xref: /openbmc/qemu/hw/intc/spapr_xive.c (revision 979a8902)
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     SpaprXiveClass *sxc = SPAPR_XIVE_GET_CLASS(xive);
290     XiveSource *xsrc = &xive->source;
291     XiveENDSource *end_xsrc = &xive->end_source;
292     Error *local_err = NULL;
293 
294     sxc->parent_realize(dev, &local_err);
295     if (local_err) {
296         error_propagate(errp, local_err);
297         return;
298     }
299 
300     if (!xive->nr_irqs) {
301         error_setg(errp, "Number of interrupt needs to be greater 0");
302         return;
303     }
304 
305     if (!xive->nr_ends) {
306         error_setg(errp, "Number of interrupt needs to be greater 0");
307         return;
308     }
309 
310     /*
311      * Initialize the internal sources, for IPIs and virtual devices.
312      */
313     object_property_set_int(OBJECT(xsrc), xive->nr_irqs, "nr-irqs",
314                             &error_fatal);
315     object_property_set_link(OBJECT(xsrc), OBJECT(xive), "xive",
316                              &error_abort);
317     object_property_set_bool(OBJECT(xsrc), true, "realized", &local_err);
318     if (local_err) {
319         error_propagate(errp, local_err);
320         return;
321     }
322     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &xsrc->esb_mmio);
323 
324     /*
325      * Initialize the END ESB source
326      */
327     object_property_set_int(OBJECT(end_xsrc), xive->nr_irqs, "nr-ends",
328                             &error_fatal);
329     object_property_set_link(OBJECT(end_xsrc), OBJECT(xive), "xive",
330                              &error_abort);
331     object_property_set_bool(OBJECT(end_xsrc), true, "realized", &local_err);
332     if (local_err) {
333         error_propagate(errp, local_err);
334         return;
335     }
336     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &end_xsrc->esb_mmio);
337 
338     /* Set the mapping address of the END ESB pages after the source ESBs */
339     xive->end_base = xive->vc_base + (1ull << xsrc->esb_shift) * xsrc->nr_irqs;
340 
341     /*
342      * Allocate the routing tables
343      */
344     xive->eat = g_new0(XiveEAS, xive->nr_irqs);
345     xive->endt = g_new0(XiveEND, xive->nr_ends);
346 
347     xive->nodename = g_strdup_printf("interrupt-controller@%" PRIx64,
348                            xive->tm_base + XIVE_TM_USER_PAGE * (1 << TM_SHIFT));
349 
350     qemu_register_reset(spapr_xive_reset, dev);
351 
352     /* TIMA initialization */
353     memory_region_init_io(&xive->tm_mmio, OBJECT(xive), &spapr_xive_tm_ops,
354                           xive, "xive.tima", 4ull << TM_SHIFT);
355     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &xive->tm_mmio);
356 
357     /*
358      * Map all regions. These will be enabled or disabled at reset and
359      * can also be overridden by KVM memory regions if active
360      */
361     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 0, xive->vc_base);
362     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 1, xive->end_base);
363     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 2, xive->tm_base);
364 }
365 
366 static int spapr_xive_get_eas(XiveRouter *xrtr, uint8_t eas_blk,
367                               uint32_t eas_idx, XiveEAS *eas)
368 {
369     SpaprXive *xive = SPAPR_XIVE(xrtr);
370 
371     if (eas_idx >= xive->nr_irqs) {
372         return -1;
373     }
374 
375     *eas = xive->eat[eas_idx];
376     return 0;
377 }
378 
379 static int spapr_xive_get_end(XiveRouter *xrtr,
380                               uint8_t end_blk, uint32_t end_idx, XiveEND *end)
381 {
382     SpaprXive *xive = SPAPR_XIVE(xrtr);
383 
384     if (end_idx >= xive->nr_ends) {
385         return -1;
386     }
387 
388     memcpy(end, &xive->endt[end_idx], sizeof(XiveEND));
389     return 0;
390 }
391 
392 static int spapr_xive_write_end(XiveRouter *xrtr, uint8_t end_blk,
393                                 uint32_t end_idx, XiveEND *end,
394                                 uint8_t word_number)
395 {
396     SpaprXive *xive = SPAPR_XIVE(xrtr);
397 
398     if (end_idx >= xive->nr_ends) {
399         return -1;
400     }
401 
402     memcpy(&xive->endt[end_idx], end, sizeof(XiveEND));
403     return 0;
404 }
405 
406 static int spapr_xive_get_nvt(XiveRouter *xrtr,
407                               uint8_t nvt_blk, uint32_t nvt_idx, XiveNVT *nvt)
408 {
409     uint32_t vcpu_id = spapr_xive_nvt_to_target(nvt_blk, nvt_idx);
410     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
411 
412     if (!cpu) {
413         /* TODO: should we assert() if we can find a NVT ? */
414         return -1;
415     }
416 
417     /*
418      * sPAPR does not maintain a NVT table. Return that the NVT is
419      * valid if we have found a matching CPU
420      */
421     nvt->w0 = cpu_to_be32(NVT_W0_VALID);
422     return 0;
423 }
424 
425 static int spapr_xive_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk,
426                                 uint32_t nvt_idx, XiveNVT *nvt,
427                                 uint8_t word_number)
428 {
429     /*
430      * We don't need to write back to the NVTs because the sPAPR
431      * machine should never hit a non-scheduled NVT. It should never
432      * get called.
433      */
434     g_assert_not_reached();
435 }
436 
437 static int spapr_xive_match_nvt(XivePresenter *xptr, uint8_t format,
438                                 uint8_t nvt_blk, uint32_t nvt_idx,
439                                 bool cam_ignore, uint8_t priority,
440                                 uint32_t logic_serv, XiveTCTXMatch *match)
441 {
442     CPUState *cs;
443     int count = 0;
444 
445     CPU_FOREACH(cs) {
446         PowerPCCPU *cpu = POWERPC_CPU(cs);
447         XiveTCTX *tctx = spapr_cpu_state(cpu)->tctx;
448         int ring;
449 
450         /*
451          * Skip partially initialized vCPUs. This can happen when
452          * vCPUs are hotplugged.
453          */
454         if (!tctx) {
455             continue;
456         }
457 
458         /*
459          * Check the thread context CAM lines and record matches.
460          */
461         ring = xive_presenter_tctx_match(xptr, tctx, format, nvt_blk, nvt_idx,
462                                          cam_ignore, logic_serv);
463         /*
464          * Save the matching thread interrupt context and follow on to
465          * check for duplicates which are invalid.
