xref: /openbmc/qemu/hw/ppc/spapr_pci.c (revision b097ba37)
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 void pci_spapr_set_irq(void *opaque, int irq_num, int level)
723 {
724     /*
725      * Here we use the number returned by pci_swizzle_map_irq_fn to find a
726      * corresponding qemu_irq.
727      */
728     SpaprPhbState *phb = opaque;
729 
730     trace_spapr_pci_lsi_set(phb->dtbusname, irq_num, phb->lsi_table[irq_num].irq);
731     qemu_set_irq(spapr_phb_lsi_qirq(phb, irq_num), level);
732 }
733 
734 static PCIINTxRoute spapr_route_intx_pin_to_irq(void *opaque, int pin)
735 {
736     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(opaque);
737     PCIINTxRoute route;
738 
739     route.mode = PCI_INTX_ENABLED;
740     route.irq = sphb->lsi_table[pin].irq;
741 
742     return route;
743 }
744 
745 /*
746  * MSI/MSIX memory region implementation.
747  * The handler handles both MSI and MSIX.
748  * The vector number is encoded in least bits in data.
749  */
750 static void spapr_msi_write(void *opaque, hwaddr addr,
751                             uint64_t data, unsigned size)
752 {
753     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
754     uint32_t irq = data;
755 
756     trace_spapr_pci_msi_write(addr, data, irq);
757 
758     qemu_irq_pulse(spapr_qirq(spapr, irq));
759 }
760 
761 static const MemoryRegionOps spapr_msi_ops = {
762     /* There is no .read as the read result is undefined by PCI spec */
763     .read = NULL,
764     .write = spapr_msi_write,
765     .endianness = DEVICE_LITTLE_ENDIAN
766 };
767 
768 /*
769  * PHB PCI device
770  */
771 static AddressSpace *spapr_pci_dma_iommu(PCIBus *bus, void *opaque, int devfn)
772 {
773     SpaprPhbState *phb = opaque;
774 
775     return &phb->iommu_as;
776 }
777 
778 static char *spapr_phb_vfio_get_loc_code(SpaprPhbState *sphb,  PCIDevice *pdev)
779 {
780     char *path = NULL, *buf = NULL, *host = NULL;
781 
782     /* Get the PCI VFIO host id */
783     host = object_property_get_str(OBJECT(pdev), "host", NULL);
784     if (!host) {
785         goto err_out;
786     }
787 
788     /* Construct the path of the file that will give us the DT location */
789     path = g_strdup_printf("/sys/bus/pci/devices/%s/devspec", host);
790     g_free(host);
791     if (!g_file_get_contents(path, &buf, NULL, NULL)) {
792         goto err_out;
793     }
794     g_free(path);
795 
796     /* Construct and read from host device tree the loc-code */
797     path = g_strdup_printf("/proc/device-tree%s/ibm,loc-code", buf);
798     g_free(buf);
799     if (!g_file_get_contents(path, &buf, NULL, NULL)) {
800         goto err_out;
801     }
802     return buf;
803 
804 err_out:
805     g_free(path);
806     return NULL;
807 }
808 
809 static char *spapr_phb_get_loc_code(SpaprPhbState *sphb, PCIDevice *pdev)
810 {
811     char *buf;
812     const char *devtype = "qemu";
813     uint32_t busnr = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(pdev))));
814 
815     if (object_dynamic_cast(OBJECT(pdev), "vfio-pci")) {
816         buf = spapr_phb_vfio_get_loc_code(sphb, pdev);
817         if (buf) {
818             return buf;
819         }
820         devtype = "vfio";
821     }
822     /*
823      * For emulated devices and VFIO-failure case, make up
824      * the loc-code.
825      */
826     buf = g_strdup_printf("%s_%s:%04x:%02x:%02x.%x",
827                           devtype, pdev->name, sphb->index, busnr,
828                           PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
829     return buf;
830 }
831 
832 /* Macros to operate with address in OF binding to PCI */
833 #define b_x(x, p, l)    (((x) & ((1<<(l))-1)) << (p))
834 #define b_n(x)          b_x((x), 31, 1) /* 0 if relocatable */
835 #define b_p(x)          b_x((x), 30, 1) /* 1 if prefetchable */
836 #define b_t(x)          b_x((x), 29, 1) /* 1 if the address is aliased */
837 #define b_ss(x)         b_x((x), 24, 2) /* the space code */
838 #define b_bbbbbbbb(x)   b_x((x), 16, 8) /* bus number */
839 #define b_ddddd(x)      b_x((x), 11, 5) /* device number */
840 #define b_fff(x)        b_x((x), 8, 3)  /* function number */
841 #define b_rrrrrrrr(x)   b_x((x), 0, 8)  /* register number */
842 
843 /* for 'reg'/'assigned-addresses' OF properties */
844 #define RESOURCE_CELLS_SIZE 2
845 #define RESOURCE_CELLS_ADDRESS 3
846 
847 typedef struct ResourceFields {
848     uint32_t phys_hi;
849     uint32_t phys_mid;
850     uint32_t phys_lo;
851     uint32_t size_hi;
852     uint32_t size_lo;
853 } QEMU_PACKED ResourceFields;
854 
855 typedef struct ResourceProps {
856     ResourceFields reg[8];
857     ResourceFields assigned[7];
858     uint32_t reg_len;
859     uint32_t assigned_len;
860 } ResourceProps;
861 
862 /* fill in the 'reg'/'assigned-resources' OF properties for
863  * a PCI device. 'reg' describes resource requirements for a
864  * device's IO/MEM regions, 'assigned-addresses' describes the
865  * actual resource assignments.
866  *
867  * the properties are arrays of ('phys-addr', 'size') pairs describing
868  * the addressable regions of the PCI device, where 'phys-addr' is a
869  * RESOURCE_CELLS_ADDRESS-tuple of 32-bit integers corresponding to
870  * (phys.hi, phys.mid, phys.lo), and 'size' is a
871  * RESOURCE_CELLS_SIZE-tuple corresponding to (size.hi, size.lo).
872  *
873  * phys.hi = 0xYYXXXXZZ, where:
874  *   0xYY = npt000ss
875  *          |||   |
876  *          |||   +-- space code
877  *          |||               |
878  *          |||               +  00 if configuration space
879  *          |||               +  01 if IO region,
880  *          |||               +  10 if 32-bit MEM region
881  *          |||               +  11 if 64-bit MEM region
882  *          |||
883  *          ||+------ for non-relocatable IO: 1 if aliased
884  *          ||        for relocatable IO: 1 if below 64KB
885  *          ||        for MEM: 1 if below 1MB
886  *          |+------- 1 if region is prefetchable
887  *          +-------- 1 if region is non-relocatable
888  *   0xXXXX = bbbbbbbb dddddfff, encoding bus, slot, and function
889  *            bits respectively
890  *   0xZZ = rrrrrrrr, the register number of the BAR corresponding
891  *          to the region
892  *
893  * phys.mid and phys.lo correspond respectively to the hi/lo portions
894  * of the actual address of the region.
895  *
896  * how the phys-addr/size values are used differ slightly between
897  * 'reg' and 'assigned-addresses' properties. namely, 'reg' has
898  * an additional description for the config space region of the
899  * device, and in the case of QEMU has n=0 and phys.mid=phys.lo=0
900  * to describe the region as relocatable, with an address-mapping
901  * that corresponds directly to the PHB's address space for the
902  * resource. 'assigned-addresses' always has n=1 set with an absolute
903  * address assigned for the resource. in general, 'assigned-addresses'
904  * won't be populated, since addresses for PCI devices are generally
905  * unmapped initially and left to the guest to assign.
906  *
907  * note also that addresses defined in these properties are, at least
908  * for PAPR guests, relative to the PHBs IO/MEM windows, and
909  * correspond directly to the addresses in the BARs.
910  *
911  * in accordance with PCI Bus Binding to Open Firmware,
912  * IEEE Std 1275-1994, section 4.1.1, as implemented by PAPR+ v2.7,
913  * Appendix C.
