xref: /qemu/hw/ppc/spapr_pci.c (revision 7a36ae7a9f4f136d40fe1da4aab66b364d4aa56d)
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 "hw/hw.h"
26 #include "hw/sysbus.h"
27 #include "hw/pci/pci.h"
28 #include "hw/pci/msi.h"
29 #include "hw/pci/msix.h"
30 #include "hw/pci/pci_host.h"
31 #include "hw/ppc/spapr.h"
32 #include "hw/pci-host/spapr.h"
33 #include "exec/address-spaces.h"
34 #include <libfdt.h>
35 #include "trace.h"
36 #include "qemu/error-report.h"
37 #include "qapi/qmp/qerror.h"
38 
39 #include "hw/pci/pci_bridge.h"
40 #include "hw/pci/pci_bus.h"
41 #include "hw/ppc/spapr_drc.h"
42 #include "sysemu/device_tree.h"
43 
44 /* Copied from the kernel arch/powerpc/platforms/pseries/msi.c */
45 #define RTAS_QUERY_FN           0
46 #define RTAS_CHANGE_FN          1
47 #define RTAS_RESET_FN           2
48 #define RTAS_CHANGE_MSI_FN      3
49 #define RTAS_CHANGE_MSIX_FN     4
50 
51 /* Interrupt types to return on RTAS_CHANGE_* */
52 #define RTAS_TYPE_MSI           1
53 #define RTAS_TYPE_MSIX          2
54 
55 #define FDT_NAME_MAX          128
56 
57 #define _FDT(exp) \
58     do { \
59         int ret = (exp);                                           \
60         if (ret < 0) {                                             \
61             return ret;                                            \
62         }                                                          \
63     } while (0)
64 
65 sPAPRPHBState *spapr_pci_find_phb(sPAPRMachineState *spapr, uint64_t buid)
66 {
67     sPAPRPHBState *sphb;
68 
69     QLIST_FOREACH(sphb, &spapr->phbs, list) {
70         if (sphb->buid != buid) {
71             continue;
72         }
73         return sphb;
74     }
75 
76     return NULL;
77 }
78 
79 PCIDevice *spapr_pci_find_dev(sPAPRMachineState *spapr, uint64_t buid,
80                               uint32_t config_addr)
81 {
82     sPAPRPHBState *sphb = spapr_pci_find_phb(spapr, buid);
83     PCIHostState *phb = PCI_HOST_BRIDGE(sphb);
84     int bus_num = (config_addr >> 16) & 0xFF;
85     int devfn = (config_addr >> 8) & 0xFF;
86 
87     if (!phb) {
88         return NULL;
89     }
90 
91     return pci_find_device(phb->bus, bus_num, devfn);
92 }
93 
94 static uint32_t rtas_pci_cfgaddr(uint32_t arg)
95 {
96     /* This handles the encoding of extended config space addresses */
97     return ((arg >> 20) & 0xf00) | (arg & 0xff);
98 }
99 
100 static void finish_read_pci_config(sPAPRMachineState *spapr, uint64_t buid,
101                                    uint32_t addr, uint32_t size,
102                                    target_ulong rets)
103 {
104     PCIDevice *pci_dev;
105     uint32_t val;
106 
107     if ((size != 1) && (size != 2) && (size != 4)) {
108         /* access must be 1, 2 or 4 bytes */
109         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
110         return;
111     }
112 
113     pci_dev = spapr_pci_find_dev(spapr, buid, addr);
114     addr = rtas_pci_cfgaddr(addr);
115 
116     if (!pci_dev || (addr % size) || (addr >= pci_config_size(pci_dev))) {
117         /* Access must be to a valid device, within bounds and
118          * naturally aligned */
119         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
120         return;
121     }
122 
123     val = pci_host_config_read_common(pci_dev, addr,
124                                       pci_config_size(pci_dev), size);
125 
126     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
127     rtas_st(rets, 1, val);
128 }
129 
130 static void rtas_ibm_read_pci_config(PowerPCCPU *cpu, sPAPRMachineState *spapr,
131                                      uint32_t token, uint32_t nargs,
132                                      target_ulong args,
