
614         { }
615     },
616 };
617 
618 static void spapr_xive_register_types(void)
619 {
620     type_register_static(&spapr_xive_info);
621 }
622 
623 type_init(spapr_xive_register_types)
624 
625 /*
626  * XIVE hcalls
627  *
628  * The terminology used by the XIVE hcalls is the following :
629  *
630  *   TARGET vCPU number
631  *   EQ     Event Queue assigned by OS to receive event data
632  *   ESB    page for source interrupt management
633  *   LISN   Logical Interrupt Source Number identifying a source in the
634  *          machine
635  *   EISN   Effective Interrupt Source Number used by guest OS to
636  *          identify source in the guest
637  *
638  * The EAS, END, NVT structures are not exposed.
639  */
640 
641 /*
642  * Linux hosts under OPAL reserve priority 7 for their own escalation
643  * interrupts (DD2.X POWER9). So we only allow the guest to use
644  * priorities [0..6].
645  */
646 static bool spapr_xive_priority_is_reserved(uint8_t priority)
647 {
648     switch (priority) {
649     case 0 ... 6:
650         return false;
651     case 7: /* OPAL escalation queue */
652     default:
653         return true;
654     }
655 }
656 
657 /*
658  * The H_INT_GET_SOURCE_INFO hcall() is used to obtain the logical
659  * real address of the MMIO page through which the Event State Buffer
660  * entry associated with the value of the "lisn" parameter is managed.
661  *
662  * Parameters:
663  * Input
664  * - R4: "flags"
665  *         Bits 0-63 reserved
666  * - R5: "lisn" is per "interrupts", "interrupt-map", or
667  *       "ibm,xive-lisn-ranges" properties, or as returned by the
668  *       ibm,query-interrupt-source-number RTAS call, or as returned
669  *       by the H_ALLOCATE_VAS_WINDOW hcall
670  *
671  * Output
672  * - R4: "flags"
673  *         Bits 0-59: Reserved
674  *         Bit 60: H_INT_ESB must be used for Event State Buffer
675  *                 management
676  *         Bit 61: 1 == LSI  0 == MSI
677  *         Bit 62: the full function page supports trigger
678  *         Bit 63: Store EOI Supported
679  * - R5: Logical Real address of full function Event State Buffer
680  *       management page, -1 if H_INT_ESB hcall flag is set to 1.
681  * - R6: Logical Real Address of trigger only Event State Buffer
682  *       management page or -1.
683  * - R7: Power of 2 page size for the ESB management pages returned in
684  *       R5 and R6.
685  */
686 
687 #define SPAPR_XIVE_SRC_H_INT_ESB     PPC_BIT(60) /* ESB manage with H_INT_ESB */
688 #define SPAPR_XIVE_SRC_LSI           PPC_BIT(61) /* Virtual LSI type */
689 #define SPAPR_XIVE_SRC_TRIGGER       PPC_BIT(62) /* Trigger and management
690                                                     on same page */
691 #define SPAPR_XIVE_SRC_STORE_EOI     PPC_BIT(63) /* Store EOI support */
692 
693 static target_ulong h_int_get_source_info(PowerPCCPU *cpu,
694                                           SpaprMachineState *spapr,
695                                           target_ulong opcode,
696                                           target_ulong *args)
697 {
698     SpaprXive *xive = spapr->xive;
699     XiveSource *xsrc = &xive->source;
700     target_ulong flags  = args[0];
701     target_ulong lisn   = args[1];
702 
703     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
704         return H_FUNCTION;
705     }
706 
707     if (flags) {
708         return H_PARAMETER;
709     }
710 
711     if (lisn >= xive->nr_irqs) {
712         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
713                       lisn);
714         return H_P2;
715     }
716 
717     if (!xive_eas_is_valid(&xive->eat[lisn])) {
718         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
719                       lisn);
720         return H_P2;
721     }
722 
723     /*
724      * All sources are emulated under the main XIVE object and share
725      * the same characteristics.
