xref: /openbmc/qemu/hw/intc/xive.c (revision 64552b6b)
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 "qemu/module.h"
13 #include "qapi/error.h"
14 #include "target/ppc/cpu.h"
15 #include "sysemu/cpus.h"
16 #include "sysemu/dma.h"
17 #include "sysemu/reset.h"
18 #include "hw/qdev-properties.h"
19 #include "monitor/monitor.h"
20 #include "hw/irq.h"
21 #include "hw/ppc/xive.h"
22 #include "hw/ppc/xive_regs.h"
23 
24 /*
25  * XIVE Thread Interrupt Management context
26  */
27 
28 /*
29  * Convert a priority number to an Interrupt Pending Buffer (IPB)
30  * register, which indicates a pending interrupt at the priority
31  * corresponding to the bit number
32  */
33 static uint8_t priority_to_ipb(uint8_t priority)
34 {
35     return priority > XIVE_PRIORITY_MAX ?
36         0 : 1 << (XIVE_PRIORITY_MAX - priority);
37 }
38 
39 /*
40  * Convert an Interrupt Pending Buffer (IPB) register to a Pending
41  * Interrupt Priority Register (PIPR), which contains the priority of
42  * the most favored pending notification.
43  */
44 static uint8_t ipb_to_pipr(uint8_t ibp)
45 {
46     return ibp ? clz32((uint32_t)ibp << 24) : 0xff;
47 }
48 
49 static void ipb_update(uint8_t *regs, uint8_t priority)
50 {
51     regs[TM_IPB] |= priority_to_ipb(priority);
52     regs[TM_PIPR] = ipb_to_pipr(regs[TM_IPB]);
53 }
54 
55 static uint8_t exception_mask(uint8_t ring)
56 {
57     switch (ring) {
58     case TM_QW1_OS:
59         return TM_QW1_NSR_EO;
60     case TM_QW3_HV_PHYS:
61         return TM_QW3_NSR_HE;
62     default:
63         g_assert_not_reached();
64     }
65 }
66 
67 static qemu_irq xive_tctx_output(XiveTCTX *tctx, uint8_t ring)
68 {
69         switch (ring) {
70         case TM_QW0_USER:
71                 return 0; /* Not supported */
72         case TM_QW1_OS:
73                 return tctx->os_output;
74         case TM_QW2_HV_POOL:
75         case TM_QW3_HV_PHYS:
76                 return tctx->hv_output;
77         default:
78                 return 0;
79         }
80 }
81 
82 static uint64_t xive_tctx_accept(XiveTCTX *tctx, uint8_t ring)
83 {
84     uint8_t *regs = &tctx->regs[ring];
85     uint8_t nsr = regs[TM_NSR];
86     uint8_t mask = exception_mask(ring);
87 
88     qemu_irq_lower(xive_tctx_output(tctx, ring));
89 
90     if (regs[TM_NSR] & mask) {
91         uint8_t cppr = regs[TM_PIPR];
92 
93         regs[TM_CPPR] = cppr;
94 
95         /* Reset the pending buffer bit */
96         regs[TM_IPB] &= ~priority_to_ipb(cppr);
97         regs[TM_PIPR] = ipb_to_pipr(regs[TM_IPB]);
98 
99         /* Drop Exception bit */
100         regs[TM_NSR] &= ~mask;
101     }
102 
103     return (nsr << 8) | regs[TM_CPPR];
104 }
105 
106 static void xive_tctx_notify(XiveTCTX *tctx, uint8_t ring)
107 {
108     uint8_t *regs = &tctx->regs[ring];
109 
110     if (regs[TM_PIPR] < regs[TM_CPPR]) {
111         switch (ring) {
112         case TM_QW1_OS:
113             regs[TM_NSR] |= TM_QW1_NSR_EO;
114             break;
115         case TM_QW3_HV_PHYS:
116             regs[TM_NSR] |= (TM_QW3_NSR_HE_PHYS << 6);
117             break;
118         default:
119             g_assert_not_reached();
120         }
121         qemu_irq_raise(xive_tctx_output(tctx, ring));
122     }
123 }
124 
125 static void xive_tctx_set_cppr(XiveTCTX *tctx, uint8_t ring, uint8_t cppr)
126 {
127     if (cppr > XIVE_PRIORITY_MAX) {
128         cppr = 0xff;
129     }
130 
131     tctx->regs[ring + TM_CPPR] = cppr;
132 
133     /* CPPR has changed, check if we need to raise a pending exception */
134     xive_tctx_notify(tctx, ring);
135 }
136 
137 static inline uint32_t xive_tctx_word2(uint8_t *ring)
138 {
139     return *((uint32_t *) &ring[TM_WORD2]);
140 }
141 
142 /*
143  * XIVE Thread Interrupt Management Area (TIMA)
144  */
145 
146 static void xive_tm_set_hv_cppr(XiveTCTX *tctx, hwaddr offset,
147                                 uint64_t value, unsigned size)
148 {
149     xive_tctx_set_cppr(tctx, TM_QW3_HV_PHYS, value & 0xff);
150 }
151 
152 static uint64_t xive_tm_ack_hv_reg(XiveTCTX *tctx, hwaddr offset, unsigned size)
153 {
154     return xive_tctx_accept(tctx, TM_QW3_HV_PHYS);
155 }
156 
157 static uint64_t xive_tm_pull_pool_ctx(XiveTCTX *tctx, hwaddr offset,
158                                       unsigned size)
159 {
160     uint32_t qw2w2_prev = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
161     uint32_t qw2w2;
162 
163     qw2w2 = xive_set_field32(TM_QW2W2_VP, qw2w2_prev, 0);
164     memcpy(&tctx->regs[TM_QW2_HV_POOL + TM_WORD2], &qw2w2, 4);
165     return qw2w2;
166 }
167 
168 static void xive_tm_vt_push(XiveTCTX *tctx, hwaddr offset,
169                             uint64_t value, unsigned size)
170 {
171     tctx->regs[TM_QW3_HV_PHYS + TM_WORD2] = value & 0xff;
172 }
173 
174 static uint64_t xive_tm_vt_poll(XiveTCTX *tctx, hwaddr offset, unsigned size)
175 {
176     return tctx->regs[TM_QW3_HV_PHYS + TM_WORD2] & 0xff;
177 }
178 
179 /*
180  * Define an access map for each page of the TIMA that we will use in
181  * the memory region ops to filter values when doing loads and stores
182  * of raw registers values
183  *
184  * Registers accessibility bits :
185  *
186  *    0x0 - no access
187  *    0x1 - write only
188  *    0x2 - read only
189  *    0x3 - read/write
190  */
191 
192 static const uint8_t xive_tm_hw_view[] = {
193     3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-0 User */
194     3, 3, 3, 3,   3, 3, 0, 2,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-1 OS   */
195     0, 0, 3, 3,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-2 POOL */
196     3, 3, 3, 3,   0, 3, 0, 2,   3, 0, 0, 3,   3, 3, 3, 0, /* QW-3 PHYS */
197 };
198 
199 static const uint8_t xive_tm_hv_view[] = {
200     3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-0 User */
201     3, 3, 3, 3,   3, 3, 0, 2,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-1 OS   */
202     0, 0, 3, 3,   0, 0, 0, 0,   0, 3, 3, 3,   0, 0, 0, 0, /* QW-2 POOL */
203     3, 3, 3, 3,   0, 3, 0, 2,   3, 0, 0, 3,   0, 0, 0, 0, /* QW-3 PHYS */
204 };
205 
206 static const uint8_t xive_tm_os_view[] = {
207     3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-0 User */
208     2, 3, 2, 2,   2, 2, 0, 2,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-1 OS   */
209     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-2 POOL */
210     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-3 PHYS */
211 };
212 
213 static const uint8_t xive_tm_user_view[] = {
214     3, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-0 User */
215     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-1 OS   */
216     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-2 POOL */
217     0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-3 PHYS */
218 };
219 
220 /*
221  * Overall TIMA access map for the thread interrupt management context
222  * registers
223  */
224 static const uint8_t *xive_tm_views[] = {
225     [XIVE_TM_HW_PAGE]   = xive_tm_hw_view,
