xref: /openbmc/qemu/hw/intc/xive.c (revision 13df93244efbd4bb8b4cf4e26104a26033178674)
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     monitor_printf(mon, "CPU[%04x]:   QW   NSR CPPR IPB LSMFB ACK# INC AGE PIPR"
497                    "  W2\n", cpu_index);
498 
499     for (i = 0; i < XIVE_TM_RING_COUNT; i++) {
500         char *s = xive_tctx_ring_print(&tctx->regs[i * XIVE_TM_RING_SIZE]);
501         monitor_printf(mon, "CPU[%04x]: %4s    %s\n", cpu_index,
502                        xive_tctx_ring_names[i], s);
503         g_free(s);
504     }
505 }
506 
507 static void xive_tctx_reset(void *dev)
508 {
509     XiveTCTX *tctx = XIVE_TCTX(dev);
510 
511     memset(tctx->regs, 0, sizeof(tctx->regs));
512 
513     /* Set some defaults */
514     tctx->regs[TM_QW1_OS + TM_LSMFB] = 0xFF;
515     tctx->regs[TM_QW1_OS + TM_ACK_CNT] = 0xFF;
516     tctx->regs[TM_QW1_OS + TM_AGE] = 0xFF;
517 
518     /*
519      * Initialize PIPR to 0xFF to avoid phantom interrupts when the
520      * CPPR is first set.
521      */
522     tctx->regs[TM_QW1_OS + TM_PIPR] =
523         ipb_to_pipr(tctx->regs[TM_QW1_OS + TM_IPB]);
524     tctx->regs[TM_QW3_HV_PHYS + TM_PIPR] =
525         ipb_to_pipr(tctx->regs[TM_QW3_HV_PHYS + TM_IPB]);
526 }
527 
528 static void xive_tctx_realize(DeviceState *dev, Error **errp)
529 {
530     XiveTCTX *tctx = XIVE_TCTX(dev);
531     PowerPCCPU *cpu;
532     CPUPPCState *env;
533     Object *obj;
534     Error *local_err = NULL;
535 
536     obj = object_property_get_link(OBJECT(dev), "cpu", &local_err);
537     if (!obj) {
538         error_propagate(errp, local_err);
539         error_prepend(errp, "required link 'cpu' not found: ");
540         return;
541     }
542 
543     cpu = POWERPC_CPU(obj);
544     tctx->cs = CPU(obj);
545 
546     env = &cpu->env;
547     switch (PPC_INPUT(env)) {
548     case PPC_FLAGS_INPUT_POWER9:
549         tctx->output = env->irq_inputs[POWER9_INPUT_INT];
550         break;
551 
552     default:
553         error_setg(errp, "XIVE interrupt controller does not support "
554                    "this CPU bus model");
555         return;
556     }
557 
558     qemu_register_reset(xive_tctx_reset, dev);
559 }
560 
561 static void xive_tctx_unrealize(DeviceState *dev, Error **errp)
562 {
563     qemu_unregister_reset(xive_tctx_reset, dev);
564 }
565 
566 static const VMStateDescription vmstate_xive_tctx = {
567     .name = TYPE_XIVE_TCTX,
568     .version_id = 1,
569     .minimum_version_id = 1,
570     .fields = (VMStateField[]) {
571         VMSTATE_BUFFER(regs, XiveTCTX),
572         VMSTATE_END_OF_LIST()
573     },
574 };
575 
576 static void xive_tctx_class_init(ObjectClass *klass, void *data)
577 {
578     DeviceClass *dc = DEVICE_CLASS(klass);
579 
580     dc->desc = "XIVE Interrupt Thread Context";
581     dc->realize = xive_tctx_realize;
582     dc->unrealize = xive_tctx_unrealize;
583     dc->vmsd = &vmstate_xive_tctx;
584 }
585 
586 static const TypeInfo xive_tctx_info = {
587     .name          = TYPE_XIVE_TCTX,
588     .parent        = TYPE_DEVICE,
589     .instance_size = sizeof(XiveTCTX),
590     .class_init    = xive_tctx_class_init,
591 };
592 
593 Object *xive_tctx_create(Object *cpu, XiveRouter *xrtr, Error **errp)
594 {
595     Error *local_err = NULL;
596     Object *obj;
597 
598     obj = object_new(TYPE_XIVE_TCTX);
599     object_property_add_child(cpu, TYPE_XIVE_TCTX, obj, &error_abort);
600     object_unref(obj);
601     object_property_add_const_link(obj, "cpu", cpu, &error_abort);
602     object_property_set_bool(obj, true, "realized", &local_err);
603     if (local_err) {
604         goto error;
605     }
606 
607     return obj;
608 
609 error:
610     object_unparent(obj);
611     error_propagate(errp, local_err);
612     return NULL;
613 }
614 
615 /*
616  * XIVE ESB helpers
617  */
618 
619 static uint8_t xive_esb_set(uint8_t *pq, uint8_t value)
620 {
621     uint8_t old_pq = *pq & 0x3;
622 
623     *pq &= ~0x3;
624     *pq |= value & 0x3;
625 
626     return old_pq;
627 }
628 
629 static bool xive_esb_trigger(uint8_t *pq)
630 {
631     uint8_t old_pq = *pq & 0x3;
632 
633     switch (old_pq) {
634     case XIVE_ESB_RESET:
635         xive_esb_set(pq, XIVE_ESB_PENDING);
636         return true;
637     case XIVE_ESB_PENDING:
638     case XIVE_ESB_QUEUED:
639         xive_esb_set(pq, XIVE_ESB_QUEUED);
640         return false;
641     case XIVE_ESB_OFF:
642         xive_esb_set(pq, XIVE_ESB_OFF);
643         return false;
644     default:
645          g_assert_not_reached();
646     }
647 }
648 
649 static bool xive_esb_eoi(uint8_t *pq)
650 {
651     uint8_t old_pq = *pq & 0x3;
652 
653     switch (old_pq) {
654     case XIVE_ESB_RESET:
655     case XIVE_ESB_PENDING:
656         xive_esb_set(pq, XIVE_ESB_RESET);
657         return false;
658     case XIVE_ESB_QUEUED:
659         xive_esb_set(pq, XIVE_ESB_PENDING);
660         return true;
661     case XIVE_ESB_OFF:
662         xive_esb_set(pq, XIVE_ESB_OFF);
663         return false;
664     default:
665          g_assert_not_reached();
666     }
667 }
668 
669 /*
670  * XIVE Interrupt Source (or IVSE)
671  */
672 
673 uint8_t xive_source_esb_get(XiveSource *xsrc, uint32_t srcno)
674 {
675     assert(srcno < xsrc->nr_irqs);
676 
677     return xsrc->status[srcno] & 0x3;
678 }
679 
680 uint8_t xive_source_esb_set(XiveSource *xsrc, uint32_t srcno, uint8_t pq)
681 {
682     assert(srcno < xsrc->nr_irqs);
683 
684     return xive_esb_set(&xsrc->status[srcno], pq);
685 }
686 
687 /*
688  * Returns whether the event notification should be forwarded.
