xref: /openbmc/qemu/hw/intc/pnv_xive.c (revision 4f7b1ecb)
1 /*
2  * QEMU PowerPC XIVE interrupt controller model
3  *
4  * Copyright (c) 2017-2019, 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/ppc/fdt.h"
19 #include "hw/ppc/pnv.h"
20 #include "hw/ppc/pnv_chip.h"
21 #include "hw/ppc/pnv_core.h"
22 #include "hw/ppc/pnv_xscom.h"
23 #include "hw/ppc/pnv_xive.h"
24 #include "hw/ppc/xive_regs.h"
25 #include "hw/qdev-properties.h"
26 #include "hw/ppc/ppc.h"
27 #include "trace.h"
28 
29 #include <libfdt.h>
30 
31 #include "pnv_xive_regs.h"
32 
33 #undef XIVE_DEBUG
34 
35 /*
36  * Virtual structures table (VST)
37  */
38 #define SBE_PER_BYTE   4
39 
40 typedef struct XiveVstInfo {
41     const char *name;
42     uint32_t    size;
43     uint32_t    max_blocks;
44 } XiveVstInfo;
45 
46 static const XiveVstInfo vst_infos[] = {
47     [VST_TSEL_IVT]  = { "EAT",  sizeof(XiveEAS), 16 },
48     [VST_TSEL_SBE]  = { "SBE",  1,               16 },
49     [VST_TSEL_EQDT] = { "ENDT", sizeof(XiveEND), 16 },
50     [VST_TSEL_VPDT] = { "VPDT", sizeof(XiveNVT), 32 },
51 
52     /*
53      *  Interrupt fifo backing store table (not modeled) :
54      *
55      * 0 - IPI,
56      * 1 - HWD,
57      * 2 - First escalate,
58      * 3 - Second escalate,
59      * 4 - Redistribution,
60      * 5 - IPI cascaded queue ?
61      */
62     [VST_TSEL_IRQ]  = { "IRQ",  1,               6  },
63 };
64 
65 #define xive_error(xive, fmt, ...)                                      \
66     qemu_log_mask(LOG_GUEST_ERROR, "XIVE[%x] - " fmt "\n",              \
67                   (xive)->chip->chip_id, ## __VA_ARGS__);
68 
69 /*
70  * When PC_TCTXT_CHIPID_OVERRIDE is configured, the PC_TCTXT_CHIPID
71  * field overrides the hardwired chip ID in the Powerbus operations
72  * and for CAM compares
73  */
74 static uint8_t pnv_xive_block_id(PnvXive *xive)
75 {
76     uint8_t blk = xive->chip->chip_id;
77     uint64_t cfg_val = xive->regs[PC_TCTXT_CFG >> 3];
78 
79     if (cfg_val & PC_TCTXT_CHIPID_OVERRIDE) {
80         blk = GETFIELD(PC_TCTXT_CHIPID, cfg_val);
81     }
82 
83     return blk;
84 }
85 
86 /*
87  * VST accessors for SBE, EAT, ENDT, NVT
88  *
89  * Indirect VST tables are arrays of VSDs pointing to a page (of same
90  * size). Each page is a direct VST table.
91  */
92 
93 #define XIVE_VSD_SIZE 8
94 
95 /* Indirect page size can be 4K, 64K, 2M, 16M. */
96 static uint64_t pnv_xive_vst_page_size_allowed(uint32_t page_shift)
97 {
98      return page_shift == 12 || page_shift == 16 ||
99          page_shift == 21 || page_shift == 24;
100 }
101 
102 static uint64_t pnv_xive_vst_addr_direct(PnvXive *xive, uint32_t type,
103                                          uint64_t vsd, uint32_t idx)
104 {
105     const XiveVstInfo *info = &vst_infos[type];
106     uint64_t vst_addr = vsd & VSD_ADDRESS_MASK;
107     uint64_t vst_tsize = 1ull << (GETFIELD(VSD_TSIZE, vsd) + 12);
108     uint32_t idx_max;
109 
110     idx_max = vst_tsize / info->size - 1;
111     if (idx > idx_max) {
112 #ifdef XIVE_DEBUG
113         xive_error(xive, "VST: %s entry %x out of range [ 0 .. %x ] !?",
114                    info->name, idx, idx_max);
115 #endif
116         return 0;
117     }
118 
119     return vst_addr + idx * info->size;
120 }
121 
122 static uint64_t pnv_xive_vst_addr_indirect(PnvXive *xive, uint32_t type,
123                                            uint64_t vsd, uint32_t idx)
124 {
125     const XiveVstInfo *info = &vst_infos[type];
126     uint64_t vsd_addr;
127     uint32_t vsd_idx;
128     uint32_t page_shift;
129     uint32_t vst_per_page;
130 
131     /* Get the page size of the indirect table. */
132     vsd_addr = vsd & VSD_ADDRESS_MASK;
133     if (ldq_be_dma(&address_space_memory, vsd_addr, &vsd,
134                     MEMTXATTRS_UNSPECIFIED)) {
135         xive_error(xive, "VST: failed to access %s entry %x @0x%" PRIx64,
136                    info->name, idx, vsd_addr);
137         return 0;
138     }
139 
140     if (!(vsd & VSD_ADDRESS_MASK)) {
141 #ifdef XIVE_DEBUG
142         xive_error(xive, "VST: invalid %s entry %x !?", info->name, idx);
143 #endif
144         return 0;
145     }
146 
147     page_shift = GETFIELD(VSD_TSIZE, vsd) + 12;
148 
149     if (!pnv_xive_vst_page_size_allowed(page_shift)) {
150         xive_error(xive, "VST: invalid %s page shift %d", info->name,
151                    page_shift);
152         return 0;
153     }
154 
155     vst_per_page = (1ull << page_shift) / info->size;
156     vsd_idx = idx / vst_per_page;
157 
158     /* Load the VSD we are looking for, if not already done */
159     if (vsd_idx) {
160         vsd_addr = vsd_addr + vsd_idx * XIVE_VSD_SIZE;
161         if (ldq_be_dma(&address_space_memory, vsd_addr, &vsd,
162                        MEMTXATTRS_UNSPECIFIED)) {
163             xive_error(xive, "VST: failed to access %s entry %x @0x%"
164                        PRIx64, info->name, vsd_idx, vsd_addr);
165             return 0;
166         }
167 
168         if (!(vsd & VSD_ADDRESS_MASK)) {
169 #ifdef XIVE_DEBUG
170             xive_error(xive, "VST: invalid %s entry %x !?", info->name, idx);
171 #endif
172             return 0;
173         }
174 
175         /*
176          * Check that the pages have a consistent size across the
177          * indirect table
178          */
179         if (page_shift != GETFIELD(VSD_TSIZE, vsd) + 12) {
180             xive_error(xive, "VST: %s entry %x indirect page size differ !?",
181                        info->name, idx);
182             return 0;
183         }
184     }
185 
186     return pnv_xive_vst_addr_direct(xive, type, vsd, (idx % vst_per_page));
187 }
188 
189 /*
190  * This is a simplified model of operation forwarding on a remote IC.
191  *
192  * A PC MMIO address is built to identify the NVT structure. The load
193  * on the remote IC will return the address of the structure in RAM,
194  * which will then be used by pnv_xive_vst_write/read to perform the
195  * RAM operation.
196  */
197 static uint64_t pnv_xive_vst_addr_remote(PnvXive *xive, uint32_t type,
198                                          uint64_t vsd, uint8_t blk,
199                                          uint32_t idx)
200 {
201     const XiveVstInfo *info = &vst_infos[type];
202     uint64_t remote_addr = vsd & VSD_ADDRESS_MASK;
203     uint64_t vst_addr;
204     MemTxResult result;
205 
206     if (type != VST_TSEL_VPDT) {
207         xive_error(xive, "VST: invalid access on remote VST %s %x/%x !?",
208                    info->name, blk, idx);
209         return 0;
210     }
211 
212     remote_addr |= ((uint64_t)idx) << xive->pc_shift;
213 
214     vst_addr = address_space_ldq_be(&address_space_memory, remote_addr,
215                                     MEMTXATTRS_UNSPECIFIED, &result);
216     if (result != MEMTX_OK) {
217         xive_error(xive, "VST: read failed at @0x%"  HWADDR_PRIx
218                    " for NVT %x/%x\n", remote_addr, blk, idx);
219         return 0;
220     }
221 
222     return vst_addr;
223 }
224 
225 static uint64_t pnv_xive_vst_addr(PnvXive *xive, uint32_t type, uint8_t blk,
226                                   uint32_t idx)
227 {
228     const XiveVstInfo *info = &vst_infos[type];
229     uint64_t vsd;
230 
231     if (blk >= info->max_blocks) {
232         xive_error(xive, "VST: invalid block id %d for VST %s %d !?",
233                    blk, info->name, idx);
234         return 0;
235     }
236 
237     vsd = xive->vsds[type][blk];
238 
239     /* Remote VST access */
240     if (GETFIELD(VSD_MODE, vsd) == VSD_MODE_FORWARD) {
241         return pnv_xive_vst_addr_remote(xive, type, vsd, blk, idx);
242     }
243 
244     if (VSD_INDIRECT & vsd) {
245         return pnv_xive_vst_addr_indirect(xive, type, vsd, idx);
246     }
247 
248     return pnv_xive_vst_addr_direct(xive, type, vsd, idx);
249 }
250 
251 static int pnv_xive_vst_read(PnvXive *xive, uint32_t type, uint8_t blk,
252                              uint32_t idx, void *data)
253 {
254     const XiveVstInfo *info = &vst_infos[type];
255     uint64_t addr = pnv_xive_vst_addr(xive, type, blk, idx);
256     MemTxResult result;
257 
258     if (!addr) {
259         return -1;
260     }
261 
262     result = address_space_read(&address_space_memory, addr,
263                                 MEMTXATTRS_UNSPECIFIED, data,
264                                 info->size);
265     if (result != MEMTX_OK) {
266         xive_error(xive, "VST: read failed at @0x%" HWADDR_PRIx
267                    " for VST %s %x/%x\n", addr, info->name, blk, idx);
268         return -1;
269     }
270     return 0;
271 }
272 
273 #define XIVE_VST_WORD_ALL -1
274 
275 static int pnv_xive_vst_write(PnvXive *xive, uint32_t type, uint8_t blk,
276                               uint32_t idx, void *data, uint32_t word_number)
277 {
278     const XiveVstInfo *info = &vst_infos[type];
279     uint64_t addr = pnv_xive_vst_addr(xive, type, blk, idx);
280     MemTxResult result;
281 
282     if (!addr) {
283         return -1;
284     }
285 
286     if (word_number == XIVE_VST_WORD_ALL) {
287         result = address_space_write(&address_space_memory, addr,
288                                      MEMTXATTRS_UNSPECIFIED, data,
289                                      info->size);
290     } else {
291         result = address_space_write(&address_space_memory,
292                                      addr + word_number * 4,
