xref: /qemu/hw/intc/xive.c (revision 09a274d8)
1 /*
2  * QEMU PowerPC 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 "qapi/error.h"
13 #include "target/ppc/cpu.h"
14 #include "sysemu/cpus.h"
15 #include "sysemu/dma.h"
16 #include "hw/qdev-properties.h"
17 #include "monitor/monitor.h"
18 #include "hw/ppc/xive.h"
19 #include "hw/ppc/xive_regs.h"
20 
21 /*
22  * XIVE Thread Interrupt Management context
23  */
24 
25 /*
26  * Convert a priority number to an Interrupt Pending Buffer (IPB)
27  * register, which indicates a pending interrupt at the priority
28  * corresponding to the bit number
29  */
30 static uint8_t priority_to_ipb(uint8_t priority)
31 {
32     return priority > XIVE_PRIORITY_MAX ?
33         0 : 1 << (XIVE_PRIORITY_MAX - priority);
34 }
35 
36 /*
37  * Convert an Interrupt Pending Buffer (IPB) register to a Pending
38  * Interrupt Priority Register (PIPR), which contains the priority of
39  * the most favored pending notification.
40  */
41 static uint8_t ipb_to_pipr(uint8_t ibp)
42 {
43     return ibp ? clz32((uint32_t)ibp << 24) : 0xff;
44 }
45 
46 static void ipb_update(uint8_t *regs, uint8_t priority)
47 {
48     regs[TM_IPB] |= priority_to_ipb(priority);
49     regs[TM_PIPR] = ipb_to_pipr(regs[TM_IPB]);
50 }
51 
52 static uint8_t exception_mask(uint8_t ring)
53 {
54     switch (ring) {
55     case TM_QW1_OS:
56         return TM_QW1_NSR_EO;
57     default:
58         g_assert_not_reached();
59     }
60 }
61 
62 static uint64_t xive_tctx_accept(XiveTCTX *tctx, uint8_t ring)
63 {
64     uint8_t *regs = &tctx->regs[ring];
65     uint8_t nsr = regs[TM_NSR];
66     uint8_t mask = exception_mask(ring);
67 
68     qemu_irq_lower(tctx->output);
69 
70     if (regs[TM_NSR] & mask) {
71         uint8_t cppr = regs[TM_PIPR];
72 
73         regs[TM_CPPR] = cppr;
74 
75         /* Reset the pending buffer bit */
76         regs[TM_IPB] &= ~priority_to_ipb(cppr);
77         regs[TM_PIPR] = ipb_to_pipr(regs[TM_IPB]);
78 
79         /* Drop Exception bit */
80         regs[TM_NSR] &= ~mask;
81     }
82 
83     return (nsr << 8) | regs[TM_CPPR];
84 }
85 
86 static void xive_tctx_notify(XiveTCTX *tctx, uint8_t ring)
87 {
88     uint8_t *regs = &tctx->regs[ring];
89 
90     if (regs[TM_PIPR] < regs[TM_CPPR]) {
91         regs[TM_NSR] |= exception_mask(ring);
92         qemu_irq_raise(tctx->output);
93     }
94 }
95 
96 static void xive_tctx_set_cppr(XiveTCTX *tctx, uint8_t ring, uint8_t cppr)
97 {
98     if (cppr > XIVE_PRIORITY_MAX) {
99         cppr = 0xff;
100     }
101 
102     tctx->regs[ring + TM_CPPR] = cppr;
103 
104     /* CPPR has changed, check if we need to raise a pending exception */
105     xive_tctx_notify(tctx, ring);
106 }
107 
108 /*
109  * XIVE Thread Interrupt Management Area (TIMA)
110  */
111 
112 /*
113  * Define an access map for each page of the TIMA that we will use in
114  * the memory region ops to filter values when doing loads and stores
115  * of raw registers values
116  *
117  * Registers accessibility bits :
118  *
119  *    0x0 - no access
120  *    0x1 - write only
121  *    0x2 - read only
122  *    0x3 - read/write
123  */
124 
125 static const uint8_t xive_tm_hw_view[] = {
126     /* QW-0 User */   3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0,
127     /* QW-1 OS   */   3, 3, 3, 3,   3, 3, 0, 3,   3, 3, 3, 3,   0, 0, 0, 0,
128     /* QW-2 POOL */   0, 0, 3, 3,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0,
129     /* QW-3 PHYS */   3, 3, 3, 3,   0, 3, 0, 3,   3, 0, 0, 3,   3, 3, 3, 0,
130 };
131 
132 static const uint8_t xive_tm_hv_view[] = {
133     /* QW-0 User */   3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0,
134     /* QW-1 OS   */   3, 3, 3, 3,   3, 3, 0, 3,   3, 3, 3, 3,   0, 0, 0, 0,
135     /* QW-2 POOL */   0, 0, 3, 3,   0, 0, 0, 0,   0, 3, 3, 3,   0, 0, 0, 0,
136     /* QW-3 PHYS */   3, 3, 3, 3,   0, 3, 0, 3,   3, 0, 0, 3,   0, 0, 0, 0,
137 };
138 
139 static const uint8_t xive_tm_os_view[] = {
140     /* QW-0 User */   3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0,
141     /* QW-1 OS   */   2, 3, 2, 2,   2, 2, 0, 2,   0, 0, 0, 0,   0, 0, 0, 0,
142     /* QW-2 POOL */   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,
143     /* QW-3 PHYS */   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,
144 };
145 
146 static const uint8_t xive_tm_user_view[] = {
147     /* QW-0 User */   3, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,
148     /* QW-1 OS   */   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,
149     /* QW-2 POOL */   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,
150     /* QW-3 PHYS */   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,
151 };
152 
153 /*
154  * Overall TIMA access map for the thread interrupt management context
155  * registers
156  */
157 static const uint8_t *xive_tm_views[] = {
158     [XIVE_TM_HW_PAGE]   = xive_tm_hw_view,
159     [XIVE_TM_HV_PAGE]   = xive_tm_hv_view,
160     [XIVE_TM_OS_PAGE]   = xive_tm_os_view,
161     [XIVE_TM_USER_PAGE] = xive_tm_user_view,
162 };
163 
164 /*
165  * Computes a register access mask for a given offset in the TIMA
166  */
167 static uint64_t xive_tm_mask(hwaddr offset, unsigned size, bool write)
168 {
169     uint8_t page_offset = (offset >> TM_SHIFT) & 0x3;
170     uint8_t reg_offset = offset & 0x3F;
171     uint8_t reg_mask = write ? 0x1 : 0x2;
172     uint64_t mask = 0x0;
173     int i;
174 
175     for (i = 0; i < size; i++) {
176         if (xive_tm_views[page_offset][reg_offset + i] & reg_mask) {
177             mask |= (uint64_t) 0xff << (8 * (size - i - 1));
178         }
179     }
180 
181     return mask;
182 }
183 
184 static void xive_tm_raw_write(XiveTCTX *tctx, hwaddr offset, uint64_t value,
185                               unsigned size)
186 {
187     uint8_t ring_offset = offset & 0x30;
188     uint8_t reg_offset = offset & 0x3F;
189     uint64_t mask = xive_tm_mask(offset, size, true);
190     int i;
191 
192     /*
193      * Only 4 or 8 bytes stores are allowed and the User ring is
194      * excluded
195      */
196     if (size < 4 || !mask || ring_offset == TM_QW0_USER) {
197         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid write access at TIMA @%"
198                       HWADDR_PRIx"\n", offset);
199         return;
200     }
201 
202     /*
203      * Use the register offset for the raw values and filter out
204      * reserved values
205      */
206     for (i = 0; i < size; i++) {
207         uint8_t byte_mask = (mask >> (8 * (size - i - 1)));
208         if (byte_mask) {
209             tctx->regs[reg_offset + i] = (value >> (8 * (size - i - 1))) &
210                 byte_mask;
211         }
212     }
213 }
214 
215 static uint64_t xive_tm_raw_read(XiveTCTX *tctx, hwaddr offset, unsigned size)
216 {
217     uint8_t ring_offset = offset & 0x30;
218     uint8_t reg_offset = offset & 0x3F;
219     uint64_t mask = xive_tm_mask(offset, size, false);
220     uint64_t ret;
221     int i;
222 
223     /*
224      * Only 4 or 8 bytes loads are allowed and the User ring is
225      * excluded
226      */
227     if (size < 4 || !mask || ring_offset == TM_QW0_USER) {
228         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid read access at TIMA @%"
229                       HWADDR_PRIx"\n", offset);
230         return -1;
231     }
232 
233     /* Use the register offset for the raw values */
234     ret = 0;
235     for (i = 0; i < size; i++) {
236         ret |= (uint64_t) tctx->regs[reg_offset + i] << (8 * (size - i - 1));
237     }
238 
239     /* filter out reserved values */
240     return ret & mask;
241 }
242 
243 /*
244  * The TM context is mapped twice within each page. Stores and loads
245  * to the first mapping below 2K write and read the specified values
246  * without modification. The second mapping above 2K performs specific
247  * state changes (side effects) in addition to setting/returning the
248  * interrupt management area context of the processor thread.