133                                      uint32_t nret, target_ulong rets)
134 {
135     uint64_t buid;
136     uint32_t size, addr;
137 
138     if ((nargs != 4) || (nret != 2)) {
139         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
140         return;
141     }
142 
143     buid = rtas_ldq(args, 1);
144     size = rtas_ld(args, 3);
145     addr = rtas_ld(args, 0);
146 
147     finish_read_pci_config(spapr, buid, addr, size, rets);
148 }
149 
150 static void rtas_read_pci_config(PowerPCCPU *cpu, sPAPRMachineState *spapr,
151                                  uint32_t token, uint32_t nargs,
152                                  target_ulong args,
153                                  uint32_t nret, target_ulong rets)
154 {
155     uint32_t size, addr;
156 
157     if ((nargs != 2) || (nret != 2)) {
158         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
159         return;
160     }
161 
162     size = rtas_ld(args, 1);
163     addr = rtas_ld(args, 0);
164 
165     finish_read_pci_config(spapr, 0, addr, size, rets);
166 }
167 
168 static void finish_write_pci_config(sPAPRMachineState *spapr, uint64_t buid,
169                                     uint32_t addr, uint32_t size,
170                                     uint32_t val, target_ulong rets)
171 {
172     PCIDevice *pci_dev;
173 
174     if ((size != 1) && (size != 2) && (size != 4)) {
175         /* access must be 1, 2 or 4 bytes */
176         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
177         return;
178     }
179 
180     pci_dev = spapr_pci_find_dev(spapr, buid, addr);
181     addr = rtas_pci_cfgaddr(addr);
182 
183     if (!pci_dev || (addr % size) || (addr >= pci_config_size(pci_dev))) {
184         /* Access must be to a valid device, within bounds and
185          * naturally aligned */
186         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
187         return;
188     }
189 
190     pci_host_config_write_common(pci_dev, addr, pci_config_size(pci_dev),
191                                  val, size);
192 
193     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
194 }
195 
196 static void rtas_ibm_write_pci_config(PowerPCCPU *cpu, sPAPRMachineState *spapr,
197                                       uint32_t token, uint32_t nargs,
198                                       target_ulong args,
199                                       uint32_t nret, target_ulong rets)
200 {
201     uint64_t buid;
202     uint32_t val, size, addr;
203 
204     if ((nargs != 5) || (nret != 1)) {
205         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
206         return;
207     }
208 
209     buid = rtas_ldq(args, 1);
210     val = rtas_ld(args, 4);
211     size = rtas_ld(args, 3);
212     addr = rtas_ld(args, 0);
213 
214     finish_write_pci_config(spapr, buid, addr, size, val, rets);
215 }
216 
217 static void rtas_write_pci_config(PowerPCCPU *cpu, sPAPRMachineState *spapr,
218                                   uint32_t token, uint32_t nargs,
219                                   target_ulong args,
220                                   uint32_t nret, target_ulong rets)
221 {
222     uint32_t val, size, addr;
223 
224     if ((nargs != 3) || (nret != 1)) {
225         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
226         return;
227     }
228 
229 
230     val = rtas_ld(args, 2);
231     size = rtas_ld(args, 1);
232     addr = rtas_ld(args, 0);
233 
234     finish_write_pci_config(spapr, 0, addr, size, val, rets);
235 }
236 
237 /*
238  * Set MSI/MSIX message data.
239  * This is required for msi_notify()/msix_notify() which
240  * will write at the addresses via spapr_msi_write().
241  *
242  * If hwaddr == 0, all entries will have .data == first_irq i.e.
243  * table will be reset.