226     [XIVE_TM_HV_PAGE]   = xive_tm_hv_view,
227     [XIVE_TM_OS_PAGE]   = xive_tm_os_view,
228     [XIVE_TM_USER_PAGE] = xive_tm_user_view,
229 };
230 
231 /*
232  * Computes a register access mask for a given offset in the TIMA
233  */
234 static uint64_t xive_tm_mask(hwaddr offset, unsigned size, bool write)
235 {
236     uint8_t page_offset = (offset >> TM_SHIFT) & 0x3;
237     uint8_t reg_offset = offset & 0x3F;
238     uint8_t reg_mask = write ? 0x1 : 0x2;
239     uint64_t mask = 0x0;
240     int i;
241 
242     for (i = 0; i < size; i++) {
243         if (xive_tm_views[page_offset][reg_offset + i] & reg_mask) {
244             mask |= (uint64_t) 0xff << (8 * (size - i - 1));
245         }
246     }
247 
248     return mask;
249 }
250 
251 static void xive_tm_raw_write(XiveTCTX *tctx, hwaddr offset, uint64_t value,
252                               unsigned size)
253 {
254     uint8_t ring_offset = offset & 0x30;
255     uint8_t reg_offset = offset & 0x3F;
256     uint64_t mask = xive_tm_mask(offset, size, true);
257     int i;
258 
259     /*
260      * Only 4 or 8 bytes stores are allowed and the User ring is
261      * excluded
262      */
263     if (size < 4 || !mask || ring_offset == TM_QW0_USER) {
264         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid write access at TIMA @%"
265                       HWADDR_PRIx"\n", offset);
266         return;
267     }
268 
269     /*
270      * Use the register offset for the raw values and filter out
271      * reserved values
272      */
273     for (i = 0; i < size; i++) {
274         uint8_t byte_mask = (mask >> (8 * (size - i - 1)));
275         if (byte_mask) {
276             tctx->regs[reg_offset + i] = (value >> (8 * (size - i - 1))) &
277                 byte_mask;
278         }
279     }
280 }
281 
282 static uint64_t xive_tm_raw_read(XiveTCTX *tctx, hwaddr offset, unsigned size)
283 {
284     uint8_t ring_offset = offset & 0x30;
285     uint8_t reg_offset = offset & 0x3F;
286     uint64_t mask = xive_tm_mask(offset, size, false);
287     uint64_t ret;
288     int i;
289 
290     /*
291      * Only 4 or 8 bytes loads are allowed and the User ring is
292      * excluded
293      */
294     if (size < 4 || !mask || ring_offset == TM_QW0_USER) {
295         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid read access at TIMA @%"
296                       HWADDR_PRIx"\n", offset);
297         return -1;
298     }
299 
300     /* Use the register offset for the raw values */
301     ret = 0;
302     for (i = 0; i < size; i++) {
303         ret |= (uint64_t) tctx->regs[reg_offset + i] << (8 * (size - i - 1));
304     }
305 
306     /* filter out reserved values */
307     return ret & mask;
308 }
309 
310 /*
311  * The TM context is mapped twice within each page. Stores and loads
312  * to the first mapping below 2K write and read the specified values
313  * without modification. The second mapping above 2K performs specific
314  * state changes (side effects) in addition to setting/returning the
315  * interrupt management area context of the processor thread.
316  */
317 static uint64_t xive_tm_ack_os_reg(XiveTCTX *tctx, hwaddr offset, unsigned size)
318 {
319     return xive_tctx_accept(tctx, TM_QW1_OS);
320 }
321 
322 static void xive_tm_set_os_cppr(XiveTCTX *tctx, hwaddr offset,
323                                 uint64_t value, unsigned size)
324 {
325     xive_tctx_set_cppr(tctx, TM_QW1_OS, value & 0xff);
326 }
327 
328 /*
329  * Adjust the IPB to allow a CPU to process event queues of other
330  * priorities during one physical interrupt cycle.
331  */
332 static void xive_tm_set_os_pending(XiveTCTX *tctx, hwaddr offset,
333                                    uint64_t value, unsigned size)
334 {
335     ipb_update(&tctx->regs[TM_QW1_OS], value & 0xff);
336     xive_tctx_notify(tctx, TM_QW1_OS);
337 }
338 
339 /*
340  * Define a mapping of "special" operations depending on the TIMA page
341  * offset and the size of the operation.
342  */
343 typedef struct XiveTmOp {
344     uint8_t  page_offset;
345     uint32_t op_offset;
346     unsigned size;
347     void     (*write_handler)(XiveTCTX *tctx, hwaddr offset, uint64_t value,
348                               unsigned size);
349     uint64_t (*read_handler)(XiveTCTX *tctx, hwaddr offset, unsigned size);
350 } XiveTmOp;
351 
352 static const XiveTmOp xive_tm_operations[] = {
353     /*
354      * MMIOs below 2K : raw values and special operations without side
355      * effects
356      */
357     { XIVE_TM_OS_PAGE, TM_QW1_OS + TM_CPPR,   1, xive_tm_set_os_cppr, NULL },
358     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_CPPR, 1, xive_tm_set_hv_cppr, NULL },
359     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_WORD2, 1, xive_tm_vt_push, NULL },
360     { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_WORD2, 1, NULL, xive_tm_vt_poll },
361 
362     /* MMIOs above 2K : special operations with side effects */
363     { XIVE_TM_OS_PAGE, TM_SPC_ACK_OS_REG,     2, NULL, xive_tm_ack_os_reg },
364     { XIVE_TM_OS_PAGE, TM_SPC_SET_OS_PENDING, 1, xive_tm_set_os_pending, NULL },
365     { XIVE_TM_HV_PAGE, TM_SPC_ACK_HV_REG,     2, NULL, xive_tm_ack_hv_reg },
366     { XIVE_TM_HV_PAGE, TM_SPC_PULL_POOL_CTX,  4, NULL, xive_tm_pull_pool_ctx },
367     { XIVE_TM_HV_PAGE, TM_SPC_PULL_POOL_CTX,  8, NULL, xive_tm_pull_pool_ctx },
368 };
369 
370 static const XiveTmOp *xive_tm_find_op(hwaddr offset, unsigned size, bool write)
371 {
372     uint8_t page_offset = (offset >> TM_SHIFT) & 0x3;
373     uint32_t op_offset = offset & 0xFFF;
374     int i;
375 
376     for (i = 0; i < ARRAY_SIZE(xive_tm_operations); i++) {
377         const XiveTmOp *xto = &xive_tm_operations[i];
378 
379         /* Accesses done from a more privileged TIMA page is allowed */
380         if (xto->page_offset >= page_offset &&
381             xto->op_offset == op_offset &&
382             xto->size == size &&
383             ((write && xto->write_handler) || (!write && xto->read_handler))) {
384             return xto;
385         }
386     }
387     return NULL;
388 }
389 
390 /*
391  * TIMA MMIO handlers
392  */
393 void xive_tctx_tm_write(XiveTCTX *tctx, hwaddr offset, uint64_t value,
394                         unsigned size)
395 {
396     const XiveTmOp *xto;
397 
398     /*
399      * TODO: check V bit in Q[0-3]W2
400      */
401 
402     /*
403      * First, check for special operations in the 2K region
404      */
405     if (offset & 0x800) {
406         xto = xive_tm_find_op(offset, size, true);
407         if (!xto) {
408             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid write access at TIMA"
409                           "@%"HWADDR_PRIx"\n", offset);
410         } else {
411             xto->write_handler(tctx, offset, value, size);
412         }
413         return;
414     }
415 
416     /*
417      * Then, for special operations in the region below 2K.