689  */
690 static bool xive_source_lsi_trigger(XiveSource *xsrc, uint32_t srcno)
691 {
692     uint8_t old_pq = xive_source_esb_get(xsrc, srcno);
693 
694     xsrc->status[srcno] |= XIVE_STATUS_ASSERTED;
695 
696     switch (old_pq) {
697     case XIVE_ESB_RESET:
698         xive_source_esb_set(xsrc, srcno, XIVE_ESB_PENDING);
699         return true;
700     default:
701         return false;
702     }
703 }
704 
705 /*
706  * Returns whether the event notification should be forwarded.
707  */
708 static bool xive_source_esb_trigger(XiveSource *xsrc, uint32_t srcno)
709 {
710     bool ret;
711 
712     assert(srcno < xsrc->nr_irqs);
713 
714     ret = xive_esb_trigger(&xsrc->status[srcno]);
715 
716     if (xive_source_irq_is_lsi(xsrc, srcno) &&
717         xive_source_esb_get(xsrc, srcno) == XIVE_ESB_QUEUED) {
718         qemu_log_mask(LOG_GUEST_ERROR,
719                       "XIVE: queued an event on LSI IRQ %d\n", srcno);
720     }
721 
722     return ret;
723 }
724 
725 /*
726  * Returns whether the event notification should be forwarded.
727  */
728 static bool xive_source_esb_eoi(XiveSource *xsrc, uint32_t srcno)
729 {
730     bool ret;
731 
732     assert(srcno < xsrc->nr_irqs);
733 
734     ret = xive_esb_eoi(&xsrc->status[srcno]);
735 
736     /*
737      * LSI sources do not set the Q bit but they can still be
738      * asserted, in which case we should forward a new event
739      * notification
740      */
741     if (xive_source_irq_is_lsi(xsrc, srcno) &&
742         xsrc->status[srcno] & XIVE_STATUS_ASSERTED) {
743         ret = xive_source_lsi_trigger(xsrc, srcno);
744     }
745 
746     return ret;
747 }
748 
749 /*
750  * Forward the source event notification to the Router
751  */
752 static void xive_source_notify(XiveSource *xsrc, int srcno)
753 {
754     XiveNotifierClass *xnc = XIVE_NOTIFIER_GET_CLASS(xsrc->xive);
755 
756     if (xnc->notify) {
757         xnc->notify(xsrc->xive, srcno);
758     }
759 }
760 
761 /*
762  * In a two pages ESB MMIO setting, even page is the trigger page, odd
763  * page is for management
764  */
765 static inline bool addr_is_even(hwaddr addr, uint32_t shift)
766 {
767     return !((addr >> shift) & 1);
768 }
769 
770 static inline bool xive_source_is_trigger_page(XiveSource *xsrc, hwaddr addr)
771 {
772     return xive_source_esb_has_2page(xsrc) &&
773         addr_is_even(addr, xsrc->esb_shift - 1);
774 }
775 
776 /*
777  * ESB MMIO loads
778  *                      Trigger page    Management/EOI page
779  *
780  * ESB MMIO setting     2 pages         1 or 2 pages
781  *
782  * 0x000 .. 0x3FF       -1              EOI and return 0|1
783  * 0x400 .. 0x7FF       -1              EOI and return 0|1
784  * 0x800 .. 0xBFF       -1              return PQ
785  * 0xC00 .. 0xCFF       -1              return PQ and atomically PQ=00
786  * 0xD00 .. 0xDFF       -1              return PQ and atomically PQ=01
787  * 0xE00 .. 0xDFF       -1              return PQ and atomically PQ=10
788  * 0xF00 .. 0xDFF       -1              return PQ and atomically PQ=11
789  */
790 static uint64_t xive_source_esb_read(void *opaque, hwaddr addr, unsigned size)
791 {
792     XiveSource *xsrc = XIVE_SOURCE(opaque);
793     uint32_t offset = addr & 0xFFF;
794     uint32_t srcno = addr >> xsrc->esb_shift;
795     uint64_t ret = -1;
796 
797     /* In a two pages ESB MMIO setting, trigger page should not be read */
798     if (xive_source_is_trigger_page(xsrc, addr)) {
799         qemu_log_mask(LOG_GUEST_ERROR,
800                       "XIVE: invalid load on IRQ %d trigger page at "
801                       "0x%"HWADDR_PRIx"\n", srcno, addr);
802         return -1;
803     }
804 
805     switch (offset) {
806     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
807         ret = xive_source_esb_eoi(xsrc, srcno);
808 
809         /* Forward the source event notification for routing */
810         if (ret) {
811             xive_source_notify(xsrc, srcno);
812         }
813         break;
814 
815     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
816         ret = xive_source_esb_get(xsrc, srcno);
817         break;
818 
819     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
820     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
821     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
822     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
823         ret = xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
824         break;
825     default:
826         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB load addr %x\n",
827                       offset);
828     }