466          */
467         if (ring != -1) {
468             if (match->tctx) {
469                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: already found a thread "
470                               "context NVT %x/%x\n", nvt_blk, nvt_idx);
471                 return -1;
472             }
473 
474             match->ring = ring;
475             match->tctx = tctx;
476             count++;
477         }
478     }
479 
480     return count;
481 }
482 
483 static uint8_t spapr_xive_get_block_id(XiveRouter *xrtr)
484 {
485     return SPAPR_XIVE_BLOCK_ID;
486 }
487 
488 static const VMStateDescription vmstate_spapr_xive_end = {
489     .name = TYPE_SPAPR_XIVE "/end",
490     .version_id = 1,
491     .minimum_version_id = 1,
492     .fields = (VMStateField []) {
493         VMSTATE_UINT32(w0, XiveEND),
494         VMSTATE_UINT32(w1, XiveEND),
495         VMSTATE_UINT32(w2, XiveEND),
496         VMSTATE_UINT32(w3, XiveEND),
497         VMSTATE_UINT32(w4, XiveEND),
498         VMSTATE_UINT32(w5, XiveEND),
499         VMSTATE_UINT32(w6, XiveEND),
500         VMSTATE_UINT32(w7, XiveEND),
501         VMSTATE_END_OF_LIST()
502     },
503 };
504 
505 static const VMStateDescription vmstate_spapr_xive_eas = {
506     .name = TYPE_SPAPR_XIVE "/eas",
507     .version_id = 1,
508     .minimum_version_id = 1,
509     .fields = (VMStateField []) {
510         VMSTATE_UINT64(w, XiveEAS),
511         VMSTATE_END_OF_LIST()
512     },
513 };
514 
515 static int vmstate_spapr_xive_pre_save(void *opaque)
516 {
517     if (kvm_irqchip_in_kernel()) {
518         return kvmppc_xive_pre_save(SPAPR_XIVE(opaque));
519     }
520 
521     return 0;
522 }
523 
524 /*
525  * Called by the sPAPR IRQ backend 'post_load' method at the machine
526  * level.
527  */
528 static int spapr_xive_post_load(SpaprInterruptController *intc, int version_id)
529 {
530     if (kvm_irqchip_in_kernel()) {
531         return kvmppc_xive_post_load(SPAPR_XIVE(intc), version_id);
532     }
533 
534     return 0;
535 }
536 
537 static const VMStateDescription vmstate_spapr_xive = {
538     .name = TYPE_SPAPR_XIVE,
539     .version_id = 1,
540     .minimum_version_id = 1,
541     .pre_save = vmstate_spapr_xive_pre_save,
542     .post_load = NULL, /* handled at the machine level */
543     .fields = (VMStateField[]) {
544         VMSTATE_UINT32_EQUAL(nr_irqs, SpaprXive, NULL),
545         VMSTATE_STRUCT_VARRAY_POINTER_UINT32(eat, SpaprXive, nr_irqs,
546                                      vmstate_spapr_xive_eas, XiveEAS),
547         VMSTATE_STRUCT_VARRAY_POINTER_UINT32(endt, SpaprXive, nr_ends,
548                                              vmstate_spapr_xive_end, XiveEND),
549         VMSTATE_END_OF_LIST()
550     },
551 };
552 
553 static int spapr_xive_claim_irq(SpaprInterruptController *intc, int lisn,
554                                 bool lsi, Error **errp)
555 {
556     SpaprXive *xive = SPAPR_XIVE(intc);
557     XiveSource *xsrc = &xive->source;
558 
559     assert(lisn < xive->nr_irqs);
560 
561     if (xive_eas_is_valid(&xive->eat[lisn])) {
562         error_setg(errp, "IRQ %d is not free", lisn);
563         return -EBUSY;
564     }
565 
566     /*
567      * Set default values when allocating an IRQ number
568      */
569     xive->eat[lisn].w |= cpu_to_be64(EAS_VALID | EAS_MASKED);
570     if (lsi) {
571         xive_source_irq_set_lsi(xsrc, lisn);
572     }
573 
574     if (kvm_irqchip_in_kernel()) {
575         return kvmppc_xive_source_reset_one(xsrc, lisn, errp);
576     }
577 
578     return 0;
579 }
580 
581 static void spapr_xive_free_irq(SpaprInterruptController *intc, int lisn)
582 {
583     SpaprXive *xive = SPAPR_XIVE(intc);
584     assert(lisn < xive->nr_irqs);
585 
586     xive->eat[lisn].w &= cpu_to_be64(~EAS_VALID);
587 }
588 
589 static Property spapr_xive_properties[] = {
590     DEFINE_PROP_UINT32("nr-irqs", SpaprXive, nr_irqs, 0),
591     DEFINE_PROP_UINT32("nr-ends", SpaprXive, nr_ends, 0),
592     DEFINE_PROP_UINT64("vc-base", SpaprXive, vc_base, SPAPR_XIVE_VC_BASE),
593     DEFINE_PROP_UINT64("tm-base", SpaprXive, tm_base, SPAPR_XIVE_TM_BASE),
594     DEFINE_PROP_END_OF_LIST(),
595 };
596 
597 static int spapr_xive_cpu_intc_create(SpaprInterruptController *intc,
598                                       PowerPCCPU *cpu, Error **errp)
599 {
600     SpaprXive *xive = SPAPR_XIVE(intc);
601     Object *obj;
602     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
603 
604     obj = xive_tctx_create(OBJECT(cpu), XIVE_PRESENTER(xive), errp);
605     if (!obj) {
606         return -1;
607     }
608 
609     spapr_cpu->tctx = XIVE_TCTX(obj);
610     return 0;
611 }
612 
613 static void xive_tctx_set_os_cam(XiveTCTX *tctx, uint32_t os_cam)
614 {
615     uint32_t qw1w2 = cpu_to_be32(TM_QW1W2_VO | os_cam);
616     memcpy(&tctx->regs[TM_QW1_OS + TM_WORD2], &qw1w2, 4);
617 }
618 
619 static void spapr_xive_cpu_intc_reset(SpaprInterruptController *intc,
620                                      PowerPCCPU *cpu)
621 {
622     XiveTCTX *tctx = spapr_cpu_state(cpu)->tctx;
623     uint8_t  nvt_blk;
624     uint32_t nvt_idx;
625 
626     xive_tctx_reset(tctx);
627 
628     /*
629      * When a Virtual Processor is scheduled to run on a HW thread,
630      * the hypervisor pushes its identifier in the OS CAM line.
631      * Emulate the same behavior under QEMU.
632      */
633     spapr_xive_cpu_to_nvt(cpu, &nvt_blk, &nvt_idx);
634 
635     xive_tctx_set_os_cam(tctx, xive_nvt_cam_line(nvt_blk, nvt_idx));
636 }
637 
638 static void spapr_xive_cpu_intc_destroy(SpaprInterruptController *intc,
639                                         PowerPCCPU *cpu)
640 {
641     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
642 
643     xive_tctx_destroy(spapr_cpu->tctx);
644     spapr_cpu->tctx = NULL;
645 }
646 
647 static void spapr_xive_set_irq(SpaprInterruptController *intc, int irq, int val)
648 {
649     SpaprXive *xive = SPAPR_XIVE(intc);
650 
651     if (kvm_irqchip_in_kernel()) {
652         kvmppc_xive_source_set_irq(&xive->source, irq, val);
653     } else {
654         xive_source_set_irq(&xive->source, irq, val);
655     }
656 }
657 
658 static void spapr_xive_print_info(SpaprInterruptController *intc, Monitor *mon)
659 {
660     SpaprXive *xive = SPAPR_XIVE(intc);
661     CPUState *cs;
662 
663     CPU_FOREACH(cs) {
664         PowerPCCPU *cpu = POWERPC_CPU(cs);
665 
666         xive_tctx_pic_print_info(spapr_cpu_state(cpu)->tctx, mon);
667     }
668 
669     spapr_xive_pic_print_info(xive, mon);
670 }
671 
672 static void spapr_xive_dt(SpaprInterruptController *intc, uint32_t nr_servers,
673                           void *fdt, uint32_t phandle)
674 {
675     SpaprXive *xive = SPAPR_XIVE(intc);
676     int node;
677     uint64_t timas[2 * 2];
678     /* Interrupt number ranges for the IPIs */
679     uint32_t lisn_ranges[] = {
680         cpu_to_be32(0),
681         cpu_to_be32(nr_servers),
682     };
683     /*
684      * EQ size - the sizes of pages supported by the system 4K, 64K,
685      * 2M, 16M. We only advertise 64K for the moment.
