xref: /qemu/hw/ppc/spapr_pci.c (revision ce266b75febdeee5b91ab630a7158ddff3e002fb)
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 "hw/hw.h"
27 #include "hw/sysbus.h"
28 #include "hw/pci/pci.h"
29 #include "hw/pci/msi.h"
30 #include "hw/pci/msix.h"
31 #include "hw/pci/pci_host.h"
32 #include "hw/ppc/spapr.h"
33 #include "hw/pci-host/spapr.h"
34 #include "exec/address-spaces.h"
35 #include <libfdt.h>
36 #include "trace.h"
37 #include "qemu/error-report.h"
38 #include "qapi/qmp/qerror.h"
39 
40 #include "hw/pci/pci_bridge.h"
41 #include "hw/pci/pci_bus.h"
42 #include "hw/ppc/spapr_drc.h"
43 #include "sysemu/device_tree.h"
44 
45 /* Copied from the kernel arch/powerpc/platforms/pseries/msi.c */
46 #define RTAS_QUERY_FN           0
47 #define RTAS_CHANGE_FN          1
48 #define RTAS_RESET_FN           2
49 #define RTAS_CHANGE_MSI_FN      3
50 #define RTAS_CHANGE_MSIX_FN     4
51 
52 /* Interrupt types to return on RTAS_CHANGE_* */
53 #define RTAS_TYPE_MSI           1
54 #define RTAS_TYPE_MSIX          2
55 
56 #define FDT_NAME_MAX          128
57 
58 #define _FDT(exp) \
59     do { \
60         int ret = (exp);                                           \
61         if (ret < 0) {                                             \
62             return ret;                                            \
63         }                                                          \
64     } while (0)
65 
66 sPAPRPHBState *spapr_pci_find_phb(sPAPRMachineState *spapr, uint64_t buid)
67 {
68     sPAPRPHBState *sphb;
69 
70     QLIST_FOREACH(sphb, &spapr->phbs, list) {
71         if (sphb->buid != buid) {
72             continue;
73         }
74         return sphb;
75     }
76 
77     return NULL;
78 }
79 
80 PCIDevice *spapr_pci_find_dev(sPAPRMachineState *spapr, uint64_t buid,
81                               uint32_t config_addr)
82 {
83     sPAPRPHBState *sphb = spapr_pci_find_phb(spapr, buid);
84     PCIHostState *phb = PCI_HOST_BRIDGE(sphb);
85     int bus_num = (config_addr >> 16) & 0xFF;
86     int devfn = (config_addr >> 8) & 0xFF;
87 
88     if (!phb) {
89         return NULL;
90     }
91 
92     return pci_find_device(phb->bus, bus_num, devfn);
93 }
94 
95 static uint32_t rtas_pci_cfgaddr(uint32_t arg)
96 {
97     /* This handles the encoding of extended config space addresses */
98     return ((arg >> 20) & 0xf00) | (arg & 0xff);
99 }
100 
101 static void finish_read_pci_config(sPAPRMachineState *spapr, uint64_t buid,
102                                    uint32_t addr, uint32_t size,
103                                    target_ulong rets)
104 {
105     PCIDevice *pci_dev;
106     uint32_t val;
107 
108     if ((size != 1) && (size != 2) && (size != 4)) {
109         /* access must be 1, 2 or 4 bytes */
110         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
111         return;
112     }
113 
114     pci_dev = spapr_pci_find_dev(spapr, buid, addr);
115     addr = rtas_pci_cfgaddr(addr);
116 
117     if (!pci_dev || (addr % size) || (addr >= pci_config_size(pci_dev))) {
118         /* Access must be to a valid device, within bounds and
119          * naturally aligned */
120         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
121         return;
122     }
123 
124     val = pci_host_config_read_common(pci_dev, addr,
125                                       pci_config_size(pci_dev), size);
126 
127     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
128     rtas_st(rets, 1, val);
129 }
130 
131 static void rtas_ibm_read_pci_config(PowerPCCPU *cpu, sPAPRMachineState *spapr,
132                                      uint32_t token, uint32_t nargs,
133                                      target_ulong args,
134                                      uint32_t nret, target_ulong rets)
135 {
136     uint64_t buid;
137     uint32_t size, addr;
138 
139     if ((nargs != 4) || (nret != 2)) {
140         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
141         return;
142     }
143 
144     buid = rtas_ldq(args, 1);
145     size = rtas_ld(args, 3);
146     addr = rtas_ld(args, 0);
147 
148     finish_read_pci_config(spapr, buid, addr, size, rets);
149 }
150 
151 static void rtas_read_pci_config(PowerPCCPU *cpu, sPAPRMachineState *spapr,
152                                  uint32_t token, uint32_t nargs,
153                                  target_ulong args,
154                                  uint32_t nret, target_ulong rets)
155 {
156     uint32_t size, addr;
157 
158     if ((nargs != 2) || (nret != 2)) {
159         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
160         return;
161     }
162 
163     size = rtas_ld(args, 1);
164     addr = rtas_ld(args, 0);
165 
166     finish_read_pci_config(spapr, 0, addr, size, rets);
167 }
168 
169 static void finish_write_pci_config(sPAPRMachineState *spapr, uint64_t buid,
170                                     uint32_t addr, uint32_t size,
171                                     uint32_t val, target_ulong rets)
172 {
173     PCIDevice *pci_dev;
174 
175     if ((size != 1) && (size != 2) && (size != 4)) {
176         /* access must be 1, 2 or 4 bytes */
177         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
178         return;
179     }
180 
181     pci_dev = spapr_pci_find_dev(spapr, buid, addr);
182     addr = rtas_pci_cfgaddr(addr);
183 
184     if (!pci_dev || (addr % size) || (addr >= pci_config_size(pci_dev))) {
185         /* Access must be to a valid device, within bounds and
186          * naturally aligned */
187         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
188         return;
189     }
190 
191     pci_host_config_write_common(pci_dev, addr, pci_config_size(pci_dev),
192                                  val, size);
193 
194     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
195 }
196 
197 static void rtas_ibm_write_pci_config(PowerPCCPU *cpu, sPAPRMachineState *spapr,
198                                       uint32_t token, uint32_t nargs,
199                                       target_ulong args,
200                                       uint32_t nret, target_ulong rets)
201 {
202     uint64_t buid;
203     uint32_t val, size, addr;
204 
205     if ((nargs != 5) || (nret != 1)) {
206         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
207         return;
208     }
209 
210     buid = rtas_ldq(args, 1);
211     val = rtas_ld(args, 4);
212     size = rtas_ld(args, 3);
213     addr = rtas_ld(args, 0);
214 
215     finish_write_pci_config(spapr, buid, addr, size, val, rets);
216 }
217 
218 static void rtas_write_pci_config(PowerPCCPU *cpu, sPAPRMachineState *spapr,
219                                   uint32_t token, uint32_t nargs,
220                                   target_ulong args,
221                                   uint32_t nret, target_ulong rets)
222 {
223     uint32_t val, size, addr;
224 
225     if ((nargs != 3) || (nret != 1)) {
226         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
227         return;
228     }
229 
230 
231     val = rtas_ld(args, 2);
232     size = rtas_ld(args, 1);
233     addr = rtas_ld(args, 0);
234 
235     finish_write_pci_config(spapr, 0, addr, size, val, rets);
236 }
237 
238 /*
239  * Set MSI/MSIX message data.
240  * This is required for msi_notify()/msix_notify() which
241  * will write at the addresses via spapr_msi_write().
242  *
243  * If hwaddr == 0, all entries will have .data == first_irq i.e.
244  * table will be reset.