244  */
245 static void spapr_msi_setmsg(PCIDevice *pdev, hwaddr addr, bool msix,
246                              unsigned first_irq, unsigned req_num)
247 {
248     unsigned i;
249     MSIMessage msg = { .address = addr, .data = first_irq };
250 
251     if (!msix) {
252         msi_set_message(pdev, msg);
253         trace_spapr_pci_msi_setup(pdev->name, 0, msg.address);
254         return;
255     }
256 
257     for (i = 0; i < req_num; ++i) {
258         msix_set_message(pdev, i, msg);
259         trace_spapr_pci_msi_setup(pdev->name, i, msg.address);
260         if (addr) {
261             ++msg.data;
262         }
263     }
264 }
265 
266 static void rtas_ibm_change_msi(PowerPCCPU *cpu, sPAPRMachineState *spapr,
267                                 uint32_t token, uint32_t nargs,
268                                 target_ulong args, uint32_t nret,
269                                 target_ulong rets)
270 {
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, num = 0;
278     sPAPRPHBState *phb = NULL;
279     PCIDevice *pdev = NULL;
280     spapr_pci_msi *msi;
281     int *config_addr_key;
282 
283     switch (func) {
284     case RTAS_CHANGE_MSI_FN:
285     case RTAS_CHANGE_FN:
286         ret_intr_type = RTAS_TYPE_MSI;
287         break;
288     case RTAS_CHANGE_MSIX_FN:
289         ret_intr_type = RTAS_TYPE_MSIX;
290         break;
291     default:
292         error_report("rtas_ibm_change_msi(%u) is not implemented", func);
293         rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
294         return;
295     }
296 
297     /* Fins sPAPRPHBState */
298     phb = spapr_pci_find_phb(spapr, buid);
299     if (phb) {
300         pdev = spapr_pci_find_dev(spapr, buid, config_addr);
301     }
302     if (!phb || !pdev) {
303         rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
304         return;
305     }
306 
307     /* Releasing MSIs */
308     if (!req_num) {
309         msi = (spapr_pci_msi *) g_hash_table_lookup(phb->msi, &config_addr);
310         if (!msi) {
311             trace_spapr_pci_msi("Releasing wrong config", config_addr);
312             rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
313             return;
314         }
315 
316         xics_free(spapr->icp, msi->first_irq, msi->num);
317         if (msi_present(pdev)) {
318             spapr_msi_setmsg(pdev, 0, false, 0, num);
319         }
320         if (msix_present(pdev)) {
321             spapr_msi_setmsg(pdev, 0, true, 0, num);
322         }
323         g_hash_table_remove(phb->msi, &config_addr);
324 
325         trace_spapr_pci_msi("Released MSIs", config_addr);
326         rtas_st(rets, 0, RTAS_OUT_SUCCESS);
327         rtas_st(rets, 1, 0);
328         return;
329     }
330 
331     /* Enabling MSI */
332 
333     /* Check if the device supports as many IRQs as requested */
334     if (ret_intr_type == RTAS_TYPE_MSI) {
335         max_irqs = msi_nr_vectors_allocated(pdev);
336     } else if (ret_intr_type == RTAS_TYPE_MSIX) {
337         max_irqs = pdev->msix_entries_nr;
338     }
339     if (!max_irqs) {
340         error_report("Requested interrupt type %d is not enabled for device %x",
341                      ret_intr_type, config_addr);
342         rtas_st(rets, 0, -1); /* Hardware error */
343         return;
344     }
345     /* Correct the number if the guest asked for too many */
346     if (req_num > max_irqs) {
347         trace_spapr_pci_msi_retry(config_addr, req_num, max_irqs);
348         req_num = max_irqs;
349         irq = 0; /* to avoid misleading trace */
350         goto out;
351     }
352 
353     /* Allocate MSIs */
354     irq = xics_alloc_block(spapr->icp, 0, req_num, false,
355                            ret_intr_type == RTAS_TYPE_MSI);