829 
830     return ret;
831 }
832 
833 /*
834  * ESB MMIO stores
835  *                      Trigger page    Management/EOI page
836  *
837  * ESB MMIO setting     2 pages         1 or 2 pages
838  *
839  * 0x000 .. 0x3FF       Trigger         Trigger
840  * 0x400 .. 0x7FF       Trigger         EOI
841  * 0x800 .. 0xBFF       Trigger         undefined
842  * 0xC00 .. 0xCFF       Trigger         PQ=00
843  * 0xD00 .. 0xDFF       Trigger         PQ=01
844  * 0xE00 .. 0xDFF       Trigger         PQ=10
845  * 0xF00 .. 0xDFF       Trigger         PQ=11
846  */
847 static void xive_source_esb_write(void *opaque, hwaddr addr,
848                                   uint64_t value, unsigned size)
849 {
850     XiveSource *xsrc = XIVE_SOURCE(opaque);
851     uint32_t offset = addr & 0xFFF;
852     uint32_t srcno = addr >> xsrc->esb_shift;
853     bool notify = false;
854 
855     /* In a two pages ESB MMIO setting, trigger page only triggers */
856     if (xive_source_is_trigger_page(xsrc, addr)) {
857         notify = xive_source_esb_trigger(xsrc, srcno);
858         goto out;
859     }
860 
861     switch (offset) {
862     case 0 ... 0x3FF:
863         notify = xive_source_esb_trigger(xsrc, srcno);
864         break;
865 
866     case XIVE_ESB_STORE_EOI ... XIVE_ESB_STORE_EOI + 0x3FF:
867         if (!(xsrc->esb_flags & XIVE_SRC_STORE_EOI)) {
868             qemu_log_mask(LOG_GUEST_ERROR,
869                           "XIVE: invalid Store EOI for IRQ %d\n", srcno);
870             return;
871         }
872 
873         notify = xive_source_esb_eoi(xsrc, srcno);
874         break;
875 
876     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
877     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
878     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
879     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
880         xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
881         break;
882 
883     default:
884         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr %x\n",
885                       offset);
886         return;
887     }
888 
889 out:
890     /* Forward the source event notification for routing */
891     if (notify) {
892         xive_source_notify(xsrc, srcno);
893     }
894 }
895 
896 static const MemoryRegionOps xive_source_esb_ops = {
897     .read = xive_source_esb_read,
898     .write = xive_source_esb_write,
899     .endianness = DEVICE_BIG_ENDIAN,
900     .valid = {
901         .min_access_size = 8,
902         .max_access_size = 8,
903     },
904     .impl = {
905         .min_access_size = 8,
906         .max_access_size = 8,
907     },
908 };
909 
910 void xive_source_set_irq(void *opaque, int srcno, int val)
911 {
912     XiveSource *xsrc = XIVE_SOURCE(opaque);
913     bool notify = false;
914 
915     if (xive_source_irq_is_lsi(xsrc, srcno)) {
916         if (val) {
917             notify = xive_source_lsi_trigger(xsrc, srcno);
918         } else {
919             xsrc->status[srcno] &= ~XIVE_STATUS_ASSERTED;
920         }
921     } else {
922         if (val) {
923             notify = xive_source_esb_trigger(xsrc, srcno);
924         }
925     }
926 
927     /* Forward the source event notification for routing */
928     if (notify) {
929         xive_source_notify(xsrc, srcno);
930     }
931 }
932 
933 void xive_source_pic_print_info(XiveSource *xsrc, uint32_t offset, Monitor *mon)
934 {
935     int i;
936 
937     for (i = 0; i < xsrc->nr_irqs; i++) {
938         uint8_t pq = xive_source_esb_get(xsrc, i);
939 
940         if (pq == XIVE_ESB_OFF) {
941             continue;
942         }
943 
944         monitor_printf(mon, "  %08x %s %c%c%c\n", i + offset,
945                        xive_source_irq_is_lsi(xsrc, i) ? "LSI" : "MSI",
946                        pq & XIVE_ESB_VAL_P ? 'P' : '-',
947                        pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
948                        xsrc->status[i] & XIVE_STATUS_ASSERTED ? 'A' : ' ');
949     }
950 }
951 
952 static void xive_source_reset(void *dev)
953 {
954     XiveSource *xsrc = XIVE_SOURCE(dev);
955 
956     /* Do not clear the LSI bitmap */
957 
958     /* PQs are initialized to 0b01 (Q=1) which corresponds to "ints off" */
959     memset(xsrc->status, XIVE_ESB_OFF, xsrc->nr_irqs);
960 }
961 
962 static void xive_source_realize(DeviceState *dev, Error **errp)
963 {
964     XiveSource *xsrc = XIVE_SOURCE(dev);
965     Object *obj;
966     Error *local_err = NULL;
967 
968     obj = object_property_get_link(OBJECT(dev), "xive", &local_err);
969     if (!obj) {
970         error_propagate(errp, local_err);