245  */
246 static void spapr_msi_setmsg(PCIDevice *pdev, hwaddr addr, bool msix,
247                              unsigned first_irq, unsigned req_num)
248 {
249     unsigned i;
250     MSIMessage msg = { .address = addr, .data = first_irq };
251 
252     if (!msix) {
253         msi_set_message(pdev, msg);
254         trace_spapr_pci_msi_setup(pdev->name, 0, msg.address);
255         return;
256     }
257 
258     for (i = 0; i < req_num; ++i) {
259         msix_set_message(pdev, i, msg);
260         trace_spapr_pci_msi_setup(pdev->name, i, msg.address);
261         if (addr) {
262             ++msg.data;
263         }
264     }
265 }
266 
267 static void rtas_ibm_change_msi(PowerPCCPU *cpu, sPAPRMachineState *spapr,
268                                 uint32_t token, uint32_t nargs,
269                                 target_ulong args, uint32_t nret,
270                                 target_ulong rets)
271 {
272     uint32_t config_addr = rtas_ld(args, 0);
273     uint64_t buid = rtas_ldq(args, 1);
274     unsigned int func = rtas_ld(args, 3);
275     unsigned int req_num = rtas_ld(args, 4); /* 0 == remove all */
276     unsigned int seq_num = rtas_ld(args, 5);
277     unsigned int ret_intr_type;
278     unsigned int irq, max_irqs = 0;
279     sPAPRPHBState *phb = NULL;
280     PCIDevice *pdev = NULL;
281     spapr_pci_msi *msi;
282     int *config_addr_key;
283 
284     switch (func) {
285     case RTAS_CHANGE_MSI_FN:
286     case RTAS_CHANGE_FN:
287         ret_intr_type = RTAS_TYPE_MSI;
288         break;
289     case RTAS_CHANGE_MSIX_FN:
290         ret_intr_type = RTAS_TYPE_MSIX;
291         break;
292     default:
293         error_report("rtas_ibm_change_msi(%u) is not implemented", func);
294         rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
295         return;
296     }
297 
298     /* Fins sPAPRPHBState */
299     phb = spapr_pci_find_phb(spapr, buid);
300     if (phb) {
301         pdev = spapr_pci_find_dev(spapr, buid, config_addr);
302     }
303     if (!phb || !pdev) {
304         rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
305         return;
306     }
307 
308     msi = (spapr_pci_msi *) g_hash_table_lookup(phb->msi, &config_addr);
309 
310     /* Releasing MSIs */
311     if (!req_num) {
312         if (!msi) {
313             trace_spapr_pci_msi("Releasing wrong config", config_addr);
314             rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
315             return;
316         }
317 
318         xics_free(spapr->icp, msi->first_irq, msi->num);
319         if (msi_present(pdev)) {
320             spapr_msi_setmsg(pdev, 0, false, 0, 0);
321         }
322         if (msix_present(pdev)) {
323             spapr_msi_setmsg(pdev, 0, true, 0, 0);
324         }
325         g_hash_table_remove(phb->msi, &config_addr);
326 
327         trace_spapr_pci_msi("Released MSIs", config_addr);
328         rtas_st(rets, 0, RTAS_OUT_SUCCESS);
329         rtas_st(rets, 1, 0);
330         return;
331     }
332 
333     /* Enabling MSI */
334 
335     /* Check if the device supports as many IRQs as requested */
336     if (ret_intr_type == RTAS_TYPE_MSI) {
337         max_irqs = msi_nr_vectors_allocated(pdev);
338     } else if (ret_intr_type == RTAS_TYPE_MSIX) {
339         max_irqs = pdev->msix_entries_nr;
340     }
341     if (!max_irqs) {
342         error_report("Requested interrupt type %d is not enabled for device %x",
343                      ret_intr_type, config_addr);
344         rtas_st(rets, 0, -1); /* Hardware error */
345         return;
346     }
347     /* Correct the number if the guest asked for too many */
348     if (req_num > max_irqs) {
349         trace_spapr_pci_msi_retry(config_addr, req_num, max_irqs);
350         req_num = max_irqs;
351         irq = 0; /* to avoid misleading trace */
352         goto out;
353     }
354 
355     /* Allocate MSIs */
356     irq = xics_alloc_block(spapr->icp, 0, req_num, false,
357                            ret_intr_type == RTAS_TYPE_MSI);
358     if (!irq) {
359         error_report("Cannot allocate MSIs for device %x", config_addr);
360         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
361         return;
362     }
363 
364     /* Release previous MSIs */
365     if (msi) {
366         xics_free(spapr->icp, msi->first_irq, msi->num);
367         g_hash_table_remove(phb->msi, &config_addr);
368     }
369 
370     /* Setup MSI/MSIX vectors in the device (via cfgspace or MSIX BAR) */
371     spapr_msi_setmsg(pdev, SPAPR_PCI_MSI_WINDOW, ret_intr_type == RTAS_TYPE_MSIX,
372                      irq, req_num);
373 
374     /* Add MSI device to cache */
375     msi = g_new(spapr_pci_msi, 1);
376     msi->first_irq = irq;
377     msi->num = req_num;
378     config_addr_key = g_new(int, 1);
379     *config_addr_key = config_addr;
380     g_hash_table_insert(phb->msi, config_addr_key, msi);
381 
382 out:
383     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
384     rtas_st(rets, 1, req_num);
385     rtas_st(rets, 2, ++seq_num);
386     if (nret > 3) {
387         rtas_st(rets, 3, ret_intr_type);
388     }
389 
390     trace_spapr_pci_rtas_ibm_change_msi(config_addr, func, req_num, irq);
391 }
392 
393 static void rtas_ibm_query_interrupt_source_number(PowerPCCPU *cpu,
394                                                    sPAPRMachineState *spapr,
395                                                    uint32_t token,
396                                                    uint32_t nargs,
397                                                    target_ulong args,
398                                                    uint32_t nret,
399                                                    target_ulong rets)
400 {
401     uint32_t config_addr = rtas_ld(args, 0);
402     uint64_t buid = rtas_ldq(args, 1);
403     unsigned int intr_src_num = -1, ioa_intr_num = rtas_ld(args, 3);
404     sPAPRPHBState *phb = NULL;
405     PCIDevice *pdev = NULL;
406     spapr_pci_msi *msi;
407 
408     /* Find sPAPRPHBState */
409     phb = spapr_pci_find_phb(spapr, buid);
410     if (phb) {
411         pdev = spapr_pci_find_dev(spapr, buid, config_addr);
412     }
413     if (!phb || !pdev) {
414         rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
415         return;
416     }
417 
418     /* Find device descriptor and start IRQ */
419     msi = (spapr_pci_msi *) g_hash_table_lookup(phb->msi, &config_addr);
420     if (!msi || !msi->first_irq || !msi->num || (ioa_intr_num >= msi->num)) {
421         trace_spapr_pci_msi("Failed to return vector", config_addr);
422         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
423         return;
424     }
425     intr_src_num = msi->first_irq + ioa_intr_num;
426     trace_spapr_pci_rtas_ibm_query_interrupt_source_number(ioa_intr_num,
427                                                            intr_src_num);
428 
429     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
430     rtas_st(rets, 1, intr_src_num);
431     rtas_st(rets, 2, 1);/* 0 == level; 1 == edge */
432 }
433 
434 static void rtas_ibm_set_eeh_option(PowerPCCPU *cpu,
435                                     sPAPRMachineState *spapr,
436                                     uint32_t token, uint32_t nargs,
437                                     target_ulong args, uint32_t nret,
438                                     target_ulong rets)
439 {
440     sPAPRPHBState *sphb;
441     sPAPRPHBClass *spc;
442     uint32_t addr, option;
443     uint64_t buid;
444     int ret;
445 
446     if ((nargs != 4) || (nret != 1)) {
447         goto param_error_exit;
448     }
449 
450     buid = rtas_ldq(args, 1);
451     addr = rtas_ld(args, 0);
452     option = rtas_ld(args, 3);
453 
454     sphb = spapr_pci_find_phb(spapr, buid);
455     if (!sphb) {
456         goto param_error_exit;
457     }
458 
459     spc = SPAPR_PCI_HOST_BRIDGE_GET_CLASS(sphb);
460     if (!spc->eeh_set_option) {
461         goto param_error_exit;
462     }
463 
464     ret = spc->eeh_set_option(sphb, addr, option);
465     rtas_st(rets, 0, ret);
466     return;
467 
468 param_error_exit:
469     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
470 }
471 
472 static void rtas_ibm_get_config_addr_info2(PowerPCCPU *cpu,
473                                            sPAPRMachineState *spapr,
474                                            uint32_t token, uint32_t nargs,
475                                            target_ulong args, uint32_t nret,
476                                            target_ulong rets)
477 {
478     sPAPRPHBState *sphb;
479     sPAPRPHBClass *spc;
480     PCIDevice *pdev;
481     uint32_t addr, option;
482     uint64_t buid;
483 
484     if ((nargs != 4) || (nret != 2)) {
485         goto param_error_exit;
486     }
487 
488     buid = rtas_ldq(args, 1);
489     sphb = spapr_pci_find_phb(spapr, buid);
490     if (!sphb) {
491         goto param_error_exit;
492     }
493 
494     spc = SPAPR_PCI_HOST_BRIDGE_GET_CLASS(sphb);
495     if (!spc->eeh_set_option) {
496         goto param_error_exit;
497     }
498 
499     /*
500      * We always have PE address of form "00BB0001". "BB"
501      * represents the bus number of PE's primary bus.
