xref: /openbmc/qemu/hw/pci-host/pnv_phb4.c (revision 1360fd832b0d41b00d15574a0e9ecd6d513dbeaa)
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 /*
33  * QEMU version of the GETFIELD/SETFIELD macros
34  *
35  * These are common with the PnvXive model.
36  */
37 static inline uint64_t GETFIELD(uint64_t mask, uint64_t word)
38 {
39     return (word & mask) >> ctz64(mask);
40 }
41 
42 static inline uint64_t SETFIELD(uint64_t mask, uint64_t word,
43                                 uint64_t value)
44 {
45     return (word & ~mask) | ((value << ctz64(mask)) & mask);
46 }
47 
48 static PCIDevice *pnv_phb4_find_cfg_dev(PnvPHB4 *phb)
49 {
50     PCIHostState *pci = PCI_HOST_BRIDGE(phb);
51     uint64_t addr = phb->regs[PHB_CONFIG_ADDRESS >> 3];
52     uint8_t bus, devfn;
53 
54     if (!(addr >> 63)) {
55         return NULL;
56     }
57     bus = (addr >> 52) & 0xff;
58     devfn = (addr >> 44) & 0xff;
59 
60     /* We don't access the root complex this way */
61     if (bus == 0 && devfn == 0) {
62         return NULL;
63     }
64     return pci_find_device(pci->bus, bus, devfn);
65 }
66 
67 /*
68  * The CONFIG_DATA register expects little endian accesses, but as the
69  * region is big endian, we have to swap the value.
70  */
71 static void pnv_phb4_config_write(PnvPHB4 *phb, unsigned off,
72                                   unsigned size, uint64_t val)
73 {
74     uint32_t cfg_addr, limit;
75     PCIDevice *pdev;
76 
77     pdev = pnv_phb4_find_cfg_dev(phb);
78     if (!pdev) {
79         return;
80     }
81     cfg_addr = (phb->regs[PHB_CONFIG_ADDRESS >> 3] >> 32) & 0xffc;
82     cfg_addr |= off;
83     limit = pci_config_size(pdev);
84     if (limit <= cfg_addr) {
85         /*
86          * conventional pci device can be behind pcie-to-pci bridge.
87          * 256 <= addr < 4K has no effects.
88          */
89         return;
90     }
91     switch (size) {
92     case 1:
93         break;
94     case 2:
95         val = bswap16(val);
96         break;
97     case 4:
98         val = bswap32(val);
99         break;
100     default:
101         g_assert_not_reached();
102     }
103     pci_host_config_write_common(pdev, cfg_addr, limit, val, size);
104 }
105 
106 static uint64_t pnv_phb4_config_read(PnvPHB4 *phb, unsigned off,
107                                      unsigned size)
108 {
109     uint32_t cfg_addr, limit;
110     PCIDevice *pdev;
111     uint64_t val;
112 
113     pdev = pnv_phb4_find_cfg_dev(phb);
114     if (!pdev) {
115         return ~0ull;
116     }
117     cfg_addr = (phb->regs[PHB_CONFIG_ADDRESS >> 3] >> 32) & 0xffc;
118     cfg_addr |= off;
119     limit = pci_config_size(pdev);
120     if (limit <= cfg_addr) {
121         /*
122          * conventional pci device can be behind pcie-to-pci bridge.
123          * 256 <= addr < 4K has no effects.
124          */
125         return ~0ull;
126     }
127     val = pci_host_config_read_common(pdev, cfg_addr, limit, size);
128     switch (size) {
129     case 1:
130         return val;
131     case 2:
132         return bswap16(val);
133     case 4:
134         return bswap32(val);
135     default:
136         g_assert_not_reached();
137     }
138 }
139 
140 /*
141  * Root complex register accesses are memory mapped.
142  */
143 static void pnv_phb4_rc_config_write(PnvPHB4 *phb, unsigned off,
144                                      unsigned size, uint64_t val)
145 {
146     PCIHostState *pci = PCI_HOST_BRIDGE(phb);
147     PCIDevice *pdev;
148 
149     if (size != 4) {
150         phb_error(phb, "rc_config_write invalid size %d\n", size);
151         return;
152     }
153 
154     pdev = pci_find_device(pci->bus, 0, 0);
155     assert(pdev);
156 
157     pci_host_config_write_common(pdev, off, PHB_RC_CONFIG_SIZE,
158                                  bswap32(val), 4);
159 }
160 
161 static uint64_t pnv_phb4_rc_config_read(PnvPHB4 *phb, unsigned off,
162                                         unsigned size)
163 {
164     PCIHostState *pci = PCI_HOST_BRIDGE(phb);
165     PCIDevice *pdev;
166     uint64_t val;
167 
168     if (size != 4) {
169         phb_error(phb, "rc_config_read invalid size %d\n", size);
170         return ~0ull;
171     }
172 
173     pdev = pci_find_device(pci->bus, 0, 0);
174     assert(pdev);
175 
176     val = pci_host_config_read_common(pdev, off, PHB_RC_CONFIG_SIZE, 4);
177     return bswap32(val);
178 }
179 
180 static void pnv_phb4_check_mbt(PnvPHB4 *phb, uint32_t index)
181 {
182     uint64_t base, start, size, mbe0, mbe1;
183     MemoryRegion *parent;
184     char name[64];
185 
186     /* Unmap first */
187     if (memory_region_is_mapped(&phb->mr_mmio[index])) {
188         /* Should we destroy it in RCU friendly way... ? */
189         memory_region_del_subregion(phb->mr_mmio[index].container,
190                                     &phb->mr_mmio[index]);
191     }
192 
193     /* Get table entry */
194     mbe0 = phb->ioda_MBT[(index << 1)];
195     mbe1 = phb->ioda_MBT[(index << 1) + 1];
196 
197     if (!(mbe0 & IODA3_MBT0_ENABLE)) {
198         return;
199     }
200 
201     /* Grab geometry from registers */
202     base = GETFIELD(IODA3_MBT0_BASE_ADDR, mbe0) << 12;
203     size = GETFIELD(IODA3_MBT1_MASK, mbe1) << 12;
204     size |= 0xff00000000000000ull;
205     size = ~size + 1;
206 
207     /* Calculate PCI side start address based on M32/M64 window type */
208     if (mbe0 & IODA3_MBT0_TYPE_M32) {
209         start = phb->regs[PHB_M32_START_ADDR >> 3];
210         if ((start + size) > 0x100000000ull) {
211             phb_error(phb, "M32 set beyond 4GB boundary !");
212             size = 0x100000000 - start;
213         }
214     } else {
215         start = base | (phb->regs[PHB_M64_UPPER_BITS >> 3]);
216     }
217 
218     /* TODO: Figure out how to implemet/decode AOMASK */
219 
220     /* Check if it matches an enabled MMIO region in the PEC stack */