293                                      MEMTXATTRS_UNSPECIFIED,
294                                      data + word_number * 4, 4);
295     }
296 
297     if (result != MEMTX_OK) {
298         xive_error(xive, "VST: write failed at @0x%" HWADDR_PRIx
299                     "for VST %s %x/%x\n", addr, info->name, blk, idx);
300         return -1;
301     }
302     return 0;
303 }
304 
305 static int pnv_xive_get_end(XiveRouter *xrtr, uint8_t blk, uint32_t idx,
306                             XiveEND *end)
307 {
308     PnvXive *xive = PNV_XIVE(xrtr);
309 
310     if (pnv_xive_block_id(xive) != blk) {
311         xive_error(xive, "VST: END %x/%x is remote !?", blk, idx);
312         return -1;
313     }
314 
315     return pnv_xive_vst_read(PNV_XIVE(xrtr), VST_TSEL_EQDT, blk, idx, end);
316 }
317 
318 static int pnv_xive_write_end(XiveRouter *xrtr, uint8_t blk, uint32_t idx,
319                               XiveEND *end, uint8_t word_number)
320 {
321     PnvXive *xive = PNV_XIVE(xrtr);
322 
323     if (pnv_xive_block_id(xive) != blk) {
324         xive_error(xive, "VST: END %x/%x is remote !?", blk, idx);
325         return -1;
326     }
327 
328     return pnv_xive_vst_write(PNV_XIVE(xrtr), VST_TSEL_EQDT, blk, idx, end,
329                               word_number);
330 }
331 
332 static int pnv_xive_end_update(PnvXive *xive)
333 {
334     uint8_t  blk = GETFIELD(VC_EQC_CWATCH_BLOCKID,
335                            xive->regs[(VC_EQC_CWATCH_SPEC >> 3)]);
336     uint32_t idx = GETFIELD(VC_EQC_CWATCH_OFFSET,
337                            xive->regs[(VC_EQC_CWATCH_SPEC >> 3)]);
338     int i;
339     uint64_t eqc_watch[4];
340 
341     for (i = 0; i < ARRAY_SIZE(eqc_watch); i++) {
342         eqc_watch[i] = cpu_to_be64(xive->regs[(VC_EQC_CWATCH_DAT0 >> 3) + i]);
343     }
344 
345     return pnv_xive_vst_write(xive, VST_TSEL_EQDT, blk, idx, eqc_watch,
346                               XIVE_VST_WORD_ALL);
347 }
348 
349 static void pnv_xive_end_cache_load(PnvXive *xive)
350 {
351     uint8_t  blk = GETFIELD(VC_EQC_CWATCH_BLOCKID,
352                            xive->regs[(VC_EQC_CWATCH_SPEC >> 3)]);
353     uint32_t idx = GETFIELD(VC_EQC_CWATCH_OFFSET,
354                            xive->regs[(VC_EQC_CWATCH_SPEC >> 3)]);
355     uint64_t eqc_watch[4] = { 0 };
356     int i;
357 
358     if (pnv_xive_vst_read(xive, VST_TSEL_EQDT, blk, idx, eqc_watch)) {
359         xive_error(xive, "VST: no END entry %x/%x !?", blk, idx);
360     }
361 
362     for (i = 0; i < ARRAY_SIZE(eqc_watch); i++) {
363         xive->regs[(VC_EQC_CWATCH_DAT0 >> 3) + i] = be64_to_cpu(eqc_watch[i]);
364     }
365 }
366 
367 static int pnv_xive_get_nvt(XiveRouter *xrtr, uint8_t blk, uint32_t idx,
368                             XiveNVT *nvt)
369 {
370     return pnv_xive_vst_read(PNV_XIVE(xrtr), VST_TSEL_VPDT, blk, idx, nvt);
371 }
372 
373 static int pnv_xive_write_nvt(XiveRouter *xrtr, uint8_t blk, uint32_t idx,
374                               XiveNVT *nvt, uint8_t word_number)
375 {
376     return pnv_xive_vst_write(PNV_XIVE(xrtr), VST_TSEL_VPDT, blk, idx, nvt,
377                               word_number);
378 }
379 
380 static int pnv_xive_nvt_update(PnvXive *xive)
381 {
382     uint8_t  blk = GETFIELD(PC_VPC_CWATCH_BLOCKID,
383                            xive->regs[(PC_VPC_CWATCH_SPEC >> 3)]);
384     uint32_t idx = GETFIELD(PC_VPC_CWATCH_OFFSET,
385                            xive->regs[(PC_VPC_CWATCH_SPEC >> 3)]);
386     int i;
387     uint64_t vpc_watch[8];
388 
389     for (i = 0; i < ARRAY_SIZE(vpc_watch); i++) {
390         vpc_watch[i] = cpu_to_be64(xive->regs[(PC_VPC_CWATCH_DAT0 >> 3) + i]);
391     }
392 
393     return pnv_xive_vst_write(xive, VST_TSEL_VPDT, blk, idx, vpc_watch,
394                               XIVE_VST_WORD_ALL);
395 }
396 
397 static void pnv_xive_nvt_cache_load(PnvXive *xive)
398 {
399     uint8_t  blk = GETFIELD(PC_VPC_CWATCH_BLOCKID,
400                            xive->regs[(PC_VPC_CWATCH_SPEC >> 3)]);
401     uint32_t idx = GETFIELD(PC_VPC_CWATCH_OFFSET,
402                            xive->regs[(PC_VPC_CWATCH_SPEC >> 3)]);
403     uint64_t vpc_watch[8] = { 0 };
404     int i;
405 
406     if (pnv_xive_vst_read(xive, VST_TSEL_VPDT, blk, idx, vpc_watch)) {
407         xive_error(xive, "VST: no NVT entry %x/%x !?", blk, idx);
408     }
409 
410     for (i = 0; i < ARRAY_SIZE(vpc_watch); i++) {
411         xive->regs[(PC_VPC_CWATCH_DAT0 >> 3) + i] = be64_to_cpu(vpc_watch[i]);
412     }
413 }
414 
415 static int pnv_xive_get_eas(XiveRouter *xrtr, uint8_t blk, uint32_t idx,
416                             XiveEAS *eas)
417 {
418     PnvXive *xive = PNV_XIVE(xrtr);
419 
420     /*
421      * EAT lookups should be local to the IC
422      */
423     if (pnv_xive_block_id(xive) != blk) {
424         xive_error(xive, "VST: EAS %x is remote !?", XIVE_EAS(blk, idx));
425         return -1;
426     }
427 
428     return pnv_xive_vst_read(xive, VST_TSEL_IVT, blk, idx, eas);
429 }
430 
431 static int pnv_xive_get_pq(XiveRouter *xrtr, uint8_t blk, uint32_t idx,
432                            uint8_t *pq)
433 {
434     PnvXive *xive = PNV_XIVE(xrtr);
435 
436     if (pnv_xive_block_id(xive) != blk) {
437         xive_error(xive, "VST: EAS %x is remote !?", XIVE_EAS(blk, idx));
438         return -1;
439     }
440 
441     *pq = xive_source_esb_get(&xive->ipi_source, idx);
442     return 0;
443 }
444 
445 static int pnv_xive_set_pq(XiveRouter *xrtr, uint8_t blk, uint32_t idx,
446                            uint8_t *pq)
447 {
448     PnvXive *xive = PNV_XIVE(xrtr);
449 
450     if (pnv_xive_block_id(xive) != blk) {
451         xive_error(xive, "VST: EAS %x is remote !?", XIVE_EAS(blk, idx));
452         return -1;
453     }
454 
455     *pq = xive_source_esb_set(&xive->ipi_source, idx, *pq);
456     return 0;
457 }
458 
459 /*
460  * One bit per thread id. The first register PC_THREAD_EN_REG0 covers
461  * the first cores 0-15 (normal) of the chip or 0-7 (fused). The
462  * second register covers cores 16-23 (normal) or 8-11 (fused).
463  */
464 static bool pnv_xive_is_cpu_enabled(PnvXive *xive, PowerPCCPU *cpu)
465 {
466     int pir = ppc_cpu_pir(cpu);
467     uint32_t fc = PNV9_PIR2FUSEDCORE(pir);
468     uint64_t reg = fc < 8 ? PC_THREAD_EN_REG0 : PC_THREAD_EN_REG1;
469     uint32_t bit = pir & 0x3f;
470 
471     return xive->regs[reg >> 3] & PPC_BIT(bit);
472 }
473 
474 static int pnv_xive_match_nvt(XivePresenter *xptr, uint8_t format,
475                               uint8_t nvt_blk, uint32_t nvt_idx,
476                               bool cam_ignore, uint8_t priority,
477                               uint32_t logic_serv, XiveTCTXMatch *match)
478 {
479     PnvXive *xive = PNV_XIVE(xptr);
480     PnvChip *chip = xive->chip;
481     int count = 0;
482     int i, j;
483 
484     for (i = 0; i < chip->nr_cores; i++) {
485         PnvCore *pc = chip->cores[i];
486         CPUCore *cc = CPU_CORE(pc);
487 
488         for (j = 0; j < cc->nr_threads; j++) {
489             PowerPCCPU *cpu = pc->threads[j];
490             XiveTCTX *tctx;
491             int ring;
492 
493             if (!pnv_xive_is_cpu_enabled(xive, cpu)) {
494                 continue;
495             }
496 
497             tctx = XIVE_TCTX(pnv_cpu_state(cpu)->intc);
498 
499             /*
500              * Check the thread context CAM lines and record matches.
501              */
502             ring = xive_presenter_tctx_match(xptr, tctx, format, nvt_blk,
503                                              nvt_idx, cam_ignore, logic_serv);
504             /*
505              * Save the context and follow on to catch duplicates, that we
506              * don't support yet.
507              */
508             if (ring != -1) {
509                 if (match->tctx) {
510                     qemu_log_mask(LOG_GUEST_ERROR, "XIVE: already found a "
511                                   "thread context NVT %x/%x\n",
512                                   nvt_blk, nvt_idx);
513                     return -1;
514                 }
515 
516                 match->ring = ring;
517                 match->tctx = tctx;
518                 count++;
519             }
520         }
521     }
522 
523     return count;
524 }
525 
526 static uint32_t pnv_xive_presenter_get_config(XivePresenter *xptr)
527 {
528     uint32_t cfg = 0;
529 
530     /* TIMA GEN1 is all P9 knows */
531     cfg |= XIVE_PRESENTER_GEN1_TIMA_OS;
532 
533     return cfg;
534 }
535 
536 static uint8_t pnv_xive_get_block_id(XiveRouter *xrtr)
537 {
538     return pnv_xive_block_id(PNV_XIVE(xrtr));
539 }
540 
541 /*
542  * The TIMA MMIO space is shared among the chips and to identify the
543  * chip from which the access is being done, we extract the chip id
544  * from the PIR.
545  */
546 static PnvXive *pnv_xive_tm_get_xive(PowerPCCPU *cpu)
547 {
548     int pir = ppc_cpu_pir(cpu);
549     XivePresenter *xptr = XIVE_TCTX(pnv_cpu_state(cpu)->intc)->xptr;
550     PnvXive *xive = PNV_XIVE(xptr);
551 
552     if (!pnv_xive_is_cpu_enabled(xive, cpu)) {
553         xive_error(xive, "IC: CPU %x is not enabled", pir);
554     }
555     return xive;
556 }
557 
558 /*
559  * The internal sources (IPIs) of the interrupt controller have no
560  * knowledge of the XIVE chip on which they reside. Encode the block
561  * id in the source interrupt number before forwarding the source
562  * event notification to the Router. This is required on a multichip
563  * system.