971         error_prepend(errp, "required link 'xive' not found: ");
972         return;
973     }
974 
975     xsrc->xive = XIVE_NOTIFIER(obj);
976 
977     if (!xsrc->nr_irqs) {
978         error_setg(errp, "Number of interrupt needs to be greater than 0");
979         return;
980     }
981 
982     if (xsrc->esb_shift != XIVE_ESB_4K &&
983         xsrc->esb_shift != XIVE_ESB_4K_2PAGE &&
984         xsrc->esb_shift != XIVE_ESB_64K &&
985         xsrc->esb_shift != XIVE_ESB_64K_2PAGE) {
986         error_setg(errp, "Invalid ESB shift setting");
987         return;
988     }
989 
990     xsrc->status = g_malloc0(xsrc->nr_irqs);
991     xsrc->lsi_map = bitmap_new(xsrc->nr_irqs);
992 
993     memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
994                           &xive_source_esb_ops, xsrc, "xive.esb",
995                           (1ull << xsrc->esb_shift) * xsrc->nr_irqs);
996 
997     qemu_register_reset(xive_source_reset, dev);
998 }
999 
1000 static const VMStateDescription vmstate_xive_source = {
1001     .name = TYPE_XIVE_SOURCE,
1002     .version_id = 1,
1003     .minimum_version_id = 1,
1004     .fields = (VMStateField[]) {
1005         VMSTATE_UINT32_EQUAL(nr_irqs, XiveSource, NULL),
1006         VMSTATE_VBUFFER_UINT32(status, XiveSource, 1, NULL, nr_irqs),
1007         VMSTATE_END_OF_LIST()
1008     },
1009 };
1010 
1011 /*
1012  * The default XIVE interrupt source setting for the ESB MMIOs is two
1013  * 64k pages without Store EOI, to be in sync with KVM.
1014  */
1015 static Property xive_source_properties[] = {
1016     DEFINE_PROP_UINT64("flags", XiveSource, esb_flags, 0),
1017     DEFINE_PROP_UINT32("nr-irqs", XiveSource, nr_irqs, 0),
1018     DEFINE_PROP_UINT32("shift", XiveSource, esb_shift, XIVE_ESB_64K_2PAGE),
1019     DEFINE_PROP_END_OF_LIST(),
1020 };
1021 
1022 static void xive_source_class_init(ObjectClass *klass, void *data)
1023 {
1024     DeviceClass *dc = DEVICE_CLASS(klass);
1025 
1026     dc->desc    = "XIVE Interrupt Source";
1027     dc->props   = xive_source_properties;
1028     dc->realize = xive_source_realize;
1029     dc->vmsd    = &vmstate_xive_source;
1030 }
1031 
1032 static const TypeInfo xive_source_info = {
1033     .name          = TYPE_XIVE_SOURCE,
1034     .parent        = TYPE_DEVICE,
1035     .instance_size = sizeof(XiveSource),
1036     .class_init    = xive_source_class_init,
1037 };
1038 
1039 /*
1040  * XiveEND helpers
1041  */
1042 
1043 void xive_end_queue_pic_print_info(XiveEND *end, uint32_t width, Monitor *mon)
1044 {
1045     uint64_t qaddr_base = xive_end_qaddr(end);
1046     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1047     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1048     uint32_t qentries = 1 << (qsize + 10);
1049     int i;
1050 
1051     /*
1052      * print out the [ (qindex - (width - 1)) .. (qindex + 1)] window
1053      */
1054     monitor_printf(mon, " [ ");
1055     qindex = (qindex - (width - 1)) & (qentries - 1);
1056     for (i = 0; i < width; i++) {
1057         uint64_t qaddr = qaddr_base + (qindex << 2);
1058         uint32_t qdata = -1;
1059 
1060         if (dma_memory_read(&address_space_memory, qaddr, &qdata,
1061                             sizeof(qdata))) {
1062             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to read EQ @0x%"
1063                           HWADDR_PRIx "\n", qaddr);
1064             return;
1065         }
1066         monitor_printf(mon, "%s%08x ", i == width - 1 ? "^" : "",
1067                        be32_to_cpu(qdata));
1068         qindex = (qindex + 1) & (qentries - 1);
1069     }
1070 }
1071 
1072 void xive_end_pic_print_info(XiveEND *end, uint32_t end_idx, Monitor *mon)
1073 {
1074     uint64_t qaddr_base = xive_end_qaddr(end);
1075     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1076     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1077     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1078     uint32_t qentries = 1 << (qsize + 10);
1079 
1080     uint32_t nvt = xive_get_field32(END_W6_NVT_INDEX, end->w6);
1081     uint8_t priority = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
1082 
1083     if (!xive_end_is_valid(end)) {
1084         return;
1085     }
1086 
1087     monitor_printf(mon, "  %08x %c%c%c%c%c prio:%d nvt:%04x eq:@%08"PRIx64
1088                    "% 6d/%5d ^%d", end_idx,
1089                    xive_end_is_valid(end)    ? 'v' : '-',
1090                    xive_end_is_enqueue(end)  ? 'q' : '-',
1091                    xive_end_is_notify(end)   ? 'n' : '-',
1092                    xive_end_is_backlog(end)  ? 'b' : '-',