418      */
419     xto = xive_tm_find_op(offset, size, true);
420     if (xto) {
421         xto->write_handler(tctx, offset, value, size);
422         return;
423     }
424 
425     /*
426      * Finish with raw access to the register values
427      */
428     xive_tm_raw_write(tctx, offset, value, size);
429 }
430 
431 uint64_t xive_tctx_tm_read(XiveTCTX *tctx, hwaddr offset, unsigned size)
432 {
433     const XiveTmOp *xto;
434 
435     /*
436      * TODO: check V bit in Q[0-3]W2
437      */
438 
439     /*
440      * First, check for special operations in the 2K region
441      */
442     if (offset & 0x800) {
443         xto = xive_tm_find_op(offset, size, false);
444         if (!xto) {
445             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid read access to TIMA"
446                           "@%"HWADDR_PRIx"\n", offset);
447             return -1;
448         }
449         return xto->read_handler(tctx, offset, size);
450     }
451 
452     /*
453      * Then, for special operations in the region below 2K.
454      */
455     xto = xive_tm_find_op(offset, size, false);
456     if (xto) {
457         return xto->read_handler(tctx, offset, size);
458     }
459 
460     /*
461      * Finish with raw access to the register values
462      */
463     return xive_tm_raw_read(tctx, offset, size);
464 }
465 
466 static void xive_tm_write(void *opaque, hwaddr offset,
467                           uint64_t value, unsigned size)
468 {
469     XiveTCTX *tctx = xive_router_get_tctx(XIVE_ROUTER(opaque), current_cpu);
470 
471     xive_tctx_tm_write(tctx, offset, value, size);
472 }
473 
474 static uint64_t xive_tm_read(void *opaque, hwaddr offset, unsigned size)
475 {
476     XiveTCTX *tctx = xive_router_get_tctx(XIVE_ROUTER(opaque), current_cpu);
477 
478     return xive_tctx_tm_read(tctx, offset, size);
479 }
480 
481 const MemoryRegionOps xive_tm_ops = {
482     .read = xive_tm_read,
483     .write = xive_tm_write,
484     .endianness = DEVICE_BIG_ENDIAN,
485     .valid = {
486         .min_access_size = 1,
487         .max_access_size = 8,
488     },
489     .impl = {
490         .min_access_size = 1,
491         .max_access_size = 8,
492     },
493 };
494 
495 static char *xive_tctx_ring_print(uint8_t *ring)
496 {
497     uint32_t w2 = xive_tctx_word2(ring);
498 
499     return g_strdup_printf("%02x   %02x  %02x    %02x   %02x  "
500                    "%02x  %02x   %02x  %08x",
501                    ring[TM_NSR], ring[TM_CPPR], ring[TM_IPB], ring[TM_LSMFB],
502                    ring[TM_ACK_CNT], ring[TM_INC], ring[TM_AGE], ring[TM_PIPR],
503                    be32_to_cpu(w2));
504 }
505 
506 static const char * const xive_tctx_ring_names[] = {
507     "USER", "OS", "POOL", "PHYS",
508 };
509 
510 void xive_tctx_pic_print_info(XiveTCTX *tctx, Monitor *mon)
511 {
512     int cpu_index = tctx->cs ? tctx->cs->cpu_index : -1;
513     int i;
514 
515     if (kvm_irqchip_in_kernel()) {
516         Error *local_err = NULL;
517 
518         kvmppc_xive_cpu_synchronize_state(tctx, &local_err);
519         if (local_err) {
520             error_report_err(local_err);
521             return;
522         }
523     }
524 
525     monitor_printf(mon, "CPU[%04x]:   QW   NSR CPPR IPB LSMFB ACK# INC AGE PIPR"
526                    "  W2\n", cpu_index);
527 
528     for (i = 0; i < XIVE_TM_RING_COUNT; i++) {
529         char *s = xive_tctx_ring_print(&tctx->regs[i * XIVE_TM_RING_SIZE]);
530         monitor_printf(mon, "CPU[%04x]: %4s    %s\n", cpu_index,
531                        xive_tctx_ring_names[i], s);
532         g_free(s);
533     }
534 }
535 
536 static void xive_tctx_reset(void *dev)
537 {
538     XiveTCTX *tctx = XIVE_TCTX(dev);
539 
540     memset(tctx->regs, 0, sizeof(tctx->regs));
541 
542     /* Set some defaults */
543     tctx->regs[TM_QW1_OS + TM_LSMFB] = 0xFF;
544     tctx->regs[TM_QW1_OS + TM_ACK_CNT] = 0xFF;
545     tctx->regs[TM_QW1_OS + TM_AGE] = 0xFF;
546 
547     /*
548      * Initialize PIPR to 0xFF to avoid phantom interrupts when the
549      * CPPR is first set.
550      */
551     tctx->regs[TM_QW1_OS + TM_PIPR] =
552         ipb_to_pipr(tctx->regs[TM_QW1_OS + TM_IPB]);
553     tctx->regs[TM_QW3_HV_PHYS + TM_PIPR] =
554         ipb_to_pipr(tctx->regs[TM_QW3_HV_PHYS + TM_IPB]);
555 }
556 
557 static void xive_tctx_realize(DeviceState *dev, Error **errp)
558 {
559     XiveTCTX *tctx = XIVE_TCTX(dev);
560     PowerPCCPU *cpu;
561     CPUPPCState *env;
562     Object *obj;
563     Error *local_err = NULL;
564 
565     obj = object_property_get_link(OBJECT(dev), "cpu", &local_err);
566     if (!obj) {
567         error_propagate(errp, local_err);
568         error_prepend(errp, "required link 'cpu' not found: ");
569         return;
570     }
571 
572     cpu = POWERPC_CPU(obj);
573     tctx->cs = CPU(obj);
574 
575     env = &cpu->env;
576     switch (PPC_INPUT(env)) {
577     case PPC_FLAGS_INPUT_POWER9:
578         tctx->hv_output = env->irq_inputs[POWER9_INPUT_HINT];
579         tctx->os_output = env->irq_inputs[POWER9_INPUT_INT];
580         break;
581 
582     default:
583         error_setg(errp, "XIVE interrupt controller does not support "
584                    "this CPU bus model");
585         return;
586     }
587 
588     /* Connect the presenter to the VCPU (required for CPU hotplug) */
589     if (kvm_irqchip_in_kernel()) {
590         kvmppc_xive_cpu_connect(tctx, &local_err);
591         if (local_err) {
592             error_propagate(errp, local_err);
593             return;
594         }
595     }
596 
597     qemu_register_reset(xive_tctx_reset, dev);
598 }
599 
600 static void xive_tctx_unrealize(DeviceState *dev, Error **errp)
601 {
602     qemu_unregister_reset(xive_tctx_reset, dev);
603 }
604 
605 static int vmstate_xive_tctx_pre_save(void *opaque)
606 {
607     Error *local_err = NULL;
608 
609     if (kvm_irqchip_in_kernel()) {
610         kvmppc_xive_cpu_get_state(XIVE_TCTX(opaque), &local_err);
611         if (local_err) {
612             error_report_err(local_err);
613             return -1;
614         }
615     }
616 
617     return 0;
618 }
619 
620 static int vmstate_xive_tctx_post_load(void *opaque, int version_id)
621 {
622     Error *local_err = NULL;
623 
624     if (kvm_irqchip_in_kernel()) {
625         /*
626          * Required for hotplugged CPU, for which the state comes
627          * after all states of the machine.