726      */
727     args[0] = 0;
728     if (!xive_source_esb_has_2page(xsrc)) {
729         args[0] |= SPAPR_XIVE_SRC_TRIGGER;
730     }
731     if (xsrc->esb_flags & XIVE_SRC_STORE_EOI) {
732         args[0] |= SPAPR_XIVE_SRC_STORE_EOI;
733     }
734 
735     /*
736      * Force the use of the H_INT_ESB hcall in case of an LSI
737      * interrupt. This is necessary under KVM to re-trigger the
738      * interrupt if the level is still asserted
739      */
740     if (xive_source_irq_is_lsi(xsrc, lisn)) {
741         args[0] |= SPAPR_XIVE_SRC_H_INT_ESB | SPAPR_XIVE_SRC_LSI;
742     }
743 
744     if (!(args[0] & SPAPR_XIVE_SRC_H_INT_ESB)) {
745         args[1] = xive->vc_base + xive_source_esb_mgmt(xsrc, lisn);
746     } else {
747         args[1] = -1;
748     }
749 
750     if (xive_source_esb_has_2page(xsrc) &&
751         !(args[0] & SPAPR_XIVE_SRC_H_INT_ESB)) {
752         args[2] = xive->vc_base + xive_source_esb_page(xsrc, lisn);
753     } else {
754         args[2] = -1;
755     }
756 
757     if (xive_source_esb_has_2page(xsrc)) {
758         args[3] = xsrc->esb_shift - 1;
759     } else {
760         args[3] = xsrc->esb_shift;
761     }
762 
763     return H_SUCCESS;
764 }
765 
766 /*
767  * The H_INT_SET_SOURCE_CONFIG hcall() is used to assign a Logical
768  * Interrupt Source to a target. The Logical Interrupt Source is
769  * designated with the "lisn" parameter and the target is designated
770  * with the "target" and "priority" parameters.  Upon return from the
771  * hcall(), no additional interrupts will be directed to the old EQ.
772  *
773  * Parameters:
774  * Input:
775  * - R4: "flags"
776  *         Bits 0-61: Reserved
777  *         Bit 62: set the "eisn" in the EAS
778  *         Bit 63: masks the interrupt source in the hardware interrupt
779  *       control structure. An interrupt masked by this mechanism will
780  *       be dropped, but it's source state bits will still be
781  *       set. There is no race-free way of unmasking and restoring the
782  *       source. Thus this should only be used in interrupts that are
783  *       also masked at the source, and only in cases where the
784  *       interrupt is not meant to be used for a large amount of time
785  *       because no valid target exists for it for example
786  * - R5: "lisn" is per "interrupts", "interrupt-map", or
787  *       "ibm,xive-lisn-ranges" properties, or as returned by the
788  *       ibm,query-interrupt-source-number RTAS call, or as returned by
789  *       the H_ALLOCATE_VAS_WINDOW hcall
790  * - R6: "target" is per "ibm,ppc-interrupt-server#s" or
791  *       "ibm,ppc-interrupt-gserver#s"
792  * - R7: "priority" is a valid priority not in
793  *       "ibm,plat-res-int-priorities"
794  * - R8: "eisn" is the guest EISN associated with the "lisn"
795  *
796  * Output:
797  * - None
798  */
799 
800 #define SPAPR_XIVE_SRC_SET_EISN PPC_BIT(62)
801 #define SPAPR_XIVE_SRC_MASK     PPC_BIT(63)
802 
803 static target_ulong h_int_set_source_config(PowerPCCPU *cpu,
804                                             SpaprMachineState *spapr,
805                                             target_ulong opcode,
806                                             target_ulong *args)
807 {
808     SpaprXive *xive = spapr->xive;
809     XiveEAS eas, new_eas;
810     target_ulong flags    = args[0];
811     target_ulong lisn     = args[1];
812     target_ulong target   = args[2];
813     target_ulong priority = args[3];
814     target_ulong eisn     = args[4];
815     uint8_t end_blk;
816     uint32_t end_idx;
817 
818     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
819         return H_FUNCTION;
820     }
821 
822     if (flags & ~(SPAPR_XIVE_SRC_SET_EISN | SPAPR_XIVE_SRC_MASK)) {
823         return H_PARAMETER;
824     }
825 
826     if (lisn >= xive->nr_irqs) {
827         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
828                       lisn);
829         return H_P2;
830     }
831 
832     eas = xive->eat[lisn];
833     if (!xive_eas_is_valid(&eas)) {
834         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
835                       lisn);
836         return H_P2;
837     }
838 
839     /* priority 0xff is used to reset the EAS */
840     if (priority == 0xff) {
841         new_eas.w = cpu_to_be64(EAS_VALID | EAS_MASKED);
842         goto out;
843     }
844 
845     if (flags & SPAPR_XIVE_SRC_MASK) {
846         new_eas.w = eas.w | cpu_to_be64(EAS_MASKED);
847     } else {
848         new_eas.w = eas.w & cpu_to_be64(~EAS_MASKED);
849     }
850 
851     if (spapr_xive_priority_is_reserved(priority)) {
852         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
853                       " is reserved\n", priority);
854         return H_P4;
855     }
856 
857     /*
858      * Validate that "target" is part of the list of threads allocated
859      * to the partition. For that, find the END corresponding to the
860      * target.