221     if (memory_region_is_mapped(&phb->stack->mmbar0) &&
222         base >= phb->stack->mmio0_base &&
223         (base + size) <= (phb->stack->mmio0_base + phb->stack->mmio0_size)) {
224         parent = &phb->stack->mmbar0;
225         base -= phb->stack->mmio0_base;
226     } else if (memory_region_is_mapped(&phb->stack->mmbar1) &&
227         base >= phb->stack->mmio1_base &&
228         (base + size) <= (phb->stack->mmio1_base + phb->stack->mmio1_size)) {
229         parent = &phb->stack->mmbar1;
230         base -= phb->stack->mmio1_base;
231     } else {
232         phb_error(phb, "PHB MBAR %d out of parent bounds", index);
233         return;
234     }
235 
236     /* Create alias (better name ?) */
237     snprintf(name, sizeof(name), "phb4-mbar%d", index);
238     memory_region_init_alias(&phb->mr_mmio[index], OBJECT(phb), name,
239                              &phb->pci_mmio, start, size);
240     memory_region_add_subregion(parent, base, &phb->mr_mmio[index]);
241 }
242 
243 static void pnv_phb4_check_all_mbt(PnvPHB4 *phb)
244 {
245     uint64_t i;
246     uint32_t num_windows = phb->big_phb ? PNV_PHB4_MAX_MMIO_WINDOWS :
247         PNV_PHB4_MIN_MMIO_WINDOWS;
248 
249     for (i = 0; i < num_windows; i++) {
250         pnv_phb4_check_mbt(phb, i);
251     }
252 }
253 
254 static uint64_t *pnv_phb4_ioda_access(PnvPHB4 *phb,
255                                       unsigned *out_table, unsigned *out_idx)
256 {
257     uint64_t adreg = phb->regs[PHB_IODA_ADDR >> 3];
258     unsigned int index = GETFIELD(PHB_IODA_AD_TADR, adreg);
259     unsigned int table = GETFIELD(PHB_IODA_AD_TSEL, adreg);
260     unsigned int mask;
261     uint64_t *tptr = NULL;
262 
263     switch (table) {
264     case IODA3_TBL_LIST:
265         tptr = phb->ioda_LIST;
266         mask = 7;
267         break;
268     case IODA3_TBL_MIST:
269         tptr = phb->ioda_MIST;
270         mask = phb->big_phb ? PNV_PHB4_MAX_MIST : (PNV_PHB4_MAX_MIST >> 1);
271         mask -= 1;
272         break;
273     case IODA3_TBL_RCAM:
274         mask = phb->big_phb ? 127 : 63;
275         break;
276     case IODA3_TBL_MRT:
277         mask = phb->big_phb ? 15 : 7;
278         break;
279     case IODA3_TBL_PESTA:
280     case IODA3_TBL_PESTB:
281         mask = phb->big_phb ? PNV_PHB4_MAX_PEs : (PNV_PHB4_MAX_PEs >> 1);
282         mask -= 1;
283         break;
284     case IODA3_TBL_TVT:
285         tptr = phb->ioda_TVT;
286         mask = phb->big_phb ? PNV_PHB4_MAX_TVEs : (PNV_PHB4_MAX_TVEs >> 1);
287         mask -= 1;
288         break;
289     case IODA3_TBL_TCR:
290     case IODA3_TBL_TDR:
291         mask = phb->big_phb ? 1023 : 511;
292         break;
293     case IODA3_TBL_MBT:
294         tptr = phb->ioda_MBT;
295         mask = phb->big_phb ? PNV_PHB4_MAX_MBEs : (PNV_PHB4_MAX_MBEs >> 1);
296         mask -= 1;
297         break;
298     case IODA3_TBL_MDT:
299         tptr = phb->ioda_MDT;
300         mask = phb->big_phb ? PNV_PHB4_MAX_PEs : (PNV_PHB4_MAX_PEs >> 1);
301         mask -= 1;
302         break;
303     case IODA3_TBL_PEEV:
304         tptr = phb->ioda_PEEV;
305         mask = phb->big_phb ? PNV_PHB4_MAX_PEEVs : (PNV_PHB4_MAX_PEEVs >> 1);
306         mask -= 1;
307         break;
308     default:
309         phb_error(phb, "invalid IODA table %d", table);
310         return NULL;
311     }
312     index &= mask;
313     if (out_idx) {
314         *out_idx = index;
315     }
316     if (out_table) {
317         *out_table = table;
318     }
319     if (tptr) {
320         tptr += index;
321     }
322     if (adreg & PHB_IODA_AD_AUTOINC) {
323         index = (index + 1) & mask;
324         adreg = SETFIELD(PHB_IODA_AD_TADR, adreg, index);
325     }
326 
327     phb->regs[PHB_IODA_ADDR >> 3] = adreg;
328     return tptr;
329 }
330 
331 static uint64_t pnv_phb4_ioda_read(PnvPHB4 *phb)
332 {
333     unsigned table, idx;
334     uint64_t *tptr;
335 
336     tptr = pnv_phb4_ioda_access(phb, &table, &idx);
337     if (!tptr) {
338         /* Special PESTA case */
339         if (table == IODA3_TBL_PESTA) {
340             return ((uint64_t)(phb->ioda_PEST_AB[idx] & 1)) << 63;
341         } else if (table == IODA3_TBL_PESTB) {
342             return ((uint64_t)(phb->ioda_PEST_AB[idx] & 2)) << 62;
343         }
344         /* Return 0 on unsupported tables, not ff's */
345         return 0;
346     }
347     return *tptr;
348 }
349 
350 static void pnv_phb4_ioda_write(PnvPHB4 *phb, uint64_t val)
351 {
352     unsigned table, idx;
353     uint64_t *tptr;
354 
355     tptr = pnv_phb4_ioda_access(phb, &table, &idx);
356     if (!tptr) {
357         /* Special PESTA case */
358         if (table == IODA3_TBL_PESTA) {
359             phb->ioda_PEST_AB[idx] &= ~1;
360             phb->ioda_PEST_AB[idx] |= (val >> 63) & 1;
361         } else if (table == IODA3_TBL_PESTB) {
362             phb->ioda_PEST_AB[idx] &= ~2;
363             phb->ioda_PEST_AB[idx] |= (val >> 62) & 2;
364         }
365         return;
366     }
367 
368     /* Handle side effects */
369     switch (table) {
370     case IODA3_TBL_LIST:
371         break;
372     case IODA3_TBL_MIST: {
373         /* Special mask for MIST partial write */
374         uint64_t adreg = phb->regs[PHB_IODA_ADDR >> 3];
375         uint32_t mmask = GETFIELD(PHB_IODA_AD_MIST_PWV, adreg);
376         uint64_t v = *tptr;
377         if (mmask == 0) {
378             mmask = 0xf;
379         }
380         if (mmask & 8) {
381             v &= 0x0000ffffffffffffull;
382             v |= 0xcfff000000000000ull & val;
383         }
384         if (mmask & 4) {
385             v &= 0xffff0000ffffffffull;
386             v |= 0x0000cfff00000000ull & val;
387         }
388         if (mmask & 2) {
389             v &= 0xffffffff0000ffffull;
390             v |= 0x00000000cfff0000ull & val;
391         }
392         if (mmask & 1) {
393             v &= 0xffffffffffff0000ull;
