xref: /openbmc/qemu/hw/ppc/spapr_pci.c (revision 3979fca4)
1 /*
2  * QEMU sPAPR PCI host originated from Uninorth PCI host
3  *
4  * Copyright (c) 2011 Alexey Kardashevskiy, IBM Corporation.
5  * Copyright (C) 2011 David Gibson, IBM Corporation.
6  *
7  * Permission is hereby granted, free of charge, to any person obtaining a copy
8  * of this software and associated documentation files (the "Software"), to deal
9  * in the Software without restriction, including without limitation the rights
10  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11  * copies of the Software, and to permit persons to whom the Software is
12  * furnished to do so, subject to the following conditions:
13  *
14  * The above copyright notice and this permission notice shall be included in
15  * all copies or substantial portions of the Software.
16  *
17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
23  * THE SOFTWARE.
24  */
25 #include "qemu/osdep.h"
26 #include "qapi/error.h"
27 #include "qemu-common.h"
28 #include "cpu.h"
29 #include "hw/hw.h"
30 #include "hw/sysbus.h"
31 #include "hw/pci/pci.h"
32 #include "hw/pci/msi.h"
33 #include "hw/pci/msix.h"
34 #include "hw/pci/pci_host.h"
35 #include "hw/ppc/spapr.h"
36 #include "hw/pci-host/spapr.h"
37 #include "exec/address-spaces.h"
38 #include "exec/ram_addr.h"
39 #include <libfdt.h>
40 #include "trace.h"
41 #include "qemu/error-report.h"
42 #include "qapi/qmp/qerror.h"
43 #include "hw/ppc/fdt.h"
44 #include "hw/pci/pci_bridge.h"
45 #include "hw/pci/pci_bus.h"
46 #include "hw/pci/pci_ids.h"
47 #include "hw/ppc/spapr_drc.h"
48 #include "sysemu/device_tree.h"
49 #include "sysemu/kvm.h"
50 #include "sysemu/hostmem.h"
51 #include "sysemu/numa.h"
52 
53 /* Copied from the kernel arch/powerpc/platforms/pseries/msi.c */
54 #define RTAS_QUERY_FN           0
55 #define RTAS_CHANGE_FN          1
56 #define RTAS_RESET_FN           2
57 #define RTAS_CHANGE_MSI_FN      3
58 #define RTAS_CHANGE_MSIX_FN     4
59 
60 /* Interrupt types to return on RTAS_CHANGE_* */
61 #define RTAS_TYPE_MSI           1
62 #define RTAS_TYPE_MSIX          2
63 
64 SpaprPhbState *spapr_pci_find_phb(SpaprMachineState *spapr, uint64_t buid)
65 {
66     SpaprPhbState *sphb;
67 
68     QLIST_FOREACH(sphb, &spapr->phbs, list) {
69         if (sphb->buid != buid) {
70             continue;
71         }
72         return sphb;
73     }
74 
75     return NULL;
76 }
77 
78 PCIDevice *spapr_pci_find_dev(SpaprMachineState *spapr, uint64_t buid,
79                               uint32_t config_addr)
80 {
81     SpaprPhbState *sphb = spapr_pci_find_phb(spapr, buid);
82     PCIHostState *phb = PCI_HOST_BRIDGE(sphb);
83     int bus_num = (config_addr >> 16) & 0xFF;
84     int devfn = (config_addr >> 8) & 0xFF;
85 
86     if (!phb) {
87         return NULL;
88     }
89 
90     return pci_find_device(phb->bus, bus_num, devfn);
91 }
92 
93 static uint32_t rtas_pci_cfgaddr(uint32_t arg)
94 {
95     /* This handles the encoding of extended config space addresses */
96     return ((arg >> 20) & 0xf00) | (arg & 0xff);
97 }
98 
99 static void finish_read_pci_config(SpaprMachineState *spapr, uint64_t buid,
100                                    uint32_t addr, uint32_t size,
101                                    target_ulong rets)
102 {
103     PCIDevice *pci_dev;
104     uint32_t val;
105 
106     if ((size != 1) && (size != 2) && (size != 4)) {
107         /* access must be 1, 2 or 4 bytes */
108         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
109         return;
110     }
111 
112     pci_dev = spapr_pci_find_dev(spapr, buid, addr);
113     addr = rtas_pci_cfgaddr(addr);
114 
115     if (!pci_dev || (addr % size) || (addr >= pci_config_size(pci_dev))) {
116         /* Access must be to a valid device, within bounds and
117          * naturally aligned */
118         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
119         return;
120     }
121 
122     val = pci_host_config_read_common(pci_dev, addr,
123                                       pci_config_size(pci_dev), size);
124 
125     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
126     rtas_st(rets, 1, val);
127 }
128 
129 static void rtas_ibm_read_pci_config(PowerPCCPU *cpu, SpaprMachineState *spapr,
130                                      uint32_t token, uint32_t nargs,
131                                      target_ulong args,
132                                      uint32_t nret, target_ulong rets)
133 {
134     uint64_t buid;
135     uint32_t size, addr;
136 
137     if ((nargs != 4) || (nret != 2)) {
138         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
139         return;
140     }
141 
142     buid = rtas_ldq(args, 1);
143     size = rtas_ld(args, 3);
144     addr = rtas_ld(args, 0);
145 
146     finish_read_pci_config(spapr, buid, addr, size, rets);
147 }
148 
149 static void rtas_read_pci_config(PowerPCCPU *cpu, SpaprMachineState *spapr,
150                                  uint32_t token, uint32_t nargs,
151                                  target_ulong args,
152                                  uint32_t nret, target_ulong rets)
153 {
154     uint32_t size, addr;
155 
156     if ((nargs != 2) || (nret != 2)) {
157         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
158         return;
159     }
160 
161     size = rtas_ld(args, 1);
162     addr = rtas_ld(args, 0);
163 
164     finish_read_pci_config(spapr, 0, addr, size, rets);
165 }
166 
167 static void finish_write_pci_config(SpaprMachineState *spapr, uint64_t buid,
168                                     uint32_t addr, uint32_t size,
169                                     uint32_t val, target_ulong rets)
170 {
171     PCIDevice *pci_dev;
172 
173     if ((size != 1) && (size != 2) && (size != 4)) {
174         /* access must be 1, 2 or 4 bytes */
175         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
176         return;
177     }
178 
179     pci_dev = spapr_pci_find_dev(spapr, buid, addr);
180     addr = rtas_pci_cfgaddr(addr);
181 
182     if (!pci_dev || (addr % size) || (addr >= pci_config_size(pci_dev))) {
183         /* Access must be to a valid device, within bounds and
184          * naturally aligned */
185         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
186         return;
187     }
188 
189     pci_host_config_write_common(pci_dev, addr, pci_config_size(pci_dev),
190                                  val, size);
191 
192     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
193 }
194 
195 static void rtas_ibm_write_pci_config(PowerPCCPU *cpu, SpaprMachineState *spapr,
196                                       uint32_t token, uint32_t nargs,
197                                       target_ulong args,
198                                       uint32_t nret, target_ulong rets)
199 {
200     uint64_t buid;
201     uint32_t val, size, addr;
202 
203     if ((nargs != 5) || (nret != 1)) {
204         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
205         return;
206     }
207 
208     buid = rtas_ldq(args, 1);
209     val = rtas_ld(args, 4);
210     size = rtas_ld(args, 3);
211     addr = rtas_ld(args, 0);
212 
213     finish_write_pci_config(spapr, buid, addr, size, val, rets);
214 }
215 
216 static void rtas_write_pci_config(PowerPCCPU *cpu, SpaprMachineState *spapr,
217                                   uint32_t token, uint32_t nargs,
218                                   target_ulong args,
219                                   uint32_t nret, target_ulong rets)
220 {
221     uint32_t val, size, addr;
222 
223     if ((nargs != 3) || (nret != 1)) {
224         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
225         return;
226     }
227 
228 
229     val = rtas_ld(args, 2);
230     size = rtas_ld(args, 1);
231     addr = rtas_ld(args, 0);
232 
233     finish_write_pci_config(spapr, 0, addr, size, val, rets);
234 }
235 
236 /*
237  * Set MSI/MSIX message data.
238  * This is required for msi_notify()/msix_notify() which
239  * will write at the addresses via spapr_msi_write().
240  *
241  * If hwaddr == 0, all entries will have .data == first_irq i.e.
242  * table will be reset.
243  */
244 static void spapr_msi_setmsg(PCIDevice *pdev, hwaddr addr, bool msix,
245                              unsigned first_irq, unsigned req_num)
246 {
247     unsigned i;
248     MSIMessage msg = { .address = addr, .data = first_irq };
249 
250     if (!msix) {
251         msi_set_message(pdev, msg);
252         trace_spapr_pci_msi_setup(pdev->name, 0, msg.address);
253         return;
254     }
255 
256     for (i = 0; i < req_num; ++i) {
257         msix_set_message(pdev, i, msg);
258         trace_spapr_pci_msi_setup(pdev->name, i, msg.address);
259         if (addr) {
260             ++msg.data;
261         }
262     }
263 }
264 
265 static void rtas_ibm_change_msi(PowerPCCPU *cpu, SpaprMachineState *spapr,
266                                 uint32_t token, uint32_t nargs,
267                                 target_ulong args, uint32_t nret,
268                                 target_ulong rets)
269 {
270     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
271     uint32_t config_addr = rtas_ld(args, 0);
272     uint64_t buid = rtas_ldq(args, 1);
273     unsigned int func = rtas_ld(args, 3);
274     unsigned int req_num = rtas_ld(args, 4); /* 0 == remove all */
275     unsigned int seq_num = rtas_ld(args, 5);
276     unsigned int ret_intr_type;
277     unsigned int irq, max_irqs = 0;
278     SpaprPhbState *phb = NULL;
279     PCIDevice *pdev = NULL;
280     spapr_pci_msi *msi;
281     int *config_addr_key;
282     Error *err = NULL;
283     int i;
284 
285     /* Fins SpaprPhbState */
286     phb = spapr_pci_find_phb(spapr, buid);
287     if (phb) {
288         pdev = spapr_pci_find_dev(spapr, buid, config_addr);
289     }
290     if (!phb || !pdev) {
291         rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
292         return;
293     }
294 
295     switch (func) {
296     case RTAS_CHANGE_FN:
297         if (msi_present(pdev)) {
298             ret_intr_type = RTAS_TYPE_MSI;
299         } else if (msix_present(pdev)) {
300             ret_intr_type = RTAS_TYPE_MSIX;
301         } else {
302             rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
303             return;
304         }
305         break;
306     case RTAS_CHANGE_MSI_FN:
307         if (msi_present(pdev)) {
308             ret_intr_type = RTAS_TYPE_MSI;
309         } else {
310             rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
311             return;
312         }
313         break;
314     case RTAS_CHANGE_MSIX_FN:
315         if (msix_present(pdev)) {
316             ret_intr_type = RTAS_TYPE_MSIX;
317         } else {
318             rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
319             return;
320         }
321         break;
322     default:
323         error_report("rtas_ibm_change_msi(%u) is not implemented", func);
324         rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
325         return;
326     }
327 
328     msi = (spapr_pci_msi *) g_hash_table_lookup(phb->msi, &config_addr);
329 
330     /* Releasing MSIs */
331     if (!req_num) {
332         if (!msi) {
333             trace_spapr_pci_msi("Releasing wrong config", config_addr);
334             rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
335             return;
336         }
337 
338         if (!smc->legacy_irq_allocation) {
339             spapr_irq_msi_free(spapr, msi->first_irq, msi->num);
340         }
341         spapr_irq_free(spapr, msi->first_irq, msi->num);
342         if (msi_present(pdev)) {
343             spapr_msi_setmsg(pdev, 0, false, 0, 0);
344         }
345         if (msix_present(pdev)) {
346             spapr_msi_setmsg(pdev, 0, true, 0, 0);
347         }
348         g_hash_table_remove(phb->msi, &config_addr);
349 
350         trace_spapr_pci_msi("Released MSIs", config_addr);
351         rtas_st(rets, 0, RTAS_OUT_SUCCESS);
352         rtas_st(rets, 1, 0);
353         return;
354     }
355 
356     /* Enabling MSI */
357 
358     /* Check if the device supports as many IRQs as requested */
359     if (ret_intr_type == RTAS_TYPE_MSI) {
360         max_irqs = msi_nr_vectors_allocated(pdev);
361     } else if (ret_intr_type == RTAS_TYPE_MSIX) {
362         max_irqs = pdev->msix_entries_nr;
363     }
364     if (!max_irqs) {
365         error_report("Requested interrupt type %d is not enabled for device %x",
366                      ret_intr_type, config_addr);