628          */
629         kvmppc_xive_cpu_set_state(XIVE_TCTX(opaque), &local_err);
630         if (local_err) {
631             error_report_err(local_err);
632             return -1;
633         }
634     }
635 
636     return 0;
637 }
638 
639 static const VMStateDescription vmstate_xive_tctx = {
640     .name = TYPE_XIVE_TCTX,
641     .version_id = 1,
642     .minimum_version_id = 1,
643     .pre_save = vmstate_xive_tctx_pre_save,
644     .post_load = vmstate_xive_tctx_post_load,
645     .fields = (VMStateField[]) {
646         VMSTATE_BUFFER(regs, XiveTCTX),
647         VMSTATE_END_OF_LIST()
648     },
649 };
650 
651 static void xive_tctx_class_init(ObjectClass *klass, void *data)
652 {
653     DeviceClass *dc = DEVICE_CLASS(klass);
654 
655     dc->desc = "XIVE Interrupt Thread Context";
656     dc->realize = xive_tctx_realize;
657     dc->unrealize = xive_tctx_unrealize;
658     dc->vmsd = &vmstate_xive_tctx;
659 }
660 
661 static const TypeInfo xive_tctx_info = {
662     .name          = TYPE_XIVE_TCTX,
663     .parent        = TYPE_DEVICE,
664     .instance_size = sizeof(XiveTCTX),
665     .class_init    = xive_tctx_class_init,
666 };
667 
668 Object *xive_tctx_create(Object *cpu, XiveRouter *xrtr, Error **errp)
669 {
670     Error *local_err = NULL;
671     Object *obj;
672 
673     obj = object_new(TYPE_XIVE_TCTX);
674     object_property_add_child(cpu, TYPE_XIVE_TCTX, obj, &error_abort);
675     object_unref(obj);
676     object_property_add_const_link(obj, "cpu", cpu, &error_abort);
677     object_property_set_bool(obj, true, "realized", &local_err);
678     if (local_err) {
679         goto error;
680     }
681 
682     return obj;
683 
684 error:
685     object_unparent(obj);
686     error_propagate(errp, local_err);
687     return NULL;
688 }
689 
690 /*
691  * XIVE ESB helpers
692  */
693 
694 static uint8_t xive_esb_set(uint8_t *pq, uint8_t value)
695 {
696     uint8_t old_pq = *pq & 0x3;
697 
698     *pq &= ~0x3;
699     *pq |= value & 0x3;
700 
701     return old_pq;
702 }
703 
704 static bool xive_esb_trigger(uint8_t *pq)
705 {
706     uint8_t old_pq = *pq & 0x3;
707 
708     switch (old_pq) {
709     case XIVE_ESB_RESET:
710         xive_esb_set(pq, XIVE_ESB_PENDING);
711         return true;
712     case XIVE_ESB_PENDING:
713     case XIVE_ESB_QUEUED:
714         xive_esb_set(pq, XIVE_ESB_QUEUED);
715         return false;
716     case XIVE_ESB_OFF:
717         xive_esb_set(pq, XIVE_ESB_OFF);
718         return false;
719     default:
720          g_assert_not_reached();
721     }
722 }
723 
724 static bool xive_esb_eoi(uint8_t *pq)
725 {
726     uint8_t old_pq = *pq & 0x3;
727 
728     switch (old_pq) {
729     case XIVE_ESB_RESET:
730     case XIVE_ESB_PENDING:
731         xive_esb_set(pq, XIVE_ESB_RESET);
732         return false;
733     case XIVE_ESB_QUEUED:
734         xive_esb_set(pq, XIVE_ESB_PENDING);
735         return true;
736     case XIVE_ESB_OFF:
737         xive_esb_set(pq, XIVE_ESB_OFF);
738         return false;
739     default:
740          g_assert_not_reached();
741     }
742 }
743 
744 /*
745  * XIVE Interrupt Source (or IVSE)
746  */
747 
748 uint8_t xive_source_esb_get(XiveSource *xsrc, uint32_t srcno)
749 {
750     assert(srcno < xsrc->nr_irqs);
751 
752     return xsrc->status[srcno] & 0x3;
753 }
754 
755 uint8_t xive_source_esb_set(XiveSource *xsrc, uint32_t srcno, uint8_t pq)
756 {
757     assert(srcno < xsrc->nr_irqs);
758 
759     return xive_esb_set(&xsrc->status[srcno], pq);
760 }
761 
762 /*
763  * Returns whether the event notification should be forwarded.
764  */
765 static bool xive_source_lsi_trigger(XiveSource *xsrc, uint32_t srcno)
766 {
767     uint8_t old_pq = xive_source_esb_get(xsrc, srcno);
768 
769     xsrc->status[srcno] |= XIVE_STATUS_ASSERTED;
770 
771     switch (old_pq) {
772     case XIVE_ESB_RESET:
773         xive_source_esb_set(xsrc, srcno, XIVE_ESB_PENDING);
774         return true;
775     default:
776         return false;
777     }
778 }
779 
780 /*
781  * Returns whether the event notification should be forwarded.
782  */
783 static bool xive_source_esb_trigger(XiveSource *xsrc, uint32_t srcno)
784 {
785     bool ret;
786 
787     assert(srcno < xsrc->nr_irqs);
788 
789     ret = xive_esb_trigger(&xsrc->status[srcno]);
790 
791     if (xive_source_irq_is_lsi(xsrc, srcno) &&
792         xive_source_esb_get(xsrc, srcno) == XIVE_ESB_QUEUED) {
793         qemu_log_mask(LOG_GUEST_ERROR,
794                       "XIVE: queued an event on LSI IRQ %d\n", srcno);
795     }
796 
797     return ret;
798 }
799 
800 /*
801  * Returns whether the event notification should be forwarded.
802  */
803 static bool xive_source_esb_eoi(XiveSource *xsrc, uint32_t srcno)
804 {
805     bool ret;
806 
807     assert(srcno < xsrc->nr_irqs);
808 
809     ret = xive_esb_eoi(&xsrc->status[srcno]);
810 
811     /*
812      * LSI sources do not set the Q bit but they can still be
813      * asserted, in which case we should forward a new event
814      * notification
815      */
816     if (xive_source_irq_is_lsi(xsrc, srcno) &&
817         xsrc->status[srcno] & XIVE_STATUS_ASSERTED) {
818         ret = xive_source_lsi_trigger(xsrc, srcno);
819     }
820 
821     return ret;
822 }
823 
824 /*
825  * Forward the source event notification to the Router
826  */
827 static void xive_source_notify(XiveSource *xsrc, int srcno)
828 {
829     XiveNotifierClass *xnc = XIVE_NOTIFIER_GET_CLASS(xsrc->xive);
830 
831     if (xnc->notify) {
832         xnc->notify(xsrc->xive, srcno);
833     }
834 }
835 
836 /*
837  * In a two pages ESB MMIO setting, even page is the trigger page, odd
838  * page is for management
839  */
840 static inline bool addr_is_even(hwaddr addr, uint32_t shift)
841 {
842     return !((addr >> shift) & 1);
843 }
844 
845 static inline bool xive_source_is_trigger_page(XiveSource *xsrc, hwaddr addr)
846 {
847     return xive_source_esb_has_2page(xsrc) &&
848         addr_is_even(addr, xsrc->esb_shift - 1);
849 }
850 
851 /*
852  * ESB MMIO loads
853  *                      Trigger page    Management/EOI page
854  *
855  * ESB MMIO setting     2 pages         1 or 2 pages
856  *
857  * 0x000 .. 0x3FF       -1              EOI and return 0|1
858  * 0x400 .. 0x7FF       -1              EOI and return 0|1
859  * 0x800 .. 0xBFF       -1              return PQ
860  * 0xC00 .. 0xCFF       -1              return PQ and atomically PQ=00
861  * 0xD00 .. 0xDFF       -1              return PQ and atomically PQ=01
862  * 0xE00 .. 0xDFF       -1              return PQ and atomically PQ=10
863  * 0xF00 .. 0xDFF       -1              return PQ and atomically PQ=11
864  */
865 static uint64_t xive_source_esb_read(void *opaque, hwaddr addr, unsigned size)
866 {
867     XiveSource *xsrc = XIVE_SOURCE(opaque);
868     uint32_t offset = addr & 0xFFF;
869     uint32_t srcno = addr >> xsrc->esb_shift;
870     uint64_t ret = -1;
871 
872     /* In a two pages ESB MMIO setting, trigger page should not be read */
873     if (xive_source_is_trigger_page(xsrc, addr)) {
874         qemu_log_mask(LOG_GUEST_ERROR,
875                       "XIVE: invalid load on IRQ %d trigger page at "
876                       "0x%"HWADDR_PRIx"\n", srcno, addr);
877         return -1;