394             v |= 0x000000000000cfffull & val;
395         }
396         *tptr = v;
397         break;
398     }
399     case IODA3_TBL_MBT:
400         *tptr = val;
401 
402         /* Copy accross the valid bit to the other half */
403         phb->ioda_MBT[idx ^ 1] &= 0x7fffffffffffffffull;
404         phb->ioda_MBT[idx ^ 1] |= 0x8000000000000000ull & val;
405 
406         /* Update mappings */
407         pnv_phb4_check_mbt(phb, idx >> 1);
408         break;
409     default:
410         *tptr = val;
411     }
412 }
413 
414 static void pnv_phb4_rtc_invalidate(PnvPHB4 *phb, uint64_t val)
415 {
416     PnvPhb4DMASpace *ds;
417 
418     /* Always invalidate all for now ... */
419     QLIST_FOREACH(ds, &phb->dma_spaces, list) {
420         ds->pe_num = PHB_INVALID_PE;
421     }
422 }
423 
424 static void pnv_phb4_update_msi_regions(PnvPhb4DMASpace *ds)
425 {
426     uint64_t cfg = ds->phb->regs[PHB_PHB4_CONFIG >> 3];
427 
428     if (cfg & PHB_PHB4C_32BIT_MSI_EN) {
429         if (!memory_region_is_mapped(MEMORY_REGION(&ds->msi32_mr))) {
430             memory_region_add_subregion(MEMORY_REGION(&ds->dma_mr),
431                                         0xffff0000, &ds->msi32_mr);
432         }
433     } else {
434         if (memory_region_is_mapped(MEMORY_REGION(&ds->msi32_mr))) {
435             memory_region_del_subregion(MEMORY_REGION(&ds->dma_mr),
436                                         &ds->msi32_mr);
437         }
438     }
439 
440     if (cfg & PHB_PHB4C_64BIT_MSI_EN) {
441         if (!memory_region_is_mapped(MEMORY_REGION(&ds->msi64_mr))) {
442             memory_region_add_subregion(MEMORY_REGION(&ds->dma_mr),
443                                         (1ull << 60), &ds->msi64_mr);
444         }
445     } else {
446         if (memory_region_is_mapped(MEMORY_REGION(&ds->msi64_mr))) {
447             memory_region_del_subregion(MEMORY_REGION(&ds->dma_mr),
448                                         &ds->msi64_mr);
449         }
450     }
451 }
452 
453 static void pnv_phb4_update_all_msi_regions(PnvPHB4 *phb)
454 {
455     PnvPhb4DMASpace *ds;
456 
457     QLIST_FOREACH(ds, &phb->dma_spaces, list) {
458         pnv_phb4_update_msi_regions(ds);
459     }
460 }
461 
462 static void pnv_phb4_update_xsrc(PnvPHB4 *phb)
463 {
464     int shift, flags, i, lsi_base;
465     XiveSource *xsrc = &phb->xsrc;
466 
467     /* The XIVE source characteristics can be set at run time */
468     if (phb->regs[PHB_CTRLR >> 3] & PHB_CTRLR_IRQ_PGSZ_64K) {
469         shift = XIVE_ESB_64K;
470     } else {
471         shift = XIVE_ESB_4K;
472     }
473     if (phb->regs[PHB_CTRLR >> 3] & PHB_CTRLR_IRQ_STORE_EOI) {
474         flags = XIVE_SRC_STORE_EOI;
475     } else {
476         flags = 0;
477     }
478 
479     phb->xsrc.esb_shift = shift;
480     phb->xsrc.esb_flags = flags;
481 
482     lsi_base = GETFIELD(PHB_LSI_SRC_ID, phb->regs[PHB_LSI_SOURCE_ID >> 3]);
483     lsi_base <<= 3;
484 
485     /* TODO: handle reset values of PHB_LSI_SRC_ID */
486     if (!lsi_base) {
487         return;
488     }
489 
490     /* TODO: need a xive_source_irq_reset_lsi() */
491     bitmap_zero(xsrc->lsi_map, xsrc->nr_irqs);
492 
493     for (i = 0; i < xsrc->nr_irqs; i++) {
494         bool msi = (i < lsi_base || i >= (lsi_base + 8));
495         if (!msi) {
496             xive_source_irq_set_lsi(xsrc, i);
497         }
498     }
499 }
500 
501 static void pnv_phb4_reg_write(void *opaque, hwaddr off, uint64_t val,
502                                unsigned size)
503 {
504     PnvPHB4 *phb = PNV_PHB4(opaque);
505     bool changed;
506 
507     /* Special case outbound configuration data */
508     if ((off & 0xfffc) == PHB_CONFIG_DATA) {
509         pnv_phb4_config_write(phb, off & 0x3, size, val);
510         return;
511     }
512 
513     /* Special case RC configuration space */
514     if ((off & 0xf800) == PHB_RC_CONFIG_BASE) {
515         pnv_phb4_rc_config_write(phb, off & 0x7ff, size, val);
516         return;
517     }
518 
519     /* Other registers are 64-bit only */
520     if (size != 8 || off & 0x7) {
521         phb_error(phb, "Invalid register access, offset: 0x%"PRIx64" size: %d",
522                    off, size);
523         return;
524     }
525 
526     /* Handle masking */
527     switch (off) {
528     case PHB_LSI_SOURCE_ID:
529         val &= PHB_LSI_SRC_ID;
530         break;
531     case PHB_M64_UPPER_BITS:
532         val &= 0xff00000000000000ull;
533         break;
534     /* TCE Kill */
535     case PHB_TCE_KILL:
536         /* Clear top 3 bits which HW does to indicate successful queuing */
537         val &= ~(PHB_TCE_KILL_ALL | PHB_TCE_KILL_PE | PHB_TCE_KILL_ONE);
538         break;
539     case PHB_Q_DMA_R:
540         /*
541          * This is enough logic to make SW happy but we aren't
542          * actually quiescing the DMAs
543          */
544         if (val & PHB_Q_DMA_R_AUTORESET) {
545             val = 0;
546         } else {
547             val &= PHB_Q_DMA_R_QUIESCE_DMA;
548         }
549         break;
550     /* LEM stuff */
551     case PHB_LEM_FIR_AND_MASK:
552         phb->regs[PHB_LEM_FIR_ACCUM >> 3] &= val;
553         return;
554     case PHB_LEM_FIR_OR_MASK:
555         phb->regs[PHB_LEM_FIR_ACCUM >> 3] |= val;
556         return;
557     case PHB_LEM_ERROR_AND_MASK:
558         phb->regs[PHB_LEM_ERROR_MASK >> 3] &= val;
559         return;
560     case PHB_LEM_ERROR_OR_MASK:
561         phb->regs[PHB_LEM_ERROR_MASK >> 3] |= val;
562         return;
563     case PHB_LEM_WOF:
564         val = 0;
565         break;
566     /* TODO: More regs ..., maybe create a table with masks... */
567 
568     /* Read only registers */
569     case PHB_CPU_LOADSTORE_STATUS:
570     case PHB_ETU_ERR_SUMMARY:
571     case PHB_PHB4_GEN_CAP:
572     case PHB_PHB4_TCE_CAP:
573     case PHB_PHB4_IRQ_CAP:
574     case PHB_PHB4_EEH_CAP:
575         return;
576     }
577 