861      */
862     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
863         return H_P3;
864     }
865 
866     new_eas.w = xive_set_field64(EAS_END_BLOCK, new_eas.w, end_blk);
867     new_eas.w = xive_set_field64(EAS_END_INDEX, new_eas.w, end_idx);
868 
869     if (flags & SPAPR_XIVE_SRC_SET_EISN) {
870         new_eas.w = xive_set_field64(EAS_END_DATA, new_eas.w, eisn);
871     }
872 
873     if (kvm_irqchip_in_kernel()) {
874         Error *local_err = NULL;
875 
876         kvmppc_xive_set_source_config(xive, lisn, &new_eas, &local_err);
877         if (local_err) {
878             error_report_err(local_err);
879             return H_HARDWARE;
880         }
881     }
882 
883 out:
884     xive->eat[lisn] = new_eas;
885     return H_SUCCESS;
886 }
887 
888 /*
889  * The H_INT_GET_SOURCE_CONFIG hcall() is used to determine to which
890  * target/priority pair is assigned to the specified Logical Interrupt
891  * Source.
892  *
893  * Parameters:
894  * Input:
895  * - R4: "flags"
896  *         Bits 0-63 Reserved
897  * - R5: "lisn" is per "interrupts", "interrupt-map", or
898  *       "ibm,xive-lisn-ranges" properties, or as returned by the
899  *       ibm,query-interrupt-source-number RTAS call, or as
900  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
901  *
902  * Output:
903  * - R4: Target to which the specified Logical Interrupt Source is
904  *       assigned
905  * - R5: Priority to which the specified Logical Interrupt Source is
906  *       assigned
907  * - R6: EISN for the specified Logical Interrupt Source (this will be
908  *       equivalent to the LISN if not changed by H_INT_SET_SOURCE_CONFIG)
909  */
910 static target_ulong h_int_get_source_config(PowerPCCPU *cpu,
911                                             SpaprMachineState *spapr,
912                                             target_ulong opcode,
913                                             target_ulong *args)
914 {
915     SpaprXive *xive = spapr->xive;
916     target_ulong flags = args[0];
917     target_ulong lisn = args[1];
918     XiveEAS eas;
919     XiveEND *end;
920     uint8_t nvt_blk;
921     uint32_t end_idx, nvt_idx;
922 
923     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
924         return H_FUNCTION;
925     }
926 
927     if (flags) {
928         return H_PARAMETER;
929     }
930 
931     if (lisn >= xive->nr_irqs) {
932         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
933                       lisn);
934         return H_P2;
935     }
936 
937     eas = xive->eat[lisn];
938     if (!xive_eas_is_valid(&eas)) {
939         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
940                       lisn);
941         return H_P2;
942     }
943 
944     /* EAS_END_BLOCK is unused on sPAPR */
945     end_idx = xive_get_field64(EAS_END_INDEX, eas.w);
946 
947     assert(end_idx < xive->nr_ends);
948     end = &xive->endt[end_idx];
949 
950     nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end->w6);
951     nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end->w6);
952     args[0] = spapr_xive_nvt_to_target(nvt_blk, nvt_idx);
953 
954     if (xive_eas_is_masked(&eas)) {
955         args[1] = 0xff;
956     } else {
957         args[1] = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
958     }
959 
960     args[2] = xive_get_field64(EAS_END_DATA, eas.w);
961 
962     return H_SUCCESS;
963 }
964 
965 /*
966  * The H_INT_GET_QUEUE_INFO hcall() is used to get the logical real
967  * address of the notification management page associated with the
968  * specified target and priority.
969  *
970  * Parameters:
971  * Input:
972  * - R4: "flags"
973  *         Bits 0-63 Reserved
974  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
975  *       "ibm,ppc-interrupt-gserver#s"
976  * - R6: "priority" is a valid priority not in
977  *       "ibm,plat-res-int-priorities"
978  *
979  * Output:
980  * - R4: Logical real address of notification page
981  * - R5: Power of 2 page size of the notification page
982  */
983 static target_ulong h_int_get_queue_info(PowerPCCPU *cpu,
984                                          SpaprMachineState *spapr,
985                                          target_ulong opcode,
986                                          target_ulong *args)
987 {
988     SpaprXive *xive = spapr->xive;
989     XiveENDSource *end_xsrc = &xive->end_source;
990     target_ulong flags = args[0];
991     target_ulong target = args[1];
992     target_ulong priority = args[2];
993     XiveEND *end;
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) {
1002         return H_PARAMETER;
1003     }
1004 
1005     /*
1006      * H_STATE should be returned if a H_INT_RESET is in progress.
1007      * This is not needed when running the emulation under QEMU
1008      */
1009 
1010     if (spapr_xive_priority_is_reserved(priority)) {
1011         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1012                       " is reserved\n", priority);
1013         return H_P3;
1014     }
1015 
1016     /*
1017      * Validate that "target" is part of the list of threads allocated
1018      * to the partition. For that, find the END corresponding to the
1019      * target.