914  */
915 static void populate_resource_props(PCIDevice *d, ResourceProps *rp)
916 {
917     int bus_num = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(d))));
918     uint32_t dev_id = (b_bbbbbbbb(bus_num) |
919                        b_ddddd(PCI_SLOT(d->devfn)) |
920                        b_fff(PCI_FUNC(d->devfn)));
921     ResourceFields *reg, *assigned;
922     int i, reg_idx = 0, assigned_idx = 0;
923 
924     /* config space region */
925     reg = &rp->reg[reg_idx++];
926     reg->phys_hi = cpu_to_be32(dev_id);
927     reg->phys_mid = 0;
928     reg->phys_lo = 0;
929     reg->size_hi = 0;
930     reg->size_lo = 0;
931 
932     for (i = 0; i < PCI_NUM_REGIONS; i++) {
933         if (!d->io_regions[i].size) {
934             continue;
935         }
936 
937         reg = &rp->reg[reg_idx++];
938 
939         reg->phys_hi = cpu_to_be32(dev_id | b_rrrrrrrr(pci_bar(d, i)));
940         if (d->io_regions[i].type & PCI_BASE_ADDRESS_SPACE_IO) {
941             reg->phys_hi |= cpu_to_be32(b_ss(1));
942         } else if (d->io_regions[i].type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
943             reg->phys_hi |= cpu_to_be32(b_ss(3));
944         } else {
945             reg->phys_hi |= cpu_to_be32(b_ss(2));
946         }
947         reg->phys_mid = 0;
948         reg->phys_lo = 0;
949         reg->size_hi = cpu_to_be32(d->io_regions[i].size >> 32);
950         reg->size_lo = cpu_to_be32(d->io_regions[i].size);
951 
952         if (d->io_regions[i].addr == PCI_BAR_UNMAPPED) {
953             continue;
954         }
955 
956         assigned = &rp->assigned[assigned_idx++];
957         assigned->phys_hi = cpu_to_be32(be32_to_cpu(reg->phys_hi) | b_n(1));
958         assigned->phys_mid = cpu_to_be32(d->io_regions[i].addr >> 32);
959         assigned->phys_lo = cpu_to_be32(d->io_regions[i].addr);
960         assigned->size_hi = reg->size_hi;
961         assigned->size_lo = reg->size_lo;
962     }
963 
964     rp->reg_len = reg_idx * sizeof(ResourceFields);
965     rp->assigned_len = assigned_idx * sizeof(ResourceFields);
966 }
967 
968 typedef struct PCIClass PCIClass;
969 typedef struct PCISubClass PCISubClass;
970 typedef struct PCIIFace PCIIFace;
971 
972 struct PCIIFace {
973     int iface;
974     const char *name;
975 };
976 
977 struct PCISubClass {
978     int subclass;
979     const char *name;
980     const PCIIFace *iface;
981 };
982 
983 struct PCIClass {
984     const char *name;
985     const PCISubClass *subc;
986 };
987 
988 static const PCISubClass undef_subclass[] = {
989     { PCI_CLASS_NOT_DEFINED_VGA, "display", NULL },
990     { 0xFF, NULL, NULL },
991 };
992 
993 static const PCISubClass mass_subclass[] = {
994     { PCI_CLASS_STORAGE_SCSI, "scsi", NULL },
995     { PCI_CLASS_STORAGE_IDE, "ide", NULL },
996     { PCI_CLASS_STORAGE_FLOPPY, "fdc", NULL },
997     { PCI_CLASS_STORAGE_IPI, "ipi", NULL },
998     { PCI_CLASS_STORAGE_RAID, "raid", NULL },
999     { PCI_CLASS_STORAGE_ATA, "ata", NULL },
1000     { PCI_CLASS_STORAGE_SATA, "sata", NULL },
1001     { PCI_CLASS_STORAGE_SAS, "sas", NULL },
1002     { 0xFF, NULL, NULL },
1003 };
1004 
1005 static const PCISubClass net_subclass[] = {
1006     { PCI_CLASS_NETWORK_ETHERNET, "ethernet", NULL },
1007     { PCI_CLASS_NETWORK_TOKEN_RING, "token-ring", NULL },
1008     { PCI_CLASS_NETWORK_FDDI, "fddi", NULL },
1009     { PCI_CLASS_NETWORK_ATM, "atm", NULL },
1010     { PCI_CLASS_NETWORK_ISDN, "isdn", NULL },
1011     { PCI_CLASS_NETWORK_WORLDFIP, "worldfip", NULL },
1012     { PCI_CLASS_NETWORK_PICMG214, "picmg", NULL },
1013     { 0xFF, NULL, NULL },
1014 };
1015 
1016 static const PCISubClass displ_subclass[] = {
1017     { PCI_CLASS_DISPLAY_VGA, "vga", NULL },
1018     { PCI_CLASS_DISPLAY_XGA, "xga", NULL },
1019     { PCI_CLASS_DISPLAY_3D, "3d-controller", NULL },
1020     { 0xFF, NULL, NULL },
1021 };
1022 
1023 static const PCISubClass media_subclass[] = {
1024     { PCI_CLASS_MULTIMEDIA_VIDEO, "video", NULL },
1025     { PCI_CLASS_MULTIMEDIA_AUDIO, "sound", NULL },
1026     { PCI_CLASS_MULTIMEDIA_PHONE, "telephony", NULL },
1027     { 0xFF, NULL, NULL },
1028 };
1029 
1030 static const PCISubClass mem_subclass[] = {
1031     { PCI_CLASS_MEMORY_RAM, "memory", NULL },
1032     { PCI_CLASS_MEMORY_FLASH, "flash", NULL },
1033     { 0xFF, NULL, NULL },
1034 };
1035 
1036 static const PCISubClass bridg_subclass[] = {
1037     { PCI_CLASS_BRIDGE_HOST, "host", NULL },
1038     { PCI_CLASS_BRIDGE_ISA, "isa", NULL },
1039     { PCI_CLASS_BRIDGE_EISA, "eisa", NULL },
1040     { PCI_CLASS_BRIDGE_MC, "mca", NULL },
1041     { PCI_CLASS_BRIDGE_PCI, "pci", NULL },
1042     { PCI_CLASS_BRIDGE_PCMCIA, "pcmcia", NULL },
1043     { PCI_CLASS_BRIDGE_NUBUS, "nubus", NULL },
1044     { PCI_CLASS_BRIDGE_CARDBUS, "cardbus", NULL },
1045     { PCI_CLASS_BRIDGE_RACEWAY, "raceway", NULL },
1046     { PCI_CLASS_BRIDGE_PCI_SEMITP, "semi-transparent-pci", NULL },
1047     { PCI_CLASS_BRIDGE_IB_PCI, "infiniband", NULL },
1048     { 0xFF, NULL, NULL },
1049 };
1050 
1051 static const PCISubClass comm_subclass[] = {
1052     { PCI_CLASS_COMMUNICATION_SERIAL, "serial", NULL },
1053     { PCI_CLASS_COMMUNICATION_PARALLEL, "parallel", NULL },
1054     { PCI_CLASS_COMMUNICATION_MULTISERIAL, "multiport-serial", NULL },
1055     { PCI_CLASS_COMMUNICATION_MODEM, "modem", NULL },
1056     { PCI_CLASS_COMMUNICATION_GPIB, "gpib", NULL },
1057     { PCI_CLASS_COMMUNICATION_SC, "smart-card", NULL },
1058     { 0xFF, NULL, NULL, },
1059 };
1060 
1061 static const PCIIFace pic_iface[] = {
1062     { PCI_CLASS_SYSTEM_PIC_IOAPIC, "io-apic" },
1063     { PCI_CLASS_SYSTEM_PIC_IOXAPIC, "io-xapic" },
1064     { 0xFF, NULL },
1065 };
1066 
1067 static const PCISubClass sys_subclass[] = {
1068     { PCI_CLASS_SYSTEM_PIC, "interrupt-controller", pic_iface },
1069     { PCI_CLASS_SYSTEM_DMA, "dma-controller", NULL },
1070     { PCI_CLASS_SYSTEM_TIMER, "timer", NULL },
1071     { PCI_CLASS_SYSTEM_RTC, "rtc", NULL },
1072     { PCI_CLASS_SYSTEM_PCI_HOTPLUG, "hot-plug-controller", NULL },
1073     { PCI_CLASS_SYSTEM_SDHCI, "sd-host-controller", NULL },
1074     { 0xFF, NULL, NULL },
1075 };
1076 
1077 static const PCISubClass inp_subclass[] = {
1078     { PCI_CLASS_INPUT_KEYBOARD, "keyboard", NULL },
1079     { PCI_CLASS_INPUT_PEN, "pen", NULL },
1080     { PCI_CLASS_INPUT_MOUSE, "mouse", NULL },
1081     { PCI_CLASS_INPUT_SCANNER, "scanner", NULL },
1082     { PCI_CLASS_INPUT_GAMEPORT, "gameport", NULL },
1083     { 0xFF, NULL, NULL },
1084 };
1085 
1086 static const PCISubClass dock_subclass[] = {
1087     { PCI_CLASS_DOCKING_GENERIC, "dock", NULL },
1088     { 0xFF, NULL, NULL },
1089 };
1090 
1091 static const PCISubClass cpu_subclass[] = {
1092     { PCI_CLASS_PROCESSOR_PENTIUM, "pentium", NULL },
1093     { PCI_CLASS_PROCESSOR_POWERPC, "powerpc", NULL },
1094     { PCI_CLASS_PROCESSOR_MIPS, "mips", NULL },
1095     { PCI_CLASS_PROCESSOR_CO, "co-processor", NULL },
1096     { 0xFF, NULL, NULL },
1097 };
1098 
1099 static const PCIIFace usb_iface[] = {
1100     { PCI_CLASS_SERIAL_USB_UHCI, "usb-uhci" },
1101     { PCI_CLASS_SERIAL_USB_OHCI, "usb-ohci", },
1102     { PCI_CLASS_SERIAL_USB_EHCI, "usb-ehci" },
1103     { PCI_CLASS_SERIAL_USB_XHCI, "usb-xhci" },
1104     { PCI_CLASS_SERIAL_USB_UNKNOWN, "usb-unknown" },
1105     { PCI_CLASS_SERIAL_USB_DEVICE, "usb-device" },
1106     { 0xFF, NULL },
1107 };
1108 
1109 static const PCISubClass ser_subclass[] = {
1110     { PCI_CLASS_SERIAL_FIREWIRE, "firewire", NULL },
1111     { PCI_CLASS_SERIAL_ACCESS, "access-bus", NULL },
