xref: /openbmc/qemu/hw/intc/xive.c (revision ca693d1c)
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     case TM_QW3_HV_PHYS:
58         return TM_QW3_NSR_HE;
59     default:
60         g_assert_not_reached();
61     }
62 }
63 
64 static uint64_t xive_tctx_accept(XiveTCTX *tctx, uint8_t ring)
65 {
66     uint8_t *regs = &tctx->regs[ring];
67     uint8_t nsr = regs[TM_NSR];
68     uint8_t mask = exception_mask(ring);
69 
70     qemu_irq_lower(tctx->output);
71 
72     if (regs[TM_NSR] & mask) {
73         uint8_t cppr = regs[TM_PIPR];
74 
75         regs[TM_CPPR] = cppr;
76 
77         /* Reset the pending buffer bit */
78         regs[TM_IPB] &= ~priority_to_ipb(cppr);
79         regs[TM_PIPR] = ipb_to_pipr(regs[TM_IPB]);
80 
81         /* Drop Exception bit */
82         regs[TM_NSR] &= ~mask;
83     }
84 
85     return (nsr << 8) | regs[TM_CPPR];
86 }
87 
88 static void xive_tctx_notify(XiveTCTX *tctx, uint8_t ring)
89 {
90     uint8_t *regs = &tctx->regs[ring];
91 
92     if (regs[TM_PIPR] < regs[TM_CPPR]) {
93         switch (ring) {
94         case TM_QW1_OS:
95             regs[TM_NSR] |= TM_QW1_NSR_EO;
96             break;
97         case TM_QW3_HV_PHYS:
98             regs[TM_NSR] |= (TM_QW3_NSR_HE_PHYS << 6);
99             break;
100         default:
101             g_assert_not_reached();
102         }
103         qemu_irq_raise(tctx->output);
104     }
105 }
106 
107 static void xive_tctx_set_cppr(XiveTCTX *tctx, uint8_t ring, uint8_t cppr)
108 {
109     if (cppr > XIVE_PRIORITY_MAX) {
110         cppr = 0xff;
111     }
112 
113     tctx->regs[ring + TM_CPPR] = cppr;
114 
115     /* CPPR has changed, check if we need to raise a pending exception */
116     xive_tctx_notify(tctx, ring);
117 }
118 
119 /*
120  * XIVE Thread Interrupt Management Area (TIMA)
121  */
122 
123 static void xive_tm_set_hv_cppr(XiveTCTX *tctx, hwaddr offset,
124                                 uint64_t value, unsigned size)
125 {
126     xive_tctx_set_cppr(tctx, TM_QW3_HV_PHYS, value & 0xff);
127 }
128 
129 static uint64_t xive_tm_ack_hv_reg(XiveTCTX *tctx, hwaddr offset, unsigned size)
130 {
131     return xive_tctx_accept(tctx, TM_QW3_HV_PHYS);
132 }
133 
134 static uint64_t xive_tm_pull_pool_ctx(XiveTCTX *tctx, hwaddr offset,
135                                       unsigned size)
136 {
137     uint64_t ret;
138 
139     ret = tctx->regs[TM_QW2_HV_POOL + TM_WORD2] & TM_QW2W2_POOL_CAM;
140     tctx->regs[TM_QW2_HV_POOL + TM_WORD2] &= ~TM_QW2W2_POOL_CAM;
141     return ret;
142 }
143 
144 static void xive_tm_vt_push(XiveTCTX *tctx, hwaddr offset,
145                             uint64_t value, unsigned size)
146 {
147     tctx->regs[TM_QW3_HV_PHYS + TM_WORD2] = value & 0xff;
148 }
149 
150 static uint64_t xive_tm_vt_poll(XiveTCTX *tctx, hwaddr offset, unsigned size)
151 {
152     return tctx->regs[TM_QW3_HV_PHYS + TM_WORD2] & 0xff;
153 }
154 
155 /*
156  * Define an access map for each page of the TIMA that we will use in
157  * the memory region ops to filter values when doing loads and stores
158  * of raw registers values
159  *
160  * Registers accessibility bits :
161  *
162  *    0x0 - no access
163  *    0x1 - write only
164  *    0x2 - read only
165  *    0x3 - read/write
166  */
167 
168 static const uint8_t xive_tm_hw_view[] = {
169     /* QW-0 User */   3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0,
170     /* QW-1 OS   */   3, 3, 3, 3,   3, 3, 0, 3,   3, 3, 3, 3,   0, 0, 0, 0,
171     /* QW-2 POOL */   0, 0, 3, 3,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0,
172     /* QW-3 PHYS */   3, 3, 3, 3,   0, 3, 0, 3,   3, 0, 0, 3,   3, 3, 3, 0,
173 };
174 
175 static const uint8_t xive_tm_hv_view[] = {
176     /* QW-0 User */   3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0,
177     /* QW-1 OS   */   3, 3, 3, 3,   3, 3, 0, 3,   3, 3, 3, 3,   0, 0, 0, 0,
178     /* QW-2 POOL */   0, 0, 3, 3,   0, 0, 0, 0,   0, 3, 3, 3,   0, 0, 0, 0,
179     /* QW-3 PHYS */   3, 3, 3, 3,   0, 3, 0, 3,   3, 0, 0, 3,   0, 0, 0, 0,
180 };
181 
182 static const uint8_t xive_tm_os_view[] = {
183     /* QW-0 User */   3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0,
184     /* QW-1 OS   */   2, 3, 2, 2,   2, 2, 0, 2,   0, 0, 0, 0,   0, 0, 0, 0,
185     /* QW-2 POOL */   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,
186     /* QW-3 PHYS */   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,
187 };
188 
189 static const uint8_t xive_tm_user_view[] = {
190     /* QW-0 User */   3, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,
191     /* QW-1 OS   */   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,
192     /* QW-2 POOL */   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,
193     /* QW-3 PHYS */   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,
194 };
195 
196 /*
197  * Overall TIMA access map for the thread interrupt management context
198  * registers
199  */
200 static const uint8_t *xive_tm_views[] = {
201     [XIVE_TM_HW_PAGE]   = xive_tm_hw_view,
202     [XIVE_TM_HV_PAGE]   = xive_tm_hv_view,
203     [XIVE_TM_OS_PAGE]   = xive_tm_os_view,
204     [XIVE_TM_USER_PAGE] = xive_tm_user_view,
205 };
206 
207 /*
208  * Computes a register access mask for a given offset in the TIMA
209  */
210 static uint64_t xive_tm_mask(hwaddr offset, unsigned size, bool write)
211 {
212     uint8_t page_offset = (offset >> TM_SHIFT) & 0x3;
213     uint8_t reg_offset = offset & 0x3F;
214     uint8_t reg_mask = write ? 0x1 : 0x2;
215     uint64_t mask = 0x0;
216     int i;
217 
218     for (i = 0; i < size; i++) {
219         if (xive_tm_views[page_offset][reg_offset + i] & reg_mask) {
220             mask |= (uint64_t) 0xff << (8 * (size - i - 1));
221         }
222     }
223 
224     return mask;
225 }
226 
227 static void xive_tm_raw_write(XiveTCTX *tctx, hwaddr offset, uint64_t value,
228                               unsigned size)
229 {
230     uint8_t ring_offset = offset & 0x30;
231     uint8_t reg_offset = offset & 0x3F;
232     uint64_t mask = xive_tm_mask(offset, size, true);
233     int i;
234 
235     /*
236      * Only 4 or 8 bytes stores are allowed and the User ring is
237      * excluded
238      */
239     if (size < 4 || !mask || ring_offset == TM_QW0_USER) {
240         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid write access at TIMA @%"
241                       HWADDR_PRIx"\n", offset);
242         return;
243     }
244 
245     /*
246      * Use the register offset for the raw values and filter out
247      * reserved values
248      */
249     for (i = 0; i < size; i++) {
250         uint8_t byte_mask = (mask >> (8 * (size - i - 1)));
251         if (byte_mask) {
252             tctx->regs[reg_offset + i] = (value >> (8 * (size - i - 1))) &
253                 byte_mask;
254         }
255     }
256 }
257 
258 static uint64_t xive_tm_raw_read(XiveTCTX *tctx, hwaddr offset, unsigned size)
259 {
260     uint8_t ring_offset = offset & 0x30;
261     uint8_t reg_offset = offset & 0x3F;
262     uint64_t mask = xive_tm_mask(offset, size, false);
263     uint64_t ret;
264     int i;
265 
266     /*
267      * Only 4 or 8 bytes loads are allowed and the User ring is
268      * excluded
269      */
270     if (size < 4 || !mask || ring_offset == TM_QW0_USER) {
271         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid read access at TIMA @%"
272                       HWADDR_PRIx"\n", offset);
273         return -1;
274     }
275 
276     /* Use the register offset for the raw values */
277     ret = 0;
278     for (i = 0; i < size; i++) {
279         ret |= (uint64_t) tctx->regs[reg_offset + i] << (8 * (size - i - 1));
280     }
281 
282     /* filter out reserved values */
283     return ret & mask;
284 }
285 
286 /*
287  * The TM context is mapped twice within each page. Stores and loads
288  * to the first mapping below 2K write and read the specified values
289  * without modification. The second mapping above 2K performs specific
290  * state changes (side effects) in addition to setting/returning the
291  * interrupt management area context of the processor thread.