356     if (!irq) {
357         error_report("Cannot allocate MSIs for device %x", config_addr);
358         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
359         return;
360     }
361 
362     /* Setup MSI/MSIX vectors in the device (via cfgspace or MSIX BAR) */
363     spapr_msi_setmsg(pdev, SPAPR_PCI_MSI_WINDOW, ret_intr_type == RTAS_TYPE_MSIX,
364                      irq, req_num);
365 
366     /* Add MSI device to cache */
367     msi = g_new(spapr_pci_msi, 1);
368     msi->first_irq = irq;
369     msi->num = req_num;
370     config_addr_key = g_new(int, 1);
371     *config_addr_key = config_addr;
372     g_hash_table_insert(phb->msi, config_addr_key, msi);
373 
374 out:
375     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
376     rtas_st(rets, 1, req_num);
377     rtas_st(rets, 2, ++seq_num);
378     if (nret > 3) {
379         rtas_st(rets, 3, ret_intr_type);
380     }
381 
382     trace_spapr_pci_rtas_ibm_change_msi(config_addr, func, req_num, irq);
383 }
384 
385 static void rtas_ibm_query_interrupt_source_number(PowerPCCPU *cpu,
386                                                    sPAPRMachineState *spapr,
387                                                    uint32_t token,
388                                                    uint32_t nargs,
389                                                    target_ulong args,
390                                                    uint32_t nret,
391                                                    target_ulong rets)
392 {
393     uint32_t config_addr = rtas_ld(args, 0);
394     uint64_t buid = rtas_ldq(args, 1);
395     unsigned int intr_src_num = -1, ioa_intr_num = rtas_ld(args, 3);
396     sPAPRPHBState *phb = NULL;
397     PCIDevice *pdev = NULL;
398     spapr_pci_msi *msi;
399 
400     /* Find sPAPRPHBState */
401     phb = spapr_pci_find_phb(spapr, buid);
402     if (phb) {
403         pdev = spapr_pci_find_dev(spapr, buid, config_addr);
404     }
405     if (!phb || !pdev) {
406         rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
407         return;
408     }
409 
410     /* Find device descriptor and start IRQ */
411     msi = (spapr_pci_msi *) g_hash_table_lookup(phb->msi, &config_addr);
412     if (!msi || !msi->first_irq || !msi->num || (ioa_intr_num >= msi->num)) {
413         trace_spapr_pci_msi("Failed to return vector", config_addr);
414         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
415         return;
416     }
417     intr_src_num = msi->first_irq + ioa_intr_num;
418     trace_spapr_pci_rtas_ibm_query_interrupt_source_number(ioa_intr_num,
419                                                            intr_src_num);
420 
421     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
422     rtas_st(rets, 1, intr_src_num);
423     rtas_st(rets, 2, 1);/* 0 == level; 1 == edge */
424 }
425 
426 static void rtas_ibm_set_eeh_option(PowerPCCPU *cpu,
427                                     sPAPRMachineState *spapr,
428                                     uint32_t token, uint32_t nargs,
429                                     target_ulong args, uint32_t nret,
430                                     target_ulong rets)
431 {
432     sPAPRPHBState *sphb;
433     sPAPRPHBClass *spc;
434     PCIDevice *pdev;
435     uint32_t addr, option;
436     uint64_t buid;
437     int ret;
438 
439     if ((nargs != 4) || (nret != 1)) {
440         goto param_error_exit;
441     }
442 
443     buid = rtas_ldq(args, 1);
444     addr = rtas_ld(args, 0);
445     option = rtas_ld(args, 3);
446 
447     sphb = spapr_pci_find_phb(spapr, buid);
448     if (!sphb) {
449         goto param_error_exit;
450     }
451 
452     pdev = pci_find_device(PCI_HOST_BRIDGE(sphb)->bus,