367         rtas_st(rets, 0, -1); /* Hardware error */
368         return;
369     }
370     /* Correct the number if the guest asked for too many */
371     if (req_num > max_irqs) {
372         trace_spapr_pci_msi_retry(config_addr, req_num, max_irqs);
373         req_num = max_irqs;
374         irq = 0; /* to avoid misleading trace */
375         goto out;
376     }
377 
378     /* Allocate MSIs */
379     if (smc->legacy_irq_allocation) {
380         irq = spapr_irq_find(spapr, req_num, ret_intr_type == RTAS_TYPE_MSI,
381                              &err);
382     } else {
383         irq = spapr_irq_msi_alloc(spapr, req_num,
384                                   ret_intr_type == RTAS_TYPE_MSI, &err);
385     }
386     if (err) {
387         error_reportf_err(err, "Can't allocate MSIs for device %x: ",
388                           config_addr);
389         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
390         return;
391     }
392 
393     for (i = 0; i < req_num; i++) {
394         spapr_irq_claim(spapr, irq + i, false, &err);
395         if (err) {
396             if (i) {
397                 spapr_irq_free(spapr, irq, i);
398             }
399             if (!smc->legacy_irq_allocation) {
400                 spapr_irq_msi_free(spapr, irq, req_num);
401             }
402             error_reportf_err(err, "Can't allocate MSIs for device %x: ",
403                               config_addr);
404             rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
405             return;
406         }
407     }
408 
409     /* Release previous MSIs */
410     if (msi) {
411         if (!smc->legacy_irq_allocation) {
412             spapr_irq_msi_free(spapr, msi->first_irq, msi->num);
413         }
414         spapr_irq_free(spapr, msi->first_irq, msi->num);
415         g_hash_table_remove(phb->msi, &config_addr);
416     }
417 
418     /* Setup MSI/MSIX vectors in the device (via cfgspace or MSIX BAR) */
419     spapr_msi_setmsg(pdev, SPAPR_PCI_MSI_WINDOW, ret_intr_type == RTAS_TYPE_MSIX,
420                      irq, req_num);
421 
422     /* Add MSI device to cache */
423     msi = g_new(spapr_pci_msi, 1);
424     msi->first_irq = irq;
425     msi->num = req_num;
426     config_addr_key = g_new(int, 1);
427     *config_addr_key = config_addr;
428     g_hash_table_insert(phb->msi, config_addr_key, msi);
429 
430 out:
431     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
432     rtas_st(rets, 1, req_num);
433     rtas_st(rets, 2, ++seq_num);
434     if (nret > 3) {
435         rtas_st(rets, 3, ret_intr_type);
436     }
437 
438     trace_spapr_pci_rtas_ibm_change_msi(config_addr, func, req_num, irq);
439 }
440 
441 static void rtas_ibm_query_interrupt_source_number(PowerPCCPU *cpu,
442                                                    SpaprMachineState *spapr,
443                                                    uint32_t token,
444                                                    uint32_t nargs,
445                                                    target_ulong args,
446                                                    uint32_t nret,
447                                                    target_ulong rets)
448 {
449     uint32_t config_addr = rtas_ld(args, 0);
450     uint64_t buid = rtas_ldq(args, 1);
451     unsigned int intr_src_num = -1, ioa_intr_num = rtas_ld(args, 3);
452     SpaprPhbState *phb = NULL;
453     PCIDevice *pdev = NULL;
454     spapr_pci_msi *msi;
455 
456     /* Find SpaprPhbState */
457     phb = spapr_pci_find_phb(spapr, buid);
458     if (phb) {
459         pdev = spapr_pci_find_dev(spapr, buid, config_addr);
460     }
461     if (!phb || !pdev) {
462         rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
463         return;
464     }
465 
466     /* Find device descriptor and start IRQ */
467     msi = (spapr_pci_msi *) g_hash_table_lookup(phb->msi, &config_addr);
468     if (!msi || !msi->first_irq || !msi->num || (ioa_intr_num >= msi->num)) {
469         trace_spapr_pci_msi("Failed to return vector", config_addr);
470         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
471         return;
472     }
473     intr_src_num = msi->first_irq + ioa_intr_num;
474     trace_spapr_pci_rtas_ibm_query_interrupt_source_number(ioa_intr_num,
475                                                            intr_src_num);
476 
477     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
478     rtas_st(rets, 1, intr_src_num);
479     rtas_st(rets, 2, 1);/* 0 == level; 1 == edge */
480 }
481 
482 static void rtas_ibm_set_eeh_option(PowerPCCPU *cpu,
483                                     SpaprMachineState *spapr,
484                                     uint32_t token, uint32_t nargs,
485                                     target_ulong args, uint32_t nret,
486                                     target_ulong rets)
487 {
488     SpaprPhbState *sphb;
489     uint32_t addr, option;
490     uint64_t buid;
491     int ret;
492 
493     if ((nargs != 4) || (nret != 1)) {
494         goto param_error_exit;
495     }
496 
497     buid = rtas_ldq(args, 1);
498     addr = rtas_ld(args, 0);
499     option = rtas_ld(args, 3);
500 
501     sphb = spapr_pci_find_phb(spapr, buid);
502     if (!sphb) {
503         goto param_error_exit;
504     }
505 
506     if (!spapr_phb_eeh_available(sphb)) {
507         goto param_error_exit;
508     }
509 
510     ret = spapr_phb_vfio_eeh_set_option(sphb, addr, option);
511     rtas_st(rets, 0, ret);
512     return;
513 
514 param_error_exit:
515     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
516 }
517 
518 static void rtas_ibm_get_config_addr_info2(PowerPCCPU *cpu,
519                                            SpaprMachineState *spapr,
520                                            uint32_t token, uint32_t nargs,
521                                            target_ulong args, uint32_t nret,
522                                            target_ulong rets)
523 {
524     SpaprPhbState *sphb;
525     PCIDevice *pdev;
526     uint32_t addr, option;
527     uint64_t buid;
528 
529     if ((nargs != 4) || (nret != 2)) {
530         goto param_error_exit;
531     }
532 
533     buid = rtas_ldq(args, 1);
534     sphb = spapr_pci_find_phb(spapr, buid);
535     if (!sphb) {
536         goto param_error_exit;
537     }
538 
539     if (!spapr_phb_eeh_available(sphb)) {
540         goto param_error_exit;
541     }
542 
543     /*
544      * We always have PE address of form "00BB0001". "BB"
545      * represents the bus number of PE's primary bus.
546      */
547     option = rtas_ld(args, 3);
548     switch (option) {
549     case RTAS_GET_PE_ADDR:
550         addr = rtas_ld(args, 0);
551         pdev = spapr_pci_find_dev(spapr, buid, addr);
552         if (!pdev) {
553             goto param_error_exit;
554         }
555 
556         rtas_st(rets, 1, (pci_bus_num(pci_get_bus(pdev)) << 16) + 1);
557         break;
558     case RTAS_GET_PE_MODE:
559         rtas_st(rets, 1, RTAS_PE_MODE_SHARED);
560         break;
561     default:
562         goto param_error_exit;
563     }
564 
565     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
566     return;
567 
568 param_error_exit:
569     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
570 }
571 
572 static void rtas_ibm_read_slot_reset_state2(PowerPCCPU *cpu,
573                                             SpaprMachineState *spapr,
574                                             uint32_t token, uint32_t nargs,
575                                             target_ulong args, uint32_t nret,
576                                             target_ulong rets)
577 {
578     SpaprPhbState *sphb;
579     uint64_t buid;
580     int state, ret;
581 
582     if ((nargs != 3) || (nret != 4 && nret != 5)) {
583         goto param_error_exit;
584     }
585 
586     buid = rtas_ldq(args, 1);
587     sphb = spapr_pci_find_phb(spapr, buid);
588     if (!sphb) {
589         goto param_error_exit;
590     }
591 
592     if (!spapr_phb_eeh_available(sphb)) {
593         goto param_error_exit;
594     }
595 
596     ret = spapr_phb_vfio_eeh_get_state(sphb, &state);
597     rtas_st(rets, 0, ret);
598     if (ret != RTAS_OUT_SUCCESS) {
599         return;
600     }
601 
602     rtas_st(rets, 1, state);
603     rtas_st(rets, 2, RTAS_EEH_SUPPORT);
604     rtas_st(rets, 3, RTAS_EEH_PE_UNAVAIL_INFO);
605     if (nret >= 5) {
606         rtas_st(rets, 4, RTAS_EEH_PE_RECOVER_INFO);
607     }
608     return;
609 
610 param_error_exit:
611     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
612 }
613 
614 static void rtas_ibm_set_slot_reset(PowerPCCPU *cpu,
615                                     SpaprMachineState *spapr,
616                                     uint32_t token, uint32_t nargs,
617                                     target_ulong args, uint32_t nret,
618                                     target_ulong rets)
619 {
620     SpaprPhbState *sphb;
621     uint32_t option;
622     uint64_t buid;
623     int ret;
624 
625     if ((nargs != 4) || (nret != 1)) {
626         goto param_error_exit;
627     }
628 
629     buid = rtas_ldq(args, 1);
630     option = rtas_ld(args, 3);
631     sphb = spapr_pci_find_phb(spapr, buid);
632     if (!sphb) {
633         goto param_error_exit;
634     }
635 
636     if (!spapr_phb_eeh_available(sphb)) {
637         goto param_error_exit;
638     }
639 
640     ret = spapr_phb_vfio_eeh_reset(sphb, option);
641     rtas_st(rets, 0, ret);
642     return;
643 
644 param_error_exit:
645     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
646 }
647 
648 static void rtas_ibm_configure_pe(PowerPCCPU *cpu,
649                                   SpaprMachineState *spapr,
650                                   uint32_t token, uint32_t nargs,
651                                   target_ulong args, uint32_t nret,
652                                   target_ulong rets)
653 {
654     SpaprPhbState *sphb;
655     uint64_t buid;
656     int ret;
657 
658     if ((nargs != 3) || (nret != 1)) {
659         goto param_error_exit;
660     }
661 
662     buid = rtas_ldq(args, 1);
663     sphb = spapr_pci_find_phb(spapr, buid);
664     if (!sphb) {
665         goto param_error_exit;
666     }
667 
668     if (!spapr_phb_eeh_available(sphb)) {
669         goto param_error_exit;
670     }
671 
672     ret = spapr_phb_vfio_eeh_configure(sphb);
673     rtas_st(rets, 0, ret);
674     return;
675 
676 param_error_exit:
677     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
678 }
679 
680 /* To support it later */
681 static void rtas_ibm_slot_error_detail(PowerPCCPU *cpu,
682                                        SpaprMachineState *spapr,
683                                        uint32_t token, uint32_t nargs,
684                                        target_ulong args, uint32_t nret,
685                                        target_ulong rets)
686 {
687     SpaprPhbState *sphb;
688     int option;
689     uint64_t buid;
690 
691     if ((nargs != 8) || (nret != 1)) {
692         goto param_error_exit;
693     }
694 
695     buid = rtas_ldq(args, 1);
696     sphb = spapr_pci_find_phb(spapr, buid);
697     if (!sphb) {
698         goto param_error_exit;
699     }
700 
701     if (!spapr_phb_eeh_available(sphb)) {
702         goto param_error_exit;
703     }
704 
705     option = rtas_ld(args, 7);
706     switch (option) {
707     case RTAS_SLOT_TEMP_ERR_LOG:
708     case RTAS_SLOT_PERM_ERR_LOG:
709         break;
710     default:
711         goto param_error_exit;
712     }
713 
714     /* We don't have error log yet */
715     rtas_st(rets, 0, RTAS_OUT_NO_ERRORS_FOUND);
716     return;
717 
718 param_error_exit:
719     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
720 }
721 
722 static int pci_spapr_swizzle(int slot, int pin)
723 {
724     return (slot + pin) % PCI_NUM_PINS;
725 }
726 
727 static int pci_spapr_map_irq(PCIDevice *pci_dev, int irq_num)
728 {
729     /*
730      * Here we need to convert pci_dev + irq_num to some unique value
731      * which is less than number of IRQs on the specific bus (4).  We
732      * use standard PCI swizzling, that is (slot number + pin number)
733      * % 4.
734      */
735     return pci_spapr_swizzle(PCI_SLOT(pci_dev->devfn), irq_num);
736 }
737 
738 static void pci_spapr_set_irq(void *opaque, int irq_num, int level)
739 {
740     /*
741      * Here we use the number returned by pci_spapr_map_irq to find a
742      * corresponding qemu_irq.