1020      */
1021     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1022         return H_P2;
1023     }
1024 
1025     assert(end_idx < xive->nr_ends);
1026     end = &xive->endt[end_idx];
1027 
1028     args[0] = xive->end_base + (1ull << (end_xsrc->esb_shift + 1)) * end_idx;
1029     if (xive_end_is_enqueue(end)) {
1030         args[1] = xive_get_field32(END_W0_QSIZE, end->w0) + 12;
1031     } else {
1032         args[1] = 0;
1033     }
1034 
1035     return H_SUCCESS;
1036 }
1037 
1038 /*
1039  * The H_INT_SET_QUEUE_CONFIG hcall() is used to set or reset a EQ for
1040  * a given "target" and "priority".  It is also used to set the
1041  * notification config associated with the EQ.  An EQ size of 0 is
1042  * used to reset the EQ config for a given target and priority. If
1043  * resetting the EQ config, the END associated with the given "target"
1044  * and "priority" will be changed to disable queueing.
1045  *
1046  * Upon return from the hcall(), no additional interrupts will be
1047  * directed to the old EQ (if one was set). The old EQ (if one was
1048  * set) should be investigated for interrupts that occurred prior to
1049  * or during the hcall().
1050  *
1051  * Parameters:
1052  * Input:
1053  * - R4: "flags"
1054  *         Bits 0-62: Reserved
1055  *         Bit 63: Unconditional Notify (n) per the XIVE spec
1056  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1057  *       "ibm,ppc-interrupt-gserver#s"
1058  * - R6: "priority" is a valid priority not in
1059  *       "ibm,plat-res-int-priorities"
1060  * - R7: "eventQueue": The logical real address of the start of the EQ
1061  * - R8: "eventQueueSize": The power of 2 EQ size per "ibm,xive-eq-sizes"
1062  *
1063  * Output:
1064  * - None
1065  */
1066 
1067 #define SPAPR_XIVE_END_ALWAYS_NOTIFY PPC_BIT(63)
1068 
1069 static target_ulong h_int_set_queue_config(PowerPCCPU *cpu,
1070                                            SpaprMachineState *spapr,
1071                                            target_ulong opcode,
1072                                            target_ulong *args)
1073 {
1074     SpaprXive *xive = spapr->xive;
1075     target_ulong flags = args[0];
1076     target_ulong target = args[1];
1077     target_ulong priority = args[2];
1078     target_ulong qpage = args[3];
1079     target_ulong qsize = args[4];
1080     XiveEND end;
1081     uint8_t end_blk, nvt_blk;
1082     uint32_t end_idx, nvt_idx;
1083 
1084     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1085         return H_FUNCTION;
1086     }
1087 
1088     if (flags & ~SPAPR_XIVE_END_ALWAYS_NOTIFY) {
1089         return H_PARAMETER;
1090     }
1091 
1092     /*
1093      * H_STATE should be returned if a H_INT_RESET is in progress.
1094      * This is not needed when running the emulation under QEMU
1095      */
1096 
1097     if (spapr_xive_priority_is_reserved(priority)) {
1098         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1099                       " is reserved\n", priority);
1100         return H_P3;
1101     }
1102 
1103     /*
1104      * Validate that "target" is part of the list of threads allocated
1105      * to the partition. For that, find the END corresponding to the
1106      * target.
1107      */
1108 
1109     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1110         return H_P2;
1111     }
1112 
1113     assert(end_idx < xive->nr_ends);
1114     memcpy(&end, &xive->endt[end_idx], sizeof(XiveEND));
1115 
1116     switch (qsize) {
1117     case 12:
1118     case 16:
1119     case 21:
1120     case 24:
1121         if (!QEMU_IS_ALIGNED(qpage, 1ul << qsize)) {
1122             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: EQ @0x%" HWADDR_PRIx
1123                           " is not naturally aligned with %" HWADDR_PRIx "\n",
1124                           qpage, (hwaddr)1 << qsize);
1125             return H_P4;
1126         }
1127         end.w2 = cpu_to_be32((qpage >> 32) & 0x0fffffff);
1128         end.w3 = cpu_to_be32(qpage & 0xffffffff);
1129         end.w0 |= cpu_to_be32(END_W0_ENQUEUE);
1130         end.w0 = xive_set_field32(END_W0_QSIZE, end.w0, qsize - 12);
1131         break;
1132     case 0:
1133         /* reset queue and disable queueing */
1134         spapr_xive_end_reset(&end);
1135         goto out;
1136 
1137     default:
1138         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid EQ size %"PRIx64"\n",
1139                       qsize);
1140         return H_P5;
1141     }
1142 
1143     if (qsize) {
1144         hwaddr plen = 1 << qsize;
1145         void *eq;
1146 
1147         /*
1148          * Validate the guest EQ. We should also check that the queue
1149          * has been zeroed by the OS.