564  */
565 static void pnv_xive_notify(XiveNotifier *xn, uint32_t srcno, bool pq_checked)
566 {
567     PnvXive *xive = PNV_XIVE(xn);
568     uint8_t blk = pnv_xive_block_id(xive);
569 
570     xive_router_notify(xn, XIVE_EAS(blk, srcno), pq_checked);
571 }
572 
573 /*
574  * XIVE helpers
575  */
576 
577 static uint64_t pnv_xive_vc_size(PnvXive *xive)
578 {
579     return (~xive->regs[CQ_VC_BARM >> 3] + 1) & CQ_VC_BARM_MASK;
580 }
581 
582 static uint64_t pnv_xive_edt_shift(PnvXive *xive)
583 {
584     return ctz64(pnv_xive_vc_size(xive) / XIVE_TABLE_EDT_MAX);
585 }
586 
587 static uint64_t pnv_xive_pc_size(PnvXive *xive)
588 {
589     return (~xive->regs[CQ_PC_BARM >> 3] + 1) & CQ_PC_BARM_MASK;
590 }
591 
592 static uint32_t pnv_xive_nr_ipis(PnvXive *xive, uint8_t blk)
593 {
594     uint64_t vsd = xive->vsds[VST_TSEL_SBE][blk];
595     uint64_t vst_tsize = 1ull << (GETFIELD(VSD_TSIZE, vsd) + 12);
596 
597     return VSD_INDIRECT & vsd ? 0 : vst_tsize * SBE_PER_BYTE;
598 }
599 
600 /*
601  * Compute the number of entries per indirect subpage.
602  */
603 static uint64_t pnv_xive_vst_per_subpage(PnvXive *xive, uint32_t type)
604 {
605     uint8_t blk = pnv_xive_block_id(xive);
606     uint64_t vsd = xive->vsds[type][blk];
607     const XiveVstInfo *info = &vst_infos[type];
608     uint64_t vsd_addr;
609     uint32_t page_shift;
610 
611     /* For direct tables, fake a valid value */
612     if (!(VSD_INDIRECT & vsd)) {
613         return 1;
614     }
615 
616     /* Get the page size of the indirect table. */
617     vsd_addr = vsd & VSD_ADDRESS_MASK;
618     if (ldq_be_dma(&address_space_memory, vsd_addr, &vsd,
619                    MEMTXATTRS_UNSPECIFIED)) {
620         xive_error(xive, "VST: failed to access %s entry @0x%" PRIx64,
621                    info->name, vsd_addr);
622         return 0;
623     }
624 
625     if (!(vsd & VSD_ADDRESS_MASK)) {
626 #ifdef XIVE_DEBUG
627         xive_error(xive, "VST: invalid %s entry %x !?", info->name, idx);
628 #endif
629         return 0;
630     }
631 
632     page_shift = GETFIELD(VSD_TSIZE, vsd) + 12;
633 
634     if (!pnv_xive_vst_page_size_allowed(page_shift)) {
635         xive_error(xive, "VST: invalid %s page shift %d", info->name,
636                    page_shift);
637         return 0;
638     }
639 
640     return (1ull << page_shift) / info->size;
641 }
642 
643 /*
644  * EDT Table
645  *
646  * The Virtualization Controller MMIO region containing the IPI ESB
647  * pages and END ESB pages is sub-divided into "sets" which map
648  * portions of the VC region to the different ESB pages. It is
649  * configured at runtime through the EDT "Domain Table" to let the
650  * firmware decide how to split the VC address space between IPI ESB
651  * pages and END ESB pages.
652  */
653 
654 /*
655  * Computes the overall size of the IPI or the END ESB pages
656  */
657 static uint64_t pnv_xive_edt_size(PnvXive *xive, uint64_t type)
658 {
659     uint64_t edt_size = 1ull << pnv_xive_edt_shift(xive);
660     uint64_t size = 0;
661     int i;
662 
663     for (i = 0; i < XIVE_TABLE_EDT_MAX; i++) {
664         uint64_t edt_type = GETFIELD(CQ_TDR_EDT_TYPE, xive->edt[i]);
665 
666         if (edt_type == type) {
667             size += edt_size;
668         }
669     }
670 
671     return size;
672 }
673 
674 /*
675  * Maps an offset of the VC region in the IPI or END region using the
676  * layout defined by the EDT "Domaine Table"
677  */
678 static uint64_t pnv_xive_edt_offset(PnvXive *xive, uint64_t vc_offset,
679                                               uint64_t type)
680 {
681     int i;
682     uint64_t edt_size = 1ull << pnv_xive_edt_shift(xive);
683     uint64_t edt_offset = vc_offset;
684 
685     for (i = 0; i < XIVE_TABLE_EDT_MAX && (i * edt_size) < vc_offset; i++) {
686         uint64_t edt_type = GETFIELD(CQ_TDR_EDT_TYPE, xive->edt[i]);
687 
688         if (edt_type != type) {
689             edt_offset -= edt_size;
690         }
691     }
692 
693     return edt_offset;
694 }
695 
696 static void pnv_xive_edt_resize(PnvXive *xive)
697 {
698     uint64_t ipi_edt_size = pnv_xive_edt_size(xive, CQ_TDR_EDT_IPI);
699     uint64_t end_edt_size = pnv_xive_edt_size(xive, CQ_TDR_EDT_EQ);
700 
701     memory_region_set_size(&xive->ipi_edt_mmio, ipi_edt_size);
702     memory_region_add_subregion(&xive->ipi_mmio, 0, &xive->ipi_edt_mmio);
703 
704     memory_region_set_size(&xive->end_edt_mmio, end_edt_size);
705     memory_region_add_subregion(&xive->end_mmio, 0, &xive->end_edt_mmio);
706 }
707 
708 /*
709  * XIVE Table configuration. Only EDT is supported.
710  */
711 static int pnv_xive_table_set_data(PnvXive *xive, uint64_t val)
712 {
713     uint64_t tsel = xive->regs[CQ_TAR >> 3] & CQ_TAR_TSEL;
714     uint8_t tsel_index = GETFIELD(CQ_TAR_TSEL_INDEX, xive->regs[CQ_TAR >> 3]);
715     uint64_t *xive_table;
716     uint8_t max_index;
717 
718     switch (tsel) {
719     case CQ_TAR_TSEL_BLK:
720         max_index = ARRAY_SIZE(xive->blk);
721         xive_table = xive->blk;
722         break;
723     case CQ_TAR_TSEL_MIG:
724         max_index = ARRAY_SIZE(xive->mig);
725         xive_table = xive->mig;
726         break;
727     case CQ_TAR_TSEL_EDT:
728         max_index = ARRAY_SIZE(xive->edt);
729         xive_table = xive->edt;
730         break;
731     case CQ_TAR_TSEL_VDT:
732         max_index = ARRAY_SIZE(xive->vdt);
733         xive_table = xive->vdt;
734         break;
735     default:
736         xive_error(xive, "IC: invalid table %d", (int) tsel);
737         return -1;
738     }
739 
740     if (tsel_index >= max_index) {
741         xive_error(xive, "IC: invalid index %d", (int) tsel_index);
742         return -1;
743     }
744 
745     xive_table[tsel_index] = val;
746 
747     if (xive->regs[CQ_TAR >> 3] & CQ_TAR_TBL_AUTOINC) {
748         xive->regs[CQ_TAR >> 3] =
749             SETFIELD(CQ_TAR_TSEL_INDEX, xive->regs[CQ_TAR >> 3], ++tsel_index);
750     }
751 
752     /*
753      * EDT configuration is complete. Resize the MMIO windows exposing
754      * the IPI and the END ESBs in the VC region.
755      */
756     if (tsel == CQ_TAR_TSEL_EDT && tsel_index == ARRAY_SIZE(xive->edt)) {
757         pnv_xive_edt_resize(xive);
758     }
759 
760     return 0;
761 }
762 
763 /*
764  * Virtual Structure Tables (VST) configuration
765  */
766 static void pnv_xive_vst_set_exclusive(PnvXive *xive, uint8_t type,
767                                        uint8_t blk, uint64_t vsd)
768 {
769     XiveENDSource *end_xsrc = &xive->end_source;
770     XiveSource *xsrc = &xive->ipi_source;
771     const XiveVstInfo *info = &vst_infos[type];
772     uint32_t page_shift = GETFIELD(VSD_TSIZE, vsd) + 12;
773     uint64_t vst_tsize = 1ull << page_shift;
774     uint64_t vst_addr = vsd & VSD_ADDRESS_MASK;
775 
776     /* Basic checks */
777 
778     if (VSD_INDIRECT & vsd) {
779         if (!(xive->regs[VC_GLOBAL_CONFIG >> 3] & VC_GCONF_INDIRECT)) {
780             xive_error(xive, "VST: %s indirect tables are not enabled",
781                        info->name);
782             return;
783         }
784 
785         if (!pnv_xive_vst_page_size_allowed(page_shift)) {
786             xive_error(xive, "VST: invalid %s page shift %d", info->name,
787                        page_shift);
788             return;
789         }
790     }
791 
792     if (!QEMU_IS_ALIGNED(vst_addr, 1ull << page_shift)) {
793         xive_error(xive, "VST: %s table address 0x%"PRIx64" is not aligned with"
794                    " page shift %d", info->name, vst_addr, page_shift);
795         return;
796     }
797 
798     /* Record the table configuration (in SRAM on HW) */
799     xive->vsds[type][blk] = vsd;
800 
801     /* Now tune the models with the configuration provided by the FW */
802 
803     switch (type) {
804     case VST_TSEL_IVT:  /* Nothing to be done */
805         break;
806 
807     case VST_TSEL_EQDT:
808         /*
809          * Backing store pages for the END.
810          *
811          * If the table is direct, we can compute the number of PQ
812          * entries provisioned by FW (such as skiboot) and resize the
813          * END ESB window accordingly.
814          */
815         if (!(VSD_INDIRECT & vsd)) {
816             memory_region_set_size(&end_xsrc->esb_mmio, (vst_tsize / info->size)
817                                    * (1ull << xsrc->esb_shift));
818         }
819         memory_region_add_subregion(&xive->end_edt_mmio, 0,
820                                     &end_xsrc->esb_mmio);
821         break;
822 
823     case VST_TSEL_SBE:
824         /*
825          * Backing store pages for the source PQ bits. The model does
826          * not use these PQ bits backed in RAM because the XiveSource
827          * model has its own.
828          *
829          * If the table is direct, we can compute the number of PQ
830          * entries provisioned by FW (such as skiboot) and resize the
831          * ESB window accordingly.
832          */
833         if (!(VSD_INDIRECT & vsd)) {
834             memory_region_set_size(&xsrc->esb_mmio, vst_tsize * SBE_PER_BYTE
835                                    * (1ull << xsrc->esb_shift));
836         }
837         memory_region_add_subregion(&xive->ipi_edt_mmio, 0, &xsrc->esb_mmio);
838         break;
839 
840     case VST_TSEL_VPDT: /* Not modeled */
841     case VST_TSEL_IRQ:  /* Not modeled */
842         /*
843          * These tables contains the backing store pages for the
844          * interrupt fifos of the VC sub-engine in case of overflow.