686      */
687     uint32_t eq_sizes[] = {
688         cpu_to_be32(16), /* 64K */
689     };
690     /*
691      * The following array is in sync with the reserved priorities
692      * defined by the 'spapr_xive_priority_is_reserved' routine.
693      */
694     uint32_t plat_res_int_priorities[] = {
695         cpu_to_be32(7),    /* start */
696         cpu_to_be32(0xf8), /* count */
697     };
698 
699     /* Thread Interrupt Management Area : User (ring 3) and OS (ring 2) */
700     timas[0] = cpu_to_be64(xive->tm_base +
701                            XIVE_TM_USER_PAGE * (1ull << TM_SHIFT));
702     timas[1] = cpu_to_be64(1ull << TM_SHIFT);
703     timas[2] = cpu_to_be64(xive->tm_base +
704                            XIVE_TM_OS_PAGE * (1ull << TM_SHIFT));
705     timas[3] = cpu_to_be64(1ull << TM_SHIFT);
706 
707     _FDT(node = fdt_add_subnode(fdt, 0, xive->nodename));
708 
709     _FDT(fdt_setprop_string(fdt, node, "device_type", "power-ivpe"));
710     _FDT(fdt_setprop(fdt, node, "reg", timas, sizeof(timas)));
711 
712     _FDT(fdt_setprop_string(fdt, node, "compatible", "ibm,power-ivpe"));
713     _FDT(fdt_setprop(fdt, node, "ibm,xive-eq-sizes", eq_sizes,
714                      sizeof(eq_sizes)));
715     _FDT(fdt_setprop(fdt, node, "ibm,xive-lisn-ranges", lisn_ranges,
716                      sizeof(lisn_ranges)));
717 
718     /* For Linux to link the LSIs to the interrupt controller. */
719     _FDT(fdt_setprop(fdt, node, "interrupt-controller", NULL, 0));
720     _FDT(fdt_setprop_cell(fdt, node, "#interrupt-cells", 2));
721 
722     /* For SLOF */
723     _FDT(fdt_setprop_cell(fdt, node, "linux,phandle", phandle));
724     _FDT(fdt_setprop_cell(fdt, node, "phandle", phandle));
725 
726     /*
727      * The "ibm,plat-res-int-priorities" property defines the priority
728      * ranges reserved by the hypervisor
729      */
730     _FDT(fdt_setprop(fdt, 0, "ibm,plat-res-int-priorities",
731                      plat_res_int_priorities, sizeof(plat_res_int_priorities)));
732 }
733 
734 static int spapr_xive_activate(SpaprInterruptController *intc,
735                                uint32_t nr_servers, Error **errp)
736 {
737     SpaprXive *xive = SPAPR_XIVE(intc);
738 
739     if (kvm_enabled()) {
740         int rc = spapr_irq_init_kvm(kvmppc_xive_connect, intc, nr_servers,
741                                     errp);
742         if (rc < 0) {
743             return rc;
744         }
745     }
746 
747     /* Activate the XIVE MMIOs */
748     spapr_xive_mmio_set_enabled(xive, true);
749 
750     return 0;
751 }
752 
753 static void spapr_xive_deactivate(SpaprInterruptController *intc)
754 {
755     SpaprXive *xive = SPAPR_XIVE(intc);
756 
757     spapr_xive_mmio_set_enabled(xive, false);
758 
759     if (kvm_irqchip_in_kernel()) {
760         kvmppc_xive_disconnect(intc);
761     }
762 }
763 
764 static void spapr_xive_class_init(ObjectClass *klass, void *data)
765 {
766     DeviceClass *dc = DEVICE_CLASS(klass);
767     XiveRouterClass *xrc = XIVE_ROUTER_CLASS(klass);
768     SpaprInterruptControllerClass *sicc = SPAPR_INTC_CLASS(klass);
769     XivePresenterClass *xpc = XIVE_PRESENTER_CLASS(klass);
770     SpaprXiveClass *sxc = SPAPR_XIVE_CLASS(klass);
771 
772     dc->desc    = "sPAPR XIVE Interrupt Controller";
773     device_class_set_props(dc, spapr_xive_properties);
774     device_class_set_parent_realize(dc, spapr_xive_realize,
775                                     &sxc->parent_realize);
776     dc->vmsd    = &vmstate_spapr_xive;
777 
778     xrc->get_eas = spapr_xive_get_eas;
779     xrc->get_end = spapr_xive_get_end;
780     xrc->write_end = spapr_xive_write_end;
781     xrc->get_nvt = spapr_xive_get_nvt;
782     xrc->write_nvt = spapr_xive_write_nvt;
783     xrc->get_block_id = spapr_xive_get_block_id;
784 
785     sicc->activate = spapr_xive_activate;
786     sicc->deactivate = spapr_xive_deactivate;
787     sicc->cpu_intc_create = spapr_xive_cpu_intc_create;
788     sicc->cpu_intc_reset = spapr_xive_cpu_intc_reset;
789     sicc->cpu_intc_destroy = spapr_xive_cpu_intc_destroy;
790     sicc->claim_irq = spapr_xive_claim_irq;
791     sicc->free_irq = spapr_xive_free_irq;
792     sicc->set_irq = spapr_xive_set_irq;
793     sicc->print_info = spapr_xive_print_info;
794     sicc->dt = spapr_xive_dt;
795     sicc->post_load = spapr_xive_post_load;
796 
797     xpc->match_nvt  = spapr_xive_match_nvt;
798 }
799 
800 static const TypeInfo spapr_xive_info = {
801     .name = TYPE_SPAPR_XIVE,
802     .parent = TYPE_XIVE_ROUTER,
803     .instance_init = spapr_xive_instance_init,
804     .instance_size = sizeof(SpaprXive),
805     .class_init = spapr_xive_class_init,
806     .class_size = sizeof(SpaprXiveClass),
807     .interfaces = (InterfaceInfo[]) {
808         { TYPE_SPAPR_INTC },
809         { }
810     },
811 };
812 
813 static void spapr_xive_register_types(void)
814 {
815     type_register_static(&spapr_xive_info);
816 }
817 
818 type_init(spapr_xive_register_types)
819 
820 /*
821  * XIVE hcalls
822  *
823  * The terminology used by the XIVE hcalls is the following :
824  *
825  *   TARGET vCPU number
826  *   EQ     Event Queue assigned by OS to receive event data
827  *   ESB    page for source interrupt management
828  *   LISN   Logical Interrupt Source Number identifying a source in the
829  *          machine
830  *   EISN   Effective Interrupt Source Number used by guest OS to
831  *          identify source in the guest
832  *
833  * The EAS, END, NVT structures are not exposed.