249  */
250 static uint64_t xive_tm_ack_os_reg(XiveTCTX *tctx, hwaddr offset, unsigned size)
251 {
252     return xive_tctx_accept(tctx, TM_QW1_OS);
253 }
254 
255 static void xive_tm_set_os_cppr(XiveTCTX *tctx, hwaddr offset,
256                                 uint64_t value, unsigned size)
257 {
258     xive_tctx_set_cppr(tctx, TM_QW1_OS, value & 0xff);
259 }
260 
261 /*
262  * Adjust the IPB to allow a CPU to process event queues of other
263  * priorities during one physical interrupt cycle.
264  */
265 static void xive_tm_set_os_pending(XiveTCTX *tctx, hwaddr offset,
266                                    uint64_t value, unsigned size)
267 {
268     ipb_update(&tctx->regs[TM_QW1_OS], value & 0xff);
269     xive_tctx_notify(tctx, TM_QW1_OS);
270 }
271 
272 /*
273  * Define a mapping of "special" operations depending on the TIMA page
274  * offset and the size of the operation.
275  */
276 typedef struct XiveTmOp {
277     uint8_t  page_offset;
278     uint32_t op_offset;
279     unsigned size;
280     void     (*write_handler)(XiveTCTX *tctx, hwaddr offset, uint64_t value,
281                               unsigned size);
282     uint64_t (*read_handler)(XiveTCTX *tctx, hwaddr offset, unsigned size);
283 } XiveTmOp;
284 
285 static const XiveTmOp xive_tm_operations[] = {
286     /*
287      * MMIOs below 2K : raw values and special operations without side
288      * effects
289      */
290     { XIVE_TM_OS_PAGE, TM_QW1_OS + TM_CPPR,   1, xive_tm_set_os_cppr, NULL },
291 
292     /* MMIOs above 2K : special operations with side effects */
293     { XIVE_TM_OS_PAGE, TM_SPC_ACK_OS_REG,     2, NULL, xive_tm_ack_os_reg },
294     { XIVE_TM_OS_PAGE, TM_SPC_SET_OS_PENDING, 1, xive_tm_set_os_pending, NULL },
295 };
296 
297 static const XiveTmOp *xive_tm_find_op(hwaddr offset, unsigned size, bool write)
298 {
299     uint8_t page_offset = (offset >> TM_SHIFT) & 0x3;
300     uint32_t op_offset = offset & 0xFFF;
301     int i;
302 
303     for (i = 0; i < ARRAY_SIZE(xive_tm_operations); i++) {
304         const XiveTmOp *xto = &xive_tm_operations[i];
305 
306         /* Accesses done from a more privileged TIMA page is allowed */
307         if (xto->page_offset >= page_offset &&
308             xto->op_offset == op_offset &&
309             xto->size == size &&
310             ((write && xto->write_handler) || (!write && xto->read_handler))) {
311             return xto;
312         }
313     }
314     return NULL;
315 }
316 
317 /*
318  * TIMA MMIO handlers
319  */
320 static void xive_tm_write(void *opaque, hwaddr offset,
321                           uint64_t value, unsigned size)
322 {
323     XiveTCTX *tctx = xive_router_get_tctx(XIVE_ROUTER(opaque), current_cpu);
324     const XiveTmOp *xto;
325 
326     /*
327      * TODO: check V bit in Q[0-3]W2, check PTER bit associated with CPU
328      */
329 
330     /*
331      * First, check for special operations in the 2K region
332      */
333     if (offset & 0x800) {
334         xto = xive_tm_find_op(offset, size, true);
335         if (!xto) {
336             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid write access at TIMA"
337                           "@%"HWADDR_PRIx"\n", offset);
338         } else {
339             xto->write_handler(tctx, offset, value, size);
340         }
341         return;
342     }
343 
344     /*
345      * Then, for special operations in the region below 2K.
346      */
347     xto = xive_tm_find_op(offset, size, true);
348     if (xto) {
349         xto->write_handler(tctx, offset, value, size);
350         return;
351     }
352 
353     /*
354      * Finish with raw access to the register values
355      */
356     xive_tm_raw_write(tctx, offset, value, size);
357 }
358 
359 static uint64_t xive_tm_read(void *opaque, hwaddr offset, unsigned size)
360 {
361     XiveTCTX *tctx = xive_router_get_tctx(XIVE_ROUTER(opaque), current_cpu);
362     const XiveTmOp *xto;
363 
364     /*
365      * TODO: check V bit in Q[0-3]W2, check PTER bit associated with CPU
366      */
367 
368     /*
369      * First, check for special operations in the 2K region
370      */
371     if (offset & 0x800) {
372         xto = xive_tm_find_op(offset, size, false);
373         if (!xto) {
374             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid read access to TIMA"
375                           "@%"HWADDR_PRIx"\n", offset);
376             return -1;
377         }
378         return xto->read_handler(tctx, offset, size);
379     }
380 
381     /*
382      * Then, for special operations in the region below 2K.