578     /* Record whether it changed */
579     changed = phb->regs[off >> 3] != val;
580 
581     /* Store in register cache first */
582     phb->regs[off >> 3] = val;
583 
584     /* Handle side effects */
585     switch (off) {
586     case PHB_PHB4_CONFIG:
587         if (changed) {
588             pnv_phb4_update_all_msi_regions(phb);
589         }
590         break;
591     case PHB_M32_START_ADDR:
592     case PHB_M64_UPPER_BITS:
593         if (changed) {
594             pnv_phb4_check_all_mbt(phb);
595         }
596         break;
597 
598     /* IODA table accesses */
599     case PHB_IODA_DATA0:
600         pnv_phb4_ioda_write(phb, val);
601         break;
602 
603     /* RTC invalidation */
604     case PHB_RTC_INVALIDATE:
605         pnv_phb4_rtc_invalidate(phb, val);
606         break;
607 
608     /* PHB Control (Affects XIVE source) */
609     case PHB_CTRLR:
610     case PHB_LSI_SOURCE_ID:
611         pnv_phb4_update_xsrc(phb);
612         break;
613 
614     /* Silent simple writes */
615     case PHB_ASN_CMPM:
616     case PHB_CONFIG_ADDRESS:
617     case PHB_IODA_ADDR:
618     case PHB_TCE_KILL:
619     case PHB_TCE_SPEC_CTL:
620     case PHB_PEST_BAR:
621     case PHB_PELTV_BAR:
622     case PHB_RTT_BAR:
623     case PHB_LEM_FIR_ACCUM:
624     case PHB_LEM_ERROR_MASK:
625     case PHB_LEM_ACTION0:
626     case PHB_LEM_ACTION1:
627     case PHB_TCE_TAG_ENABLE:
628     case PHB_INT_NOTIFY_ADDR:
629     case PHB_INT_NOTIFY_INDEX:
630     case PHB_DMARD_SYNC:
631        break;
632 
633     /* Noise on anything else */
634     default:
635         qemu_log_mask(LOG_UNIMP, "phb4: reg_write 0x%"PRIx64"=%"PRIx64"\n",
636                       off, val);
637     }
638 }
639 
640 static uint64_t pnv_phb4_reg_read(void *opaque, hwaddr off, unsigned size)
641 {
642     PnvPHB4 *phb = PNV_PHB4(opaque);
643     uint64_t val;
644 
645     if ((off & 0xfffc) == PHB_CONFIG_DATA) {
646         return pnv_phb4_config_read(phb, off & 0x3, size);
647     }
648 
649     /* Special case RC configuration space */
650     if ((off & 0xf800) == PHB_RC_CONFIG_BASE) {
651         return pnv_phb4_rc_config_read(phb, off & 0x7ff, size);
652     }
653 
654     /* Other registers are 64-bit only */
655     if (size != 8 || off & 0x7) {
656         phb_error(phb, "Invalid register access, offset: 0x%"PRIx64" size: %d",
657                    off, size);
658         return ~0ull;
659     }
660 
661     /* Default read from cache */
662     val = phb->regs[off >> 3];
663 
664     switch (off) {
665     case PHB_VERSION:
666         return phb->version;
667 
668         /* Read-only */
669     case PHB_PHB4_GEN_CAP:
670         return 0xe4b8000000000000ull;
671     case PHB_PHB4_TCE_CAP:
672         return phb->big_phb ? 0x4008440000000400ull : 0x2008440000000200ull;
673     case PHB_PHB4_IRQ_CAP:
674         return phb->big_phb ? 0x0800000000001000ull : 0x0800000000000800ull;
675     case PHB_PHB4_EEH_CAP:
676         return phb->big_phb ? 0x2000000000000000ull : 0x1000000000000000ull;
677 
678     /* IODA table accesses */
679     case PHB_IODA_DATA0:
680         return pnv_phb4_ioda_read(phb);
681 
682     /* Link training always appears trained */
683     case PHB_PCIE_DLP_TRAIN_CTL:
684         /* TODO: Do something sensible with speed ? */
685         return PHB_PCIE_DLP_INBAND_PRESENCE | PHB_PCIE_DLP_TL_LINKACT;
686 
687     /* DMA read sync: make it look like it's complete */
688     case PHB_DMARD_SYNC:
689         return PHB_DMARD_SYNC_COMPLETE;
690 
691     /* Silent simple reads */
692     case PHB_LSI_SOURCE_ID:
693     case PHB_CPU_LOADSTORE_STATUS:
694     case PHB_ASN_CMPM:
695     case PHB_PHB4_CONFIG:
696     case PHB_M32_START_ADDR:
697     case PHB_CONFIG_ADDRESS:
698     case PHB_IODA_ADDR:
699     case PHB_RTC_INVALIDATE:
700     case PHB_TCE_KILL:
701     case PHB_TCE_SPEC_CTL:
702     case PHB_PEST_BAR:
703     case PHB_PELTV_BAR:
704     case PHB_RTT_BAR:
705     case PHB_M64_UPPER_BITS:
706     case PHB_CTRLR:
707     case PHB_LEM_FIR_ACCUM:
708     case PHB_LEM_ERROR_MASK:
709     case PHB_LEM_ACTION0:
710     case PHB_LEM_ACTION1:
711     case PHB_TCE_TAG_ENABLE:
712     case PHB_INT_NOTIFY_ADDR:
713     case PHB_INT_NOTIFY_INDEX:
714     case PHB_Q_DMA_R:
715     case PHB_ETU_ERR_SUMMARY:
716         break;
717 
718     /* Noise on anything else */
719     default:
720         qemu_log_mask(LOG_UNIMP, "phb4: reg_read 0x%"PRIx64"=%"PRIx64"\n",
721                       off, val);
722     }
723     return val;
724 }
725 
726 static const MemoryRegionOps pnv_phb4_reg_ops = {
727     .read = pnv_phb4_reg_read,
728     .write = pnv_phb4_reg_write,
729     .valid.min_access_size = 1,
730     .valid.max_access_size = 8,
731     .impl.min_access_size = 1,
732     .impl.max_access_size = 8,
733     .endianness = DEVICE_BIG_ENDIAN,
734 };
735 
736 static uint64_t pnv_phb4_xscom_read(void *opaque, hwaddr addr, unsigned size)
737 {
738     PnvPHB4 *phb = PNV_PHB4(opaque);
739     uint32_t reg = addr >> 3;
740     uint64_t val;
741     hwaddr offset;
742 
743     switch (reg) {
744     case PHB_SCOM_HV_IND_ADDR:
745         return phb->scom_hv_ind_addr_reg;
746 
747     case PHB_SCOM_HV_IND_DATA:
748         if (!(phb->scom_hv_ind_addr_reg & PHB_SCOM_HV_IND_ADDR_VALID)) {
749             phb_error(phb, "Invalid indirect address");
750             return ~0ull;
751         }
752         size = (phb->scom_hv_ind_addr_reg & PHB_SCOM_HV_IND_ADDR_4B) ? 4 : 8;
753         offset = GETFIELD(PHB_SCOM_HV_IND_ADDR_ADDR, phb->scom_hv_ind_addr_reg);
754         val = pnv_phb4_reg_read(phb, offset, size);
755         if (phb->scom_hv_ind_addr_reg & PHB_SCOM_HV_IND_ADDR_AUTOINC) {
756             offset += size;
757             offset &= 0x3fff;
758             phb->scom_hv_ind_addr_reg = SETFIELD(PHB_SCOM_HV_IND_ADDR_ADDR,
759                                                  phb->scom_hv_ind_addr_reg,