845          */
846         break;
847 
848     default:
849         g_assert_not_reached();
850     }
851 }
852 
853 /*
854  * Both PC and VC sub-engines are configured as each use the Virtual
855  * Structure Tables : SBE, EAS, END and NVT.
856  */
857 static void pnv_xive_vst_set_data(PnvXive *xive, uint64_t vsd, bool pc_engine)
858 {
859     uint8_t mode = GETFIELD(VSD_MODE, vsd);
860     uint8_t type = GETFIELD(VST_TABLE_SELECT,
861                             xive->regs[VC_VSD_TABLE_ADDR >> 3]);
862     uint8_t blk = GETFIELD(VST_TABLE_BLOCK,
863                            xive->regs[VC_VSD_TABLE_ADDR >> 3]);
864     uint64_t vst_addr = vsd & VSD_ADDRESS_MASK;
865 
866     if (type > VST_TSEL_IRQ) {
867         xive_error(xive, "VST: invalid table type %d", type);
868         return;
869     }
870 
871     if (blk >= vst_infos[type].max_blocks) {
872         xive_error(xive, "VST: invalid block id %d for"
873                       " %s table", blk, vst_infos[type].name);
874         return;
875     }
876 
877     /*
878      * Only take the VC sub-engine configuration into account because
879      * the XiveRouter model combines both VC and PC sub-engines
880      */
881     if (pc_engine) {
882         return;
883     }
884 
885     if (!vst_addr) {
886         xive_error(xive, "VST: invalid %s table address", vst_infos[type].name);
887         return;
888     }
889 
890     switch (mode) {
891     case VSD_MODE_FORWARD:
892         xive->vsds[type][blk] = vsd;
893         break;
894 
895     case VSD_MODE_EXCLUSIVE:
896         pnv_xive_vst_set_exclusive(xive, type, blk, vsd);
897         break;
898 
899     default:
900         xive_error(xive, "VST: unsupported table mode %d", mode);
901         return;
902     }
903 }
904 
905 /*
906  * Interrupt controller MMIO region. The layout is compatible between
907  * 4K and 64K pages :
908  *
909  * Page 0           sub-engine BARs
910  *  0x000 - 0x3FF   IC registers
911  *  0x400 - 0x7FF   PC registers
912  *  0x800 - 0xFFF   VC registers
913  *
914  * Page 1           Notify page (writes only)
915  *  0x000 - 0x7FF   HW interrupt triggers (PSI, PHB)
916  *  0x800 - 0xFFF   forwards and syncs
917  *
918  * Page 2           LSI Trigger page (writes only) (not modeled)
919  * Page 3           LSI SB EOI page (reads only) (not modeled)
920  *
921  * Page 4-7         indirect TIMA
922  */
923 
924 /*
925  * IC - registers MMIO
926  */
927 static void pnv_xive_ic_reg_write(void *opaque, hwaddr offset,
928                                   uint64_t val, unsigned size)
929 {
930     PnvXive *xive = PNV_XIVE(opaque);
931     MemoryRegion *sysmem = get_system_memory();
932     uint32_t reg = offset >> 3;
933     bool is_chip0 = xive->chip->chip_id == 0;
934 
935     switch (offset) {
936 
937     /*
938      * XIVE CQ (PowerBus bridge) settings
939      */
940     case CQ_MSGSND:     /* msgsnd for doorbells */
941     case CQ_FIRMASK_OR: /* FIR error reporting */
942         break;
943     case CQ_PBI_CTL:
944         if (val & CQ_PBI_PC_64K) {
945             xive->pc_shift = 16;
946         }
947         if (val & CQ_PBI_VC_64K) {
948             xive->vc_shift = 16;
949         }
950         break;
951     case CQ_CFG_PB_GEN: /* PowerBus General Configuration */
952         /*
953          * TODO: CQ_INT_ADDR_OPT for 1-block-per-chip mode
954          */
955         break;
956 
957     /*
958      * XIVE Virtualization Controller settings
959      */
960     case VC_GLOBAL_CONFIG:
961         break;
962 
963     /*
964      * XIVE Presenter Controller settings
965      */
966     case PC_GLOBAL_CONFIG:
967         /*
968          * PC_GCONF_CHIPID_OVR
969          *   Overrides Int command Chip ID with the Chip ID field (DEBUG)
970          */
971         break;
972     case PC_TCTXT_CFG:
973         /*
974          * TODO: block group support
975          */
976         break;
977     case PC_TCTXT_TRACK:
978         /*
979          * PC_TCTXT_TRACK_EN:
980          *   enable block tracking and exchange of block ownership
981          *   information between Interrupt controllers
982          */
983         break;
984 
985     /*
986      * Misc settings
987      */
988     case VC_SBC_CONFIG: /* Store EOI configuration */
989         /*
990          * Configure store EOI if required by firmware (skiboot has removed
991          * support recently though)
992          */
993         if (val & (VC_SBC_CONF_CPLX_CIST | VC_SBC_CONF_CIST_BOTH)) {
994             xive->ipi_source.esb_flags |= XIVE_SRC_STORE_EOI;
995         }
996         break;
997 
998     case VC_EQC_CONFIG: /* TODO: silent escalation */
999     case VC_AIB_TX_ORDER_TAG2: /* relax ordering */
1000         break;
1001 
1002     /*
1003      * XIVE BAR settings (XSCOM only)
1004      */
1005     case CQ_RST_CTL:
1006         /* bit4: resets all BAR registers */
1007         break;
1008 
1009     case CQ_IC_BAR: /* IC BAR. 8 pages */
1010         xive->ic_shift = val & CQ_IC_BAR_64K ? 16 : 12;
1011         if (!(val & CQ_IC_BAR_VALID)) {
1012             xive->ic_base = 0;
1013             if (xive->regs[reg] & CQ_IC_BAR_VALID) {
1014                 memory_region_del_subregion(&xive->ic_mmio,
1015                                             &xive->ic_reg_mmio);
1016                 memory_region_del_subregion(&xive->ic_mmio,
1017                                             &xive->ic_notify_mmio);
1018                 memory_region_del_subregion(&xive->ic_mmio,
1019                                             &xive->ic_lsi_mmio);
1020                 memory_region_del_subregion(&xive->ic_mmio,
1021                                             &xive->tm_indirect_mmio);
1022 
1023                 memory_region_del_subregion(sysmem, &xive->ic_mmio);
1024             }
1025         } else {
1026             xive->ic_base = val & ~(CQ_IC_BAR_VALID | CQ_IC_BAR_64K);
1027             if (!(xive->regs[reg] & CQ_IC_BAR_VALID)) {
1028                 memory_region_add_subregion(sysmem, xive->ic_base,
1029                                             &xive->ic_mmio);
1030 
1031                 memory_region_add_subregion(&xive->ic_mmio,  0,
1032                                             &xive->ic_reg_mmio);
1033                 memory_region_add_subregion(&xive->ic_mmio,
1034                                             1ul << xive->ic_shift,
1035                                             &xive->ic_notify_mmio);
1036                 memory_region_add_subregion(&xive->ic_mmio,
1037                                             2ul << xive->ic_shift,
1038                                             &xive->ic_lsi_mmio);
1039                 memory_region_add_subregion(&xive->ic_mmio,
1040                                             4ull << xive->ic_shift,
1041                                             &xive->tm_indirect_mmio);
1042             }
1043         }
1044         break;
1045 
1046     case CQ_TM1_BAR: /* TM BAR. 4 pages. Map only once */
1047     case CQ_TM2_BAR: /* second TM BAR. for hotplug. Not modeled */
1048         xive->tm_shift = val & CQ_TM_BAR_64K ? 16 : 12;
1049         if (!(val & CQ_TM_BAR_VALID)) {
1050             xive->tm_base = 0;
1051             if (xive->regs[reg] & CQ_TM_BAR_VALID && is_chip0) {
1052                 memory_region_del_subregion(sysmem, &xive->tm_mmio);
1053             }
1054         } else {
1055             xive->tm_base = val & ~(CQ_TM_BAR_VALID | CQ_TM_BAR_64K);
1056             if (!(xive->regs[reg] & CQ_TM_BAR_VALID) && is_chip0) {
1057                 memory_region_add_subregion(sysmem, xive->tm_base,
1058                                             &xive->tm_mmio);
1059             }
1060         }
1061         break;
1062 
1063     case CQ_PC_BARM:
1064         xive->regs[reg] = val;
1065         memory_region_set_size(&xive->pc_mmio, pnv_xive_pc_size(xive));
1066         break;
1067     case CQ_PC_BAR: /* From 32M to 512G */
1068         if (!(val & CQ_PC_BAR_VALID)) {
1069             xive->pc_base = 0;
1070             if (xive->regs[reg] & CQ_PC_BAR_VALID) {
1071                 memory_region_del_subregion(sysmem, &xive->pc_mmio);
1072             }
1073         } else {
1074             xive->pc_base = val & ~(CQ_PC_BAR_VALID);
1075             if (!(xive->regs[reg] & CQ_PC_BAR_VALID)) {
1076                 memory_region_add_subregion(sysmem, xive->pc_base,
1077                                             &xive->pc_mmio);
1078             }
1079         }
1080         break;
1081 
1082     case CQ_VC_BARM:
1083         xive->regs[reg] = val;
1084         memory_region_set_size(&xive->vc_mmio, pnv_xive_vc_size(xive));
1085         break;
1086     case CQ_VC_BAR: /* From 64M to 4TB */
1087         if (!(val & CQ_VC_BAR_VALID)) {
1088             xive->vc_base = 0;
1089             if (xive->regs[reg] & CQ_VC_BAR_VALID) {
1090                 memory_region_del_subregion(sysmem, &xive->vc_mmio);
1091             }
1092         } else {
1093             xive->vc_base = val & ~(CQ_VC_BAR_VALID);
1094             if (!(xive->regs[reg] & CQ_VC_BAR_VALID)) {
1095                 memory_region_add_subregion(sysmem, xive->vc_base,
1096                                             &xive->vc_mmio);
1097             }
1098         }
1099         break;
1100 
1101     /*
1102      * XIVE Table settings.