383      */
384     xto = xive_tm_find_op(offset, size, false);
385     if (xto) {
386         return xto->read_handler(tctx, offset, size);
387     }
388 
389     /*
390      * Finish with raw access to the register values
391      */
392     return xive_tm_raw_read(tctx, offset, size);
393 }
394 
395 const MemoryRegionOps xive_tm_ops = {
396     .read = xive_tm_read,
397     .write = xive_tm_write,
398     .endianness = DEVICE_BIG_ENDIAN,
399     .valid = {
400         .min_access_size = 1,
401         .max_access_size = 8,
402     },
403     .impl = {
404         .min_access_size = 1,
405         .max_access_size = 8,
406     },
407 };
408 
409 static inline uint32_t xive_tctx_word2(uint8_t *ring)
410 {
411     return *((uint32_t *) &ring[TM_WORD2]);
412 }
413 
414 static char *xive_tctx_ring_print(uint8_t *ring)
415 {
416     uint32_t w2 = xive_tctx_word2(ring);
417 
418     return g_strdup_printf("%02x   %02x  %02x    %02x   %02x  "
419                    "%02x  %02x   %02x  %08x",
420                    ring[TM_NSR], ring[TM_CPPR], ring[TM_IPB], ring[TM_LSMFB],
421                    ring[TM_ACK_CNT], ring[TM_INC], ring[TM_AGE], ring[TM_PIPR],
422                    be32_to_cpu(w2));
423 }
424 
425 static const char * const xive_tctx_ring_names[] = {
426     "USER", "OS", "POOL", "PHYS",
427 };
428 
429 void xive_tctx_pic_print_info(XiveTCTX *tctx, Monitor *mon)
430 {
431     int cpu_index = tctx->cs ? tctx->cs->cpu_index : -1;
432     int i;
433 
434     monitor_printf(mon, "CPU[%04x]:   QW   NSR CPPR IPB LSMFB ACK# INC AGE PIPR"
435                    "  W2\n", cpu_index);
436 
437     for (i = 0; i < XIVE_TM_RING_COUNT; i++) {
438         char *s = xive_tctx_ring_print(&tctx->regs[i * XIVE_TM_RING_SIZE]);
439         monitor_printf(mon, "CPU[%04x]: %4s    %s\n", cpu_index,
440                        xive_tctx_ring_names[i], s);
441         g_free(s);
442     }
443 }
444 
445 static void xive_tctx_reset(void *dev)
446 {
447     XiveTCTX *tctx = XIVE_TCTX(dev);
448 
449     memset(tctx->regs, 0, sizeof(tctx->regs));
450 
451     /* Set some defaults */
452     tctx->regs[TM_QW1_OS + TM_LSMFB] = 0xFF;
453     tctx->regs[TM_QW1_OS + TM_ACK_CNT] = 0xFF;
454     tctx->regs[TM_QW1_OS + TM_AGE] = 0xFF;
455 
456     /*
457      * Initialize PIPR to 0xFF to avoid phantom interrupts when the
458      * CPPR is first set.
459      */
460     tctx->regs[TM_QW1_OS + TM_PIPR] =
461         ipb_to_pipr(tctx->regs[TM_QW1_OS + TM_IPB]);
462 }
463 
464 static void xive_tctx_realize(DeviceState *dev, Error **errp)
465 {
466     XiveTCTX *tctx = XIVE_TCTX(dev);
467     PowerPCCPU *cpu;
468     CPUPPCState *env;
469     Object *obj;
470     Error *local_err = NULL;
471 
472     obj = object_property_get_link(OBJECT(dev), "cpu", &local_err);
473     if (!obj) {
474         error_propagate(errp, local_err);
475         error_prepend(errp, "required link 'cpu' not found: ");
476         return;
477     }
478 
479     cpu = POWERPC_CPU(obj);
480     tctx->cs = CPU(obj);
481 
482     env = &cpu->env;
483     switch (PPC_INPUT(env)) {
484     case PPC_FLAGS_INPUT_POWER7:
485         tctx->output = env->irq_inputs[POWER7_INPUT_INT];
486         break;
487 
488     default:
489         error_setg(errp, "XIVE interrupt controller does not support "
490                    "this CPU bus model");
491         return;
492     }
493 
494     qemu_register_reset(xive_tctx_reset, dev);
495 }
496 
497 static void xive_tctx_unrealize(DeviceState *dev, Error **errp)
498 {
499     qemu_unregister_reset(xive_tctx_reset, dev);
500 }
501 
502 static const VMStateDescription vmstate_xive_tctx = {
503     .name = TYPE_XIVE_TCTX,
504     .version_id = 1,
505     .minimum_version_id = 1,
506     .fields = (VMStateField[]) {
507         VMSTATE_BUFFER(regs, XiveTCTX),
508         VMSTATE_END_OF_LIST()
509     },
510 };
511 
512 static void xive_tctx_class_init(ObjectClass *klass, void *data)
513 {
514     DeviceClass *dc = DEVICE_CLASS(klass);
515 
516     dc->desc = "XIVE Interrupt Thread Context";
517     dc->realize = xive_tctx_realize;
518     dc->unrealize = xive_tctx_unrealize;
519     dc->vmsd = &vmstate_xive_tctx;
520 }
521 
522 static const TypeInfo xive_tctx_info = {
523     .name          = TYPE_XIVE_TCTX,
524     .parent        = TYPE_DEVICE,
525     .instance_size = sizeof(XiveTCTX),
526     .class_init    = xive_tctx_class_init,
527 };
528 
529 Object *xive_tctx_create(Object *cpu, XiveRouter *xrtr, Error **errp)
530 {
531     Error *local_err = NULL;
532     Object *obj;
533 
534     obj = object_new(TYPE_XIVE_TCTX);
535     object_property_add_child(cpu, TYPE_XIVE_TCTX, obj, &error_abort);
536     object_unref(obj);
537     object_property_add_const_link(obj, "cpu", cpu, &error_abort);
538     object_property_set_bool(obj, true, "realized", &local_err);
539     if (local_err) {
540         goto error;
541     }
542 
543     return obj;
544 
545 error:
546     object_unparent(obj);
547     error_propagate(errp, local_err);
548     return NULL;
549 }
550 
551 /*
552  * XIVE ESB helpers
553  */
554 
555 static uint8_t xive_esb_set(uint8_t *pq, uint8_t value)
556 {
557     uint8_t old_pq = *pq & 0x3;
558 
559     *pq &= ~0x3;
560     *pq |= value & 0x3;
561 
562     return old_pq;
563 }
564 
565 static bool xive_esb_trigger(uint8_t *pq)
566 {
567     uint8_t old_pq = *pq & 0x3;
568 
569     switch (old_pq) {
570     case XIVE_ESB_RESET:
571         xive_esb_set(pq, XIVE_ESB_PENDING);
572         return true;
573     case XIVE_ESB_PENDING:
574     case XIVE_ESB_QUEUED:
575         xive_esb_set(pq, XIVE_ESB_QUEUED);
576         return false;
577     case XIVE_ESB_OFF:
578         xive_esb_set(pq, XIVE_ESB_OFF);
579         return false;
580     default:
581          g_assert_not_reached();
582     }
583 }
584 
585 static bool xive_esb_eoi(uint8_t *pq)
586 {
587     uint8_t old_pq = *pq & 0x3;
588 
589     switch (old_pq) {
590     case XIVE_ESB_RESET:
591     case XIVE_ESB_PENDING:
592         xive_esb_set(pq, XIVE_ESB_RESET);
593         return false;
594     case XIVE_ESB_QUEUED:
595         xive_esb_set(pq, XIVE_ESB_PENDING);
596         return true;
597     case XIVE_ESB_OFF:
598         xive_esb_set(pq, XIVE_ESB_OFF);
599         return false;
600     default:
601          g_assert_not_reached();
602     }
603 }
604 
605 /*
606  * XIVE Interrupt Source (or IVSE)
607  */
608 
609 uint8_t xive_source_esb_get(XiveSource *xsrc, uint32_t srcno)
610 {
611     assert(srcno < xsrc->nr_irqs);
612 
613     return xsrc->status[srcno] & 0x3;
614 }
615 
616 uint8_t xive_source_esb_set(XiveSource *xsrc, uint32_t srcno, uint8_t pq)
617 {
618     assert(srcno < xsrc->nr_irqs);
619 
620     return xive_esb_set(&xsrc->status[srcno], pq);
621 }
622 
623 /*
624  * Returns whether the event notification should be forwarded.