760                                                  offset);
761         }
762         return val;
763     case PHB_SCOM_ETU_LEM_FIR:
764     case PHB_SCOM_ETU_LEM_FIR_AND:
765     case PHB_SCOM_ETU_LEM_FIR_OR:
766     case PHB_SCOM_ETU_LEM_FIR_MSK:
767     case PHB_SCOM_ETU_LEM_ERR_MSK_AND:
768     case PHB_SCOM_ETU_LEM_ERR_MSK_OR:
769     case PHB_SCOM_ETU_LEM_ACT0:
770     case PHB_SCOM_ETU_LEM_ACT1:
771     case PHB_SCOM_ETU_LEM_WOF:
772         offset = ((reg - PHB_SCOM_ETU_LEM_FIR) << 3) + PHB_LEM_FIR_ACCUM;
773         return pnv_phb4_reg_read(phb, offset, size);
774     case PHB_SCOM_ETU_PMON_CONFIG:
775     case PHB_SCOM_ETU_PMON_CTR0:
776     case PHB_SCOM_ETU_PMON_CTR1:
777     case PHB_SCOM_ETU_PMON_CTR2:
778     case PHB_SCOM_ETU_PMON_CTR3:
779         offset = ((reg - PHB_SCOM_ETU_PMON_CONFIG) << 3) + PHB_PERFMON_CONFIG;
780         return pnv_phb4_reg_read(phb, offset, size);
781 
782     default:
783         qemu_log_mask(LOG_UNIMP, "phb4: xscom_read 0x%"HWADDR_PRIx"\n", addr);
784         return ~0ull;
785     }
786 }
787 
788 static void pnv_phb4_xscom_write(void *opaque, hwaddr addr,
789                                  uint64_t val, unsigned size)
790 {
791     PnvPHB4 *phb = PNV_PHB4(opaque);
792     uint32_t reg = addr >> 3;
793     hwaddr offset;
794 
795     switch (reg) {
796     case PHB_SCOM_HV_IND_ADDR:
797         phb->scom_hv_ind_addr_reg = val & 0xe000000000001fff;
798         break;
799     case PHB_SCOM_HV_IND_DATA:
800         if (!(phb->scom_hv_ind_addr_reg & PHB_SCOM_HV_IND_ADDR_VALID)) {
801             phb_error(phb, "Invalid indirect address");
802             break;
803         }
804         size = (phb->scom_hv_ind_addr_reg & PHB_SCOM_HV_IND_ADDR_4B) ? 4 : 8;
805         offset = GETFIELD(PHB_SCOM_HV_IND_ADDR_ADDR, phb->scom_hv_ind_addr_reg);
806         pnv_phb4_reg_write(phb, offset, val, size);
807         if (phb->scom_hv_ind_addr_reg & PHB_SCOM_HV_IND_ADDR_AUTOINC) {
808             offset += size;
809             offset &= 0x3fff;
810             phb->scom_hv_ind_addr_reg = SETFIELD(PHB_SCOM_HV_IND_ADDR_ADDR,
811                                                  phb->scom_hv_ind_addr_reg,
812                                                  offset);
813         }
814         break;
815     case PHB_SCOM_ETU_LEM_FIR:
816     case PHB_SCOM_ETU_LEM_FIR_AND:
817     case PHB_SCOM_ETU_LEM_FIR_OR:
818     case PHB_SCOM_ETU_LEM_FIR_MSK:
819     case PHB_SCOM_ETU_LEM_ERR_MSK_AND:
820     case PHB_SCOM_ETU_LEM_ERR_MSK_OR:
821     case PHB_SCOM_ETU_LEM_ACT0:
822     case PHB_SCOM_ETU_LEM_ACT1:
823     case PHB_SCOM_ETU_LEM_WOF:
824         offset = ((reg - PHB_SCOM_ETU_LEM_FIR) << 3) + PHB_LEM_FIR_ACCUM;
825         pnv_phb4_reg_write(phb, offset, val, size);
826         break;
827     case PHB_SCOM_ETU_PMON_CONFIG:
828     case PHB_SCOM_ETU_PMON_CTR0:
829     case PHB_SCOM_ETU_PMON_CTR1:
830     case PHB_SCOM_ETU_PMON_CTR2:
831     case PHB_SCOM_ETU_PMON_CTR3:
832         offset = ((reg - PHB_SCOM_ETU_PMON_CONFIG) << 3) + PHB_PERFMON_CONFIG;
833         pnv_phb4_reg_write(phb, offset, val, size);
834         break;
835     default:
836         qemu_log_mask(LOG_UNIMP, "phb4: xscom_write 0x%"HWADDR_PRIx
837                       "=%"PRIx64"\n", addr, val);
838     }
839 }
840 
841 const MemoryRegionOps pnv_phb4_xscom_ops = {
842     .read = pnv_phb4_xscom_read,
843     .write = pnv_phb4_xscom_write,
844     .valid.min_access_size = 8,
845     .valid.max_access_size = 8,
846     .impl.min_access_size = 8,
847     .impl.max_access_size = 8,
848     .endianness = DEVICE_BIG_ENDIAN,
849 };
850 
851 static int pnv_phb4_map_irq(PCIDevice *pci_dev, int irq_num)
852 {
853     /* Check that out properly ... */
854     return irq_num & 3;
855 }
856 
857 static void pnv_phb4_set_irq(void *opaque, int irq_num, int level)
858 {
859     PnvPHB4 *phb = PNV_PHB4(opaque);
860     uint32_t lsi_base;
861 
862     /* LSI only ... */
863     if (irq_num > 3) {
864         phb_error(phb, "IRQ %x is not an LSI", irq_num);
865     }
866     lsi_base = GETFIELD(PHB_LSI_SRC_ID, phb->regs[PHB_LSI_SOURCE_ID >> 3]);
867     lsi_base <<= 3;
868     qemu_set_irq(phb->qirqs[lsi_base + irq_num], level);
869 }
870 
871 static bool pnv_phb4_resolve_pe(PnvPhb4DMASpace *ds)
872 {
873     uint64_t rtt, addr;
874     uint16_t rte;
875     int bus_num;
876     int num_PEs;
877 
878     /* Already resolved ? */
879     if (ds->pe_num != PHB_INVALID_PE) {
880         return true;
881     }
882 
883     /* We need to lookup the RTT */
884     rtt = ds->phb->regs[PHB_RTT_BAR >> 3];
885     if (!(rtt & PHB_RTT_BAR_ENABLE)) {
886         phb_error(ds->phb, "DMA with RTT BAR disabled !");
887         /* Set error bits ? fence ? ... */
888         return false;
889     }
890 
891     /* Read RTE */
892     bus_num = pci_bus_num(ds->bus);
893     addr = rtt & PHB_RTT_BASE_ADDRESS_MASK;
894     addr += 2 * PCI_BUILD_BDF(bus_num, ds->devfn);
895     if (dma_memory_read(&address_space_memory, addr, &rte,
896                         sizeof(rte), MEMTXATTRS_UNSPECIFIED)) {
897         phb_error(ds->phb, "Failed to read RTT entry at 0x%"PRIx64, addr);
898         /* Set error bits ? fence ? ... */
899         return false;
900     }
901     rte = be16_to_cpu(rte);
902 
903     /* Fail upon reading of invalid PE# */
904     num_PEs = ds->phb->big_phb ? PNV_PHB4_MAX_PEs : (PNV_PHB4_MAX_PEs >> 1);
905     if (rte >= num_PEs) {
906         phb_error(ds->phb, "RTE for RID 0x%x invalid (%04x", ds->devfn, rte);
907         rte &= num_PEs - 1;
908     }
909     ds->pe_num = rte;
910     return true;
911 }
912 
913 static void pnv_phb4_translate_tve(PnvPhb4DMASpace *ds, hwaddr addr,
914                                    bool is_write, uint64_t tve,
915                                    IOMMUTLBEntry *tlb)
916 {
917     uint64_t tta = GETFIELD(IODA3_TVT_TABLE_ADDR, tve);
918     int32_t  lev = GETFIELD(IODA3_TVT_NUM_LEVELS, tve);