1103      */
1104     case CQ_TAR: /* Table Address */
1105         break;
1106     case CQ_TDR: /* Table Data */
1107         pnv_xive_table_set_data(xive, val);
1108         break;
1109 
1110     /*
1111      * XIVE VC & PC Virtual Structure Table settings
1112      */
1113     case VC_VSD_TABLE_ADDR:
1114     case PC_VSD_TABLE_ADDR: /* Virtual table selector */
1115         break;
1116     case VC_VSD_TABLE_DATA: /* Virtual table setting */
1117     case PC_VSD_TABLE_DATA:
1118         pnv_xive_vst_set_data(xive, val, offset == PC_VSD_TABLE_DATA);
1119         break;
1120 
1121     /*
1122      * Interrupt fifo overflow in memory backing store (Not modeled)
1123      */
1124     case VC_IRQ_CONFIG_IPI:
1125     case VC_IRQ_CONFIG_HW:
1126     case VC_IRQ_CONFIG_CASCADE1:
1127     case VC_IRQ_CONFIG_CASCADE2:
1128     case VC_IRQ_CONFIG_REDIST:
1129     case VC_IRQ_CONFIG_IPI_CASC:
1130         break;
1131 
1132     /*
1133      * XIVE hardware thread enablement
1134      */
1135     case PC_THREAD_EN_REG0: /* Physical Thread Enable */
1136     case PC_THREAD_EN_REG1: /* Physical Thread Enable (fused core) */
1137         break;
1138 
1139     case PC_THREAD_EN_REG0_SET:
1140         xive->regs[PC_THREAD_EN_REG0 >> 3] |= val;
1141         break;
1142     case PC_THREAD_EN_REG1_SET:
1143         xive->regs[PC_THREAD_EN_REG1 >> 3] |= val;
1144         break;
1145     case PC_THREAD_EN_REG0_CLR:
1146         xive->regs[PC_THREAD_EN_REG0 >> 3] &= ~val;
1147         break;
1148     case PC_THREAD_EN_REG1_CLR:
1149         xive->regs[PC_THREAD_EN_REG1 >> 3] &= ~val;
1150         break;
1151 
1152     /*
1153      * Indirect TIMA access set up. Defines the PIR of the HW thread
1154      * to use.
1155      */
1156     case PC_TCTXT_INDIR0 ... PC_TCTXT_INDIR3:
1157         break;
1158 
1159     /*
1160      * XIVE PC & VC cache updates for EAS, NVT and END
1161      */
1162     case VC_IVC_SCRUB_MASK:
1163     case VC_IVC_SCRUB_TRIG:
1164         break;
1165 
1166     case VC_EQC_CWATCH_SPEC:
1167         val &= ~VC_EQC_CWATCH_CONFLICT; /* HW resets this bit */
1168         break;
1169     case VC_EQC_CWATCH_DAT1 ... VC_EQC_CWATCH_DAT3:
1170         break;
1171     case VC_EQC_CWATCH_DAT0:
1172         /* writing to DATA0 triggers the cache write */
1173         xive->regs[reg] = val;
1174         pnv_xive_end_update(xive);
1175         break;
1176     case VC_EQC_SCRUB_MASK:
1177     case VC_EQC_SCRUB_TRIG:
1178         /*
1179          * The scrubbing registers flush the cache in RAM and can also
1180          * invalidate.
1181          */
1182         break;
1183 
1184     case PC_VPC_CWATCH_SPEC:
1185         val &= ~PC_VPC_CWATCH_CONFLICT; /* HW resets this bit */
1186         break;
1187     case PC_VPC_CWATCH_DAT1 ... PC_VPC_CWATCH_DAT7:
1188         break;
1189     case PC_VPC_CWATCH_DAT0:
1190         /* writing to DATA0 triggers the cache write */
1191         xive->regs[reg] = val;
1192         pnv_xive_nvt_update(xive);
1193         break;
1194     case PC_VPC_SCRUB_MASK:
1195     case PC_VPC_SCRUB_TRIG:
1196         /*
1197          * The scrubbing registers flush the cache in RAM and can also
1198          * invalidate.
1199          */
1200         break;
1201 
1202 
1203     /*
1204      * XIVE PC & VC cache invalidation
1205      */
1206     case PC_AT_KILL:
1207         break;
1208     case VC_AT_MACRO_KILL:
1209         break;
1210     case PC_AT_KILL_MASK:
1211     case VC_AT_MACRO_KILL_MASK:
1212         break;
1213 
1214     default:
1215         xive_error(xive, "IC: invalid write to reg=0x%"HWADDR_PRIx, offset);
1216         return;
1217     }
1218 
1219     xive->regs[reg] = val;
1220 }
1221 
1222 static uint64_t pnv_xive_ic_reg_read(void *opaque, hwaddr offset, unsigned size)
1223 {
1224     PnvXive *xive = PNV_XIVE(opaque);
1225     uint64_t val = 0;
1226     uint32_t reg = offset >> 3;
1227 
1228     switch (offset) {
1229     case CQ_CFG_PB_GEN:
1230     case CQ_IC_BAR:
1231     case CQ_TM1_BAR:
1232     case CQ_TM2_BAR:
1233     case CQ_PC_BAR:
1234     case CQ_PC_BARM:
1235     case CQ_VC_BAR:
1236     case CQ_VC_BARM:
1237     case CQ_TAR:
1238     case CQ_TDR:
1239     case CQ_PBI_CTL:
1240 
1241     case PC_TCTXT_CFG:
1242     case PC_TCTXT_TRACK:
1243     case PC_TCTXT_INDIR0:
1244     case PC_TCTXT_INDIR1:
1245     case PC_TCTXT_INDIR2:
1246     case PC_TCTXT_INDIR3:
1247     case PC_GLOBAL_CONFIG:
1248 
1249     case PC_VPC_SCRUB_MASK:
1250 
1251     case VC_GLOBAL_CONFIG:
1252     case VC_AIB_TX_ORDER_TAG2:
1253 
1254     case VC_IRQ_CONFIG_IPI:
1255     case VC_IRQ_CONFIG_HW:
1256     case VC_IRQ_CONFIG_CASCADE1:
1257     case VC_IRQ_CONFIG_CASCADE2:
1258     case VC_IRQ_CONFIG_REDIST:
1259     case VC_IRQ_CONFIG_IPI_CASC:
1260 
1261     case VC_EQC_SCRUB_MASK:
1262     case VC_IVC_SCRUB_MASK:
1263     case VC_SBC_CONFIG:
1264     case VC_AT_MACRO_KILL_MASK:
1265     case VC_VSD_TABLE_ADDR:
1266     case PC_VSD_TABLE_ADDR:
1267     case VC_VSD_TABLE_DATA:
1268     case PC_VSD_TABLE_DATA:
1269     case PC_THREAD_EN_REG0:
1270     case PC_THREAD_EN_REG1:
1271         val = xive->regs[reg];
1272         break;
1273 
1274     /*
1275      * XIVE hardware thread enablement
1276      */
1277     case PC_THREAD_EN_REG0_SET:
1278     case PC_THREAD_EN_REG0_CLR:
1279         val = xive->regs[PC_THREAD_EN_REG0 >> 3];
1280         break;
1281     case PC_THREAD_EN_REG1_SET:
1282     case PC_THREAD_EN_REG1_CLR:
1283         val = xive->regs[PC_THREAD_EN_REG1 >> 3];
1284         break;
1285 
1286     case CQ_MSGSND: /* Identifies which cores have msgsnd enabled. */
1287         val = 0xffffff0000000000;
1288         break;
1289 
1290     /*
1291      * XIVE PC & VC cache updates for EAS, NVT and END
1292      */
1293     case VC_EQC_CWATCH_SPEC:
1294         xive->regs[reg] = ~(VC_EQC_CWATCH_FULL | VC_EQC_CWATCH_CONFLICT);
1295         val = xive->regs[reg];
1296         break;
1297     case VC_EQC_CWATCH_DAT0:
1298         /*
1299          * Load DATA registers from cache with data requested by the
1300          * SPEC register
1301          */
1302         pnv_xive_end_cache_load(xive);
1303         val = xive->regs[reg];
1304         break;
1305     case VC_EQC_CWATCH_DAT1 ... VC_EQC_CWATCH_DAT3:
1306         val = xive->regs[reg];
1307         break;
1308 
1309     case PC_VPC_CWATCH_SPEC:
1310         xive->regs[reg] = ~(PC_VPC_CWATCH_FULL | PC_VPC_CWATCH_CONFLICT);
1311         val = xive->regs[reg];
1312         break;
1313     case PC_VPC_CWATCH_DAT0:
1314         /*
1315          * Load DATA registers from cache with data requested by the
1316          * SPEC register
1317          */
1318         pnv_xive_nvt_cache_load(xive);
1319         val = xive->regs[reg];
1320         break;
1321     case PC_VPC_CWATCH_DAT1 ... PC_VPC_CWATCH_DAT7:
1322         val = xive->regs[reg];
1323         break;
1324 
1325     case PC_VPC_SCRUB_TRIG:
1326     case VC_IVC_SCRUB_TRIG:
1327     case VC_EQC_SCRUB_TRIG:
1328         xive->regs[reg] &= ~VC_SCRUB_VALID;
1329         val = xive->regs[reg];
1330         break;
1331 
1332     /*
1333      * XIVE PC & VC cache invalidation
1334      */
1335     case PC_AT_KILL:
1336         xive->regs[reg] &= ~PC_AT_KILL_VALID;
1337         val = xive->regs[reg];
1338         break;
1339     case VC_AT_MACRO_KILL:
1340         xive->regs[reg] &= ~VC_KILL_VALID;
1341         val = xive->regs[reg];
1342         break;
1343 
1344     /*
1345      * XIVE synchronisation
1346      */
1347     case VC_EQC_CONFIG:
1348         val = VC_EQC_SYNC_MASK;
1349         break;
1350 
1351     default:
1352         xive_error(xive, "IC: invalid read reg=0x%"HWADDR_PRIx, offset);
1353     }
1354 
1355     return val;
1356 }
1357 
1358 static const MemoryRegionOps pnv_xive_ic_reg_ops = {
1359     .read = pnv_xive_ic_reg_read,
1360     .write = pnv_xive_ic_reg_write,
1361     .endianness = DEVICE_BIG_ENDIAN,
1362     .valid = {
1363         .min_access_size = 8,
1364         .max_access_size = 8,
1365     },
1366     .impl = {
1367         .min_access_size = 8,
1368         .max_access_size = 8,
1369     },
1370 };
1371 
1372 /*
1373  * IC - Notify MMIO port page (write only)
1374  */
1375 #define PNV_XIVE_FORWARD_IPI        0x800 /* Forward IPI */
1376 #define PNV_XIVE_FORWARD_HW         0x880 /* Forward HW */
1377 #define PNV_XIVE_FORWARD_OS_ESC     0x900 /* Forward OS escalation */
1378 #define PNV_XIVE_FORWARD_HW_ESC     0x980 /* Forward Hyp escalation */
1379 #define PNV_XIVE_FORWARD_REDIS      0xa00 /* Forward Redistribution */
1380 #define PNV_XIVE_RESERVED5          0xa80 /* Cache line 5 PowerBUS operation */
1381 #define PNV_XIVE_RESERVED6          0xb00 /* Cache line 6 PowerBUS operation */
1382 #define PNV_XIVE_RESERVED7          0xb80 /* Cache line 7 PowerBUS operation */
1383 
1384 /* VC synchronisation */
1385 #define PNV_XIVE_SYNC_IPI           0xc00 /* Sync IPI */
1386 #define PNV_XIVE_SYNC_HW            0xc80 /* Sync HW */
1387 #define PNV_XIVE_SYNC_OS_ESC        0xd00 /* Sync OS escalation */
1388 #define PNV_XIVE_SYNC_HW_ESC        0xd80 /* Sync Hyp escalation */
1389 #define PNV_XIVE_SYNC_REDIS         0xe00 /* Sync Redistribution */
1390 
1391 /* PC synchronisation */
1392 #define PNV_XIVE_SYNC_PULL          0xe80 /* Sync pull context */
1393 #define PNV_XIVE_SYNC_PUSH          0xf00 /* Sync push context */
1394 #define PNV_XIVE_SYNC_VPC           0xf80 /* Sync remove VPC store */
1395 
1396 static void pnv_xive_end_notify(XiveRouter *xrtr, XiveEAS *eas)
1397 {
1398     PnvXive *xive = PNV_XIVE(xrtr);
1399     uint8_t end_blk = xive_get_field64(EAS_END_BLOCK, eas->w);
1400     uint32_t end_idx = xive_get_field64(EAS_END_INDEX, eas->w);
1401     uint32_t end_data = xive_get_field64(EAS_END_DATA, eas->w);
1402     uint64_t end_vsd = xive->vsds[VST_TSEL_EQDT][end_blk];
1403 
1404     switch (GETFIELD(VSD_MODE, end_vsd)) {
1405     case VSD_MODE_EXCLUSIVE:
1406         /* Perform the END notification on the local IC. */
1407         xive_router_end_notify(xrtr, eas);
1408         break;
1409 
1410     case VSD_MODE_FORWARD: {
1411         MemTxResult result;
1412         uint64_t notif_port = end_vsd & VSD_ADDRESS_MASK;
1413         uint64_t data = XIVE_TRIGGER_END | XIVE_TRIGGER_PQ |
1414             be64_to_cpu(eas->w);
1415 
1416         /* Forward the store on the remote IC notify page. */
1417         address_space_stq_be(&address_space_memory, notif_port, data,
1418                              MEMTXATTRS_UNSPECIFIED, &result);
1419         if (result != MEMTX_OK) {
1420             xive_error(xive, "IC: Forward notif END %x/%x [%x] failed @%"
1421                        HWADDR_PRIx, end_blk, end_idx, end_data, notif_port);
1422             return;
1423         }
1424         break;
1425     }
1426 
1427     case VSD_MODE_INVALID:
1428     default:
1429         /* Set FIR */
1430         xive_error(xive, "IC: Invalid END VSD for block %x", end_blk);
1431         return;
1432     }
1433 }
1434 
1435 /*
1436  * The notify page can either be used to receive trigger events from
1437  * the HW controllers (PHB, PSI) or to reroute interrupts between
1438  * Interrupt controllers.