1112     { PCI_CLASS_SERIAL_SSA, "ssa", NULL },
1113     { PCI_CLASS_SERIAL_USB, "usb", usb_iface },
1114     { PCI_CLASS_SERIAL_FIBER, "fibre-channel", NULL },
1115     { PCI_CLASS_SERIAL_SMBUS, "smb", NULL },
1116     { PCI_CLASS_SERIAL_IB, "infiniband", NULL },
1117     { PCI_CLASS_SERIAL_IPMI, "ipmi", NULL },
1118     { PCI_CLASS_SERIAL_SERCOS, "sercos", NULL },
1119     { PCI_CLASS_SERIAL_CANBUS, "canbus", NULL },
1120     { 0xFF, NULL, NULL },
1121 };
1122 
1123 static const PCISubClass wrl_subclass[] = {
1124     { PCI_CLASS_WIRELESS_IRDA, "irda", NULL },
1125     { PCI_CLASS_WIRELESS_CIR, "consumer-ir", NULL },
1126     { PCI_CLASS_WIRELESS_RF_CONTROLLER, "rf-controller", NULL },
1127     { PCI_CLASS_WIRELESS_BLUETOOTH, "bluetooth", NULL },
1128     { PCI_CLASS_WIRELESS_BROADBAND, "broadband", NULL },
1129     { 0xFF, NULL, NULL },
1130 };
1131 
1132 static const PCISubClass sat_subclass[] = {
1133     { PCI_CLASS_SATELLITE_TV, "satellite-tv", NULL },
1134     { PCI_CLASS_SATELLITE_AUDIO, "satellite-audio", NULL },
1135     { PCI_CLASS_SATELLITE_VOICE, "satellite-voice", NULL },
1136     { PCI_CLASS_SATELLITE_DATA, "satellite-data", NULL },
1137     { 0xFF, NULL, NULL },
1138 };
1139 
1140 static const PCISubClass crypt_subclass[] = {
1141     { PCI_CLASS_CRYPT_NETWORK, "network-encryption", NULL },
1142     { PCI_CLASS_CRYPT_ENTERTAINMENT,
1143       "entertainment-encryption", NULL },
1144     { 0xFF, NULL, NULL },
1145 };
1146 
1147 static const PCISubClass spc_subclass[] = {
1148     { PCI_CLASS_SP_DPIO, "dpio", NULL },
1149     { PCI_CLASS_SP_PERF, "counter", NULL },
1150     { PCI_CLASS_SP_SYNCH, "measurement", NULL },
1151     { PCI_CLASS_SP_MANAGEMENT, "management-card", NULL },
1152     { 0xFF, NULL, NULL },
1153 };
1154 
1155 static const PCIClass pci_classes[] = {
1156     { "legacy-device", undef_subclass },
1157     { "mass-storage",  mass_subclass },
1158     { "network", net_subclass },
1159     { "display", displ_subclass, },
1160     { "multimedia-device", media_subclass },
1161     { "memory-controller", mem_subclass },
1162     { "unknown-bridge", bridg_subclass },
1163     { "communication-controller", comm_subclass},
1164     { "system-peripheral", sys_subclass },
1165     { "input-controller", inp_subclass },
1166     { "docking-station", dock_subclass },
1167     { "cpu", cpu_subclass },
1168     { "serial-bus", ser_subclass },
1169     { "wireless-controller", wrl_subclass },
1170     { "intelligent-io", NULL },
1171     { "satellite-device", sat_subclass },
1172     { "encryption", crypt_subclass },
1173     { "data-processing-controller", spc_subclass },
1174 };
1175 
1176 static const char *pci_find_device_name(uint8_t class, uint8_t subclass,
1177                                         uint8_t iface)
1178 {
1179     const PCIClass *pclass;
1180     const PCISubClass *psubclass;
1181     const PCIIFace *piface;
1182     const char *name;
1183 
1184     if (class >= ARRAY_SIZE(pci_classes)) {
1185         return "pci";
1186     }
1187 
1188     pclass = pci_classes + class;
1189     name = pclass->name;
1190 
1191     if (pclass->subc == NULL) {
1192         return name;
1193     }
1194 
1195     psubclass = pclass->subc;
1196     while ((psubclass->subclass & 0xff) != 0xff) {
1197         if ((psubclass->subclass & 0xff) == subclass) {
1198             name = psubclass->name;
1199             break;
1200         }
1201         psubclass++;
1202     }
1203 
1204     piface = psubclass->iface;
1205     if (piface == NULL) {
1206         return name;
1207     }
1208     while ((piface->iface & 0xff) != 0xff) {
1209         if ((piface->iface & 0xff) == iface) {
1210             name = piface->name;
1211             break;
1212         }
1213         piface++;
1214     }
1215 
1216     return name;
1217 }
1218 
1219 static gchar *pci_get_node_name(PCIDevice *dev)
1220 {
1221     int slot = PCI_SLOT(dev->devfn);
1222     int func = PCI_FUNC(dev->devfn);
1223     uint32_t ccode = pci_default_read_config(dev, PCI_CLASS_PROG, 3);
1224     const char *name;
1225 
1226     name = pci_find_device_name((ccode >> 16) & 0xff, (ccode >> 8) & 0xff,
1227                                 ccode & 0xff);
1228 
1229     if (func != 0) {
1230         return g_strdup_printf("%s@%x,%x", name, slot, func);
1231     } else {
1232         return g_strdup_printf("%s@%x", name, slot);
1233     }
1234 }
1235 
1236 static uint32_t spapr_phb_get_pci_drc_index(SpaprPhbState *phb,
1237                                             PCIDevice *pdev);
1238 
1239 static void spapr_populate_pci_child_dt(PCIDevice *dev, void *fdt, int offset,
1240                                        SpaprPhbState *sphb)
1241 {
1242     ResourceProps rp;
1243     bool is_bridge = false;
1244     int pci_status;
1245     char *buf = NULL;
1246     uint32_t drc_index = spapr_phb_get_pci_drc_index(sphb, dev);
1247     uint32_t ccode = pci_default_read_config(dev, PCI_CLASS_PROG, 3);
1248     uint32_t max_msi, max_msix;
1249 
1250     if (pci_default_read_config(dev, PCI_HEADER_TYPE, 1) ==
1251         PCI_HEADER_TYPE_BRIDGE) {
1252         is_bridge = true;
1253     }
1254 
1255     /* in accordance with PAPR+ v2.7 13.6.3, Table 181 */
1256     _FDT(fdt_setprop_cell(fdt, offset, "vendor-id",
1257                           pci_default_read_config(dev, PCI_VENDOR_ID, 2)));
1258     _FDT(fdt_setprop_cell(fdt, offset, "device-id",
1259                           pci_default_read_config(dev, PCI_DEVICE_ID, 2)));
1260     _FDT(fdt_setprop_cell(fdt, offset, "revision-id",
1261                           pci_default_read_config(dev, PCI_REVISION_ID, 1)));
1262     _FDT(fdt_setprop_cell(fdt, offset, "class-code", ccode));
1263     if (pci_default_read_config(dev, PCI_INTERRUPT_PIN, 1)) {
1264         _FDT(fdt_setprop_cell(fdt, offset, "interrupts",
1265                  pci_default_read_config(dev, PCI_INTERRUPT_PIN, 1)));
1266     }
1267 
1268     if (!is_bridge) {
1269         _FDT(fdt_setprop_cell(fdt, offset, "min-grant",
1270             pci_default_read_config(dev, PCI_MIN_GNT, 1)));
1271         _FDT(fdt_setprop_cell(fdt, offset, "max-latency",
1272             pci_default_read_config(dev, PCI_MAX_LAT, 1)));
1273     }
1274 
1275     if (pci_default_read_config(dev, PCI_SUBSYSTEM_ID, 2)) {
1276         _FDT(fdt_setprop_cell(fdt, offset, "subsystem-id",
1277                  pci_default_read_config(dev, PCI_SUBSYSTEM_ID, 2)));
1278     }
1279 
1280     if (pci_default_read_config(dev, PCI_SUBSYSTEM_VENDOR_ID, 2)) {
1281         _FDT(fdt_setprop_cell(fdt, offset, "subsystem-vendor-id",
1282                  pci_default_read_config(dev, PCI_SUBSYSTEM_VENDOR_ID, 2)));
1283     }
1284 
1285     _FDT(fdt_setprop_cell(fdt, offset, "cache-line-size",
1286         pci_default_read_config(dev, PCI_CACHE_LINE_SIZE, 1)));
1287 
1288     /* the following fdt cells are masked off the pci status register */
1289     pci_status = pci_default_read_config(dev, PCI_STATUS, 2);
1290     _FDT(fdt_setprop_cell(fdt, offset, "devsel-speed",
1291                           PCI_STATUS_DEVSEL_MASK & pci_status));
1292 
1293     if (pci_status & PCI_STATUS_FAST_BACK) {
1294         _FDT(fdt_setprop(fdt, offset, "fast-back-to-back", NULL, 0));
1295     }
1296     if (pci_status & PCI_STATUS_66MHZ) {
1297         _FDT(fdt_setprop(fdt, offset, "66mhz-capable", NULL, 0));
1298     }
1299     if (pci_status & PCI_STATUS_UDF) {
1300         _FDT(fdt_setprop(fdt, offset, "udf-supported", NULL, 0));
1301     }
1302 
1303     _FDT(fdt_setprop_string(fdt, offset, "name",