502      */
503     option = rtas_ld(args, 3);
504     switch (option) {
505     case RTAS_GET_PE_ADDR:
506         addr = rtas_ld(args, 0);
507         pdev = spapr_pci_find_dev(spapr, buid, addr);
508         if (!pdev) {
509             goto param_error_exit;
510         }
511 
512         rtas_st(rets, 1, (pci_bus_num(pdev->bus) << 16) + 1);
513         break;
514     case RTAS_GET_PE_MODE:
515         rtas_st(rets, 1, RTAS_PE_MODE_SHARED);
516         break;
517     default:
518         goto param_error_exit;
519     }
520 
521     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
522     return;
523 
524 param_error_exit:
525     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
526 }
527 
528 static void rtas_ibm_read_slot_reset_state2(PowerPCCPU *cpu,
529                                             sPAPRMachineState *spapr,
530                                             uint32_t token, uint32_t nargs,
531                                             target_ulong args, uint32_t nret,
532                                             target_ulong rets)
533 {
534     sPAPRPHBState *sphb;
535     sPAPRPHBClass *spc;
536     uint64_t buid;
537     int state, ret;
538 
539     if ((nargs != 3) || (nret != 4 && nret != 5)) {
540         goto param_error_exit;
541     }
542 
543     buid = rtas_ldq(args, 1);
544     sphb = spapr_pci_find_phb(spapr, buid);
545     if (!sphb) {
546         goto param_error_exit;
547     }
548 
549     spc = SPAPR_PCI_HOST_BRIDGE_GET_CLASS(sphb);
550     if (!spc->eeh_get_state) {
551         goto param_error_exit;
552     }
553 
554     ret = spc->eeh_get_state(sphb, &state);
555     rtas_st(rets, 0, ret);
556     if (ret != RTAS_OUT_SUCCESS) {
557         return;
558     }
559 
560     rtas_st(rets, 1, state);
561     rtas_st(rets, 2, RTAS_EEH_SUPPORT);
562     rtas_st(rets, 3, RTAS_EEH_PE_UNAVAIL_INFO);
563     if (nret >= 5) {
564         rtas_st(rets, 4, RTAS_EEH_PE_RECOVER_INFO);
565     }
566     return;
567 
568 param_error_exit:
569     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
570 }
571 
572 static void rtas_ibm_set_slot_reset(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     sPAPRPHBClass *spc;
580     uint32_t option;
581     uint64_t buid;
582     int ret;
583 
584     if ((nargs != 4) || (nret != 1)) {
585         goto param_error_exit;
586     }
587 
588     buid = rtas_ldq(args, 1);
589     option = rtas_ld(args, 3);
590     sphb = spapr_pci_find_phb(spapr, buid);
591     if (!sphb) {
592         goto param_error_exit;
593     }
594 
595     spc = SPAPR_PCI_HOST_BRIDGE_GET_CLASS(sphb);
596     if (!spc->eeh_reset) {
597         goto param_error_exit;
598     }
599 
600     ret = spc->eeh_reset(sphb, option);
601     rtas_st(rets, 0, ret);
602     return;
603 
604 param_error_exit:
605     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
606 }
607 
608 static void rtas_ibm_configure_pe(PowerPCCPU *cpu,
609                                   sPAPRMachineState *spapr,
610                                   uint32_t token, uint32_t nargs,
611                                   target_ulong args, uint32_t nret,
612                                   target_ulong rets)
613 {
614     sPAPRPHBState *sphb;
615     sPAPRPHBClass *spc;
616     uint64_t buid;
617     int ret;
618 
619     if ((nargs != 3) || (nret != 1)) {
620         goto param_error_exit;
621     }
622 
623     buid = rtas_ldq(args, 1);
624     sphb = spapr_pci_find_phb(spapr, buid);
625     if (!sphb) {
626         goto param_error_exit;
627     }
628 
629     spc = SPAPR_PCI_HOST_BRIDGE_GET_CLASS(sphb);
630     if (!spc->eeh_configure) {
631         goto param_error_exit;
632     }
633 
634     ret = spc->eeh_configure(sphb);
635     rtas_st(rets, 0, ret);
636     return;
637 
638 param_error_exit:
639     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
640 }
641 
642 /* To support it later */
643 static void rtas_ibm_slot_error_detail(PowerPCCPU *cpu,
644                                        sPAPRMachineState *spapr,
645                                        uint32_t token, uint32_t nargs,
646                                        target_ulong args, uint32_t nret,
647                                        target_ulong rets)
648 {
649     sPAPRPHBState *sphb;
650     sPAPRPHBClass *spc;
651     int option;
652     uint64_t buid;
653 
654     if ((nargs != 8) || (nret != 1)) {
655         goto param_error_exit;
656     }
657 
658     buid = rtas_ldq(args, 1);
659     sphb = spapr_pci_find_phb(spapr, buid);
660     if (!sphb) {
661         goto param_error_exit;
662     }
663 
664     spc = SPAPR_PCI_HOST_BRIDGE_GET_CLASS(sphb);
665     if (!spc->eeh_set_option) {
666         goto param_error_exit;
667     }
668 
669     option = rtas_ld(args, 7);
670     switch (option) {
671     case RTAS_SLOT_TEMP_ERR_LOG:
672     case RTAS_SLOT_PERM_ERR_LOG:
673         break;
674     default:
675         goto param_error_exit;
676     }
677 
678     /* We don't have error log yet */
679     rtas_st(rets, 0, RTAS_OUT_NO_ERRORS_FOUND);
680     return;
681 
682 param_error_exit:
683     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
684 }
685 
686 static int pci_spapr_swizzle(int slot, int pin)
687 {
688     return (slot + pin) % PCI_NUM_PINS;
689 }
690 
691 static int pci_spapr_map_irq(PCIDevice *pci_dev, int irq_num)
692 {
693     /*
694      * Here we need to convert pci_dev + irq_num to some unique value
695      * which is less than number of IRQs on the specific bus (4).  We
696      * use standard PCI swizzling, that is (slot number + pin number)
697      * % 4.
698      */
699     return pci_spapr_swizzle(PCI_SLOT(pci_dev->devfn), irq_num);
700 }
701 
702 static void pci_spapr_set_irq(void *opaque, int irq_num, int level)
703 {
704     /*
705      * Here we use the number returned by pci_spapr_map_irq to find a
706      * corresponding qemu_irq.
707      */
708     sPAPRPHBState *phb = opaque;
709 
710     trace_spapr_pci_lsi_set(phb->dtbusname, irq_num, phb->lsi_table[irq_num].irq);
711     qemu_set_irq(spapr_phb_lsi_qirq(phb, irq_num), level);
712 }
713 
714 static PCIINTxRoute spapr_route_intx_pin_to_irq(void *opaque, int pin)
715 {
716     sPAPRPHBState *sphb = SPAPR_PCI_HOST_BRIDGE(opaque);
717     PCIINTxRoute route;
718 
719     route.mode = PCI_INTX_ENABLED;
720     route.irq = sphb->lsi_table[pin].irq;
721 
722     return route;
723 }
724 
725 /*
726  * MSI/MSIX memory region implementation.
727  * The handler handles both MSI and MSIX.
728  * For MSI-X, the vector number is encoded as a part of the address,
729  * data is set to 0.
730  * For MSI, the vector number is encoded in least bits in data.
731  */
732 static void spapr_msi_write(void *opaque, hwaddr addr,
733                             uint64_t data, unsigned size)
734 {
735     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
736     uint32_t irq = data;
737 
738     trace_spapr_pci_msi_write(addr, data, irq);
739 
740     qemu_irq_pulse(xics_get_qirq(spapr->icp, irq));
741 }
742 
743 static const MemoryRegionOps spapr_msi_ops = {
744     /* There is no .read as the read result is undefined by PCI spec */
745     .read = NULL,
746     .write = spapr_msi_write,
747     .endianness = DEVICE_LITTLE_ENDIAN
748 };
749 
750 /*
751  * PHB PCI device
752  */
753 static AddressSpace *spapr_pci_dma_iommu(PCIBus *bus, void *opaque, int devfn)
754 {
755     sPAPRPHBState *phb = opaque;
756 
757     return &phb->iommu_as;
758 }
759 
760 static char *spapr_phb_vfio_get_loc_code(sPAPRPHBState *sphb,  PCIDevice *pdev)
761 {
762     char *path = NULL, *buf = NULL, *host = NULL;
763 
764     /* Get the PCI VFIO host id */
765     host = object_property_get_str(OBJECT(pdev), "host", NULL);
766     if (!host) {
767         goto err_out;
768     }
769 
770     /* Construct the path of the file that will give us the DT location */
771     path = g_strdup_printf("/sys/bus/pci/devices/%s/devspec", host);
772     g_free(host);
773     if (!path || !g_file_get_contents(path, &buf, NULL, NULL)) {
774         goto err_out;
775     }
776     g_free(path);
777 
778     /* Construct and read from host device tree the loc-code */
779     path = g_strdup_printf("/proc/device-tree%s/ibm,loc-code", buf);
780     g_free(buf);
781     if (!path || !g_file_get_contents(path, &buf, NULL, NULL)) {
782         goto err_out;
783     }
784     return buf;
785 
786 err_out:
787     g_free(path);
788     return NULL;
789 }
790 
791 static char *spapr_phb_get_loc_code(sPAPRPHBState *sphb, PCIDevice *pdev)
792 {
793     char *buf;
794     const char *devtype = "qemu";
795     uint32_t busnr = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(pdev))));
796 
797     if (object_dynamic_cast(OBJECT(pdev), "vfio-pci")) {
798         buf = spapr_phb_vfio_get_loc_code(sphb, pdev);
799         if (buf) {
800             return buf;
801         }
802         devtype = "vfio";
803     }
804     /*
805      * For emulated devices and VFIO-failure case, make up
806      * the loc-code.
807      */
808     buf = g_strdup_printf("%s_%s:%04x:%02x:%02x.%x",
809                           devtype, pdev->name, sphb->index, busnr,
810                           PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
811     return buf;
812 }
813 
814 /* Macros to operate with address in OF binding to PCI */
815 #define b_x(x, p, l)    (((x) & ((1<<(l))-1)) << (p))
816 #define b_n(x)          b_x((x), 31, 1) /* 0 if relocatable */
817 #define b_p(x)          b_x((x), 30, 1) /* 1 if prefetchable */
818 #define b_t(x)          b_x((x), 29, 1) /* 1 if the address is aliased */
819 #define b_ss(x)         b_x((x), 24, 2) /* the space code */
820 #define b_bbbbbbbb(x)   b_x((x), 16, 8) /* bus number */
821 #define b_ddddd(x)      b_x((x), 11, 5) /* device number */
822 #define b_fff(x)        b_x((x), 8, 3)  /* function number */
823 #define b_rrrrrrrr(x)   b_x((x), 0, 8)  /* register number */
824 
825 /* for 'reg'/'assigned-addresses' OF properties */
826 #define RESOURCE_CELLS_SIZE 2
827 #define RESOURCE_CELLS_ADDRESS 3
828 
829 typedef struct ResourceFields {
830     uint32_t phys_hi;
831     uint32_t phys_mid;
832     uint32_t phys_lo;
833     uint32_t size_hi;
834     uint32_t size_lo;
835 } QEMU_PACKED ResourceFields;
836 
837 typedef struct ResourceProps {
838     ResourceFields reg[8];
839     ResourceFields assigned[7];
840     uint32_t reg_len;
841     uint32_t assigned_len;
842 } ResourceProps;
843 
844 /* fill in the 'reg'/'assigned-resources' OF properties for
845  * a PCI device. 'reg' describes resource requirements for a
846  * device's IO/MEM regions, 'assigned-addresses' describes the
847  * actual resource assignments.