878     }
879 
880     switch (offset) {
881     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
882         ret = xive_source_esb_eoi(xsrc, srcno);
883 
884         /* Forward the source event notification for routing */
885         if (ret) {
886             xive_source_notify(xsrc, srcno);
887         }
888         break;
889 
890     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
891         ret = xive_source_esb_get(xsrc, srcno);
892         break;
893 
894     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
895     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
896     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
897     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
898         ret = xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
899         break;
900     default:
901         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB load addr %x\n",
902                       offset);
903     }
904 
905     return ret;
906 }
907 
908 /*
909  * ESB MMIO stores
910  *                      Trigger page    Management/EOI page
911  *
912  * ESB MMIO setting     2 pages         1 or 2 pages
913  *
914  * 0x000 .. 0x3FF       Trigger         Trigger
915  * 0x400 .. 0x7FF       Trigger         EOI
916  * 0x800 .. 0xBFF       Trigger         undefined
917  * 0xC00 .. 0xCFF       Trigger         PQ=00
918  * 0xD00 .. 0xDFF       Trigger         PQ=01
919  * 0xE00 .. 0xDFF       Trigger         PQ=10
920  * 0xF00 .. 0xDFF       Trigger         PQ=11
921  */
922 static void xive_source_esb_write(void *opaque, hwaddr addr,
923                                   uint64_t value, unsigned size)
924 {
925     XiveSource *xsrc = XIVE_SOURCE(opaque);
926     uint32_t offset = addr & 0xFFF;
927     uint32_t srcno = addr >> xsrc->esb_shift;
928     bool notify = false;
929 
930     /* In a two pages ESB MMIO setting, trigger page only triggers */
931     if (xive_source_is_trigger_page(xsrc, addr)) {
932         notify = xive_source_esb_trigger(xsrc, srcno);
933         goto out;
934     }
935 
936     switch (offset) {
937     case 0 ... 0x3FF:
938         notify = xive_source_esb_trigger(xsrc, srcno);
939         break;
940 
941     case XIVE_ESB_STORE_EOI ... XIVE_ESB_STORE_EOI + 0x3FF:
942         if (!(xsrc->esb_flags & XIVE_SRC_STORE_EOI)) {
943             qemu_log_mask(LOG_GUEST_ERROR,
944                           "XIVE: invalid Store EOI for IRQ %d\n", srcno);
945             return;
946         }
947 
948         notify = xive_source_esb_eoi(xsrc, srcno);
949         break;
950 
951     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
952     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
953     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
954     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
955         xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
956         break;
957 
958     default:
959         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr %x\n",
960                       offset);
961         return;
962     }
963 
964 out:
965     /* Forward the source event notification for routing */
966     if (notify) {
967         xive_source_notify(xsrc, srcno);
968     }
969 }
970 
971 static const MemoryRegionOps xive_source_esb_ops = {
972     .read = xive_source_esb_read,
973     .write = xive_source_esb_write,
974     .endianness = DEVICE_BIG_ENDIAN,
975     .valid = {
976         .min_access_size = 8,
977         .max_access_size = 8,
978     },
979     .impl = {
980         .min_access_size = 8,
981         .max_access_size = 8,
982     },
983 };
984 
985 void xive_source_set_irq(void *opaque, int srcno, int val)
986 {
987     XiveSource *xsrc = XIVE_SOURCE(opaque);
988     bool notify = false;
989 
990     if (xive_source_irq_is_lsi(xsrc, srcno)) {
991         if (val) {
992             notify = xive_source_lsi_trigger(xsrc, srcno);
993         } else {
994             xsrc->status[srcno] &= ~XIVE_STATUS_ASSERTED;
995         }
996     } else {
997         if (val) {
998             notify = xive_source_esb_trigger(xsrc, srcno);
999         }
1000     }
1001 
1002     /* Forward the source event notification for routing */
1003     if (notify) {
1004         xive_source_notify(xsrc, srcno);
1005     }
1006 }
1007 
1008 void xive_source_pic_print_info(XiveSource *xsrc, uint32_t offset, Monitor *mon)
1009 {
1010     int i;
1011 
1012     for (i = 0; i < xsrc->nr_irqs; i++) {
1013         uint8_t pq = xive_source_esb_get(xsrc, i);
1014 
1015         if (pq == XIVE_ESB_OFF) {
1016             continue;
1017         }
1018 
1019         monitor_printf(mon, "  %08x %s %c%c%c\n", i + offset,
1020                        xive_source_irq_is_lsi(xsrc, i) ? "LSI" : "MSI",
1021                        pq & XIVE_ESB_VAL_P ? 'P' : '-',
1022                        pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1023                        xsrc->status[i] & XIVE_STATUS_ASSERTED ? 'A' : ' ');
1024     }
1025 }
1026 
1027 static void xive_source_reset(void *dev)
1028 {
1029     XiveSource *xsrc = XIVE_SOURCE(dev);
1030 
1031     /* Do not clear the LSI bitmap */
1032 
1033     /* PQs are initialized to 0b01 (Q=1) which corresponds to "ints off" */
1034     memset(xsrc->status, XIVE_ESB_OFF, xsrc->nr_irqs);
1035 }
1036 
1037 static void xive_source_realize(DeviceState *dev, Error **errp)
1038 {
1039     XiveSource *xsrc = XIVE_SOURCE(dev);
1040     Object *obj;
1041     Error *local_err = NULL;
1042 
1043     obj = object_property_get_link(OBJECT(dev), "xive", &local_err);
1044     if (!obj) {
1045         error_propagate(errp, local_err);
1046         error_prepend(errp, "required link 'xive' not found: ");
1047         return;
1048     }
1049 
1050     xsrc->xive = XIVE_NOTIFIER(obj);
1051 
1052     if (!xsrc->nr_irqs) {
1053         error_setg(errp, "Number of interrupt needs to be greater than 0");
1054         return;
1055     }
1056 
1057     if (xsrc->esb_shift != XIVE_ESB_4K &&
1058         xsrc->esb_shift != XIVE_ESB_4K_2PAGE &&
1059         xsrc->esb_shift != XIVE_ESB_64K &&
1060         xsrc->esb_shift != XIVE_ESB_64K_2PAGE) {
1061         error_setg(errp, "Invalid ESB shift setting");
1062         return;
1063     }
1064 
1065     xsrc->status = g_malloc0(xsrc->nr_irqs);
1066     xsrc->lsi_map = bitmap_new(xsrc->nr_irqs);
1067 
1068     if (!kvm_irqchip_in_kernel()) {
1069         memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
1070                               &xive_source_esb_ops, xsrc, "xive.esb",
1071                               (1ull << xsrc->esb_shift) * xsrc->nr_irqs);
1072     }
1073 
1074     qemu_register_reset(xive_source_reset, dev);
1075 }
1076 
1077 static const VMStateDescription vmstate_xive_source = {
1078     .name = TYPE_XIVE_SOURCE,
1079     .version_id = 1,
1080     .minimum_version_id = 1,
1081     .fields = (VMStateField[]) {
1082         VMSTATE_UINT32_EQUAL(nr_irqs, XiveSource, NULL),
1083         VMSTATE_VBUFFER_UINT32(status, XiveSource, 1, NULL, nr_irqs),
1084         VMSTATE_END_OF_LIST()
1085     },
1086 };
1087 
1088 /*
1089  * The default XIVE interrupt source setting for the ESB MMIOs is two
1090  * 64k pages without Store EOI, to be in sync with KVM.