919     uint32_t tts = GETFIELD(IODA3_TVT_TCE_TABLE_SIZE, tve);
920     uint32_t tps = GETFIELD(IODA3_TVT_IO_PSIZE, tve);
921 
922     /* Invalid levels */
923     if (lev > 4) {
924         phb_error(ds->phb, "Invalid #levels in TVE %d", lev);
925         return;
926     }
927 
928     /* Invalid entry */
929     if (tts == 0) {
930         phb_error(ds->phb, "Access to invalid TVE");
931         return;
932     }
933 
934     /* IO Page Size of 0 means untranslated, else use TCEs */
935     if (tps == 0) {
936         /* TODO: Handle boundaries */
937 
938         /* Use 4k pages like q35 ... for now */
939         tlb->iova = addr & 0xfffffffffffff000ull;
940         tlb->translated_addr = addr & 0x0003fffffffff000ull;
941         tlb->addr_mask = 0xfffull;
942         tlb->perm = IOMMU_RW;
943     } else {
944         uint32_t tce_shift, tbl_shift, sh;
945         uint64_t base, taddr, tce, tce_mask;
946 
947         /* Address bits per bottom level TCE entry */
948         tce_shift = tps + 11;
949 
950         /* Address bits per table level */
951         tbl_shift = tts + 8;
952 
953         /* Top level table base address */
954         base = tta << 12;
955 
956         /* Total shift to first level */
957         sh = tbl_shift * lev + tce_shift;
958 
959         /* TODO: Limit to support IO page sizes */
960 
961         /* TODO: Multi-level untested */
962         while ((lev--) >= 0) {
963             /* Grab the TCE address */
964             taddr = base | (((addr >> sh) & ((1ul << tbl_shift) - 1)) << 3);
965             if (dma_memory_read(&address_space_memory, taddr, &tce,
966                                 sizeof(tce), MEMTXATTRS_UNSPECIFIED)) {
967                 phb_error(ds->phb, "Failed to read TCE at 0x%"PRIx64, taddr);
968                 return;
969             }
970             tce = be64_to_cpu(tce);
971 
972             /* Check permission for indirect TCE */
973             if ((lev >= 0) && !(tce & 3)) {
974                 phb_error(ds->phb, "Invalid indirect TCE at 0x%"PRIx64, taddr);
975                 phb_error(ds->phb, " xlate %"PRIx64":%c TVE=%"PRIx64, addr,
976                            is_write ? 'W' : 'R', tve);
977                 phb_error(ds->phb, " tta=%"PRIx64" lev=%d tts=%d tps=%d",
978                            tta, lev, tts, tps);
979                 return;
980             }
981             sh -= tbl_shift;
982             base = tce & ~0xfffull;
983         }
984 
985         /* We exit the loop with TCE being the final TCE */
986         tce_mask = ~((1ull << tce_shift) - 1);
987         tlb->iova = addr & tce_mask;
988         tlb->translated_addr = tce & tce_mask;
989         tlb->addr_mask = ~tce_mask;
990         tlb->perm = tce & 3;
991         if ((is_write & !(tce & 2)) || ((!is_write) && !(tce & 1))) {
992             phb_error(ds->phb, "TCE access fault at 0x%"PRIx64, taddr);
993             phb_error(ds->phb, " xlate %"PRIx64":%c TVE=%"PRIx64, addr,
994                        is_write ? 'W' : 'R', tve);
995             phb_error(ds->phb, " tta=%"PRIx64" lev=%d tts=%d tps=%d",
996                        tta, lev, tts, tps);
997         }
998     }
999 }
1000 
1001 static IOMMUTLBEntry pnv_phb4_translate_iommu(IOMMUMemoryRegion *iommu,
1002                                               hwaddr addr,
1003                                               IOMMUAccessFlags flag,
1004                                               int iommu_idx)
1005 {
1006     PnvPhb4DMASpace *ds = container_of(iommu, PnvPhb4DMASpace, dma_mr);
1007     int tve_sel;
1008     uint64_t tve, cfg;
1009     IOMMUTLBEntry ret = {
1010         .target_as = &address_space_memory,
1011         .iova = addr,
1012         .translated_addr = 0,
1013         .addr_mask = ~(hwaddr)0,
1014         .perm = IOMMU_NONE,
1015     };
1016 
1017     /* Resolve PE# */
1018     if (!pnv_phb4_resolve_pe(ds)) {
1019         phb_error(ds->phb, "Failed to resolve PE# for bus @%p (%d) devfn 0x%x",
1020                    ds->bus, pci_bus_num(ds->bus), ds->devfn);
1021         return ret;
1022     }
1023 
1024     /* Check top bits */
1025     switch (addr >> 60) {
1026     case 00:
1027         /* DMA or 32-bit MSI ? */
1028         cfg = ds->phb->regs[PHB_PHB4_CONFIG >> 3];
1029         if ((cfg & PHB_PHB4C_32BIT_MSI_EN) &&
1030             ((addr & 0xffffffffffff0000ull) == 0xffff0000ull)) {
1031             phb_error(ds->phb, "xlate on 32-bit MSI region");
1032             return ret;
1033         }
1034         /* Choose TVE XXX Use PHB4 Control Register */
1035         tve_sel = (addr >> 59) & 1;
1036         tve = ds->phb->ioda_TVT[ds->pe_num * 2 + tve_sel];
1037         pnv_phb4_translate_tve(ds, addr, flag & IOMMU_WO, tve, &ret);
1038         break;
1039     case 01:
1040         phb_error(ds->phb, "xlate on 64-bit MSI region");
1041         break;
1042     default:
1043         phb_error(ds->phb, "xlate on unsupported address 0x%"PRIx64, addr);
1044     }
1045     return ret;
1046 }
1047 
1048 #define TYPE_PNV_PHB4_IOMMU_MEMORY_REGION "pnv-phb4-iommu-memory-region"
1049 DECLARE_INSTANCE_CHECKER(IOMMUMemoryRegion, PNV_PHB4_IOMMU_MEMORY_REGION,
1050                          TYPE_PNV_PHB4_IOMMU_MEMORY_REGION)
1051 
1052 static void pnv_phb4_iommu_memory_region_class_init(ObjectClass *klass,
1053                                                     void *data)
1054 {
1055     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_CLASS(klass);
1056 
1057     imrc->translate = pnv_phb4_translate_iommu;
1058 }
1059 
1060 static const TypeInfo pnv_phb4_iommu_memory_region_info = {
1061     .parent = TYPE_IOMMU_MEMORY_REGION,
1062     .name = TYPE_PNV_PHB4_IOMMU_MEMORY_REGION,
1063     .class_init = pnv_phb4_iommu_memory_region_class_init,
1064 };
1065 
1066 /*
1067  * MSI/MSIX memory region implementation.
1068  * The handler handles both MSI and MSIX.