1093                    xive_end_is_escalate(end) ? 'e' : '-',
1094                    priority, nvt, qaddr_base, qindex, qentries, qgen);
1095 
1096     xive_end_queue_pic_print_info(end, 6, mon);
1097     monitor_printf(mon, "]\n");
1098 }
1099 
1100 static void xive_end_enqueue(XiveEND *end, uint32_t data)
1101 {
1102     uint64_t qaddr_base = xive_end_qaddr(end);
1103     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1104     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1105     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1106 
1107     uint64_t qaddr = qaddr_base + (qindex << 2);
1108     uint32_t qdata = cpu_to_be32((qgen << 31) | (data & 0x7fffffff));
1109     uint32_t qentries = 1 << (qsize + 10);
1110 
1111     if (dma_memory_write(&address_space_memory, qaddr, &qdata, sizeof(qdata))) {
1112         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to write END data @0x%"
1113                       HWADDR_PRIx "\n", qaddr);
1114         return;
1115     }
1116 
1117     qindex = (qindex + 1) & (qentries - 1);
1118     if (qindex == 0) {
1119         qgen ^= 1;
1120         end->w1 = xive_set_field32(END_W1_GENERATION, end->w1, qgen);
1121     }
1122     end->w1 = xive_set_field32(END_W1_PAGE_OFF, end->w1, qindex);
1123 }
1124 
1125 /*
1126  * XIVE Router (aka. Virtualization Controller or IVRE)
1127  */
1128 
1129 int xive_router_get_eas(XiveRouter *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1130                         XiveEAS *eas)
1131 {
1132     XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1133 
1134     return xrc->get_eas(xrtr, eas_blk, eas_idx, eas);
1135 }
1136 
1137 int xive_router_get_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1138                         XiveEND *end)
1139 {
1140    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1141 
1142    return xrc->get_end(xrtr, end_blk, end_idx, end);
1143 }
1144 
1145 int xive_router_write_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1146                           XiveEND *end, uint8_t word_number)
1147 {
1148    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1149 
1150    return xrc->write_end(xrtr, end_blk, end_idx, end, word_number);
1151 }
1152 
1153 int xive_router_get_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1154                         XiveNVT *nvt)
1155 {
1156    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1157 
1158    return xrc->get_nvt(xrtr, nvt_blk, nvt_idx, nvt);
1159 }
1160 
1161 int xive_router_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1162                         XiveNVT *nvt, uint8_t word_number)
1163 {
1164    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1165 
1166    return xrc->write_nvt(xrtr, nvt_blk, nvt_idx, nvt, word_number);
1167 }
1168 
1169 XiveTCTX *xive_router_get_tctx(XiveRouter *xrtr, CPUState *cs)
1170 {
1171     XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1172 
1173     return xrc->get_tctx(xrtr, cs);
1174 }
1175 
1176 /*
1177  * By default on P9, the HW CAM line (23bits) is hardwired to :
1178  *
1179  *   0x000||0b1||4Bit chip number||7Bit Thread number.
1180  *
1181  * When the block grouping is enabled, the CAM line is changed to :
1182  *
1183  *   4Bit chip number||0x001||7Bit Thread number.
1184  */
1185 static uint32_t hw_cam_line(uint8_t chip_id, uint8_t tid)
1186 {
1187     return 1 << 11 | (chip_id & 0xf) << 7 | (tid & 0x7f);
1188 }
1189 
1190 static bool xive_presenter_tctx_match_hw(XiveTCTX *tctx,
1191                                          uint8_t nvt_blk, uint32_t nvt_idx)
1192 {
1193     CPUPPCState *env = &POWERPC_CPU(tctx->cs)->env;
1194     uint32_t pir = env->spr_cb[SPR_PIR].default_value;
1195 
1196     return hw_cam_line((pir >> 8) & 0xf, pir & 0x7f) ==
1197         hw_cam_line(nvt_blk, nvt_idx);
1198 }
1199 
1200 /*
1201  * The thread context register words are in big-endian format.
1202  */
1203 static int xive_presenter_tctx_match(XiveTCTX *tctx, uint8_t format,
1204                                      uint8_t nvt_blk, uint32_t nvt_idx,
1205                                      bool cam_ignore, uint32_t logic_serv)
1206 {
1207     uint32_t cam = xive_nvt_cam_line(nvt_blk, nvt_idx);
1208     uint32_t qw3w2 = xive_tctx_word2(&tctx->regs[TM_QW3_HV_PHYS]);
1209     uint32_t qw2w2 = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
1210     uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
1211     uint32_t qw0w2 = xive_tctx_word2(&tctx->regs[TM_QW0_USER]);
1212 
1213     /*
1214      * TODO (PowerNV): ignore mode. The low order bits of the NVT
1215      * identifier are ignored in the "CAM" match.