625  */
626 static bool xive_source_lsi_trigger(XiveSource *xsrc, uint32_t srcno)
627 {
628     uint8_t old_pq = xive_source_esb_get(xsrc, srcno);
629 
630     xsrc->status[srcno] |= XIVE_STATUS_ASSERTED;
631 
632     switch (old_pq) {
633     case XIVE_ESB_RESET:
634         xive_source_esb_set(xsrc, srcno, XIVE_ESB_PENDING);
635         return true;
636     default:
637         return false;
638     }
639 }
640 
641 /*
642  * Returns whether the event notification should be forwarded.
643  */
644 static bool xive_source_esb_trigger(XiveSource *xsrc, uint32_t srcno)
645 {
646     bool ret;
647 
648     assert(srcno < xsrc->nr_irqs);
649 
650     ret = xive_esb_trigger(&xsrc->status[srcno]);
651 
652     if (xive_source_irq_is_lsi(xsrc, srcno) &&
653         xive_source_esb_get(xsrc, srcno) == XIVE_ESB_QUEUED) {
654         qemu_log_mask(LOG_GUEST_ERROR,
655                       "XIVE: queued an event on LSI IRQ %d\n", srcno);
656     }
657 
658     return ret;
659 }
660 
661 /*
662  * Returns whether the event notification should be forwarded.
663  */
664 static bool xive_source_esb_eoi(XiveSource *xsrc, uint32_t srcno)
665 {
666     bool ret;
667 
668     assert(srcno < xsrc->nr_irqs);
669 
670     ret = xive_esb_eoi(&xsrc->status[srcno]);
671 
672     /*
673      * LSI sources do not set the Q bit but they can still be
674      * asserted, in which case we should forward a new event
675      * notification
676      */
677     if (xive_source_irq_is_lsi(xsrc, srcno) &&
678         xsrc->status[srcno] & XIVE_STATUS_ASSERTED) {
679         ret = xive_source_lsi_trigger(xsrc, srcno);
680     }
681 
682     return ret;
683 }
684 
685 /*
686  * Forward the source event notification to the Router
687  */
688 static void xive_source_notify(XiveSource *xsrc, int srcno)
689 {
690     XiveNotifierClass *xnc = XIVE_NOTIFIER_GET_CLASS(xsrc->xive);
691 
692     if (xnc->notify) {
693         xnc->notify(xsrc->xive, srcno);
694     }
695 }
696 
697 /*
698  * In a two pages ESB MMIO setting, even page is the trigger page, odd
699  * page is for management
700  */
701 static inline bool addr_is_even(hwaddr addr, uint32_t shift)
702 {
703     return !((addr >> shift) & 1);
704 }
705 
706 static inline bool xive_source_is_trigger_page(XiveSource *xsrc, hwaddr addr)
707 {
708     return xive_source_esb_has_2page(xsrc) &&
709         addr_is_even(addr, xsrc->esb_shift - 1);
710 }
711 
712 /*
713  * ESB MMIO loads
714  *                      Trigger page    Management/EOI page
715  *
716  * ESB MMIO setting     2 pages         1 or 2 pages
717  *
718  * 0x000 .. 0x3FF       -1              EOI and return 0|1
719  * 0x400 .. 0x7FF       -1              EOI and return 0|1
720  * 0x800 .. 0xBFF       -1              return PQ
721  * 0xC00 .. 0xCFF       -1              return PQ and atomically PQ=00
722  * 0xD00 .. 0xDFF       -1              return PQ and atomically PQ=01
723  * 0xE00 .. 0xDFF       -1              return PQ and atomically PQ=10
724  * 0xF00 .. 0xDFF       -1              return PQ and atomically PQ=11
725  */
726 static uint64_t xive_source_esb_read(void *opaque, hwaddr addr, unsigned size)
727 {
728     XiveSource *xsrc = XIVE_SOURCE(opaque);
729     uint32_t offset = addr & 0xFFF;
730     uint32_t srcno = addr >> xsrc->esb_shift;
731     uint64_t ret = -1;
732 
733     /* In a two pages ESB MMIO setting, trigger page should not be read */
734     if (xive_source_is_trigger_page(xsrc, addr)) {
735         qemu_log_mask(LOG_GUEST_ERROR,
736                       "XIVE: invalid load on IRQ %d trigger page at "
737                       "0x%"HWADDR_PRIx"\n", srcno, addr);
738         return -1;
739     }
740 
741     switch (offset) {
742     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
743         ret = xive_source_esb_eoi(xsrc, srcno);
744 
745         /* Forward the source event notification for routing */
746         if (ret) {
747             xive_source_notify(xsrc, srcno);
748         }
749         break;
750 
751     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
752         ret = xive_source_esb_get(xsrc, srcno);
753         break;
754 
755     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
756     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
757     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
758     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
759         ret = xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
760         break;
761     default:
762         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB load addr %x\n",
763                       offset);
764     }
765 
766     return ret;
767 }
768 
769 /*
770  * ESB MMIO stores
771  *                      Trigger page    Management/EOI page
772  *
773  * ESB MMIO setting     2 pages         1 or 2 pages
774  *
775  * 0x000 .. 0x3FF       Trigger         Trigger
776  * 0x400 .. 0x7FF       Trigger         EOI
777  * 0x800 .. 0xBFF       Trigger         undefined
778  * 0xC00 .. 0xCFF       Trigger         PQ=00
779  * 0xD00 .. 0xDFF       Trigger         PQ=01
780  * 0xE00 .. 0xDFF       Trigger         PQ=10
781  * 0xF00 .. 0xDFF       Trigger         PQ=11
782  */
783 static void xive_source_esb_write(void *opaque, hwaddr addr,
784                                   uint64_t value, unsigned size)
785 {
786     XiveSource *xsrc = XIVE_SOURCE(opaque);
787     uint32_t offset = addr & 0xFFF;
788     uint32_t srcno = addr >> xsrc->esb_shift;
789     bool notify = false;
790 
791     /* In a two pages ESB MMIO setting, trigger page only triggers */
792     if (xive_source_is_trigger_page(xsrc, addr)) {
793         notify = xive_source_esb_trigger(xsrc, srcno);
794         goto out;
795     }
796 
797     switch (offset) {
798     case 0 ... 0x3FF:
799         notify = xive_source_esb_trigger(xsrc, srcno);
800         break;
801 
802     case XIVE_ESB_STORE_EOI ... XIVE_ESB_STORE_EOI + 0x3FF:
803         if (!(xsrc->esb_flags & XIVE_SRC_STORE_EOI)) {
804             qemu_log_mask(LOG_GUEST_ERROR,
805                           "XIVE: invalid Store EOI for IRQ %d\n", srcno);
806             return;
807         }
808 
809         notify = xive_source_esb_eoi(xsrc, srcno);
810         break;
811 
812     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
813     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
814     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
815     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
816         xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
817         break;
818 
819     default:
820         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr %x\n",
821                       offset);
822         return;
823     }
824 
825 out:
826     /* Forward the source event notification for routing */
827     if (notify) {