1091  */
1092 static Property xive_source_properties[] = {
1093     DEFINE_PROP_UINT64("flags", XiveSource, esb_flags, 0),
1094     DEFINE_PROP_UINT32("nr-irqs", XiveSource, nr_irqs, 0),
1095     DEFINE_PROP_UINT32("shift", XiveSource, esb_shift, XIVE_ESB_64K_2PAGE),
1096     DEFINE_PROP_END_OF_LIST(),
1097 };
1098 
1099 static void xive_source_class_init(ObjectClass *klass, void *data)
1100 {
1101     DeviceClass *dc = DEVICE_CLASS(klass);
1102 
1103     dc->desc    = "XIVE Interrupt Source";
1104     dc->props   = xive_source_properties;
1105     dc->realize = xive_source_realize;
1106     dc->vmsd    = &vmstate_xive_source;
1107 }
1108 
1109 static const TypeInfo xive_source_info = {
1110     .name          = TYPE_XIVE_SOURCE,
1111     .parent        = TYPE_DEVICE,
1112     .instance_size = sizeof(XiveSource),
1113     .class_init    = xive_source_class_init,
1114 };
1115 
1116 /*
1117  * XiveEND helpers
1118  */
1119 
1120 void xive_end_queue_pic_print_info(XiveEND *end, uint32_t width, Monitor *mon)
1121 {
1122     uint64_t qaddr_base = xive_end_qaddr(end);
1123     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1124     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1125     uint32_t qentries = 1 << (qsize + 10);
1126     int i;
1127 
1128     /*
1129      * print out the [ (qindex - (width - 1)) .. (qindex + 1)] window
1130      */
1131     monitor_printf(mon, " [ ");
1132     qindex = (qindex - (width - 1)) & (qentries - 1);
1133     for (i = 0; i < width; i++) {
1134         uint64_t qaddr = qaddr_base + (qindex << 2);
1135         uint32_t qdata = -1;
1136 
1137         if (dma_memory_read(&address_space_memory, qaddr, &qdata,
1138                             sizeof(qdata))) {
1139             qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to read EQ @0x%"
1140                           HWADDR_PRIx "\n", qaddr);
1141             return;
1142         }
1143         monitor_printf(mon, "%s%08x ", i == width - 1 ? "^" : "",
1144                        be32_to_cpu(qdata));
1145         qindex = (qindex + 1) & (qentries - 1);
1146     }
1147 }
1148 
1149 void xive_end_pic_print_info(XiveEND *end, uint32_t end_idx, Monitor *mon)
1150 {
1151     uint64_t qaddr_base = xive_end_qaddr(end);
1152     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1153     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1154     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1155     uint32_t qentries = 1 << (qsize + 10);
1156 
1157     uint32_t nvt = xive_get_field32(END_W6_NVT_INDEX, end->w6);
1158     uint8_t priority = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
1159 
1160     if (!xive_end_is_valid(end)) {
1161         return;
1162     }
1163 
1164     monitor_printf(mon, "  %08x %c%c%c%c%c prio:%d nvt:%04x eq:@%08"PRIx64
1165                    "% 6d/%5d ^%d", end_idx,
1166                    xive_end_is_valid(end)    ? 'v' : '-',
1167                    xive_end_is_enqueue(end)  ? 'q' : '-',
1168                    xive_end_is_notify(end)   ? 'n' : '-',
1169                    xive_end_is_backlog(end)  ? 'b' : '-',
1170                    xive_end_is_escalate(end) ? 'e' : '-',
1171                    priority, nvt, qaddr_base, qindex, qentries, qgen);
1172 
1173     xive_end_queue_pic_print_info(end, 6, mon);
1174     monitor_printf(mon, "]\n");
1175 }
1176 
1177 static void xive_end_enqueue(XiveEND *end, uint32_t data)
1178 {
1179     uint64_t qaddr_base = xive_end_qaddr(end);
1180     uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1181     uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1182     uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1183 
1184     uint64_t qaddr = qaddr_base + (qindex << 2);
1185     uint32_t qdata = cpu_to_be32((qgen << 31) | (data & 0x7fffffff));
1186     uint32_t qentries = 1 << (qsize + 10);
1187 
1188     if (dma_memory_write(&address_space_memory, qaddr, &qdata, sizeof(qdata))) {
1189         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to write END data @0x%"
1190                       HWADDR_PRIx "\n", qaddr);
1191         return;
1192     }
1193 
1194     qindex = (qindex + 1) & (qentries - 1);
1195     if (qindex == 0) {
1196         qgen ^= 1;
1197         end->w1 = xive_set_field32(END_W1_GENERATION, end->w1, qgen);
1198     }
1199     end->w1 = xive_set_field32(END_W1_PAGE_OFF, end->w1, qindex);
1200 }
1201 
1202 /*
1203  * XIVE Router (aka. Virtualization Controller or IVRE)
1204  */
1205 
1206 int xive_router_get_eas(XiveRouter *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1207                         XiveEAS *eas)
1208 {
1209     XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1210 
1211     return xrc->get_eas(xrtr, eas_blk, eas_idx, eas);
1212 }
1213 
1214 int xive_router_get_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1215                         XiveEND *end)
1216 {
1217    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1218 
1219    return xrc->get_end(xrtr, end_blk, end_idx, end);
1220 }
1221 
1222 int xive_router_write_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1223                           XiveEND *end, uint8_t word_number)
1224 {
1225    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1226 
1227    return xrc->write_end(xrtr, end_blk, end_idx, end, word_number);
1228 }
1229 
1230 int xive_router_get_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1231                         XiveNVT *nvt)
1232 {
1233    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1234 
1235    return xrc->get_nvt(xrtr, nvt_blk, nvt_idx, nvt);
1236 }
1237 
1238 int xive_router_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1239                         XiveNVT *nvt, uint8_t word_number)
1240 {
1241    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1242 
1243    return xrc->write_nvt(xrtr, nvt_blk, nvt_idx, nvt, word_number);
1244 }
1245 
1246 XiveTCTX *xive_router_get_tctx(XiveRouter *xrtr, CPUState *cs)
1247 {
1248     XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1249 
1250     return xrc->get_tctx(xrtr, cs);
1251 }
1252 
1253 /*
1254  * Encode the HW CAM line in the block group mode format :
1255  *
1256  *   chip << 19 | 0000000 0 0001 thread (7Bit)
1257  */
1258 static uint32_t xive_tctx_hw_cam_line(XiveTCTX *tctx)
1259 {
1260     CPUPPCState *env = &POWERPC_CPU(tctx->cs)->env;
1261     uint32_t pir = env->spr_cb[SPR_PIR].default_value;
1262 
1263     return xive_nvt_cam_line((pir >> 8) & 0xf, 1 << 7 | (pir & 0x7f));
1264 }
1265 
1266 /*
1267  * The thread context register words are in big-endian format.
1268  */
1269 static int xive_presenter_tctx_match(XiveTCTX *tctx, uint8_t format,
1270                                      uint8_t nvt_blk, uint32_t nvt_idx,
1271                                      bool cam_ignore, uint32_t logic_serv)
1272 {
1273     uint32_t cam = xive_nvt_cam_line(nvt_blk, nvt_idx);
1274     uint32_t qw3w2 = xive_tctx_word2(&tctx->regs[TM_QW3_HV_PHYS]);
1275     uint32_t qw2w2 = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
1276     uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
1277     uint32_t qw0w2 = xive_tctx_word2(&tctx->regs[TM_QW0_USER]);
1278 
1279     /*
1280      * TODO (PowerNV): ignore mode. The low order bits of the NVT
1281      * identifier are ignored in the "CAM" match.