1150          */
1151         eq = address_space_map(CPU(cpu)->as, qpage, &plen, true,
1152                                MEMTXATTRS_UNSPECIFIED);
1153         if (plen != 1 << qsize) {
1154             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to map EQ @0x%"
1155                           HWADDR_PRIx "\n", qpage);
1156             return H_P4;
1157         }
1158         address_space_unmap(CPU(cpu)->as, eq, plen, true, plen);
1159     }
1160 
1161     /* "target" should have been validated above */
1162     if (spapr_xive_target_to_nvt(target, &nvt_blk, &nvt_idx)) {
1163         g_assert_not_reached();
1164     }
1165 
1166     /*
1167      * Ensure the priority and target are correctly set (they will not
1168      * be right after allocation)
1169      */
1170     end.w6 = xive_set_field32(END_W6_NVT_BLOCK, 0ul, nvt_blk) |
1171         xive_set_field32(END_W6_NVT_INDEX, 0ul, nvt_idx);
1172     end.w7 = xive_set_field32(END_W7_F0_PRIORITY, 0ul, priority);
1173 
1174     if (flags & SPAPR_XIVE_END_ALWAYS_NOTIFY) {
1175         end.w0 |= cpu_to_be32(END_W0_UCOND_NOTIFY);
1176     } else {
1177         end.w0 &= cpu_to_be32((uint32_t)~END_W0_UCOND_NOTIFY);
1178     }
1179 
1180     /*
1181      * The generation bit for the END starts at 1 and The END page
1182      * offset counter starts at 0.
1183      */
1184     end.w1 = cpu_to_be32(END_W1_GENERATION) |
1185         xive_set_field32(END_W1_PAGE_OFF, 0ul, 0ul);
1186     end.w0 |= cpu_to_be32(END_W0_VALID);
1187 
1188     /*
1189      * TODO: issue syncs required to ensure all in-flight interrupts
1190      * are complete on the old END
1191      */
1192 
1193 out:
1194     if (kvm_irqchip_in_kernel()) {
1195         Error *local_err = NULL;
1196 
1197         kvmppc_xive_set_queue_config(xive, end_blk, end_idx, &end, &local_err);
1198         if (local_err) {
1199             error_report_err(local_err);
1200             return H_HARDWARE;
1201         }
1202     }
1203 
1204     /* Update END */
1205     memcpy(&xive->endt[end_idx], &end, sizeof(XiveEND));
1206     return H_SUCCESS;
1207 }
1208 
1209 /*
1210  * The H_INT_GET_QUEUE_CONFIG hcall() is used to get a EQ for a given
1211  * target and priority.
1212  *
1213  * Parameters:
1214  * Input:
1215  * - R4: "flags"
1216  *         Bits 0-62: Reserved
1217  *         Bit 63: Debug: Return debug data
1218  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1219  *       "ibm,ppc-interrupt-gserver#s"
1220  * - R6: "priority" is a valid priority not in
1221  *       "ibm,plat-res-int-priorities"
1222  *
1223  * Output:
1224  * - R4: "flags":
1225  *       Bits 0-61: Reserved
1226  *       Bit 62: The value of Event Queue Generation Number (g) per
1227  *              the XIVE spec if "Debug" = 1
1228  *       Bit 63: The value of Unconditional Notify (n) per the XIVE spec
1229  * - R5: The logical real address of the start of the EQ
1230  * - R6: The power of 2 EQ size per "ibm,xive-eq-sizes"
1231  * - R7: The value of Event Queue Offset Counter per XIVE spec
1232  *       if "Debug" = 1, else 0
1233  *
1234  */
1235 
1236 #define SPAPR_XIVE_END_DEBUG     PPC_BIT(63)
1237 
1238 static target_ulong h_int_get_queue_config(PowerPCCPU *cpu,
1239                                            SpaprMachineState *spapr,
1240                                            target_ulong opcode,
1241                                            target_ulong *args)
1242 {
1243     SpaprXive *xive = spapr->xive;
1244     target_ulong flags = args[0];
1245     target_ulong target = args[1];
1246     target_ulong priority = args[2];
1247     XiveEND *end;
1248     uint8_t end_blk;
1249     uint32_t end_idx;
1250 
1251     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1252         return H_FUNCTION;
1253     }
1254 
1255     if (flags & ~SPAPR_XIVE_END_DEBUG) {
1256         return H_PARAMETER;
1257     }
1258 
1259     /*
1260      * H_STATE should be returned if a H_INT_RESET is in progress.
1261      * This is not needed when running the emulation under QEMU
1262      */
1263 
1264     if (spapr_xive_priority_is_reserved(priority)) {
1265         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1266                       " is reserved\n", priority);
1267         return H_P3;
1268     }
1269 
1270     /*
1271      * Validate that "target" is part of the list of threads allocated
1272      * to the partition. For that, find the END corresponding to the
1273      * target.