848  *
849  * the properties are arrays of ('phys-addr', 'size') pairs describing
850  * the addressable regions of the PCI device, where 'phys-addr' is a
851  * RESOURCE_CELLS_ADDRESS-tuple of 32-bit integers corresponding to
852  * (phys.hi, phys.mid, phys.lo), and 'size' is a
853  * RESOURCE_CELLS_SIZE-tuple corresponding to (size.hi, size.lo).
854  *
855  * phys.hi = 0xYYXXXXZZ, where:
856  *   0xYY = npt000ss
857  *          |||   |
858  *          |||   +-- space code
859  *          |||               |
860  *          |||               +  00 if configuration space
861  *          |||               +  01 if IO region,
862  *          |||               +  10 if 32-bit MEM region
863  *          |||               +  11 if 64-bit MEM region
864  *          |||
865  *          ||+------ for non-relocatable IO: 1 if aliased
866  *          ||        for relocatable IO: 1 if below 64KB
867  *          ||        for MEM: 1 if below 1MB
868  *          |+------- 1 if region is prefetchable
869  *          +-------- 1 if region is non-relocatable
870  *   0xXXXX = bbbbbbbb dddddfff, encoding bus, slot, and function
871  *            bits respectively
872  *   0xZZ = rrrrrrrr, the register number of the BAR corresponding
873  *          to the region
874  *
875  * phys.mid and phys.lo correspond respectively to the hi/lo portions
876  * of the actual address of the region.
877  *
878  * how the phys-addr/size values are used differ slightly between
879  * 'reg' and 'assigned-addresses' properties. namely, 'reg' has
880  * an additional description for the config space region of the
881  * device, and in the case of QEMU has n=0 and phys.mid=phys.lo=0
882  * to describe the region as relocatable, with an address-mapping
883  * that corresponds directly to the PHB's address space for the
884  * resource. 'assigned-addresses' always has n=1 set with an absolute
885  * address assigned for the resource. in general, 'assigned-addresses'
886  * won't be populated, since addresses for PCI devices are generally
887  * unmapped initially and left to the guest to assign.
888  *
889  * note also that addresses defined in these properties are, at least
890  * for PAPR guests, relative to the PHBs IO/MEM windows, and
891  * correspond directly to the addresses in the BARs.
892  *
893  * in accordance with PCI Bus Binding to Open Firmware,
894  * IEEE Std 1275-1994, section 4.1.1, as implemented by PAPR+ v2.7,
895  * Appendix C.
896  */
897 static void populate_resource_props(PCIDevice *d, ResourceProps *rp)
898 {
899     int bus_num = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(d))));
900     uint32_t dev_id = (b_bbbbbbbb(bus_num) |
901                        b_ddddd(PCI_SLOT(d->devfn)) |
902                        b_fff(PCI_FUNC(d->devfn)));
903     ResourceFields *reg, *assigned;
904     int i, reg_idx = 0, assigned_idx = 0;
905 
906     /* config space region */
907     reg = &rp->reg[reg_idx++];
908     reg->phys_hi = cpu_to_be32(dev_id);
909     reg->phys_mid = 0;
910     reg->phys_lo = 0;
911     reg->size_hi = 0;
912     reg->size_lo = 0;
913 
914     for (i = 0; i < PCI_NUM_REGIONS; i++) {
915         if (!d->io_regions[i].size) {
916             continue;
917         }
918 
919         reg = &rp->reg[reg_idx++];
920 
921         reg->phys_hi = cpu_to_be32(dev_id | b_rrrrrrrr(pci_bar(d, i)));
922         if (d->io_regions[i].type & PCI_BASE_ADDRESS_SPACE_IO) {
923             reg->phys_hi |= cpu_to_be32(b_ss(1));
924         } else if (d->io_regions[i].type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
925             reg->phys_hi |= cpu_to_be32(b_ss(3));
926         } else {
927             reg->phys_hi |= cpu_to_be32(b_ss(2));
928         }
929         reg->phys_mid = 0;
930         reg->phys_lo = 0;
931         reg->size_hi = cpu_to_be32(d->io_regions[i].size >> 32);
932         reg->size_lo = cpu_to_be32(d->io_regions[i].size);
933 
934         if (d->io_regions[i].addr == PCI_BAR_UNMAPPED) {
935             continue;
936         }
937 
938         assigned = &rp->assigned[assigned_idx++];
939         assigned->phys_hi = cpu_to_be32(reg->phys_hi | b_n(1));
940         assigned->phys_mid = cpu_to_be32(d->io_regions[i].addr >> 32);
941         assigned->phys_lo = cpu_to_be32(d->io_regions[i].addr);
942         assigned->size_hi = reg->size_hi;
943         assigned->size_lo = reg->size_lo;
944     }
945 
946     rp->reg_len = reg_idx * sizeof(ResourceFields);
947     rp->assigned_len = assigned_idx * sizeof(ResourceFields);
948 }
949 
950 static uint32_t spapr_phb_get_pci_drc_index(sPAPRPHBState *phb,
951                                             PCIDevice *pdev);
952 
953 static int spapr_populate_pci_child_dt(PCIDevice *dev, void *fdt, int offset,
954                                        sPAPRPHBState *sphb)
955 {
956     ResourceProps rp;
957     bool is_bridge = false;
958     int pci_status, err;
959     char *buf = NULL;
960     uint32_t drc_index = spapr_phb_get_pci_drc_index(sphb, dev);
961     uint32_t max_msi, max_msix;
962 
963     if (pci_default_read_config(dev, PCI_HEADER_TYPE, 1) ==
964         PCI_HEADER_TYPE_BRIDGE) {
965         is_bridge = true;
966     }
967 
968     /* in accordance with PAPR+ v2.7 13.6.3, Table 181 */
969     _FDT(fdt_setprop_cell(fdt, offset, "vendor-id",
970                           pci_default_read_config(dev, PCI_VENDOR_ID, 2)));
971     _FDT(fdt_setprop_cell(fdt, offset, "device-id",
972                           pci_default_read_config(dev, PCI_DEVICE_ID, 2)));
973     _FDT(fdt_setprop_cell(fdt, offset, "revision-id",
974                           pci_default_read_config(dev, PCI_REVISION_ID, 1)));
975     _FDT(fdt_setprop_cell(fdt, offset, "class-code",
976                           pci_default_read_config(dev, PCI_CLASS_PROG, 3)));
977     if (pci_default_read_config(dev, PCI_INTERRUPT_PIN, 1)) {
978         _FDT(fdt_setprop_cell(fdt, offset, "interrupts",
979                  pci_default_read_config(dev, PCI_INTERRUPT_PIN, 1)));
980     }
981 
982     if (!is_bridge) {
983         _FDT(fdt_setprop_cell(fdt, offset, "min-grant",
984             pci_default_read_config(dev, PCI_MIN_GNT, 1)));
985         _FDT(fdt_setprop_cell(fdt, offset, "max-latency",
986             pci_default_read_config(dev, PCI_MAX_LAT, 1)));
987     }
988 
989     if (pci_default_read_config(dev, PCI_SUBSYSTEM_ID, 2)) {
990         _FDT(fdt_setprop_cell(fdt, offset, "subsystem-id",
991                  pci_default_read_config(dev, PCI_SUBSYSTEM_ID, 2)));
992     }
993 
994     if (pci_default_read_config(dev, PCI_SUBSYSTEM_VENDOR_ID, 2)) {
995         _FDT(fdt_setprop_cell(fdt, offset, "subsystem-vendor-id",
996                  pci_default_read_config(dev, PCI_SUBSYSTEM_VENDOR_ID, 2)));
997     }
998 
999     _FDT(fdt_setprop_cell(fdt, offset, "cache-line-size",
1000         pci_default_read_config(dev, PCI_CACHE_LINE_SIZE, 1)));
1001 
1002     /* the following fdt cells are masked off the pci status register */
1003     pci_status = pci_default_read_config(dev, PCI_STATUS, 2);
1004     _FDT(fdt_setprop_cell(fdt, offset, "devsel-speed",
1005                           PCI_STATUS_DEVSEL_MASK & pci_status));
1006 
1007     if (pci_status & PCI_STATUS_FAST_BACK) {
1008         _FDT(fdt_setprop(fdt, offset, "fast-back-to-back", NULL, 0));
1009     }
1010     if (pci_status & PCI_STATUS_66MHZ) {
1011         _FDT(fdt_setprop(fdt, offset, "66mhz-capable", NULL, 0));
1012     }
1013     if (pci_status & PCI_STATUS_UDF) {
1014         _FDT(fdt_setprop(fdt, offset, "udf-supported", NULL, 0));
1015     }
1016 
1017     /* NOTE: this is normally generated by firmware via path/unit name,
1018      * but in our case we must set it manually since it does not get
1019      * processed by OF beforehand
1020      */
1021     _FDT(fdt_setprop_string(fdt, offset, "name", "pci"));
1022     buf = spapr_phb_get_loc_code(sphb, dev);
1023     if (!buf) {
1024         error_report("Failed setting the ibm,loc-code");
1025         return -1;
1026     }
1027 
1028     err = fdt_setprop_string(fdt, offset, "ibm,loc-code", buf);