1216      */
1217 
1218     if (format == 0) {
1219         if (cam_ignore == true) {
1220             /*
1221              * F=0 & i=1: Logical server notification (bits ignored at
1222              * the end of the NVT identifier)
1223              */
1224             qemu_log_mask(LOG_UNIMP, "XIVE: no support for LS NVT %x/%x\n",
1225                           nvt_blk, nvt_idx);
1226              return -1;
1227         }
1228 
1229         /* F=0 & i=0: Specific NVT notification */
1230 
1231         /* PHYS ring */
1232         if ((be32_to_cpu(qw3w2) & TM_QW3W2_VT) &&
1233             xive_presenter_tctx_match_hw(tctx, nvt_blk, nvt_idx)) {
1234             return TM_QW3_HV_PHYS;
1235         }
1236 
1237         /* HV POOL ring */
1238         if ((be32_to_cpu(qw2w2) & TM_QW2W2_VP) &&
1239             cam == xive_get_field32(TM_QW2W2_POOL_CAM, qw2w2)) {
1240             return TM_QW2_HV_POOL;
1241         }
1242 
1243         /* OS ring */
1244         if ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1245             cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) {
1246             return TM_QW1_OS;
1247         }
1248     } else {
1249         /* F=1 : User level Event-Based Branch (EBB) notification */
1250 
1251         /* USER ring */
1252         if  ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1253              (cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) &&
1254              (be32_to_cpu(qw0w2) & TM_QW0W2_VU) &&
1255              (logic_serv == xive_get_field32(TM_QW0W2_LOGIC_SERV, qw0w2))) {
1256             return TM_QW0_USER;
1257         }
1258     }
1259     return -1;
1260 }
1261 
1262 typedef struct XiveTCTXMatch {
1263     XiveTCTX *tctx;
1264     uint8_t ring;
1265 } XiveTCTXMatch;
1266 
1267 static bool xive_presenter_match(XiveRouter *xrtr, uint8_t format,
1268                                  uint8_t nvt_blk, uint32_t nvt_idx,
1269                                  bool cam_ignore, uint8_t priority,
1270                                  uint32_t logic_serv, XiveTCTXMatch *match)
1271 {
1272     CPUState *cs;
1273 
1274     /*
1275      * TODO (PowerNV): handle chip_id overwrite of block field for
1276      * hardwired CAM compares
1277      */
1278 
1279     CPU_FOREACH(cs) {
1280         XiveTCTX *tctx = xive_router_get_tctx(xrtr, cs);
1281         int ring;
1282 
1283         /*
1284          * HW checks that the CPU is enabled in the Physical Thread
1285          * Enable Register (PTER).
1286          */
1287 
1288         /*
1289          * Check the thread context CAM lines and record matches. We
1290          * will handle CPU exception delivery later
1291          */
1292         ring = xive_presenter_tctx_match(tctx, format, nvt_blk, nvt_idx,
1293                                          cam_ignore, logic_serv);
1294         /*
1295          * Save the context and follow on to catch duplicates, that we
1296          * don't support yet.
1297          */
1298         if (ring != -1) {
1299             if (match->tctx) {
1300                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: already found a thread "
1301                               "context NVT %x/%x\n", nvt_blk, nvt_idx);
1302                 return false;
1303             }
1304 
1305             match->ring = ring;
1306             match->tctx = tctx;
1307         }
1308     }
1309 
1310     if (!match->tctx) {
1311         qemu_log_mask(LOG_UNIMP, "XIVE: NVT %x/%x is not dispatched\n",
1312                       nvt_blk, nvt_idx);
1313         return false;
1314     }
1315 
1316     return true;
1317 }
1318 
1319 /*
1320  * This is our simple Xive Presenter Engine model. It is merged in the
1321  * Router as it does not require an extra object.
1322  *
1323  * It receives notification requests sent by the IVRE to find one
1324  * matching NVT (or more) dispatched on the processor threads. In case
1325  * of a single NVT notification, the process is abreviated and the
1326  * thread is signaled if a match is found. In case of a logical server
1327  * notification (bits ignored at the end of the NVT identifier), the
1328  * IVPE and IVRE select a winning thread using different filters. This
1329  * involves 2 or 3 exchanges on the PowerBus that the model does not
1330  * support.
1331  *
1332  * The parameters represent what is sent on the PowerBus
1333  */
1334 static void xive_presenter_notify(XiveRouter *xrtr, uint8_t format,
1335                                   uint8_t nvt_blk, uint32_t nvt_idx,
1336                                   bool cam_ignore, uint8_t priority,
1337                                   uint32_t logic_serv)
1338 {
1339     XiveNVT nvt;
1340     XiveTCTXMatch match = { .tctx = NULL, .ring = 0 };
1341     bool found;
1342 
1343     /* NVT cache lookup */
1344     if (xive_router_get_nvt(xrtr, nvt_blk, nvt_idx, &nvt)) {
1345         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no NVT %x/%x\n",
1346                       nvt_blk, nvt_idx);
1347         return;
1348     }
1349 
1350     if (!xive_nvt_is_valid(&nvt)) {
1351         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is invalid\n",
1352                       nvt_blk, nvt_idx);
1353         return;
1354     }
1355 
1356     found = xive_presenter_match(xrtr, format, nvt_blk, nvt_idx, cam_ignore,
1357                                  priority, logic_serv, &match);
1358     if (found) {
1359         ipb_update(&match.tctx->regs[match.ring], priority);
1360         xive_tctx_notify(match.tctx, match.ring);
1361         return;
1362     }
1363 
1364     /* Record the IPB in the associated NVT structure */
1365     ipb_update((uint8_t *) &nvt.w4, priority);
1366     xive_router_write_nvt(xrtr, nvt_blk, nvt_idx, &nvt, 4);
1367 
1368     /*
1369      * If no matching NVT is dispatched on a HW thread :
1370      * - update the NVT structure if backlog is activated
1371      * - escalate (ESe PQ bits and EAS in w4-5) if escalation is
1372      *   activated
1373      */
1374 }
1375 
1376 /*
1377  * An END trigger can come from an event trigger (IPI or HW) or from
1378  * another chip. We don't model the PowerBus but the END trigger
1379  * message has the same parameters than in the function below.