828         xive_source_notify(xsrc, srcno);
829     }
830 }
831 
832 static const MemoryRegionOps xive_source_esb_ops = {
833     .read = xive_source_esb_read,
834     .write = xive_source_esb_write,
835     .endianness = DEVICE_BIG_ENDIAN,
836     .valid = {
837         .min_access_size = 8,
838         .max_access_size = 8,
839     },
840     .impl = {
841         .min_access_size = 8,
842         .max_access_size = 8,
843     },
844 };
845 
846 void xive_source_set_irq(void *opaque, int srcno, int val)
847 {
848     XiveSource *xsrc = XIVE_SOURCE(opaque);
849     bool notify = false;
850 
851     if (xive_source_irq_is_lsi(xsrc, srcno)) {
852         if (val) {
853             notify = xive_source_lsi_trigger(xsrc, srcno);
854         } else {
855             xsrc->status[srcno] &= ~XIVE_STATUS_ASSERTED;
856         }
857     } else {
858         if (val) {
859             notify = xive_source_esb_trigger(xsrc, srcno);
860         }
861     }
862 
863     /* Forward the source event notification for routing */
864     if (notify) {
865         xive_source_notify(xsrc, srcno);
866     }
867 }
868 
869 void xive_source_pic_print_info(XiveSource *xsrc, uint32_t offset, Monitor *mon)
870 {
871     int i;
872 
873     for (i = 0; i < xsrc->nr_irqs; i++) {
874         uint8_t pq = xive_source_esb_get(xsrc, i);
875 
876         if (pq == XIVE_ESB_OFF) {
877             continue;
878         }
879 
880         monitor_printf(mon, "  %08x %s %c%c%c\n", i + offset,
881                        xive_source_irq_is_lsi(xsrc, i) ? "LSI" : "MSI",
882                        pq & XIVE_ESB_VAL_P ? 'P' : '-',
883                        pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
884                        xsrc->status[i] & XIVE_STATUS_ASSERTED ? 'A' : ' ');
885     }
886 }
887 
888 static void xive_source_reset(void *dev)
889 {
890     XiveSource *xsrc = XIVE_SOURCE(dev);
891 
892     /* Do not clear the LSI bitmap */
893 
894     /* PQs are initialized to 0b01 (Q=1) which corresponds to "ints off" */
895     memset(xsrc->status, XIVE_ESB_OFF, xsrc->nr_irqs);
896 }
897 
898 static void xive_source_realize(DeviceState *dev, Error **errp)
899 {
900     XiveSource *xsrc = XIVE_SOURCE(dev);
901     Object *obj;
902     Error *local_err = NULL;
903 
904     obj = object_property_get_link(OBJECT(dev), "xive", &local_err);
905     if (!obj) {
906         error_propagate(errp, local_err);
907         error_prepend(errp, "required link 'xive' not found: ");
908         return;
909     }
910 
911     xsrc->xive = XIVE_NOTIFIER(obj);
912 
913     if (!xsrc->nr_irqs) {
914         error_setg(errp, "Number of interrupt needs to be greater than 0");
915         return;
916     }
917 
918     if (xsrc->esb_shift != XIVE_ESB_4K &&
919         xsrc->esb_shift != XIVE_ESB_4K_2PAGE &&
920         xsrc->esb_shift != XIVE_ESB_64K &&
921         xsrc->esb_shift != XIVE_ESB_64K_2PAGE) {
922         error_setg(errp, "Invalid ESB shift setting");
923         return;
924     }
925 
926     xsrc->status = g_malloc0(xsrc->nr_irqs);
927     xsrc->lsi_map = bitmap_new(xsrc->nr_irqs);
928 
929     memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
930                           &xive_source_esb_ops, xsrc, "xive.esb",
931                           (1ull << xsrc->esb_shift) * xsrc->nr_irqs);
932 
933     qemu_register_reset(xive_source_reset, dev);
934 }
935 
936 static const VMStateDescription vmstate_xive_source = {
937     .name = TYPE_XIVE_SOURCE,
938     .version_id = 1,
939     .minimum_version_id = 1,
940     .fields = (VMStateField[]) {
941         VMSTATE_UINT32_EQUAL(nr_irqs, XiveSource, NULL),
942         VMSTATE_VBUFFER_UINT32(status, XiveSource, 1, NULL, nr_irqs),
943         VMSTATE_END_OF_LIST()
944     },
945 };
946 
947 /*
948  * The default XIVE interrupt source setting for the ESB MMIOs is two
949  * 64k pages without Store EOI, to be in sync with KVM.
950  */
951 static Property xive_source_properties[] = {
952     DEFINE_PROP_UINT64("flags", XiveSource, esb_flags, 0),
953     DEFINE_PROP_UINT32("nr-irqs", XiveSource, nr_irqs, 0),
954     DEFINE_PROP_UINT32("shift", XiveSource, esb_shift, XIVE_ESB_64K_2PAGE),
955     DEFINE_PROP_END_OF_LIST(),
956 };
957 
958 static void xive_source_class_init(ObjectClass *klass, void *data)
959 {
960     DeviceClass *dc = DEVICE_CLASS(klass);
961 
962     dc->desc    = "XIVE Interrupt Source";
963     dc->props   = xive_source_properties;
964     dc->realize = xive_source_realize;
965     dc->vmsd    = &vmstate_xive_source;
966 }
967 
968 static const TypeInfo xive_source_info = {
969     .name          = TYPE_XIVE_SOURCE,
970     .parent        = TYPE_DEVICE,
971     .instance_size = sizeof(XiveSource),
972     .class_init    = xive_source_class_init,
973 };
974 
975 /*
976  * XiveEND helpers
977  */
978 
979 void xive_end_queue_pic_print_info(XiveEND *end, uint32_t width, Monitor *mon)
980 {
981     uint64_t qaddr_base = (uint64_t) be32_to_cpu(end->w2 & 0x0fffffff) << 32
982         | be32_to_cpu(end->w3);
983     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
984     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
985     uint32_t qentries = 1 << (qsize + 10);
986     int i;
987 
988     /*
989      * print out the [ (qindex - (width - 1)) .. (qindex + 1)] window
990      */
991     monitor_printf(mon, " [ ");
992     qindex = (qindex - (width - 1)) & (qentries - 1);
993     for (i = 0; i < width; i++) {
994         uint64_t qaddr = qaddr_base + (qindex << 2);
995         uint32_t qdata = -1;
996 
997         if (dma_memory_read(&address_space_memory, qaddr, &qdata,
998                             sizeof(qdata))) {
999             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to read EQ @0x%"
1000                           HWADDR_PRIx "\n", qaddr);
1001             return;
1002         }
1003         monitor_printf(mon, "%s%08x ", i == width - 1 ? "^" : "",
1004                        be32_to_cpu(qdata));
1005         qindex = (qindex + 1) & (qentries - 1);
1006     }
1007 }
1008 
1009 void xive_end_pic_print_info(XiveEND *end, uint32_t end_idx, Monitor *mon)
1010 {
1011     uint64_t qaddr_base = (uint64_t) be32_to_cpu(end->w2 & 0x0fffffff) << 32
1012         | be32_to_cpu(end->w3);
1013     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1014     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1015     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1016     uint32_t qentries = 1 << (qsize + 10);
1017 
1018     uint32_t nvt = xive_get_field32(END_W6_NVT_INDEX, end->w6);
1019     uint8_t priority = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
1020 
1021     if (!xive_end_is_valid(end)) {
1022         return;
1023     }
1024 
1025     monitor_printf(mon, "  %08x %c%c%c%c%c prio:%d nvt:%04x eq:@%08"PRIx64
1026                    "% 6d/%5d ^%d", end_idx,