1069  */
1070 static void pnv_phb4_msi_write(void *opaque, hwaddr addr,
1071                                uint64_t data, unsigned size)
1072 {
1073     PnvPhb4DMASpace *ds = opaque;
1074     PnvPHB4 *phb = ds->phb;
1075 
1076     uint32_t src = ((addr >> 4) & 0xffff) | (data & 0x1f);
1077 
1078     /* Resolve PE# */
1079     if (!pnv_phb4_resolve_pe(ds)) {
1080         phb_error(phb, "Failed to resolve PE# for bus @%p (%d) devfn 0x%x",
1081                    ds->bus, pci_bus_num(ds->bus), ds->devfn);
1082         return;
1083     }
1084 
1085     /* TODO: Check it doesn't collide with LSIs */
1086     if (src >= phb->xsrc.nr_irqs) {
1087         phb_error(phb, "MSI %d out of bounds", src);
1088         return;
1089     }
1090 
1091     /* TODO: check PE/MSI assignement */
1092 
1093     qemu_irq_pulse(phb->qirqs[src]);
1094 }
1095 
1096 /* There is no .read as the read result is undefined by PCI spec */
1097 static uint64_t pnv_phb4_msi_read(void *opaque, hwaddr addr, unsigned size)
1098 {
1099     PnvPhb4DMASpace *ds = opaque;
1100 
1101     phb_error(ds->phb, "Invalid MSI read @ 0x%" HWADDR_PRIx, addr);
1102     return -1;
1103 }
1104 
1105 static const MemoryRegionOps pnv_phb4_msi_ops = {
1106     .read = pnv_phb4_msi_read,
1107     .write = pnv_phb4_msi_write,
1108     .endianness = DEVICE_LITTLE_ENDIAN
1109 };
1110 
1111 static PnvPhb4DMASpace *pnv_phb4_dma_find(PnvPHB4 *phb, PCIBus *bus, int devfn)
1112 {
1113     PnvPhb4DMASpace *ds;
1114 
1115     QLIST_FOREACH(ds, &phb->dma_spaces, list) {
1116         if (ds->bus == bus && ds->devfn == devfn) {
1117             break;
1118         }
1119     }
1120     return ds;
1121 }
1122 
1123 static AddressSpace *pnv_phb4_dma_iommu(PCIBus *bus, void *opaque, int devfn)
1124 {
1125     PnvPHB4 *phb = opaque;
1126     PnvPhb4DMASpace *ds;
1127     char name[32];
1128 
1129     ds = pnv_phb4_dma_find(phb, bus, devfn);
1130 
1131     if (ds == NULL) {
1132         ds = g_malloc0(sizeof(PnvPhb4DMASpace));
1133         ds->bus = bus;
1134         ds->devfn = devfn;
1135         ds->pe_num = PHB_INVALID_PE;
1136         ds->phb = phb;
1137         snprintf(name, sizeof(name), "phb4-%d.%d-iommu", phb->chip_id,
1138                  phb->phb_id);
1139         memory_region_init_iommu(&ds->dma_mr, sizeof(ds->dma_mr),
1140                                  TYPE_PNV_PHB4_IOMMU_MEMORY_REGION,
1141                                  OBJECT(phb), name, UINT64_MAX);
1142         address_space_init(&ds->dma_as, MEMORY_REGION(&ds->dma_mr),
1143                            name);
1144         memory_region_init_io(&ds->msi32_mr, OBJECT(phb), &pnv_phb4_msi_ops,
1145                               ds, "msi32", 0x10000);
1146         memory_region_init_io(&ds->msi64_mr, OBJECT(phb), &pnv_phb4_msi_ops,
1147                               ds, "msi64", 0x100000);
1148         pnv_phb4_update_msi_regions(ds);
1149 
1150         QLIST_INSERT_HEAD(&phb->dma_spaces, ds, list);
1151     }
1152     return &ds->dma_as;
1153 }
1154 
1155 static void pnv_phb4_instance_init(Object *obj)
1156 {
1157     PnvPHB4 *phb = PNV_PHB4(obj);
1158 
1159     QLIST_INIT(&phb->dma_spaces);
1160 
1161     /* XIVE interrupt source object */
1162     object_initialize_child(obj, "source", &phb->xsrc, TYPE_XIVE_SOURCE);
1163 }
1164 
1165 static void pnv_phb4_realize(DeviceState *dev, Error **errp)
1166 {
1167     PnvPHB4 *phb = PNV_PHB4(dev);
1168     PCIHostState *pci = PCI_HOST_BRIDGE(dev);
1169     XiveSource *xsrc = &phb->xsrc;
1170     int nr_irqs;
1171     char name[32];
1172 
1173     assert(phb->stack);
1174 
1175     /* Set the "big_phb" flag */
1176     phb->big_phb = phb->phb_id == 0 || phb->phb_id == 3;
1177 
1178     /* Controller Registers */
1179     snprintf(name, sizeof(name), "phb4-%d.%d-regs", phb->chip_id,
1180              phb->phb_id);
1181     memory_region_init_io(&phb->mr_regs, OBJECT(phb), &pnv_phb4_reg_ops, phb,
1182                           name, 0x2000);
1183 
1184     /*
1185      * PHB4 doesn't support IO space. However, qemu gets very upset if
1186      * we don't have an IO region to anchor IO BARs onto so we just
1187      * initialize one which we never hook up to anything
1188      */
1189 
1190     snprintf(name, sizeof(name), "phb4-%d.%d-pci-io", phb->chip_id,
1191              phb->phb_id);
1192     memory_region_init(&phb->pci_io, OBJECT(phb), name, 0x10000);
1193 
1194     snprintf(name, sizeof(name), "phb4-%d.%d-pci-mmio", phb->chip_id,
1195              phb->phb_id);
1196     memory_region_init(&phb->pci_mmio, OBJECT(phb), name,
1197                        PCI_MMIO_TOTAL_SIZE);
1198 
1199     pci->bus = pci_register_root_bus(dev, dev->id,
1200                                      pnv_phb4_set_irq, pnv_phb4_map_irq, phb,
1201                                      &phb->pci_mmio, &phb->pci_io,
1202                                      0, 4, TYPE_PNV_PHB4_ROOT_BUS);
1203     pci_setup_iommu(pci->bus, pnv_phb4_dma_iommu, phb);
1204     pci->bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
1205 
1206     /* Add a single Root port if running with defaults */
1207     if (defaults_enabled()) {
1208         pnv_phb_attach_root_port(PCI_HOST_BRIDGE(phb),
1209                                  TYPE_PNV_PHB4_ROOT_PORT);
1210     }
1211 
1212     /* Setup XIVE Source */
1213     if (phb->big_phb) {
1214         nr_irqs = PNV_PHB4_MAX_INTs;
1215     } else {
1216         nr_irqs = PNV_PHB4_MAX_INTs >> 1;
1217     }
1218     object_property_set_int(OBJECT(xsrc), "nr-irqs", nr_irqs, &error_fatal);
1219     object_property_set_link(OBJECT(xsrc), "xive", OBJECT(phb), &error_fatal);
1220     if (!qdev_realize(DEVICE(xsrc), NULL, errp)) {
1221         return;
1222     }
1223 
1224     pnv_phb4_update_xsrc(phb);
1225 
1226     phb->qirqs = qemu_allocate_irqs(xive_source_set_irq, xsrc, xsrc->nr_irqs);
1227 }
1228 
1229 static const char *pnv_phb4_root_bus_path(PCIHostState *host_bridge,
1230                                           PCIBus *rootbus)
1231 {
1232     PnvPHB4 *phb = PNV_PHB4(host_bridge);
1233 
1234     snprintf(phb->bus_path, sizeof(phb->bus_path), "00%02x:%02x",
1235              phb->chip_id, phb->phb_id);
1236     return phb->bus_path;
1237 }
1238 
1239 static void pnv_phb4_xive_notify(XiveNotifier *xf, uint32_t srcno)
1240 {
1241     PnvPHB4 *phb = PNV_PHB4(xf);
1242     uint64_t notif_port = phb->regs[PHB_INT_NOTIFY_ADDR >> 3];
1243     uint32_t offset = phb->regs[PHB_INT_NOTIFY_INDEX >> 3];
1244     uint64_t data = XIVE_TRIGGER_PQ | offset | srcno;
1245     MemTxResult result;
1246 
1247     trace_pnv_phb4_xive_notify(notif_port, data);
1248 
1249     address_space_stq_be(&address_space_memory, notif_port, data,
1250                          MEMTXATTRS_UNSPECIFIED, &result);
1251     if (result != MEMTX_OK) {