1439  */
1440 static void pnv_xive_ic_hw_trigger(PnvXive *xive, hwaddr addr, uint64_t val)
1441 {
1442     uint8_t blk;
1443     uint32_t idx;
1444 
1445     trace_pnv_xive_ic_hw_trigger(addr, val);
1446 
1447     if (val & XIVE_TRIGGER_END) {
1448         val = cpu_to_be64(val);
1449         pnv_xive_end_notify(XIVE_ROUTER(xive), (XiveEAS *) &val);
1450         return;
1451     }
1452 
1453     /*
1454      * Forward the source event notification directly to the Router.
1455      * The source interrupt number should already be correctly encoded
1456      * with the chip block id by the sending device (PHB, PSI).
1457      */
1458     blk = XIVE_EAS_BLOCK(val);
1459     idx = XIVE_EAS_INDEX(val);
1460 
1461     xive_router_notify(XIVE_NOTIFIER(xive), XIVE_EAS(blk, idx),
1462                        !!(val & XIVE_TRIGGER_PQ));
1463 }
1464 
1465 static void pnv_xive_ic_notify_write(void *opaque, hwaddr addr, uint64_t val,
1466                                      unsigned size)
1467 {
1468     PnvXive *xive = PNV_XIVE(opaque);
1469 
1470     /* VC: HW triggers */
1471     switch (addr) {
1472     case 0x000 ... 0x7FF:
1473         pnv_xive_ic_hw_trigger(opaque, addr, val);
1474         break;
1475 
1476     /* VC: Forwarded IRQs */
1477     case PNV_XIVE_FORWARD_IPI:
1478     case PNV_XIVE_FORWARD_HW:
1479     case PNV_XIVE_FORWARD_OS_ESC:
1480     case PNV_XIVE_FORWARD_HW_ESC:
1481     case PNV_XIVE_FORWARD_REDIS:
1482         /* TODO: forwarded IRQs. Should be like HW triggers */
1483         xive_error(xive, "IC: forwarded at @0x%"HWADDR_PRIx" IRQ 0x%"PRIx64,
1484                    addr, val);
1485         break;
1486 
1487     /* VC syncs */
1488     case PNV_XIVE_SYNC_IPI:
1489     case PNV_XIVE_SYNC_HW:
1490     case PNV_XIVE_SYNC_OS_ESC:
1491     case PNV_XIVE_SYNC_HW_ESC:
1492     case PNV_XIVE_SYNC_REDIS:
1493         break;
1494 
1495     /* PC syncs */
1496     case PNV_XIVE_SYNC_PULL:
1497     case PNV_XIVE_SYNC_PUSH:
1498     case PNV_XIVE_SYNC_VPC:
1499         break;
1500 
1501     default:
1502         xive_error(xive, "IC: invalid notify write @%"HWADDR_PRIx, addr);
1503     }
1504 }
1505 
1506 static uint64_t pnv_xive_ic_notify_read(void *opaque, hwaddr addr,
1507                                         unsigned size)
1508 {
1509     PnvXive *xive = PNV_XIVE(opaque);
1510 
1511     /* loads are invalid */
1512     xive_error(xive, "IC: invalid notify read @%"HWADDR_PRIx, addr);
1513     return -1;
1514 }
1515 
1516 static const MemoryRegionOps pnv_xive_ic_notify_ops = {
1517     .read = pnv_xive_ic_notify_read,
1518     .write = pnv_xive_ic_notify_write,
1519     .endianness = DEVICE_BIG_ENDIAN,
1520     .valid = {
1521         .min_access_size = 8,
1522         .max_access_size = 8,
1523     },
1524     .impl = {
1525         .min_access_size = 8,
1526         .max_access_size = 8,
1527     },
1528 };
1529 
1530 /*
1531  * IC - LSI MMIO handlers (not modeled)
1532  */
1533 
1534 static void pnv_xive_ic_lsi_write(void *opaque, hwaddr addr,
1535                               uint64_t val, unsigned size)
1536 {
1537     PnvXive *xive = PNV_XIVE(opaque);
1538 
1539     xive_error(xive, "IC: LSI invalid write @%"HWADDR_PRIx, addr);
1540 }
1541 
1542 static uint64_t pnv_xive_ic_lsi_read(void *opaque, hwaddr addr, unsigned size)
1543 {
1544     PnvXive *xive = PNV_XIVE(opaque);
1545 
1546     xive_error(xive, "IC: LSI invalid read @%"HWADDR_PRIx, addr);
1547     return -1;
1548 }
1549 
1550 static const MemoryRegionOps pnv_xive_ic_lsi_ops = {
1551     .read = pnv_xive_ic_lsi_read,
1552     .write = pnv_xive_ic_lsi_write,
1553     .endianness = DEVICE_BIG_ENDIAN,
1554     .valid = {
1555         .min_access_size = 8,
1556         .max_access_size = 8,
1557     },
1558     .impl = {
1559         .min_access_size = 8,
1560         .max_access_size = 8,
1561     },
1562 };
1563 
1564 /*
1565  * IC - Indirect TIMA MMIO handlers
1566  */
1567 
1568 /*
1569  * When the TIMA is accessed from the indirect page, the thread id of
1570  * the target CPU is configured in the PC_TCTXT_INDIR0 register before
1571  * use. This is used for resets and for debug purpose also.
1572  */
1573 static XiveTCTX *pnv_xive_get_indirect_tctx(PnvXive *xive)
1574 {
1575     PnvChip *chip = xive->chip;
1576     uint64_t tctxt_indir = xive->regs[PC_TCTXT_INDIR0 >> 3];
1577     PowerPCCPU *cpu = NULL;
1578     int pir;
1579 
1580     if (!(tctxt_indir & PC_TCTXT_INDIR_VALID)) {
1581         xive_error(xive, "IC: no indirect TIMA access in progress");
1582         return NULL;
1583     }
1584 
1585     pir = (chip->chip_id << 8) | GETFIELD(PC_TCTXT_INDIR_THRDID, tctxt_indir);
1586     cpu = pnv_chip_find_cpu(chip, pir);
1587     if (!cpu) {
1588         xive_error(xive, "IC: invalid PIR %x for indirect access", pir);
1589         return NULL;
1590     }
1591 
1592     /* Check that HW thread is XIVE enabled */
1593     if (!pnv_xive_is_cpu_enabled(xive, cpu)) {
1594         xive_error(xive, "IC: CPU %x is not enabled", pir);
1595     }
1596 
1597     return XIVE_TCTX(pnv_cpu_state(cpu)->intc);
1598 }
1599 
1600 static void xive_tm_indirect_write(void *opaque, hwaddr offset,
1601                                    uint64_t value, unsigned size)
1602 {
1603     XiveTCTX *tctx = pnv_xive_get_indirect_tctx(PNV_XIVE(opaque));
1604 
1605     xive_tctx_tm_write(XIVE_PRESENTER(opaque), tctx, offset, value, size);
1606 }
1607 
1608 static uint64_t xive_tm_indirect_read(void *opaque, hwaddr offset,
1609                                       unsigned size)
1610 {
1611     XiveTCTX *tctx = pnv_xive_get_indirect_tctx(PNV_XIVE(opaque));
1612 
1613     return xive_tctx_tm_read(XIVE_PRESENTER(opaque), tctx, offset, size);
1614 }
1615 
1616 static const MemoryRegionOps xive_tm_indirect_ops = {
1617     .read = xive_tm_indirect_read,
1618     .write = xive_tm_indirect_write,
1619     .endianness = DEVICE_BIG_ENDIAN,
1620     .valid = {
1621         .min_access_size = 1,
1622         .max_access_size = 8,
1623     },
1624     .impl = {
1625         .min_access_size = 1,
1626         .max_access_size = 8,
1627     },
1628 };
1629 
1630 static void pnv_xive_tm_write(void *opaque, hwaddr offset,
1631                               uint64_t value, unsigned size)
1632 {
1633     PowerPCCPU *cpu = POWERPC_CPU(current_cpu);
1634     PnvXive *xive = pnv_xive_tm_get_xive(cpu);
1635     XiveTCTX *tctx = XIVE_TCTX(pnv_cpu_state(cpu)->intc);
1636 
1637     xive_tctx_tm_write(XIVE_PRESENTER(xive), tctx, offset, value, size);
1638 }
1639 
1640 static uint64_t pnv_xive_tm_read(void *opaque, hwaddr offset, unsigned size)
1641 {
1642     PowerPCCPU *cpu = POWERPC_CPU(current_cpu);
1643     PnvXive *xive = pnv_xive_tm_get_xive(cpu);
1644     XiveTCTX *tctx = XIVE_TCTX(pnv_cpu_state(cpu)->intc);
1645 
1646     return xive_tctx_tm_read(XIVE_PRESENTER(xive), tctx, offset, size);
1647 }
1648 
1649 const MemoryRegionOps pnv_xive_tm_ops = {
1650     .read = pnv_xive_tm_read,
1651     .write = pnv_xive_tm_write,
1652     .endianness = DEVICE_BIG_ENDIAN,
1653     .valid = {
1654         .min_access_size = 1,
1655         .max_access_size = 8,
1656     },
1657     .impl = {
1658         .min_access_size = 1,
1659         .max_access_size = 8,
1660     },
1661 };
1662 
1663 /*
1664  * Interrupt controller XSCOM region.