292  */
293 static uint64_t xive_tm_ack_os_reg(XiveTCTX *tctx, hwaddr offset, unsigned size)
294 {
295     return xive_tctx_accept(tctx, TM_QW1_OS);
296 }
297 
298 static void xive_tm_set_os_cppr(XiveTCTX *tctx, hwaddr offset,
299                                 uint64_t value, unsigned size)
300 {
301     xive_tctx_set_cppr(tctx, TM_QW1_OS, value & 0xff);
302 }
303 
304 /*
305  * Adjust the IPB to allow a CPU to process event queues of other
306  * priorities during one physical interrupt cycle.
307  */
308 static void xive_tm_set_os_pending(XiveTCTX *tctx, hwaddr offset,
309                                    uint64_t value, unsigned size)
310 {
311     ipb_update(&tctx->regs[TM_QW1_OS], value & 0xff);
312     xive_tctx_notify(tctx, TM_QW1_OS);
313 }
314 
315 /*
316  * Define a mapping of "special" operations depending on the TIMA page
317  * offset and the size of the operation.
318  */
319 typedef struct XiveTmOp {
320     uint8_t  page_offset;
321     uint32_t op_offset;
322     unsigned size;
323     void     (*write_handler)(XiveTCTX *tctx, hwaddr offset, uint64_t value,
324                               unsigned size);
325     uint64_t (*read_handler)(XiveTCTX *tctx, hwaddr offset, unsigned size);
326 } XiveTmOp;
327 
328 static const XiveTmOp xive_tm_operations[] = {
329     /*
330      * MMIOs below 2K : raw values and special operations without side
331      * effects
332      */
333     { XIVE_TM_OS_PAGE, TM_QW1_OS + TM_CPPR,   1, xive_tm_set_os_cppr, NULL },
334     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_CPPR, 1, xive_tm_set_hv_cppr, NULL },
335     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_WORD2, 1, xive_tm_vt_push, NULL },
336     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_WORD2, 1, NULL, xive_tm_vt_poll },
337 
338     /* MMIOs above 2K : special operations with side effects */
339     { XIVE_TM_OS_PAGE, TM_SPC_ACK_OS_REG,     2, NULL, xive_tm_ack_os_reg },
340     { XIVE_TM_OS_PAGE, TM_SPC_SET_OS_PENDING, 1, xive_tm_set_os_pending, NULL },
341     { XIVE_TM_HV_PAGE, TM_SPC_ACK_HV_REG,     2, NULL, xive_tm_ack_hv_reg },
342     { XIVE_TM_HV_PAGE, TM_SPC_PULL_POOL_CTX,  4, NULL, xive_tm_pull_pool_ctx },
343     { XIVE_TM_HV_PAGE, TM_SPC_PULL_POOL_CTX,  8, NULL, xive_tm_pull_pool_ctx },
344 };
345 
346 static const XiveTmOp *xive_tm_find_op(hwaddr offset, unsigned size, bool write)
347 {
348     uint8_t page_offset = (offset >> TM_SHIFT) & 0x3;
349     uint32_t op_offset = offset & 0xFFF;
350     int i;
351 
352     for (i = 0; i < ARRAY_SIZE(xive_tm_operations); i++) {
353         const XiveTmOp *xto = &xive_tm_operations[i];
354 
355         /* Accesses done from a more privileged TIMA page is allowed */
356         if (xto->page_offset >= page_offset &&
357             xto->op_offset == op_offset &&
358             xto->size == size &&
359             ((write && xto->write_handler) || (!write && xto->read_handler))) {
360             return xto;
361         }
362     }
363     return NULL;
364 }
365 
366 /*
367  * TIMA MMIO handlers
368  */
369 void xive_tctx_tm_write(XiveTCTX *tctx, hwaddr offset, uint64_t value,
370                         unsigned size)
371 {
372     const XiveTmOp *xto;
373 
374     /*
375      * TODO: check V bit in Q[0-3]W2
376      */
377 
378     /*
379      * First, check for special operations in the 2K region
380      */
381     if (offset & 0x800) {
382         xto = xive_tm_find_op(offset, size, true);
383         if (!xto) {
384             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid write access at TIMA"
385                           "@%"HWADDR_PRIx"\n", offset);
386         } else {
387             xto->write_handler(tctx, offset, value, size);
388         }
389         return;
390     }
391 
392     /*
393      * Then, for special operations in the region below 2K.
394      */
395     xto = xive_tm_find_op(offset, size, true);
396     if (xto) {
397         xto->write_handler(tctx, offset, value, size);
398         return;
399     }
400 
401     /*
402      * Finish with raw access to the register values
403      */
404     xive_tm_raw_write(tctx, offset, value, size);
405 }
406 
407 uint64_t xive_tctx_tm_read(XiveTCTX *tctx, hwaddr offset, unsigned size)
408 {
409     const XiveTmOp *xto;
410 
411     /*
412      * TODO: check V bit in Q[0-3]W2
413      */
414 
415     /*
416      * First, check for special operations in the 2K region
417      */
418     if (offset & 0x800) {
419         xto = xive_tm_find_op(offset, size, false);
420         if (!xto) {
421             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid read access to TIMA"
422                           "@%"HWADDR_PRIx"\n", offset);
423             return -1;
424         }
425         return xto->read_handler(tctx, offset, size);
426     }
427 
428     /*
429      * Then, for special operations in the region below 2K.