1304                             pci_find_device_name((ccode >> 16) & 0xff,
1305                                                  (ccode >> 8) & 0xff,
1306                                                  ccode & 0xff)));
1307 
1308     buf = spapr_phb_get_loc_code(sphb, dev);
1309     _FDT(fdt_setprop_string(fdt, offset, "ibm,loc-code", buf));
1310     g_free(buf);
1311 
1312     if (drc_index) {
1313         _FDT(fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index));
1314     }
1315 
1316     _FDT(fdt_setprop_cell(fdt, offset, "#address-cells",
1317                           RESOURCE_CELLS_ADDRESS));
1318     _FDT(fdt_setprop_cell(fdt, offset, "#size-cells",
1319                           RESOURCE_CELLS_SIZE));
1320 
1321     if (msi_present(dev)) {
1322         max_msi = msi_nr_vectors_allocated(dev);
1323         if (max_msi) {
1324             _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi", max_msi));
1325         }
1326     }
1327     if (msix_present(dev)) {
1328         max_msix = dev->msix_entries_nr;
1329         if (max_msix) {
1330             _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi-x", max_msix));
1331         }
1332     }
1333 
1334     populate_resource_props(dev, &rp);
1335     _FDT(fdt_setprop(fdt, offset, "reg", (uint8_t *)rp.reg, rp.reg_len));
1336     _FDT(fdt_setprop(fdt, offset, "assigned-addresses",
1337                      (uint8_t *)rp.assigned, rp.assigned_len));
1338 
1339     if (sphb->pcie_ecs && pci_is_express(dev)) {
1340         _FDT(fdt_setprop_cell(fdt, offset, "ibm,pci-config-space-type", 0x1));
1341     }
1342 
1343     spapr_phb_nvgpu_populate_pcidev_dt(dev, fdt, offset, sphb);
1344 }
1345 
1346 /* create OF node for pci device and required OF DT properties */
1347 static int spapr_create_pci_child_dt(SpaprPhbState *phb, PCIDevice *dev,
1348                                      void *fdt, int node_offset)
1349 {
1350     int offset;
1351     gchar *nodename;
1352 
1353     nodename = pci_get_node_name(dev);
1354     _FDT(offset = fdt_add_subnode(fdt, node_offset, nodename));
1355     g_free(nodename);
1356 
1357     spapr_populate_pci_child_dt(dev, fdt, offset, phb);
1358 
1359     return offset;
1360 }
1361 
1362 /* Callback to be called during DRC release. */
1363 void spapr_phb_remove_pci_device_cb(DeviceState *dev)
1364 {
1365     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
1366 
1367     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
1368     object_unparent(OBJECT(dev));
1369 }
1370 
1371 static SpaprDrc *spapr_phb_get_pci_func_drc(SpaprPhbState *phb,
1372                                                     uint32_t busnr,
1373                                                     int32_t devfn)
1374 {
1375     return spapr_drc_by_id(TYPE_SPAPR_DRC_PCI,
1376                            (phb->index << 16) | (busnr << 8) | devfn);
1377 }
1378 
1379 static SpaprDrc *spapr_phb_get_pci_drc(SpaprPhbState *phb,
1380                                                PCIDevice *pdev)
1381 {
1382     uint32_t busnr = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(pdev))));
1383     return spapr_phb_get_pci_func_drc(phb, busnr, pdev->devfn);
1384 }
1385 
1386 static uint32_t spapr_phb_get_pci_drc_index(SpaprPhbState *phb,
1387                                             PCIDevice *pdev)
1388 {
1389     SpaprDrc *drc = spapr_phb_get_pci_drc(phb, pdev);
1390 
1391     if (!drc) {
1392         return 0;
1393     }
1394 
1395     return spapr_drc_index(drc);
1396 }
1397 
1398 int spapr_pci_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
1399                           void *fdt, int *fdt_start_offset, Error **errp)
1400 {
1401     HotplugHandler *plug_handler = qdev_get_hotplug_handler(drc->dev);
1402     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(plug_handler);
1403     PCIDevice *pdev = PCI_DEVICE(drc->dev);
1404 
1405     *fdt_start_offset = spapr_create_pci_child_dt(sphb, pdev, fdt, 0);
1406     return 0;
1407 }
1408 
1409 static void spapr_pci_plug(HotplugHandler *plug_handler,
1410                            DeviceState *plugged_dev, Error **errp)
1411 {
1412     SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1413     PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1414     SpaprDrc *drc = spapr_phb_get_pci_drc(phb, pdev);
1415     Error *local_err = NULL;
1416     PCIBus *bus = PCI_BUS(qdev_get_parent_bus(DEVICE(pdev)));
1417     uint32_t slotnr = PCI_SLOT(pdev->devfn);
1418 
1419     /* if DR is disabled we don't need to do anything in the case of
1420      * hotplug or coldplug callbacks
1421      */
1422     if (!phb->dr_enabled) {
1423         /* if this is a hotplug operation initiated by the user
1424          * we need to let them know it's not enabled
1425          */
1426         if (plugged_dev->hotplugged) {
1427             error_setg(&local_err, QERR_BUS_NO_HOTPLUG,
1428                        object_get_typename(OBJECT(phb)));
1429         }
1430         goto out;
1431     }
1432 
1433     g_assert(drc);
1434 
1435     /* Following the QEMU convention used for PCIe multifunction
1436      * hotplug, we do not allow functions to be hotplugged to a
1437      * slot that already has function 0 present
1438      */
1439     if (plugged_dev->hotplugged && bus->devices[PCI_DEVFN(slotnr, 0)] &&
1440         PCI_FUNC(pdev->devfn) != 0) {
1441         error_setg(&local_err, "PCI: slot %d function 0 already ocuppied by %s,"
1442                    " additional functions can no longer be exposed to guest.",
1443                    slotnr, bus->devices[PCI_DEVFN(slotnr, 0)]->name);
1444         goto out;
1445     }
1446 
1447     spapr_drc_attach(drc, DEVICE(pdev), &local_err);
1448     if (local_err) {
1449         goto out;
1450     }
1451 
1452     /* If this is function 0, signal hotplug for all the device functions.
1453      * Otherwise defer sending the hotplug event.
1454      */
1455     if (!spapr_drc_hotplugged(plugged_dev)) {
1456         spapr_drc_reset(drc);
1457     } else if (PCI_FUNC(pdev->devfn) == 0) {
1458         int i;
1459 
1460         for (i = 0; i < 8; i++) {
1461             SpaprDrc *func_drc;
1462             SpaprDrcClass *func_drck;
1463             SpaprDREntitySense state;
1464 
1465             func_drc = spapr_phb_get_pci_func_drc(phb, pci_bus_num(bus),
1466                                                   PCI_DEVFN(slotnr, i));
1467             func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1468             state = func_drck->dr_entity_sense(func_drc);
1469 
1470             if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) {
1471                 spapr_hotplug_req_add_by_index(func_drc);
1472             }
1473         }
1474     }
1475 
1476 out:
1477     error_propagate(errp, local_err);
1478 }
1479 
1480 static void spapr_pci_unplug(HotplugHandler *plug_handler,
1481                              DeviceState *plugged_dev, Error **errp)
1482 {
1483     /* some version guests do not wait for completion of a device
1484      * cleanup (generally done asynchronously by the kernel) before
1485      * signaling to QEMU that the device is safe, but instead sleep
1486      * for some 'safe' period of time. unfortunately on a busy host
1487      * this sleep isn't guaranteed to be long enough, resulting in
1488      * bad things like IRQ lines being left asserted during final
1489      * device removal. to deal with this we call reset just prior
1490      * to finalizing the device, which will put the device back into
1491      * an 'idle' state, as the device cleanup code expects.