453                            (addr >> 16) & 0xFF, (addr >> 8) & 0xFF);
454     if (!pdev || !object_dynamic_cast(OBJECT(pdev), "vfio-pci")) {
455         goto param_error_exit;
456     }
457 
458     spc = SPAPR_PCI_HOST_BRIDGE_GET_CLASS(sphb);
459     if (!spc->eeh_set_option) {
460         goto param_error_exit;
461     }
462 
463     ret = spc->eeh_set_option(sphb, addr, option);
464     rtas_st(rets, 0, ret);
465     return;
466 
467 param_error_exit:
468     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
469 }
470 
471 static void rtas_ibm_get_config_addr_info2(PowerPCCPU *cpu,
472                                            sPAPRMachineState *spapr,
473                                            uint32_t token, uint32_t nargs,
474                                            target_ulong args, uint32_t nret,
475                                            target_ulong rets)
476 {
477     sPAPRPHBState *sphb;
478     sPAPRPHBClass *spc;
479     PCIDevice *pdev;
480     uint32_t addr, option;
481     uint64_t buid;
482 
483     if ((nargs != 4) || (nret != 2)) {
484         goto param_error_exit;
485     }
486 
487     buid = rtas_ldq(args, 1);
488     sphb = spapr_pci_find_phb(spapr, buid);
489     if (!sphb) {
490         goto param_error_exit;
491     }
492 
493     spc = SPAPR_PCI_HOST_BRIDGE_GET_CLASS(sphb);
494     if (!spc->eeh_set_option) {
495         goto param_error_exit;
496     }
497 
498     /*
499      * We always have PE address of form "00BB0001". "BB"
500      * represents the bus number of PE's primary bus.
501      */
502     option = rtas_ld(args, 3);
503     switch (option) {
504     case RTAS_GET_PE_ADDR:
505         addr = rtas_ld(args, 0);
506         pdev = spapr_pci_find_dev(spapr, buid, addr);
507         if (!pdev) {
508             goto param_error_exit;
509         }
510 
511         rtas_st(rets, 1, (pci_bus_num(pdev->bus) << 16) + 1);
512         break;
513     case RTAS_GET_PE_MODE:
514         rtas_st(rets, 1, RTAS_PE_MODE_SHARED);
515         break;
516     default:
517         goto param_error_exit;
518     }
519 
520     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
521     return;
522 
523 param_error_exit:
524     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
525 }
526 
527 static void rtas_ibm_read_slot_reset_state2(PowerPCCPU *cpu,
528                                             sPAPRMachineState *spapr,
529                                             uint32_t token, uint32_t nargs,
530                                             target_ulong args, uint32_t nret,
531                                             target_ulong rets)
532 {
533     sPAPRPHBState *sphb;
534     sPAPRPHBClass *spc;
535     uint64_t buid;
536     int state, ret;
537 
538     if ((nargs != 3) || (nret != 4 && nret != 5)) {
539         goto param_error_exit;
540     }
541 
542     buid = rtas_ldq(args, 1);
543     sphb = spapr_pci_find_phb(spapr, buid);
544     if (!sphb) {
545         goto param_error_exit;
546     }
547 
548     spc = SPAPR_PCI_HOST_BRIDGE_GET_CLASS(sphb);
549     if (!spc->eeh_get_state) {
550         goto param_error_exit;
551     }
552 
553     ret = spc->eeh_get_state(sphb, &state);
554     rtas_st(rets, 0, ret);
555     if (ret != RTAS_OUT_SUCCESS) {
556         return;
557     }
558 
559     rtas_st(rets, 1, state);
560     rtas_st(rets, 2, RTAS_EEH_SUPPORT);
561     rtas_st(rets, 3, RTAS_EEH_PE_UNAVAIL_INFO);
562     if (nret >= 5) {
563         rtas_st(rets, 4, RTAS_EEH_PE_RECOVER_INFO);
564     }
565     return;
566 
567 param_error_exit:
568     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
569 }
570 
571 static void rtas_ibm_set_slot_reset(PowerPCCPU *cpu,