430      */
431     xto = xive_tm_find_op(offset, size, false);
432     if (xto) {
433         return xto->read_handler(tctx, offset, size);
434     }
435 
436     /*
437      * Finish with raw access to the register values
438      */
439     return xive_tm_raw_read(tctx, offset, size);
440 }
441 
442 static void xive_tm_write(void *opaque, hwaddr offset,
443                           uint64_t value, unsigned size)
444 {
445     XiveTCTX *tctx = xive_router_get_tctx(XIVE_ROUTER(opaque), current_cpu);
446 
447     xive_tctx_tm_write(tctx, offset, value, size);
448 }
449 
450 static uint64_t xive_tm_read(void *opaque, hwaddr offset, unsigned size)
451 {
452     XiveTCTX *tctx = xive_router_get_tctx(XIVE_ROUTER(opaque), current_cpu);
453 
454     return xive_tctx_tm_read(tctx, offset, size);
455 }
456 
457 const MemoryRegionOps xive_tm_ops = {
458     .read = xive_tm_read,
459     .write = xive_tm_write,
460     .endianness = DEVICE_BIG_ENDIAN,
461     .valid = {
462         .min_access_size = 1,
463         .max_access_size = 8,
464     },
465     .impl = {
466         .min_access_size = 1,
467         .max_access_size = 8,
468     },
469 };
470 
471 static inline uint32_t xive_tctx_word2(uint8_t *ring)
472 {
473     return *((uint32_t *) &ring[TM_WORD2]);
474 }
475 
476 static char *xive_tctx_ring_print(uint8_t *ring)
477 {
478     uint32_t w2 = xive_tctx_word2(ring);
479 
480     return g_strdup_printf("%02x   %02x  %02x    %02x   %02x  "
481                    "%02x  %02x   %02x  %08x",
482                    ring[TM_NSR], ring[TM_CPPR], ring[TM_IPB], ring[TM_LSMFB],
483                    ring[TM_ACK_CNT], ring[TM_INC], ring[TM_AGE], ring[TM_PIPR],
484                    be32_to_cpu(w2));
485 }
486 
487 static const char * const xive_tctx_ring_names[] = {
488     "USER", "OS", "POOL", "PHYS",
489 };
490 
491 void xive_tctx_pic_print_info(XiveTCTX *tctx, Monitor *mon)
492 {
493     int cpu_index = tctx->cs ? tctx->cs->cpu_index : -1;
494     int i;
495 
496     if (kvm_irqchip_in_kernel()) {
497         Error *local_err = NULL;
498 
499         kvmppc_xive_cpu_synchronize_state(tctx, &local_err);
500         if (local_err) {
501             error_report_err(local_err);
502             return;
503         }
504     }
505 
506     monitor_printf(mon, "CPU[%04x]:   QW   NSR CPPR IPB LSMFB ACK# INC AGE PIPR"
507                    "  W2\n", cpu_index);
508 
509     for (i = 0; i < XIVE_TM_RING_COUNT; i++) {
510         char *s = xive_tctx_ring_print(&tctx->regs[i * XIVE_TM_RING_SIZE]);
511         monitor_printf(mon, "CPU[%04x]: %4s    %s\n", cpu_index,
512                        xive_tctx_ring_names[i], s);
513         g_free(s);
514     }
515 }
516 
517 static void xive_tctx_reset(void *dev)
518 {
519     XiveTCTX *tctx = XIVE_TCTX(dev);
520 
521     memset(tctx->regs, 0, sizeof(tctx->regs));
522 
523     /* Set some defaults */
524     tctx->regs[TM_QW1_OS + TM_LSMFB] = 0xFF;
525     tctx->regs[TM_QW1_OS + TM_ACK_CNT] = 0xFF;
526     tctx->regs[TM_QW1_OS + TM_AGE] = 0xFF;
527 
528     /*
529      * Initialize PIPR to 0xFF to avoid phantom interrupts when the
530      * CPPR is first set.
531      */
532     tctx->regs[TM_QW1_OS + TM_PIPR] =
533         ipb_to_pipr(tctx->regs[TM_QW1_OS + TM_IPB]);
534     tctx->regs[TM_QW3_HV_PHYS + TM_PIPR] =
535         ipb_to_pipr(tctx->regs[TM_QW3_HV_PHYS + TM_IPB]);
536 }
537 
538 static void xive_tctx_realize(DeviceState *dev, Error **errp)
539 {
540     XiveTCTX *tctx = XIVE_TCTX(dev);
541     PowerPCCPU *cpu;
542     CPUPPCState *env;
543     Object *obj;
544     Error *local_err = NULL;
545 
546     obj = object_property_get_link(OBJECT(dev), "cpu", &local_err);
547     if (!obj) {
548         error_propagate(errp, local_err);
549         error_prepend(errp, "required link 'cpu' not found: ");
550         return;
551     }
552 
553     cpu = POWERPC_CPU(obj);
554     tctx->cs = CPU(obj);
555 
556     env = &cpu->env;
557     switch (PPC_INPUT(env)) {
558     case PPC_FLAGS_INPUT_POWER9:
559         tctx->output = env->irq_inputs[POWER9_INPUT_INT];
560         break;
561 
562     default:
563         error_setg(errp, "XIVE interrupt controller does not support "
564                    "this CPU bus model");
565         return;
566     }
567 
568     /* Connect the presenter to the VCPU (required for CPU hotplug) */
569     if (kvm_irqchip_in_kernel()) {
570         kvmppc_xive_cpu_connect(tctx, &local_err);
571         if (local_err) {
572             error_propagate(errp, local_err);
573             return;
574         }
575     }
576 
577     qemu_register_reset(xive_tctx_reset, dev);
578 }
579 
580 static void xive_tctx_unrealize(DeviceState *dev, Error **errp)
581 {
582     qemu_unregister_reset(xive_tctx_reset, dev);
583 }
584 
585 static int vmstate_xive_tctx_pre_save(void *opaque)
586 {
587     Error *local_err = NULL;
588 
589     if (kvm_irqchip_in_kernel()) {
590         kvmppc_xive_cpu_get_state(XIVE_TCTX(opaque), &local_err);
591         if (local_err) {
592             error_report_err(local_err);
593             return -1;
594         }
595     }
596 
597     return 0;
598 }
599 
600 static const VMStateDescription vmstate_xive_tctx = {
601     .name = TYPE_XIVE_TCTX,
602     .version_id = 1,
603     .minimum_version_id = 1,
604     .pre_save = vmstate_xive_tctx_pre_save,
605     .post_load = NULL, /* handled by the sPAPRxive model */
606     .fields = (VMStateField[]) {
607         VMSTATE_BUFFER(regs, XiveTCTX),
608         VMSTATE_END_OF_LIST()
609     },
610 };
611 
612 static void xive_tctx_class_init(ObjectClass *klass, void *data)
613 {
614     DeviceClass *dc = DEVICE_CLASS(klass);
615 
616     dc->desc = "XIVE Interrupt Thread Context";
617     dc->realize = xive_tctx_realize;
618     dc->unrealize = xive_tctx_unrealize;
619     dc->vmsd = &vmstate_xive_tctx;
620 }
621 
622 static const TypeInfo xive_tctx_info = {
623     .name          = TYPE_XIVE_TCTX,
624     .parent        = TYPE_DEVICE,
625     .instance_size = sizeof(XiveTCTX),
626     .class_init    = xive_tctx_class_init,
627 };
628 
629 Object *xive_tctx_create(Object *cpu, XiveRouter *xrtr, Error **errp)
630 {
631     Error *local_err = NULL;
632     Object *obj;
633 
634     obj = object_new(TYPE_XIVE_TCTX);
635     object_property_add_child(cpu, TYPE_XIVE_TCTX, obj, &error_abort);
636     object_unref(obj);
637     object_property_add_const_link(obj, "cpu", cpu, &error_abort);
638     object_property_set_bool(obj, true, "realized", &local_err);
639     if (local_err) {
640         goto error;
641     }
642 
643     return obj;
644 
645 error:
646     object_unparent(obj);
647     error_propagate(errp, local_err);
648     return NULL;
649 }
650 
651 /*
652  * XIVE ESB helpers
653  */
654 
655 static uint8_t xive_esb_set(uint8_t *pq, uint8_t value)
656 {
657     uint8_t old_pq = *pq & 0x3;
658 
659     *pq &= ~0x3;
660     *pq |= value & 0x3;
661 
662     return old_pq;
663 }
664 
665 static bool xive_esb_trigger(uint8_t *pq)
666 {
667     uint8_t old_pq = *pq & 0x3;
668 
669     switch (old_pq) {
670     case XIVE_ESB_RESET:
671         xive_esb_set(pq, XIVE_ESB_PENDING);
672         return true;
673     case XIVE_ESB_PENDING:
674     case XIVE_ESB_QUEUED:
675         xive_esb_set(pq, XIVE_ESB_QUEUED);
676         return false;
677     case XIVE_ESB_OFF:
678         xive_esb_set(pq, XIVE_ESB_OFF);
679         return false;
680     default:
681          g_assert_not_reached();
682     }
683 }
684 
685 static bool xive_esb_eoi(uint8_t *pq)
686 {
687     uint8_t old_pq = *pq & 0x3;
688 
689     switch (old_pq) {
690     case XIVE_ESB_RESET:
691     case XIVE_ESB_PENDING:
692         xive_esb_set(pq, XIVE_ESB_RESET);
693         return false;
694     case XIVE_ESB_QUEUED:
695         xive_esb_set(pq, XIVE_ESB_PENDING);
696         return true;
697     case XIVE_ESB_OFF:
698         xive_esb_set(pq, XIVE_ESB_OFF);
699         return false;
700     default:
701          g_assert_not_reached();
702     }
703 }
704 
705 /*
706  * XIVE Interrupt Source (or IVSE)
707  */
708 
709 uint8_t xive_source_esb_get(XiveSource *xsrc, uint32_t srcno)
710 {
711     assert(srcno < xsrc->nr_irqs);
712 
713     return xsrc->status[srcno] & 0x3;
714 }
715 
716 uint8_t xive_source_esb_set(XiveSource *xsrc, uint32_t srcno, uint8_t pq)
717 {
718     assert(srcno < xsrc->nr_irqs);
719 
720     return xive_esb_set(&xsrc->status[srcno], pq);
721 }
722 
723 /*
724  * Returns whether the event notification should be forwarded.