1027                    xive_end_is_valid(end)    ? 'v' : '-',
1028                    xive_end_is_enqueue(end)  ? 'q' : '-',
1029                    xive_end_is_notify(end)   ? 'n' : '-',
1030                    xive_end_is_backlog(end)  ? 'b' : '-',
1031                    xive_end_is_escalate(end) ? 'e' : '-',
1032                    priority, nvt, qaddr_base, qindex, qentries, qgen);
1033 
1034     xive_end_queue_pic_print_info(end, 6, mon);
1035     monitor_printf(mon, "]\n");
1036 }
1037 
1038 static void xive_end_enqueue(XiveEND *end, uint32_t data)
1039 {
1040     uint64_t qaddr_base = (uint64_t) be32_to_cpu(end->w2 & 0x0fffffff) << 32
1041         | be32_to_cpu(end->w3);
1042     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1043     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1044     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1045 
1046     uint64_t qaddr = qaddr_base + (qindex << 2);
1047     uint32_t qdata = cpu_to_be32((qgen << 31) | (data & 0x7fffffff));
1048     uint32_t qentries = 1 << (qsize + 10);
1049 
1050     if (dma_memory_write(&address_space_memory, qaddr, &qdata, sizeof(qdata))) {
1051         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to write END data @0x%"
1052                       HWADDR_PRIx "\n", qaddr);
1053         return;
1054     }
1055 
1056     qindex = (qindex + 1) & (qentries - 1);
1057     if (qindex == 0) {
1058         qgen ^= 1;
1059         end->w1 = xive_set_field32(END_W1_GENERATION, end->w1, qgen);
1060     }
1061     end->w1 = xive_set_field32(END_W1_PAGE_OFF, end->w1, qindex);
1062 }
1063 
1064 /*
1065  * XIVE Router (aka. Virtualization Controller or IVRE)
1066  */
1067 
1068 int xive_router_get_eas(XiveRouter *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1069                         XiveEAS *eas)
1070 {
1071     XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1072 
1073     return xrc->get_eas(xrtr, eas_blk, eas_idx, eas);
1074 }
1075 
1076 int xive_router_get_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1077                         XiveEND *end)
1078 {
1079    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1080 
1081    return xrc->get_end(xrtr, end_blk, end_idx, end);
1082 }
1083 
1084 int xive_router_write_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1085                           XiveEND *end, uint8_t word_number)
1086 {
1087    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1088 
1089    return xrc->write_end(xrtr, end_blk, end_idx, end, word_number);
1090 }
1091 
1092 int xive_router_get_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1093                         XiveNVT *nvt)
1094 {
1095    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1096 
1097    return xrc->get_nvt(xrtr, nvt_blk, nvt_idx, nvt);
1098 }
1099 
1100 int xive_router_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1101                         XiveNVT *nvt, uint8_t word_number)
1102 {
1103    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1104 
1105    return xrc->write_nvt(xrtr, nvt_blk, nvt_idx, nvt, word_number);
1106 }
1107 
1108 XiveTCTX *xive_router_get_tctx(XiveRouter *xrtr, CPUState *cs)
1109 {
1110     XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1111 
1112     return xrc->get_tctx(xrtr, cs);
1113 }
1114 
1115 /*
1116  * The thread context register words are in big-endian format.
1117  */
1118 static int xive_presenter_tctx_match(XiveTCTX *tctx, uint8_t format,
1119                                      uint8_t nvt_blk, uint32_t nvt_idx,
1120                                      bool cam_ignore, uint32_t logic_serv)
1121 {
1122     uint32_t cam = xive_nvt_cam_line(nvt_blk, nvt_idx);
1123     uint32_t qw2w2 = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
1124     uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
1125     uint32_t qw0w2 = xive_tctx_word2(&tctx->regs[TM_QW0_USER]);
1126 
1127     /*
1128      * TODO (PowerNV): ignore mode. The low order bits of the NVT
1129      * identifier are ignored in the "CAM" match.
1130      */
1131 
1132     if (format == 0) {
1133         if (cam_ignore == true) {
1134             /*
1135              * F=0 & i=1: Logical server notification (bits ignored at
1136              * the end of the NVT identifier)
1137              */
1138             qemu_log_mask(LOG_UNIMP, "XIVE: no support for LS NVT %x/%x\n",
1139                           nvt_blk, nvt_idx);
1140              return -1;
1141         }
1142 
1143         /* F=0 & i=0: Specific NVT notification */
1144 
1145         /* TODO (PowerNV) : PHYS ring */
1146 
1147         /* HV POOL ring */
1148         if ((be32_to_cpu(qw2w2) & TM_QW2W2_VP) &&
1149             cam == xive_get_field32(TM_QW2W2_POOL_CAM, qw2w2)) {
1150             return TM_QW2_HV_POOL;
1151         }
1152 
1153         /* OS ring */
1154         if ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1155             cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) {
1156             return TM_QW1_OS;
1157         }
1158     } else {
1159         /* F=1 : User level Event-Based Branch (EBB) notification */
1160 
1161         /* USER ring */
1162         if  ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1163              (cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) &&
1164              (be32_to_cpu(qw0w2) & TM_QW0W2_VU) &&
1165              (logic_serv == xive_get_field32(TM_QW0W2_LOGIC_SERV, qw0w2))) {
1166             return TM_QW0_USER;
1167         }
1168     }
1169     return -1;
1170 }
1171 
1172 typedef struct XiveTCTXMatch {
1173     XiveTCTX *tctx;
1174     uint8_t ring;
1175 } XiveTCTXMatch;
1176 
1177 static bool xive_presenter_match(XiveRouter *xrtr, uint8_t format,
1178                                  uint8_t nvt_blk, uint32_t nvt_idx,
1179                                  bool cam_ignore, uint8_t priority,
1180                                  uint32_t logic_serv, XiveTCTXMatch *match)
1181 {
1182     CPUState *cs;
1183 
1184     /*
1185      * TODO (PowerNV): handle chip_id overwrite of block field for
1186      * hardwired CAM compares
1187      */
1188 
1189     CPU_FOREACH(cs) {
1190         XiveTCTX *tctx = xive_router_get_tctx(xrtr, cs);
1191         int ring;
1192 
1193         /*
1194          * HW checks that the CPU is enabled in the Physical Thread
1195          * Enable Register (PTER).
1196          */
1197 
1198         /*
1199          * Check the thread context CAM lines and record matches. We
1200          * will handle CPU exception delivery later
1201          */
1202         ring = xive_presenter_tctx_match(tctx, format, nvt_blk, nvt_idx,
1203                                          cam_ignore, logic_serv);
1204         /*
1205          * Save the context and follow on to catch duplicates, that we
1206          * don't support yet.