834  */
835 
836 /*
837  * Linux hosts under OPAL reserve priority 7 for their own escalation
838  * interrupts (DD2.X POWER9). So we only allow the guest to use
839  * priorities [0..6].
840  */
841 static bool spapr_xive_priority_is_reserved(uint8_t priority)
842 {
843     switch (priority) {
844     case 0 ... 6:
845         return false;
846     case 7: /* OPAL escalation queue */
847     default:
848         return true;
849     }
850 }
851 
852 /*
853  * The H_INT_GET_SOURCE_INFO hcall() is used to obtain the logical
854  * real address of the MMIO page through which the Event State Buffer
855  * entry associated with the value of the "lisn" parameter is managed.
856  *
857  * Parameters:
858  * Input
859  * - R4: "flags"
860  *         Bits 0-63 reserved
861  * - R5: "lisn" is per "interrupts", "interrupt-map", or
862  *       "ibm,xive-lisn-ranges" properties, or as returned by the
863  *       ibm,query-interrupt-source-number RTAS call, or as returned
864  *       by the H_ALLOCATE_VAS_WINDOW hcall
865  *
866  * Output
867  * - R4: "flags"
868  *         Bits 0-59: Reserved
869  *         Bit 60: H_INT_ESB must be used for Event State Buffer
870  *                 management
871  *         Bit 61: 1 == LSI  0 == MSI
872  *         Bit 62: the full function page supports trigger
873  *         Bit 63: Store EOI Supported
874  * - R5: Logical Real address of full function Event State Buffer
875  *       management page, -1 if H_INT_ESB hcall flag is set to 1.
876  * - R6: Logical Real Address of trigger only Event State Buffer
877  *       management page or -1.
878  * - R7: Power of 2 page size for the ESB management pages returned in
879  *       R5 and R6.
880  */
881 
882 #define SPAPR_XIVE_SRC_H_INT_ESB     PPC_BIT(60) /* ESB manage with H_INT_ESB */
883 #define SPAPR_XIVE_SRC_LSI           PPC_BIT(61) /* Virtual LSI type */
884 #define SPAPR_XIVE_SRC_TRIGGER       PPC_BIT(62) /* Trigger and management
885                                                     on same page */
886 #define SPAPR_XIVE_SRC_STORE_EOI     PPC_BIT(63) /* Store EOI support */
887 
888 static target_ulong h_int_get_source_info(PowerPCCPU *cpu,
889                                           SpaprMachineState *spapr,
890                                           target_ulong opcode,
891                                           target_ulong *args)
892 {
893     SpaprXive *xive = spapr->xive;
894     XiveSource *xsrc = &xive->source;
895     target_ulong flags  = args[0];
896     target_ulong lisn   = args[1];
897 
898     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
899         return H_FUNCTION;
900     }
901 
902     if (flags) {
903         return H_PARAMETER;
904     }
905 
906     if (lisn >= xive->nr_irqs) {
907         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
908                       lisn);
909         return H_P2;
910     }
911 
912     if (!xive_eas_is_valid(&xive->eat[lisn])) {
913         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
914                       lisn);
915         return H_P2;
916     }
917 
918     /*
919      * All sources are emulated under the main XIVE object and share
920      * the same characteristics.
921      */
922     args[0] = 0;
923     if (!xive_source_esb_has_2page(xsrc)) {
924         args[0] |= SPAPR_XIVE_SRC_TRIGGER;
925     }
926     if (xsrc->esb_flags & XIVE_SRC_STORE_EOI) {
927         args[0] |= SPAPR_XIVE_SRC_STORE_EOI;
928     }
929 
930     /*
931      * Force the use of the H_INT_ESB hcall in case of an LSI
932      * interrupt. This is necessary under KVM to re-trigger the
933      * interrupt if the level is still asserted
934      */
935     if (xive_source_irq_is_lsi(xsrc, lisn)) {
936         args[0] |= SPAPR_XIVE_SRC_H_INT_ESB | SPAPR_XIVE_SRC_LSI;
937     }
938 
939     if (!(args[0] & SPAPR_XIVE_SRC_H_INT_ESB)) {
940         args[1] = xive->vc_base + xive_source_esb_mgmt(xsrc, lisn);
941     } else {
942         args[1] = -1;
943     }
944 
945     if (xive_source_esb_has_2page(xsrc) &&
946         !(args[0] & SPAPR_XIVE_SRC_H_INT_ESB)) {
947         args[2] = xive->vc_base + xive_source_esb_page(xsrc, lisn);
948     } else {
949         args[2] = -1;
950     }
951 
952     if (xive_source_esb_has_2page(xsrc)) {
953         args[3] = xsrc->esb_shift - 1;
954     } else {
955         args[3] = xsrc->esb_shift;
956     }
957 
958     return H_SUCCESS;
959 }
960 
961 /*
962  * The H_INT_SET_SOURCE_CONFIG hcall() is used to assign a Logical
963  * Interrupt Source to a target. The Logical Interrupt Source is
964  * designated with the "lisn" parameter and the target is designated
965  * with the "target" and "priority" parameters.  Upon return from the
966  * hcall(), no additional interrupts will be directed to the old EQ.