1029     g_free(buf);
1030     if (err < 0) {
1031         return err;
1032     }
1033 
1034     if (drc_index) {
1035         _FDT(fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index));
1036     }
1037 
1038     _FDT(fdt_setprop_cell(fdt, offset, "#address-cells",
1039                           RESOURCE_CELLS_ADDRESS));
1040     _FDT(fdt_setprop_cell(fdt, offset, "#size-cells",
1041                           RESOURCE_CELLS_SIZE));
1042 
1043     max_msi = msi_nr_vectors_allocated(dev);
1044     if (max_msi) {
1045         _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi", max_msi));
1046     }
1047     max_msix = dev->msix_entries_nr;
1048     if (max_msix) {
1049         _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi-x", max_msix));
1050     }
1051 
1052     populate_resource_props(dev, &rp);
1053     _FDT(fdt_setprop(fdt, offset, "reg", (uint8_t *)rp.reg, rp.reg_len));
1054     _FDT(fdt_setprop(fdt, offset, "assigned-addresses",
1055                      (uint8_t *)rp.assigned, rp.assigned_len));
1056 
1057     return 0;
1058 }
1059 
1060 /* create OF node for pci device and required OF DT properties */
1061 static int spapr_create_pci_child_dt(sPAPRPHBState *phb, PCIDevice *dev,
1062                                      void *fdt, int node_offset)
1063 {
1064     int offset, ret;
1065     int slot = PCI_SLOT(dev->devfn);
1066     int func = PCI_FUNC(dev->devfn);
1067     char nodename[FDT_NAME_MAX];
1068 
1069     if (func != 0) {
1070         snprintf(nodename, FDT_NAME_MAX, "pci@%x,%x", slot, func);
1071     } else {
1072         snprintf(nodename, FDT_NAME_MAX, "pci@%x", slot);
1073     }
1074     offset = fdt_add_subnode(fdt, node_offset, nodename);
1075     ret = spapr_populate_pci_child_dt(dev, fdt, offset, phb);
1076 
1077     g_assert(!ret);
1078     if (ret) {
1079         return 0;
1080     }
1081     return offset;
1082 }
1083 
1084 static void spapr_phb_add_pci_device(sPAPRDRConnector *drc,
1085                                      sPAPRPHBState *phb,
1086                                      PCIDevice *pdev,
1087                                      Error **errp)
1088 {
1089     sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
1090     DeviceState *dev = DEVICE(pdev);
1091     void *fdt = NULL;
1092     int fdt_start_offset = 0, fdt_size;
1093 
1094     if (object_dynamic_cast(OBJECT(pdev), "vfio-pci")) {
1095         sPAPRTCETable *tcet = spapr_tce_find_by_liobn(phb->dma_liobn);
1096 
1097         spapr_tce_set_need_vfio(tcet, true);
1098     }
1099 
1100     if (dev->hotplugged) {
1101         fdt = create_device_tree(&fdt_size);
1102         fdt_start_offset = spapr_create_pci_child_dt(phb, pdev, fdt, 0);
1103         if (!fdt_start_offset) {
1104             error_setg(errp, "Failed to create pci child device tree node");
1105             goto out;
1106         }
1107     }
1108 
1109     drck->attach(drc, DEVICE(pdev),
1110                  fdt, fdt_start_offset, !dev->hotplugged, errp);
1111 out:
1112     if (*errp) {
1113         g_free(fdt);
1114     }
1115 }
1116 
1117 static void spapr_phb_remove_pci_device_cb(DeviceState *dev, void *opaque)
1118 {
1119     /* some version guests do not wait for completion of a device
1120      * cleanup (generally done asynchronously by the kernel) before
1121      * signaling to QEMU that the device is safe, but instead sleep
1122      * for some 'safe' period of time. unfortunately on a busy host
1123      * this sleep isn't guaranteed to be long enough, resulting in
1124      * bad things like IRQ lines being left asserted during final
1125      * device removal. to deal with this we call reset just prior
1126      * to finalizing the device, which will put the device back into
1127      * an 'idle' state, as the device cleanup code expects.
1128      */
1129     pci_device_reset(PCI_DEVICE(dev));
1130     object_unparent(OBJECT(dev));
1131 }
1132 
1133 static void spapr_phb_remove_pci_device(sPAPRDRConnector *drc,
1134                                         sPAPRPHBState *phb,
1135                                         PCIDevice *pdev,
1136                                         Error **errp)
1137 {
1138     sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
1139 
1140     drck->detach(drc, DEVICE(pdev), spapr_phb_remove_pci_device_cb, phb, errp);
1141 }
1142 
1143 static sPAPRDRConnector *spapr_phb_get_pci_drc(sPAPRPHBState *phb,
1144                                                PCIDevice *pdev)
1145 {
1146     uint32_t busnr = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(pdev))));
1147     return spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_PCI,
1148                                     (phb->index << 16) |
1149                                     (busnr << 8) |
1150                                     pdev->devfn);
1151 }
1152 
1153 static uint32_t spapr_phb_get_pci_drc_index(sPAPRPHBState *phb,
1154                                             PCIDevice *pdev)
1155 {
1156     sPAPRDRConnector *drc = spapr_phb_get_pci_drc(phb, pdev);
1157     sPAPRDRConnectorClass *drck;
1158 
1159     if (!drc) {
1160         return 0;
1161     }
1162 
1163     drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
1164     return drck->get_index(drc);
1165 }
1166 
1167 static void spapr_phb_hot_plug_child(HotplugHandler *plug_handler,
1168                                      DeviceState *plugged_dev, Error **errp)
1169 {
1170     sPAPRPHBState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1171     PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1172     sPAPRDRConnector *drc = spapr_phb_get_pci_drc(phb, pdev);
1173     Error *local_err = NULL;
1174 
1175     /* if DR is disabled we don't need to do anything in the case of
1176      * hotplug or coldplug callbacks
1177      */
1178     if (!phb->dr_enabled) {
1179         /* if this is a hotplug operation initiated by the user
1180          * we need to let them know it's not enabled
1181          */
1182         if (plugged_dev->hotplugged) {
1183             error_setg(errp, QERR_BUS_NO_HOTPLUG,
1184                        object_get_typename(OBJECT(phb)));
1185         }
1186         return;
1187     }
1188 
1189     g_assert(drc);
1190 
1191     spapr_phb_add_pci_device(drc, phb, pdev, &local_err);
1192     if (local_err) {
1193         error_propagate(errp, local_err);
1194         return;
1195     }
1196     if (plugged_dev->hotplugged) {
1197         spapr_hotplug_req_add_by_index(drc);
1198     }
1199 }
1200 
1201 static void spapr_phb_hot_unplug_child(HotplugHandler *plug_handler,
1202                                        DeviceState *plugged_dev, Error **errp)
1203 {
1204     sPAPRPHBState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1205     PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1206     sPAPRDRConnectorClass *drck;
1207     sPAPRDRConnector *drc = spapr_phb_get_pci_drc(phb, pdev);
1208     Error *local_err = NULL;
1209 
1210     if (!phb->dr_enabled) {
1211         error_setg(errp, QERR_BUS_NO_HOTPLUG,
1212                    object_get_typename(OBJECT(phb)));
1213         return;
1214     }
1215 
1216     g_assert(drc);
1217 
1218     drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
1219     if (!drck->release_pending(drc)) {
1220         spapr_phb_remove_pci_device(drc, phb, pdev, &local_err);
1221         if (local_err) {
1222             error_propagate(errp, local_err);
1223             return;
1224         }
1225         spapr_hotplug_req_remove_by_index(drc);
1226     }
1227 }
1228 
1229 static void spapr_phb_realize(DeviceState *dev, Error **errp)
1230 {
1231     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
1232     SysBusDevice *s = SYS_BUS_DEVICE(dev);
1233     sPAPRPHBState *sphb = SPAPR_PCI_HOST_BRIDGE(s);
1234     PCIHostState *phb = PCI_HOST_BRIDGE(s);
1235     sPAPRPHBClass *info = SPAPR_PCI_HOST_BRIDGE_GET_CLASS(s);
1236     char *namebuf;
1237     int i;
1238     PCIBus *bus;
1239     uint64_t msi_window_size = 4096;
1240 
1241     if (sphb->index != (uint32_t)-1) {