1282      */
1283 
1284     if (format == 0) {
1285         if (cam_ignore == true) {
1286             /*
1287              * F=0 & i=1: Logical server notification (bits ignored at
1288              * the end of the NVT identifier)
1289              */
1290             qemu_log_mask(LOG_UNIMP, "XIVE: no support for LS NVT %x/%x\n",
1291                           nvt_blk, nvt_idx);
1292              return -1;
1293         }
1294 
1295         /* F=0 & i=0: Specific NVT notification */
1296 
1297         /* PHYS ring */
1298         if ((be32_to_cpu(qw3w2) & TM_QW3W2_VT) &&
1299             cam == xive_tctx_hw_cam_line(tctx)) {
1300             return TM_QW3_HV_PHYS;
1301         }
1302 
1303         /* HV POOL ring */
1304         if ((be32_to_cpu(qw2w2) & TM_QW2W2_VP) &&
1305             cam == xive_get_field32(TM_QW2W2_POOL_CAM, qw2w2)) {
1306             return TM_QW2_HV_POOL;
1307         }
1308 
1309         /* OS ring */
1310         if ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1311             cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) {
1312             return TM_QW1_OS;
1313         }
1314     } else {
1315         /* F=1 : User level Event-Based Branch (EBB) notification */
1316 
1317         /* USER ring */
1318         if  ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1319              (cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) &&
1320              (be32_to_cpu(qw0w2) & TM_QW0W2_VU) &&
1321              (logic_serv == xive_get_field32(TM_QW0W2_LOGIC_SERV, qw0w2))) {
1322             return TM_QW0_USER;
1323         }
1324     }
1325     return -1;
1326 }
1327 
1328 typedef struct XiveTCTXMatch {
1329     XiveTCTX *tctx;
1330     uint8_t ring;
1331 } XiveTCTXMatch;
1332 
1333 static bool xive_presenter_match(XiveRouter *xrtr, uint8_t format,
1334                                  uint8_t nvt_blk, uint32_t nvt_idx,
1335                                  bool cam_ignore, uint8_t priority,
1336                                  uint32_t logic_serv, XiveTCTXMatch *match)
1337 {
1338     CPUState *cs;
1339 
1340     /*
1341      * TODO (PowerNV): handle chip_id overwrite of block field for
1342      * hardwired CAM compares
1343      */
1344 
1345     CPU_FOREACH(cs) {
1346         XiveTCTX *tctx = xive_router_get_tctx(xrtr, cs);
1347         int ring;
1348 
1349         /*
1350          * HW checks that the CPU is enabled in the Physical Thread
1351          * Enable Register (PTER).
1352          */
1353 
1354         /*
1355          * Check the thread context CAM lines and record matches. We
1356          * will handle CPU exception delivery later
1357          */
1358         ring = xive_presenter_tctx_match(tctx, format, nvt_blk, nvt_idx,
1359                                          cam_ignore, logic_serv);
1360         /*
1361          * Save the context and follow on to catch duplicates, that we
1362          * don't support yet.
1363          */
1364         if (ring != -1) {
1365             if (match->tctx) {
1366                 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: already found a thread "
1367                               "context NVT %x/%x\n", nvt_blk, nvt_idx);
1368                 return false;
1369             }
1370 
1371             match->ring = ring;
1372             match->tctx = tctx;
1373         }
1374     }
1375 
1376     if (!match->tctx) {
1377         qemu_log_mask(LOG_UNIMP, "XIVE: NVT %x/%x is not dispatched\n",
1378                       nvt_blk, nvt_idx);
1379         return false;
1380     }
1381 
1382     return true;
1383 }
1384 
1385 /*
1386  * This is our simple Xive Presenter Engine model. It is merged in the
1387  * Router as it does not require an extra object.
1388  *
1389  * It receives notification requests sent by the IVRE to find one
1390  * matching NVT (or more) dispatched on the processor threads. In case
1391  * of a single NVT notification, the process is abreviated and the
1392  * thread is signaled if a match is found. In case of a logical server
1393  * notification (bits ignored at the end of the NVT identifier), the
1394  * IVPE and IVRE select a winning thread using different filters. This
1395  * involves 2 or 3 exchanges on the PowerBus that the model does not
1396  * support.
1397  *
1398  * The parameters represent what is sent on the PowerBus
1399  */
1400 static void xive_presenter_notify(XiveRouter *xrtr, uint8_t format,
1401                                   uint8_t nvt_blk, uint32_t nvt_idx,
1402                                   bool cam_ignore, uint8_t priority,
1403                                   uint32_t logic_serv)
1404 {
1405     XiveNVT nvt;
1406     XiveTCTXMatch match = { .tctx = NULL, .ring = 0 };
1407     bool found;
1408 
1409     /* NVT cache lookup */
1410     if (xive_router_get_nvt(xrtr, nvt_blk, nvt_idx, &nvt)) {
1411         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no NVT %x/%x\n",
1412                       nvt_blk, nvt_idx);
1413         return;
1414     }
1415 
1416     if (!xive_nvt_is_valid(&nvt)) {
1417         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is invalid\n",
1418                       nvt_blk, nvt_idx);
1419         return;
1420     }
1421 
1422     found = xive_presenter_match(xrtr, format, nvt_blk, nvt_idx, cam_ignore,
1423                                  priority, logic_serv, &match);
1424     if (found) {
1425         ipb_update(&match.tctx->regs[match.ring], priority);
1426         xive_tctx_notify(match.tctx, match.ring);
1427         return;
1428     }
1429 
1430     /* Record the IPB in the associated NVT structure */
1431     ipb_update((uint8_t *) &nvt.w4, priority);
1432     xive_router_write_nvt(xrtr, nvt_blk, nvt_idx, &nvt, 4);
1433 
1434     /*
1435      * If no matching NVT is dispatched on a HW thread :
1436      * - update the NVT structure if backlog is activated
1437      * - escalate (ESe PQ bits and EAS in w4-5) if escalation is
1438      *   activated
1439      */
1440 }
1441 
1442 /*
1443  * An END trigger can come from an event trigger (IPI or HW) or from
1444  * another chip. We don't model the PowerBus but the END trigger
1445  * message has the same parameters than in the function below.
1446  */
1447 static void xive_router_end_notify(XiveRouter *xrtr, uint8_t end_blk,
1448                                    uint32_t end_idx, uint32_t end_data)
1449 {
1450     XiveEND end;
1451     uint8_t priority;
1452     uint8_t format;
1453 
1454     /* END cache lookup */
1455     if (xive_router_get_end(xrtr, end_blk, end_idx, &end)) {
1456         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1457                       end_idx);
1458         return;
1459     }
1460 
1461     if (!xive_end_is_valid(&end)) {
1462         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1463                       end_blk, end_idx);
1464         return;
1465     }
1466 
1467     if (xive_end_is_enqueue(&end)) {
1468         xive_end_enqueue(&end, end_data);
1469         /* Enqueuing event data modifies the EQ toggle and index */
1470         xive_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1471     }
1472 
1473     /*
1474      * The W7 format depends on the F bit in W6. It defines the type
1475      * of the notification :
1476      *
1477      *   F=0 : single or multiple NVT notification
1478      *   F=1 : User level Event-Based Branch (EBB) notification, no
1479      *         priority
1480      */
1481     format = xive_get_field32(END_W6_FORMAT_BIT, end.w6);
1482     priority = xive_get_field32(END_W7_F0_PRIORITY, end.w7);
1483 
1484     /* The END is masked */
1485     if (format == 0 && priority == 0xff) {
1486         return;
1487     }
1488 
1489     /*
1490      * Check the END ESn (Event State Buffer for notification) for
1491      * even futher coalescing in the Router
1492      */
1493     if (!xive_end_is_notify(&end)) {
1494         uint8_t pq = xive_get_field32(END_W1_ESn, end.w1);
1495         bool notify = xive_esb_trigger(&pq);
1496 
1497         if (pq != xive_get_field32(END_W1_ESn, end.w1)) {
1498             end.w1 = xive_set_field32(END_W1_ESn, end.w1, pq);
1499             xive_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1500         }
1501 
1502         /* ESn[Q]=1 : end of notification */
1503         if (!notify) {
1504             return;
1505         }
1506     }
1507 
1508     /*
1509      * Follows IVPE notification
1510      */
1511     xive_presenter_notify(xrtr, format,
1512                           xive_get_field32(END_W6_NVT_BLOCK, end.w6),
1513                           xive_get_field32(END_W6_NVT_INDEX, end.w6),
1514                           xive_get_field32(END_W7_F0_IGNORE, end.w7),
1515                           priority,
1516                           xive_get_field32(END_W7_F1_LOG_SERVER_ID, end.w7));
1517 
1518     /* TODO: Auto EOI. */
1519 }
1520 
1521 void xive_router_notify(XiveNotifier *xn, uint32_t lisn)
1522 {
1523     XiveRouter *xrtr = XIVE_ROUTER(xn);
1524     uint8_t eas_blk = XIVE_SRCNO_BLOCK(lisn);
1525     uint32_t eas_idx = XIVE_SRCNO_INDEX(lisn);
1526     XiveEAS eas;
1527 
1528     /* EAS cache lookup */
1529     if (xive_router_get_eas(xrtr, eas_blk, eas_idx, &eas)) {
1530         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN %x\n", lisn);
1531         return;
1532     }
1533 
1534     /*
1535      * The IVRE checks the State Bit Cache at this point. We skip the
1536      * SBC lookup because the state bits of the sources are modeled
1537      * internally in QEMU.