1274      */
1275     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1276         return H_P2;
1277     }
1278 
1279     assert(end_idx < xive->nr_ends);
1280     end = &xive->endt[end_idx];
1281 
1282     args[0] = 0;
1283     if (xive_end_is_notify(end)) {
1284         args[0] |= SPAPR_XIVE_END_ALWAYS_NOTIFY;
1285     }
1286 
1287     if (xive_end_is_enqueue(end)) {
1288         args[1] = xive_end_qaddr(end);
1289         args[2] = xive_get_field32(END_W0_QSIZE, end->w0) + 12;
1290     } else {
1291         args[1] = 0;
1292         args[2] = 0;
1293     }
1294 
1295     if (kvm_irqchip_in_kernel()) {
1296         Error *local_err = NULL;
1297 
1298         kvmppc_xive_get_queue_config(xive, end_blk, end_idx, end, &local_err);
1299         if (local_err) {
1300             error_report_err(local_err);
1301             return H_HARDWARE;
1302         }
1303     }
1304 
1305     /* TODO: do we need any locking on the END ? */
1306     if (flags & SPAPR_XIVE_END_DEBUG) {
1307         /* Load the event queue generation number into the return flags */
1308         args[0] |= (uint64_t)xive_get_field32(END_W1_GENERATION, end->w1) << 62;
1309 
1310         /* Load R7 with the event queue offset counter */
1311         args[3] = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1312     } else {
1313         args[3] = 0;
1314     }
1315 
1316     return H_SUCCESS;
1317 }
1318 
1319 /*
1320  * The H_INT_SET_OS_REPORTING_LINE hcall() is used to set the
1321  * reporting cache line pair for the calling thread.  The reporting
1322  * cache lines will contain the OS interrupt context when the OS
1323  * issues a CI store byte to @TIMA+0xC10 to acknowledge the OS
1324  * interrupt. The reporting cache lines can be reset by inputting -1
1325  * in "reportingLine".  Issuing the CI store byte without reporting
1326  * cache lines registered will result in the data not being accessible
1327  * to the OS.
1328  *
1329  * Parameters:
1330  * Input:
1331  * - R4: "flags"
1332  *         Bits 0-63: Reserved
1333  * - R5: "reportingLine": The logical real address of the reporting cache
1334  *       line pair
1335  *
1336  * Output:
1337  * - None
1338  */
1339 static target_ulong h_int_set_os_reporting_line(PowerPCCPU *cpu,
1340                                                 SpaprMachineState *spapr,
1341                                                 target_ulong opcode,
1342                                                 target_ulong *args)
1343 {
1344     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1345         return H_FUNCTION;
1346     }
1347 
1348     /*
1349      * H_STATE should be returned if a H_INT_RESET is in progress.
1350      * This is not needed when running the emulation under QEMU
1351      */
1352 
1353     /* TODO: H_INT_SET_OS_REPORTING_LINE */
1354     return H_FUNCTION;
1355 }
1356 
1357 /*
1358  * The H_INT_GET_OS_REPORTING_LINE hcall() is used to get the logical
1359  * real address of the reporting cache line pair set for the input
1360  * "target".  If no reporting cache line pair has been set, -1 is
1361  * returned.
1362  *
1363  * Parameters:
1364  * Input:
1365  * - R4: "flags"
1366  *         Bits 0-63: Reserved
1367  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1368  *       "ibm,ppc-interrupt-gserver#s"
1369  * - R6: "reportingLine": The logical real address of the reporting
1370  *        cache line pair
1371  *
1372  * Output:
1373  * - R4: The logical real address of the reporting line if set, else -1
1374  */
1375 static target_ulong h_int_get_os_reporting_line(PowerPCCPU *cpu,
1376                                                 SpaprMachineState *spapr,
1377                                                 target_ulong opcode,
1378                                                 target_ulong *args)
1379 {
1380     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1381         return H_FUNCTION;
1382     }
1383 
1384     /*
1385      * H_STATE should be returned if a H_INT_RESET is in progress.
1386      * This is not needed when running the emulation under QEMU
1387      */
1388 
1389     /* TODO: H_INT_GET_OS_REPORTING_LINE */
1390     return H_FUNCTION;
1391 }
1392 
1393 /*
1394  * The H_INT_ESB hcall() is used to issue a load or store to the ESB
1395  * page for the input "lisn".  This hcall is only supported for LISNs
1396  * that have the ESB hcall flag set to 1 when returned from hcall()
1397  * H_INT_GET_SOURCE_INFO.
