xref: /openbmc/qemu/hw/pci-host/pnv_phb4.c (revision d2704eb3fd5093224fb14ecbfdc916d42d42416b)
1 /*
2  * QEMU PowerPC PowerNV (POWER9) PHB4 model
3  *
4  * Copyright (c) 2018-2020, 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 #include "qemu/osdep.h"
10 #include "qemu/log.h"
11 #include "qapi/visitor.h"
12 #include "qapi/error.h"
13 #include "qemu-common.h"
14 #include "monitor/monitor.h"
15 #include "target/ppc/cpu.h"
16 #include "hw/pci-host/pnv_phb4_regs.h"
17 #include "hw/pci-host/pnv_phb4.h"
18 #include "hw/pci/pcie_host.h"
19 #include "hw/pci/pcie_port.h"
20 #include "hw/ppc/pnv.h"
21 #include "hw/ppc/pnv_xscom.h"
22 #include "hw/irq.h"
23 #include "hw/qdev-properties.h"
24 #include "qom/object.h"
25 #include "sysemu/sysemu.h"
26 #include "trace.h"
27 
28 #define phb_error(phb, fmt, ...)                                        \
29     qemu_log_mask(LOG_GUEST_ERROR, "phb4[%d:%d]: " fmt "\n",            \
30                   (phb)->chip_id, (phb)->phb_id, ## __VA_ARGS__)
31 
32 #define phb_pec_error(pec, fmt, ...)                                    \
33     qemu_log_mask(LOG_GUEST_ERROR, "phb4_pec[%d:%d]: " fmt "\n",        \
34                   (pec)->chip_id, (pec)->index, ## __VA_ARGS__)
35 
36 /*
37  * QEMU version of the GETFIELD/SETFIELD macros
38  *
39  * These are common with the PnvXive model.
40  */
41 static inline uint64_t GETFIELD(uint64_t mask, uint64_t word)
42 {
43     return (word & mask) >> ctz64(mask);
44 }
45 
46 static inline uint64_t SETFIELD(uint64_t mask, uint64_t word,
47                                 uint64_t value)
48 {
49     return (word & ~mask) | ((value << ctz64(mask)) & mask);
50 }
51 
52 static PCIDevice *pnv_phb4_find_cfg_dev(PnvPHB4 *phb)
53 {
54     PCIHostState *pci = PCI_HOST_BRIDGE(phb);
55     uint64_t addr = phb->regs[PHB_CONFIG_ADDRESS >> 3];
56     uint8_t bus, devfn;
57 
58     if (!(addr >> 63)) {
59         return NULL;
60     }
61     bus = (addr >> 52) & 0xff;
62     devfn = (addr >> 44) & 0xff;
63 
64     /* We don't access the root complex this way */
65     if (bus == 0 && devfn == 0) {
66         return NULL;
67     }
68     return pci_find_device(pci->bus, bus, devfn);
69 }
70 
71 /*
72  * The CONFIG_DATA register expects little endian accesses, but as the
73  * region is big endian, we have to swap the value.
74  */
75 static void pnv_phb4_config_write(PnvPHB4 *phb, unsigned off,
76                                   unsigned size, uint64_t val)
77 {
78     uint32_t cfg_addr, limit;
79     PCIDevice *pdev;
80 
81     pdev = pnv_phb4_find_cfg_dev(phb);
82     if (!pdev) {
83         return;
84     }
85     cfg_addr = (phb->regs[PHB_CONFIG_ADDRESS >> 3] >> 32) & 0xffc;
86     cfg_addr |= off;
87     limit = pci_config_size(pdev);
88     if (limit <= cfg_addr) {
89         /*
90          * conventional pci device can be behind pcie-to-pci bridge.
91          * 256 <= addr < 4K has no effects.
92          */
93         return;
94     }
95     switch (size) {
96     case 1:
97         break;
98     case 2:
99         val = bswap16(val);
100         break;
101     case 4:
102         val = bswap32(val);
103         break;
104     default:
105         g_assert_not_reached();
106     }
107     pci_host_config_write_common(pdev, cfg_addr, limit, val, size);
108 }
109 
110 static uint64_t pnv_phb4_config_read(PnvPHB4 *phb, unsigned off,
111                                      unsigned size)
112 {
113     uint32_t cfg_addr, limit;
114     PCIDevice *pdev;
115     uint64_t val;
116 
117     pdev = pnv_phb4_find_cfg_dev(phb);
118     if (!pdev) {
119         return ~0ull;
120     }
121     cfg_addr = (phb->regs[PHB_CONFIG_ADDRESS >> 3] >> 32) & 0xffc;
122     cfg_addr |= off;
123     limit = pci_config_size(pdev);
124     if (limit <= cfg_addr) {
125         /*
126          * conventional pci device can be behind pcie-to-pci bridge.
127          * 256 <= addr < 4K has no effects.
128          */
129         return ~0ull;
130     }
131     val = pci_host_config_read_common(pdev, cfg_addr, limit, size);
132     switch (size) {
133     case 1:
134         return val;
135     case 2:
136         return bswap16(val);
137     case 4:
138         return bswap32(val);
139     default:
140         g_assert_not_reached();
141     }
142 }
143 
144 /*
145  * Root complex register accesses are memory mapped.
146  */
147 static void pnv_phb4_rc_config_write(PnvPHB4 *phb, unsigned off,
148                                      unsigned size, uint64_t val)
149 {
150     PCIHostState *pci = PCI_HOST_BRIDGE(phb);
151     PCIDevice *pdev;
152 
153     if (size != 4) {
154         phb_error(phb, "rc_config_write invalid size %d\n", size);
155         return;
156     }
157 
158     pdev = pci_find_device(pci->bus, 0, 0);
159     if (!pdev) {
160         phb_error(phb, "rc_config_write device not found\n");
161         return;
162     }
163 
164     pci_host_config_write_common(pdev, off, PHB_RC_CONFIG_SIZE,
165                                  bswap32(val), 4);
166 }
167 
168 static uint64_t pnv_phb4_rc_config_read(PnvPHB4 *phb, unsigned off,
169                                         unsigned size)
170 {
171     PCIHostState *pci = PCI_HOST_BRIDGE(phb);
172     PCIDevice *pdev;
173     uint64_t val;
174 
175     if (size != 4) {
176         phb_error(phb, "rc_config_read invalid size %d\n", size);
177         return ~0ull;
178     }
179 
180     pdev = pci_find_device(pci->bus, 0, 0);
181     if (!pdev) {
182         phb_error(phb, "rc_config_read device not found\n");
183         return ~0ull;
184     }
185 
186     val = pci_host_config_read_common(pdev, off, PHB_RC_CONFIG_SIZE, 4);
187     return bswap32(val);
188 }
189 
190 static void pnv_phb4_check_mbt(PnvPHB4 *phb, uint32_t index)
191 {
192     uint64_t base, start, size, mbe0, mbe1;
193     MemoryRegion *parent;
194     char name[64];
195 
196     /* Unmap first */
197     if (memory_region_is_mapped(&phb->mr_mmio[index])) {
198         /* Should we destroy it in RCU friendly way... ? */
199         memory_region_del_subregion(phb->mr_mmio[index].container,
200                                     &phb->mr_mmio[index]);
201     }
202 
203     /* Get table entry */
204     mbe0 = phb->ioda_MBT[(index << 1)];
205     mbe1 = phb->ioda_MBT[(index << 1) + 1];
206 
207     if (!(mbe0 & IODA3_MBT0_ENABLE)) {
208         return;
209     }
210 
211     /* Grab geometry from registers */
212     base = GETFIELD(IODA3_MBT0_BASE_ADDR, mbe0) << 12;
213     size = GETFIELD(IODA3_MBT1_MASK, mbe1) << 12;
214     size |= 0xff00000000000000ull;
215     size = ~size + 1;
216 
217     /* Calculate PCI side start address based on M32/M64 window type */
218     if (mbe0 & IODA3_MBT0_TYPE_M32) {
219         start = phb->regs[PHB_M32_START_ADDR >> 3];
220         if ((start + size) > 0x100000000ull) {
221             phb_error(phb, "M32 set beyond 4GB boundary !");
222             size = 0x100000000 - start;
223         }
224     } else {
225         start = base | (phb->regs[PHB_M64_UPPER_BITS >> 3]);
226     }
227 
228     /* TODO: Figure out how to implemet/decode AOMASK */
229 
230     /* Check if it matches an enabled MMIO region in the PEC stack */
231     if (memory_region_is_mapped(&phb->mmbar0) &&
232         base >= phb->mmio0_base &&
233         (base + size) <= (phb->mmio0_base + phb->mmio0_size)) {
234         parent = &phb->mmbar0;
235         base -= phb->mmio0_base;
236     } else if (memory_region_is_mapped(&phb->mmbar1) &&
237         base >= phb->mmio1_base &&
238         (base + size) <= (phb->mmio1_base + phb->mmio1_size)) {
239         parent = &phb->mmbar1;
240         base -= phb->mmio1_base;
241     } else {
242         phb_error(phb, "PHB MBAR %d out of parent bounds", index);
243         return;
244     }
245 
246     /* Create alias (better name ?) */
247     snprintf(name, sizeof(name), "phb4-mbar%d", index);
248     memory_region_init_alias(&phb->mr_mmio[index], OBJECT(phb), name,
249                              &phb->pci_mmio, start, size);