572                                     sPAPRMachineState *spapr,
573                                     uint32_t token, uint32_t nargs,
574                                     target_ulong args, uint32_t nret,
575                                     target_ulong rets)
576 {
577     sPAPRPHBState *sphb;
578     sPAPRPHBClass *spc;
579     uint32_t option;
580     uint64_t buid;
581     int ret;
582 
583     if ((nargs != 4) || (nret != 1)) {
584         goto param_error_exit;
585     }
586 
587     buid = rtas_ldq(args, 1);
588     option = rtas_ld(args, 3);
589     sphb = spapr_pci_find_phb(spapr, buid);
590     if (!sphb) {
591         goto param_error_exit;
592     }
593 
594     spc = SPAPR_PCI_HOST_BRIDGE_GET_CLASS(sphb);
595     if (!spc->eeh_reset) {
596         goto param_error_exit;
597     }
598 
599     ret = spc->eeh_reset(sphb, option);
600     rtas_st(rets, 0, ret);
601     return;
602 
603 param_error_exit:
604     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
605 }
606 
607 static void rtas_ibm_configure_pe(PowerPCCPU *cpu,
608                                   sPAPRMachineState *spapr,
609                                   uint32_t token, uint32_t nargs,
610                                   target_ulong args, uint32_t nret,
611                                   target_ulong rets)
612 {
613     sPAPRPHBState *sphb;
614     sPAPRPHBClass *spc;
615     uint64_t buid;
616     int ret;
617 
618     if ((nargs != 3) || (nret != 1)) {
619         goto param_error_exit;
620     }
621 
622     buid = rtas_ldq(args, 1);
623     sphb = spapr_pci_find_phb(spapr, buid);
624     if (!sphb) {
625         goto param_error_exit;
626     }
627 
628     spc = SPAPR_PCI_HOST_BRIDGE_GET_CLASS(sphb);
629     if (!spc->eeh_configure) {
630         goto param_error_exit;
631     }
632 
633     ret = spc->eeh_configure(sphb);
634     rtas_st(rets, 0, ret);
635     return;
636 
637 param_error_exit:
638     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
639 }
640 
641 /* To support it later */
642 static void rtas_ibm_slot_error_detail(PowerPCCPU *cpu,
643                                        sPAPRMachineState *spapr,
644                                        uint32_t token, uint32_t nargs,
645                                        target_ulong args, uint32_t nret,
646                                        target_ulong rets)
647 {
648     sPAPRPHBState *sphb;
649     sPAPRPHBClass *spc;
650     int option;
651     uint64_t buid;
652 
653     if ((nargs != 8) || (nret != 1)) {
654         goto param_error_exit;
655     }
656 
657     buid = rtas_ldq(args, 1);
658     sphb = spapr_pci_find_phb(spapr, buid);
659     if (!sphb) {
660         goto param_error_exit;
661     }
662 
663     spc = SPAPR_PCI_HOST_BRIDGE_GET_CLASS(sphb);
664     if (!spc->eeh_set_option) {
665         goto param_error_exit;
666     }
667 
668     option = rtas_ld(args, 7);
669     switch (option) {
670     case RTAS_SLOT_TEMP_ERR_LOG:
671     case RTAS_SLOT_PERM_ERR_LOG:
672         break;
673     default:
674         goto param_error_exit;
675     }
676 
677     /* We don't have error log yet */
678     rtas_st(rets, 0, RTAS_OUT_NO_ERRORS_FOUND);
679     return;
680 
681 param_error_exit:
682     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
683 }
684 
685 static int pci_spapr_swizzle(int slot, int pin)
686 {
687     return (slot + pin) % PCI_NUM_PINS;
688 }
689 
690 static int pci_spapr_map_irq(PCIDevice *pci_dev, int irq_num)
691 {
692     /*
693      * Here we need to convert pci_dev + irq_num to some unique value
694      * which is less than number of IRQs on the specific bus (4).  We
695      * use standard PCI swizzling, that is (slot number + pin number)
696      * % 4.