1242         hwaddr windows_base;
1243 
1244         if ((sphb->buid != (uint64_t)-1) || (sphb->dma_liobn != (uint32_t)-1)
1245             || (sphb->mem_win_addr != (hwaddr)-1)
1246             || (sphb->io_win_addr != (hwaddr)-1)) {
1247             error_setg(errp, "Either \"index\" or other parameters must"
1248                        " be specified for PAPR PHB, not both");
1249             return;
1250         }
1251 
1252         if (sphb->index > SPAPR_PCI_MAX_INDEX) {
1253             error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
1254                        SPAPR_PCI_MAX_INDEX);
1255             return;
1256         }
1257 
1258         sphb->buid = SPAPR_PCI_BASE_BUID + sphb->index;
1259         sphb->dma_liobn = SPAPR_PCI_LIOBN(sphb->index, 0);
1260 
1261         windows_base = SPAPR_PCI_WINDOW_BASE
1262             + sphb->index * SPAPR_PCI_WINDOW_SPACING;
1263         sphb->mem_win_addr = windows_base + SPAPR_PCI_MMIO_WIN_OFF;
1264         sphb->io_win_addr = windows_base + SPAPR_PCI_IO_WIN_OFF;
1265     }
1266 
1267     if (sphb->buid == (uint64_t)-1) {
1268         error_setg(errp, "BUID not specified for PHB");
1269         return;
1270     }
1271 
1272     if (sphb->dma_liobn == (uint32_t)-1) {
1273         error_setg(errp, "LIOBN not specified for PHB");
1274         return;
1275     }
1276 
1277     if (sphb->mem_win_addr == (hwaddr)-1) {
1278         error_setg(errp, "Memory window address not specified for PHB");
1279         return;
1280     }
1281 
1282     if (sphb->io_win_addr == (hwaddr)-1) {
1283         error_setg(errp, "IO window address not specified for PHB");
1284         return;
1285     }
1286 
1287     if (spapr_pci_find_phb(spapr, sphb->buid)) {
1288         error_setg(errp, "PCI host bridges must have unique BUIDs");
1289         return;
1290     }
1291 
1292     sphb->dtbusname = g_strdup_printf("pci@%" PRIx64, sphb->buid);
1293 
1294     namebuf = alloca(strlen(sphb->dtbusname) + 32);
1295 
1296     /* Initialize memory regions */
1297     sprintf(namebuf, "%s.mmio", sphb->dtbusname);
1298     memory_region_init(&sphb->memspace, OBJECT(sphb), namebuf, UINT64_MAX);
1299 
1300     sprintf(namebuf, "%s.mmio-alias", sphb->dtbusname);
1301     memory_region_init_alias(&sphb->memwindow, OBJECT(sphb),
1302                              namebuf, &sphb->memspace,
1303                              SPAPR_PCI_MEM_WIN_BUS_OFFSET, sphb->mem_win_size);
1304     memory_region_add_subregion(get_system_memory(), sphb->mem_win_addr,
1305                                 &sphb->memwindow);
1306 
1307     /* Initialize IO regions */
1308     sprintf(namebuf, "%s.io", sphb->dtbusname);
1309     memory_region_init(&sphb->iospace, OBJECT(sphb),
1310                        namebuf, SPAPR_PCI_IO_WIN_SIZE);
1311 
1312     sprintf(namebuf, "%s.io-alias", sphb->dtbusname);
1313     memory_region_init_alias(&sphb->iowindow, OBJECT(sphb), namebuf,
1314                              &sphb->iospace, 0, SPAPR_PCI_IO_WIN_SIZE);
1315     memory_region_add_subregion(get_system_memory(), sphb->io_win_addr,
1316                                 &sphb->iowindow);
1317 
1318     bus = pci_register_bus(dev, NULL,
1319                            pci_spapr_set_irq, pci_spapr_map_irq, sphb,
1320                            &sphb->memspace, &sphb->iospace,
1321                            PCI_DEVFN(0, 0), PCI_NUM_PINS, TYPE_PCI_BUS);
1322     phb->bus = bus;
1323     qbus_set_hotplug_handler(BUS(phb->bus), DEVICE(sphb), NULL);
1324 
1325     /*
1326      * Initialize PHB address space.
1327      * By default there will be at least one subregion for default
1328      * 32bit DMA window.
1329      * Later the guest might want to create another DMA window
1330      * which will become another memory subregion.
1331      */
1332     sprintf(namebuf, "%s.iommu-root", sphb->dtbusname);
1333 
1334     memory_region_init(&sphb->iommu_root, OBJECT(sphb),
1335                        namebuf, UINT64_MAX);
1336     address_space_init(&sphb->iommu_as, &sphb->iommu_root,
1337                        sphb->dtbusname);
1338 
1339     /*
1340      * As MSI/MSIX interrupts trigger by writing at MSI/MSIX vectors,
1341      * we need to allocate some memory to catch those writes coming
1342      * from msi_notify()/msix_notify().
1343      * As MSIMessage:addr is going to be the same and MSIMessage:data
1344      * is going to be a VIRQ number, 4 bytes of the MSI MR will only
1345      * be used.
1346      *
1347      * For KVM we want to ensure that this memory is a full page so that
1348      * our memory slot is of page size granularity.
1349      */
1350 #ifdef CONFIG_KVM
1351     if (kvm_enabled()) {
1352         msi_window_size = getpagesize();
1353     }
1354 #endif
1355 
1356     memory_region_init_io(&sphb->msiwindow, NULL, &spapr_msi_ops, spapr,
1357                           "msi", msi_window_size);
1358     memory_region_add_subregion(&sphb->iommu_root, SPAPR_PCI_MSI_WINDOW,
1359                                 &sphb->msiwindow);
1360 
1361     pci_setup_iommu(bus, spapr_pci_dma_iommu, sphb);
1362 
1363     pci_bus_set_route_irq_fn(bus, spapr_route_intx_pin_to_irq);
1364 
1365     QLIST_INSERT_HEAD(&spapr->phbs, sphb, list);
1366 
1367     /* Initialize the LSI table */
1368     for (i = 0; i < PCI_NUM_PINS; i++) {
1369         uint32_t irq;
1370 
1371         irq = xics_alloc_block(spapr->icp, 0, 1, true, false);
1372         if (!irq) {
1373             error_setg(errp, "spapr_allocate_lsi failed");
1374             return;
1375         }
1376 
1377         sphb->lsi_table[i].irq = irq;
1378     }
1379 
1380     /* allocate connectors for child PCI devices */
1381     if (sphb->dr_enabled) {
1382         for (i = 0; i < PCI_SLOT_MAX * 8; i++) {
1383             spapr_dr_connector_new(OBJECT(phb),
1384                                    SPAPR_DR_CONNECTOR_TYPE_PCI,
1385                                    (sphb->index << 16) | i);
1386         }
1387     }
1388 
1389     if (!info->finish_realize) {
1390         error_setg(errp, "finish_realize not defined");
1391         return;
1392     }
1393 
1394     info->finish_realize(sphb, errp);
1395 
1396     sphb->msi = g_hash_table_new_full(g_int_hash, g_int_equal, g_free, g_free);
1397 }
1398 
1399 static void spapr_phb_finish_realize(sPAPRPHBState *sphb, Error **errp)
1400 {
1401     sPAPRTCETable *tcet;
1402     uint32_t nb_table;
1403 
1404     nb_table = sphb->dma_win_size >> SPAPR_TCE_PAGE_SHIFT;
1405     tcet = spapr_tce_new_table(DEVICE(sphb), sphb->dma_liobn,
1406                                0, SPAPR_TCE_PAGE_SHIFT, nb_table, false);
1407     if (!tcet) {
1408         error_setg(errp, "Unable to create TCE table for %s",
1409                    sphb->dtbusname);
1410         return ;
1411     }
1412 
1413     /* Register default 32bit DMA window */
1414     memory_region_add_subregion(&sphb->iommu_root, sphb->dma_win_addr,
1415                                 spapr_tce_get_iommu(tcet));
1416 }
1417 
1418 static int spapr_phb_children_reset(Object *child, void *opaque)
1419 {
1420     DeviceState *dev = (DeviceState *) object_dynamic_cast(child, TYPE_DEVICE);
1421 
1422     if (dev) {
1423         device_reset(dev);
1424     }
1425 
1426     return 0;
1427 }
1428 
1429 static void spapr_phb_reset(DeviceState *qdev)
1430 {
1431     /* Reset the IOMMU state */
1432     object_child_foreach(OBJECT(qdev), spapr_phb_children_reset, NULL);
1433 }
1434 
1435 static Property spapr_phb_properties[] = {
1436     DEFINE_PROP_UINT32("index", sPAPRPHBState, index, -1),
1437     DEFINE_PROP_UINT64("buid", sPAPRPHBState, buid, -1),
1438     DEFINE_PROP_UINT32("liobn", sPAPRPHBState, dma_liobn, -1),
1439     DEFINE_PROP_UINT64("mem_win_addr", sPAPRPHBState, mem_win_addr, -1),
1440     DEFINE_PROP_UINT64("mem_win_size", sPAPRPHBState, mem_win_size,
1441                        SPAPR_PCI_MMIO_WIN_SIZE),
1442     DEFINE_PROP_UINT64("io_win_addr", sPAPRPHBState, io_win_addr, -1),
1443     DEFINE_PROP_UINT64("io_win_size", sPAPRPHBState, io_win_size,