1538      */
1539 
1540     if (!xive_eas_is_valid(&eas)) {
1541         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid LISN %x\n", lisn);
1542         return;
1543     }
1544 
1545     if (xive_eas_is_masked(&eas)) {
1546         /* Notification completed */
1547         return;
1548     }
1549 
1550     /*
1551      * The event trigger becomes an END trigger
1552      */
1553     xive_router_end_notify(xrtr,
1554                            xive_get_field64(EAS_END_BLOCK, eas.w),
1555                            xive_get_field64(EAS_END_INDEX, eas.w),
1556                            xive_get_field64(EAS_END_DATA,  eas.w));
1557 }
1558 
1559 static void xive_router_class_init(ObjectClass *klass, void *data)
1560 {
1561     DeviceClass *dc = DEVICE_CLASS(klass);
1562     XiveNotifierClass *xnc = XIVE_NOTIFIER_CLASS(klass);
1563 
1564     dc->desc    = "XIVE Router Engine";
1565     xnc->notify = xive_router_notify;
1566 }
1567 
1568 static const TypeInfo xive_router_info = {
1569     .name          = TYPE_XIVE_ROUTER,
1570     .parent        = TYPE_SYS_BUS_DEVICE,
1571     .abstract      = true,
1572     .class_size    = sizeof(XiveRouterClass),
1573     .class_init    = xive_router_class_init,
1574     .interfaces    = (InterfaceInfo[]) {
1575         { TYPE_XIVE_NOTIFIER },
1576         { }
1577     }
1578 };
1579 
1580 void xive_eas_pic_print_info(XiveEAS *eas, uint32_t lisn, Monitor *mon)
1581 {
1582     if (!xive_eas_is_valid(eas)) {
1583         return;
1584     }
1585 
1586     monitor_printf(mon, "  %08x %s end:%02x/%04x data:%08x\n",
1587                    lisn, xive_eas_is_masked(eas) ? "M" : " ",
1588                    (uint8_t)  xive_get_field64(EAS_END_BLOCK, eas->w),
1589                    (uint32_t) xive_get_field64(EAS_END_INDEX, eas->w),
1590                    (uint32_t) xive_get_field64(EAS_END_DATA, eas->w));
1591 }
1592 
1593 /*
1594  * END ESB MMIO loads
1595  */
1596 static uint64_t xive_end_source_read(void *opaque, hwaddr addr, unsigned size)
1597 {
1598     XiveENDSource *xsrc = XIVE_END_SOURCE(opaque);
1599     uint32_t offset = addr & 0xFFF;
1600     uint8_t end_blk;
1601     uint32_t end_idx;
1602     XiveEND end;
1603     uint32_t end_esmask;
1604     uint8_t pq;
1605     uint64_t ret = -1;
1606 
1607     end_blk = xsrc->block_id;
1608     end_idx = addr >> (xsrc->esb_shift + 1);
1609 
1610     if (xive_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
1611         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1612                       end_idx);
1613         return -1;
1614     }
1615 
1616     if (!xive_end_is_valid(&end)) {
1617         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1618                       end_blk, end_idx);
1619         return -1;
1620     }
1621 
1622     end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END_W1_ESn : END_W1_ESe;
1623     pq = xive_get_field32(end_esmask, end.w1);
1624 
1625     switch (offset) {
1626     case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
1627         ret = xive_esb_eoi(&pq);
1628 
1629         /* Forward the source event notification for routing ?? */
1630         break;
1631 
1632     case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
1633         ret = pq;
1634         break;
1635 
1636     case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1637     case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1638     case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1639     case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1640         ret = xive_esb_set(&pq, (offset >> 8) & 0x3);
1641         break;
1642     default:
1643         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB load addr %d\n",
1644                       offset);
1645         return -1;
1646     }
1647 
1648     if (pq != xive_get_field32(end_esmask, end.w1)) {
1649         end.w1 = xive_set_field32(end_esmask, end.w1, pq);
1650         xive_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
1651     }
1652 
1653     return ret;
1654 }
1655 
1656 /*
1657  * END ESB MMIO stores are invalid
1658  */
1659 static void xive_end_source_write(void *opaque, hwaddr addr,
1660                                   uint64_t value, unsigned size)
1661 {
1662     qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr 0x%"
1663                   HWADDR_PRIx"\n", addr);
1664 }
1665 
1666 static const MemoryRegionOps xive_end_source_ops = {
1667     .read = xive_end_source_read,
1668     .write = xive_end_source_write,
1669     .endianness = DEVICE_BIG_ENDIAN,
1670     .valid = {
1671         .min_access_size = 8,
1672         .max_access_size = 8,
1673     },
1674     .impl = {
1675         .min_access_size = 8,
1676         .max_access_size = 8,
1677     },
1678 };
1679 
1680 static void xive_end_source_realize(DeviceState *dev, Error **errp)
1681 {
1682     XiveENDSource *xsrc = XIVE_END_SOURCE(dev);
1683     Object *obj;
1684     Error *local_err = NULL;
1685 
1686     obj = object_property_get_link(OBJECT(dev), "xive", &local_err);
1687     if (!obj) {
1688         error_propagate(errp, local_err);
1689         error_prepend(errp, "required link 'xive' not found: ");
1690         return;
1691     }
1692 
1693     xsrc->xrtr = XIVE_ROUTER(obj);
1694 
1695     if (!xsrc->nr_ends) {
1696         error_setg(errp, "Number of interrupt needs to be greater than 0");
1697         return;
1698     }
1699 
1700     if (xsrc->esb_shift != XIVE_ESB_4K &&
1701         xsrc->esb_shift != XIVE_ESB_64K) {
1702         error_setg(errp, "Invalid ESB shift setting");
1703         return;
1704     }
1705 
1706     /*
1707      * Each END is assigned an even/odd pair of MMIO pages, the even page
1708      * manages the ESn field while the odd page manages the ESe field.
1709      */
1710     memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
1711                           &xive_end_source_ops, xsrc, "xive.end",
1712                           (1ull << (xsrc->esb_shift + 1)) * xsrc->nr_ends);
1713 }
1714 
1715 static Property xive_end_source_properties[] = {
1716     DEFINE_PROP_UINT8("block-id", XiveENDSource, block_id, 0),
1717     DEFINE_PROP_UINT32("nr-ends", XiveENDSource, nr_ends, 0),
1718     DEFINE_PROP_UINT32("shift", XiveENDSource, esb_shift, XIVE_ESB_64K),
1719     DEFINE_PROP_END_OF_LIST(),
1720 };
1721 
1722 static void xive_end_source_class_init(ObjectClass *klass, void *data)
1723 {
1724     DeviceClass *dc = DEVICE_CLASS(klass);
1725 
1726     dc->desc    = "XIVE END Source";
1727     dc->props   = xive_end_source_properties;
1728     dc->realize = xive_end_source_realize;
1729 }
1730 
1731 static const TypeInfo xive_end_source_info = {
1732     .name          = TYPE_XIVE_END_SOURCE,
1733     .parent        = TYPE_DEVICE,
1734     .instance_size = sizeof(XiveENDSource),
1735     .class_init    = xive_end_source_class_init,
1736 };
1737 
1738 /*
1739  * XIVE Notifier
1740  */
1741 static const TypeInfo xive_notifier_info = {
1742     .name = TYPE_XIVE_NOTIFIER,
1743     .parent = TYPE_INTERFACE,
1744     .class_size = sizeof(XiveNotifierClass),
1745 };
1746 
1747 static void xive_register_types(void)
1748 {
1749     type_register_static(&xive_source_info);
1750     type_register_static(&xive_notifier_info);
1751     type_register_static(&xive_router_info);
1752     type_register_static(&xive_end_source_info);
1753     type_register_static(&xive_tctx_info);
1754 }
1755 
1756 type_init(xive_register_types)
1757