1492      */
1493     pci_device_reset(PCI_DEVICE(plugged_dev));
1494     object_property_set_bool(OBJECT(plugged_dev), false, "realized", NULL);
1495 }
1496 
1497 static void spapr_pci_unplug_request(HotplugHandler *plug_handler,
1498                                      DeviceState *plugged_dev, Error **errp)
1499 {
1500     SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1501     PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1502     SpaprDrc *drc = spapr_phb_get_pci_drc(phb, pdev);
1503 
1504     if (!phb->dr_enabled) {
1505         error_setg(errp, QERR_BUS_NO_HOTPLUG,
1506                    object_get_typename(OBJECT(phb)));
1507         return;
1508     }
1509 
1510     g_assert(drc);
1511     g_assert(drc->dev == plugged_dev);
1512 
1513     if (!spapr_drc_unplug_requested(drc)) {
1514         PCIBus *bus = PCI_BUS(qdev_get_parent_bus(DEVICE(pdev)));
1515         uint32_t slotnr = PCI_SLOT(pdev->devfn);
1516         SpaprDrc *func_drc;
1517         SpaprDrcClass *func_drck;
1518         SpaprDREntitySense state;
1519         int i;
1520 
1521         /* ensure any other present functions are pending unplug */
1522         if (PCI_FUNC(pdev->devfn) == 0) {
1523             for (i = 1; i < 8; i++) {
1524                 func_drc = spapr_phb_get_pci_func_drc(phb, pci_bus_num(bus),
1525                                                       PCI_DEVFN(slotnr, i));
1526                 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1527                 state = func_drck->dr_entity_sense(func_drc);
1528                 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT
1529                     && !spapr_drc_unplug_requested(func_drc)) {
1530                     error_setg(errp,
1531                                "PCI: slot %d, function %d still present. "
1532                                "Must unplug all non-0 functions first.",
1533                                slotnr, i);
1534                     return;
1535                 }
1536             }
1537         }
1538 
1539         spapr_drc_detach(drc);
1540 
1541         /* if this isn't func 0, defer unplug event. otherwise signal removal
1542          * for all present functions
1543          */
1544         if (PCI_FUNC(pdev->devfn) == 0) {
1545             for (i = 7; i >= 0; i--) {
1546                 func_drc = spapr_phb_get_pci_func_drc(phb, pci_bus_num(bus),
1547                                                       PCI_DEVFN(slotnr, i));
1548                 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1549                 state = func_drck->dr_entity_sense(func_drc);
1550                 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) {
1551                     spapr_hotplug_req_remove_by_index(func_drc);
1552                 }
1553             }
1554         }
1555     }
1556 }
1557 
1558 static void spapr_phb_finalizefn(Object *obj)
1559 {
1560     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(obj);
1561 
1562     g_free(sphb->dtbusname);
1563     sphb->dtbusname = NULL;
1564 }
1565 
1566 static void spapr_phb_unrealize(DeviceState *dev, Error **errp)
1567 {
1568     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
1569     SysBusDevice *s = SYS_BUS_DEVICE(dev);
1570     PCIHostState *phb = PCI_HOST_BRIDGE(s);
1571     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(phb);
1572     SpaprTceTable *tcet;
1573     int i;
1574     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
1575 
1576     spapr_phb_nvgpu_free(sphb);
1577 
1578     if (sphb->msi) {
1579         g_hash_table_unref(sphb->msi);
1580         sphb->msi = NULL;
1581     }
1582 
1583     /*
1584      * Remove IO/MMIO subregions and aliases, rest should get cleaned
1585      * via PHB's unrealize->object_finalize
1586      */
1587     for (i = windows_supported - 1; i >= 0; i--) {
1588         tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[i]);
1589         if (tcet) {
1590             memory_region_del_subregion(&sphb->iommu_root,
1591                                         spapr_tce_get_iommu(tcet));
1592         }
1593     }
1594 
1595     if (sphb->dr_enabled) {
1596         for (i = PCI_SLOT_MAX * 8 - 1; i >= 0; i--) {
1597             SpaprDrc *drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PCI,
1598                                                     (sphb->index << 16) | i);
1599 
1600             if (drc) {
1601                 object_unparent(OBJECT(drc));
1602             }
1603         }
1604     }
1605 
1606     for (i = PCI_NUM_PINS - 1; i >= 0; i--) {
1607         if (sphb->lsi_table[i].irq) {
1608             spapr_irq_free(spapr, sphb->lsi_table[i].irq, 1);
1609             sphb->lsi_table[i].irq = 0;
1610         }
1611     }
1612 
1613     QLIST_REMOVE(sphb, list);
1614 
1615     memory_region_del_subregion(&sphb->iommu_root, &sphb->msiwindow);
1616 
1617     address_space_destroy(&sphb->iommu_as);
1618 
1619     qbus_set_hotplug_handler(BUS(phb->bus), NULL, &error_abort);
1620     pci_unregister_root_bus(phb->bus);
1621 
1622     memory_region_del_subregion(get_system_memory(), &sphb->iowindow);
1623     if (sphb->mem64_win_pciaddr != (hwaddr)-1) {
1624         memory_region_del_subregion(get_system_memory(), &sphb->mem64window);
1625     }
1626     memory_region_del_subregion(get_system_memory(), &sphb->mem32window);
1627 }
1628 
1629 static bool spapr_phb_allows_extended_config_space(PCIBus *bus)
1630 {
1631     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(BUS(bus)->parent);
1632 
1633     return sphb->pcie_ecs;
1634 }
1635 
1636 static void spapr_phb_root_bus_class_init(ObjectClass *klass, void *data)
1637 {
1638     PCIBusClass *pbc = PCI_BUS_CLASS(klass);
1639 
1640     pbc->allows_extended_config_space = spapr_phb_allows_extended_config_space;
1641 }
1642 
1643 #define TYPE_SPAPR_PHB_ROOT_BUS "pci"
1644 
1645 static const TypeInfo spapr_phb_root_bus_info = {
1646     .name = TYPE_SPAPR_PHB_ROOT_BUS,
1647     .parent = TYPE_PCI_BUS,
1648     .class_init = spapr_phb_root_bus_class_init,
1649 };
1650 
1651 static void spapr_phb_realize(DeviceState *dev, Error **errp)
1652 {
1653     /* We don't use SPAPR_MACHINE() in order to exit gracefully if the user
1654      * tries to add a sPAPR PHB to a non-pseries machine.
1655      */
1656     SpaprMachineState *spapr =
1657         (SpaprMachineState *) object_dynamic_cast(qdev_get_machine(),
1658                                                   TYPE_SPAPR_MACHINE);
1659     SpaprMachineClass *smc = spapr ? SPAPR_MACHINE_GET_CLASS(spapr) : NULL;
1660     SysBusDevice *s = SYS_BUS_DEVICE(dev);
1661     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(s);
1662     PCIHostState *phb = PCI_HOST_BRIDGE(s);
1663     char *namebuf;
1664     int i;
1665     PCIBus *bus;
1666     uint64_t msi_window_size = 4096;
1667     SpaprTceTable *tcet;
1668     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
1669 
1670     if (!spapr) {
1671         error_setg(errp, TYPE_SPAPR_PCI_HOST_BRIDGE " needs a pseries machine");
1672         return;
1673     }
1674 
1675     assert(sphb->index != (uint32_t)-1); /* checked in spapr_phb_pre_plug() */
1676 
1677     if (sphb->mem64_win_size != 0) {
1678         if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) {
1679             error_setg(errp, "32-bit memory window of size 0x%"HWADDR_PRIx
1680                        " (max 2 GiB)", sphb->mem_win_size);
1681             return;
1682         }
1683 
1684         /* 64-bit window defaults to identity mapping */
1685         sphb->mem64_win_pciaddr = sphb->mem64_win_addr;
1686     } else if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) {
1687         /*
1688          * For compatibility with old configuration, if no 64-bit MMIO
1689          * window is specified, but the ordinary (32-bit) memory
1690          * window is specified as > 2GiB, we treat it as a 2GiB 32-bit
1691          * window, with a 64-bit MMIO window following on immediately
1692          * afterwards
1693          */
1694         sphb->mem64_win_size = sphb->mem_win_size - SPAPR_PCI_MEM32_WIN_SIZE;
1695         sphb->mem64_win_addr = sphb->mem_win_addr + SPAPR_PCI_MEM32_WIN_SIZE;
1696         sphb->mem64_win_pciaddr =
1697             SPAPR_PCI_MEM_WIN_BUS_OFFSET + SPAPR_PCI_MEM32_WIN_SIZE;
1698         sphb->mem_win_size = SPAPR_PCI_MEM32_WIN_SIZE;
1699     }
1700 
1701     if (spapr_pci_find_phb(spapr, sphb->buid)) {
1702         error_setg(errp, "PCI host bridges must have unique BUIDs");
1703         return;
1704     }
1705 
1706     if (sphb->numa_node != -1 &&
1707         (sphb->numa_node >= MAX_NODES || !numa_info[sphb->numa_node].present)) {
1708         error_setg(errp, "Invalid NUMA node ID for PCI host bridge");
1709         return;
1710     }
1711 
1712     sphb->dtbusname = g_strdup_printf("pci@%" PRIx64, sphb->buid);
1713 
1714     /* Initialize memory regions */
1715     namebuf = g_strdup_printf("%s.mmio", sphb->dtbusname);
1716     memory_region_init(&sphb->memspace, OBJECT(sphb), namebuf, UINT64_MAX);
1717     g_free(namebuf);
1718 
1719     namebuf = g_strdup_printf("%s.mmio32-alias", sphb->dtbusname);
1720     memory_region_init_alias(&sphb->mem32window, OBJECT(sphb),
1721                              namebuf, &sphb->memspace,
1722                              SPAPR_PCI_MEM_WIN_BUS_OFFSET, sphb->mem_win_size);
1723     g_free(namebuf);
1724     memory_region_add_subregion(get_system_memory(), sphb->mem_win_addr,
1725                                 &sphb->mem32window);
1726 
1727     if (sphb->mem64_win_size != 0) {
1728         namebuf = g_strdup_printf("%s.mmio64-alias", sphb->dtbusname);
1729         memory_region_init_alias(&sphb->mem64window, OBJECT(sphb),
1730                                  namebuf, &sphb->memspace,
1731                                  sphb->mem64_win_pciaddr, sphb->mem64_win_size);
1732         g_free(namebuf);
1733 
1734         memory_region_add_subregion(get_system_memory(),
1735                                     sphb->mem64_win_addr,
1736                                     &sphb->mem64window);
1737     }
1738 
1739     /* Initialize IO regions */
1740     namebuf = g_strdup_printf("%s.io", sphb->dtbusname);
1741     memory_region_init(&sphb->iospace, OBJECT(sphb),
1742                        namebuf, SPAPR_PCI_IO_WIN_SIZE);
1743     g_free(namebuf);
1744 
1745     namebuf = g_strdup_printf("%s.io-alias", sphb->dtbusname);
1746     memory_region_init_alias(&sphb->iowindow, OBJECT(sphb), namebuf,
1747                              &sphb->iospace, 0, SPAPR_PCI_IO_WIN_SIZE);
1748     g_free(namebuf);
1749     memory_region_add_subregion(get_system_memory(), sphb->io_win_addr,
1750                                 &sphb->iowindow);
1751 
1752     bus = pci_register_root_bus(dev, NULL,
1753                                 pci_spapr_set_irq, pci_swizzle_map_irq_fn, sphb,
1754                                 &sphb->memspace, &sphb->iospace,
1755                                 PCI_DEVFN(0, 0), PCI_NUM_PINS,
1756                                 TYPE_SPAPR_PHB_ROOT_BUS);
1757     phb->bus = bus;
1758     qbus_set_hotplug_handler(BUS(phb->bus), OBJECT(sphb), NULL);
1759 
1760     /*
1761      * Initialize PHB address space.