697      */
698     return pci_spapr_swizzle(PCI_SLOT(pci_dev->devfn), irq_num);
699 }
700 
701 static void pci_spapr_set_irq(void *opaque, int irq_num, int level)
702 {
703     /*
704      * Here we use the number returned by pci_spapr_map_irq to find a
705      * corresponding qemu_irq.
706      */
707     sPAPRPHBState *phb = opaque;
708 
709     trace_spapr_pci_lsi_set(phb->dtbusname, irq_num, phb->lsi_table[irq_num].irq);
710     qemu_set_irq(spapr_phb_lsi_qirq(phb, irq_num), level);
711 }
712 
713 static PCIINTxRoute spapr_route_intx_pin_to_irq(void *opaque, int pin)
714 {
715     sPAPRPHBState *sphb = SPAPR_PCI_HOST_BRIDGE(opaque);
716     PCIINTxRoute route;
717 
718     route.mode = PCI_INTX_ENABLED;
719     route.irq = sphb->lsi_table[pin].irq;
720 
721     return route;
722 }
723 
724 /*
725  * MSI/MSIX memory region implementation.
726  * The handler handles both MSI and MSIX.
727  * For MSI-X, the vector number is encoded as a part of the address,
728  * data is set to 0.
729  * For MSI, the vector number is encoded in least bits in data.
730  */
731 static void spapr_msi_write(void *opaque, hwaddr addr,
732                             uint64_t data, unsigned size)
733 {
734     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
735     uint32_t irq = data;
736 
737     trace_spapr_pci_msi_write(addr, data, irq);
738 
739     qemu_irq_pulse(xics_get_qirq(spapr->icp, irq));
740 }
741 
742 static const MemoryRegionOps spapr_msi_ops = {
743     /* There is no .read as the read result is undefined by PCI spec */
744     .read = NULL,
745     .write = spapr_msi_write,
746     .endianness = DEVICE_LITTLE_ENDIAN
747 };
748 
749 /*
750  * PHB PCI device
751  */
752 static AddressSpace *spapr_pci_dma_iommu(PCIBus *bus, void *opaque, int devfn)
753 {
754     sPAPRPHBState *phb = opaque;
755 
756     return &phb->iommu_as;
757 }
758 
759 static char *spapr_phb_vfio_get_loc_code(sPAPRPHBState *sphb,  PCIDevice *pdev)
760 {
761     char *path = NULL, *buf = NULL, *host = NULL;
762 
763     /* Get the PCI VFIO host id */
764     host = object_property_get_str(OBJECT(pdev), "host", NULL);
765     if (!host) {
766         goto err_out;
767     }
768 
769     /* Construct the path of the file that will give us the DT location */
770     path = g_strdup_printf("/sys/bus/pci/devices/%s/devspec", host);
771     g_free(host);
772     if (!path || !g_file_get_contents(path, &buf, NULL, NULL)) {
773         goto err_out;
774     }
775     g_free(path);
776 
777     /* Construct and read from host device tree the loc-code */
778     path = g_strdup_printf("/proc/device-tree%s/ibm,loc-code", buf);
779     g_free(buf);
780     if (!path || !g_file_get_contents(path, &buf, NULL, NULL)) {
781         goto err_out;
782     }
783     return buf;
784 
785 err_out:
786     g_free(path);
787     return NULL;
788 }
789 
790 static char *spapr_phb_get_loc_code(sPAPRPHBState *sphb, PCIDevice *pdev)
791 {
792     char *buf;
793     const char *devtype = "qemu";
794     uint32_t busnr = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(pdev))));
795 
796     if (object_dynamic_cast(OBJECT(pdev), "vfio-pci")) {
797         buf = spapr_phb_vfio_get_loc_code(sphb, pdev);
798         if (buf) {
799             return buf;
800         }
801         devtype = "vfio";
802     }
803     /*
804      * For emulated devices and VFIO-failure case, make up
805      * the loc-code.