967  *
968  * Parameters:
969  * Input:
970  * - R4: "flags"
971  *         Bits 0-61: Reserved
972  *         Bit 62: set the "eisn" in the EAS
973  *         Bit 63: masks the interrupt source in the hardware interrupt
974  *       control structure. An interrupt masked by this mechanism will
975  *       be dropped, but it's source state bits will still be
976  *       set. There is no race-free way of unmasking and restoring the
977  *       source. Thus this should only be used in interrupts that are
978  *       also masked at the source, and only in cases where the
979  *       interrupt is not meant to be used for a large amount of time
980  *       because no valid target exists for it for example
981  * - R5: "lisn" is per "interrupts", "interrupt-map", or
982  *       "ibm,xive-lisn-ranges" properties, or as returned by the
983  *       ibm,query-interrupt-source-number RTAS call, or as returned by
984  *       the H_ALLOCATE_VAS_WINDOW hcall
985  * - R6: "target" is per "ibm,ppc-interrupt-server#s" or
986  *       "ibm,ppc-interrupt-gserver#s"
987  * - R7: "priority" is a valid priority not in
988  *       "ibm,plat-res-int-priorities"
989  * - R8: "eisn" is the guest EISN associated with the "lisn"
990  *
991  * Output:
992  * - None
993  */
994 
995 #define SPAPR_XIVE_SRC_SET_EISN PPC_BIT(62)
996 #define SPAPR_XIVE_SRC_MASK     PPC_BIT(63)
997 
998 static target_ulong h_int_set_source_config(PowerPCCPU *cpu,
999                                             SpaprMachineState *spapr,
1000                                             target_ulong opcode,
1001                                             target_ulong *args)
1002 {
1003     SpaprXive *xive = spapr->xive;
1004     XiveEAS eas, new_eas;
1005     target_ulong flags    = args[0];
1006     target_ulong lisn     = args[1];
1007     target_ulong target   = args[2];
1008     target_ulong priority = args[3];
1009     target_ulong eisn     = args[4];
1010     uint8_t end_blk;
1011     uint32_t end_idx;
1012 
1013     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1014         return H_FUNCTION;
1015     }
1016 
1017     if (flags & ~(SPAPR_XIVE_SRC_SET_EISN | SPAPR_XIVE_SRC_MASK)) {
1018         return H_PARAMETER;
1019     }
1020 
1021     if (lisn >= xive->nr_irqs) {
1022         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1023                       lisn);
1024         return H_P2;
1025     }
1026 
1027     eas = xive->eat[lisn];
1028     if (!xive_eas_is_valid(&eas)) {
1029         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1030                       lisn);
1031         return H_P2;
1032     }
1033 
1034     /* priority 0xff is used to reset the EAS */
1035     if (priority == 0xff) {
1036         new_eas.w = cpu_to_be64(EAS_VALID | EAS_MASKED);
1037         goto out;
1038     }
1039 
1040     if (flags & SPAPR_XIVE_SRC_MASK) {
1041         new_eas.w = eas.w | cpu_to_be64(EAS_MASKED);
1042     } else {
1043         new_eas.w = eas.w & cpu_to_be64(~EAS_MASKED);
1044     }
1045 
1046     if (spapr_xive_priority_is_reserved(priority)) {
1047         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1048                       " is reserved\n", priority);
1049         return H_P4;
1050     }
1051 
1052     /*
1053      * Validate that "target" is part of the list of threads allocated
1054      * to the partition. For that, find the END corresponding to the
1055      * target.
1056      */
1057     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1058         return H_P3;
1059     }
1060 
1061     new_eas.w = xive_set_field64(EAS_END_BLOCK, new_eas.w, end_blk);
1062     new_eas.w = xive_set_field64(EAS_END_INDEX, new_eas.w, end_idx);
1063 
1064     if (flags & SPAPR_XIVE_SRC_SET_EISN) {
1065         new_eas.w = xive_set_field64(EAS_END_DATA, new_eas.w, eisn);
1066     }
1067 
1068     if (kvm_irqchip_in_kernel()) {
1069         Error *local_err = NULL;
1070 
1071         kvmppc_xive_set_source_config(xive, lisn, &new_eas, &local_err);
1072         if (local_err) {
1073             error_report_err(local_err);
1074             return H_HARDWARE;
1075         }
1076     }
1077 
1078 out:
1079     xive->eat[lisn] = new_eas;
1080     return H_SUCCESS;
1081 }
1082 
1083 /*
1084  * The H_INT_GET_SOURCE_CONFIG hcall() is used to determine to which
1085  * target/priority pair is assigned to the specified Logical Interrupt
1086  * Source.
1087  *
1088  * Parameters:
1089  * Input:
1090  * - R4: "flags"
1091  *         Bits 0-63 Reserved
1092  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1093  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1094  *       ibm,query-interrupt-source-number RTAS call, or as
1095  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1096  *
1097  * Output:
1098  * - R4: Target to which the specified Logical Interrupt Source is
1099  *       assigned
1100  * - R5: Priority to which the specified Logical Interrupt Source is
1101  *       assigned
1102  * - R6: EISN for the specified Logical Interrupt Source (this will be
1103  *       equivalent to the LISN if not changed by H_INT_SET_SOURCE_CONFIG)
1104  */
1105 static target_ulong h_int_get_source_config(PowerPCCPU *cpu,
1106                                             SpaprMachineState *spapr,
1107                                             target_ulong opcode,
1108                                             target_ulong *args)
1109 {
1110     SpaprXive *xive = spapr->xive;
1111     target_ulong flags = args[0];
1112     target_ulong lisn = args[1];
1113     XiveEAS eas;
1114     XiveEND *end;
1115     uint8_t nvt_blk;
1116     uint32_t end_idx, nvt_idx;
1117 
1118     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1119         return H_FUNCTION;
1120     }
1121 
1122     if (flags) {
1123         return H_PARAMETER;
1124     }
1125 
1126     if (lisn >= xive->nr_irqs) {
1127         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1128                       lisn);
1129         return H_P2;
1130     }
1131 
1132     eas = xive->eat[lisn];
1133     if (!xive_eas_is_valid(&eas)) {
1134         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1135                       lisn);
1136         return H_P2;
1137     }
1138 
1139     /* EAS_END_BLOCK is unused on sPAPR */
1140     end_idx = xive_get_field64(EAS_END_INDEX, eas.w);
1141 
1142     assert(end_idx < xive->nr_ends);
1143     end = &xive->endt[end_idx];
1144 
1145     nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end->w6);
1146     nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end->w6);
1147     args[0] = spapr_xive_nvt_to_target(nvt_blk, nvt_idx);
1148 
1149     if (xive_eas_is_masked(&eas)) {
1150         args[1] = 0xff;
1151     } else {
1152         args[1] = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
1153     }
1154 
1155     args[2] = xive_get_field64(EAS_END_DATA, eas.w);
1156 
1157     return H_SUCCESS;
1158 }
1159 
1160 /*
1161  * The H_INT_GET_QUEUE_INFO hcall() is used to get the logical real
1162  * address of the notification management page associated with the
1163  * specified target and priority.
1164  *
1165  * Parameters:
1166  * Input:
1167  * - R4: "flags"
1168  *         Bits 0-63 Reserved
1169  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1170  *       "ibm,ppc-interrupt-gserver#s"
1171  * - R6: "priority" is a valid priority not in
1172  *       "ibm,plat-res-int-priorities"
1173  *
1174  * Output:
1175  * - R4: Logical real address of notification page
1176  * - R5: Power of 2 page size of the notification page
1177  */
1178 static target_ulong h_int_get_queue_info(PowerPCCPU *cpu,
1179                                          SpaprMachineState *spapr,
1180                                          target_ulong opcode,
1181                                          target_ulong *args)
1182 {
1183     SpaprXive *xive = spapr->xive;
1184     XiveENDSource *end_xsrc = &xive->end_source;
1185     target_ulong flags = args[0];
1186     target_ulong target = args[1];
1187     target_ulong priority = args[2];
1188     XiveEND *end;
1189     uint8_t end_blk;
1190     uint32_t end_idx;
1191 
1192     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1193         return H_FUNCTION;
1194     }
1195 
1196     if (flags) {
1197         return H_PARAMETER;
1198     }
1199 
1200     /*
1201      * H_STATE should be returned if a H_INT_RESET is in progress.
1202      * This is not needed when running the emulation under QEMU
1203      */
1204 
1205     if (spapr_xive_priority_is_reserved(priority)) {
1206         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1207                       " is reserved\n", priority);
1208         return H_P3;
1209     }
1210 
1211     /*
1212      * Validate that "target" is part of the list of threads allocated
1213      * to the partition. For that, find the END corresponding to the
1214      * target.