1398  *
1399  * Parameters:
1400  * Input:
1401  * - R4: "flags"
1402  *         Bits 0-62: Reserved
1403  *         bit 63: Store: Store=1, store operation, else load operation
1404  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1405  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1406  *       ibm,query-interrupt-source-number RTAS call, or as
1407  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1408  * - R6: "esbOffset" is the offset into the ESB page for the load or
1409  *       store operation
1410  * - R7: "storeData" is the data to write for a store operation
1411  *
1412  * Output:
1413  * - R4: The value of the load if load operation, else -1
1414  */
1415 
1416 #define SPAPR_XIVE_ESB_STORE PPC_BIT(63)
1417 
1418 static target_ulong h_int_esb(PowerPCCPU *cpu,
1419                               SpaprMachineState *spapr,
1420                               target_ulong opcode,
1421                               target_ulong *args)
1422 {
1423     SpaprXive *xive = spapr->xive;
1424     XiveEAS eas;
1425     target_ulong flags  = args[0];
1426     target_ulong lisn   = args[1];
1427     target_ulong offset = args[2];
1428     target_ulong data   = args[3];
1429     hwaddr mmio_addr;
1430     XiveSource *xsrc = &xive->source;
1431 
1432     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1433         return H_FUNCTION;
1434     }
1435 
1436     if (flags & ~SPAPR_XIVE_ESB_STORE) {
1437         return H_PARAMETER;
1438     }
1439 
1440     if (lisn >= xive->nr_irqs) {
1441         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1442                       lisn);
1443         return H_P2;
1444     }
1445 
1446     eas = xive->eat[lisn];
1447     if (!xive_eas_is_valid(&eas)) {
1448         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1449                       lisn);
1450         return H_P2;
1451     }
1452 
1453     if (offset > (1ull << xsrc->esb_shift)) {
1454         return H_P3;
1455     }
1456 
1457     if (kvm_irqchip_in_kernel()) {
1458         args[0] = kvmppc_xive_esb_rw(xsrc, lisn, offset, data,
1459                                      flags & SPAPR_XIVE_ESB_STORE);
1460     } else {
1461         mmio_addr = xive->vc_base + xive_source_esb_mgmt(xsrc, lisn) + offset;
1462 
1463         if (dma_memory_rw(&address_space_memory, mmio_addr, &data, 8,
1464                           (flags & SPAPR_XIVE_ESB_STORE))) {
1465             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to access ESB @0x%"
1466                           HWADDR_PRIx "\n", mmio_addr);
1467             return H_HARDWARE;
1468         }
1469         args[0] = (flags & SPAPR_XIVE_ESB_STORE) ? -1 : data;
1470     }
1471     return H_SUCCESS;
1472 }
1473 
1474 /*
1475  * The H_INT_SYNC hcall() is used to issue hardware syncs that will
1476  * ensure any in flight events for the input lisn are in the event
1477  * queue.
1478  *
1479  * Parameters:
1480  * Input:
1481  * - R4: "flags"
1482  *         Bits 0-63: Reserved
1483  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1484  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1485  *       ibm,query-interrupt-source-number RTAS call, or as
1486  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1487  *
1488  * Output:
1489  * - None
1490  */
1491 static target_ulong h_int_sync(PowerPCCPU *cpu,
1492                                SpaprMachineState *spapr,
1493                                target_ulong opcode,
1494                                target_ulong *args)
1495 {
1496     SpaprXive *xive = spapr->xive;
1497     XiveEAS eas;
1498     target_ulong flags = args[0];
1499     target_ulong lisn = args[1];
1500 
1501     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1502         return H_FUNCTION;
1503     }
1504 
1505     if (flags) {
1506         return H_PARAMETER;
1507     }
1508 
1509     if (lisn >= xive->nr_irqs) {
1510         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1511                       lisn);
1512         return H_P2;
1513     }
1514 
1515     eas = xive->eat[lisn];
1516     if (!xive_eas_is_valid(&eas)) {
1517         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1518                       lisn);
1519         return H_P2;
1520     }
1521 
1522     /*
1523      * H_STATE should be returned if a H_INT_RESET is in progress.
1524      * This is not needed when running the emulation under QEMU
1525      */
1526 
1527     /*
1528      * This is not real hardware. Nothing to be done unless when
1529      * under KVM
1530      */
1531 
1532     if (kvm_irqchip_in_kernel()) {
1533         Error *local_err = NULL;
1534 
1535         kvmppc_xive_sync_source(xive, lisn, &local_err);
1536         if (local_err) {
1537             error_report_err(local_err);
1538             return H_HARDWARE;
1539         }
1540     }
1541     return H_SUCCESS;
1542 }
1543 
1544 /*
1545  * The H_INT_RESET hcall() is used to reset all of the partition's
1546  * interrupt exploitation structures to their initial state.  This
1547  * means losing all previously set interrupt state set via
1548  * H_INT_SET_SOURCE_CONFIG and H_INT_SET_QUEUE_CONFIG.