1207          */
1208         if (ring != -1) {
1209             if (match->tctx) {
1210                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: already found a thread "
1211                               "context NVT %x/%x\n", nvt_blk, nvt_idx);
1212                 return false;
1213             }
1214 
1215             match->ring = ring;
1216             match->tctx = tctx;
1217         }
1218     }
1219 
1220     if (!match->tctx) {
1221         qemu_log_mask(LOG_UNIMP, "XIVE: NVT %x/%x is not dispatched\n",
1222                       nvt_blk, nvt_idx);
1223         return false;
1224     }
1225 
1226     return true;
1227 }
1228 
1229 /*
1230  * This is our simple Xive Presenter Engine model. It is merged in the
1231  * Router as it does not require an extra object.
1232  *
1233  * It receives notification requests sent by the IVRE to find one
1234  * matching NVT (or more) dispatched on the processor threads. In case
1235  * of a single NVT notification, the process is abreviated and the
1236  * thread is signaled if a match is found. In case of a logical server
1237  * notification (bits ignored at the end of the NVT identifier), the
1238  * IVPE and IVRE select a winning thread using different filters. This
1239  * involves 2 or 3 exchanges on the PowerBus that the model does not
1240  * support.
1241  *
1242  * The parameters represent what is sent on the PowerBus
1243  */
1244 static void xive_presenter_notify(XiveRouter *xrtr, uint8_t format,
1245                                   uint8_t nvt_blk, uint32_t nvt_idx,
1246                                   bool cam_ignore, uint8_t priority,
1247                                   uint32_t logic_serv)
1248 {
1249     XiveNVT nvt;
1250     XiveTCTXMatch match = { .tctx = NULL, .ring = 0 };
1251     bool found;
1252 
1253     /* NVT cache lookup */
1254     if (xive_router_get_nvt(xrtr, nvt_blk, nvt_idx, &nvt)) {
1255         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no NVT %x/%x\n",
1256                       nvt_blk, nvt_idx);
1257         return;
1258     }
1259 
1260     if (!xive_nvt_is_valid(&nvt)) {
1261         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is invalid\n",
1262                       nvt_blk, nvt_idx);
1263         return;
1264     }
1265 
1266     found = xive_presenter_match(xrtr, format, nvt_blk, nvt_idx, cam_ignore,
1267                                  priority, logic_serv, &match);
1268     if (found) {
1269         ipb_update(&match.tctx->regs[match.ring], priority);
1270         xive_tctx_notify(match.tctx, match.ring);
1271         return;
1272     }
1273 
1274     /* Record the IPB in the associated NVT structure */
1275     ipb_update((uint8_t *) &nvt.w4, priority);
1276     xive_router_write_nvt(xrtr, nvt_blk, nvt_idx, &nvt, 4);
1277 
1278     /*
1279      * If no matching NVT is dispatched on a HW thread :
1280      * - update the NVT structure if backlog is activated
1281      * - escalate (ESe PQ bits and EAS in w4-5) if escalation is
1282      *   activated
1283      */
1284 }
1285 
1286 /*
1287  * An END trigger can come from an event trigger (IPI or HW) or from
1288  * another chip. We don't model the PowerBus but the END trigger
1289  * message has the same parameters than in the function below.
1290  */
1291 static void xive_router_end_notify(XiveRouter *xrtr, uint8_t end_blk,
1292                                    uint32_t end_idx, uint32_t end_data)
1293 {
1294     XiveEND end;
1295     uint8_t priority;
1296     uint8_t format;
1297 
1298     /* END cache lookup */
1299     if (xive_router_get_end(xrtr, end_blk, end_idx, &end)) {
1300         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1301                       end_idx);
1302         return;
1303     }
1304 
1305     if (!xive_end_is_valid(&end)) {
1306         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1307                       end_blk, end_idx);
1308         return;
1309     }
1310 
1311     if (xive_end_is_enqueue(&end)) {
1312         xive_end_enqueue(&end, end_data);
1313         /* Enqueuing event data modifies the EQ toggle and index */
1314         xive_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1315     }
1316 
1317     /*
1318      * The W7 format depends on the F bit in W6. It defines the type
1319      * of the notification :
1320      *
1321      *   F=0 : single or multiple NVT notification
1322      *   F=1 : User level Event-Based Branch (EBB) notification, no
1323      *         priority
1324      */
1325     format = xive_get_field32(END_W6_FORMAT_BIT, end.w6);
1326     priority = xive_get_field32(END_W7_F0_PRIORITY, end.w7);
1327 
1328     /* The END is masked */
1329     if (format == 0 && priority == 0xff) {
1330         return;
1331     }
1332 
1333     /*
1334      * Check the END ESn (Event State Buffer for notification) for
1335      * even futher coalescing in the Router
1336      */
1337     if (!xive_end_is_notify(&end)) {
1338         uint8_t pq = xive_get_field32(END_W1_ESn, end.w1);
1339         bool notify = xive_esb_trigger(&pq);
1340 
1341         if (pq != xive_get_field32(END_W1_ESn, end.w1)) {
1342             end.w1 = xive_set_field32(END_W1_ESn, end.w1, pq);
1343             xive_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1344         }
1345 
1346         /* ESn[Q]=1 : end of notification */
1347         if (!notify) {
1348             return;
1349         }
1350     }
1351 
1352     /*
1353      * Follows IVPE notification
1354      */
1355     xive_presenter_notify(xrtr, format,
1356                           xive_get_field32(END_W6_NVT_BLOCK, end.w6),
1357                           xive_get_field32(END_W6_NVT_INDEX, end.w6),
1358                           xive_get_field32(END_W7_F0_IGNORE, end.w7),
1359                           priority,
1360                           xive_get_field32(END_W7_F1_LOG_SERVER_ID, end.w7));
1361 
1362     /* TODO: Auto EOI. */
1363 }
1364 
1365 static void xive_router_notify(XiveNotifier *xn, uint32_t lisn)
1366 {
1367     XiveRouter *xrtr = XIVE_ROUTER(xn);
1368     uint8_t eas_blk = XIVE_SRCNO_BLOCK(lisn);
1369     uint32_t eas_idx = XIVE_SRCNO_INDEX(lisn);
1370     XiveEAS eas;
1371 
1372     /* EAS cache lookup */
1373     if (xive_router_get_eas(xrtr, eas_blk, eas_idx, &eas)) {
1374         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN %x\n", lisn);
1375         return;
1376     }
1377 
1378     /*
1379      * The IVRE checks the State Bit Cache at this point. We skip the
1380      * SBC lookup because the state bits of the sources are modeled
1381      * internally in QEMU.