725  */
726 static bool xive_source_lsi_trigger(XiveSource *xsrc, uint32_t srcno)
727 {
728     uint8_t old_pq = xive_source_esb_get(xsrc, srcno);
729 
730     xsrc->status[srcno] |= XIVE_STATUS_ASSERTED;
731 
732     switch (old_pq) {
733     case XIVE_ESB_RESET:
734         xive_source_esb_set(xsrc, srcno, XIVE_ESB_PENDING);
735         return true;
736     default:
737         return false;
738     }
739 }
740 
741 /*
742  * Returns whether the event notification should be forwarded.
743  */
744 static bool xive_source_esb_trigger(XiveSource *xsrc, uint32_t srcno)
745 {
746     bool ret;
747 
748     assert(srcno < xsrc->nr_irqs);
749 
750     ret = xive_esb_trigger(&xsrc->status[srcno]);
751 
752     if (xive_source_irq_is_lsi(xsrc, srcno) &&
753         xive_source_esb_get(xsrc, srcno) == XIVE_ESB_QUEUED) {
754         qemu_log_mask(LOG_GUEST_ERROR,
755                       "XIVE: queued an event on LSI IRQ %d\n", srcno);
756     }
757 
758     return ret;
759 }
760 
761 /*
762  * Returns whether the event notification should be forwarded.
763  */
764 static bool xive_source_esb_eoi(XiveSource *xsrc, uint32_t srcno)
765 {
766     bool ret;
767 
768     assert(srcno < xsrc->nr_irqs);
769 
770     ret = xive_esb_eoi(&xsrc->status[srcno]);
771 
772     /*
773      * LSI sources do not set the Q bit but they can still be
774      * asserted, in which case we should forward a new event
775      * notification
776      */
777     if (xive_source_irq_is_lsi(xsrc, srcno) &&
778         xsrc->status[srcno] & XIVE_STATUS_ASSERTED) {
779         ret = xive_source_lsi_trigger(xsrc, srcno);
780     }
781 
782     return ret;
783 }
784 
785 /*
786  * Forward the source event notification to the Router
787  */
788 static void xive_source_notify(XiveSource *xsrc, int srcno)
789 {
790     XiveNotifierClass *xnc = XIVE_NOTIFIER_GET_CLASS(xsrc->xive);
791 
792     if (xnc->notify) {
793         xnc->notify(xsrc->xive, srcno);
794     }
795 }
796 
797 /*
798  * In a two pages ESB MMIO setting, even page is the trigger page, odd
799  * page is for management
800  */
801 static inline bool addr_is_even(hwaddr addr, uint32_t shift)
802 {
803     return !((addr >> shift) & 1);
804 }
805 
806 static inline bool xive_source_is_trigger_page(XiveSource *xsrc, hwaddr addr)
807 {
808     return xive_source_esb_has_2page(xsrc) &&
809         addr_is_even(addr, xsrc->esb_shift - 1);
810 }
811 
812 /*
813  * ESB MMIO loads
814  *                      Trigger page    Management/EOI page
815  *
816  * ESB MMIO setting     2 pages         1 or 2 pages
817  *
818  * 0x000 .. 0x3FF       -1              EOI and return 0|1
819  * 0x400 .. 0x7FF       -1              EOI and return 0|1
820  * 0x800 .. 0xBFF       -1              return PQ
821  * 0xC00 .. 0xCFF       -1              return PQ and atomically PQ=00
822  * 0xD00 .. 0xDFF       -1              return PQ and atomically PQ=01
823  * 0xE00 .. 0xDFF       -1              return PQ and atomically PQ=10
824  * 0xF00 .. 0xDFF       -1              return PQ and atomically PQ=11
825  */
826 static uint64_t xive_source_esb_read(void *opaque, hwaddr addr, unsigned size)
827 {
828     XiveSource *xsrc = XIVE_SOURCE(opaque);
829     uint32_t offset = addr & 0xFFF;
830     uint32_t srcno = addr >> xsrc->esb_shift;
831     uint64_t ret = -1;
832 
833     /* In a two pages ESB MMIO setting, trigger page should not be read */
834     if (xive_source_is_trigger_page(xsrc, addr)) {
835         qemu_log_mask(LOG_GUEST_ERROR,
836                       "XIVE: invalid load on IRQ %d trigger page at "
837                       "0x%"HWADDR_PRIx"\n", srcno, addr);
838         return -1;
839     }
840 
841     switch (offset) {
842     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
843         ret = xive_source_esb_eoi(xsrc, srcno);
844 
845         /* Forward the source event notification for routing */
846         if (ret) {
847             xive_source_notify(xsrc, srcno);
848         }
849         break;
850 
851     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
852         ret = xive_source_esb_get(xsrc, srcno);
853         break;
854 
855     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
856     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
857     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
858     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
859         ret = xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
860         break;
861     default:
862         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB load addr %x\n",
863                       offset);
864     }
865 
866     return ret;
867 }
868 
869 /*
870  * ESB MMIO stores
871  *                      Trigger page    Management/EOI page
872  *
873  * ESB MMIO setting     2 pages         1 or 2 pages
874  *
875  * 0x000 .. 0x3FF       Trigger         Trigger
876  * 0x400 .. 0x7FF       Trigger         EOI
877  * 0x800 .. 0xBFF       Trigger         undefined
878  * 0xC00 .. 0xCFF       Trigger         PQ=00
879  * 0xD00 .. 0xDFF       Trigger         PQ=01
880  * 0xE00 .. 0xDFF       Trigger         PQ=10
881  * 0xF00 .. 0xDFF       Trigger         PQ=11
882  */
883 static void xive_source_esb_write(void *opaque, hwaddr addr,
884                                   uint64_t value, unsigned size)
885 {
886     XiveSource *xsrc = XIVE_SOURCE(opaque);
887     uint32_t offset = addr & 0xFFF;