1665  */
1666 static uint64_t pnv_xive_xscom_read(void *opaque, hwaddr addr, unsigned size)
1667 {
1668     switch (addr >> 3) {
1669     case X_VC_EQC_CONFIG:
1670         /* FIXME (skiboot): This is the only XSCOM load. Bizarre. */
1671         return VC_EQC_SYNC_MASK;
1672     default:
1673         return pnv_xive_ic_reg_read(opaque, addr, size);
1674     }
1675 }
1676 
1677 static void pnv_xive_xscom_write(void *opaque, hwaddr addr,
1678                                 uint64_t val, unsigned size)
1679 {
1680     pnv_xive_ic_reg_write(opaque, addr, val, size);
1681 }
1682 
1683 static const MemoryRegionOps pnv_xive_xscom_ops = {
1684     .read = pnv_xive_xscom_read,
1685     .write = pnv_xive_xscom_write,
1686     .endianness = DEVICE_BIG_ENDIAN,
1687     .valid = {
1688         .min_access_size = 8,
1689         .max_access_size = 8,
1690     },
1691     .impl = {
1692         .min_access_size = 8,
1693         .max_access_size = 8,
1694     }
1695 };
1696 
1697 /*
1698  * Virtualization Controller MMIO region containing the IPI and END ESB pages
1699  */
1700 static uint64_t pnv_xive_vc_read(void *opaque, hwaddr offset,
1701                                  unsigned size)
1702 {
1703     PnvXive *xive = PNV_XIVE(opaque);
1704     uint64_t edt_index = offset >> pnv_xive_edt_shift(xive);
1705     uint64_t edt_type = 0;
1706     uint64_t edt_offset;
1707     MemTxResult result;
1708     AddressSpace *edt_as = NULL;
1709     uint64_t ret = -1;
1710 
1711     if (edt_index < XIVE_TABLE_EDT_MAX) {
1712         edt_type = GETFIELD(CQ_TDR_EDT_TYPE, xive->edt[edt_index]);
1713     }
1714 
1715     switch (edt_type) {
1716     case CQ_TDR_EDT_IPI:
1717         edt_as = &xive->ipi_as;
1718         break;
1719     case CQ_TDR_EDT_EQ:
1720         edt_as = &xive->end_as;
1721         break;
1722     default:
1723         xive_error(xive, "VC: invalid EDT type for read @%"HWADDR_PRIx, offset);
1724         return -1;
1725     }
1726 
1727     /* Remap the offset for the targeted address space */
1728     edt_offset = pnv_xive_edt_offset(xive, offset, edt_type);
1729 
1730     ret = address_space_ldq(edt_as, edt_offset, MEMTXATTRS_UNSPECIFIED,
1731                             &result);
1732 
1733     if (result != MEMTX_OK) {
1734         xive_error(xive, "VC: %s read failed at @0x%"HWADDR_PRIx " -> @0x%"
1735                    HWADDR_PRIx, edt_type == CQ_TDR_EDT_IPI ? "IPI" : "END",
1736                    offset, edt_offset);
1737         return -1;
1738     }
1739 
1740     return ret;
1741 }
1742 
1743 static void pnv_xive_vc_write(void *opaque, hwaddr offset,
1744                               uint64_t val, unsigned size)
1745 {
1746     PnvXive *xive = PNV_XIVE(opaque);
1747     uint64_t edt_index = offset >> pnv_xive_edt_shift(xive);
1748     uint64_t edt_type = 0;
1749     uint64_t edt_offset;
1750     MemTxResult result;
1751     AddressSpace *edt_as = NULL;
1752 
1753     if (edt_index < XIVE_TABLE_EDT_MAX) {
1754         edt_type = GETFIELD(CQ_TDR_EDT_TYPE, xive->edt[edt_index]);
1755     }
1756 
1757     switch (edt_type) {
1758     case CQ_TDR_EDT_IPI:
1759         edt_as = &xive->ipi_as;
1760         break;
1761     case CQ_TDR_EDT_EQ:
1762         edt_as = &xive->end_as;
1763         break;
1764     default:
1765         xive_error(xive, "VC: invalid EDT type for write @%"HWADDR_PRIx,
1766                    offset);
1767         return;
1768     }
1769 
1770     /* Remap the offset for the targeted address space */
1771     edt_offset = pnv_xive_edt_offset(xive, offset, edt_type);
1772 
1773     address_space_stq(edt_as, edt_offset, val, MEMTXATTRS_UNSPECIFIED, &result);
1774     if (result != MEMTX_OK) {
1775         xive_error(xive, "VC: write failed at @0x%"HWADDR_PRIx, edt_offset);
1776     }
1777 }
1778 
1779 static const MemoryRegionOps pnv_xive_vc_ops = {
1780     .read = pnv_xive_vc_read,
1781     .write = pnv_xive_vc_write,
1782     .endianness = DEVICE_BIG_ENDIAN,
1783     .valid = {
1784         .min_access_size = 8,
1785         .max_access_size = 8,
1786     },
1787     .impl = {
1788         .min_access_size = 8,
1789         .max_access_size = 8,
1790     },
1791 };
1792 
1793 /*
1794  * Presenter Controller MMIO region. Points to the NVT sets.
1795  *
1796  * HW implements all possible mem ops to the underlying NVT structure
1797  * but QEMU does not need to be so precise. The model implementation
1798  * simply returns the RAM address of the NVT structure which is then
1799  * used by pnv_xive_vst_write/read to perform the RAM operation.
1800  */
1801 static uint64_t pnv_xive_pc_read(void *opaque, hwaddr offset, unsigned size)
1802 {
1803     PnvXive *xive = PNV_XIVE(opaque);
1804     uint32_t nvt_idx = offset >> xive->pc_shift;
1805     uint8_t blk = pnv_xive_block_id(xive); /* TODO: VDT -> block xlate */
1806 
1807     return pnv_xive_vst_addr(xive, VST_TSEL_VPDT, blk, nvt_idx);
1808 }
1809 
1810 static void pnv_xive_pc_write(void *opaque, hwaddr addr,
1811                               uint64_t value, unsigned size)
1812 {
1813     PnvXive *xive = PNV_XIVE(opaque);
1814 
1815     xive_error(xive, "PC: invalid write to VC @%"HWADDR_PRIx, addr);
1816 }
1817 
1818 static const MemoryRegionOps pnv_xive_pc_ops = {
1819     .read = pnv_xive_pc_read,
1820     .write = pnv_xive_pc_write,
1821     .endianness = DEVICE_BIG_ENDIAN,
1822     .valid = {
1823         .min_access_size = 8,
1824         .max_access_size = 8,
1825     },
1826     .impl = {
1827         .min_access_size = 8,
1828         .max_access_size = 8,
1829     },
1830 };
1831 
1832 static void xive_nvt_pic_print_info(XiveNVT *nvt, uint32_t nvt_idx,
1833                                     GString *buf)
1834 {
1835     uint8_t  eq_blk = xive_get_field32(NVT_W1_EQ_BLOCK, nvt->w1);
1836     uint32_t eq_idx = xive_get_field32(NVT_W1_EQ_INDEX, nvt->w1);
1837 
1838     if (!xive_nvt_is_valid(nvt)) {
1839         return;
1840     }
1841 
1842     g_string_append_printf(buf, "  %08x end:%02x/%04x IPB:%02x\n",
1843                            nvt_idx, eq_blk, eq_idx,
1844                            xive_get_field32(NVT_W4_IPB, nvt->w4));
1845 }
1846 
1847 void pnv_xive_pic_print_info(PnvXive *xive, GString *buf)
1848 {
1849     XiveRouter *xrtr = XIVE_ROUTER(xive);
1850     uint8_t blk = pnv_xive_block_id(xive);
1851     uint8_t chip_id = xive->chip->chip_id;
1852     uint32_t srcno0 = XIVE_EAS(blk, 0);
1853     uint32_t nr_ipis = pnv_xive_nr_ipis(xive, blk);
1854     XiveEAS eas;
1855     XiveEND end;
1856     XiveNVT nvt;
1857     int i;
1858     uint64_t xive_nvt_per_subpage;
1859 
1860     g_string_append_printf(buf, "XIVE[%x] #%d Source %08x .. %08x\n",
1861                            chip_id, blk, srcno0, srcno0 + nr_ipis - 1);
1862     xive_source_pic_print_info(&xive->ipi_source, srcno0, buf);
1863 
1864     g_string_append_printf(buf, "XIVE[%x] #%d EAT %08x .. %08x\n",
1865                            chip_id, blk, srcno0, srcno0 + nr_ipis - 1);
1866     for (i = 0; i < nr_ipis; i++) {
1867         if (xive_router_get_eas(xrtr, blk, i, &eas)) {
1868             break;
1869         }
1870         if (!xive_eas_is_masked(&eas)) {
1871             xive_eas_pic_print_info(&eas, i, buf);
1872         }
1873     }
1874 
1875     g_string_append_printf(buf, "XIVE[%x] #%d ENDT\n", chip_id, blk);
1876     i = 0;
1877     while (!xive_router_get_end(xrtr, blk, i, &end)) {
1878         xive_end_pic_print_info(&end, i++, buf);
1879     }
1880 
1881     g_string_append_printf(buf, "XIVE[%x] #%d END Escalation EAT\n",
1882                            chip_id, blk);
1883     i = 0;
1884     while (!xive_router_get_end(xrtr, blk, i, &end)) {
1885         xive_end_eas_pic_print_info(&end, i++, buf);
1886     }
1887 
1888     g_string_append_printf(buf, "XIVE[%x] #%d NVTT %08x .. %08x\n",
1889                            chip_id, blk, 0, XIVE_NVT_COUNT - 1);
1890     xive_nvt_per_subpage = pnv_xive_vst_per_subpage(xive, VST_TSEL_VPDT);
1891     for (i = 0; i < XIVE_NVT_COUNT; i += xive_nvt_per_subpage) {
1892         while (!xive_router_get_nvt(xrtr, blk, i, &nvt)) {
1893             xive_nvt_pic_print_info(&nvt, i++, buf);
1894         }
1895     }
1896 }
1897 
1898 static void pnv_xive_reset(void *dev)
1899 {
1900     PnvXive *xive = PNV_XIVE(dev);
1901     XiveSource *xsrc = &xive->ipi_source;
1902     XiveENDSource *end_xsrc = &xive->end_source;
1903 
1904     /* Default page size (Should be changed at runtime to 64k) */
1905     xive->ic_shift = xive->vc_shift = xive->pc_shift = 12;
1906 
1907     /* Clear subregions */
1908     if (memory_region_is_mapped(&xsrc->esb_mmio)) {
1909         memory_region_del_subregion(&xive->ipi_edt_mmio, &xsrc->esb_mmio);
1910     }
1911 
1912     if (memory_region_is_mapped(&xive->ipi_edt_mmio)) {
1913         memory_region_del_subregion(&xive->ipi_mmio, &xive->ipi_edt_mmio);
1914     }
1915 
1916     if (memory_region_is_mapped(&end_xsrc->esb_mmio)) {
1917         memory_region_del_subregion(&xive->end_edt_mmio, &end_xsrc->esb_mmio);
1918     }
1919 
1920     if (memory_region_is_mapped(&xive->end_edt_mmio)) {
1921         memory_region_del_subregion(&xive->end_mmio, &xive->end_edt_mmio);
1922     }
1923 }
1924 
1925 static void pnv_xive_init(Object *obj)
1926 {
1927     PnvXive *xive = PNV_XIVE(obj);
1928 
1929     object_initialize_child(obj, "ipi_source", &xive->ipi_source,
1930                             TYPE_XIVE_SOURCE);
1931     object_initialize_child(obj, "end_source", &xive->end_source,
1932                             TYPE_XIVE_END_SOURCE);
1933 }
1934 
1935 /*
1936  *  Maximum number of IRQs and ENDs supported by HW
1937  */
1938 #define PNV_XIVE_NR_IRQS (PNV9_XIVE_VC_SIZE / (1ull << XIVE_ESB_64K_2PAGE))
1939 #define PNV_XIVE_NR_ENDS (PNV9_XIVE_VC_SIZE / (1ull << XIVE_ESB_64K_2PAGE))
1940 
1941 static void pnv_xive_realize(DeviceState *dev, Error **errp)
1942 {
1943     PnvXive *xive = PNV_XIVE(dev);
1944     PnvXiveClass *pxc = PNV_XIVE_GET_CLASS(dev);
1945     XiveSource *xsrc = &xive->ipi_source;
1946     XiveENDSource *end_xsrc = &xive->end_source;
1947     Error *local_err = NULL;
1948 
1949     pxc->parent_realize(dev, &local_err);
1950     if (local_err) {
1951         error_propagate(errp, local_err);
1952         return;
1953     }
1954 
1955     assert(xive->chip);
1956 
1957     /*
1958      * The XiveSource and XiveENDSource objects are realized with the
1959      * maximum allowed HW configuration. The ESB MMIO regions will be
1960      * resized dynamically when the controller is configured by the FW
1961      * to limit accesses to resources not provisioned.