250     memory_region_add_subregion(parent, base, &phb->mr_mmio[index]);
251 }
252 
253 static void pnv_phb4_check_all_mbt(PnvPHB4 *phb)
254 {
255     uint64_t i;
256     uint32_t num_windows = phb->big_phb ? PNV_PHB4_MAX_MMIO_WINDOWS :
257         PNV_PHB4_MIN_MMIO_WINDOWS;
258 
259     for (i = 0; i < num_windows; i++) {
260         pnv_phb4_check_mbt(phb, i);
261     }
262 }
263 
264 static uint64_t *pnv_phb4_ioda_access(PnvPHB4 *phb,
265                                       unsigned *out_table, unsigned *out_idx)
266 {
267     uint64_t adreg = phb->regs[PHB_IODA_ADDR >> 3];
268     unsigned int index = GETFIELD(PHB_IODA_AD_TADR, adreg);
269     unsigned int table = GETFIELD(PHB_IODA_AD_TSEL, adreg);
270     unsigned int mask;
271     uint64_t *tptr = NULL;
272 
273     switch (table) {
274     case IODA3_TBL_LIST:
275         tptr = phb->ioda_LIST;
276         mask = 7;
277         break;
278     case IODA3_TBL_MIST:
279         tptr = phb->ioda_MIST;
280         mask = phb->big_phb ? PNV_PHB4_MAX_MIST : (PNV_PHB4_MAX_MIST >> 1);
281         mask -= 1;
282         break;
283     case IODA3_TBL_RCAM:
284         mask = phb->big_phb ? 127 : 63;
285         break;
286     case IODA3_TBL_MRT:
287         mask = phb->big_phb ? 15 : 7;
288         break;
289     case IODA3_TBL_PESTA:
290     case IODA3_TBL_PESTB:
291         mask = phb->big_phb ? PNV_PHB4_MAX_PEs : (PNV_PHB4_MAX_PEs >> 1);
292         mask -= 1;
293         break;
294     case IODA3_TBL_TVT:
295         tptr = phb->ioda_TVT;
296         mask = phb->big_phb ? PNV_PHB4_MAX_TVEs : (PNV_PHB4_MAX_TVEs >> 1);
297         mask -= 1;
298         break;
299     case IODA3_TBL_TCR:
300     case IODA3_TBL_TDR:
301         mask = phb->big_phb ? 1023 : 511;
302         break;
303     case IODA3_TBL_MBT:
304         tptr = phb->ioda_MBT;
305         mask = phb->big_phb ? PNV_PHB4_MAX_MBEs : (PNV_PHB4_MAX_MBEs >> 1);
306         mask -= 1;
307         break;
308     case IODA3_TBL_MDT:
309         tptr = phb->ioda_MDT;
310         mask = phb->big_phb ? PNV_PHB4_MAX_PEs : (PNV_PHB4_MAX_PEs >> 1);
311         mask -= 1;
312         break;
313     case IODA3_TBL_PEEV:
314         tptr = phb->ioda_PEEV;
315         mask = phb->big_phb ? PNV_PHB4_MAX_PEEVs : (PNV_PHB4_MAX_PEEVs >> 1);
316         mask -= 1;
317         break;
318     default:
319         phb_error(phb, "invalid IODA table %d", table);
320         return NULL;
321     }
322     index &= mask;
323     if (out_idx) {
324         *out_idx = index;
325     }
326     if (out_table) {
327         *out_table = table;
328     }
329     if (tptr) {
330         tptr += index;
331     }
332     if (adreg & PHB_IODA_AD_AUTOINC) {
333         index = (index + 1) & mask;
334         adreg = SETFIELD(PHB_IODA_AD_TADR, adreg, index);
335     }
336 
337     phb->regs[PHB_IODA_ADDR >> 3] = adreg;
338     return tptr;
339 }
340 
341 static uint64_t pnv_phb4_ioda_read(PnvPHB4 *phb)
342 {
343     unsigned table, idx;
344     uint64_t *tptr;
345 
346     tptr = pnv_phb4_ioda_access(phb, &table, &idx);
347     if (!tptr) {
348         /* Special PESTA case */
349         if (table == IODA3_TBL_PESTA) {
350             return ((uint64_t)(phb->ioda_PEST_AB[idx] & 1)) << 63;
351         } else if (table == IODA3_TBL_PESTB) {
352             return ((uint64_t)(phb->ioda_PEST_AB[idx] & 2)) << 62;
353         }
354         /* Return 0 on unsupported tables, not ff's */
355         return 0;
356     }
357     return *tptr;
358 }
359 
360 static void pnv_phb4_ioda_write(PnvPHB4 *phb, uint64_t val)
361 {
362     unsigned table, idx;
363     uint64_t *tptr;
364 
365     tptr = pnv_phb4_ioda_access(phb, &table, &idx);
366     if (!tptr) {
367         /* Special PESTA case */
368         if (table == IODA3_TBL_PESTA) {
369             phb->ioda_PEST_AB[idx] &= ~1;
370             phb->ioda_PEST_AB[idx] |= (val >> 63) & 1;
371         } else if (table == IODA3_TBL_PESTB) {
372             phb->ioda_PEST_AB[idx] &= ~2;
373             phb->ioda_PEST_AB[idx] |= (val >> 62) & 2;
374         }
375         return;
376     }
377 
378     /* Handle side effects */
379     switch (table) {
380     case IODA3_TBL_LIST:
381         break;
382     case IODA3_TBL_MIST: {
383         /* Special mask for MIST partial write */
384         uint64_t adreg = phb->regs[PHB_IODA_ADDR >> 3];
385         uint32_t mmask = GETFIELD(PHB_IODA_AD_MIST_PWV, adreg);
386         uint64_t v = *tptr;
387         if (mmask == 0) {
388             mmask = 0xf;
389         }
390         if (mmask & 8) {
391             v &= 0x0000ffffffffffffull;
392             v |= 0xcfff000000000000ull & val;
393         }
394         if (mmask & 4) {
395             v &= 0xffff0000ffffffffull;
396             v |= 0x0000cfff00000000ull & val;
397         }
398         if (mmask & 2) {
399             v &= 0xffffffff0000ffffull;
400             v |= 0x00000000cfff0000ull & val;
401         }
402         if (mmask & 1) {
403             v &= 0xffffffffffff0000ull;
404             v |= 0x000000000000cfffull & val;
405         }
406         *tptr = v;
407         break;
408     }
409     case IODA3_TBL_MBT:
410         *tptr = val;
411 
412         /* Copy accross the valid bit to the other half */
413         phb->ioda_MBT[idx ^ 1] &= 0x7fffffffffffffffull;
414         phb->ioda_MBT[idx ^ 1] |= 0x8000000000000000ull & val;
415 
416         /* Update mappings */
417         pnv_phb4_check_mbt(phb, idx >> 1);
418         break;
419     default:
420         *tptr = val;
421     }
422 }
423 
424 static void pnv_phb4_rtc_invalidate(PnvPHB4 *phb, uint64_t val)
425 {
426     PnvPhb4DMASpace *ds;
427 
428     /* Always invalidate all for now ... */
429     QLIST_FOREACH(ds, &phb->dma_spaces, list) {
430         ds->pe_num = PHB_INVALID_PE;
431     }
432 }
433 
434 static void pnv_phb4_update_msi_regions(PnvPhb4DMASpace *ds)
435 {
436     uint64_t cfg = ds->phb->regs[PHB_PHB4_CONFIG >> 3];
437 
438     if (cfg & PHB_PHB4C_32BIT_MSI_EN) {
439         if (!memory_region_is_mapped(MEMORY_REGION(&ds->msi32_mr))) {
440             memory_region_add_subregion(MEMORY_REGION(&ds->dma_mr),
441                                         0xffff0000, &ds->msi32_mr);
442         }
443     } else {
444         if (memory_region_is_mapped(MEMORY_REGION(&ds->msi32_mr))) {
445             memory_region_del_subregion(MEMORY_REGION(&ds->dma_mr),
446                                         &ds->msi32_mr);
447         }
448     }
449 
450     if (cfg & PHB_PHB4C_64BIT_MSI_EN) {
451         if (!memory_region_is_mapped(MEMORY_REGION(&ds->msi64_mr))) {
452             memory_region_add_subregion(MEMORY_REGION(&ds->dma_mr),
453                                         (1ull << 60), &ds->msi64_mr);
454         }
455     } else {
456         if (memory_region_is_mapped(MEMORY_REGION(&ds->msi64_mr))) {
457             memory_region_del_subregion(MEMORY_REGION(&ds->dma_mr),
458                                         &ds->msi64_mr);
459         }
460     }
461 }
462 
463 static void pnv_phb4_update_all_msi_regions(PnvPHB4 *phb)
464 {
465     PnvPhb4DMASpace *ds;
466 
467     QLIST_FOREACH(ds, &phb->dma_spaces, list) {
468         pnv_phb4_update_msi_regions(ds);
469     }
470 }
471 
472 static void pnv_phb4_update_xsrc(PnvPHB4 *phb)
473 {
474     int shift, flags, i, lsi_base;
475     XiveSource *xsrc = &phb->xsrc;
476 
477     /* The XIVE source characteristics can be set at run time */
478     if (phb->regs[PHB_CTRLR >> 3] & PHB_CTRLR_IRQ_PGSZ_64K) {
479         shift = XIVE_ESB_64K;
480     } else {
481         shift = XIVE_ESB_4K;
482     }
483     if (phb->regs[PHB_CTRLR >> 3] & PHB_CTRLR_IRQ_STORE_EOI) {
484         flags = XIVE_SRC_STORE_EOI;
485     } else {
486         flags = 0;
487     }
488 
489     phb->xsrc.esb_shift = shift;
490     phb->xsrc.esb_flags = flags;
491 
492     lsi_base = GETFIELD(PHB_LSI_SRC_ID, phb->regs[PHB_LSI_SOURCE_ID >> 3]);
493     lsi_base <<= 3;
494 
495     /* TODO: handle reset values of PHB_LSI_SRC_ID */
496     if (!lsi_base) {
497         return;
498     }
499 