888     uint32_t srcno = addr >> xsrc->esb_shift;
889     bool notify = false;
890 
891     /* In a two pages ESB MMIO setting, trigger page only triggers */
892     if (xive_source_is_trigger_page(xsrc, addr)) {
893         notify = xive_source_esb_trigger(xsrc, srcno);
894         goto out;
895     }
896 
897     switch (offset) {
898     case 0 ... 0x3FF:
899         notify = xive_source_esb_trigger(xsrc, srcno);
900         break;
901 
902     case XIVE_ESB_STORE_EOI ... XIVE_ESB_STORE_EOI + 0x3FF:
903         if (!(xsrc->esb_flags & XIVE_SRC_STORE_EOI)) {
904             qemu_log_mask(LOG_GUEST_ERROR,
905                           "XIVE: invalid Store EOI for IRQ %d\n", srcno);
906             return;
907         }
908 
909         notify = xive_source_esb_eoi(xsrc, srcno);
910         break;
911 
912     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
913     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
914     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
915     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
916         xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
917         break;
918 
919     default:
920         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr %x\n",
921                       offset);
922         return;
923     }
924 
925 out:
926     /* Forward the source event notification for routing */
927     if (notify) {
928         xive_source_notify(xsrc, srcno);
929     }
930 }
931 
932 static const MemoryRegionOps xive_source_esb_ops = {
933     .read = xive_source_esb_read,
934     .write = xive_source_esb_write,
935     .endianness = DEVICE_BIG_ENDIAN,
936     .valid = {
937         .min_access_size = 8,
938         .max_access_size = 8,
939     },
940     .impl = {
941         .min_access_size = 8,
942         .max_access_size = 8,
943     },
944 };
945 
946 void xive_source_set_irq(void *opaque, int srcno, int val)
947 {
948     XiveSource *xsrc = XIVE_SOURCE(opaque);
949     bool notify = false;
950 
951     if (xive_source_irq_is_lsi(xsrc, srcno)) {
952         if (val) {
953             notify = xive_source_lsi_trigger(xsrc, srcno);
954         } else {
955             xsrc->status[srcno] &= ~XIVE_STATUS_ASSERTED;
956         }
957     } else {
958         if (val) {
959             notify = xive_source_esb_trigger(xsrc, srcno);
960         }
961     }
962 
963     /* Forward the source event notification for routing */
964     if (notify) {
965         xive_source_notify(xsrc, srcno);
966     }
967 }
968 
969 void xive_source_pic_print_info(XiveSource *xsrc, uint32_t offset, Monitor *mon)
970 {
971     int i;
972 
973     for (i = 0; i < xsrc->nr_irqs; i++) {
974         uint8_t pq = xive_source_esb_get(xsrc, i);
975 
976         if (pq == XIVE_ESB_OFF) {
977             continue;
978         }
979 
980         monitor_printf(mon, "  %08x %s %c%c%c\n", i + offset,
981                        xive_source_irq_is_lsi(xsrc, i) ? "LSI" : "MSI",
982                        pq & XIVE_ESB_VAL_P ? 'P' : '-',
983                        pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
984                        xsrc->status[i] & XIVE_STATUS_ASSERTED ? 'A' : ' ');
985     }
986 }
987 
988 static void xive_source_reset(void *dev)
989 {
990     XiveSource *xsrc = XIVE_SOURCE(dev);
991 
992     /* Do not clear the LSI bitmap */
993 
994     /* PQs are initialized to 0b01 (Q=1) which corresponds to "ints off" */
995     memset(xsrc->status, XIVE_ESB_OFF, xsrc->nr_irqs);
996 }
997 
998 static void xive_source_realize(DeviceState *dev, Error **errp)
999 {
1000     XiveSource *xsrc = XIVE_SOURCE(dev);
1001     Object *obj;
1002     Error *local_err = NULL;
1003 
1004     obj = object_property_get_link(OBJECT(dev), "xive", &local_err);
1005     if (!obj) {
1006         error_propagate(errp, local_err);
1007         error_prepend(errp, "required link 'xive' not found: ");
1008         return;
1009     }
1010 
1011     xsrc->xive = XIVE_NOTIFIER(obj);
1012 
1013     if (!xsrc->nr_irqs) {
1014         error_setg(errp, "Number of interrupt needs to be greater than 0");
1015         return;
1016     }
1017 
1018     if (xsrc->esb_shift != XIVE_ESB_4K &&
1019         xsrc->esb_shift != XIVE_ESB_4K_2PAGE &&
1020         xsrc->esb_shift != XIVE_ESB_64K &&
1021         xsrc->esb_shift != XIVE_ESB_64K_2PAGE) {
1022         error_setg(errp, "Invalid ESB shift setting");
1023         return;
1024     }
1025 
1026     xsrc->status = g_malloc0(xsrc->nr_irqs);
1027     xsrc->lsi_map = bitmap_new(xsrc->nr_irqs);
1028 
1029     if (!kvm_irqchip_in_kernel()) {
1030         memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
1031                               &xive_source_esb_ops, xsrc, "xive.esb",
1032                               (1ull << xsrc->esb_shift) * xsrc->nr_irqs);
1033     }
1034 
1035     qemu_register_reset(xive_source_reset, dev);
1036 }
1037 
1038 static const VMStateDescription vmstate_xive_source = {
1039     .name = TYPE_XIVE_SOURCE,
1040     .version_id = 1,
1041     .minimum_version_id = 1,
1042     .fields = (VMStateField[]) {
1043         VMSTATE_UINT32_EQUAL(nr_irqs, XiveSource, NULL),
1044         VMSTATE_VBUFFER_UINT32(status, XiveSource, 1, NULL, nr_irqs),
1045         VMSTATE_END_OF_LIST()
1046     },
1047 };
1048 
1049 /*
1050  * The default XIVE interrupt source setting for the ESB MMIOs is two
1051  * 64k pages without Store EOI, to be in sync with KVM.