1252         phb_error(phb, "trigger failed @%"HWADDR_PRIx "\n", notif_port);
1253         return;
1254     }
1255 }
1256 
1257 static Property pnv_phb4_properties[] = {
1258         DEFINE_PROP_UINT32("index", PnvPHB4, phb_id, 0),
1259         DEFINE_PROP_UINT32("chip-id", PnvPHB4, chip_id, 0),
1260         DEFINE_PROP_UINT64("version", PnvPHB4, version, 0),
1261         DEFINE_PROP_LINK("stack", PnvPHB4, stack, TYPE_PNV_PHB4_PEC_STACK,
1262                          PnvPhb4PecStack *),
1263         DEFINE_PROP_END_OF_LIST(),
1264 };
1265 
1266 static void pnv_phb4_class_init(ObjectClass *klass, void *data)
1267 {
1268     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(klass);
1269     DeviceClass *dc = DEVICE_CLASS(klass);
1270     XiveNotifierClass *xfc = XIVE_NOTIFIER_CLASS(klass);
1271 
1272     hc->root_bus_path   = pnv_phb4_root_bus_path;
1273     dc->realize         = pnv_phb4_realize;
1274     device_class_set_props(dc, pnv_phb4_properties);
1275     set_bit(DEVICE_CATEGORY_BRIDGE, dc->categories);
1276     dc->user_creatable  = false;
1277 
1278     xfc->notify         = pnv_phb4_xive_notify;
1279 }
1280 
1281 static const TypeInfo pnv_phb4_type_info = {
1282     .name          = TYPE_PNV_PHB4,
1283     .parent        = TYPE_PCIE_HOST_BRIDGE,
1284     .instance_init = pnv_phb4_instance_init,
1285     .instance_size = sizeof(PnvPHB4),
1286     .class_init    = pnv_phb4_class_init,
1287     .interfaces = (InterfaceInfo[]) {
1288             { TYPE_XIVE_NOTIFIER },
1289             { },
1290     }
1291 };
1292 
1293 static void pnv_phb4_root_bus_class_init(ObjectClass *klass, void *data)
1294 {
1295     BusClass *k = BUS_CLASS(klass);
1296 
1297     /*
1298      * PHB4 has only a single root complex. Enforce the limit on the
1299      * parent bus
1300      */
1301     k->max_dev = 1;
1302 }
1303 
1304 static const TypeInfo pnv_phb4_root_bus_info = {
1305     .name = TYPE_PNV_PHB4_ROOT_BUS,
1306     .parent = TYPE_PCIE_BUS,
1307     .class_init = pnv_phb4_root_bus_class_init,
1308     .interfaces = (InterfaceInfo[]) {
1309         { INTERFACE_PCIE_DEVICE },
1310         { }
1311     },
1312 };
1313 
1314 static void pnv_phb4_root_port_reset(DeviceState *dev)
1315 {
1316     PCIERootPortClass *rpc = PCIE_ROOT_PORT_GET_CLASS(dev);
1317     PCIDevice *d = PCI_DEVICE(dev);
1318     uint8_t *conf = d->config;
1319 
1320     rpc->parent_reset(dev);
1321 
1322     pci_byte_test_and_set_mask(conf + PCI_IO_BASE,
1323                                PCI_IO_RANGE_MASK & 0xff);
1324     pci_byte_test_and_clear_mask(conf + PCI_IO_LIMIT,
1325                                  PCI_IO_RANGE_MASK & 0xff);
1326     pci_set_word(conf + PCI_MEMORY_BASE, 0);
1327     pci_set_word(conf + PCI_MEMORY_LIMIT, 0xfff0);
1328     pci_set_word(conf + PCI_PREF_MEMORY_BASE, 0x1);
1329     pci_set_word(conf + PCI_PREF_MEMORY_LIMIT, 0xfff1);
1330     pci_set_long(conf + PCI_PREF_BASE_UPPER32, 0x1); /* Hack */
1331     pci_set_long(conf + PCI_PREF_LIMIT_UPPER32, 0xffffffff);
1332 }
1333 
1334 static void pnv_phb4_root_port_realize(DeviceState *dev, Error **errp)
1335 {
1336     PCIERootPortClass *rpc = PCIE_ROOT_PORT_GET_CLASS(dev);
1337     PCIDevice *pci = PCI_DEVICE(dev);
1338     PCIBus *bus = pci_get_bus(pci);
1339     PnvPHB4 *phb = NULL;
1340     Error *local_err = NULL;
1341 
1342     phb = (PnvPHB4 *) object_dynamic_cast(OBJECT(bus->qbus.parent),
1343                                           TYPE_PNV_PHB4);
1344 
1345     if (!phb) {
1346         error_setg(errp, "%s must be connected to pnv-phb4 buses", dev->id);
1347         return;
1348     }
1349 
1350     /* Set unique chassis/slot values for the root port */
1351     qdev_prop_set_uint8(&pci->qdev, "chassis", phb->chip_id);
1352     qdev_prop_set_uint16(&pci->qdev, "slot", phb->phb_id);
1353 
1354     rpc->parent_realize(dev, &local_err);
1355     if (local_err) {
1356         error_propagate(errp, local_err);
1357         return;
1358     }
1359 }
1360 
1361 static void pnv_phb4_root_port_class_init(ObjectClass *klass, void *data)
1362 {
1363     DeviceClass *dc = DEVICE_CLASS(klass);
1364     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
1365     PCIERootPortClass *rpc = PCIE_ROOT_PORT_CLASS(klass);
1366 
1367     dc->desc     = "IBM PHB4 PCIE Root Port";
1368     dc->user_creatable = true;
1369 
1370     device_class_set_parent_realize(dc, pnv_phb4_root_port_realize,
1371                                     &rpc->parent_realize);
1372     device_class_set_parent_reset(dc, pnv_phb4_root_port_reset,
1373                                   &rpc->parent_reset);
1374 
1375     k->vendor_id = PCI_VENDOR_ID_IBM;
1376     k->device_id = PNV_PHB4_DEVICE_ID;
1377     k->revision  = 0;
1378 
1379     rpc->exp_offset = 0x48;
1380     rpc->aer_offset = 0x100;
1381 
1382     dc->reset = &pnv_phb4_root_port_reset;
1383 }
1384 
1385 static const TypeInfo pnv_phb4_root_port_info = {
1386     .name          = TYPE_PNV_PHB4_ROOT_PORT,
1387     .parent        = TYPE_PCIE_ROOT_PORT,
1388     .instance_size = sizeof(PnvPHB4RootPort),
1389     .class_init    = pnv_phb4_root_port_class_init,
1390 };
1391 
1392 static void pnv_phb4_register_types(void)
1393 {
1394     type_register_static(&pnv_phb4_root_bus_info);
1395     type_register_static(&pnv_phb4_root_port_info);
1396     type_register_static(&pnv_phb4_type_info);
1397     type_register_static(&pnv_phb4_iommu_memory_region_info);
1398 }
1399 
1400 type_init(pnv_phb4_register_types);
1401 
1402 void pnv_phb4_update_regions(PnvPhb4PecStack *stack)
1403 {
1404     PnvPHB4 *phb = &stack->phb;
1405 
1406     /* Unmap first always */
1407     if (memory_region_is_mapped(&phb->mr_regs)) {
1408         memory_region_del_subregion(&stack->phbbar, &phb->mr_regs);
1409     }
1410     if (memory_region_is_mapped(&phb->xsrc.esb_mmio)) {
1411         memory_region_del_subregion(&stack->intbar, &phb->xsrc.esb_mmio);
1412     }
1413 
1414     /* Map registers if enabled */
1415     if (memory_region_is_mapped(&stack->phbbar)) {
1416         memory_region_add_subregion(&stack->phbbar, 0, &phb->mr_regs);
1417     }
1418 
1419     /* Map ESB if enabled */
1420     if (memory_region_is_mapped(&stack->intbar)) {
1421         memory_region_add_subregion(&stack->intbar, 0, &phb->xsrc.esb_mmio);
1422     }
1423 
1424     /* Check/update m32 */
1425     pnv_phb4_check_all_mbt(phb);
1426 }
1427 
1428 void pnv_phb4_pic_print_info(PnvPHB4 *phb, Monitor *mon)
1429 {
1430     uint32_t offset = phb->regs[PHB_INT_NOTIFY_INDEX >> 3];
1431 
1432     monitor_printf(mon, "PHB4[%x:%x] Source %08x .. %08x\n",
1433                    phb->chip_id, phb->phb_id,
1434                    offset, offset + phb->xsrc.nr_irqs - 1);
1435     xive_source_pic_print_info(&phb->xsrc, 0, mon);
1436 }
1437