1762      * By default there will be at least one subregion for default
1763      * 32bit DMA window.
1764      * Later the guest might want to create another DMA window
1765      * which will become another memory subregion.
1766      */
1767     namebuf = g_strdup_printf("%s.iommu-root", sphb->dtbusname);
1768     memory_region_init(&sphb->iommu_root, OBJECT(sphb),
1769                        namebuf, UINT64_MAX);
1770     g_free(namebuf);
1771     address_space_init(&sphb->iommu_as, &sphb->iommu_root,
1772                        sphb->dtbusname);
1773 
1774     /*
1775      * As MSI/MSIX interrupts trigger by writing at MSI/MSIX vectors,
1776      * we need to allocate some memory to catch those writes coming
1777      * from msi_notify()/msix_notify().
1778      * As MSIMessage:addr is going to be the same and MSIMessage:data
1779      * is going to be a VIRQ number, 4 bytes of the MSI MR will only
1780      * be used.
1781      *
1782      * For KVM we want to ensure that this memory is a full page so that
1783      * our memory slot is of page size granularity.
1784      */
1785 #ifdef CONFIG_KVM
1786     if (kvm_enabled()) {
1787         msi_window_size = getpagesize();
1788     }
1789 #endif
1790 
1791     memory_region_init_io(&sphb->msiwindow, OBJECT(sphb), &spapr_msi_ops, spapr,
1792                           "msi", msi_window_size);
1793     memory_region_add_subregion(&sphb->iommu_root, SPAPR_PCI_MSI_WINDOW,
1794                                 &sphb->msiwindow);
1795 
1796     pci_setup_iommu(bus, spapr_pci_dma_iommu, sphb);
1797 
1798     pci_bus_set_route_irq_fn(bus, spapr_route_intx_pin_to_irq);
1799 
1800     QLIST_INSERT_HEAD(&spapr->phbs, sphb, list);
1801 
1802     /* Initialize the LSI table */
1803     for (i = 0; i < PCI_NUM_PINS; i++) {
1804         uint32_t irq = SPAPR_IRQ_PCI_LSI + sphb->index * PCI_NUM_PINS + i;
1805         Error *local_err = NULL;
1806 
1807         if (smc->legacy_irq_allocation) {
1808             irq = spapr_irq_findone(spapr, &local_err);
1809             if (local_err) {
1810                 error_propagate_prepend(errp, local_err,
1811                                         "can't allocate LSIs: ");
1812                 /*
1813                  * Older machines will never support PHB hotplug, ie, this is an
1814                  * init only path and QEMU will terminate. No need to rollback.
1815                  */
1816                 return;
1817             }
1818         }
1819 
1820         spapr_irq_claim(spapr, irq, true, &local_err);
1821         if (local_err) {
1822             error_propagate_prepend(errp, local_err, "can't allocate LSIs: ");
1823             goto unrealize;
1824         }
1825 
1826         sphb->lsi_table[i].irq = irq;
1827     }
1828 
1829     /* allocate connectors for child PCI devices */
1830     if (sphb->dr_enabled) {
1831         for (i = 0; i < PCI_SLOT_MAX * 8; i++) {
1832             spapr_dr_connector_new(OBJECT(phb), TYPE_SPAPR_DRC_PCI,
1833                                    (sphb->index << 16) | i);
1834         }
1835     }
1836 
1837     /* DMA setup */
1838     for (i = 0; i < windows_supported; ++i) {
1839         tcet = spapr_tce_new_table(DEVICE(sphb), sphb->dma_liobn[i]);
1840         if (!tcet) {
1841             error_setg(errp, "Creating window#%d failed for %s",
1842                        i, sphb->dtbusname);
1843             goto unrealize;
1844         }
1845         memory_region_add_subregion(&sphb->iommu_root, 0,
1846                                     spapr_tce_get_iommu(tcet));
1847     }
1848 
1849     sphb->msi = g_hash_table_new_full(g_int_hash, g_int_equal, g_free, g_free);
1850     return;
1851 
1852 unrealize:
1853     spapr_phb_unrealize(dev, NULL);
1854 }
1855 
1856 static int spapr_phb_children_reset(Object *child, void *opaque)
1857 {
1858     DeviceState *dev = (DeviceState *) object_dynamic_cast(child, TYPE_DEVICE);
1859 
1860     if (dev) {
1861         device_reset(dev);
1862     }
1863 
1864     return 0;
1865 }
1866 
1867 void spapr_phb_dma_reset(SpaprPhbState *sphb)
1868 {
1869     int i;
1870     SpaprTceTable *tcet;
1871 
1872     for (i = 0; i < SPAPR_PCI_DMA_MAX_WINDOWS; ++i) {
1873         tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[i]);
1874 
1875         if (tcet && tcet->nb_table) {
1876             spapr_tce_table_disable(tcet);
1877         }
1878     }
1879 
1880     /* Register default 32bit DMA window */
1881     tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[0]);
1882     spapr_tce_table_enable(tcet, SPAPR_TCE_PAGE_SHIFT, sphb->dma_win_addr,
1883                            sphb->dma_win_size >> SPAPR_TCE_PAGE_SHIFT);
1884 }
1885 
1886 static void spapr_phb_reset(DeviceState *qdev)
1887 {
1888     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(qdev);
1889     Error *errp = NULL;
1890 
1891     spapr_phb_dma_reset(sphb);
1892     spapr_phb_nvgpu_free(sphb);
1893     spapr_phb_nvgpu_setup(sphb, &errp);
1894     if (errp) {
1895         error_report_err(errp);
1896     }
1897 
1898     /* Reset the IOMMU state */
1899     object_child_foreach(OBJECT(qdev), spapr_phb_children_reset, NULL);
1900 
1901     if (spapr_phb_eeh_available(SPAPR_PCI_HOST_BRIDGE(qdev))) {
1902         spapr_phb_vfio_reset(qdev);
1903     }
1904 }
1905 
1906 static Property spapr_phb_properties[] = {
1907     DEFINE_PROP_UINT32("index", SpaprPhbState, index, -1),
1908     DEFINE_PROP_UINT64("mem_win_size", SpaprPhbState, mem_win_size,
1909                        SPAPR_PCI_MEM32_WIN_SIZE),
1910     DEFINE_PROP_UINT64("mem64_win_size", SpaprPhbState, mem64_win_size,
1911                        SPAPR_PCI_MEM64_WIN_SIZE),
1912     DEFINE_PROP_UINT64("io_win_size", SpaprPhbState, io_win_size,
1913                        SPAPR_PCI_IO_WIN_SIZE),
1914     DEFINE_PROP_BOOL("dynamic-reconfiguration", SpaprPhbState, dr_enabled,
1915                      true),
1916     /* Default DMA window is 0..1GB */
1917     DEFINE_PROP_UINT64("dma_win_addr", SpaprPhbState, dma_win_addr, 0),
1918     DEFINE_PROP_UINT64("dma_win_size", SpaprPhbState, dma_win_size, 0x40000000),
1919     DEFINE_PROP_UINT64("dma64_win_addr", SpaprPhbState, dma64_win_addr,
1920                        0x800000000000000ULL),
1921     DEFINE_PROP_BOOL("ddw", SpaprPhbState, ddw_enabled, true),
1922     DEFINE_PROP_UINT64("pgsz", SpaprPhbState, page_size_mask,
1923                        (1ULL << 12) | (1ULL << 16)),
1924     DEFINE_PROP_UINT32("numa_node", SpaprPhbState, numa_node, -1),
1925     DEFINE_PROP_BOOL("pre-2.8-migration", SpaprPhbState,
1926                      pre_2_8_migration, false),
1927     DEFINE_PROP_BOOL("pcie-extended-configuration-space", SpaprPhbState,
1928                      pcie_ecs, true),
1929     DEFINE_PROP_UINT64("gpa", SpaprPhbState, nv2_gpa_win_addr, 0),
1930     DEFINE_PROP_UINT64("atsd", SpaprPhbState, nv2_atsd_win_addr, 0),
1931     DEFINE_PROP_END_OF_LIST(),
1932 };
1933 
1934 static const VMStateDescription vmstate_spapr_pci_lsi = {
1935     .name = "spapr_pci/lsi",
1936     .version_id = 1,
1937     .minimum_version_id = 1,
1938     .fields = (VMStateField[]) {