743      */
744     SpaprPhbState *phb = opaque;
745 
746     trace_spapr_pci_lsi_set(phb->dtbusname, irq_num, phb->lsi_table[irq_num].irq);
747     qemu_set_irq(spapr_phb_lsi_qirq(phb, irq_num), level);
748 }
749 
750 static PCIINTxRoute spapr_route_intx_pin_to_irq(void *opaque, int pin)
751 {
752     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(opaque);
753     PCIINTxRoute route;
754 
755     route.mode = PCI_INTX_ENABLED;
756     route.irq = sphb->lsi_table[pin].irq;
757 
758     return route;
759 }
760 
761 /*
762  * MSI/MSIX memory region implementation.
763  * The handler handles both MSI and MSIX.
764  * The vector number is encoded in least bits in data.
765  */
766 static void spapr_msi_write(void *opaque, hwaddr addr,
767                             uint64_t data, unsigned size)
768 {
769     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
770     uint32_t irq = data;
771 
772     trace_spapr_pci_msi_write(addr, data, irq);
773 
774     qemu_irq_pulse(spapr_qirq(spapr, irq));
775 }
776 
777 static const MemoryRegionOps spapr_msi_ops = {
778     /* There is no .read as the read result is undefined by PCI spec */
779     .read = NULL,
780     .write = spapr_msi_write,
781     .endianness = DEVICE_LITTLE_ENDIAN
782 };
783 
784 /*
785  * PHB PCI device
786  */
787 static AddressSpace *spapr_pci_dma_iommu(PCIBus *bus, void *opaque, int devfn)
788 {
789     SpaprPhbState *phb = opaque;
790 
791     return &phb->iommu_as;
792 }
793 
794 static char *spapr_phb_vfio_get_loc_code(SpaprPhbState *sphb,  PCIDevice *pdev)
795 {
796     char *path = NULL, *buf = NULL, *host = NULL;
797 
798     /* Get the PCI VFIO host id */
799     host = object_property_get_str(OBJECT(pdev), "host", NULL);
800     if (!host) {
801         goto err_out;
802     }
803 
804     /* Construct the path of the file that will give us the DT location */
805     path = g_strdup_printf("/sys/bus/pci/devices/%s/devspec", host);
806     g_free(host);
807     if (!g_file_get_contents(path, &buf, NULL, NULL)) {
808         goto err_out;
809     }
810     g_free(path);
811 
812     /* Construct and read from host device tree the loc-code */
813     path = g_strdup_printf("/proc/device-tree%s/ibm,loc-code", buf);
814     g_free(buf);
815     if (!g_file_get_contents(path, &buf, NULL, NULL)) {
816         goto err_out;
817     }
818     return buf;
819 
820 err_out:
821     g_free(path);
822     return NULL;
823 }
824 
825 static char *spapr_phb_get_loc_code(SpaprPhbState *sphb, PCIDevice *pdev)
826 {
827     char *buf;
828     const char *devtype = "qemu";
829     uint32_t busnr = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(pdev))));
830 
831     if (object_dynamic_cast(OBJECT(pdev), "vfio-pci")) {
832         buf = spapr_phb_vfio_get_loc_code(sphb, pdev);
833         if (buf) {
834             return buf;
835         }
836         devtype = "vfio";
837     }
838     /*
839      * For emulated devices and VFIO-failure case, make up
840      * the loc-code.
841      */
842     buf = g_strdup_printf("%s_%s:%04x:%02x:%02x.%x",
843                           devtype, pdev->name, sphb->index, busnr,
844                           PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
845     return buf;
846 }
847 
848 /* Macros to operate with address in OF binding to PCI */
849 #define b_x(x, p, l)    (((x) & ((1<<(l))-1)) << (p))
850 #define b_n(x)          b_x((x), 31, 1) /* 0 if relocatable */
851 #define b_p(x)          b_x((x), 30, 1) /* 1 if prefetchable */
852 #define b_t(x)          b_x((x), 29, 1) /* 1 if the address is aliased */
853 #define b_ss(x)         b_x((x), 24, 2) /* the space code */
854 #define b_bbbbbbbb(x)   b_x((x), 16, 8) /* bus number */
855 #define b_ddddd(x)      b_x((x), 11, 5) /* device number */
856 #define b_fff(x)        b_x((x), 8, 3)  /* function number */
857 #define b_rrrrrrrr(x)   b_x((x), 0, 8)  /* register number */
858 
859 /* for 'reg'/'assigned-addresses' OF properties */
860 #define RESOURCE_CELLS_SIZE 2
861 #define RESOURCE_CELLS_ADDRESS 3
862 
863 typedef struct ResourceFields {
864     uint32_t phys_hi;
865     uint32_t phys_mid;
866     uint32_t phys_lo;
867     uint32_t size_hi;
868     uint32_t size_lo;
869 } QEMU_PACKED ResourceFields;
870 
871 typedef struct ResourceProps {
872     ResourceFields reg[8];
873     ResourceFields assigned[7];
874     uint32_t reg_len;
875     uint32_t assigned_len;
876 } ResourceProps;
877 
878 /* fill in the 'reg'/'assigned-resources' OF properties for
879  * a PCI device. 'reg' describes resource requirements for a
880  * device's IO/MEM regions, 'assigned-addresses' describes the
881  * actual resource assignments.
882  *
883  * the properties are arrays of ('phys-addr', 'size') pairs describing
884  * the addressable regions of the PCI device, where 'phys-addr' is a
885  * RESOURCE_CELLS_ADDRESS-tuple of 32-bit integers corresponding to
886  * (phys.hi, phys.mid, phys.lo), and 'size' is a
887  * RESOURCE_CELLS_SIZE-tuple corresponding to (size.hi, size.lo).
888  *
889  * phys.hi = 0xYYXXXXZZ, where:
890  *   0xYY = npt000ss
891  *          |||   |
892  *          |||   +-- space code
893  *          |||               |
894  *          |||               +  00 if configuration space
895  *          |||               +  01 if IO region,
896  *          |||               +  10 if 32-bit MEM region
897  *          |||               +  11 if 64-bit MEM region
898  *          |||
899  *          ||+------ for non-relocatable IO: 1 if aliased
900  *          ||        for relocatable IO: 1 if below 64KB
901  *          ||        for MEM: 1 if below 1MB
902  *          |+------- 1 if region is prefetchable
903  *          +-------- 1 if region is non-relocatable
904  *   0xXXXX = bbbbbbbb dddddfff, encoding bus, slot, and function
905  *            bits respectively
906  *   0xZZ = rrrrrrrr, the register number of the BAR corresponding
907  *          to the region
908  *
909  * phys.mid and phys.lo correspond respectively to the hi/lo portions
910  * of the actual address of the region.
911  *
912  * how the phys-addr/size values are used differ slightly between
913  * 'reg' and 'assigned-addresses' properties. namely, 'reg' has
914  * an additional description for the config space region of the
915  * device, and in the case of QEMU has n=0 and phys.mid=phys.lo=0
916  * to describe the region as relocatable, with an address-mapping
917  * that corresponds directly to the PHB's address space for the
918  * resource. 'assigned-addresses' always has n=1 set with an absolute
919  * address assigned for the resource. in general, 'assigned-addresses'
920  * won't be populated, since addresses for PCI devices are generally
921  * unmapped initially and left to the guest to assign.
922  *
923  * note also that addresses defined in these properties are, at least
924  * for PAPR guests, relative to the PHBs IO/MEM windows, and
925  * correspond directly to the addresses in the BARs.
926  *
927  * in accordance with PCI Bus Binding to Open Firmware,
928  * IEEE Std 1275-1994, section 4.1.1, as implemented by PAPR+ v2.7,
929  * Appendix C.
930  */
931 static void populate_resource_props(PCIDevice *d, ResourceProps *rp)
932 {
933     int bus_num = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(d))));
934     uint32_t dev_id = (b_bbbbbbbb(bus_num) |
935                        b_ddddd(PCI_SLOT(d->devfn)) |
936                        b_fff(PCI_FUNC(d->devfn)));
937     ResourceFields *reg, *assigned;
938     int i, reg_idx = 0, assigned_idx = 0;
939 
940     /* config space region */
941     reg = &rp->reg[reg_idx++];
942     reg->phys_hi = cpu_to_be32(dev_id);
943     reg->phys_mid = 0;
944     reg->phys_lo = 0;
945     reg->size_hi = 0;
946     reg->size_lo = 0;
947 
948     for (i = 0; i < PCI_NUM_REGIONS; i++) {
949         if (!d->io_regions[i].size) {
950             continue;
951         }
952 
953         reg = &rp->reg[reg_idx++];
954 
955         reg->phys_hi = cpu_to_be32(dev_id | b_rrrrrrrr(pci_bar(d, i)));
956         if (d->io_regions[i].type & PCI_BASE_ADDRESS_SPACE_IO) {
957             reg->phys_hi |= cpu_to_be32(b_ss(1));
958         } else if (d->io_regions[i].type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
959             reg->phys_hi |= cpu_to_be32(b_ss(3));
960         } else {
961             reg->phys_hi |= cpu_to_be32(b_ss(2));
962         }
963         reg->phys_mid = 0;
964         reg->phys_lo = 0;
965         reg->size_hi = cpu_to_be32(d->io_regions[i].size >> 32);
966         reg->size_lo = cpu_to_be32(d->io_regions[i].size);
967 
968         if (d->io_regions[i].addr == PCI_BAR_UNMAPPED) {
969             continue;
970         }
971 
972         assigned = &rp->assigned[assigned_idx++];
973         assigned->phys_hi = cpu_to_be32(be32_to_cpu(reg->phys_hi) | b_n(1));
974         assigned->phys_mid = cpu_to_be32(d->io_regions[i].addr >> 32);
975         assigned->phys_lo = cpu_to_be32(d->io_regions[i].addr);
976         assigned->size_hi = reg->size_hi;
977         assigned->size_lo = reg->size_lo;
978     }
979 
980     rp->reg_len = reg_idx * sizeof(ResourceFields);
981     rp->assigned_len = assigned_idx * sizeof(ResourceFields);
982 }
983 
984 typedef struct PCIClass PCIClass;
985 typedef struct PCISubClass PCISubClass;
986 typedef struct PCIIFace PCIIFace;
987 
988 struct PCIIFace {
989     int iface;
990     const char *name;
991 };
992 
993 struct PCISubClass {
994     int subclass;
995     const char *name;
996     const PCIIFace *iface;
997 };
998 
999 struct PCIClass {
1000     const char *name;
1001     const PCISubClass *subc;
1002 };
1003 
1004 static const PCISubClass undef_subclass[] = {
1005     { PCI_CLASS_NOT_DEFINED_VGA, "display", NULL },
1006     { 0xFF, NULL, NULL },
1007 };
1008 
1009 static const PCISubClass mass_subclass[] = {
1010     { PCI_CLASS_STORAGE_SCSI, "scsi", NULL },
1011     { PCI_CLASS_STORAGE_IDE, "ide", NULL },