500     /* TODO: need a xive_source_irq_reset_lsi() */
501     bitmap_zero(xsrc->lsi_map, xsrc->nr_irqs);
502 
503     for (i = 0; i < xsrc->nr_irqs; i++) {
504         bool msi = (i < lsi_base || i >= (lsi_base + 8));
505         if (!msi) {
506             xive_source_irq_set_lsi(xsrc, i);
507         }
508     }
509 }
510 
511 static void pnv_phb4_reg_write(void *opaque, hwaddr off, uint64_t val,
512                                unsigned size)
513 {
514     PnvPHB4 *phb = PNV_PHB4(opaque);
515     bool changed;
516 
517     /* Special case outbound configuration data */
518     if ((off & 0xfffc) == PHB_CONFIG_DATA) {
519         pnv_phb4_config_write(phb, off & 0x3, size, val);
520         return;
521     }
522 
523     /* Special case RC configuration space */
524     if ((off & 0xf800) == PHB_RC_CONFIG_BASE) {
525         pnv_phb4_rc_config_write(phb, off & 0x7ff, size, val);
526         return;
527     }
528 
529     /* Other registers are 64-bit only */
530     if (size != 8 || off & 0x7) {
531         phb_error(phb, "Invalid register access, offset: 0x%"PRIx64" size: %d",
532                    off, size);
533         return;
534     }
535 
536     /* Handle masking */
537     switch (off) {
538     case PHB_LSI_SOURCE_ID:
539         val &= PHB_LSI_SRC_ID;
540         break;
541     case PHB_M64_UPPER_BITS:
542         val &= 0xff00000000000000ull;
543         break;
544     /* TCE Kill */
545     case PHB_TCE_KILL:
546         /* Clear top 3 bits which HW does to indicate successful queuing */
547         val &= ~(PHB_TCE_KILL_ALL | PHB_TCE_KILL_PE | PHB_TCE_KILL_ONE);
548         break;
549     case PHB_Q_DMA_R:
550         /*
551          * This is enough logic to make SW happy but we aren't
552          * actually quiescing the DMAs
553          */
554         if (val & PHB_Q_DMA_R_AUTORESET) {
555             val = 0;
556         } else {
557             val &= PHB_Q_DMA_R_QUIESCE_DMA;
558         }
559         break;
560     /* LEM stuff */
561     case PHB_LEM_FIR_AND_MASK:
562         phb->regs[PHB_LEM_FIR_ACCUM >> 3] &= val;
563         return;
564     case PHB_LEM_FIR_OR_MASK:
565         phb->regs[PHB_LEM_FIR_ACCUM >> 3] |= val;
566         return;
567     case PHB_LEM_ERROR_AND_MASK:
568         phb->regs[PHB_LEM_ERROR_MASK >> 3] &= val;
569         return;
570     case PHB_LEM_ERROR_OR_MASK:
571         phb->regs[PHB_LEM_ERROR_MASK >> 3] |= val;
572         return;
573     case PHB_LEM_WOF:
574         val = 0;
575         break;
576     /* TODO: More regs ..., maybe create a table with masks... */
577 
578     /* Read only registers */
579     case PHB_CPU_LOADSTORE_STATUS:
580     case PHB_ETU_ERR_SUMMARY:
581     case PHB_PHB4_GEN_CAP:
582     case PHB_PHB4_TCE_CAP:
583     case PHB_PHB4_IRQ_CAP:
584     case PHB_PHB4_EEH_CAP:
585         return;
586     }
587 
588     /* Record whether it changed */
589     changed = phb->regs[off >> 3] != val;
590 
591     /* Store in register cache first */
592     phb->regs[off >> 3] = val;
593 
594     /* Handle side effects */
595     switch (off) {
596     case PHB_PHB4_CONFIG:
597         if (changed) {
598             pnv_phb4_update_all_msi_regions(phb);
599         }
600         break;
601     case PHB_M32_START_ADDR:
602     case PHB_M64_UPPER_BITS:
603         if (changed) {
604             pnv_phb4_check_all_mbt(phb);
605         }
606         break;
607 
608     /* IODA table accesses */
609     case PHB_IODA_DATA0:
610         pnv_phb4_ioda_write(phb, val);
611         break;
612 
613     /* RTC invalidation */
614     case PHB_RTC_INVALIDATE:
615         pnv_phb4_rtc_invalidate(phb, val);
616         break;
617 
618     /* PHB Control (Affects XIVE source) */
619     case PHB_CTRLR:
620     case PHB_LSI_SOURCE_ID:
621         pnv_phb4_update_xsrc(phb);
622         break;
623 
624     /* Silent simple writes */
625     case PHB_ASN_CMPM:
626     case PHB_CONFIG_ADDRESS:
627     case PHB_IODA_ADDR:
628     case PHB_TCE_KILL:
629     case PHB_TCE_SPEC_CTL:
630     case PHB_PEST_BAR:
631     case PHB_PELTV_BAR:
632     case PHB_RTT_BAR:
633     case PHB_LEM_FIR_ACCUM:
634     case PHB_LEM_ERROR_MASK:
635     case PHB_LEM_ACTION0:
636     case PHB_LEM_ACTION1:
637     case PHB_TCE_TAG_ENABLE:
638     case PHB_INT_NOTIFY_ADDR:
639     case PHB_INT_NOTIFY_INDEX:
640     case PHB_DMARD_SYNC:
641        break;
642 
643     /* Noise on anything else */
644     default:
645         qemu_log_mask(LOG_UNIMP, "phb4: reg_write 0x%"PRIx64"=%"PRIx64"\n",
646                       off, val);
647     }
648 }
649 
650 static uint64_t pnv_phb4_reg_read(void *opaque, hwaddr off, unsigned size)
651 {
652     PnvPHB4 *phb = PNV_PHB4(opaque);
653     uint64_t val;
654 
655     if ((off & 0xfffc) == PHB_CONFIG_DATA) {
656         return pnv_phb4_config_read(phb, off & 0x3, size);
657     }
658 
659     /* Special case RC configuration space */
660     if ((off & 0xf800) == PHB_RC_CONFIG_BASE) {
661         return pnv_phb4_rc_config_read(phb, off & 0x7ff, size);
662     }
663 
664     /* Other registers are 64-bit only */
665     if (size != 8 || off & 0x7) {
666         phb_error(phb, "Invalid register access, offset: 0x%"PRIx64" size: %d",
667                    off, size);
668         return ~0ull;
669     }
670 
671     /* Default read from cache */
672     val = phb->regs[off >> 3];
673 
674     switch (off) {
675     case PHB_VERSION:
676         return phb->version;
677 
678         /* Read-only */
679     case PHB_PHB4_GEN_CAP:
680         return 0xe4b8000000000000ull;
681     case PHB_PHB4_TCE_CAP:
682         return phb->big_phb ? 0x4008440000000400ull : 0x2008440000000200ull;
683     case PHB_PHB4_IRQ_CAP:
684         return phb->big_phb ? 0x0800000000001000ull : 0x0800000000000800ull;
685     case PHB_PHB4_EEH_CAP:
686         return phb->big_phb ? 0x2000000000000000ull : 0x1000000000000000ull;
687 
688     /* IODA table accesses */
689     case PHB_IODA_DATA0:
690         return pnv_phb4_ioda_read(phb);
691 
692     /* Link training always appears trained */
693     case PHB_PCIE_DLP_TRAIN_CTL:
694         /* TODO: Do something sensible with speed ? */
695         return PHB_PCIE_DLP_INBAND_PRESENCE | PHB_PCIE_DLP_TL_LINKACT;
696 
697     /* DMA read sync: make it look like it's complete */
698     case PHB_DMARD_SYNC:
699         return PHB_DMARD_SYNC_COMPLETE;
700 
701     /* Silent simple reads */
702     case PHB_LSI_SOURCE_ID:
703     case PHB_CPU_LOADSTORE_STATUS:
704     case PHB_ASN_CMPM:
705     case PHB_PHB4_CONFIG:
706     case PHB_M32_START_ADDR:
707     case PHB_CONFIG_ADDRESS:
708     case PHB_IODA_ADDR:
709     case PHB_RTC_INVALIDATE:
710     case PHB_TCE_KILL:
711     case PHB_TCE_SPEC_CTL:
712     case PHB_PEST_BAR:
713     case PHB_PELTV_BAR:
714     case PHB_RTT_BAR:
715     case PHB_M64_UPPER_BITS:
716     case PHB_CTRLR:
717     case PHB_LEM_FIR_ACCUM:
718     case PHB_LEM_ERROR_MASK:
719     case PHB_LEM_ACTION0:
720     case PHB_LEM_ACTION1:
721     case PHB_TCE_TAG_ENABLE:
722     case PHB_INT_NOTIFY_ADDR:
723     case PHB_INT_NOTIFY_INDEX:
724     case PHB_Q_DMA_R:
725     case PHB_ETU_ERR_SUMMARY:
726         break;
727 
728     /* Noise on anything else */
729     default:
730         qemu_log_mask(LOG_UNIMP, "phb4: reg_read 0x%"PRIx64"=%"PRIx64"\n",
731                       off, val);
732     }
733     return val;
734 }
735 
736 static const MemoryRegionOps pnv_phb4_reg_ops = {
737     .read = pnv_phb4_reg_read,
738     .write = pnv_phb4_reg_write,
739     .valid.min_access_size = 1,
740     .valid.max_access_size = 8,
741     .impl.min_access_size = 1,
742     .impl.max_access_size = 8,
743     .endianness = DEVICE_BIG_ENDIAN,
744 };
745 
746 static uint64_t pnv_phb4_xscom_read(void *opaque, hwaddr addr, unsigned size)
747 {
748     PnvPHB4 *phb = PNV_PHB4(opaque);
749     uint32_t reg = addr >> 3;
750     uint64_t val;
751     hwaddr offset;
752 
753     switch (reg) {
754     case PHB_SCOM_HV_IND_ADDR:
755         return phb->scom_hv_ind_addr_reg;
756 
757     case PHB_SCOM_HV_IND_DATA:
758         if (!(phb->scom_hv_ind_addr_reg & PHB_SCOM_HV_IND_ADDR_VALID)) {
759             phb_error(phb, "Invalid indirect address");
760             return ~0ull;
761         }
762         size = (phb->scom_hv_ind_addr_reg & PHB_SCOM_HV_IND_ADDR_4B) ? 4 : 8;
763         offset = GETFIELD(PHB_SCOM_HV_IND_ADDR_ADDR, phb->scom_hv_ind_addr_reg);
764         val = pnv_phb4_reg_read(phb, offset, size);
765         if (phb->scom_hv_ind_addr_reg & PHB_SCOM_HV_IND_ADDR_AUTOINC) {
766             offset += size;
767             offset &= 0x3fff;
768             phb->scom_hv_ind_addr_reg = SETFIELD(PHB_SCOM_HV_IND_ADDR_ADDR,
769                                                  phb->scom_hv_ind_addr_reg,
770                                                  offset);
771         }
772         return val;
773     case PHB_SCOM_ETU_LEM_FIR:
774     case PHB_SCOM_ETU_LEM_FIR_AND:
775     case PHB_SCOM_ETU_LEM_FIR_OR:
776     case PHB_SCOM_ETU_LEM_FIR_MSK:
777     case PHB_SCOM_ETU_LEM_ERR_MSK_AND:
778     case PHB_SCOM_ETU_LEM_ERR_MSK_OR:
779     case PHB_SCOM_ETU_LEM_ACT0:
780     case PHB_SCOM_ETU_LEM_ACT1:
781     case PHB_SCOM_ETU_LEM_WOF:
782         offset = ((reg - PHB_SCOM_ETU_LEM_FIR) << 3) + PHB_LEM_FIR_ACCUM;
783         return pnv_phb4_reg_read(phb, offset, size);
784     case PHB_SCOM_ETU_PMON_CONFIG:
785     case PHB_SCOM_ETU_PMON_CTR0:
786     case PHB_SCOM_ETU_PMON_CTR1:
787     case PHB_SCOM_ETU_PMON_CTR2:
788     case PHB_SCOM_ETU_PMON_CTR3:
789         offset = ((reg - PHB_SCOM_ETU_PMON_CONFIG) << 3) + PHB_PERFMON_CONFIG;
790         return pnv_phb4_reg_read(phb, offset, size);
791 
792     default:
793         qemu_log_mask(LOG_UNIMP, "phb4: xscom_read 0x%"HWADDR_PRIx"\n", addr);
794         return ~0ull;
795     }
796 }
797 
798 static void pnv_phb4_xscom_write(void *opaque, hwaddr addr,
799                                  uint64_t val, unsigned size)
800 {
801     PnvPHB4 *phb = PNV_PHB4(opaque);
802     uint32_t reg = addr >> 3;
803     hwaddr offset;
804 
805     switch (reg) {
806     case PHB_SCOM_HV_IND_ADDR:
807         phb->scom_hv_ind_addr_reg = val & 0xe000000000001fff;
808         break;
809     case PHB_SCOM_HV_IND_DATA:
810         if (!(phb->scom_hv_ind_addr_reg & PHB_SCOM_HV_IND_ADDR_VALID)) {
811             phb_error(phb, "Invalid indirect address");
812             break;
813         }
814         size = (phb->scom_hv_ind_addr_reg & PHB_SCOM_HV_IND_ADDR_4B) ? 4 : 8;
815         offset = GETFIELD(PHB_SCOM_HV_IND_ADDR_ADDR, phb->scom_hv_ind_addr_reg);
816         pnv_phb4_reg_write(phb, offset, val, size);
817         if (phb->scom_hv_ind_addr_reg & PHB_SCOM_HV_IND_ADDR_AUTOINC) {
818             offset += size;
819             offset &= 0x3fff;
820             phb->scom_hv_ind_addr_reg = SETFIELD(PHB_SCOM_HV_IND_ADDR_ADDR,
821                                                  phb->scom_hv_ind_addr_reg,
822                                                  offset);
823         }
824         break;
825     case PHB_SCOM_ETU_LEM_FIR:
826     case PHB_SCOM_ETU_LEM_FIR_AND:
827     case PHB_SCOM_ETU_LEM_FIR_OR:
828     case PHB_SCOM_ETU_LEM_FIR_MSK:
829     case PHB_SCOM_ETU_LEM_ERR_MSK_AND:
830     case PHB_SCOM_ETU_LEM_ERR_MSK_OR:
831     case PHB_SCOM_ETU_LEM_ACT0:
832     case PHB_SCOM_ETU_LEM_ACT1:
833     case PHB_SCOM_ETU_LEM_WOF:
834         offset = ((reg - PHB_SCOM_ETU_LEM_FIR) << 3) + PHB_LEM_FIR_ACCUM;
835         pnv_phb4_reg_write(phb, offset, val, size);
836         break;
837     case PHB_SCOM_ETU_PMON_CONFIG:
838     case PHB_SCOM_ETU_PMON_CTR0:
839     case PHB_SCOM_ETU_PMON_CTR1:
840     case PHB_SCOM_ETU_PMON_CTR2:
841     case PHB_SCOM_ETU_PMON_CTR3:
842         offset = ((reg - PHB_SCOM_ETU_PMON_CONFIG) << 3) + PHB_PERFMON_CONFIG;
843         pnv_phb4_reg_write(phb, offset, val, size);
844         break;
845     default:
846         qemu_log_mask(LOG_UNIMP, "phb4: xscom_write 0x%"HWADDR_PRIx
847                       "=%"PRIx64"\n", addr, val);
848     }
849 }
850 
851 const MemoryRegionOps pnv_phb4_xscom_ops = {
852     .read = pnv_phb4_xscom_read,
853     .write = pnv_phb4_xscom_write,
854     .valid.min_access_size = 8,
855     .valid.max_access_size = 8,
856     .impl.min_access_size = 8,
857     .impl.max_access_size = 8,
858     .endianness = DEVICE_BIG_ENDIAN,
859 };
860 
861 static uint64_t pnv_pec_stk_nest_xscom_read(void *opaque, hwaddr addr,
862                                             unsigned size)
863 {
864     PnvPHB4 *phb = PNV_PHB4(opaque);
865     uint32_t reg = addr >> 3;
866 
867     /* TODO: add list of allowed registers and error out if not */
868     return phb->nest_regs[reg];
869 }
870 
871 static void pnv_phb4_update_regions(PnvPHB4 *phb)
872 {
873     /* Unmap first always */
874     if (memory_region_is_mapped(&phb->mr_regs)) {
875         memory_region_del_subregion(&phb->phbbar, &phb->mr_regs);
876     }
877     if (memory_region_is_mapped(&phb->xsrc.esb_mmio)) {
878         memory_region_del_subregion(&phb->intbar, &phb->xsrc.esb_mmio);
879     }
880 
881     /* Map registers if enabled */
882     if (memory_region_is_mapped(&phb->phbbar)) {
883         memory_region_add_subregion(&phb->phbbar, 0, &phb->mr_regs);
884     }
885 
886     /* Map ESB if enabled */
887     if (memory_region_is_mapped(&phb->intbar)) {
888         memory_region_add_subregion(&phb->intbar, 0, &phb->xsrc.esb_mmio);
889     }
890 
891     /* Check/update m32 */
892     pnv_phb4_check_all_mbt(phb);
893 }
894 
895 static void pnv_pec_stk_update_map(PnvPHB4 *phb)
896 {
897     PnvPhb4PecStack *stack = phb->stack;
898     PnvPhb4PecState *pec = phb->pec;
899     MemoryRegion *sysmem = get_system_memory();
900     uint64_t bar_en = phb->nest_regs[PEC_NEST_STK_BAR_EN];
901     uint64_t bar, mask, size;
902     char name[64];
903 
904     /*
905      * NOTE: This will really not work well if those are remapped
906      * after the PHB has created its sub regions. We could do better
907      * if we had a way to resize regions but we don't really care
908      * that much in practice as the stuff below really only happens
909      * once early during boot
910      */
911 
912     /* Handle unmaps */
913     if (memory_region_is_mapped(&phb->mmbar0) &&
914         !(bar_en & PEC_NEST_STK_BAR_EN_MMIO0)) {
915         memory_region_del_subregion(sysmem, &phb->mmbar0);
916     }
917     if (memory_region_is_mapped(&phb->mmbar1) &&
918         !(bar_en & PEC_NEST_STK_BAR_EN_MMIO1)) {
919         memory_region_del_subregion(sysmem, &phb->mmbar1);
920     }
921     if (memory_region_is_mapped(&phb->phbbar) &&
922         !(bar_en & PEC_NEST_STK_BAR_EN_PHB)) {
923         memory_region_del_subregion(sysmem, &phb->phbbar);
924     }
925     if (memory_region_is_mapped(&phb->intbar) &&
926         !(bar_en & PEC_NEST_STK_BAR_EN_INT)) {
927         memory_region_del_subregion(sysmem, &phb->intbar);
928     }
929 
930     /* Update PHB */
931     pnv_phb4_update_regions(phb);
932 
933     /* Handle maps */
934     if (!memory_region_is_mapped(&phb->mmbar0) &&
935         (bar_en & PEC_NEST_STK_BAR_EN_MMIO0)) {
936         bar = phb->nest_regs[PEC_NEST_STK_MMIO_BAR0] >> 8;
937         mask = phb->nest_regs[PEC_NEST_STK_MMIO_BAR0_MASK];
938         size = ((~mask) >> 8) + 1;
939         snprintf(name, sizeof(name), "pec-%d.%d-phb-%d-mmio0",
940                  pec->chip_id, pec->index, stack->stack_no);
941         memory_region_init(&phb->mmbar0, OBJECT(phb), name, size);
942         memory_region_add_subregion(sysmem, bar, &phb->mmbar0);
943         phb->mmio0_base = bar;
944         phb->mmio0_size = size;
945     }
946     if (!memory_region_is_mapped(&phb->mmbar1) &&
947         (bar_en & PEC_NEST_STK_BAR_EN_MMIO1)) {
948         bar = phb->nest_regs[PEC_NEST_STK_MMIO_BAR1] >> 8;
949         mask = phb->nest_regs[PEC_NEST_STK_MMIO_BAR1_MASK];
950         size = ((~mask) >> 8) + 1;
951         snprintf(name, sizeof(name), "pec-%d.%d-phb-%d-mmio1",
952                  pec->chip_id, pec->index, stack->stack_no);