1052  */
1053 static Property xive_source_properties[] = {
1054     DEFINE_PROP_UINT64("flags", XiveSource, esb_flags, 0),
1055     DEFINE_PROP_UINT32("nr-irqs", XiveSource, nr_irqs, 0),
1056     DEFINE_PROP_UINT32("shift", XiveSource, esb_shift, XIVE_ESB_64K_2PAGE),
1057     DEFINE_PROP_END_OF_LIST(),
1058 };
1059 
1060 static void xive_source_class_init(ObjectClass *klass, void *data)
1061 {
1062     DeviceClass *dc = DEVICE_CLASS(klass);
1063 
1064     dc->desc    = "XIVE Interrupt Source";
1065     dc->props   = xive_source_properties;
1066     dc->realize = xive_source_realize;
1067     dc->vmsd    = &vmstate_xive_source;
1068 }
1069 
1070 static const TypeInfo xive_source_info = {
1071     .name          = TYPE_XIVE_SOURCE,
1072     .parent        = TYPE_DEVICE,
1073     .instance_size = sizeof(XiveSource),
1074     .class_init    = xive_source_class_init,
1075 };
1076 
1077 /*
1078  * XiveEND helpers
1079  */
1080 
1081 void xive_end_queue_pic_print_info(XiveEND *end, uint32_t width, Monitor *mon)
1082 {
1083     uint64_t qaddr_base = xive_end_qaddr(end);
1084     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1085     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1086     uint32_t qentries = 1 << (qsize + 10);
1087     int i;
1088 
1089     /*
1090      * print out the [ (qindex - (width - 1)) .. (qindex + 1)] window
1091      */
1092     monitor_printf(mon, " [ ");
1093     qindex = (qindex - (width - 1)) & (qentries - 1);
1094     for (i = 0; i < width; i++) {
1095         uint64_t qaddr = qaddr_base + (qindex << 2);
1096         uint32_t qdata = -1;
1097 
1098         if (dma_memory_read(&address_space_memory, qaddr, &qdata,
1099                             sizeof(qdata))) {
1100             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to read EQ @0x%"
1101                           HWADDR_PRIx "\n", qaddr);
1102             return;
1103         }
1104         monitor_printf(mon, "%s%08x ", i == width - 1 ? "^" : "",
1105                        be32_to_cpu(qdata));
1106         qindex = (qindex + 1) & (qentries - 1);
1107     }
1108 }
1109 
1110 void xive_end_pic_print_info(XiveEND *end, uint32_t end_idx, Monitor *mon)
1111 {
1112     uint64_t qaddr_base = xive_end_qaddr(end);
1113     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1114     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1115     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1116     uint32_t qentries = 1 << (qsize + 10);
1117 
1118     uint32_t nvt = xive_get_field32(END_W6_NVT_INDEX, end->w6);
1119     uint8_t priority = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
1120 
1121     if (!xive_end_is_valid(end)) {
1122         return;
1123     }
1124 
1125     monitor_printf(mon, "  %08x %c%c%c%c%c prio:%d nvt:%04x eq:@%08"PRIx64
1126                    "% 6d/%5d ^%d", end_idx,
1127                    xive_end_is_valid(end)    ? 'v' : '-',
1128                    xive_end_is_enqueue(end)  ? 'q' : '-',
1129                    xive_end_is_notify(end)   ? 'n' : '-',
1130                    xive_end_is_backlog(end)  ? 'b' : '-',
1131                    xive_end_is_escalate(end) ? 'e' : '-',
1132                    priority, nvt, qaddr_base, qindex, qentries, qgen);
1133 
1134     xive_end_queue_pic_print_info(end, 6, mon);
1135     monitor_printf(mon, "]\n");
1136 }
1137 
1138 static void xive_end_enqueue(XiveEND *end, uint32_t data)
1139 {
1140     uint64_t qaddr_base = xive_end_qaddr(end);
1141     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1142     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1143     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1144 
1145     uint64_t qaddr = qaddr_base + (qindex << 2);
1146     uint32_t qdata = cpu_to_be32((qgen << 31) | (data & 0x7fffffff));
1147     uint32_t qentries = 1 << (qsize + 10);
1148 
1149     if (dma_memory_write(&address_space_memory, qaddr, &qdata, sizeof(qdata))) {
1150         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to write END data @0x%"
1151                       HWADDR_PRIx "\n", qaddr);
1152         return;
1153     }
1154 
1155     qindex = (qindex + 1) & (qentries - 1);
1156     if (qindex == 0) {
1157         qgen ^= 1;
1158         end->w1 = xive_set_field32(END_W1_GENERATION, end->w1, qgen);
1159     }
1160     end->w1 = xive_set_field32(END_W1_PAGE_OFF, end->w1, qindex);
1161 }
1162 
1163 /*
1164  * XIVE Router (aka. Virtualization Controller or IVRE)
1165  */
1166 
1167 int xive_router_get_eas(XiveRouter *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1168                         XiveEAS *eas)
1169 {
1170     XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1171 
1172     return xrc->get_eas(xrtr, eas_blk, eas_idx, eas);
1173 }
1174 
1175 int xive_router_get_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1176                         XiveEND *end)
1177 {
1178    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1179 
1180    return xrc->get_end(xrtr, end_blk, end_idx, end);
1181 }
1182 
1183 int xive_router_write_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1184                           XiveEND *end, uint8_t word_number)
1185 {
1186    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1187 
1188    return xrc->write_end(xrtr, end_blk, end_idx, end, word_number);
1189 }
1190 
1191 int xive_router_get_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1192                         XiveNVT *nvt)
1193 {
1194    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1195 
1196    return xrc->get_nvt(xrtr, nvt_blk, nvt_idx, nvt);
1197 }
1198 
1199 int xive_router_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1200                         XiveNVT *nvt, uint8_t word_number)
1201 {
1202    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1203 
1204    return xrc->write_nvt(xrtr, nvt_blk, nvt_idx, nvt, word_number);
1205 }
1206 
1207 XiveTCTX *xive_router_get_tctx(XiveRouter *xrtr, CPUState *cs)
1208 {
1209     XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1210 
1211     return xrc->get_tctx(xrtr, cs);
1212 }
1213 
1214 /*
1215  * By default on P9, the HW CAM line (23bits) is hardwired to :
1216  *
1217  *   0x000||0b1||4Bit chip number||7Bit Thread number.
1218  *
1219  * When the block grouping is enabled, the CAM line is changed to :
1220  *
1221  *   4Bit chip number||0x001||7Bit Thread number.
1222  */
1223 static uint32_t hw_cam_line(uint8_t chip_id, uint8_t tid)
1224 {
1225     return 1 << 11 | (chip_id & 0xf) << 7 | (tid & 0x7f);
1226 }
1227 
1228 static bool xive_presenter_tctx_match_hw(XiveTCTX *tctx,
1229                                          uint8_t nvt_blk, uint32_t nvt_idx)
1230 {
1231     CPUPPCState *env = &POWERPC_CPU(tctx->cs)->env;
1232     uint32_t pir = env->spr_cb[SPR_PIR].default_value;
1233 
1234     return hw_cam_line((pir >> 8) & 0xf, pir & 0x7f) ==
1235         hw_cam_line(nvt_blk, nvt_idx);
1236 }
1237 
1238 /*
1239  * The thread context register words are in big-endian format.
1240  */
1241 static int xive_presenter_tctx_match(XiveTCTX *tctx, uint8_t format,
1242                                      uint8_t nvt_blk, uint32_t nvt_idx,
1243                                      bool cam_ignore, uint32_t logic_serv)
1244 {
1245     uint32_t cam = xive_nvt_cam_line(nvt_blk, nvt_idx);
1246     uint32_t qw3w2 = xive_tctx_word2(&tctx->regs[TM_QW3_HV_PHYS]);
1247     uint32_t qw2w2 = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
1248     uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
1249     uint32_t qw0w2 = xive_tctx_word2(&tctx->regs[TM_QW0_USER]);
1250 
1251     /*
1252      * TODO (PowerNV): ignore mode. The low order bits of the NVT
1253      * identifier are ignored in the "CAM" match.
1254      */
1255 
1256     if (format == 0) {
1257         if (cam_ignore == true) {
1258             /*
1259              * F=0 & i=1: Logical server notification (bits ignored at
1260              * the end of the NVT identifier)
1261              */
1262             qemu_log_mask(LOG_UNIMP, "XIVE: no support for LS NVT %x/%x\n",
1263                           nvt_blk, nvt_idx);
1264              return -1;
1265         }
1266 
1267         /* F=0 & i=0: Specific NVT notification */
1268 
1269         /* PHYS ring */
1270         if ((be32_to_cpu(qw3w2) & TM_QW3W2_VT) &&
1271             xive_presenter_tctx_match_hw(tctx, nvt_blk, nvt_idx)) {
1272             return TM_QW3_HV_PHYS;
1273         }
1274 
1275         /* HV POOL ring */
1276         if ((be32_to_cpu(qw2w2) & TM_QW2W2_VP) &&
1277             cam == xive_get_field32(TM_QW2W2_POOL_CAM, qw2w2)) {
1278             return TM_QW2_HV_POOL;
1279         }
1280 
1281         /* OS ring */
1282         if ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1283             cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) {
1284             return TM_QW1_OS;
1285         }
1286     } else {
1287         /* F=1 : User level Event-Based Branch (EBB) notification */
1288 
1289         /* USER ring */
1290         if  ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1291              (cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) &&
1292              (be32_to_cpu(qw0w2) & TM_QW0W2_VU) &&
1293              (logic_serv == xive_get_field32(TM_QW0W2_LOGIC_SERV, qw0w2))) {
1294             return TM_QW0_USER;
1295         }
1296     }
1297     return -1;
1298 }
1299 
1300 typedef struct XiveTCTXMatch {
1301     XiveTCTX *tctx;
1302     uint8_t ring;
1303 } XiveTCTXMatch;
1304 
1305 static bool xive_presenter_match(XiveRouter *xrtr, uint8_t format,
1306                                  uint8_t nvt_blk, uint32_t nvt_idx,
1307                                  bool cam_ignore, uint8_t priority,
1308                                  uint32_t logic_serv, XiveTCTXMatch *match)
1309 {
1310     CPUState *cs;
1311 
1312     /*
1313      * TODO (PowerNV): handle chip_id overwrite of block field for
1314      * hardwired CAM compares
1315      */
1316 
1317     CPU_FOREACH(cs) {
1318         XiveTCTX *tctx = xive_router_get_tctx(xrtr, cs);
1319         int ring;
1320 
1321         /*
1322          * HW checks that the CPU is enabled in the Physical Thread
1323          * Enable Register (PTER).