1012     { PCI_CLASS_STORAGE_FLOPPY, "fdc", NULL },
1013     { PCI_CLASS_STORAGE_IPI, "ipi", NULL },
1014     { PCI_CLASS_STORAGE_RAID, "raid", NULL },
1015     { PCI_CLASS_STORAGE_ATA, "ata", NULL },
1016     { PCI_CLASS_STORAGE_SATA, "sata", NULL },
1017     { PCI_CLASS_STORAGE_SAS, "sas", NULL },
1018     { 0xFF, NULL, NULL },
1019 };
1020 
1021 static const PCISubClass net_subclass[] = {
1022     { PCI_CLASS_NETWORK_ETHERNET, "ethernet", NULL },
1023     { PCI_CLASS_NETWORK_TOKEN_RING, "token-ring", NULL },
1024     { PCI_CLASS_NETWORK_FDDI, "fddi", NULL },
1025     { PCI_CLASS_NETWORK_ATM, "atm", NULL },
1026     { PCI_CLASS_NETWORK_ISDN, "isdn", NULL },
1027     { PCI_CLASS_NETWORK_WORLDFIP, "worldfip", NULL },
1028     { PCI_CLASS_NETWORK_PICMG214, "picmg", NULL },
1029     { 0xFF, NULL, NULL },
1030 };
1031 
1032 static const PCISubClass displ_subclass[] = {
1033     { PCI_CLASS_DISPLAY_VGA, "vga", NULL },
1034     { PCI_CLASS_DISPLAY_XGA, "xga", NULL },
1035     { PCI_CLASS_DISPLAY_3D, "3d-controller", NULL },
1036     { 0xFF, NULL, NULL },
1037 };
1038 
1039 static const PCISubClass media_subclass[] = {
1040     { PCI_CLASS_MULTIMEDIA_VIDEO, "video", NULL },
1041     { PCI_CLASS_MULTIMEDIA_AUDIO, "sound", NULL },
1042     { PCI_CLASS_MULTIMEDIA_PHONE, "telephony", NULL },
1043     { 0xFF, NULL, NULL },
1044 };
1045 
1046 static const PCISubClass mem_subclass[] = {
1047     { PCI_CLASS_MEMORY_RAM, "memory", NULL },
1048     { PCI_CLASS_MEMORY_FLASH, "flash", NULL },
1049     { 0xFF, NULL, NULL },
1050 };
1051 
1052 static const PCISubClass bridg_subclass[] = {
1053     { PCI_CLASS_BRIDGE_HOST, "host", NULL },
1054     { PCI_CLASS_BRIDGE_ISA, "isa", NULL },
1055     { PCI_CLASS_BRIDGE_EISA, "eisa", NULL },
1056     { PCI_CLASS_BRIDGE_MC, "mca", NULL },
1057     { PCI_CLASS_BRIDGE_PCI, "pci", NULL },
1058     { PCI_CLASS_BRIDGE_PCMCIA, "pcmcia", NULL },
1059     { PCI_CLASS_BRIDGE_NUBUS, "nubus", NULL },
1060     { PCI_CLASS_BRIDGE_CARDBUS, "cardbus", NULL },
1061     { PCI_CLASS_BRIDGE_RACEWAY, "raceway", NULL },
1062     { PCI_CLASS_BRIDGE_PCI_SEMITP, "semi-transparent-pci", NULL },
1063     { PCI_CLASS_BRIDGE_IB_PCI, "infiniband", NULL },
1064     { 0xFF, NULL, NULL },
1065 };
1066 
1067 static const PCISubClass comm_subclass[] = {
1068     { PCI_CLASS_COMMUNICATION_SERIAL, "serial", NULL },
1069     { PCI_CLASS_COMMUNICATION_PARALLEL, "parallel", NULL },
1070     { PCI_CLASS_COMMUNICATION_MULTISERIAL, "multiport-serial", NULL },
1071     { PCI_CLASS_COMMUNICATION_MODEM, "modem", NULL },
1072     { PCI_CLASS_COMMUNICATION_GPIB, "gpib", NULL },
1073     { PCI_CLASS_COMMUNICATION_SC, "smart-card", NULL },
1074     { 0xFF, NULL, NULL, },
1075 };
1076 
1077 static const PCIIFace pic_iface[] = {
1078     { PCI_CLASS_SYSTEM_PIC_IOAPIC, "io-apic" },
1079     { PCI_CLASS_SYSTEM_PIC_IOXAPIC, "io-xapic" },
1080     { 0xFF, NULL },
1081 };
1082 
1083 static const PCISubClass sys_subclass[] = {
1084     { PCI_CLASS_SYSTEM_PIC, "interrupt-controller", pic_iface },
1085     { PCI_CLASS_SYSTEM_DMA, "dma-controller", NULL },
1086     { PCI_CLASS_SYSTEM_TIMER, "timer", NULL },
1087     { PCI_CLASS_SYSTEM_RTC, "rtc", NULL },
1088     { PCI_CLASS_SYSTEM_PCI_HOTPLUG, "hot-plug-controller", NULL },
1089     { PCI_CLASS_SYSTEM_SDHCI, "sd-host-controller", NULL },
1090     { 0xFF, NULL, NULL },
1091 };
1092 
1093 static const PCISubClass inp_subclass[] = {
1094     { PCI_CLASS_INPUT_KEYBOARD, "keyboard", NULL },
1095     { PCI_CLASS_INPUT_PEN, "pen", NULL },
1096     { PCI_CLASS_INPUT_MOUSE, "mouse", NULL },
1097     { PCI_CLASS_INPUT_SCANNER, "scanner", NULL },
1098     { PCI_CLASS_INPUT_GAMEPORT, "gameport", NULL },
1099     { 0xFF, NULL, NULL },
1100 };
1101 
1102 static const PCISubClass dock_subclass[] = {
1103     { PCI_CLASS_DOCKING_GENERIC, "dock", NULL },
1104     { 0xFF, NULL, NULL },
1105 };
1106 
1107 static const PCISubClass cpu_subclass[] = {
1108     { PCI_CLASS_PROCESSOR_PENTIUM, "pentium", NULL },
1109     { PCI_CLASS_PROCESSOR_POWERPC, "powerpc", NULL },
1110     { PCI_CLASS_PROCESSOR_MIPS, "mips", NULL },
1111     { PCI_CLASS_PROCESSOR_CO, "co-processor", NULL },
1112     { 0xFF, NULL, NULL },
1113 };
1114 
1115 static const PCIIFace usb_iface[] = {
1116     { PCI_CLASS_SERIAL_USB_UHCI, "usb-uhci" },
1117     { PCI_CLASS_SERIAL_USB_OHCI, "usb-ohci", },
1118     { PCI_CLASS_SERIAL_USB_EHCI, "usb-ehci" },
1119     { PCI_CLASS_SERIAL_USB_XHCI, "usb-xhci" },
1120     { PCI_CLASS_SERIAL_USB_UNKNOWN, "usb-unknown" },
1121     { PCI_CLASS_SERIAL_USB_DEVICE, "usb-device" },
1122     { 0xFF, NULL },
1123 };
1124 
1125 static const PCISubClass ser_subclass[] = {
1126     { PCI_CLASS_SERIAL_FIREWIRE, "firewire", NULL },
1127     { PCI_CLASS_SERIAL_ACCESS, "access-bus", NULL },
1128     { PCI_CLASS_SERIAL_SSA, "ssa", NULL },
1129     { PCI_CLASS_SERIAL_USB, "usb", usb_iface },
1130     { PCI_CLASS_SERIAL_FIBER, "fibre-channel", NULL },
1131     { PCI_CLASS_SERIAL_SMBUS, "smb", NULL },
1132     { PCI_CLASS_SERIAL_IB, "infiniband", NULL },
1133     { PCI_CLASS_SERIAL_IPMI, "ipmi", NULL },
1134     { PCI_CLASS_SERIAL_SERCOS, "sercos", NULL },
1135     { PCI_CLASS_SERIAL_CANBUS, "canbus", NULL },
1136     { 0xFF, NULL, NULL },
1137 };
1138 
1139 static const PCISubClass wrl_subclass[] = {
1140     { PCI_CLASS_WIRELESS_IRDA, "irda", NULL },
1141     { PCI_CLASS_WIRELESS_CIR, "consumer-ir", NULL },
1142     { PCI_CLASS_WIRELESS_RF_CONTROLLER, "rf-controller", NULL },
1143     { PCI_CLASS_WIRELESS_BLUETOOTH, "bluetooth", NULL },
1144     { PCI_CLASS_WIRELESS_BROADBAND, "broadband", NULL },
1145     { 0xFF, NULL, NULL },
1146 };
1147 
1148 static const PCISubClass sat_subclass[] = {
1149     { PCI_CLASS_SATELLITE_TV, "satellite-tv", NULL },
1150     { PCI_CLASS_SATELLITE_AUDIO, "satellite-audio", NULL },
1151     { PCI_CLASS_SATELLITE_VOICE, "satellite-voice", NULL },
1152     { PCI_CLASS_SATELLITE_DATA, "satellite-data", NULL },
1153     { 0xFF, NULL, NULL },
1154 };
1155 
1156 static const PCISubClass crypt_subclass[] = {
1157     { PCI_CLASS_CRYPT_NETWORK, "network-encryption", NULL },
1158     { PCI_CLASS_CRYPT_ENTERTAINMENT,
1159       "entertainment-encryption", NULL },
1160     { 0xFF, NULL, NULL },
1161 };
1162 
1163 static const PCISubClass spc_subclass[] = {
1164     { PCI_CLASS_SP_DPIO, "dpio", NULL },
1165     { PCI_CLASS_SP_PERF, "counter", NULL },
1166     { PCI_CLASS_SP_SYNCH, "measurement", NULL },
1167     { PCI_CLASS_SP_MANAGEMENT, "management-card", NULL },
1168     { 0xFF, NULL, NULL },
1169 };
1170 
1171 static const PCIClass pci_classes[] = {
1172     { "legacy-device", undef_subclass },
1173     { "mass-storage",  mass_subclass },
1174     { "network", net_subclass },
1175     { "display", displ_subclass, },
1176     { "multimedia-device", media_subclass },
1177     { "memory-controller", mem_subclass },
1178     { "unknown-bridge", bridg_subclass },
1179     { "communication-controller", comm_subclass},
1180     { "system-peripheral", sys_subclass },
1181     { "input-controller", inp_subclass },
1182     { "docking-station", dock_subclass },
1183     { "cpu", cpu_subclass },
1184     { "serial-bus", ser_subclass },
1185     { "wireless-controller", wrl_subclass },
1186     { "intelligent-io", NULL },
1187     { "satellite-device", sat_subclass },
1188     { "encryption", crypt_subclass },
1189     { "data-processing-controller", spc_subclass },
1190 };
1191 
1192 static const char *pci_find_device_name(uint8_t class, uint8_t subclass,
1193                                         uint8_t iface)
1194 {
1195     const PCIClass *pclass;
1196     const PCISubClass *psubclass;
1197     const PCIIFace *piface;
1198     const char *name;
1199 
1200     if (class >= ARRAY_SIZE(pci_classes)) {
1201         return "pci";
1202     }
1203 
1204     pclass = pci_classes + class;
1205     name = pclass->name;
1206 
1207     if (pclass->subc == NULL) {
1208         return name;
1209     }
1210 
1211     psubclass = pclass->subc;
1212     while ((psubclass->subclass & 0xff) != 0xff) {
1213         if ((psubclass->subclass & 0xff) == subclass) {
1214             name = psubclass->name;
1215             break;
1216         }
1217         psubclass++;
1218     }
1219 
1220     piface = psubclass->iface;
1221     if (piface == NULL) {
1222         return name;
1223     }
1224     while ((piface->iface & 0xff) != 0xff) {
1225         if ((piface->iface & 0xff) == iface) {
1226             name = piface->name;
1227             break;
1228         }
1229         piface++;
1230     }
1231 
1232     return name;
1233 }
1234 
1235 static gchar *pci_get_node_name(PCIDevice *dev)
1236 {
1237     int slot = PCI_SLOT(dev->devfn);
1238     int func = PCI_FUNC(dev->devfn);
1239     uint32_t ccode = pci_default_read_config(dev, PCI_CLASS_PROG, 3);
1240     const char *name;
1241 
1242     name = pci_find_device_name((ccode >> 16) & 0xff, (ccode >> 8) & 0xff,
1243                                 ccode & 0xff);
1244 
1245     if (func != 0) {
1246         return g_strdup_printf("%s@%x,%x", name, slot, func);
1247     } else {
1248         return g_strdup_printf("%s@%x", name, slot);
1249     }
1250 }
1251 
1252 static uint32_t spapr_phb_get_pci_drc_index(SpaprPhbState *phb,
1253                                             PCIDevice *pdev);
1254 
1255 static void spapr_populate_pci_child_dt(PCIDevice *dev, void *fdt, int offset,
1256                                        SpaprPhbState *sphb)
1257 {
1258     ResourceProps rp;
1259     bool is_bridge = false;
1260     int pci_status;
1261     char *buf = NULL;
1262     uint32_t drc_index = spapr_phb_get_pci_drc_index(sphb, dev);
1263     uint32_t ccode = pci_default_read_config(dev, PCI_CLASS_PROG, 3);
1264     uint32_t max_msi, max_msix;
1265 
1266     if (pci_default_read_config(dev, PCI_HEADER_TYPE, 1) ==
1267         PCI_HEADER_TYPE_BRIDGE) {
1268         is_bridge = true;
1269     }
1270 