1444                        SPAPR_PCI_IO_WIN_SIZE),
1445     DEFINE_PROP_BOOL("dynamic-reconfiguration", sPAPRPHBState, dr_enabled,
1446                      true),
1447     /* Default DMA window is 0..1GB */
1448     DEFINE_PROP_UINT64("dma_win_addr", sPAPRPHBState, dma_win_addr, 0),
1449     DEFINE_PROP_UINT64("dma_win_size", sPAPRPHBState, dma_win_size, 0x40000000),
1450     DEFINE_PROP_END_OF_LIST(),
1451 };
1452 
1453 static const VMStateDescription vmstate_spapr_pci_lsi = {
1454     .name = "spapr_pci/lsi",
1455     .version_id = 1,
1456     .minimum_version_id = 1,
1457     .fields = (VMStateField[]) {
1458         VMSTATE_UINT32_EQUAL(irq, struct spapr_pci_lsi),
1459 
1460         VMSTATE_END_OF_LIST()
1461     },
1462 };
1463 
1464 static const VMStateDescription vmstate_spapr_pci_msi = {
1465     .name = "spapr_pci/msi",
1466     .version_id = 1,
1467     .minimum_version_id = 1,
1468     .fields = (VMStateField []) {
1469         VMSTATE_UINT32(key, spapr_pci_msi_mig),
1470         VMSTATE_UINT32(value.first_irq, spapr_pci_msi_mig),
1471         VMSTATE_UINT32(value.num, spapr_pci_msi_mig),
1472         VMSTATE_END_OF_LIST()
1473     },
1474 };
1475 
1476 static void spapr_pci_pre_save(void *opaque)
1477 {
1478     sPAPRPHBState *sphb = opaque;
1479     GHashTableIter iter;
1480     gpointer key, value;
1481     int i;
1482 
1483     g_free(sphb->msi_devs);
1484     sphb->msi_devs = NULL;
1485     sphb->msi_devs_num = g_hash_table_size(sphb->msi);
1486     if (!sphb->msi_devs_num) {
1487         return;
1488     }
1489     sphb->msi_devs = g_malloc(sphb->msi_devs_num * sizeof(spapr_pci_msi_mig));
1490 
1491     g_hash_table_iter_init(&iter, sphb->msi);
1492     for (i = 0; g_hash_table_iter_next(&iter, &key, &value); ++i) {
1493         sphb->msi_devs[i].key = *(uint32_t *) key;
1494         sphb->msi_devs[i].value = *(spapr_pci_msi *) value;
1495     }
1496 }
1497 
1498 static int spapr_pci_post_load(void *opaque, int version_id)
1499 {
1500     sPAPRPHBState *sphb = opaque;
1501     gpointer key, value;
1502     int i;
1503 
1504     for (i = 0; i < sphb->msi_devs_num; ++i) {
1505         key = g_memdup(&sphb->msi_devs[i].key,
1506                        sizeof(sphb->msi_devs[i].key));
1507         value = g_memdup(&sphb->msi_devs[i].value,
1508                          sizeof(sphb->msi_devs[i].value));
1509         g_hash_table_insert(sphb->msi, key, value);
1510     }
1511     g_free(sphb->msi_devs);
1512     sphb->msi_devs = NULL;
1513     sphb->msi_devs_num = 0;
1514 
1515     return 0;
1516 }
1517 
1518 static const VMStateDescription vmstate_spapr_pci = {
1519     .name = "spapr_pci",
1520     .version_id = 2,
1521     .minimum_version_id = 2,
1522     .pre_save = spapr_pci_pre_save,
1523     .post_load = spapr_pci_post_load,
1524     .fields = (VMStateField[]) {
1525         VMSTATE_UINT64_EQUAL(buid, sPAPRPHBState),
1526         VMSTATE_UINT32_EQUAL(dma_liobn, sPAPRPHBState),
1527         VMSTATE_UINT64_EQUAL(mem_win_addr, sPAPRPHBState),
1528         VMSTATE_UINT64_EQUAL(mem_win_size, sPAPRPHBState),
1529         VMSTATE_UINT64_EQUAL(io_win_addr, sPAPRPHBState),
1530         VMSTATE_UINT64_EQUAL(io_win_size, sPAPRPHBState),
1531         VMSTATE_STRUCT_ARRAY(lsi_table, sPAPRPHBState, PCI_NUM_PINS, 0,
1532                              vmstate_spapr_pci_lsi, struct spapr_pci_lsi),
1533         VMSTATE_INT32(msi_devs_num, sPAPRPHBState),
1534         VMSTATE_STRUCT_VARRAY_ALLOC(msi_devs, sPAPRPHBState, msi_devs_num, 0,
1535                                     vmstate_spapr_pci_msi, spapr_pci_msi_mig),
1536         VMSTATE_END_OF_LIST()
1537     },
1538 };
1539 
1540 static const char *spapr_phb_root_bus_path(PCIHostState *host_bridge,
1541                                            PCIBus *rootbus)
1542 {
1543     sPAPRPHBState *sphb = SPAPR_PCI_HOST_BRIDGE(host_bridge);
1544 
1545     return sphb->dtbusname;
1546 }
1547 
1548 static void spapr_phb_class_init(ObjectClass *klass, void *data)
1549 {
1550     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(klass);
1551     DeviceClass *dc = DEVICE_CLASS(klass);
1552     sPAPRPHBClass *spc = SPAPR_PCI_HOST_BRIDGE_CLASS(klass);
1553     HotplugHandlerClass *hp = HOTPLUG_HANDLER_CLASS(klass);
1554 
1555     hc->root_bus_path = spapr_phb_root_bus_path;
1556     dc->realize = spapr_phb_realize;
1557     dc->props = spapr_phb_properties;
1558     dc->reset = spapr_phb_reset;
1559     dc->vmsd = &vmstate_spapr_pci;
1560     set_bit(DEVICE_CATEGORY_BRIDGE, dc->categories);
1561     dc->cannot_instantiate_with_device_add_yet = false;
1562     spc->finish_realize = spapr_phb_finish_realize;
1563     hp->plug = spapr_phb_hot_plug_child;
1564     hp->unplug = spapr_phb_hot_unplug_child;
1565 }
1566 
1567 static const TypeInfo spapr_phb_info = {
1568     .name          = TYPE_SPAPR_PCI_HOST_BRIDGE,
1569     .parent        = TYPE_PCI_HOST_BRIDGE,
1570     .instance_size = sizeof(sPAPRPHBState),
1571     .class_init    = spapr_phb_class_init,
1572     .class_size    = sizeof(sPAPRPHBClass),
1573     .interfaces    = (InterfaceInfo[]) {
1574         { TYPE_HOTPLUG_HANDLER },
1575         { }
1576     }
1577 };
1578 
1579 PCIHostState *spapr_create_phb(sPAPRMachineState *spapr, int index)
1580 {
1581     DeviceState *dev;
1582 
1583     dev = qdev_create(NULL, TYPE_SPAPR_PCI_HOST_BRIDGE);
1584     qdev_prop_set_uint32(dev, "index", index);
1585     qdev_init_nofail(dev);
1586 
1587     return PCI_HOST_BRIDGE(dev);
1588 }
1589 
1590 typedef struct sPAPRFDT {
1591     void *fdt;
1592     int node_off;
1593     sPAPRPHBState *sphb;
1594 } sPAPRFDT;
1595 
1596 static void spapr_populate_pci_devices_dt(PCIBus *bus, PCIDevice *pdev,
1597                                           void *opaque)
1598 {
1599     PCIBus *sec_bus;
1600     sPAPRFDT *p = opaque;
1601     int offset;
1602     sPAPRFDT s_fdt;
1603 
1604     offset = spapr_create_pci_child_dt(p->sphb, pdev, p->fdt, p->node_off);
1605     if (!offset) {
1606         error_report("Failed to create pci child device tree node");
1607         return;
1608     }
1609 
1610     if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) !=
1611          PCI_HEADER_TYPE_BRIDGE)) {
1612         return;
1613     }
1614 
1615     sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev));
1616     if (!sec_bus) {
1617         return;
1618     }
1619 
1620     s_fdt.fdt = p->fdt;
1621     s_fdt.node_off = offset;
1622     s_fdt.sphb = p->sphb;
1623     pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
1624                         spapr_populate_pci_devices_dt,
1625                         &s_fdt);
1626 }
1627 
1628 static void spapr_phb_pci_enumerate_bridge(PCIBus *bus, PCIDevice *pdev,
1629                                            void *opaque)
1630 {
1631     unsigned int *bus_no = opaque;
1632     unsigned int primary = *bus_no;
1633     unsigned int subordinate = 0xff;
1634     PCIBus *sec_bus = NULL;
1635 
1636     if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) !=
1637          PCI_HEADER_TYPE_BRIDGE)) {
1638         return;
1639     }
1640 
1641     (*bus_no)++;
1642     pci_default_write_config(pdev, PCI_PRIMARY_BUS, primary, 1);
1643     pci_default_write_config(pdev, PCI_SECONDARY_BUS, *bus_no, 1);
1644     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
1645 
1646     sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev));
1647     if (!sec_bus) {
1648         return;
1649     }
1650 
1651     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, subordinate, 1);
1652     pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
1653                         spapr_phb_pci_enumerate_bridge, bus_no);
1654     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
1655 }
1656 
1657 static void spapr_phb_pci_enumerate(sPAPRPHBState *phb)
1658 {
1659     PCIBus *bus = PCI_HOST_BRIDGE(phb)->bus;
1660     unsigned int bus_no = 0;
1661 
1662     pci_for_each_device(bus, pci_bus_num(bus),
1663                         spapr_phb_pci_enumerate_bridge,
1664                         &bus_no);
1665 
1666 }
1667 