1380  */
1381 static void xive_router_end_notify(XiveRouter *xrtr, uint8_t end_blk,
1382                                    uint32_t end_idx, uint32_t end_data)
1383 {
1384     XiveEND end;
1385     uint8_t priority;
1386     uint8_t format;
1387 
1388     /* END cache lookup */
1389     if (xive_router_get_end(xrtr, end_blk, end_idx, &end)) {
1390         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1391                       end_idx);
1392         return;
1393     }
1394 
1395     if (!xive_end_is_valid(&end)) {
1396         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1397                       end_blk, end_idx);
1398         return;
1399     }
1400 
1401     if (xive_end_is_enqueue(&end)) {
1402         xive_end_enqueue(&end, end_data);
1403         /* Enqueuing event data modifies the EQ toggle and index */
1404         xive_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1405     }
1406 
1407     /*
1408      * The W7 format depends on the F bit in W6. It defines the type
1409      * of the notification :
1410      *
1411      *   F=0 : single or multiple NVT notification
1412      *   F=1 : User level Event-Based Branch (EBB) notification, no
1413      *         priority
1414      */
1415     format = xive_get_field32(END_W6_FORMAT_BIT, end.w6);
1416     priority = xive_get_field32(END_W7_F0_PRIORITY, end.w7);
1417 
1418     /* The END is masked */
1419     if (format == 0 && priority == 0xff) {
1420         return;
1421     }
1422 
1423     /*
1424      * Check the END ESn (Event State Buffer for notification) for
1425      * even futher coalescing in the Router
1426      */
1427     if (!xive_end_is_notify(&end)) {
1428         uint8_t pq = xive_get_field32(END_W1_ESn, end.w1);
1429         bool notify = xive_esb_trigger(&pq);
1430 
1431         if (pq != xive_get_field32(END_W1_ESn, end.w1)) {
1432             end.w1 = xive_set_field32(END_W1_ESn, end.w1, pq);
1433             xive_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1434         }
1435 
1436         /* ESn[Q]=1 : end of notification */
1437         if (!notify) {
1438             return;
1439         }
1440     }
1441 
1442     /*
1443      * Follows IVPE notification
1444      */
1445     xive_presenter_notify(xrtr, format,
1446                           xive_get_field32(END_W6_NVT_BLOCK, end.w6),
1447                           xive_get_field32(END_W6_NVT_INDEX, end.w6),
1448                           xive_get_field32(END_W7_F0_IGNORE, end.w7),
1449                           priority,
1450                           xive_get_field32(END_W7_F1_LOG_SERVER_ID, end.w7));
1451 
1452     /* TODO: Auto EOI. */
1453 }
1454 
1455 void xive_router_notify(XiveNotifier *xn, uint32_t lisn)
1456 {
1457     XiveRouter *xrtr = XIVE_ROUTER(xn);
1458     uint8_t eas_blk = XIVE_SRCNO_BLOCK(lisn);
1459     uint32_t eas_idx = XIVE_SRCNO_INDEX(lisn);
1460     XiveEAS eas;
1461 
1462     /* EAS cache lookup */
1463     if (xive_router_get_eas(xrtr, eas_blk, eas_idx, &eas)) {
1464         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN %x\n", lisn);
1465         return;
1466     }
1467 
1468     /*
1469      * The IVRE checks the State Bit Cache at this point. We skip the
1470      * SBC lookup because the state bits of the sources are modeled
1471      * internally in QEMU.
1472      */
1473 
1474     if (!xive_eas_is_valid(&eas)) {
1475         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid LISN %x\n", lisn);
1476         return;
1477     }
1478 
1479     if (xive_eas_is_masked(&eas)) {
1480         /* Notification completed */
1481         return;
1482     }
1483 
1484     /*
1485      * The event trigger becomes an END trigger
1486      */
1487     xive_router_end_notify(xrtr,
1488                            xive_get_field64(EAS_END_BLOCK, eas.w),
1489                            xive_get_field64(EAS_END_INDEX, eas.w),
1490                            xive_get_field64(EAS_END_DATA,  eas.w));
1491 }
1492 
1493 static void xive_router_class_init(ObjectClass *klass, void *data)
1494 {
1495     DeviceClass *dc = DEVICE_CLASS(klass);
1496     XiveNotifierClass *xnc = XIVE_NOTIFIER_CLASS(klass);
1497 
1498     dc->desc    = "XIVE Router Engine";
1499     xnc->notify = xive_router_notify;
1500 }
1501 
1502 static const TypeInfo xive_router_info = {
1503     .name          = TYPE_XIVE_ROUTER,
1504     .parent        = TYPE_SYS_BUS_DEVICE,
1505     .abstract      = true,
1506     .class_size    = sizeof(XiveRouterClass),
1507     .class_init    = xive_router_class_init,
1508     .interfaces    = (InterfaceInfo[]) {
1509         { TYPE_XIVE_NOTIFIER },
1510         { }
1511     }
1512 };
1513 
1514 void xive_eas_pic_print_info(XiveEAS *eas, uint32_t lisn, Monitor *mon)
1515 {
1516     if (!xive_eas_is_valid(eas)) {
1517         return;
1518     }
1519 
1520     monitor_printf(mon, "  %08x %s end:%02x/%04x data:%08x\n",
1521                    lisn, xive_eas_is_masked(eas) ? "M" : " ",
1522                    (uint8_t)  xive_get_field64(EAS_END_BLOCK, eas->w),
1523                    (uint32_t) xive_get_field64(EAS_END_INDEX, eas->w),
1524                    (uint32_t) xive_get_field64(EAS_END_DATA, eas->w));
1525 }
1526 
1527 /*
1528  * END ESB MMIO loads
1529  */
1530 static uint64_t xive_end_source_read(void *opaque, hwaddr addr, unsigned size)
1531 {
1532     XiveENDSource *xsrc = XIVE_END_SOURCE(opaque);