1271     /* in accordance with PAPR+ v2.7 13.6.3, Table 181 */
1272     _FDT(fdt_setprop_cell(fdt, offset, "vendor-id",
1273                           pci_default_read_config(dev, PCI_VENDOR_ID, 2)));
1274     _FDT(fdt_setprop_cell(fdt, offset, "device-id",
1275                           pci_default_read_config(dev, PCI_DEVICE_ID, 2)));
1276     _FDT(fdt_setprop_cell(fdt, offset, "revision-id",
1277                           pci_default_read_config(dev, PCI_REVISION_ID, 1)));
1278     _FDT(fdt_setprop_cell(fdt, offset, "class-code", ccode));
1279     if (pci_default_read_config(dev, PCI_INTERRUPT_PIN, 1)) {
1280         _FDT(fdt_setprop_cell(fdt, offset, "interrupts",
1281                  pci_default_read_config(dev, PCI_INTERRUPT_PIN, 1)));
1282     }
1283 
1284     if (!is_bridge) {
1285         _FDT(fdt_setprop_cell(fdt, offset, "min-grant",
1286             pci_default_read_config(dev, PCI_MIN_GNT, 1)));
1287         _FDT(fdt_setprop_cell(fdt, offset, "max-latency",
1288             pci_default_read_config(dev, PCI_MAX_LAT, 1)));
1289     }
1290 
1291     if (pci_default_read_config(dev, PCI_SUBSYSTEM_ID, 2)) {
1292         _FDT(fdt_setprop_cell(fdt, offset, "subsystem-id",
1293                  pci_default_read_config(dev, PCI_SUBSYSTEM_ID, 2)));
1294     }
1295 
1296     if (pci_default_read_config(dev, PCI_SUBSYSTEM_VENDOR_ID, 2)) {
1297         _FDT(fdt_setprop_cell(fdt, offset, "subsystem-vendor-id",
1298                  pci_default_read_config(dev, PCI_SUBSYSTEM_VENDOR_ID, 2)));
1299     }
1300 
1301     _FDT(fdt_setprop_cell(fdt, offset, "cache-line-size",
1302         pci_default_read_config(dev, PCI_CACHE_LINE_SIZE, 1)));
1303 
1304     /* the following fdt cells are masked off the pci status register */
1305     pci_status = pci_default_read_config(dev, PCI_STATUS, 2);
1306     _FDT(fdt_setprop_cell(fdt, offset, "devsel-speed",
1307                           PCI_STATUS_DEVSEL_MASK & pci_status));
1308 
1309     if (pci_status & PCI_STATUS_FAST_BACK) {
1310         _FDT(fdt_setprop(fdt, offset, "fast-back-to-back", NULL, 0));
1311     }
1312     if (pci_status & PCI_STATUS_66MHZ) {
1313         _FDT(fdt_setprop(fdt, offset, "66mhz-capable", NULL, 0));
1314     }
1315     if (pci_status & PCI_STATUS_UDF) {
1316         _FDT(fdt_setprop(fdt, offset, "udf-supported", NULL, 0));
1317     }
1318 
1319     _FDT(fdt_setprop_string(fdt, offset, "name",
1320                             pci_find_device_name((ccode >> 16) & 0xff,
1321                                                  (ccode >> 8) & 0xff,
1322                                                  ccode & 0xff)));
1323 
1324     buf = spapr_phb_get_loc_code(sphb, dev);
1325     _FDT(fdt_setprop_string(fdt, offset, "ibm,loc-code", buf));
1326     g_free(buf);
1327 
1328     if (drc_index) {
1329         _FDT(fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index));
1330     }
1331 
1332     _FDT(fdt_setprop_cell(fdt, offset, "#address-cells",
1333                           RESOURCE_CELLS_ADDRESS));
1334     _FDT(fdt_setprop_cell(fdt, offset, "#size-cells",
1335                           RESOURCE_CELLS_SIZE));
1336 
1337     if (msi_present(dev)) {
1338         max_msi = msi_nr_vectors_allocated(dev);
1339         if (max_msi) {
1340             _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi", max_msi));
1341         }
1342     }
1343     if (msix_present(dev)) {
1344         max_msix = dev->msix_entries_nr;
1345         if (max_msix) {
1346             _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi-x", max_msix));
1347         }
1348     }
1349 
1350     populate_resource_props(dev, &rp);
1351     _FDT(fdt_setprop(fdt, offset, "reg", (uint8_t *)rp.reg, rp.reg_len));
1352     _FDT(fdt_setprop(fdt, offset, "assigned-addresses",
1353                      (uint8_t *)rp.assigned, rp.assigned_len));
1354 
1355     if (sphb->pcie_ecs && pci_is_express(dev)) {
1356         _FDT(fdt_setprop_cell(fdt, offset, "ibm,pci-config-space-type", 0x1));
1357     }
1358 }
1359 
1360 /* create OF node for pci device and required OF DT properties */
1361 static int spapr_create_pci_child_dt(SpaprPhbState *phb, PCIDevice *dev,
1362                                      void *fdt, int node_offset)
1363 {
1364     int offset;
1365     gchar *nodename;
1366 
1367     nodename = pci_get_node_name(dev);
1368     _FDT(offset = fdt_add_subnode(fdt, node_offset, nodename));
1369     g_free(nodename);
1370 
1371     spapr_populate_pci_child_dt(dev, fdt, offset, phb);
1372 
1373     return offset;
1374 }
1375 
1376 /* Callback to be called during DRC release. */
1377 void spapr_phb_remove_pci_device_cb(DeviceState *dev)
1378 {
1379     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
1380 
1381     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
1382     object_unparent(OBJECT(dev));
1383 }
1384 
1385 static SpaprDrc *spapr_phb_get_pci_func_drc(SpaprPhbState *phb,
1386                                                     uint32_t busnr,
1387                                                     int32_t devfn)
1388 {
1389     return spapr_drc_by_id(TYPE_SPAPR_DRC_PCI,
1390                            (phb->index << 16) | (busnr << 8) | devfn);
1391 }
1392 
1393 static SpaprDrc *spapr_phb_get_pci_drc(SpaprPhbState *phb,
1394                                                PCIDevice *pdev)
1395 {
1396     uint32_t busnr = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(pdev))));
1397     return spapr_phb_get_pci_func_drc(phb, busnr, pdev->devfn);
1398 }
1399 
1400 static uint32_t spapr_phb_get_pci_drc_index(SpaprPhbState *phb,
1401                                             PCIDevice *pdev)
1402 {
1403     SpaprDrc *drc = spapr_phb_get_pci_drc(phb, pdev);
1404 
1405     if (!drc) {
1406         return 0;
1407     }
1408 
1409     return spapr_drc_index(drc);
1410 }
1411 
1412 int spapr_pci_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
1413                           void *fdt, int *fdt_start_offset, Error **errp)
1414 {
1415     HotplugHandler *plug_handler = qdev_get_hotplug_handler(drc->dev);
1416     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(plug_handler);
1417     PCIDevice *pdev = PCI_DEVICE(drc->dev);
1418 
1419     *fdt_start_offset = spapr_create_pci_child_dt(sphb, pdev, fdt, 0);
1420     return 0;
1421 }
1422 
1423 static void spapr_pci_plug(HotplugHandler *plug_handler,
1424                            DeviceState *plugged_dev, Error **errp)
1425 {
1426     SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1427     PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1428     SpaprDrc *drc = spapr_phb_get_pci_drc(phb, pdev);
1429     Error *local_err = NULL;
1430     PCIBus *bus = PCI_BUS(qdev_get_parent_bus(DEVICE(pdev)));
1431     uint32_t slotnr = PCI_SLOT(pdev->devfn);
1432 
1433     /* if DR is disabled we don't need to do anything in the case of
1434      * hotplug or coldplug callbacks
1435      */
1436     if (!phb->dr_enabled) {
1437         /* if this is a hotplug operation initiated by the user
1438          * we need to let them know it's not enabled
1439          */
1440         if (plugged_dev->hotplugged) {
1441             error_setg(&local_err, QERR_BUS_NO_HOTPLUG,
1442                        object_get_typename(OBJECT(phb)));
1443         }
1444         goto out;
1445     }
1446 
1447     g_assert(drc);
1448 
1449     /* Following the QEMU convention used for PCIe multifunction
1450      * hotplug, we do not allow functions to be hotplugged to a
1451      * slot that already has function 0 present
1452      */
1453     if (plugged_dev->hotplugged && bus->devices[PCI_DEVFN(slotnr, 0)] &&
1454         PCI_FUNC(pdev->devfn) != 0) {
1455         error_setg(&local_err, "PCI: slot %d function 0 already ocuppied by %s,"
1456                    " additional functions can no longer be exposed to guest.",
1457                    slotnr, bus->devices[PCI_DEVFN(slotnr, 0)]->name);
1458         goto out;
1459     }
1460 
1461     spapr_drc_attach(drc, DEVICE(pdev), &local_err);
1462     if (local_err) {
1463         goto out;
1464     }
1465 
1466     /* If this is function 0, signal hotplug for all the device functions.
1467      * Otherwise defer sending the hotplug event.
1468      */
1469     if (!spapr_drc_hotplugged(plugged_dev)) {
1470         spapr_drc_reset(drc);
1471     } else if (PCI_FUNC(pdev->devfn) == 0) {
1472         int i;
1473 
1474         for (i = 0; i < 8; i++) {
1475             SpaprDrc *func_drc;
1476             SpaprDrcClass *func_drck;
1477             SpaprDREntitySense state;
1478 
1479             func_drc = spapr_phb_get_pci_func_drc(phb, pci_bus_num(bus),
1480                                                   PCI_DEVFN(slotnr, i));
1481             func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1482             state = func_drck->dr_entity_sense(func_drc);
1483 
1484             if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) {
1485                 spapr_hotplug_req_add_by_index(func_drc);
1486             }
1487         }
1488     }
1489 
1490 out:
1491     error_propagate(errp, local_err);
1492 }
1493 
1494 static void spapr_pci_unplug(HotplugHandler *plug_handler,
1495                              DeviceState *plugged_dev, Error **errp)
1496 {
1497     /* some version guests do not wait for completion of a device
1498      * cleanup (generally done asynchronously by the kernel) before
1499      * signaling to QEMU that the device is safe, but instead sleep
1500      * for some 'safe' period of time. unfortunately on a busy host
1501      * this sleep isn't guaranteed to be long enough, resulting in
1502      * bad things like IRQ lines being left asserted during final
1503      * device removal. to deal with this we call reset just prior
1504      * to finalizing the device, which will put the device back into
1505      * an 'idle' state, as the device cleanup code expects.