1215      */
1216     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1217         return H_P2;
1218     }
1219 
1220     assert(end_idx < xive->nr_ends);
1221     end = &xive->endt[end_idx];
1222 
1223     args[0] = xive->end_base + (1ull << (end_xsrc->esb_shift + 1)) * end_idx;
1224     if (xive_end_is_enqueue(end)) {
1225         args[1] = xive_get_field32(END_W0_QSIZE, end->w0) + 12;
1226     } else {
1227         args[1] = 0;
1228     }
1229 
1230     return H_SUCCESS;
1231 }
1232 
1233 /*
1234  * The H_INT_SET_QUEUE_CONFIG hcall() is used to set or reset a EQ for
1235  * a given "target" and "priority".  It is also used to set the
1236  * notification config associated with the EQ.  An EQ size of 0 is
1237  * used to reset the EQ config for a given target and priority. If
1238  * resetting the EQ config, the END associated with the given "target"
1239  * and "priority" will be changed to disable queueing.
1240  *
1241  * Upon return from the hcall(), no additional interrupts will be
1242  * directed to the old EQ (if one was set). The old EQ (if one was
1243  * set) should be investigated for interrupts that occurred prior to
1244  * or during the hcall().
1245  *
1246  * Parameters:
1247  * Input:
1248  * - R4: "flags"
1249  *         Bits 0-62: Reserved
1250  *         Bit 63: Unconditional Notify (n) per the XIVE spec
1251  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1252  *       "ibm,ppc-interrupt-gserver#s"
1253  * - R6: "priority" is a valid priority not in
1254  *       "ibm,plat-res-int-priorities"
1255  * - R7: "eventQueue": The logical real address of the start of the EQ
1256  * - R8: "eventQueueSize": The power of 2 EQ size per "ibm,xive-eq-sizes"
1257  *
1258  * Output:
1259  * - None
1260  */
1261 
1262 #define SPAPR_XIVE_END_ALWAYS_NOTIFY PPC_BIT(63)
1263 
1264 static target_ulong h_int_set_queue_config(PowerPCCPU *cpu,
1265                                            SpaprMachineState *spapr,
1266                                            target_ulong opcode,
1267                                            target_ulong *args)
1268 {
1269     SpaprXive *xive = spapr->xive;
1270     target_ulong flags = args[0];
1271     target_ulong target = args[1];
1272     target_ulong priority = args[2];
1273     target_ulong qpage = args[3];
1274     target_ulong qsize = args[4];
1275     XiveEND end;
1276     uint8_t end_blk, nvt_blk;
1277     uint32_t end_idx, nvt_idx;
1278 
1279     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1280         return H_FUNCTION;
1281     }
1282 
1283     if (flags & ~SPAPR_XIVE_END_ALWAYS_NOTIFY) {
1284         return H_PARAMETER;
1285     }
1286 
1287     /*
1288      * H_STATE should be returned if a H_INT_RESET is in progress.
1289      * This is not needed when running the emulation under QEMU
1290      */
1291 
1292     if (spapr_xive_priority_is_reserved(priority)) {
1293         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1294                       " is reserved\n", priority);
1295         return H_P3;
1296     }
1297 
1298     /*
1299      * Validate that "target" is part of the list of threads allocated
1300      * to the partition. For that, find the END corresponding to the
1301      * target.
1302      */
1303 
1304     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1305         return H_P2;
1306     }
1307 
1308     assert(end_idx < xive->nr_ends);
1309     memcpy(&end, &xive->endt[end_idx], sizeof(XiveEND));
1310 
1311     switch (qsize) {
1312     case 12:
1313     case 16:
1314     case 21:
1315     case 24:
1316         if (!QEMU_IS_ALIGNED(qpage, 1ul << qsize)) {
1317             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: EQ @0x%" HWADDR_PRIx
1318                           " is not naturally aligned with %" HWADDR_PRIx "\n",
1319                           qpage, (hwaddr)1 << qsize);
1320             return H_P4;
1321         }
1322         end.w2 = cpu_to_be32((qpage >> 32) & 0x0fffffff);
1323         end.w3 = cpu_to_be32(qpage & 0xffffffff);
1324         end.w0 |= cpu_to_be32(END_W0_ENQUEUE);
1325         end.w0 = xive_set_field32(END_W0_QSIZE, end.w0, qsize - 12);
1326         break;
1327     case 0:
1328         /* reset queue and disable queueing */
1329         spapr_xive_end_reset(&end);
1330         goto out;
1331 
1332     default:
1333         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid EQ size %"PRIx64"\n",
1334                       qsize);
1335         return H_P5;
1336     }
1337 
1338     if (qsize) {
1339         hwaddr plen = 1 << qsize;
1340         void *eq;
1341 
1342         /*
1343          * Validate the guest EQ. We should also check that the queue
1344          * has been zeroed by the OS.
1345          */
1346         eq = address_space_map(CPU(cpu)->as, qpage, &plen, true,
1347                                MEMTXATTRS_UNSPECIFIED);
1348         if (plen != 1 << qsize) {
1349             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to map EQ @0x%"
1350                           HWADDR_PRIx "\n", qpage);
1351             return H_P4;
1352         }
1353         address_space_unmap(CPU(cpu)->as, eq, plen, true, plen);
1354     }
1355 
1356     /* "target" should have been validated above */
1357     if (spapr_xive_target_to_nvt(target, &nvt_blk, &nvt_idx)) {
1358         g_assert_not_reached();
1359     }
1360 
1361     /*
1362      * Ensure the priority and target are correctly set (they will not
1363      * be right after allocation)
1364      */
1365     end.w6 = xive_set_field32(END_W6_NVT_BLOCK, 0ul, nvt_blk) |
1366         xive_set_field32(END_W6_NVT_INDEX, 0ul, nvt_idx);
1367     end.w7 = xive_set_field32(END_W7_F0_PRIORITY, 0ul, priority);
1368 
1369     if (flags & SPAPR_XIVE_END_ALWAYS_NOTIFY) {
1370         end.w0 |= cpu_to_be32(END_W0_UCOND_NOTIFY);
1371     } else {
1372         end.w0 &= cpu_to_be32((uint32_t)~END_W0_UCOND_NOTIFY);
1373     }
1374 
1375     /*
1376      * The generation bit for the END starts at 1 and The END page
1377      * offset counter starts at 0.
1378      */
1379     end.w1 = cpu_to_be32(END_W1_GENERATION) |
1380         xive_set_field32(END_W1_PAGE_OFF, 0ul, 0ul);
1381     end.w0 |= cpu_to_be32(END_W0_VALID);
1382 
1383     /*
1384      * TODO: issue syncs required to ensure all in-flight interrupts
1385      * are complete on the old END
1386      */
1387 
1388 out:
1389     if (kvm_irqchip_in_kernel()) {
1390         Error *local_err = NULL;
1391 
1392         kvmppc_xive_set_queue_config(xive, end_blk, end_idx, &end, &local_err);
1393         if (local_err) {
1394             error_report_err(local_err);
1395             return H_HARDWARE;
1396         }
1397     }
1398 
1399     /* Update END */
1400     memcpy(&xive->endt[end_idx], &end, sizeof(XiveEND));
1401     return H_SUCCESS;
1402 }
1403 
1404 /*
1405  * The H_INT_GET_QUEUE_CONFIG hcall() is used to get a EQ for a given
1406  * target and priority.