953         memory_region_init(&phb->mmbar1, OBJECT(phb), name, size);
954         memory_region_add_subregion(sysmem, bar, &phb->mmbar1);
955         phb->mmio1_base = bar;
956         phb->mmio1_size = size;
957     }
958     if (!memory_region_is_mapped(&phb->phbbar) &&
959         (bar_en & PEC_NEST_STK_BAR_EN_PHB)) {
960         bar = phb->nest_regs[PEC_NEST_STK_PHB_REGS_BAR] >> 8;
961         size = PNV_PHB4_NUM_REGS << 3;
962         snprintf(name, sizeof(name), "pec-%d.%d-phb-%d",
963                  pec->chip_id, pec->index, stack->stack_no);
964         memory_region_init(&phb->phbbar, OBJECT(phb), name, size);
965         memory_region_add_subregion(sysmem, bar, &phb->phbbar);
966     }
967     if (!memory_region_is_mapped(&phb->intbar) &&
968         (bar_en & PEC_NEST_STK_BAR_EN_INT)) {
969         bar = phb->nest_regs[PEC_NEST_STK_INT_BAR] >> 8;
970         size = PNV_PHB4_MAX_INTs << 16;
971         snprintf(name, sizeof(name), "pec-%d.%d-phb-%d-int",
972                  phb->pec->chip_id, phb->pec->index, stack->stack_no);
973         memory_region_init(&phb->intbar, OBJECT(phb), name, size);
974         memory_region_add_subregion(sysmem, bar, &phb->intbar);
975     }
976 
977     /* Update PHB */
978     pnv_phb4_update_regions(phb);
979 }
980 
981 static void pnv_pec_stk_nest_xscom_write(void *opaque, hwaddr addr,
982                                          uint64_t val, unsigned size)
983 {
984     PnvPHB4 *phb = PNV_PHB4(opaque);
985     PnvPhb4PecState *pec = phb->pec;
986     uint32_t reg = addr >> 3;
987 
988     switch (reg) {
989     case PEC_NEST_STK_PCI_NEST_FIR:
990         phb->nest_regs[PEC_NEST_STK_PCI_NEST_FIR] = val;
991         break;
992     case PEC_NEST_STK_PCI_NEST_FIR_CLR:
993         phb->nest_regs[PEC_NEST_STK_PCI_NEST_FIR] &= val;
994         break;
995     case PEC_NEST_STK_PCI_NEST_FIR_SET:
996         phb->nest_regs[PEC_NEST_STK_PCI_NEST_FIR] |= val;
997         break;
998     case PEC_NEST_STK_PCI_NEST_FIR_MSK:
999         phb->nest_regs[PEC_NEST_STK_PCI_NEST_FIR_MSK] = val;
1000         break;
1001     case PEC_NEST_STK_PCI_NEST_FIR_MSKC:
1002         phb->nest_regs[PEC_NEST_STK_PCI_NEST_FIR_MSK] &= val;
1003         break;
1004     case PEC_NEST_STK_PCI_NEST_FIR_MSKS:
1005         phb->nest_regs[PEC_NEST_STK_PCI_NEST_FIR_MSK] |= val;
1006         break;
1007     case PEC_NEST_STK_PCI_NEST_FIR_ACT0:
1008     case PEC_NEST_STK_PCI_NEST_FIR_ACT1:
1009         phb->nest_regs[reg] = val;
1010         break;
1011     case PEC_NEST_STK_PCI_NEST_FIR_WOF:
1012         phb->nest_regs[reg] = 0;
1013         break;
1014     case PEC_NEST_STK_ERR_REPORT_0:
1015     case PEC_NEST_STK_ERR_REPORT_1:
1016     case PEC_NEST_STK_PBCQ_GNRL_STATUS:
1017         /* Flag error ? */
1018         break;
1019     case PEC_NEST_STK_PBCQ_MODE:
1020         phb->nest_regs[reg] = val & 0xff00000000000000ull;
1021         break;
1022     case PEC_NEST_STK_MMIO_BAR0:
1023     case PEC_NEST_STK_MMIO_BAR0_MASK:
1024     case PEC_NEST_STK_MMIO_BAR1:
1025     case PEC_NEST_STK_MMIO_BAR1_MASK:
1026         if (phb->nest_regs[PEC_NEST_STK_BAR_EN] &
1027             (PEC_NEST_STK_BAR_EN_MMIO0 |
1028              PEC_NEST_STK_BAR_EN_MMIO1)) {
1029             phb_pec_error(pec, "Changing enabled BAR unsupported\n");
1030         }
1031         phb->nest_regs[reg] = val & 0xffffffffff000000ull;
1032         break;
1033     case PEC_NEST_STK_PHB_REGS_BAR:
1034         if (phb->nest_regs[PEC_NEST_STK_BAR_EN] & PEC_NEST_STK_BAR_EN_PHB) {
1035             phb_pec_error(pec, "Changing enabled BAR unsupported\n");
1036         }
1037         phb->nest_regs[reg] = val & 0xffffffffffc00000ull;
1038         break;
1039     case PEC_NEST_STK_INT_BAR:
1040         if (phb->nest_regs[PEC_NEST_STK_BAR_EN] & PEC_NEST_STK_BAR_EN_INT) {
1041             phb_pec_error(pec, "Changing enabled BAR unsupported\n");
1042         }
1043         phb->nest_regs[reg] = val & 0xfffffff000000000ull;
1044         break;
1045     case PEC_NEST_STK_BAR_EN:
1046         phb->nest_regs[reg] = val & 0xf000000000000000ull;
1047         pnv_pec_stk_update_map(phb);
1048         break;
1049     case PEC_NEST_STK_DATA_FRZ_TYPE:
1050     case PEC_NEST_STK_PBCQ_TUN_BAR:
1051         /* Not used for now */
1052         phb->nest_regs[reg] = val;
1053         break;
1054     default:
1055         qemu_log_mask(LOG_UNIMP, "phb4_pec: nest_xscom_write 0x%"HWADDR_PRIx
1056                       "=%"PRIx64"\n", addr, val);
1057     }
1058 }
1059 
1060 static const MemoryRegionOps pnv_pec_stk_nest_xscom_ops = {
1061     .read = pnv_pec_stk_nest_xscom_read,
1062     .write = pnv_pec_stk_nest_xscom_write,
1063     .valid.min_access_size = 8,
1064     .valid.max_access_size = 8,
1065     .impl.min_access_size = 8,
1066     .impl.max_access_size = 8,
1067     .endianness = DEVICE_BIG_ENDIAN,
1068 };
1069 
1070 static uint64_t pnv_pec_stk_pci_xscom_read(void *opaque, hwaddr addr,
1071                                            unsigned size)
1072 {
1073     PnvPHB4 *phb = PNV_PHB4(opaque);
1074     uint32_t reg = addr >> 3;
1075 
1076     /* TODO: add list of allowed registers and error out if not */
1077     return phb->pci_regs[reg];
1078 }
1079 
1080 static void pnv_pec_stk_pci_xscom_write(void *opaque, hwaddr addr,
1081                                         uint64_t val, unsigned size)
1082 {
1083     PnvPHB4 *phb = PNV_PHB4(opaque);
1084     uint32_t reg = addr >> 3;
1085 
1086     switch (reg) {
1087     case PEC_PCI_STK_PCI_FIR:
1088         phb->pci_regs[reg] = val;
1089         break;
1090     case PEC_PCI_STK_PCI_FIR_CLR:
1091         phb->pci_regs[PEC_PCI_STK_PCI_FIR] &= val;
1092         break;
1093     case PEC_PCI_STK_PCI_FIR_SET:
1094         phb->pci_regs[PEC_PCI_STK_PCI_FIR] |= val;
1095         break;
1096     case PEC_PCI_STK_PCI_FIR_MSK:
1097         phb->pci_regs[reg] = val;
1098         break;
1099     case PEC_PCI_STK_PCI_FIR_MSKC:
1100         phb->pci_regs[PEC_PCI_STK_PCI_FIR_MSK] &= val;
1101         break;
1102     case PEC_PCI_STK_PCI_FIR_MSKS:
1103         phb->pci_regs[PEC_PCI_STK_PCI_FIR_MSK] |= val;
1104         break;
1105     case PEC_PCI_STK_PCI_FIR_ACT0:
1106     case PEC_PCI_STK_PCI_FIR_ACT1:
1107         phb->pci_regs[reg] = val;
1108         break;
1109     case PEC_PCI_STK_PCI_FIR_WOF:
1110         phb->pci_regs[reg] = 0;
1111         break;
1112     case PEC_PCI_STK_ETU_RESET:
1113         phb->pci_regs[reg] = val & 0x8000000000000000ull;
1114         /* TODO: Implement reset */
1115         break;
1116     case PEC_PCI_STK_PBAIB_ERR_REPORT:
1117         break;
1118     case PEC_PCI_STK_PBAIB_TX_CMD_CRED:
1119     case PEC_PCI_STK_PBAIB_TX_DAT_CRED:
1120         phb->pci_regs[reg] = val;
1121         break;
1122     default:
1123         qemu_log_mask(LOG_UNIMP, "phb4_pec_stk: pci_xscom_write 0x%"HWADDR_PRIx
1124                       "=%"PRIx64"\n", addr, val);
1125     }
1126 }
1127 
1128 static const MemoryRegionOps pnv_pec_stk_pci_xscom_ops = {
1129     .read = pnv_pec_stk_pci_xscom_read,
1130     .write = pnv_pec_stk_pci_xscom_write,
1131     .valid.min_access_size = 8,
1132     .valid.max_access_size = 8,
1133     .impl.min_access_size = 8,
1134     .impl.max_access_size = 8,
1135     .endianness = DEVICE_BIG_ENDIAN,
1136 };
1137 
1138 static int pnv_phb4_map_irq(PCIDevice *pci_dev, int irq_num)
1139 {
1140     /* Check that out properly ... */
1141     return irq_num & 3;
1142 }
1143 
1144 static void pnv_phb4_set_irq(void *opaque, int irq_num, int level)
1145 {
1146     PnvPHB4 *phb = PNV_PHB4(opaque);
1147     uint32_t lsi_base;
1148 
1149     /* LSI only ... */
1150     if (irq_num > 3) {
1151         phb_error(phb, "IRQ %x is not an LSI", irq_num);
1152     }
1153     lsi_base = GETFIELD(PHB_LSI_SRC_ID, phb->regs[PHB_LSI_SOURCE_ID >> 3]);
1154     lsi_base <<= 3;
1155     qemu_set_irq(phb->qirqs[lsi_base + irq_num], level);
1156 }
1157 
1158 static bool pnv_phb4_resolve_pe(PnvPhb4DMASpace *ds)
1159 {
1160     uint64_t rtt, addr;
1161     uint16_t rte;
1162     int bus_num;
1163     int num_PEs;