1939         VMSTATE_UINT32_EQUAL(irq, struct spapr_pci_lsi, NULL),
1940 
1941         VMSTATE_END_OF_LIST()
1942     },
1943 };
1944 
1945 static const VMStateDescription vmstate_spapr_pci_msi = {
1946     .name = "spapr_pci/msi",
1947     .version_id = 1,
1948     .minimum_version_id = 1,
1949     .fields = (VMStateField []) {
1950         VMSTATE_UINT32(key, spapr_pci_msi_mig),
1951         VMSTATE_UINT32(value.first_irq, spapr_pci_msi_mig),
1952         VMSTATE_UINT32(value.num, spapr_pci_msi_mig),
1953         VMSTATE_END_OF_LIST()
1954     },
1955 };
1956 
1957 static int spapr_pci_pre_save(void *opaque)
1958 {
1959     SpaprPhbState *sphb = opaque;
1960     GHashTableIter iter;
1961     gpointer key, value;
1962     int i;
1963 
1964     if (sphb->pre_2_8_migration) {
1965         sphb->mig_liobn = sphb->dma_liobn[0];
1966         sphb->mig_mem_win_addr = sphb->mem_win_addr;
1967         sphb->mig_mem_win_size = sphb->mem_win_size;
1968         sphb->mig_io_win_addr = sphb->io_win_addr;
1969         sphb->mig_io_win_size = sphb->io_win_size;
1970 
1971         if ((sphb->mem64_win_size != 0)
1972             && (sphb->mem64_win_addr
1973                 == (sphb->mem_win_addr + sphb->mem_win_size))) {
1974             sphb->mig_mem_win_size += sphb->mem64_win_size;
1975         }
1976     }
1977 
1978     g_free(sphb->msi_devs);
1979     sphb->msi_devs = NULL;
1980     sphb->msi_devs_num = g_hash_table_size(sphb->msi);
1981     if (!sphb->msi_devs_num) {
1982         return 0;
1983     }
1984     sphb->msi_devs = g_new(spapr_pci_msi_mig, sphb->msi_devs_num);
1985 
1986     g_hash_table_iter_init(&iter, sphb->msi);
1987     for (i = 0; g_hash_table_iter_next(&iter, &key, &value); ++i) {
1988         sphb->msi_devs[i].key = *(uint32_t *) key;
1989         sphb->msi_devs[i].value = *(spapr_pci_msi *) value;
1990     }
1991 
1992     return 0;
1993 }
1994 
1995 static int spapr_pci_post_load(void *opaque, int version_id)
1996 {
1997     SpaprPhbState *sphb = opaque;
1998     gpointer key, value;
1999     int i;
2000 
2001     for (i = 0; i < sphb->msi_devs_num; ++i) {
2002         key = g_memdup(&sphb->msi_devs[i].key,
2003                        sizeof(sphb->msi_devs[i].key));
2004         value = g_memdup(&sphb->msi_devs[i].value,
2005                          sizeof(sphb->msi_devs[i].value));
2006         g_hash_table_insert(sphb->msi, key, value);
2007     }
2008     g_free(sphb->msi_devs);
2009     sphb->msi_devs = NULL;
2010     sphb->msi_devs_num = 0;
2011 
2012     return 0;
2013 }
2014 
2015 static bool pre_2_8_migration(void *opaque, int version_id)
2016 {
2017     SpaprPhbState *sphb = opaque;
2018 
2019     return sphb->pre_2_8_migration;
2020 }
2021 
2022 static const VMStateDescription vmstate_spapr_pci = {
2023     .name = "spapr_pci",
2024     .version_id = 2,
2025     .minimum_version_id = 2,
2026     .pre_save = spapr_pci_pre_save,
2027     .post_load = spapr_pci_post_load,
2028     .fields = (VMStateField[]) {
2029         VMSTATE_UINT64_EQUAL(buid, SpaprPhbState, NULL),
2030         VMSTATE_UINT32_TEST(mig_liobn, SpaprPhbState, pre_2_8_migration),
2031         VMSTATE_UINT64_TEST(mig_mem_win_addr, SpaprPhbState, pre_2_8_migration),
2032         VMSTATE_UINT64_TEST(mig_mem_win_size, SpaprPhbState, pre_2_8_migration),
2033         VMSTATE_UINT64_TEST(mig_io_win_addr, SpaprPhbState, pre_2_8_migration),
2034         VMSTATE_UINT64_TEST(mig_io_win_size, SpaprPhbState, pre_2_8_migration),
2035         VMSTATE_STRUCT_ARRAY(lsi_table, SpaprPhbState, PCI_NUM_PINS, 0,
2036                              vmstate_spapr_pci_lsi, struct spapr_pci_lsi),
2037         VMSTATE_INT32(msi_devs_num, SpaprPhbState),
2038         VMSTATE_STRUCT_VARRAY_ALLOC(msi_devs, SpaprPhbState, msi_devs_num, 0,
2039                                     vmstate_spapr_pci_msi, spapr_pci_msi_mig),
2040         VMSTATE_END_OF_LIST()
2041     },
2042 };
2043 
2044 static const char *spapr_phb_root_bus_path(PCIHostState *host_bridge,
2045                                            PCIBus *rootbus)
2046 {
2047     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(host_bridge);
2048 
2049     return sphb->dtbusname;
2050 }
2051 
2052 static void spapr_phb_class_init(ObjectClass *klass, void *data)
2053 {
2054     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(klass);
2055     DeviceClass *dc = DEVICE_CLASS(klass);
2056     HotplugHandlerClass *hp = HOTPLUG_HANDLER_CLASS(klass);
2057 
2058     hc->root_bus_path = spapr_phb_root_bus_path;
2059     dc->realize = spapr_phb_realize;
2060     dc->unrealize = spapr_phb_unrealize;
2061     dc->props = spapr_phb_properties;
2062     dc->reset = spapr_phb_reset;
2063     dc->vmsd = &vmstate_spapr_pci;
2064     /* Supported by TYPE_SPAPR_MACHINE */
2065     dc->user_creatable = true;
2066     set_bit(DEVICE_CATEGORY_BRIDGE, dc->categories);
2067     hp->plug = spapr_pci_plug;
2068     hp->unplug = spapr_pci_unplug;
2069     hp->unplug_request = spapr_pci_unplug_request;
2070 }
2071 
2072 static const TypeInfo spapr_phb_info = {
2073     .name          = TYPE_SPAPR_PCI_HOST_BRIDGE,
2074     .parent        = TYPE_PCI_HOST_BRIDGE,
2075     .instance_size = sizeof(SpaprPhbState),
2076     .instance_finalize = spapr_phb_finalizefn,
2077     .class_init    = spapr_phb_class_init,
2078     .interfaces    = (InterfaceInfo[]) {
2079         { TYPE_HOTPLUG_HANDLER },
2080         { }
2081     }
2082 };
2083 
2084 typedef struct SpaprFdt {
2085     void *fdt;
2086     int node_off;
2087     SpaprPhbState *sphb;
2088 } SpaprFdt;
2089 
2090 static void spapr_populate_pci_devices_dt(PCIBus *bus, PCIDevice *pdev,
2091                                           void *opaque)
2092 {
2093     PCIBus *sec_bus;
2094     SpaprFdt *p = opaque;
2095     int offset;
2096     SpaprFdt s_fdt;
2097 
2098     offset = spapr_create_pci_child_dt(p->sphb, pdev, p->fdt, p->node_off);
2099     if (!offset) {
2100         error_report("Failed to create pci child device tree node");
2101         return;
2102     }
2103 
2104     if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) !=
2105          PCI_HEADER_TYPE_BRIDGE)) {
2106         return;
2107     }
2108 
2109     sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev));
2110     if (!sec_bus) {
2111         return;
2112     }
2113 
2114     s_fdt.fdt = p->fdt;
2115     s_fdt.node_off = offset;
2116     s_fdt.sphb = p->sphb;
2117     pci_for_each_device_reverse(sec_bus, pci_bus_num(sec_bus),
2118                                 spapr_populate_pci_devices_dt,
2119                                 &s_fdt);
2120 }
2121 
2122 static void spapr_phb_pci_enumerate_bridge(PCIBus *bus, PCIDevice *pdev,
2123                                            void *opaque)
2124 {
2125     unsigned int *bus_no = opaque;
2126     PCIBus *sec_bus = NULL;
2127 
2128     if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) !=
2129          PCI_HEADER_TYPE_BRIDGE)) {
2130         return;
2131     }
2132 
2133     (*bus_no)++;
2134     pci_default_write_config(pdev, PCI_PRIMARY_BUS, pci_dev_bus_num(pdev), 1);
2135     pci_default_write_config(pdev, PCI_SECONDARY_BUS, *bus_no, 1);
2136     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