1407  *
1408  * Parameters:
1409  * Input:
1410  * - R4: "flags"
1411  *         Bits 0-62: Reserved
1412  *         Bit 63: Debug: Return debug data
1413  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1414  *       "ibm,ppc-interrupt-gserver#s"
1415  * - R6: "priority" is a valid priority not in
1416  *       "ibm,plat-res-int-priorities"
1417  *
1418  * Output:
1419  * - R4: "flags":
1420  *       Bits 0-61: Reserved
1421  *       Bit 62: The value of Event Queue Generation Number (g) per
1422  *              the XIVE spec if "Debug" = 1
1423  *       Bit 63: The value of Unconditional Notify (n) per the XIVE spec
1424  * - R5: The logical real address of the start of the EQ
1425  * - R6: The power of 2 EQ size per "ibm,xive-eq-sizes"
1426  * - R7: The value of Event Queue Offset Counter per XIVE spec
1427  *       if "Debug" = 1, else 0
1428  *
1429  */
1430 
1431 #define SPAPR_XIVE_END_DEBUG     PPC_BIT(63)
1432 
1433 static target_ulong h_int_get_queue_config(PowerPCCPU *cpu,
1434                                            SpaprMachineState *spapr,
1435                                            target_ulong opcode,
1436                                            target_ulong *args)
1437 {
1438     SpaprXive *xive = spapr->xive;
1439     target_ulong flags = args[0];
1440     target_ulong target = args[1];
1441     target_ulong priority = args[2];
1442     XiveEND *end;
1443     uint8_t end_blk;
1444     uint32_t end_idx;
1445 
1446     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1447         return H_FUNCTION;
1448     }
1449 
1450     if (flags & ~SPAPR_XIVE_END_DEBUG) {
1451         return H_PARAMETER;
1452     }
1453 
1454     /*
1455      * H_STATE should be returned if a H_INT_RESET is in progress.
1456      * This is not needed when running the emulation under QEMU
1457      */
1458 
1459     if (spapr_xive_priority_is_reserved(priority)) {
1460         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1461                       " is reserved\n", priority);
1462         return H_P3;
1463     }
1464 
1465     /*
1466      * Validate that "target" is part of the list of threads allocated
1467      * to the partition. For that, find the END corresponding to the
1468      * target.
1469      */
1470     if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1471         return H_P2;
1472     }
1473 
1474     assert(end_idx < xive->nr_ends);
1475     end = &xive->endt[end_idx];
1476 
1477     args[0] = 0;
1478     if (xive_end_is_notify(end)) {
1479         args[0] |= SPAPR_XIVE_END_ALWAYS_NOTIFY;
1480     }
1481 
1482     if (xive_end_is_enqueue(end)) {
1483         args[1] = xive_end_qaddr(end);
1484         args[2] = xive_get_field32(END_W0_QSIZE, end->w0) + 12;
1485     } else {
1486         args[1] = 0;
1487         args[2] = 0;
1488     }
1489 
1490     if (kvm_irqchip_in_kernel()) {
1491         Error *local_err = NULL;
1492 
1493         kvmppc_xive_get_queue_config(xive, end_blk, end_idx, end, &local_err);
1494         if (local_err) {
1495             error_report_err(local_err);
1496             return H_HARDWARE;
1497         }
1498     }
1499 
1500     /* TODO: do we need any locking on the END ? */
1501     if (flags & SPAPR_XIVE_END_DEBUG) {
1502         /* Load the event queue generation number into the return flags */
1503         args[0] |= (uint64_t)xive_get_field32(END_W1_GENERATION, end->w1) << 62;
1504 
1505         /* Load R7 with the event queue offset counter */
1506         args[3] = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1507     } else {
1508         args[3] = 0;
1509     }
1510 
1511     return H_SUCCESS;
1512 }
1513 
1514 /*
1515  * The H_INT_SET_OS_REPORTING_LINE hcall() is used to set the
1516  * reporting cache line pair for the calling thread.  The reporting
1517  * cache lines will contain the OS interrupt context when the OS
1518  * issues a CI store byte to @TIMA+0xC10 to acknowledge the OS
1519  * interrupt. The reporting cache lines can be reset by inputting -1
1520  * in "reportingLine".  Issuing the CI store byte without reporting
1521  * cache lines registered will result in the data not being accessible
1522  * to the OS.
1523  *
1524  * Parameters:
1525  * Input:
1526  * - R4: "flags"
1527  *         Bits 0-63: Reserved
1528  * - R5: "reportingLine": The logical real address of the reporting cache
1529  *       line pair
1530  *
1531  * Output:
1532  * - None
1533  */
1534 static target_ulong h_int_set_os_reporting_line(PowerPCCPU *cpu,
1535                                                 SpaprMachineState *spapr,
1536                                                 target_ulong opcode,
1537                                                 target_ulong *args)
1538 {
1539     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1540         return H_FUNCTION;
1541     }
1542 
1543     /*
1544      * H_STATE should be returned if a H_INT_RESET is in progress.
1545      * This is not needed when running the emulation under QEMU
1546      */
1547 
1548     /* TODO: H_INT_SET_OS_REPORTING_LINE */
1549     return H_FUNCTION;
1550 }
1551 
1552 /*
1553  * The H_INT_GET_OS_REPORTING_LINE hcall() is used to get the logical
1554  * real address of the reporting cache line pair set for the input
1555  * "target".  If no reporting cache line pair has been set, -1 is
1556  * returned.
1557  *
1558  * Parameters:
1559  * Input:
1560  * - R4: "flags"
1561  *         Bits 0-63: Reserved
1562  * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1563  *       "ibm,ppc-interrupt-gserver#s"
1564  * - R6: "reportingLine": The logical real address of the reporting
1565  *        cache line pair
1566  *
1567  * Output:
1568  * - R4: The logical real address of the reporting line if set, else -1
1569  */
1570 static target_ulong h_int_get_os_reporting_line(PowerPCCPU *cpu,
1571                                                 SpaprMachineState *spapr,
1572                                                 target_ulong opcode,
1573                                                 target_ulong *args)
1574 {
1575     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1576         return H_FUNCTION;
1577     }
1578 
1579     /*
1580      * H_STATE should be returned if a H_INT_RESET is in progress.
1581      * This is not needed when running the emulation under QEMU
1582      */
1583 
1584     /* TODO: H_INT_GET_OS_REPORTING_LINE */
1585     return H_FUNCTION;
1586 }
1587 
1588 /*
1589  * The H_INT_ESB hcall() is used to issue a load or store to the ESB
1590  * page for the input "lisn".  This hcall is only supported for LISNs
1591  * that have the ESB hcall flag set to 1 when returned from hcall()
1592  * H_INT_GET_SOURCE_INFO.