1506      */
1507     pci_device_reset(PCI_DEVICE(plugged_dev));
1508     object_property_set_bool(OBJECT(plugged_dev), false, "realized", NULL);
1509 }
1510 
1511 static void spapr_pci_unplug_request(HotplugHandler *plug_handler,
1512                                      DeviceState *plugged_dev, Error **errp)
1513 {
1514     SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1515     PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1516     SpaprDrc *drc = spapr_phb_get_pci_drc(phb, pdev);
1517 
1518     if (!phb->dr_enabled) {
1519         error_setg(errp, QERR_BUS_NO_HOTPLUG,
1520                    object_get_typename(OBJECT(phb)));
1521         return;
1522     }
1523 
1524     g_assert(drc);
1525     g_assert(drc->dev == plugged_dev);
1526 
1527     if (!spapr_drc_unplug_requested(drc)) {
1528         PCIBus *bus = PCI_BUS(qdev_get_parent_bus(DEVICE(pdev)));
1529         uint32_t slotnr = PCI_SLOT(pdev->devfn);
1530         SpaprDrc *func_drc;
1531         SpaprDrcClass *func_drck;
1532         SpaprDREntitySense state;
1533         int i;
1534 
1535         /* ensure any other present functions are pending unplug */
1536         if (PCI_FUNC(pdev->devfn) == 0) {
1537             for (i = 1; i < 8; i++) {
1538                 func_drc = spapr_phb_get_pci_func_drc(phb, pci_bus_num(bus),
1539                                                       PCI_DEVFN(slotnr, i));
1540                 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1541                 state = func_drck->dr_entity_sense(func_drc);
1542                 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT
1543                     && !spapr_drc_unplug_requested(func_drc)) {
1544                     error_setg(errp,
1545                                "PCI: slot %d, function %d still present. "
1546                                "Must unplug all non-0 functions first.",
1547                                slotnr, i);
1548                     return;
1549                 }
1550             }
1551         }
1552 
1553         spapr_drc_detach(drc);
1554 
1555         /* if this isn't func 0, defer unplug event. otherwise signal removal
1556          * for all present functions
1557          */
1558         if (PCI_FUNC(pdev->devfn) == 0) {
1559             for (i = 7; i >= 0; i--) {
1560                 func_drc = spapr_phb_get_pci_func_drc(phb, pci_bus_num(bus),
1561                                                       PCI_DEVFN(slotnr, i));
1562                 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1563                 state = func_drck->dr_entity_sense(func_drc);
1564                 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) {
1565                     spapr_hotplug_req_remove_by_index(func_drc);
1566                 }
1567             }
1568         }
1569     }
1570 }
1571 
1572 static void spapr_phb_finalizefn(Object *obj)
1573 {
1574     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(obj);
1575 
1576     g_free(sphb->dtbusname);
1577     sphb->dtbusname = NULL;
1578 }
1579 
1580 static void spapr_phb_unrealize(DeviceState *dev, Error **errp)
1581 {
1582     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
1583     SysBusDevice *s = SYS_BUS_DEVICE(dev);
1584     PCIHostState *phb = PCI_HOST_BRIDGE(s);
1585     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(phb);
1586     SpaprTceTable *tcet;
1587     int i;
1588     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
1589 
1590     if (sphb->msi) {
1591         g_hash_table_unref(sphb->msi);
1592         sphb->msi = NULL;
1593     }
1594 
1595     /*
1596      * Remove IO/MMIO subregions and aliases, rest should get cleaned
1597      * via PHB's unrealize->object_finalize
1598      */
1599     for (i = windows_supported - 1; i >= 0; i--) {
1600         tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[i]);
1601         if (tcet) {
1602             memory_region_del_subregion(&sphb->iommu_root,
1603                                         spapr_tce_get_iommu(tcet));
1604         }
1605     }
1606 
1607     if (sphb->dr_enabled) {
1608         for (i = PCI_SLOT_MAX * 8 - 1; i >= 0; i--) {
1609             SpaprDrc *drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PCI,
1610                                                     (sphb->index << 16) | i);
1611 
1612             if (drc) {
1613                 object_unparent(OBJECT(drc));
1614             }
1615         }
1616     }
1617 
1618     for (i = PCI_NUM_PINS - 1; i >= 0; i--) {
1619         if (sphb->lsi_table[i].irq) {
1620             spapr_irq_free(spapr, sphb->lsi_table[i].irq, 1);
1621             sphb->lsi_table[i].irq = 0;
1622         }
1623     }
1624 
1625     QLIST_REMOVE(sphb, list);
1626 
1627     memory_region_del_subregion(&sphb->iommu_root, &sphb->msiwindow);
1628 
1629     address_space_destroy(&sphb->iommu_as);
1630 
1631     qbus_set_hotplug_handler(BUS(phb->bus), NULL, &error_abort);
1632     pci_unregister_root_bus(phb->bus);
1633 
1634     memory_region_del_subregion(get_system_memory(), &sphb->iowindow);
1635     if (sphb->mem64_win_pciaddr != (hwaddr)-1) {
1636         memory_region_del_subregion(get_system_memory(), &sphb->mem64window);
1637     }
1638     memory_region_del_subregion(get_system_memory(), &sphb->mem32window);
1639 }
1640 
1641 static bool spapr_phb_allows_extended_config_space(PCIBus *bus)
1642 {
1643     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(BUS(bus)->parent);
1644 
1645     return sphb->pcie_ecs;
1646 }
1647 
1648 static void spapr_phb_root_bus_class_init(ObjectClass *klass, void *data)
1649 {
1650     PCIBusClass *pbc = PCI_BUS_CLASS(klass);
1651 
1652     pbc->allows_extended_config_space = spapr_phb_allows_extended_config_space;
1653 }
1654 
1655 #define TYPE_SPAPR_PHB_ROOT_BUS "pci"
1656 
1657 static const TypeInfo spapr_phb_root_bus_info = {
1658     .name = TYPE_SPAPR_PHB_ROOT_BUS,
1659     .parent = TYPE_PCI_BUS,
1660     .class_init = spapr_phb_root_bus_class_init,
1661 };
1662 
1663 static void spapr_phb_realize(DeviceState *dev, Error **errp)
1664 {
1665     /* We don't use SPAPR_MACHINE() in order to exit gracefully if the user
1666      * tries to add a sPAPR PHB to a non-pseries machine.
1667      */
1668     SpaprMachineState *spapr =
1669         (SpaprMachineState *) object_dynamic_cast(qdev_get_machine(),
1670                                                   TYPE_SPAPR_MACHINE);
1671     SpaprMachineClass *smc = spapr ? SPAPR_MACHINE_GET_CLASS(spapr) : NULL;
1672     SysBusDevice *s = SYS_BUS_DEVICE(dev);
1673     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(s);
1674     PCIHostState *phb = PCI_HOST_BRIDGE(s);
1675     char *namebuf;
1676     int i;
1677     PCIBus *bus;
1678     uint64_t msi_window_size = 4096;
1679     SpaprTceTable *tcet;
1680     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
1681 
1682     if (!spapr) {
1683         error_setg(errp, TYPE_SPAPR_PCI_HOST_BRIDGE " needs a pseries machine");
1684         return;
1685     }
1686 
1687     assert(sphb->index != (uint32_t)-1); /* checked in spapr_phb_pre_plug() */
1688 
1689     if (sphb->mem64_win_size != 0) {
1690         if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) {
1691             error_setg(errp, "32-bit memory window of size 0x%"HWADDR_PRIx
1692                        " (max 2 GiB)", sphb->mem_win_size);
1693             return;
1694         }
1695 
1696         /* 64-bit window defaults to identity mapping */
1697         sphb->mem64_win_pciaddr = sphb->mem64_win_addr;
1698     } else if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) {
1699         /*
1700          * For compatibility with old configuration, if no 64-bit MMIO
1701          * window is specified, but the ordinary (32-bit) memory
1702          * window is specified as > 2GiB, we treat it as a 2GiB 32-bit
1703          * window, with a 64-bit MMIO window following on immediately
1704          * afterwards
1705          */
1706         sphb->mem64_win_size = sphb->mem_win_size - SPAPR_PCI_MEM32_WIN_SIZE;
1707         sphb->mem64_win_addr = sphb->mem_win_addr + SPAPR_PCI_MEM32_WIN_SIZE;
1708         sphb->mem64_win_pciaddr =
1709             SPAPR_PCI_MEM_WIN_BUS_OFFSET + SPAPR_PCI_MEM32_WIN_SIZE;
1710         sphb->mem_win_size = SPAPR_PCI_MEM32_WIN_SIZE;
1711     }
1712 
1713     if (spapr_pci_find_phb(spapr, sphb->buid)) {
1714         error_setg(errp, "PCI host bridges must have unique BUIDs");
1715         return;
1716     }
1717 
1718     if (sphb->numa_node != -1 &&
1719         (sphb->numa_node >= MAX_NODES || !numa_info[sphb->numa_node].present)) {
1720         error_setg(errp, "Invalid NUMA node ID for PCI host bridge");
1721         return;
1722     }
1723 
1724     sphb->dtbusname = g_strdup_printf("pci@%" PRIx64, sphb->buid);
1725 
1726     /* Initialize memory regions */
1727     namebuf = g_strdup_printf("%s.mmio", sphb->dtbusname);
1728     memory_region_init(&sphb->memspace, OBJECT(sphb), namebuf, UINT64_MAX);
1729     g_free(namebuf);
1730 
1731     namebuf = g_strdup_printf("%s.mmio32-alias", sphb->dtbusname);
1732     memory_region_init_alias(&sphb->mem32window, OBJECT(sphb),
1733                              namebuf, &sphb->memspace,
1734                              SPAPR_PCI_MEM_WIN_BUS_OFFSET, sphb->mem_win_size);
1735     g_free(namebuf);
1736     memory_region_add_subregion(get_system_memory(), sphb->mem_win_addr,
1737                                 &sphb->mem32window);
1738 
1739     if (sphb->mem64_win_size != 0) {
1740         namebuf = g_strdup_printf("%s.mmio64-alias", sphb->dtbusname);
1741         memory_region_init_alias(&sphb->mem64window, OBJECT(sphb),
1742                                  namebuf, &sphb->memspace,
1743                                  sphb->mem64_win_pciaddr, sphb->mem64_win_size);
1744         g_free(namebuf);
1745 
1746         memory_region_add_subregion(get_system_memory(),
1747                                     sphb->mem64_win_addr,
1748                                     &sphb->mem64window);
1749     }
1750 
1751     /* Initialize IO regions */
1752     namebuf = g_strdup_printf("%s.io", sphb->dtbusname);
1753     memory_region_init(&sphb->iospace, OBJECT(sphb),
1754                        namebuf, SPAPR_PCI_IO_WIN_SIZE);
1755     g_free(namebuf);
1756 
1757     namebuf = g_strdup_printf("%s.io-alias", sphb->dtbusname);
1758     memory_region_init_alias(&sphb->iowindow, OBJECT(sphb), namebuf,
1759                              &sphb->iospace, 0, SPAPR_PCI_IO_WIN_SIZE);
1760     g_free(namebuf);
1761     memory_region_add_subregion(get_system_memory(), sphb->io_win_addr,
1762                                 &sphb->iowindow);
1763 
1764     bus = pci_register_root_bus(dev, NULL,
1765                                 pci_spapr_set_irq, pci_spapr_map_irq, sphb,
1766                                 &sphb->memspace, &sphb->iospace,
1767                                 PCI_DEVFN(0, 0), PCI_NUM_PINS,
1768                                 TYPE_SPAPR_PHB_ROOT_BUS);
1769     phb->bus = bus;
1770     qbus_set_hotplug_handler(BUS(phb->bus), OBJECT(sphb), NULL);
1771 
1772     /*
1773      * Initialize PHB address space.
1774      * By default there will be at least one subregion for default
1775      * 32bit DMA window.
1776      * Later the guest might want to create another DMA window
1777      * which will become another memory subregion.
1778      */
1779     namebuf = g_strdup_printf("%s.iommu-root", sphb->dtbusname);
1780     memory_region_init(&sphb->iommu_root, OBJECT(sphb),
1781                        namebuf, UINT64_MAX);
1782     g_free(namebuf);
1783     address_space_init(&sphb->iommu_as, &sphb->iommu_root,
1784                        sphb->dtbusname);
1785 
1786     /*
1787      * As MSI/MSIX interrupts trigger by writing at MSI/MSIX vectors,
1788      * we need to allocate some memory to catch those writes coming
1789      * from msi_notify()/msix_notify().
1790      * As MSIMessage:addr is going to be the same and MSIMessage:data
1791      * is going to be a VIRQ number, 4 bytes of the MSI MR will only
1792      * be used.
1793      *
1794      * For KVM we want to ensure that this memory is a full page so that
1795      * our memory slot is of page size granularity.
1796      */
1797 #ifdef CONFIG_KVM
1798     if (kvm_enabled()) {
1799         msi_window_size = getpagesize();
1800     }
1801 #endif
1802 
1803     memory_region_init_io(&sphb->msiwindow, OBJECT(sphb), &spapr_msi_ops, spapr,
1804                           "msi", msi_window_size);
1805     memory_region_add_subregion(&sphb->iommu_root, SPAPR_PCI_MSI_WINDOW,
1806                                 &sphb->msiwindow);
1807 
1808     pci_setup_iommu(bus, spapr_pci_dma_iommu, sphb);
1809 
1810     pci_bus_set_route_irq_fn(bus, spapr_route_intx_pin_to_irq);
1811 
1812     QLIST_INSERT_HEAD(&spapr->phbs, sphb, list);
1813 
1814     /* Initialize the LSI table */
1815     for (i = 0; i < PCI_NUM_PINS; i++) {
1816         uint32_t irq = SPAPR_IRQ_PCI_LSI + sphb->index * PCI_NUM_PINS + i;
1817         Error *local_err = NULL;
1818 
1819         if (smc->legacy_irq_allocation) {
1820             irq = spapr_irq_findone(spapr, &local_err);
1821             if (local_err) {
1822                 error_propagate_prepend(errp, local_err,
1823                                         "can't allocate LSIs: ");
1824                 /*
1825                  * Older machines will never support PHB hotplug, ie, this is an
1826                  * init only path and QEMU will terminate. No need to rollback.