1324          */
1325 
1326         /*
1327          * Check the thread context CAM lines and record matches. We
1328          * will handle CPU exception delivery later
1329          */
1330         ring = xive_presenter_tctx_match(tctx, format, nvt_blk, nvt_idx,
1331                                          cam_ignore, logic_serv);
1332         /*
1333          * Save the context and follow on to catch duplicates, that we
1334          * don't support yet.
1335          */
1336         if (ring != -1) {
1337             if (match->tctx) {
1338                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: already found a thread "
1339                               "context NVT %x/%x\n", nvt_blk, nvt_idx);
1340                 return false;
1341             }
1342 
1343             match->ring = ring;
1344             match->tctx = tctx;
1345         }
1346     }
1347 
1348     if (!match->tctx) {
1349         qemu_log_mask(LOG_UNIMP, "XIVE: NVT %x/%x is not dispatched\n",
1350                       nvt_blk, nvt_idx);
1351         return false;
1352     }
1353 
1354     return true;
1355 }
1356 
1357 /*
1358  * This is our simple Xive Presenter Engine model. It is merged in the
1359  * Router as it does not require an extra object.
1360  *
1361  * It receives notification requests sent by the IVRE to find one
1362  * matching NVT (or more) dispatched on the processor threads. In case
1363  * of a single NVT notification, the process is abreviated and the
1364  * thread is signaled if a match is found. In case of a logical server
1365  * notification (bits ignored at the end of the NVT identifier), the
1366  * IVPE and IVRE select a winning thread using different filters. This
1367  * involves 2 or 3 exchanges on the PowerBus that the model does not
1368  * support.
1369  *
1370  * The parameters represent what is sent on the PowerBus
1371  */
1372 static void xive_presenter_notify(XiveRouter *xrtr, uint8_t format,
1373                                   uint8_t nvt_blk, uint32_t nvt_idx,
1374                                   bool cam_ignore, uint8_t priority,
1375                                   uint32_t logic_serv)
1376 {
1377     XiveNVT nvt;
1378     XiveTCTXMatch match = { .tctx = NULL, .ring = 0 };
1379     bool found;
1380 
1381     /* NVT cache lookup */
1382     if (xive_router_get_nvt(xrtr, nvt_blk, nvt_idx, &nvt)) {
1383         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no NVT %x/%x\n",
1384                       nvt_blk, nvt_idx);
1385         return;
1386     }
1387 
1388     if (!xive_nvt_is_valid(&nvt)) {
1389         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is invalid\n",
1390                       nvt_blk, nvt_idx);
1391         return;
1392     }
1393 
1394     found = xive_presenter_match(xrtr, format, nvt_blk, nvt_idx, cam_ignore,
1395                                  priority, logic_serv, &match);
1396     if (found) {
1397         ipb_update(&match.tctx->regs[match.ring], priority);
1398         xive_tctx_notify(match.tctx, match.ring);
1399         return;
1400     }
1401 
1402     /* Record the IPB in the associated NVT structure */
1403     ipb_update((uint8_t *) &nvt.w4, priority);
1404     xive_router_write_nvt(xrtr, nvt_blk, nvt_idx, &nvt, 4);
1405 
1406     /*
1407      * If no matching NVT is dispatched on a HW thread :
1408      * - update the NVT structure if backlog is activated
1409      * - escalate (ESe PQ bits and EAS in w4-5) if escalation is
1410      *   activated
1411      */
1412 }
1413 
1414 /*
1415  * An END trigger can come from an event trigger (IPI or HW) or from
1416  * another chip. We don't model the PowerBus but the END trigger
1417  * message has the same parameters than in the function below.
1418  */
1419 static void xive_router_end_notify(XiveRouter *xrtr, uint8_t end_blk,
1420                                    uint32_t end_idx, uint32_t end_data)
1421 {
1422     XiveEND end;
1423     uint8_t priority;
1424     uint8_t format;
1425 
1426     /* END cache lookup */
1427     if (xive_router_get_end(xrtr, end_blk, end_idx, &end)) {
1428         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1429                       end_idx);
1430         return;
1431     }
1432 
1433     if (!xive_end_is_valid(&end)) {
1434         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1435                       end_blk, end_idx);
1436         return;
1437     }
1438 
1439     if (xive_end_is_enqueue(&end)) {
1440         xive_end_enqueue(&end, end_data);
1441         /* Enqueuing event data modifies the EQ toggle and index */
1442         xive_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1443     }
1444 
1445     /*
1446      * The W7 format depends on the F bit in W6. It defines the type
1447      * of the notification :
1448      *
1449      *   F=0 : single or multiple NVT notification
1450      *   F=1 : User level Event-Based Branch (EBB) notification, no
1451      *         priority
1452      */
1453     format = xive_get_field32(END_W6_FORMAT_BIT, end.w6);
1454     priority = xive_get_field32(END_W7_F0_PRIORITY, end.w7);
1455 
1456     /* The END is masked */
1457     if (format == 0 && priority == 0xff) {
1458         return;
1459     }
1460 
1461     /*
1462      * Check the END ESn (Event State Buffer for notification) for
1463      * even futher coalescing in the Router
1464      */
1465     if (!xive_end_is_notify(&end)) {
1466         uint8_t pq = xive_get_field32(END_W1_ESn, end.w1);
1467         bool notify = xive_esb_trigger(&pq);
1468 
1469         if (pq != xive_get_field32(END_W1_ESn, end.w1)) {
1470             end.w1 = xive_set_field32(END_W1_ESn, end.w1, pq);
1471             xive_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1472         }
1473 
1474         /* ESn[Q]=1 : end of notification */
1475         if (!notify) {
1476             return;
1477         }
1478     }
1479 
1480     /*
1481      * Follows IVPE notification
1482      */
1483     xive_presenter_notify(xrtr, format,
1484                           xive_get_field32(END_W6_NVT_BLOCK, end.w6),
1485                           xive_get_field32(END_W6_NVT_INDEX, end.w6),
1486                           xive_get_field32(END_W7_F0_IGNORE, end.w7),
1487                           priority,
1488                           xive_get_field32(END_W7_F1_LOG_SERVER_ID, end.w7));
1489 
1490     /* TODO: Auto EOI. */
1491 }
1492 
1493 void xive_router_notify(XiveNotifier *xn, uint32_t lisn)
1494 {
1495     XiveRouter *xrtr = XIVE_ROUTER(xn);
1496     uint8_t eas_blk = XIVE_SRCNO_BLOCK(lisn);
1497     uint32_t eas_idx = XIVE_SRCNO_INDEX(lisn);
1498     XiveEAS eas;
1499 
1500     /* EAS cache lookup */
1501     if (xive_router_get_eas(xrtr, eas_blk, eas_idx, &eas)) {
1502         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN %x\n", lisn);
1503         return;
1504     }
1505 
1506     /*
1507      * The IVRE checks the State Bit Cache at this point. We skip the
1508      * SBC lookup because the state bits of the sources are modeled
1509      * internally in QEMU.