1593  *
1594  * Parameters:
1595  * Input:
1596  * - R4: "flags"
1597  *         Bits 0-62: Reserved
1598  *         bit 63: Store: Store=1, store operation, else load operation
1599  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1600  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1601  *       ibm,query-interrupt-source-number RTAS call, or as
1602  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1603  * - R6: "esbOffset" is the offset into the ESB page for the load or
1604  *       store operation
1605  * - R7: "storeData" is the data to write for a store operation
1606  *
1607  * Output:
1608  * - R4: The value of the load if load operation, else -1
1609  */
1610 
1611 #define SPAPR_XIVE_ESB_STORE PPC_BIT(63)
1612 
1613 static target_ulong h_int_esb(PowerPCCPU *cpu,
1614                               SpaprMachineState *spapr,
1615                               target_ulong opcode,
1616                               target_ulong *args)
1617 {
1618     SpaprXive *xive = spapr->xive;
1619     XiveEAS eas;
1620     target_ulong flags  = args[0];
1621     target_ulong lisn   = args[1];
1622     target_ulong offset = args[2];
1623     target_ulong data   = args[3];
1624     hwaddr mmio_addr;
1625     XiveSource *xsrc = &xive->source;
1626 
1627     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1628         return H_FUNCTION;
1629     }
1630 
1631     if (flags & ~SPAPR_XIVE_ESB_STORE) {
1632         return H_PARAMETER;
1633     }
1634 
1635     if (lisn >= xive->nr_irqs) {
1636         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1637                       lisn);
1638         return H_P2;
1639     }
1640 
1641     eas = xive->eat[lisn];
1642     if (!xive_eas_is_valid(&eas)) {
1643         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1644                       lisn);
1645         return H_P2;
1646     }
1647 
1648     if (offset > (1ull << xsrc->esb_shift)) {
1649         return H_P3;
1650     }
1651 
1652     if (kvm_irqchip_in_kernel()) {
1653         args[0] = kvmppc_xive_esb_rw(xsrc, lisn, offset, data,
1654                                      flags & SPAPR_XIVE_ESB_STORE);
1655     } else {
1656         mmio_addr = xive->vc_base + xive_source_esb_mgmt(xsrc, lisn) + offset;
1657 
1658         if (dma_memory_rw(&address_space_memory, mmio_addr, &data, 8,
1659                           (flags & SPAPR_XIVE_ESB_STORE))) {
1660             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to access ESB @0x%"
1661                           HWADDR_PRIx "\n", mmio_addr);
1662             return H_HARDWARE;
1663         }
1664         args[0] = (flags & SPAPR_XIVE_ESB_STORE) ? -1 : data;
1665     }
1666     return H_SUCCESS;
1667 }
1668 
1669 /*
1670  * The H_INT_SYNC hcall() is used to issue hardware syncs that will
1671  * ensure any in flight events for the input lisn are in the event
1672  * queue.
1673  *
1674  * Parameters:
1675  * Input:
1676  * - R4: "flags"
1677  *         Bits 0-63: Reserved
1678  * - R5: "lisn" is per "interrupts", "interrupt-map", or
1679  *       "ibm,xive-lisn-ranges" properties, or as returned by the
1680  *       ibm,query-interrupt-source-number RTAS call, or as
1681  *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1682  *
1683  * Output:
1684  * - None
1685  */
1686 static target_ulong h_int_sync(PowerPCCPU *cpu,
1687                                SpaprMachineState *spapr,
1688                                target_ulong opcode,
1689                                target_ulong *args)
1690 {
1691     SpaprXive *xive = spapr->xive;
1692     XiveEAS eas;
1693     target_ulong flags = args[0];
1694     target_ulong lisn = args[1];
1695 
1696     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1697         return H_FUNCTION;
1698     }
1699 
1700     if (flags) {
1701         return H_PARAMETER;
1702     }
1703 
1704     if (lisn >= xive->nr_irqs) {
1705         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1706                       lisn);
1707         return H_P2;
1708     }
1709 
1710     eas = xive->eat[lisn];
1711     if (!xive_eas_is_valid(&eas)) {
1712         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1713                       lisn);
1714         return H_P2;
1715     }
1716 
1717     /*
1718      * H_STATE should be returned if a H_INT_RESET is in progress.
1719      * This is not needed when running the emulation under QEMU
1720      */
1721 
1722     /*
1723      * This is not real hardware. Nothing to be done unless when
1724      * under KVM
1725      */
1726 
1727     if (kvm_irqchip_in_kernel()) {
1728         Error *local_err = NULL;
1729 
1730         kvmppc_xive_sync_source(xive, lisn, &local_err);
1731         if (local_err) {
1732             error_report_err(local_err);
1733             return H_HARDWARE;
1734         }
1735     }
1736     return H_SUCCESS;
1737 }
1738 
1739 /*
1740  * The H_INT_RESET hcall() is used to reset all of the partition's
1741  * interrupt exploitation structures to their initial state.  This
1742  * means losing all previously set interrupt state set via
1743  * H_INT_SET_SOURCE_CONFIG and H_INT_SET_QUEUE_CONFIG.
1744  *
1745  * Parameters:
1746  * Input:
1747  * - R4: "flags"
1748  *         Bits 0-63: Reserved
1749  *
1750  * Output:
1751  * - None
1752  */
1753 static target_ulong h_int_reset(PowerPCCPU *cpu,
1754                                 SpaprMachineState *spapr,
1755                                 target_ulong opcode,
1756                                 target_ulong *args)
1757 {
1758     SpaprXive *xive = spapr->xive;
1759     target_ulong flags   = args[0];
1760 
1761     if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1762         return H_FUNCTION;
1763     }
1764 
1765     if (flags) {
1766         return H_PARAMETER;
1767     }
1768 
1769     device_legacy_reset(DEVICE(xive));
1770 
1771     if (kvm_irqchip_in_kernel()) {
1772         Error *local_err = NULL;
1773 
1774         kvmppc_xive_reset(xive, &local_err);
1775         if (local_err) {
1776             error_report_err(local_err);
1777             return H_HARDWARE;
1778         }
1779     }
1780     return H_SUCCESS;
1781 }
1782 
1783 void spapr_xive_hcall_init(SpaprMachineState *spapr)
1784 {
1785     spapr_register_hypercall(H_INT_GET_SOURCE_INFO, h_int_get_source_info);
1786     spapr_register_hypercall(H_INT_SET_SOURCE_CONFIG, h_int_set_source_config);
1787     spapr_register_hypercall(H_INT_GET_SOURCE_CONFIG, h_int_get_source_config);
1788     spapr_register_hypercall(H_INT_GET_QUEUE_INFO, h_int_get_queue_info);
1789     spapr_register_hypercall(H_INT_SET_QUEUE_CONFIG, h_int_set_queue_config);
1790     spapr_register_hypercall(H_INT_GET_QUEUE_CONFIG, h_int_get_queue_config);
1791     spapr_register_hypercall(H_INT_SET_OS_REPORTING_LINE,
1792                              h_int_set_os_reporting_line);
1793     spapr_register_hypercall(H_INT_GET_OS_REPORTING_LINE,
1794                              h_int_get_os_reporting_line);
1795     spapr_register_hypercall(H_INT_ESB, h_int_esb);
1796     spapr_register_hypercall(H_INT_SYNC, h_int_sync);
1797     spapr_register_hypercall(H_INT_RESET, h_int_reset);
1798 }
1799