1549  *
1550  * Parameters:
1551  * Input:
1552  * - R4: "flags"
1553  *         Bits 0-63: Reserved
1554  *
1555  * Output:
1556  * - None
1557  */
1558 static target_ulong h_int_reset(PowerPCCPU *cpu,
1559                                 SpaprMachineState *spapr,
1560                                 target_ulong opcode,
1561                                 target_ulong *args)
1562 {
1563     SpaprXive *xive = spapr->xive;
1564     target_ulong flags   = args[0];
1565 
1566     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1567         return H_FUNCTION;
1568     }
1569 
1570     if (flags) {
1571         return H_PARAMETER;
1572     }
1573 
1574     device_reset(DEVICE(xive));
1575 
1576     if (kvm_irqchip_in_kernel()) {
1577         Error *local_err = NULL;
1578 
1579         kvmppc_xive_reset(xive, &local_err);
1580         if (local_err) {
1581             error_report_err(local_err);
1582             return H_HARDWARE;
1583         }
1584     }
1585     return H_SUCCESS;
1586 }
1587 
1588 void spapr_xive_hcall_init(SpaprMachineState *spapr)
1589 {
1590     spapr_register_hypercall(H_INT_GET_SOURCE_INFO, h_int_get_source_info);
1591     spapr_register_hypercall(H_INT_SET_SOURCE_CONFIG, h_int_set_source_config);
1592     spapr_register_hypercall(H_INT_GET_SOURCE_CONFIG, h_int_get_source_config);
1593     spapr_register_hypercall(H_INT_GET_QUEUE_INFO, h_int_get_queue_info);
1594     spapr_register_hypercall(H_INT_SET_QUEUE_CONFIG, h_int_set_queue_config);
1595     spapr_register_hypercall(H_INT_GET_QUEUE_CONFIG, h_int_get_queue_config);
1596     spapr_register_hypercall(H_INT_SET_OS_REPORTING_LINE,
1597                              h_int_set_os_reporting_line);
1598     spapr_register_hypercall(H_INT_GET_OS_REPORTING_LINE,
1599                              h_int_get_os_reporting_line);
1600     spapr_register_hypercall(H_INT_ESB, h_int_esb);
1601     spapr_register_hypercall(H_INT_SYNC, h_int_sync);
1602     spapr_register_hypercall(H_INT_RESET, h_int_reset);
1603 }
1604 
1605 void spapr_dt_xive(SpaprMachineState *spapr, uint32_t nr_servers, void *fdt,
1606                    uint32_t phandle)
1607 {
1608     SpaprXive *xive = spapr->xive;
1609     int node;
1610     uint64_t timas[2 * 2];
1611     /* Interrupt number ranges for the IPIs */
1612     uint32_t lisn_ranges[] = {
1613         cpu_to_be32(0),
1614         cpu_to_be32(nr_servers),
1615     };
1616     /*
1617      * EQ size - the sizes of pages supported by the system 4K, 64K,
1618      * 2M, 16M. We only advertise 64K for the moment.
1619      */
1620     uint32_t eq_sizes[] = {
1621         cpu_to_be32(16), /* 64K */
1622     };
1623     /*
1624      * The following array is in sync with the reserved priorities
1625      * defined by the 'spapr_xive_priority_is_reserved' routine.
1626      */
1627     uint32_t plat_res_int_priorities[] = {
1628         cpu_to_be32(7),    /* start */
1629         cpu_to_be32(0xf8), /* count */
1630     };
1631 
1632     /* Thread Interrupt Management Area : User (ring 3) and OS (ring 2) */
1633     timas[0] = cpu_to_be64(xive->tm_base +
1634                            XIVE_TM_USER_PAGE * (1ull << TM_SHIFT));
1635     timas[1] = cpu_to_be64(1ull << TM_SHIFT);
1636     timas[2] = cpu_to_be64(xive->tm_base +
1637                            XIVE_TM_OS_PAGE * (1ull << TM_SHIFT));
1638     timas[3] = cpu_to_be64(1ull << TM_SHIFT);
1639 
1640     _FDT(node = fdt_add_subnode(fdt, 0, xive->nodename));
1641 
1642     _FDT(fdt_setprop_string(fdt, node, "device_type", "power-ivpe"));
1643     _FDT(fdt_setprop(fdt, node, "reg", timas, sizeof(timas)));
1644 
1645     _FDT(fdt_setprop_string(fdt, node, "compatible", "ibm,power-ivpe"));
1646     _FDT(fdt_setprop(fdt, node, "ibm,xive-eq-sizes", eq_sizes,
1647                      sizeof(eq_sizes)));
1648     _FDT(fdt_setprop(fdt, node, "ibm,xive-lisn-ranges", lisn_ranges,
1649                      sizeof(lisn_ranges)));
1650 
1651     /* For Linux to link the LSIs to the interrupt controller. */
1652     _FDT(fdt_setprop(fdt, node, "interrupt-controller", NULL, 0));
1653     _FDT(fdt_setprop_cell(fdt, node, "#interrupt-cells", 2));
1654 
1655     /* For SLOF */
1656     _FDT(fdt_setprop_cell(fdt, node, "linux,phandle", phandle));
1657     _FDT(fdt_setprop_cell(fdt, node, "phandle", phandle));
1658 
1659     /*
1660      * The "ibm,plat-res-int-priorities" property defines the priority
1661      * ranges reserved by the hypervisor
1662      */
1663     _FDT(fdt_setprop(fdt, 0, "ibm,plat-res-int-priorities",
1664                      plat_res_int_priorities, sizeof(plat_res_int_priorities)));
1665 }
1666