1668 int spapr_populate_pci_dt(sPAPRPHBState *phb,
1669                           uint32_t xics_phandle,
1670                           void *fdt)
1671 {
1672     int bus_off, i, j, ret;
1673     char nodename[FDT_NAME_MAX];
1674     uint32_t bus_range[] = { cpu_to_be32(0), cpu_to_be32(0xff) };
1675     const uint64_t mmiosize = memory_region_size(&phb->memwindow);
1676     const uint64_t w32max = (1ULL << 32) - SPAPR_PCI_MEM_WIN_BUS_OFFSET;
1677     const uint64_t w32size = MIN(w32max, mmiosize);
1678     const uint64_t w64size = (mmiosize > w32size) ? (mmiosize - w32size) : 0;
1679     struct {
1680         uint32_t hi;
1681         uint64_t child;
1682         uint64_t parent;
1683         uint64_t size;
1684     } QEMU_PACKED ranges[] = {
1685         {
1686             cpu_to_be32(b_ss(1)), cpu_to_be64(0),
1687             cpu_to_be64(phb->io_win_addr),
1688             cpu_to_be64(memory_region_size(&phb->iospace)),
1689         },
1690         {
1691             cpu_to_be32(b_ss(2)), cpu_to_be64(SPAPR_PCI_MEM_WIN_BUS_OFFSET),
1692             cpu_to_be64(phb->mem_win_addr),
1693             cpu_to_be64(w32size),
1694         },
1695         {
1696             cpu_to_be32(b_ss(3)), cpu_to_be64(1ULL << 32),
1697             cpu_to_be64(phb->mem_win_addr + w32size),
1698             cpu_to_be64(w64size)
1699         },
1700     };
1701     const unsigned sizeof_ranges = (w64size ? 3 : 2) * sizeof(ranges[0]);
1702     uint64_t bus_reg[] = { cpu_to_be64(phb->buid), 0 };
1703     uint32_t interrupt_map_mask[] = {
1704         cpu_to_be32(b_ddddd(-1)|b_fff(0)), 0x0, 0x0, cpu_to_be32(-1)};
1705     uint32_t interrupt_map[PCI_SLOT_MAX * PCI_NUM_PINS][7];
1706     sPAPRTCETable *tcet;
1707     PCIBus *bus = PCI_HOST_BRIDGE(phb)->bus;
1708     sPAPRFDT s_fdt;
1709 
1710     /* Start populating the FDT */
1711     snprintf(nodename, FDT_NAME_MAX, "pci@%" PRIx64, phb->buid);
1712     bus_off = fdt_add_subnode(fdt, 0, nodename);
1713     if (bus_off < 0) {
1714         return bus_off;
1715     }
1716 
1717     /* Write PHB properties */
1718     _FDT(fdt_setprop_string(fdt, bus_off, "device_type", "pci"));
1719     _FDT(fdt_setprop_string(fdt, bus_off, "compatible", "IBM,Logical_PHB"));
1720     _FDT(fdt_setprop_cell(fdt, bus_off, "#address-cells", 0x3));
1721     _FDT(fdt_setprop_cell(fdt, bus_off, "#size-cells", 0x2));
1722     _FDT(fdt_setprop_cell(fdt, bus_off, "#interrupt-cells", 0x1));
1723     _FDT(fdt_setprop(fdt, bus_off, "used-by-rtas", NULL, 0));
1724     _FDT(fdt_setprop(fdt, bus_off, "bus-range", &bus_range, sizeof(bus_range)));
1725     _FDT(fdt_setprop(fdt, bus_off, "ranges", &ranges, sizeof_ranges));
1726     _FDT(fdt_setprop(fdt, bus_off, "reg", &bus_reg, sizeof(bus_reg)));
1727     _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pci-config-space-type", 0x1));
1728     _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pe-total-#msi", XICS_IRQS));
1729 
1730     /* Build the interrupt-map, this must matches what is done
1731      * in pci_spapr_map_irq
1732      */
1733     _FDT(fdt_setprop(fdt, bus_off, "interrupt-map-mask",
1734                      &interrupt_map_mask, sizeof(interrupt_map_mask)));
1735     for (i = 0; i < PCI_SLOT_MAX; i++) {
1736         for (j = 0; j < PCI_NUM_PINS; j++) {
1737             uint32_t *irqmap = interrupt_map[i*PCI_NUM_PINS + j];
1738             int lsi_num = pci_spapr_swizzle(i, j);
1739 
1740             irqmap[0] = cpu_to_be32(b_ddddd(i)|b_fff(0));
1741             irqmap[1] = 0;
1742             irqmap[2] = 0;
1743             irqmap[3] = cpu_to_be32(j+1);
1744             irqmap[4] = cpu_to_be32(xics_phandle);
1745             irqmap[5] = cpu_to_be32(phb->lsi_table[lsi_num].irq);
1746             irqmap[6] = cpu_to_be32(0x8);
1747         }
1748     }
1749     /* Write interrupt map */
1750     _FDT(fdt_setprop(fdt, bus_off, "interrupt-map", &interrupt_map,
1751                      sizeof(interrupt_map)));
1752 
1753     tcet = spapr_tce_find_by_liobn(SPAPR_PCI_LIOBN(phb->index, 0));
1754     spapr_dma_dt(fdt, bus_off, "ibm,dma-window",
1755                  tcet->liobn, tcet->bus_offset,
1756                  tcet->nb_table << tcet->page_shift);
1757 
1758     /* Walk the bridges and program the bus numbers*/
1759     spapr_phb_pci_enumerate(phb);
1760     _FDT(fdt_setprop_cell(fdt, bus_off, "qemu,phb-enumerated", 0x1));
1761 
1762     /* Populate tree nodes with PCI devices attached */
1763     s_fdt.fdt = fdt;
1764     s_fdt.node_off = bus_off;
1765     s_fdt.sphb = phb;
1766     pci_for_each_device(bus, pci_bus_num(bus),
1767                         spapr_populate_pci_devices_dt,
1768                         &s_fdt);
1769 
1770     ret = spapr_drc_populate_dt(fdt, bus_off, OBJECT(phb),
1771                                 SPAPR_DR_CONNECTOR_TYPE_PCI);
1772     if (ret) {
1773         return ret;
1774     }
1775 
1776     return 0;
1777 }
1778 
1779 void spapr_pci_rtas_init(void)
1780 {
1781     spapr_rtas_register(RTAS_READ_PCI_CONFIG, "read-pci-config",
1782                         rtas_read_pci_config);
1783     spapr_rtas_register(RTAS_WRITE_PCI_CONFIG, "write-pci-config",
1784                         rtas_write_pci_config);
1785     spapr_rtas_register(RTAS_IBM_READ_PCI_CONFIG, "ibm,read-pci-config",
1786                         rtas_ibm_read_pci_config);
1787     spapr_rtas_register(RTAS_IBM_WRITE_PCI_CONFIG, "ibm,write-pci-config",
1788                         rtas_ibm_write_pci_config);
1789     if (msi_supported) {
1790         spapr_rtas_register(RTAS_IBM_QUERY_INTERRUPT_SOURCE_NUMBER,
1791                             "ibm,query-interrupt-source-number",
1792                             rtas_ibm_query_interrupt_source_number);
1793         spapr_rtas_register(RTAS_IBM_CHANGE_MSI, "ibm,change-msi",
1794                             rtas_ibm_change_msi);
1795     }
1796 
1797     spapr_rtas_register(RTAS_IBM_SET_EEH_OPTION,
1798                         "ibm,set-eeh-option",
1799                         rtas_ibm_set_eeh_option);
1800     spapr_rtas_register(RTAS_IBM_GET_CONFIG_ADDR_INFO2,
1801                         "ibm,get-config-addr-info2",
1802                         rtas_ibm_get_config_addr_info2);
1803     spapr_rtas_register(RTAS_IBM_READ_SLOT_RESET_STATE2,
1804                         "ibm,read-slot-reset-state2",
1805                         rtas_ibm_read_slot_reset_state2);
1806     spapr_rtas_register(RTAS_IBM_SET_SLOT_RESET,
1807                         "ibm,set-slot-reset",
1808                         rtas_ibm_set_slot_reset);
1809     spapr_rtas_register(RTAS_IBM_CONFIGURE_PE,
1810                         "ibm,configure-pe",
1811                         rtas_ibm_configure_pe);
1812     spapr_rtas_register(RTAS_IBM_SLOT_ERROR_DETAIL,
1813                         "ibm,slot-error-detail",
1814                         rtas_ibm_slot_error_detail);
1815 }
1816 
1817 static void spapr_pci_register_types(void)
1818 {
1819     type_register_static(&spapr_phb_info);
1820 }
1821 
1822 type_init(spapr_pci_register_types)
1823 
1824 static int spapr_switch_one_vga(DeviceState *dev, void *opaque)
1825 {
1826     bool be = *(bool *)opaque;
1827 
1828     if (object_dynamic_cast(OBJECT(dev), "VGA")
1829         || object_dynamic_cast(OBJECT(dev), "secondary-vga")) {
1830         object_property_set_bool(OBJECT(dev), be, "big-endian-framebuffer",
1831                                  &error_abort);
1832     }
1833     return 0;
1834 }
1835 
1836 void spapr_pci_switch_vga(bool big_endian)
1837 {
1838     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
1839     sPAPRPHBState *sphb;
1840 
1841     /*
1842      * For backward compatibility with existing guests, we switch
1843      * the endianness of the VGA controller when changing the guest
1844      * interrupt mode
1845      */
1846     QLIST_FOREACH(sphb, &spapr->phbs, list) {
1847         BusState *bus = &PCI_HOST_BRIDGE(sphb)->bus->qbus;
1848         qbus_walk_children(bus, spapr_switch_one_vga, NULL, NULL, NULL,
1849                            &big_endian);
1850     }
1851 }
1852