1164 
1165     /* Already resolved ? */
1166     if (ds->pe_num != PHB_INVALID_PE) {
1167         return true;
1168     }
1169 
1170     /* We need to lookup the RTT */
1171     rtt = ds->phb->regs[PHB_RTT_BAR >> 3];
1172     if (!(rtt & PHB_RTT_BAR_ENABLE)) {
1173         phb_error(ds->phb, "DMA with RTT BAR disabled !");
1174         /* Set error bits ? fence ? ... */
1175         return false;
1176     }
1177 
1178     /* Read RTE */
1179     bus_num = pci_bus_num(ds->bus);
1180     addr = rtt & PHB_RTT_BASE_ADDRESS_MASK;
1181     addr += 2 * PCI_BUILD_BDF(bus_num, ds->devfn);
1182     if (dma_memory_read(&address_space_memory, addr, &rte,
1183                         sizeof(rte), MEMTXATTRS_UNSPECIFIED)) {
1184         phb_error(ds->phb, "Failed to read RTT entry at 0x%"PRIx64, addr);
1185         /* Set error bits ? fence ? ... */
1186         return false;
1187     }
1188     rte = be16_to_cpu(rte);
1189 
1190     /* Fail upon reading of invalid PE# */
1191     num_PEs = ds->phb->big_phb ? PNV_PHB4_MAX_PEs : (PNV_PHB4_MAX_PEs >> 1);
1192     if (rte >= num_PEs) {
1193         phb_error(ds->phb, "RTE for RID 0x%x invalid (%04x", ds->devfn, rte);
1194         rte &= num_PEs - 1;
1195     }
1196     ds->pe_num = rte;
1197     return true;
1198 }
1199 
1200 static void pnv_phb4_translate_tve(PnvPhb4DMASpace *ds, hwaddr addr,
1201                                    bool is_write, uint64_t tve,
1202                                    IOMMUTLBEntry *tlb)
1203 {
1204     uint64_t tta = GETFIELD(IODA3_TVT_TABLE_ADDR, tve);
1205     int32_t  lev = GETFIELD(IODA3_TVT_NUM_LEVELS, tve);
1206     uint32_t tts = GETFIELD(IODA3_TVT_TCE_TABLE_SIZE, tve);
1207     uint32_t tps = GETFIELD(IODA3_TVT_IO_PSIZE, tve);
1208 
1209     /* Invalid levels */
1210     if (lev > 4) {
1211         phb_error(ds->phb, "Invalid #levels in TVE %d", lev);
1212         return;
1213     }
1214 
1215     /* Invalid entry */
1216     if (tts == 0) {
1217         phb_error(ds->phb, "Access to invalid TVE");
1218         return;
1219     }
1220 
1221     /* IO Page Size of 0 means untranslated, else use TCEs */
1222     if (tps == 0) {
1223         /* TODO: Handle boundaries */
1224 
1225         /* Use 4k pages like q35 ... for now */
1226         tlb->iova = addr & 0xfffffffffffff000ull;
1227         tlb->translated_addr = addr & 0x0003fffffffff000ull;
1228         tlb->addr_mask = 0xfffull;
1229         tlb->perm = IOMMU_RW;
1230     } else {
1231         uint32_t tce_shift, tbl_shift, sh;
1232         uint64_t base, taddr, tce, tce_mask;
1233 
1234         /* Address bits per bottom level TCE entry */
1235         tce_shift = tps + 11;
1236 
1237         /* Address bits per table level */
1238         tbl_shift = tts + 8;
1239 
1240         /* Top level table base address */
1241         base = tta << 12;
1242 
1243         /* Total shift to first level */
1244         sh = tbl_shift * lev + tce_shift;
1245 
1246         /* TODO: Limit to support IO page sizes */
1247 
1248         /* TODO: Multi-level untested */
1249         while ((lev--) >= 0) {
1250             /* Grab the TCE address */
1251             taddr = base | (((addr >> sh) & ((1ul << tbl_shift) - 1)) << 3);
1252             if (dma_memory_read(&address_space_memory, taddr, &tce,
1253                                 sizeof(tce), MEMTXATTRS_UNSPECIFIED)) {
1254                 phb_error(ds->phb, "Failed to read TCE at 0x%"PRIx64, taddr);
1255                 return;
1256             }
1257             tce = be64_to_cpu(tce);
1258 
1259             /* Check permission for indirect TCE */
1260             if ((lev >= 0) && !(tce & 3)) {
1261                 phb_error(ds->phb, "Invalid indirect TCE at 0x%"PRIx64, taddr);
1262                 phb_error(ds->phb, " xlate %"PRIx64":%c TVE=%"PRIx64, addr,
1263                            is_write ? 'W' : 'R', tve);
1264                 phb_error(ds->phb, " tta=%"PRIx64" lev=%d tts=%d tps=%d",
1265                            tta, lev, tts, tps);
1266                 return;
1267             }
1268             sh -= tbl_shift;
1269             base = tce & ~0xfffull;
1270         }
1271 
1272         /* We exit the loop with TCE being the final TCE */
1273         tce_mask = ~((1ull << tce_shift) - 1);
1274         tlb->iova = addr & tce_mask;
1275         tlb->translated_addr = tce & tce_mask;
1276         tlb->addr_mask = ~tce_mask;
1277         tlb->perm = tce & 3;
1278         if ((is_write & !(tce & 2)) || ((!is_write) && !(tce & 1))) {
1279             phb_error(ds->phb, "TCE access fault at 0x%"PRIx64, taddr);
1280             phb_error(ds->phb, " xlate %"PRIx64":%c TVE=%"PRIx64, addr,
1281                        is_write ? 'W' : 'R', tve);
1282             phb_error(ds->phb, " tta=%"PRIx64" lev=%d tts=%d tps=%d",
1283                        tta, lev, tts, tps);
1284         }
1285     }
1286 }
1287 
1288 static IOMMUTLBEntry pnv_phb4_translate_iommu(IOMMUMemoryRegion *iommu,
1289                                               hwaddr addr,
1290                                               IOMMUAccessFlags flag,
1291                                               int iommu_idx)
1292 {
1293     PnvPhb4DMASpace *ds = container_of(iommu, PnvPhb4DMASpace, dma_mr);
1294     int tve_sel;
1295     uint64_t tve, cfg;
1296     IOMMUTLBEntry ret = {
1297         .target_as = &address_space_memory,
1298         .iova = addr,
1299         .translated_addr = 0,
1300         .addr_mask = ~(hwaddr)0,
1301         .perm = IOMMU_NONE,
1302     };
1303 
1304     /* Resolve PE# */
1305     if (!pnv_phb4_resolve_pe(ds)) {
1306         phb_error(ds->phb, "Failed to resolve PE# for bus @%p (%d) devfn 0x%x",
1307                    ds->bus, pci_bus_num(ds->bus), ds->devfn);
1308         return ret;
1309     }
1310 
1311     /* Check top bits */
1312     switch (addr >> 60) {
1313     case 00:
1314         /* DMA or 32-bit MSI ? */
1315         cfg = ds->phb->regs[PHB_PHB4_CONFIG >> 3];
1316         if ((cfg & PHB_PHB4C_32BIT_MSI_EN) &&
1317             ((addr & 0xffffffffffff0000ull) == 0xffff0000ull)) {
1318             phb_error(ds->phb, "xlate on 32-bit MSI region");
1319             return ret;
1320         }
1321         /* Choose TVE XXX Use PHB4 Control Register */
1322         tve_sel = (addr >> 59) & 1;
1323         tve = ds->phb->ioda_TVT[ds->pe_num * 2 + tve_sel];
1324         pnv_phb4_translate_tve(ds, addr, flag & IOMMU_WO, tve, &ret);
1325         break;
1326     case 01:
1327         phb_error(ds->phb, "xlate on 64-bit MSI region");
1328         break;
1329     default:
1330         phb_error(ds->phb, "xlate on unsupported address 0x%"PRIx64, addr);
1331     }
1332     return ret;
1333 }
1334 
1335 #define TYPE_PNV_PHB4_IOMMU_MEMORY_REGION "pnv-phb4-iommu-memory-region"
1336 DECLARE_INSTANCE_CHECKER(IOMMUMemoryRegion, PNV_PHB4_IOMMU_MEMORY_REGION,
1337                          TYPE_PNV_PHB4_IOMMU_MEMORY_REGION)
1338 
1339 static void pnv_phb4_iommu_memory_region_class_init(ObjectClass *klass,
1340                                                     void *data)
1341 {
1342     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_CLASS(klass);
1343 
1344     imrc->translate = pnv_phb4_translate_iommu;
1345 }
1346 
1347 static const TypeInfo pnv_phb4_iommu_memory_region_info = {
1348     .parent = TYPE_IOMMU_MEMORY_REGION,
1349     .name = TYPE_PNV_PHB4_IOMMU_MEMORY_REGION,
1350     .class_init = pnv_phb4_iommu_memory_region_class_init,
1351 };
1352 
1353 /*
1354  * Return the index/phb-id of a PHB4 that belongs to a
1355  * pec->stacks[stack_index] stack.
1356  */
1357 int pnv_phb4_pec_get_phb_id(PnvPhb4PecState *pec, int stack_index)
1358 {
1359     PnvPhb4PecClass *pecc = PNV_PHB4_PEC_GET_CLASS(pec);
1360     int index = pec->index;
1361     int offset = 0;
1362 
1363     while (index--) {
1364         offset += pecc->num_stacks[index];
1365     }
1366 
1367     return offset + stack_index;
1368 }
1369 
1370 /*
1371  * MSI/MSIX memory region implementation.
1372  * The handler handles both MSI and MSIX.