1827                  */
1828                 return;
1829             }
1830         }
1831 
1832         spapr_irq_claim(spapr, irq, true, &local_err);
1833         if (local_err) {
1834             error_propagate_prepend(errp, local_err, "can't allocate LSIs: ");
1835             goto unrealize;
1836         }
1837 
1838         sphb->lsi_table[i].irq = irq;
1839     }
1840 
1841     /* allocate connectors for child PCI devices */
1842     if (sphb->dr_enabled) {
1843         for (i = 0; i < PCI_SLOT_MAX * 8; i++) {
1844             spapr_dr_connector_new(OBJECT(phb), TYPE_SPAPR_DRC_PCI,
1845                                    (sphb->index << 16) | i);
1846         }
1847     }
1848 
1849     /* DMA setup */
1850     for (i = 0; i < windows_supported; ++i) {
1851         tcet = spapr_tce_new_table(DEVICE(sphb), sphb->dma_liobn[i]);
1852         if (!tcet) {
1853             error_setg(errp, "Creating window#%d failed for %s",
1854                        i, sphb->dtbusname);
1855             goto unrealize;
1856         }
1857         memory_region_add_subregion(&sphb->iommu_root, 0,
1858                                     spapr_tce_get_iommu(tcet));
1859     }
1860 
1861     sphb->msi = g_hash_table_new_full(g_int_hash, g_int_equal, g_free, g_free);
1862     return;
1863 
1864 unrealize:
1865     spapr_phb_unrealize(dev, NULL);
1866 }
1867 
1868 static int spapr_phb_children_reset(Object *child, void *opaque)
1869 {
1870     DeviceState *dev = (DeviceState *) object_dynamic_cast(child, TYPE_DEVICE);
1871 
1872     if (dev) {
1873         device_reset(dev);
1874     }
1875 
1876     return 0;
1877 }
1878 
1879 void spapr_phb_dma_reset(SpaprPhbState *sphb)
1880 {
1881     int i;
1882     SpaprTceTable *tcet;
1883 
1884     for (i = 0; i < SPAPR_PCI_DMA_MAX_WINDOWS; ++i) {
1885         tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[i]);
1886 
1887         if (tcet && tcet->nb_table) {
1888             spapr_tce_table_disable(tcet);
1889         }
1890     }
1891 
1892     /* Register default 32bit DMA window */
1893     tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[0]);
1894     spapr_tce_table_enable(tcet, SPAPR_TCE_PAGE_SHIFT, sphb->dma_win_addr,
1895                            sphb->dma_win_size >> SPAPR_TCE_PAGE_SHIFT);
1896 }
1897 
1898 static void spapr_phb_reset(DeviceState *qdev)
1899 {
1900     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(qdev);
1901 
1902     spapr_phb_dma_reset(sphb);
1903 
1904     /* Reset the IOMMU state */
1905     object_child_foreach(OBJECT(qdev), spapr_phb_children_reset, NULL);
1906 
1907     if (spapr_phb_eeh_available(SPAPR_PCI_HOST_BRIDGE(qdev))) {
1908         spapr_phb_vfio_reset(qdev);
1909     }
1910 }
1911 
1912 static Property spapr_phb_properties[] = {
1913     DEFINE_PROP_UINT32("index", SpaprPhbState, index, -1),
1914     DEFINE_PROP_UINT64("mem_win_size", SpaprPhbState, mem_win_size,
1915                        SPAPR_PCI_MEM32_WIN_SIZE),
1916     DEFINE_PROP_UINT64("mem64_win_size", SpaprPhbState, mem64_win_size,
1917                        SPAPR_PCI_MEM64_WIN_SIZE),
1918     DEFINE_PROP_UINT64("io_win_size", SpaprPhbState, io_win_size,
1919                        SPAPR_PCI_IO_WIN_SIZE),
1920     DEFINE_PROP_BOOL("dynamic-reconfiguration", SpaprPhbState, dr_enabled,
1921                      true),
1922     /* Default DMA window is 0..1GB */
1923     DEFINE_PROP_UINT64("dma_win_addr", SpaprPhbState, dma_win_addr, 0),
1924     DEFINE_PROP_UINT64("dma_win_size", SpaprPhbState, dma_win_size, 0x40000000),
1925     DEFINE_PROP_UINT64("dma64_win_addr", SpaprPhbState, dma64_win_addr,
1926                        0x800000000000000ULL),
1927     DEFINE_PROP_BOOL("ddw", SpaprPhbState, ddw_enabled, true),
1928     DEFINE_PROP_UINT64("pgsz", SpaprPhbState, page_size_mask,
1929                        (1ULL << 12) | (1ULL << 16)),
1930     DEFINE_PROP_UINT32("numa_node", SpaprPhbState, numa_node, -1),
1931     DEFINE_PROP_BOOL("pre-2.8-migration", SpaprPhbState,
1932                      pre_2_8_migration, false),
1933     DEFINE_PROP_BOOL("pcie-extended-configuration-space", SpaprPhbState,
1934                      pcie_ecs, true),
1935     DEFINE_PROP_END_OF_LIST(),
1936 };
1937 
1938 static const VMStateDescription vmstate_spapr_pci_lsi = {
1939     .name = "spapr_pci/lsi",
1940     .version_id = 1,
1941     .minimum_version_id = 1,
1942     .fields = (VMStateField[]) {
1943         VMSTATE_UINT32_EQUAL(irq, struct spapr_pci_lsi, NULL),
1944 
1945         VMSTATE_END_OF_LIST()
1946     },
1947 };
1948 
1949 static const VMStateDescription vmstate_spapr_pci_msi = {
1950     .name = "spapr_pci/msi",
1951     .version_id = 1,
1952     .minimum_version_id = 1,
1953     .fields = (VMStateField []) {
1954         VMSTATE_UINT32(key, spapr_pci_msi_mig),
1955         VMSTATE_UINT32(value.first_irq, spapr_pci_msi_mig),
1956         VMSTATE_UINT32(value.num, spapr_pci_msi_mig),
1957         VMSTATE_END_OF_LIST()
1958     },
1959 };
1960 
1961 static int spapr_pci_pre_save(void *opaque)
1962 {
1963     SpaprPhbState *sphb = opaque;
1964     GHashTableIter iter;
1965     gpointer key, value;
1966     int i;
1967 
1968     if (sphb->pre_2_8_migration) {
1969         sphb->mig_liobn = sphb->dma_liobn[0];
1970         sphb->mig_mem_win_addr = sphb->mem_win_addr;
1971         sphb->mig_mem_win_size = sphb->mem_win_size;
1972         sphb->mig_io_win_addr = sphb->io_win_addr;
1973         sphb->mig_io_win_size = sphb->io_win_size;
1974 
1975         if ((sphb->mem64_win_size != 0)
1976             && (sphb->mem64_win_addr
1977                 == (sphb->mem_win_addr + sphb->mem_win_size))) {
1978             sphb->mig_mem_win_size += sphb->mem64_win_size;
1979         }
1980     }
1981 
1982     g_free(sphb->msi_devs);
1983     sphb->msi_devs = NULL;
1984     sphb->msi_devs_num = g_hash_table_size(sphb->msi);
1985     if (!sphb->msi_devs_num) {
1986         return 0;
1987     }
1988     sphb->msi_devs = g_new(spapr_pci_msi_mig, sphb->msi_devs_num);
1989 
1990     g_hash_table_iter_init(&iter, sphb->msi);
1991     for (i = 0; g_hash_table_iter_next(&iter, &key, &value); ++i) {
1992         sphb->msi_devs[i].key = *(uint32_t *) key;
1993         sphb->msi_devs[i].value = *(spapr_pci_msi *) value;
1994     }
1995 
1996     return 0;
1997 }
1998 
1999 static int spapr_pci_post_load(void *opaque, int version_id)
2000 {
2001     SpaprPhbState *sphb = opaque;
2002     gpointer key, value;
2003     int i;
2004 
2005     for (i = 0; i < sphb->msi_devs_num; ++i) {
2006         key = g_memdup(&sphb->msi_devs[i].key,
2007                        sizeof(sphb->msi_devs[i].key));
2008         value = g_memdup(&sphb->msi_devs[i].value,
2009                          sizeof(sphb->msi_devs[i].value));
2010         g_hash_table_insert(sphb->msi, key, value);
2011     }
2012     g_free(sphb->msi_devs);
2013     sphb->msi_devs = NULL;
2014     sphb->msi_devs_num = 0;
2015 
2016     return 0;
2017 }
2018 
2019 static bool pre_2_8_migration(void *opaque, int version_id)
2020 {
2021     SpaprPhbState *sphb = opaque;
2022 
2023     return sphb->pre_2_8_migration;
2024 }
2025 
2026 static const VMStateDescription vmstate_spapr_pci = {
2027     .name = "spapr_pci",
2028     .version_id = 2,
2029     .minimum_version_id = 2,
2030     .pre_save = spapr_pci_pre_save,
2031     .post_load = spapr_pci_post_load,
2032     .fields = (VMStateField[]) {
2033         VMSTATE_UINT64_EQUAL(buid, SpaprPhbState, NULL),
2034         VMSTATE_UINT32_TEST(mig_liobn, SpaprPhbState, pre_2_8_migration),
2035         VMSTATE_UINT64_TEST(mig_mem_win_addr, SpaprPhbState, pre_2_8_migration),
2036         VMSTATE_UINT64_TEST(mig_mem_win_size, SpaprPhbState, pre_2_8_migration),
2037         VMSTATE_UINT64_TEST(mig_io_win_addr, SpaprPhbState, pre_2_8_migration),
2038         VMSTATE_UINT64_TEST(mig_io_win_size, SpaprPhbState, pre_2_8_migration),
2039         VMSTATE_STRUCT_ARRAY(lsi_table, SpaprPhbState, PCI_NUM_PINS, 0,
2040                              vmstate_spapr_pci_lsi, struct spapr_pci_lsi),
2041         VMSTATE_INT32(msi_devs_num, SpaprPhbState),
2042         VMSTATE_STRUCT_VARRAY_ALLOC(msi_devs, SpaprPhbState, msi_devs_num, 0,
2043                                     vmstate_spapr_pci_msi, spapr_pci_msi_mig),
2044         VMSTATE_END_OF_LIST()
2045     },
2046 };
2047 
2048 static const char *spapr_phb_root_bus_path(PCIHostState *host_bridge,
2049                                            PCIBus *rootbus)
2050 {
2051     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(host_bridge);
2052 
2053     return sphb->dtbusname;
2054 }
2055 
2056 static void spapr_phb_class_init(ObjectClass *klass, void *data)
2057 {
2058     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(klass);
2059     DeviceClass *dc = DEVICE_CLASS(klass);
2060     HotplugHandlerClass *hp = HOTPLUG_HANDLER_CLASS(klass);
2061 
2062     hc->root_bus_path = spapr_phb_root_bus_path;
2063     dc->realize = spapr_phb_realize;
2064     dc->unrealize = spapr_phb_unrealize;
2065     dc->props = spapr_phb_properties;
2066     dc->reset = spapr_phb_reset;
2067     dc->vmsd = &vmstate_spapr_pci;
2068     /* Supported by TYPE_SPAPR_MACHINE */
2069     dc->user_creatable = true;
2070     set_bit(DEVICE_CATEGORY_BRIDGE, dc->categories);
2071     hp->plug = spapr_pci_plug;
2072     hp->unplug = spapr_pci_unplug;
2073     hp->unplug_request = spapr_pci_unplug_request;
2074 }
2075 
2076 static const TypeInfo spapr_phb_info = {
2077     .name          = TYPE_SPAPR_PCI_HOST_BRIDGE,
2078     .parent        = TYPE_PCI_HOST_BRIDGE,
2079     .instance_size = sizeof(SpaprPhbState),
2080     .instance_finalize = spapr_phb_finalizefn,
2081     .class_init    = spapr_phb_class_init,
2082     .interfaces    = (InterfaceInfo[]) {
2083         { TYPE_HOTPLUG_HANDLER },
2084         { }
2085     }
2086 };
2087 
2088 typedef struct SpaprFdt {
2089     void *fdt;
2090     int node_off;
2091     SpaprPhbState *sphb;
2092 } SpaprFdt;
2093 
2094 static void spapr_populate_pci_devices_dt(PCIBus *bus, PCIDevice *pdev,
2095                                           void *opaque)
2096 {
2097     PCIBus *sec_bus;
2098     SpaprFdt *p = opaque;
2099     int offset;
2100     SpaprFdt s_fdt;
2101 
2102     offset = spapr_create_pci_child_dt(p->sphb, pdev, p->fdt, p->node_off);
2103     if (!offset) {
2104         error_report("Failed to create pci child device tree node");
2105         return;
2106     }
2107 
2108     if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) !=
2109          PCI_HEADER_TYPE_BRIDGE)) {
2110         return;
2111     }
2112 
2113     sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev));
2114     if (!sec_bus) {
2115         return;
2116     }
2117 
2118     s_fdt.fdt = p->fdt;
2119     s_fdt.node_off = offset;
2120     s_fdt.sphb = p->sphb;
2121     pci_for_each_device_reverse(sec_bus, pci_bus_num(sec_bus),
2122                                 spapr_populate_pci_devices_dt,