1533     uint32_t offset = addr & 0xFFF;
1534     uint8_t end_blk;
1535     uint32_t end_idx;
1536     XiveEND end;
1537     uint32_t end_esmask;
1538     uint8_t pq;
1539     uint64_t ret = -1;
1540 
1541     end_blk = xsrc->block_id;
1542     end_idx = addr >> (xsrc->esb_shift + 1);
1543 
1544     if (xive_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
1545         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1546                       end_idx);
1547         return -1;
1548     }
1549 
1550     if (!xive_end_is_valid(&end)) {
1551         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1552                       end_blk, end_idx);
1553         return -1;
1554     }
1555 
1556     end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END_W1_ESn : END_W1_ESe;
1557     pq = xive_get_field32(end_esmask, end.w1);
1558 
1559     switch (offset) {
1560     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
1561         ret = xive_esb_eoi(&pq);
1562 
1563         /* Forward the source event notification for routing ?? */
1564         break;
1565 
1566     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
1567         ret = pq;
1568         break;
1569 
1570     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1571     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1572     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1573     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1574         ret = xive_esb_set(&pq, (offset >> 8) & 0x3);
1575         break;
1576     default:
1577         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB load addr %d\n",
1578                       offset);
1579         return -1;
1580     }
1581 
1582     if (pq != xive_get_field32(end_esmask, end.w1)) {
1583         end.w1 = xive_set_field32(end_esmask, end.w1, pq);
1584         xive_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
1585     }
1586 
1587     return ret;
1588 }
1589 
1590 /*
1591  * END ESB MMIO stores are invalid
1592  */
1593 static void xive_end_source_write(void *opaque, hwaddr addr,
1594                                   uint64_t value, unsigned size)
1595 {
1596     qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr 0x%"
1597                   HWADDR_PRIx"\n", addr);
1598 }
1599 
1600 static const MemoryRegionOps xive_end_source_ops = {
1601     .read = xive_end_source_read,
1602     .write = xive_end_source_write,
1603     .endianness = DEVICE_BIG_ENDIAN,
1604     .valid = {
1605         .min_access_size = 8,
1606         .max_access_size = 8,
1607     },
1608     .impl = {
1609         .min_access_size = 8,
1610         .max_access_size = 8,
1611     },
1612 };
1613 
1614 static void xive_end_source_realize(DeviceState *dev, Error **errp)
1615 {
1616     XiveENDSource *xsrc = XIVE_END_SOURCE(dev);
1617     Object *obj;
1618     Error *local_err = NULL;
1619 
1620     obj = object_property_get_link(OBJECT(dev), "xive", &local_err);
1621     if (!obj) {
1622         error_propagate(errp, local_err);
1623         error_prepend(errp, "required link 'xive' not found: ");
1624         return;
1625     }
1626 
1627     xsrc->xrtr = XIVE_ROUTER(obj);
1628 
1629     if (!xsrc->nr_ends) {
1630         error_setg(errp, "Number of interrupt needs to be greater than 0");
1631         return;
1632     }
1633 
1634     if (xsrc->esb_shift != XIVE_ESB_4K &&
1635         xsrc->esb_shift != XIVE_ESB_64K) {
1636         error_setg(errp, "Invalid ESB shift setting");
1637         return;
1638     }
1639 
1640     /*
1641      * Each END is assigned an even/odd pair of MMIO pages, the even page
1642      * manages the ESn field while the odd page manages the ESe field.
1643      */
1644     memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
1645                           &xive_end_source_ops, xsrc, "xive.end",
1646                           (1ull << (xsrc->esb_shift + 1)) * xsrc->nr_ends);
1647 }
1648 
1649 static Property xive_end_source_properties[] = {
1650     DEFINE_PROP_UINT8("block-id", XiveENDSource, block_id, 0),
1651     DEFINE_PROP_UINT32("nr-ends", XiveENDSource, nr_ends, 0),
1652     DEFINE_PROP_UINT32("shift", XiveENDSource, esb_shift, XIVE_ESB_64K),
1653     DEFINE_PROP_END_OF_LIST(),
1654 };
1655 
1656 static void xive_end_source_class_init(ObjectClass *klass, void *data)
1657 {
1658     DeviceClass *dc = DEVICE_CLASS(klass);
1659 
1660     dc->desc    = "XIVE END Source";
1661     dc->props   = xive_end_source_properties;
1662     dc->realize = xive_end_source_realize;
1663 }
1664 
1665 static const TypeInfo xive_end_source_info = {
1666     .name          = TYPE_XIVE_END_SOURCE,
1667     .parent        = TYPE_DEVICE,
1668     .instance_size = sizeof(XiveENDSource),
1669     .class_init    = xive_end_source_class_init,
1670 };
1671 
1672 /*
1673  * XIVE Notifier
1674  */
1675 static const TypeInfo xive_notifier_info = {
1676     .name = TYPE_XIVE_NOTIFIER,
1677     .parent = TYPE_INTERFACE,
1678     .class_size = sizeof(XiveNotifierClass),
1679 };
1680 
1681 static void xive_register_types(void)
1682 {
1683     type_register_static(&xive_source_info);
1684     type_register_static(&xive_notifier_info);
1685     type_register_static(&xive_router_info);
1686     type_register_static(&xive_end_source_info);
1687     type_register_static(&xive_tctx_info);
1688 }
1689 
1690 type_init(xive_register_types)
1691