2137 
2138     sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev));
2139     if (!sec_bus) {
2140         return;
2141     }
2142 
2143     pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
2144                         spapr_phb_pci_enumerate_bridge, bus_no);
2145     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
2146 }
2147 
2148 static void spapr_phb_pci_enumerate(SpaprPhbState *phb)
2149 {
2150     PCIBus *bus = PCI_HOST_BRIDGE(phb)->bus;
2151     unsigned int bus_no = 0;
2152 
2153     pci_for_each_device(bus, pci_bus_num(bus),
2154                         spapr_phb_pci_enumerate_bridge,
2155                         &bus_no);
2156 
2157 }
2158 
2159 int spapr_populate_pci_dt(SpaprPhbState *phb, uint32_t intc_phandle, void *fdt,
2160                           uint32_t nr_msis, int *node_offset)
2161 {
2162     int bus_off, i, j, ret;
2163     uint32_t bus_range[] = { cpu_to_be32(0), cpu_to_be32(0xff) };
2164     struct {
2165         uint32_t hi;
2166         uint64_t child;
2167         uint64_t parent;
2168         uint64_t size;
2169     } QEMU_PACKED ranges[] = {
2170         {
2171             cpu_to_be32(b_ss(1)), cpu_to_be64(0),
2172             cpu_to_be64(phb->io_win_addr),
2173             cpu_to_be64(memory_region_size(&phb->iospace)),
2174         },
2175         {
2176             cpu_to_be32(b_ss(2)), cpu_to_be64(SPAPR_PCI_MEM_WIN_BUS_OFFSET),
2177             cpu_to_be64(phb->mem_win_addr),
2178             cpu_to_be64(phb->mem_win_size),
2179         },
2180         {
2181             cpu_to_be32(b_ss(3)), cpu_to_be64(phb->mem64_win_pciaddr),
2182             cpu_to_be64(phb->mem64_win_addr),
2183             cpu_to_be64(phb->mem64_win_size),
2184         },
2185     };
2186     const unsigned sizeof_ranges =
2187         (phb->mem64_win_size ? 3 : 2) * sizeof(ranges[0]);
2188     uint64_t bus_reg[] = { cpu_to_be64(phb->buid), 0 };
2189     uint32_t interrupt_map_mask[] = {
2190         cpu_to_be32(b_ddddd(-1)|b_fff(0)), 0x0, 0x0, cpu_to_be32(-1)};
2191     uint32_t interrupt_map[PCI_SLOT_MAX * PCI_NUM_PINS][7];
2192     uint32_t ddw_applicable[] = {
2193         cpu_to_be32(RTAS_IBM_QUERY_PE_DMA_WINDOW),
2194         cpu_to_be32(RTAS_IBM_CREATE_PE_DMA_WINDOW),
2195         cpu_to_be32(RTAS_IBM_REMOVE_PE_DMA_WINDOW)
2196     };
2197     uint32_t ddw_extensions[] = {
2198         cpu_to_be32(1),
2199         cpu_to_be32(RTAS_IBM_RESET_PE_DMA_WINDOW)
2200     };
2201     uint32_t associativity[] = {cpu_to_be32(0x4),
2202                                 cpu_to_be32(0x0),
2203                                 cpu_to_be32(0x0),
2204                                 cpu_to_be32(0x0),
2205                                 cpu_to_be32(phb->numa_node)};
2206     SpaprTceTable *tcet;
2207     PCIBus *bus = PCI_HOST_BRIDGE(phb)->bus;
2208     SpaprFdt s_fdt;
2209     SpaprDrc *drc;
2210     Error *errp = NULL;
2211 
2212     /* Start populating the FDT */
2213     _FDT(bus_off = fdt_add_subnode(fdt, 0, phb->dtbusname));
2214     if (node_offset) {
2215         *node_offset = bus_off;
2216     }
2217 
2218     /* Write PHB properties */
2219     _FDT(fdt_setprop_string(fdt, bus_off, "device_type", "pci"));
2220     _FDT(fdt_setprop_string(fdt, bus_off, "compatible", "IBM,Logical_PHB"));
2221     _FDT(fdt_setprop_cell(fdt, bus_off, "#address-cells", 0x3));
2222     _FDT(fdt_setprop_cell(fdt, bus_off, "#size-cells", 0x2));
2223     _FDT(fdt_setprop_cell(fdt, bus_off, "#interrupt-cells", 0x1));
2224     _FDT(fdt_setprop(fdt, bus_off, "used-by-rtas", NULL, 0));
2225     _FDT(fdt_setprop(fdt, bus_off, "bus-range", &bus_range, sizeof(bus_range)));
2226     _FDT(fdt_setprop(fdt, bus_off, "ranges", &ranges, sizeof_ranges));
2227     _FDT(fdt_setprop(fdt, bus_off, "reg", &bus_reg, sizeof(bus_reg)));
2228     _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pci-config-space-type", 0x1));
2229     _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pe-total-#msi", nr_msis));
2230 
2231     /* Dynamic DMA window */
2232     if (phb->ddw_enabled) {
2233         _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-applicable", &ddw_applicable,
2234                          sizeof(ddw_applicable)));
2235         _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-extensions",
2236                          &ddw_extensions, sizeof(ddw_extensions)));
2237     }
2238 
2239     /* Advertise NUMA via ibm,associativity */
2240     if (phb->numa_node != -1) {
2241         _FDT(fdt_setprop(fdt, bus_off, "ibm,associativity", associativity,
2242                          sizeof(associativity)));
2243     }
2244 
2245     /* Build the interrupt-map, this must matches what is done
2246      * in pci_swizzle_map_irq_fn
2247      */
2248     _FDT(fdt_setprop(fdt, bus_off, "interrupt-map-mask",
2249                      &interrupt_map_mask, sizeof(interrupt_map_mask)));
2250     for (i = 0; i < PCI_SLOT_MAX; i++) {
2251         for (j = 0; j < PCI_NUM_PINS; j++) {
2252             uint32_t *irqmap = interrupt_map[i*PCI_NUM_PINS + j];
2253             int lsi_num = pci_swizzle(i, j);
2254 
2255             irqmap[0] = cpu_to_be32(b_ddddd(i)|b_fff(0));
2256             irqmap[1] = 0;
2257             irqmap[2] = 0;
2258             irqmap[3] = cpu_to_be32(j+1);
2259             irqmap[4] = cpu_to_be32(intc_phandle);
2260             spapr_dt_irq(&irqmap[5], phb->lsi_table[lsi_num].irq, true);
2261         }
2262     }
2263     /* Write interrupt map */
2264     _FDT(fdt_setprop(fdt, bus_off, "interrupt-map", &interrupt_map,
2265                      sizeof(interrupt_map)));
2266 
2267     tcet = spapr_tce_find_by_liobn(phb->dma_liobn[0]);
2268     if (!tcet) {
2269         return -1;
2270     }
2271     spapr_dma_dt(fdt, bus_off, "ibm,dma-window",
2272                  tcet->liobn, tcet->bus_offset,
2273                  tcet->nb_table << tcet->page_shift);
2274 
2275     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, phb->index);
2276     if (drc) {
2277         uint32_t drc_index = cpu_to_be32(spapr_drc_index(drc));
2278 
2279         _FDT(fdt_setprop(fdt, bus_off, "ibm,my-drc-index", &drc_index,
2280                          sizeof(drc_index)));
2281     }
2282 
2283     /* Walk the bridges and program the bus numbers*/
2284     spapr_phb_pci_enumerate(phb);
2285     _FDT(fdt_setprop_cell(fdt, bus_off, "qemu,phb-enumerated", 0x1));
2286 
2287     /* Populate tree nodes with PCI devices attached */
2288     s_fdt.fdt = fdt;
2289     s_fdt.node_off = bus_off;
2290     s_fdt.sphb = phb;
2291     pci_for_each_device_reverse(bus, pci_bus_num(bus),
2292                                 spapr_populate_pci_devices_dt,
2293                                 &s_fdt);
2294 
2295     ret = spapr_drc_populate_dt(fdt, bus_off, OBJECT(phb),
2296                                 SPAPR_DR_CONNECTOR_TYPE_PCI);
2297     if (ret) {
2298         return ret;
2299     }
2300 
2301     spapr_phb_nvgpu_populate_dt(phb, fdt, bus_off, &errp);
2302     if (errp) {
2303         error_report_err(errp);
2304     }
2305     spapr_phb_nvgpu_ram_populate_dt(phb, fdt);
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