2123                                 &s_fdt);
2124 }
2125 
2126 static void spapr_phb_pci_enumerate_bridge(PCIBus *bus, PCIDevice *pdev,
2127                                            void *opaque)
2128 {
2129     unsigned int *bus_no = opaque;
2130     PCIBus *sec_bus = NULL;
2131 
2132     if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) !=
2133          PCI_HEADER_TYPE_BRIDGE)) {
2134         return;
2135     }
2136 
2137     (*bus_no)++;
2138     pci_default_write_config(pdev, PCI_PRIMARY_BUS, pci_dev_bus_num(pdev), 1);
2139     pci_default_write_config(pdev, PCI_SECONDARY_BUS, *bus_no, 1);
2140     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
2141 
2142     sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev));
2143     if (!sec_bus) {
2144         return;
2145     }
2146 
2147     pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
2148                         spapr_phb_pci_enumerate_bridge, bus_no);
2149     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
2150 }
2151 
2152 static void spapr_phb_pci_enumerate(SpaprPhbState *phb)
2153 {
2154     PCIBus *bus = PCI_HOST_BRIDGE(phb)->bus;
2155     unsigned int bus_no = 0;
2156 
2157     pci_for_each_device(bus, pci_bus_num(bus),
2158                         spapr_phb_pci_enumerate_bridge,
2159                         &bus_no);
2160 
2161 }
2162 
2163 int spapr_populate_pci_dt(SpaprPhbState *phb, uint32_t intc_phandle, void *fdt,
2164                           uint32_t nr_msis, int *node_offset)
2165 {
2166     int bus_off, i, j, ret;
2167     gchar *nodename;
2168     uint32_t bus_range[] = { cpu_to_be32(0), cpu_to_be32(0xff) };
2169     struct {
2170         uint32_t hi;
2171         uint64_t child;
2172         uint64_t parent;
2173         uint64_t size;
2174     } QEMU_PACKED ranges[] = {
2175         {
2176             cpu_to_be32(b_ss(1)), cpu_to_be64(0),
2177             cpu_to_be64(phb->io_win_addr),
2178             cpu_to_be64(memory_region_size(&phb->iospace)),
2179         },
2180         {
2181             cpu_to_be32(b_ss(2)), cpu_to_be64(SPAPR_PCI_MEM_WIN_BUS_OFFSET),
2182             cpu_to_be64(phb->mem_win_addr),
2183             cpu_to_be64(phb->mem_win_size),
2184         },
2185         {
2186             cpu_to_be32(b_ss(3)), cpu_to_be64(phb->mem64_win_pciaddr),
2187             cpu_to_be64(phb->mem64_win_addr),
2188             cpu_to_be64(phb->mem64_win_size),
2189         },
2190     };
2191     const unsigned sizeof_ranges =
2192         (phb->mem64_win_size ? 3 : 2) * sizeof(ranges[0]);
2193     uint64_t bus_reg[] = { cpu_to_be64(phb->buid), 0 };
2194     uint32_t interrupt_map_mask[] = {
2195         cpu_to_be32(b_ddddd(-1)|b_fff(0)), 0x0, 0x0, cpu_to_be32(-1)};
2196     uint32_t interrupt_map[PCI_SLOT_MAX * PCI_NUM_PINS][7];
2197     uint32_t ddw_applicable[] = {
2198         cpu_to_be32(RTAS_IBM_QUERY_PE_DMA_WINDOW),
2199         cpu_to_be32(RTAS_IBM_CREATE_PE_DMA_WINDOW),
2200         cpu_to_be32(RTAS_IBM_REMOVE_PE_DMA_WINDOW)
2201     };
2202     uint32_t ddw_extensions[] = {
2203         cpu_to_be32(1),
2204         cpu_to_be32(RTAS_IBM_RESET_PE_DMA_WINDOW)
2205     };
2206     uint32_t associativity[] = {cpu_to_be32(0x4),
2207                                 cpu_to_be32(0x0),
2208                                 cpu_to_be32(0x0),
2209                                 cpu_to_be32(0x0),
2210                                 cpu_to_be32(phb->numa_node)};
2211     SpaprTceTable *tcet;
2212     PCIBus *bus = PCI_HOST_BRIDGE(phb)->bus;
2213     SpaprFdt s_fdt;
2214     SpaprDrc *drc;
2215 
2216     /* Start populating the FDT */
2217     nodename = g_strdup_printf("pci@%" PRIx64, phb->buid);
2218     _FDT(bus_off = fdt_add_subnode(fdt, 0, nodename));
2219     g_free(nodename);
2220     if (node_offset) {
2221         *node_offset = bus_off;
2222     }
2223 
2224     /* Write PHB properties */
2225     _FDT(fdt_setprop_string(fdt, bus_off, "device_type", "pci"));
2226     _FDT(fdt_setprop_string(fdt, bus_off, "compatible", "IBM,Logical_PHB"));
2227     _FDT(fdt_setprop_cell(fdt, bus_off, "#address-cells", 0x3));
2228     _FDT(fdt_setprop_cell(fdt, bus_off, "#size-cells", 0x2));
2229     _FDT(fdt_setprop_cell(fdt, bus_off, "#interrupt-cells", 0x1));
2230     _FDT(fdt_setprop(fdt, bus_off, "used-by-rtas", NULL, 0));
2231     _FDT(fdt_setprop(fdt, bus_off, "bus-range", &bus_range, sizeof(bus_range)));
2232     _FDT(fdt_setprop(fdt, bus_off, "ranges", &ranges, sizeof_ranges));
2233     _FDT(fdt_setprop(fdt, bus_off, "reg", &bus_reg, sizeof(bus_reg)));
2234     _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pci-config-space-type", 0x1));
2235     _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pe-total-#msi", nr_msis));
2236 
2237     /* Dynamic DMA window */
2238     if (phb->ddw_enabled) {
2239         _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-applicable", &ddw_applicable,
2240                          sizeof(ddw_applicable)));
2241         _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-extensions",
2242                          &ddw_extensions, sizeof(ddw_extensions)));
2243     }
2244 
2245     /* Advertise NUMA via ibm,associativity */
2246     if (phb->numa_node != -1) {
2247         _FDT(fdt_setprop(fdt, bus_off, "ibm,associativity", associativity,
2248                          sizeof(associativity)));
2249     }
2250 
2251     /* Build the interrupt-map, this must matches what is done
2252      * in pci_spapr_map_irq
2253      */
2254     _FDT(fdt_setprop(fdt, bus_off, "interrupt-map-mask",
2255                      &interrupt_map_mask, sizeof(interrupt_map_mask)));
2256     for (i = 0; i < PCI_SLOT_MAX; i++) {
2257         for (j = 0; j < PCI_NUM_PINS; j++) {
2258             uint32_t *irqmap = interrupt_map[i*PCI_NUM_PINS + j];
2259             int lsi_num = pci_spapr_swizzle(i, j);
2260 
2261             irqmap[0] = cpu_to_be32(b_ddddd(i)|b_fff(0));
2262             irqmap[1] = 0;
2263             irqmap[2] = 0;
2264             irqmap[3] = cpu_to_be32(j+1);
2265             irqmap[4] = cpu_to_be32(intc_phandle);
2266             spapr_dt_irq(&irqmap[5], phb->lsi_table[lsi_num].irq, true);
2267         }
2268     }
2269     /* Write interrupt map */
2270     _FDT(fdt_setprop(fdt, bus_off, "interrupt-map", &interrupt_map,
2271                      sizeof(interrupt_map)));
2272 
2273     tcet = spapr_tce_find_by_liobn(phb->dma_liobn[0]);
2274     if (!tcet) {
2275         return -1;
2276     }
2277     spapr_dma_dt(fdt, bus_off, "ibm,dma-window",
2278                  tcet->liobn, tcet->bus_offset,
2279                  tcet->nb_table << tcet->page_shift);
2280 
2281     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, phb->index);
2282     if (drc) {
2283         uint32_t drc_index = cpu_to_be32(spapr_drc_index(drc));
2284 
2285         _FDT(fdt_setprop(fdt, bus_off, "ibm,my-drc-index", &drc_index,
2286                          sizeof(drc_index)));
2287     }
2288 
2289     /* Walk the bridges and program the bus numbers*/
2290     spapr_phb_pci_enumerate(phb);
2291     _FDT(fdt_setprop_cell(fdt, bus_off, "qemu,phb-enumerated", 0x1));
2292 
2293     /* Populate tree nodes with PCI devices attached */
2294     s_fdt.fdt = fdt;
2295     s_fdt.node_off = bus_off;
2296     s_fdt.sphb = phb;
2297     pci_for_each_device_reverse(bus, pci_bus_num(bus),
2298                                 spapr_populate_pci_devices_dt,
2299                                 &s_fdt);
2300 
2301     ret = spapr_drc_populate_dt(fdt, bus_off, OBJECT(phb),
2302                                 SPAPR_DR_CONNECTOR_TYPE_PCI);
2303     if (ret) {
2304         return ret;
2305     }
2306 
2307     return 0;
2308 }
2309 
2310 void spapr_pci_rtas_init(void)
2311 {
2312     spapr_rtas_register(RTAS_READ_PCI_CONFIG, "read-pci-config",
2313                         rtas_read_pci_config);
2314     spapr_rtas_register(RTAS_WRITE_PCI_CONFIG, "write-pci-config",
2315                         rtas_write_pci_config);
2316     spapr_rtas_register(RTAS_IBM_READ_PCI_CONFIG, "ibm,read-pci-config",
2317                         rtas_ibm_read_pci_config);
2318     spapr_rtas_register(RTAS_IBM_WRITE_PCI_CONFIG, "ibm,write-pci-config",
2319                         rtas_ibm_write_pci_config);
2320     if (msi_nonbroken) {
2321         spapr_rtas_register(RTAS_IBM_QUERY_INTERRUPT_SOURCE_NUMBER,
2322                             "ibm,query-interrupt-source-number",
2323                             rtas_ibm_query_interrupt_source_number);
2324         spapr_rtas_register(RTAS_IBM_CHANGE_MSI, "ibm,change-msi",
2325                             rtas_ibm_change_msi);
2326     }
2327 
2328     spapr_rtas_register(RTAS_IBM_SET_EEH_OPTION,
2329                         "ibm,set-eeh-option",
2330                         rtas_ibm_set_eeh_option);
2331     spapr_rtas_register(RTAS_IBM_GET_CONFIG_ADDR_INFO2,
2332                         "ibm,get-config-addr-info2",
2333                         rtas_ibm_get_config_addr_info2);
2334     spapr_rtas_register(RTAS_IBM_READ_SLOT_RESET_STATE2,
2335                         "ibm,read-slot-reset-state2",
2336                         rtas_ibm_read_slot_reset_state2);
2337     spapr_rtas_register(RTAS_IBM_SET_SLOT_RESET,
2338                         "ibm,set-slot-reset",
2339                         rtas_ibm_set_slot_reset);
2340     spapr_rtas_register(RTAS_IBM_CONFIGURE_PE,
2341                         "ibm,configure-pe",
2342                         rtas_ibm_configure_pe);
2343     spapr_rtas_register(RTAS_IBM_SLOT_ERROR_DETAIL,
2344                         "ibm,slot-error-detail",
2345                         rtas_ibm_slot_error_detail);
2346 }
2347 
2348 static void spapr_pci_register_types(void)
2349 {
2350     type_register_static(&spapr_phb_info);
2351     type_register_static(&spapr_phb_root_bus_info);
2352 }
2353 
2354 type_init(spapr_pci_register_types)
2355 
2356 static int spapr_switch_one_vga(DeviceState *dev, void *opaque)
2357 {
2358     bool be = *(bool *)opaque;
2359 
2360     if (object_dynamic_cast(OBJECT(dev), "VGA")
2361         || object_dynamic_cast(OBJECT(dev), "secondary-vga")) {
2362         object_property_set_bool(OBJECT(dev), be, "big-endian-framebuffer",
2363                                  &error_abort);
2364     }
2365     return 0;
2366 }
2367 
2368 void spapr_pci_switch_vga(bool big_endian)
2369 {
2370     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
2371     SpaprPhbState *sphb;
2372 
2373     /*
2374      * For backward compatibility with existing guests, we switch
2375      * the endianness of the VGA controller when changing the guest
2376      * interrupt mode
2377      */
2378     QLIST_FOREACH(sphb, &spapr->phbs, list) {
2379         BusState *bus = &PCI_HOST_BRIDGE(sphb)->bus->qbus;
2380         qbus_walk_children(bus, spapr_switch_one_vga, NULL, NULL, NULL,
2381                            &big_endian);
2382     }
2383 }
2384