1382      */
1383 
1384     if (!xive_eas_is_valid(&eas)) {
1385         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid LISN %x\n", lisn);
1386         return;
1387     }
1388 
1389     if (xive_eas_is_masked(&eas)) {
1390         /* Notification completed */
1391         return;
1392     }
1393 
1394     /*
1395      * The event trigger becomes an END trigger
1396      */
1397     xive_router_end_notify(xrtr,
1398                            xive_get_field64(EAS_END_BLOCK, eas.w),
1399                            xive_get_field64(EAS_END_INDEX, eas.w),
1400                            xive_get_field64(EAS_END_DATA,  eas.w));
1401 }
1402 
1403 static void xive_router_class_init(ObjectClass *klass, void *data)
1404 {
1405     DeviceClass *dc = DEVICE_CLASS(klass);
1406     XiveNotifierClass *xnc = XIVE_NOTIFIER_CLASS(klass);
1407 
1408     dc->desc    = "XIVE Router Engine";
1409     xnc->notify = xive_router_notify;
1410 }
1411 
1412 static const TypeInfo xive_router_info = {
1413     .name          = TYPE_XIVE_ROUTER,
1414     .parent        = TYPE_SYS_BUS_DEVICE,
1415     .abstract      = true,
1416     .class_size    = sizeof(XiveRouterClass),
1417     .class_init    = xive_router_class_init,
1418     .interfaces    = (InterfaceInfo[]) {
1419         { TYPE_XIVE_NOTIFIER },
1420         { }
1421     }
1422 };
1423 
1424 void xive_eas_pic_print_info(XiveEAS *eas, uint32_t lisn, Monitor *mon)
1425 {
1426     if (!xive_eas_is_valid(eas)) {
1427         return;
1428     }
1429 
1430     monitor_printf(mon, "  %08x %s end:%02x/%04x data:%08x\n",
1431                    lisn, xive_eas_is_masked(eas) ? "M" : " ",
1432                    (uint8_t)  xive_get_field64(EAS_END_BLOCK, eas->w),
1433                    (uint32_t) xive_get_field64(EAS_END_INDEX, eas->w),
1434                    (uint32_t) xive_get_field64(EAS_END_DATA, eas->w));
1435 }
1436 
1437 /*
1438  * END ESB MMIO loads
1439  */
1440 static uint64_t xive_end_source_read(void *opaque, hwaddr addr, unsigned size)
1441 {
1442     XiveENDSource *xsrc = XIVE_END_SOURCE(opaque);
1443     uint32_t offset = addr & 0xFFF;
1444     uint8_t end_blk;
1445     uint32_t end_idx;
1446     XiveEND end;
1447     uint32_t end_esmask;
1448     uint8_t pq;
1449     uint64_t ret = -1;
1450 
1451     end_blk = xsrc->block_id;
1452     end_idx = addr >> (xsrc->esb_shift + 1);
1453 
1454     if (xive_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
1455         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1456                       end_idx);
1457         return -1;
1458     }
1459 
1460     if (!xive_end_is_valid(&end)) {
1461         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1462                       end_blk, end_idx);
1463         return -1;
1464     }
1465 
1466     end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END_W1_ESn : END_W1_ESe;
1467     pq = xive_get_field32(end_esmask, end.w1);
1468 
1469     switch (offset) {
1470     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
1471         ret = xive_esb_eoi(&pq);
1472 
1473         /* Forward the source event notification for routing ?? */
1474         break;
1475 
1476     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
1477         ret = pq;
1478         break;
1479 
1480     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1481     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1482     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1483     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1484         ret = xive_esb_set(&pq, (offset >> 8) & 0x3);
1485         break;
1486     default:
1487         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB load addr %d\n",
1488                       offset);
1489         return -1;
1490     }
1491 
1492     if (pq != xive_get_field32(end_esmask, end.w1)) {
1493         end.w1 = xive_set_field32(end_esmask, end.w1, pq);
1494         xive_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
1495     }
1496 
1497     return ret;
1498 }
1499 
1500 /*
1501  * END ESB MMIO stores are invalid
1502  */
1503 static void xive_end_source_write(void *opaque, hwaddr addr,
1504                                   uint64_t value, unsigned size)
1505 {
1506     qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr 0x%"
1507                   HWADDR_PRIx"\n", addr);
1508 }
1509 
1510 static const MemoryRegionOps xive_end_source_ops = {
1511     .read = xive_end_source_read,
1512     .write = xive_end_source_write,
1513     .endianness = DEVICE_BIG_ENDIAN,
1514     .valid = {
1515         .min_access_size = 8,
1516         .max_access_size = 8,
1517     },
1518     .impl = {
1519         .min_access_size = 8,
1520         .max_access_size = 8,
1521     },
1522 };
1523 
1524 static void xive_end_source_realize(DeviceState *dev, Error **errp)
1525 {
1526     XiveENDSource *xsrc = XIVE_END_SOURCE(dev);
1527     Object *obj;
1528     Error *local_err = NULL;
1529 
1530     obj = object_property_get_link(OBJECT(dev), "xive", &local_err);
1531     if (!obj) {
1532         error_propagate(errp, local_err);
1533         error_prepend(errp, "required link 'xive' not found: ");
1534         return;
1535     }
1536 
1537     xsrc->xrtr = XIVE_ROUTER(obj);
1538 
1539     if (!xsrc->nr_ends) {
1540         error_setg(errp, "Number of interrupt needs to be greater than 0");
1541         return;
1542     }
1543 
1544     if (xsrc->esb_shift != XIVE_ESB_4K &&
1545         xsrc->esb_shift != XIVE_ESB_64K) {
1546         error_setg(errp, "Invalid ESB shift setting");
1547         return;
1548     }
1549 
1550     /*
1551      * Each END is assigned an even/odd pair of MMIO pages, the even page
1552      * manages the ESn field while the odd page manages the ESe field.
1553      */
1554     memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
1555                           &xive_end_source_ops, xsrc, "xive.end",
1556                           (1ull << (xsrc->esb_shift + 1)) * xsrc->nr_ends);
1557 }
1558 
1559 static Property xive_end_source_properties[] = {
1560     DEFINE_PROP_UINT8("block-id", XiveENDSource, block_id, 0),
1561     DEFINE_PROP_UINT32("nr-ends", XiveENDSource, nr_ends, 0),
1562     DEFINE_PROP_UINT32("shift", XiveENDSource, esb_shift, XIVE_ESB_64K),
1563     DEFINE_PROP_END_OF_LIST(),
1564 };
1565 
1566 static void xive_end_source_class_init(ObjectClass *klass, void *data)
1567 {
1568     DeviceClass *dc = DEVICE_CLASS(klass);
1569 
1570     dc->desc    = "XIVE END Source";
1571     dc->props   = xive_end_source_properties;
1572     dc->realize = xive_end_source_realize;
1573 }
1574 
1575 static const TypeInfo xive_end_source_info = {
1576     .name          = TYPE_XIVE_END_SOURCE,
1577     .parent        = TYPE_DEVICE,
1578     .instance_size = sizeof(XiveENDSource),
1579     .class_init    = xive_end_source_class_init,
1580 };
1581 
1582 /*
1583  * XIVE Notifier
1584  */
1585 static const TypeInfo xive_notifier_info = {
1586     .name = TYPE_XIVE_NOTIFIER,
1587     .parent = TYPE_INTERFACE,
1588     .class_size = sizeof(XiveNotifierClass),
1589 };
1590 
1591 static void xive_register_types(void)
1592 {
1593     type_register_static(&xive_source_info);
1594     type_register_static(&xive_notifier_info);
1595     type_register_static(&xive_router_info);
1596     type_register_static(&xive_end_source_info);
1597     type_register_static(&xive_tctx_info);
1598 }
1599 
1600 type_init(xive_register_types)
1601