1373  */
1374 static void pnv_phb4_msi_write(void *opaque, hwaddr addr,
1375                                uint64_t data, unsigned size)
1376 {
1377     PnvPhb4DMASpace *ds = opaque;
1378     PnvPHB4 *phb = ds->phb;
1379 
1380     uint32_t src = ((addr >> 4) & 0xffff) | (data & 0x1f);
1381 
1382     /* Resolve PE# */
1383     if (!pnv_phb4_resolve_pe(ds)) {
1384         phb_error(phb, "Failed to resolve PE# for bus @%p (%d) devfn 0x%x",
1385                    ds->bus, pci_bus_num(ds->bus), ds->devfn);
1386         return;
1387     }
1388 
1389     /* TODO: Check it doesn't collide with LSIs */
1390     if (src >= phb->xsrc.nr_irqs) {
1391         phb_error(phb, "MSI %d out of bounds", src);
1392         return;
1393     }
1394 
1395     /* TODO: check PE/MSI assignement */
1396 
1397     qemu_irq_pulse(phb->qirqs[src]);
1398 }
1399 
1400 /* There is no .read as the read result is undefined by PCI spec */
1401 static uint64_t pnv_phb4_msi_read(void *opaque, hwaddr addr, unsigned size)
1402 {
1403     PnvPhb4DMASpace *ds = opaque;
1404 
1405     phb_error(ds->phb, "Invalid MSI read @ 0x%" HWADDR_PRIx, addr);
1406     return -1;
1407 }
1408 
1409 static const MemoryRegionOps pnv_phb4_msi_ops = {
1410     .read = pnv_phb4_msi_read,
1411     .write = pnv_phb4_msi_write,
1412     .endianness = DEVICE_LITTLE_ENDIAN
1413 };
1414 
1415 static PnvPhb4DMASpace *pnv_phb4_dma_find(PnvPHB4 *phb, PCIBus *bus, int devfn)
1416 {
1417     PnvPhb4DMASpace *ds;
1418 
1419     QLIST_FOREACH(ds, &phb->dma_spaces, list) {
1420         if (ds->bus == bus && ds->devfn == devfn) {
1421             break;
1422         }
1423     }
1424     return ds;
1425 }
1426 
1427 static AddressSpace *pnv_phb4_dma_iommu(PCIBus *bus, void *opaque, int devfn)
1428 {
1429     PnvPHB4 *phb = opaque;
1430     PnvPhb4DMASpace *ds;
1431     char name[32];
1432 
1433     ds = pnv_phb4_dma_find(phb, bus, devfn);
1434 
1435     if (ds == NULL) {
1436         ds = g_malloc0(sizeof(PnvPhb4DMASpace));
1437         ds->bus = bus;
1438         ds->devfn = devfn;
1439         ds->pe_num = PHB_INVALID_PE;
1440         ds->phb = phb;
1441         snprintf(name, sizeof(name), "phb4-%d.%d-iommu", phb->chip_id,
1442                  phb->phb_id);
1443         memory_region_init_iommu(&ds->dma_mr, sizeof(ds->dma_mr),
1444                                  TYPE_PNV_PHB4_IOMMU_MEMORY_REGION,
1445                                  OBJECT(phb), name, UINT64_MAX);
1446         address_space_init(&ds->dma_as, MEMORY_REGION(&ds->dma_mr),
1447                            name);
1448         memory_region_init_io(&ds->msi32_mr, OBJECT(phb), &pnv_phb4_msi_ops,
1449                               ds, "msi32", 0x10000);
1450         memory_region_init_io(&ds->msi64_mr, OBJECT(phb), &pnv_phb4_msi_ops,
1451                               ds, "msi64", 0x100000);
1452         pnv_phb4_update_msi_regions(ds);
1453 
1454         QLIST_INSERT_HEAD(&phb->dma_spaces, ds, list);
1455     }
1456     return &ds->dma_as;
1457 }
1458 
1459 static void pnv_phb4_xscom_realize(PnvPHB4 *phb)
1460 {
1461     PnvPhb4PecStack *stack = phb->stack;
1462     PnvPhb4PecState *pec = phb->pec;
1463     PnvPhb4PecClass *pecc = PNV_PHB4_PEC_GET_CLASS(pec);
1464     uint32_t pec_nest_base;
1465     uint32_t pec_pci_base;
1466     char name[64];
1467 
1468     assert(pec);
1469 
1470     /* Initialize the XSCOM regions for the stack registers */
1471     snprintf(name, sizeof(name), "xscom-pec-%d.%d-nest-phb-%d",
1472              pec->chip_id, pec->index, stack->stack_no);
1473     pnv_xscom_region_init(&phb->nest_regs_mr, OBJECT(phb),
1474                           &pnv_pec_stk_nest_xscom_ops, phb, name,
1475                           PHB4_PEC_NEST_STK_REGS_COUNT);
1476 
1477     snprintf(name, sizeof(name), "xscom-pec-%d.%d-pci-phb-%d",
1478              pec->chip_id, pec->index, stack->stack_no);
1479     pnv_xscom_region_init(&phb->pci_regs_mr, OBJECT(phb),
1480                           &pnv_pec_stk_pci_xscom_ops, phb, name,
1481                           PHB4_PEC_PCI_STK_REGS_COUNT);
1482 
1483     /* PHB pass-through */
1484     snprintf(name, sizeof(name), "xscom-pec-%d.%d-pci-phb-%d",
1485              pec->chip_id, pec->index, stack->stack_no);
1486     pnv_xscom_region_init(&phb->phb_regs_mr, OBJECT(phb),
1487                           &pnv_phb4_xscom_ops, phb, name, 0x40);
1488 
1489     pec_nest_base = pecc->xscom_nest_base(pec);
1490     pec_pci_base = pecc->xscom_pci_base(pec);
1491 
1492     /* Populate the XSCOM address space. */
1493     pnv_xscom_add_subregion(pec->chip,
1494                             pec_nest_base + 0x40 * (stack->stack_no + 1),
1495                             &phb->nest_regs_mr);
1496     pnv_xscom_add_subregion(pec->chip,
1497                             pec_pci_base + 0x40 * (stack->stack_no + 1),
1498                             &phb->pci_regs_mr);
1499     pnv_xscom_add_subregion(pec->chip,
1500                             pec_pci_base + PNV9_XSCOM_PEC_PCI_STK0 +
1501                             0x40 * stack->stack_no,
1502                             &phb->phb_regs_mr);
1503 }
1504 
1505 static void pnv_phb4_instance_init(Object *obj)
1506 {
1507     PnvPHB4 *phb = PNV_PHB4(obj);
1508 
1509     QLIST_INIT(&phb->dma_spaces);
1510 
1511     /* XIVE interrupt source object */
1512     object_initialize_child(obj, "source", &phb->xsrc, TYPE_XIVE_SOURCE);
1513 }
1514 
1515 static PnvPhb4PecStack *pnv_phb4_get_stack(PnvChip *chip, PnvPHB4 *phb,
1516                                            Error **errp)
1517 {
1518     Pnv9Chip *chip9 = PNV9_CHIP(chip);
1519     int chip_id = phb->chip_id;
1520     int index = phb->phb_id;
1521     int i, j;
1522 
1523     for (i = 0; i < chip->num_pecs; i++) {
1524         /*
1525          * For each PEC, check the amount of stacks it supports
1526          * and see if the given phb4 index matches a stack.
1527          */
1528         PnvPhb4PecState *pec = &chip9->pecs[i];
1529 
1530         for (j = 0; j < pec->num_stacks; j++) {
1531             if (index == pnv_phb4_pec_get_phb_id(pec, j)) {
1532                 return &pec->stacks[j];
1533             }
1534         }
1535     }
1536 
1537     error_setg(errp,
1538                "pnv-phb4 chip-id %d index %d didn't match any existing PEC",
1539                chip_id, index);
1540 
1541     return NULL;
1542 }
1543 
1544 static void pnv_phb4_realize(DeviceState *dev, Error **errp)
1545 {
1546     PnvPHB4 *phb = PNV_PHB4(dev);
1547     PCIHostState *pci = PCI_HOST_BRIDGE(dev);
1548     XiveSource *xsrc = &phb->xsrc;
1549     Error *local_err = NULL;
1550     int nr_irqs;
1551     char name[32];
1552 
1553     /* User created PHB */
1554     if (!phb->stack) {
1555         PnvMachineState *pnv = PNV_MACHINE(qdev_get_machine());
1556         PnvChip *chip = pnv_get_chip(pnv, phb->chip_id);
1557         PnvPhb4PecClass *pecc;
1558         BusState *s;
1559 
1560         if (!chip) {
1561             error_setg(errp, "invalid chip id: %d", phb->chip_id);
1562             return;
1563         }
1564 
1565         phb->stack = pnv_phb4_get_stack(chip, phb, &local_err);
1566         if (local_err) {
1567             error_propagate(errp, local_err);
1568             return;
1569         }
1570 
1571         /*
1572          * All other phb properties but 'pec' ad 'version' are
1573          * already set.
1574          */
1575         object_property_set_link(OBJECT(phb), "pec", OBJECT(phb->stack->pec),
1576                                  &error_abort);
1577         pecc = PNV_PHB4_PEC_GET_CLASS(phb->pec);
1578         object_property_set_int(OBJECT(phb), "version", pecc->version,
1579                                 &error_fatal);
1580 
1581         /*
1582          * Assign stack->phb since pnv_phb4_update_regions() uses it
1583          * to access the phb.
1584          */
1585         phb->stack->phb = phb;
1586 
1587         /*
1588          * Reparent user created devices to the chip to build
1589          * correctly the device tree.