1962      */
1963     object_property_set_int(OBJECT(xsrc), "nr-irqs", PNV_XIVE_NR_IRQS,
1964                             &error_fatal);
1965     object_property_set_link(OBJECT(xsrc), "xive", OBJECT(xive), &error_abort);
1966     if (!qdev_realize(DEVICE(xsrc), NULL, errp)) {
1967         return;
1968     }
1969 
1970     object_property_set_int(OBJECT(end_xsrc), "nr-ends", PNV_XIVE_NR_ENDS,
1971                             &error_fatal);
1972     object_property_set_link(OBJECT(end_xsrc), "xive", OBJECT(xive),
1973                              &error_abort);
1974     if (!qdev_realize(DEVICE(end_xsrc), NULL, errp)) {
1975         return;
1976     }
1977 
1978     /* Default page size. Generally changed at runtime to 64k */
1979     xive->ic_shift = xive->vc_shift = xive->pc_shift = 12;
1980 
1981     /* XSCOM region, used for initial configuration of the BARs */
1982     memory_region_init_io(&xive->xscom_regs, OBJECT(dev), &pnv_xive_xscom_ops,
1983                           xive, "xscom-xive", PNV9_XSCOM_XIVE_SIZE << 3);
1984 
1985     /* Interrupt controller MMIO regions */
1986     memory_region_init(&xive->ic_mmio, OBJECT(dev), "xive-ic",
1987                        PNV9_XIVE_IC_SIZE);
1988 
1989     memory_region_init_io(&xive->ic_reg_mmio, OBJECT(dev), &pnv_xive_ic_reg_ops,
1990                           xive, "xive-ic-reg", 1 << xive->ic_shift);
1991     memory_region_init_io(&xive->ic_notify_mmio, OBJECT(dev),
1992                           &pnv_xive_ic_notify_ops,
1993                           xive, "xive-ic-notify", 1 << xive->ic_shift);
1994     xive->ic_notify_mmio.disable_reentrancy_guard = true;
1995 
1996     /* The Pervasive LSI trigger and EOI pages (not modeled) */
1997     memory_region_init_io(&xive->ic_lsi_mmio, OBJECT(dev), &pnv_xive_ic_lsi_ops,
1998                           xive, "xive-ic-lsi", 2 << xive->ic_shift);
1999 
2000     /* Thread Interrupt Management Area (Indirect) */
2001     memory_region_init_io(&xive->tm_indirect_mmio, OBJECT(dev),
2002                           &xive_tm_indirect_ops,
2003                           xive, "xive-tima-indirect", PNV9_XIVE_TM_SIZE);
2004     /*
2005      * Overall Virtualization Controller MMIO region containing the
2006      * IPI ESB pages and END ESB pages. The layout is defined by the
2007      * EDT "Domain table" and the accesses are dispatched using
2008      * address spaces for each.
2009      */
2010     memory_region_init_io(&xive->vc_mmio, OBJECT(xive), &pnv_xive_vc_ops, xive,
2011                           "xive-vc", PNV9_XIVE_VC_SIZE);
2012 
2013     memory_region_init(&xive->ipi_mmio, OBJECT(xive), "xive-vc-ipi",
2014                        PNV9_XIVE_VC_SIZE);
2015     address_space_init(&xive->ipi_as, &xive->ipi_mmio, "xive-vc-ipi");
2016     memory_region_init(&xive->end_mmio, OBJECT(xive), "xive-vc-end",
2017                        PNV9_XIVE_VC_SIZE);
2018     address_space_init(&xive->end_as, &xive->end_mmio, "xive-vc-end");
2019 
2020     /*
2021      * The MMIO windows exposing the IPI ESBs and the END ESBs in the
2022      * VC region. Their size is configured by the FW in the EDT table.
2023      */
2024     memory_region_init(&xive->ipi_edt_mmio, OBJECT(xive), "xive-vc-ipi-edt", 0);
2025     memory_region_init(&xive->end_edt_mmio, OBJECT(xive), "xive-vc-end-edt", 0);
2026 
2027     /* Presenter Controller MMIO region (not modeled) */
2028     memory_region_init_io(&xive->pc_mmio, OBJECT(xive), &pnv_xive_pc_ops, xive,
2029                           "xive-pc", PNV9_XIVE_PC_SIZE);
2030     xive->pc_mmio.disable_reentrancy_guard = true;
2031 
2032     /* Thread Interrupt Management Area (Direct) */
2033     memory_region_init_io(&xive->tm_mmio, OBJECT(xive), &pnv_xive_tm_ops,
2034                           xive, "xive-tima", PNV9_XIVE_TM_SIZE);
2035 
2036     qemu_register_reset(pnv_xive_reset, dev);
2037 }
2038 
2039 static int pnv_xive_dt_xscom(PnvXScomInterface *dev, void *fdt,
2040                              int xscom_offset)
2041 {
2042     const char compat[] = "ibm,power9-xive-x";
2043     char *name;
2044     int offset;
2045     uint32_t lpc_pcba = PNV9_XSCOM_XIVE_BASE;
2046     uint32_t reg[] = {
2047         cpu_to_be32(lpc_pcba),
2048         cpu_to_be32(PNV9_XSCOM_XIVE_SIZE)
2049     };
2050 
2051     name = g_strdup_printf("xive@%x", lpc_pcba);
2052     offset = fdt_add_subnode(fdt, xscom_offset, name);
2053     _FDT(offset);
2054     g_free(name);
2055 
2056     _FDT((fdt_setprop(fdt, offset, "reg", reg, sizeof(reg))));
2057     _FDT((fdt_setprop(fdt, offset, "compatible", compat,
2058                       sizeof(compat))));
2059     return 0;
2060 }
2061 
2062 static Property pnv_xive_properties[] = {
2063     DEFINE_PROP_UINT64("ic-bar", PnvXive, ic_base, 0),
2064     DEFINE_PROP_UINT64("vc-bar", PnvXive, vc_base, 0),
2065     DEFINE_PROP_UINT64("pc-bar", PnvXive, pc_base, 0),
2066     DEFINE_PROP_UINT64("tm-bar", PnvXive, tm_base, 0),
2067     /* The PnvChip id identifies the XIVE interrupt controller. */
2068     DEFINE_PROP_LINK("chip", PnvXive, chip, TYPE_PNV_CHIP, PnvChip *),
2069     DEFINE_PROP_END_OF_LIST(),
2070 };
2071 
2072 static void pnv_xive_class_init(ObjectClass *klass, void *data)
2073 {
2074     DeviceClass *dc = DEVICE_CLASS(klass);
2075     PnvXScomInterfaceClass *xdc = PNV_XSCOM_INTERFACE_CLASS(klass);
2076     XiveRouterClass *xrc = XIVE_ROUTER_CLASS(klass);
2077     XiveNotifierClass *xnc = XIVE_NOTIFIER_CLASS(klass);
2078     XivePresenterClass *xpc = XIVE_PRESENTER_CLASS(klass);
2079     PnvXiveClass *pxc = PNV_XIVE_CLASS(klass);
2080 
2081     xdc->dt_xscom = pnv_xive_dt_xscom;
2082 
2083     dc->desc = "PowerNV XIVE Interrupt Controller";
2084     device_class_set_parent_realize(dc, pnv_xive_realize, &pxc->parent_realize);
2085     dc->realize = pnv_xive_realize;
2086     device_class_set_props(dc, pnv_xive_properties);
2087 
2088     xrc->get_eas = pnv_xive_get_eas;
2089     xrc->get_pq = pnv_xive_get_pq;
2090     xrc->set_pq = pnv_xive_set_pq;
2091     xrc->get_end = pnv_xive_get_end;
2092     xrc->write_end = pnv_xive_write_end;
2093     xrc->get_nvt = pnv_xive_get_nvt;
2094     xrc->write_nvt = pnv_xive_write_nvt;
2095     xrc->get_block_id = pnv_xive_get_block_id;
2096     xrc->end_notify = pnv_xive_end_notify;
2097 
2098     xnc->notify = pnv_xive_notify;
2099     xpc->match_nvt  = pnv_xive_match_nvt;
2100     xpc->get_config = pnv_xive_presenter_get_config;
2101 };
2102 
2103 static const TypeInfo pnv_xive_info = {
2104     .name          = TYPE_PNV_XIVE,
2105     .parent        = TYPE_XIVE_ROUTER,
2106     .instance_init = pnv_xive_init,
2107     .instance_size = sizeof(PnvXive),
2108     .class_init    = pnv_xive_class_init,
2109     .class_size    = sizeof(PnvXiveClass),
2110     .interfaces    = (InterfaceInfo[]) {
2111         { TYPE_PNV_XSCOM_INTERFACE },
2112         { }
2113     }
2114 };
2115 
2116 static void pnv_xive_register_types(void)
2117 {
2118     type_register_static(&pnv_xive_info);
2119 }
2120 
2121 type_init(pnv_xive_register_types)
2122