1510      */
1511 
1512     if (!xive_eas_is_valid(&eas)) {
1513         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid LISN %x\n", lisn);
1514         return;
1515     }
1516 
1517     if (xive_eas_is_masked(&eas)) {
1518         /* Notification completed */
1519         return;
1520     }
1521 
1522     /*
1523      * The event trigger becomes an END trigger
1524      */
1525     xive_router_end_notify(xrtr,
1526                            xive_get_field64(EAS_END_BLOCK, eas.w),
1527                            xive_get_field64(EAS_END_INDEX, eas.w),
1528                            xive_get_field64(EAS_END_DATA,  eas.w));
1529 }
1530 
1531 static void xive_router_class_init(ObjectClass *klass, void *data)
1532 {
1533     DeviceClass *dc = DEVICE_CLASS(klass);
1534     XiveNotifierClass *xnc = XIVE_NOTIFIER_CLASS(klass);
1535 
1536     dc->desc    = "XIVE Router Engine";
1537     xnc->notify = xive_router_notify;
1538 }
1539 
1540 static const TypeInfo xive_router_info = {
1541     .name          = TYPE_XIVE_ROUTER,
1542     .parent        = TYPE_SYS_BUS_DEVICE,
1543     .abstract      = true,
1544     .class_size    = sizeof(XiveRouterClass),
1545     .class_init    = xive_router_class_init,
1546     .interfaces    = (InterfaceInfo[]) {
1547         { TYPE_XIVE_NOTIFIER },
1548         { }
1549     }
1550 };
1551 
1552 void xive_eas_pic_print_info(XiveEAS *eas, uint32_t lisn, Monitor *mon)
1553 {
1554     if (!xive_eas_is_valid(eas)) {
1555         return;
1556     }
1557 
1558     monitor_printf(mon, "  %08x %s end:%02x/%04x data:%08x\n",
1559                    lisn, xive_eas_is_masked(eas) ? "M" : " ",
1560                    (uint8_t)  xive_get_field64(EAS_END_BLOCK, eas->w),
1561                    (uint32_t) xive_get_field64(EAS_END_INDEX, eas->w),
1562                    (uint32_t) xive_get_field64(EAS_END_DATA, eas->w));
1563 }
1564 
1565 /*
1566  * END ESB MMIO loads
1567  */
1568 static uint64_t xive_end_source_read(void *opaque, hwaddr addr, unsigned size)
1569 {
1570     XiveENDSource *xsrc = XIVE_END_SOURCE(opaque);
1571     uint32_t offset = addr & 0xFFF;
1572     uint8_t end_blk;
1573     uint32_t end_idx;
1574     XiveEND end;
1575     uint32_t end_esmask;
1576     uint8_t pq;
1577     uint64_t ret = -1;
1578 
1579     end_blk = xsrc->block_id;
1580     end_idx = addr >> (xsrc->esb_shift + 1);
1581 
1582     if (xive_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
1583         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1584                       end_idx);
1585         return -1;
1586     }
1587 
1588     if (!xive_end_is_valid(&end)) {
1589         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1590                       end_blk, end_idx);
1591         return -1;
1592     }
1593 
1594     end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END_W1_ESn : END_W1_ESe;
1595     pq = xive_get_field32(end_esmask, end.w1);
1596 
1597     switch (offset) {
1598     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
1599         ret = xive_esb_eoi(&pq);
1600 
1601         /* Forward the source event notification for routing ?? */
1602         break;
1603 
1604     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
1605         ret = pq;
1606         break;
1607 
1608     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1609     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1610     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1611     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1612         ret = xive_esb_set(&pq, (offset >> 8) & 0x3);
1613         break;
1614     default:
1615         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB load addr %d\n",
1616                       offset);
1617         return -1;
1618     }
1619 
1620     if (pq != xive_get_field32(end_esmask, end.w1)) {
1621         end.w1 = xive_set_field32(end_esmask, end.w1, pq);
1622         xive_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
1623     }
1624 
1625     return ret;
1626 }
1627 
1628 /*
1629  * END ESB MMIO stores are invalid
1630  */
1631 static void xive_end_source_write(void *opaque, hwaddr addr,
1632                                   uint64_t value, unsigned size)
1633 {
1634     qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr 0x%"
1635                   HWADDR_PRIx"\n", addr);
1636 }
1637 
1638 static const MemoryRegionOps xive_end_source_ops = {
1639     .read = xive_end_source_read,
1640     .write = xive_end_source_write,
1641     .endianness = DEVICE_BIG_ENDIAN,
1642     .valid = {
1643         .min_access_size = 8,
1644         .max_access_size = 8,
1645     },
1646     .impl = {
1647         .min_access_size = 8,
1648         .max_access_size = 8,
1649     },
1650 };
1651 
1652 static void xive_end_source_realize(DeviceState *dev, Error **errp)
1653 {
1654     XiveENDSource *xsrc = XIVE_END_SOURCE(dev);
1655     Object *obj;
1656     Error *local_err = NULL;
1657 
1658     obj = object_property_get_link(OBJECT(dev), "xive", &local_err);
1659     if (!obj) {
1660         error_propagate(errp, local_err);
1661         error_prepend(errp, "required link 'xive' not found: ");
1662         return;
1663     }
1664 
1665     xsrc->xrtr = XIVE_ROUTER(obj);
1666 
1667     if (!xsrc->nr_ends) {
1668         error_setg(errp, "Number of interrupt needs to be greater than 0");
1669         return;
1670     }
1671 
1672     if (xsrc->esb_shift != XIVE_ESB_4K &&
1673         xsrc->esb_shift != XIVE_ESB_64K) {
1674         error_setg(errp, "Invalid ESB shift setting");
1675         return;
1676     }
1677 
1678     /*
1679      * Each END is assigned an even/odd pair of MMIO pages, the even page
1680      * manages the ESn field while the odd page manages the ESe field.
1681      */
1682     memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
1683                           &xive_end_source_ops, xsrc, "xive.end",
1684                           (1ull << (xsrc->esb_shift + 1)) * xsrc->nr_ends);
1685 }
1686 
1687 static Property xive_end_source_properties[] = {
1688     DEFINE_PROP_UINT8("block-id", XiveENDSource, block_id, 0),
1689     DEFINE_PROP_UINT32("nr-ends", XiveENDSource, nr_ends, 0),
1690     DEFINE_PROP_UINT32("shift", XiveENDSource, esb_shift, XIVE_ESB_64K),
1691     DEFINE_PROP_END_OF_LIST(),
1692 };
1693 
1694 static void xive_end_source_class_init(ObjectClass *klass, void *data)
1695 {
1696     DeviceClass *dc = DEVICE_CLASS(klass);
1697 
1698     dc->desc    = "XIVE END Source";
1699     dc->props   = xive_end_source_properties;
1700     dc->realize = xive_end_source_realize;
1701 }
1702 
1703 static const TypeInfo xive_end_source_info = {
1704     .name          = TYPE_XIVE_END_SOURCE,
1705     .parent        = TYPE_DEVICE,
1706     .instance_size = sizeof(XiveENDSource),
1707     .class_init    = xive_end_source_class_init,
1708 };
1709 
1710 /*
1711  * XIVE Notifier
1712  */
1713 static const TypeInfo xive_notifier_info = {
1714     .name = TYPE_XIVE_NOTIFIER,
1715     .parent = TYPE_INTERFACE,
1716     .class_size = sizeof(XiveNotifierClass),
1717 };
1718 
1719 static void xive_register_types(void)
1720 {
1721     type_register_static(&xive_source_info);
1722     type_register_static(&xive_notifier_info);
1723     type_register_static(&xive_router_info);
1724     type_register_static(&xive_end_source_info);
1725     type_register_static(&xive_tctx_info);
1726 }
1727 
1728 type_init(xive_register_types)
1729