1590          */
1591         pnv_chip_parent_fixup(chip, OBJECT(phb), phb->phb_id);
1592 
1593         s = qdev_get_parent_bus(DEVICE(chip));
1594         if (!qdev_set_parent_bus(DEVICE(phb), s, &local_err)) {
1595             error_propagate(errp, local_err);
1596             return;
1597         }
1598     }
1599 
1600     /* Set the "big_phb" flag */
1601     phb->big_phb = phb->phb_id == 0 || phb->phb_id == 3;
1602 
1603     /* Controller Registers */
1604     snprintf(name, sizeof(name), "phb4-%d.%d-regs", phb->chip_id,
1605              phb->phb_id);
1606     memory_region_init_io(&phb->mr_regs, OBJECT(phb), &pnv_phb4_reg_ops, phb,
1607                           name, 0x2000);
1608 
1609     /*
1610      * PHB4 doesn't support IO space. However, qemu gets very upset if
1611      * we don't have an IO region to anchor IO BARs onto so we just
1612      * initialize one which we never hook up to anything
1613      */
1614 
1615     snprintf(name, sizeof(name), "phb4-%d.%d-pci-io", phb->chip_id,
1616              phb->phb_id);
1617     memory_region_init(&phb->pci_io, OBJECT(phb), name, 0x10000);
1618 
1619     snprintf(name, sizeof(name), "phb4-%d.%d-pci-mmio", phb->chip_id,
1620              phb->phb_id);
1621     memory_region_init(&phb->pci_mmio, OBJECT(phb), name,
1622                        PCI_MMIO_TOTAL_SIZE);
1623 
1624     pci->bus = pci_register_root_bus(dev, dev->id,
1625                                      pnv_phb4_set_irq, pnv_phb4_map_irq, phb,
1626                                      &phb->pci_mmio, &phb->pci_io,
1627                                      0, 4, TYPE_PNV_PHB4_ROOT_BUS);
1628     pci_setup_iommu(pci->bus, pnv_phb4_dma_iommu, phb);
1629     pci->bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
1630 
1631     /* Add a single Root port if running with defaults */
1632     if (defaults_enabled()) {
1633         pnv_phb_attach_root_port(PCI_HOST_BRIDGE(phb),
1634                                  TYPE_PNV_PHB4_ROOT_PORT);
1635     }
1636 
1637     /* Setup XIVE Source */
1638     if (phb->big_phb) {
1639         nr_irqs = PNV_PHB4_MAX_INTs;
1640     } else {
1641         nr_irqs = PNV_PHB4_MAX_INTs >> 1;
1642     }
1643     object_property_set_int(OBJECT(xsrc), "nr-irqs", nr_irqs, &error_fatal);
1644     object_property_set_link(OBJECT(xsrc), "xive", OBJECT(phb), &error_fatal);
1645     if (!qdev_realize(DEVICE(xsrc), NULL, errp)) {
1646         return;
1647     }
1648 
1649     pnv_phb4_update_xsrc(phb);
1650 
1651     phb->qirqs = qemu_allocate_irqs(xive_source_set_irq, xsrc, xsrc->nr_irqs);
1652 
1653     pnv_phb4_xscom_realize(phb);
1654 }
1655 
1656 static const char *pnv_phb4_root_bus_path(PCIHostState *host_bridge,
1657                                           PCIBus *rootbus)
1658 {
1659     PnvPHB4 *phb = PNV_PHB4(host_bridge);
1660 
1661     snprintf(phb->bus_path, sizeof(phb->bus_path), "00%02x:%02x",
1662              phb->chip_id, phb->phb_id);
1663     return phb->bus_path;
1664 }
1665 
1666 static void pnv_phb4_xive_notify(XiveNotifier *xf, uint32_t srcno)
1667 {
1668     PnvPHB4 *phb = PNV_PHB4(xf);
1669     uint64_t notif_port = phb->regs[PHB_INT_NOTIFY_ADDR >> 3];
1670     uint32_t offset = phb->regs[PHB_INT_NOTIFY_INDEX >> 3];
1671     uint64_t data = XIVE_TRIGGER_PQ | offset | srcno;
1672     MemTxResult result;
1673 
1674     trace_pnv_phb4_xive_notify(notif_port, data);
1675 
1676     address_space_stq_be(&address_space_memory, notif_port, data,
1677                          MEMTXATTRS_UNSPECIFIED, &result);
1678     if (result != MEMTX_OK) {
1679         phb_error(phb, "trigger failed @%"HWADDR_PRIx "\n", notif_port);
1680         return;
1681     }
1682 }
1683 
1684 static Property pnv_phb4_properties[] = {
1685         DEFINE_PROP_UINT32("index", PnvPHB4, phb_id, 0),
1686         DEFINE_PROP_UINT32("chip-id", PnvPHB4, chip_id, 0),
1687         DEFINE_PROP_UINT64("version", PnvPHB4, version, 0),
1688         DEFINE_PROP_LINK("stack", PnvPHB4, stack, TYPE_PNV_PHB4_PEC_STACK,
1689                          PnvPhb4PecStack *),
1690         DEFINE_PROP_LINK("pec", PnvPHB4, pec, TYPE_PNV_PHB4_PEC,
1691                          PnvPhb4PecState *),
1692         DEFINE_PROP_END_OF_LIST(),
1693 };
1694 
1695 static void pnv_phb4_class_init(ObjectClass *klass, void *data)
1696 {
1697     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(klass);
1698     DeviceClass *dc = DEVICE_CLASS(klass);
1699     XiveNotifierClass *xfc = XIVE_NOTIFIER_CLASS(klass);
1700 
1701     hc->root_bus_path   = pnv_phb4_root_bus_path;
1702     dc->realize         = pnv_phb4_realize;
1703     device_class_set_props(dc, pnv_phb4_properties);
1704     set_bit(DEVICE_CATEGORY_BRIDGE, dc->categories);
1705     dc->user_creatable  = true;
1706 
1707     xfc->notify         = pnv_phb4_xive_notify;
1708 }
1709 
1710 static const TypeInfo pnv_phb4_type_info = {
1711     .name          = TYPE_PNV_PHB4,
1712     .parent        = TYPE_PCIE_HOST_BRIDGE,
1713     .instance_init = pnv_phb4_instance_init,
1714     .instance_size = sizeof(PnvPHB4),
1715     .class_init    = pnv_phb4_class_init,
1716     .interfaces = (InterfaceInfo[]) {
1717             { TYPE_XIVE_NOTIFIER },
1718             { },
1719     }
1720 };
1721 
1722 static void pnv_phb4_root_bus_class_init(ObjectClass *klass, void *data)
1723 {
1724     BusClass *k = BUS_CLASS(klass);
1725 
1726     /*
1727      * PHB4 has only a single root complex. Enforce the limit on the
1728      * parent bus
1729      */
1730     k->max_dev = 1;
1731 }
1732 
1733 static const TypeInfo pnv_phb4_root_bus_info = {
1734     .name = TYPE_PNV_PHB4_ROOT_BUS,
1735     .parent = TYPE_PCIE_BUS,
1736     .class_init = pnv_phb4_root_bus_class_init,
1737     .interfaces = (InterfaceInfo[]) {
1738         { INTERFACE_PCIE_DEVICE },
1739         { }
1740     },
1741 };
1742 
1743 static void pnv_phb4_root_port_reset(DeviceState *dev)
1744 {
1745     PCIERootPortClass *rpc = PCIE_ROOT_PORT_GET_CLASS(dev);
1746     PCIDevice *d = PCI_DEVICE(dev);
1747     uint8_t *conf = d->config;
1748 
1749     rpc->parent_reset(dev);
1750 
1751     pci_byte_test_and_set_mask(conf + PCI_IO_BASE,
1752                                PCI_IO_RANGE_MASK & 0xff);
1753     pci_byte_test_and_clear_mask(conf + PCI_IO_LIMIT,
1754                                  PCI_IO_RANGE_MASK & 0xff);
1755     pci_set_word(conf + PCI_MEMORY_BASE, 0);
1756     pci_set_word(conf + PCI_MEMORY_LIMIT, 0xfff0);
1757     pci_set_word(conf + PCI_PREF_MEMORY_BASE, 0x1);
1758     pci_set_word(conf + PCI_PREF_MEMORY_LIMIT, 0xfff1);
1759     pci_set_long(conf + PCI_PREF_BASE_UPPER32, 0x1); /* Hack */
1760     pci_set_long(conf + PCI_PREF_LIMIT_UPPER32, 0xffffffff);
1761 }
1762 
1763 static void pnv_phb4_root_port_realize(DeviceState *dev, Error **errp)
1764 {
1765     PCIERootPortClass *rpc = PCIE_ROOT_PORT_GET_CLASS(dev);
1766     PCIDevice *pci = PCI_DEVICE(dev);
1767     PCIBus *bus = pci_get_bus(pci);
1768     PnvPHB4 *phb = NULL;
1769     Error *local_err = NULL;
1770 
1771     phb = (PnvPHB4 *) object_dynamic_cast(OBJECT(bus->qbus.parent),
1772                                           TYPE_PNV_PHB4);
1773 
1774     if (!phb) {
1775         error_setg(errp, "%s must be connected to pnv-phb4 buses", dev->id);
1776         return;
1777     }
1778 
1779     /* Set unique chassis/slot values for the root port */
1780     qdev_prop_set_uint8(&pci->qdev, "chassis", phb->chip_id);
1781     qdev_prop_set_uint16(&pci->qdev, "slot", phb->phb_id);
1782 
1783     rpc->parent_realize(dev, &local_err);
1784     if (local_err) {
1785         error_propagate(errp, local_err);
1786         return;
1787     }
1788 }
1789 
1790 static void pnv_phb4_root_port_class_init(ObjectClass *klass, void *data)
1791 {
1792     DeviceClass *dc = DEVICE_CLASS(klass);
1793     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
1794     PCIERootPortClass *rpc = PCIE_ROOT_PORT_CLASS(klass);
1795 
1796     dc->desc     = "IBM PHB4 PCIE Root Port";
1797     dc->user_creatable = true;
1798 
1799     device_class_set_parent_realize(dc, pnv_phb4_root_port_realize,
1800                                     &rpc->parent_realize);
1801     device_class_set_parent_reset(dc, pnv_phb4_root_port_reset,
1802                                   &rpc->parent_reset);
1803 
1804     k->vendor_id = PCI_VENDOR_ID_IBM;
1805     k->device_id = PNV_PHB4_DEVICE_ID;
1806     k->revision  = 0;
1807 
1808     rpc->exp_offset = 0x48;
1809     rpc->aer_offset = 0x100;
1810 
1811     dc->reset = &pnv_phb4_root_port_reset;
1812 }
1813 
1814 static const TypeInfo pnv_phb4_root_port_info = {
1815     .name          = TYPE_PNV_PHB4_ROOT_PORT,
1816     .parent        = TYPE_PCIE_ROOT_PORT,
1817     .instance_size = sizeof(PnvPHB4RootPort),
1818     .class_init    = pnv_phb4_root_port_class_init,
1819 };
1820 
1821 static void pnv_phb4_register_types(void)
1822 {
1823     type_register_static(&pnv_phb4_root_bus_info);
1824     type_register_static(&pnv_phb4_root_port_info);
1825     type_register_static(&pnv_phb4_type_info);
1826     type_register_static(&pnv_phb4_iommu_memory_region_info);
1827 }
1828 
1829 type_init(pnv_phb4_register_types);
1830 
1831 void pnv_phb4_pic_print_info(PnvPHB4 *phb, Monitor *mon)
1832 {
1833     uint32_t offset = phb->regs[PHB_INT_NOTIFY_INDEX >> 3];
1834 
1835     monitor_printf(mon, "PHB4[%x:%x] Source %08x .. %08x\n",
1836                    phb->chip_id, phb->phb_id,
1837                    offset, offset + phb->xsrc.nr_irqs - 1);
1838     xive_source_pic_print_info(&phb->xsrc, 0, mon);
1839 }
1840