806      */
807     buf = g_strdup_printf("%s_%s:%04x:%02x:%02x.%x",
808                           devtype, pdev->name, sphb->index, busnr,
809                           PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
810     return buf;
811 }
812 
813 /* Macros to operate with address in OF binding to PCI */
814 #define b_x(x, p, l)    (((x) & ((1<<(l))-1)) << (p))
815 #define b_n(x)          b_x((x), 31, 1) /* 0 if relocatable */
816 #define b_p(x)          b_x((x), 30, 1) /* 1 if prefetchable */
817 #define b_t(x)          b_x((x), 29, 1) /* 1 if the address is aliased */
818 #define b_ss(x)         b_x((x), 24, 2) /* the space code */
819 #define b_bbbbbbbb(x)   b_x((x), 16, 8) /* bus number */
820 #define b_ddddd(x)      b_x((x), 11, 5) /* device number */
821 #define b_fff(x)        b_x((x), 8, 3)  /* function number */
822 #define b_rrrrrrrr(x)   b_x((x), 0, 8)  /* register number */
823 
824 /* for 'reg'/'assigned-addresses' OF properties */
825 #define RESOURCE_CELLS_SIZE 2
826 #define RESOURCE_CELLS_ADDRESS 3
827 
828 typedef struct ResourceFields {
829     uint32_t phys_hi;
830     uint32_t phys_mid;
831     uint32_t phys_lo;
832     uint32_t size_hi;
833     uint32_t size_lo;
834 } QEMU_PACKED ResourceFields;
835 
836 typedef struct ResourceProps {
837     ResourceFields reg[8];
838     ResourceFields assigned[7];
839     uint32_t reg_len;
840     uint32_t assigned_len;
841 } ResourceProps;
842 
843 /* fill in the 'reg'/'assigned-resources' OF properties for
844  * a PCI device. 'reg' describes resource requirements for a
845  * device's IO/MEM regions, 'assigned-addresses' describes the
846  * actual resource assignments.
847  *
848  * the properties are arrays of ('phys-addr', 'size') pairs describing
849  * the addressable regions of the PCI device, where 'phys-addr' is a
850  * RESOURCE_CELLS_ADDRESS-tuple of 32-bit integers corresponding to
851  * (phys.hi, phys.mid, phys.lo), and 'size' is a
852  * RESOURCE_CELLS_SIZE-tuple corresponding to (size.hi, size.lo).
853  *
854  * phys.hi = 0xYYXXXXZZ, where:
855  *   0xYY = npt000ss
856  *          |||   |
857  *          |||   +-- space code
858  *          |||               |
859  *          |||               +  00 if configuration space
860  *          |||               +  01 if IO region,
861  *          |||               +  10 if 32-bit MEM region
862  *          |||               +  11 if 64-bit MEM region
863  *          |||
864  *          ||+------ for non-relocatable IO: 1 if aliased
865  *          ||        for relocatable IO: 1 if below 64KB
866  *          ||        for MEM: 1 if below 1MB
867  *          |+------- 1 if region is prefetchable
868  *          +-------- 1 if region is non-relocatable
869  *   0xXXXX = bbbbbbbb dddddfff, encoding bus, slot, and function
870  *            bits respectively
871  *   0xZZ = rrrrrrrr, the register number of the BAR corresponding
872  *          to the region
873  *
874  * phys.mid and phys.lo correspond respectively to the hi/lo portions
875  * of the actual address of the region.
876  *
877  * how the phys-addr/size values are used differ slightly between
878  * 'reg' and 'assigned-addresses' properties. namely, 'reg' has
879  * an additional description for the config space region of the
880  * device, and in the case of QEMU has n=0 and phys.mid=phys.lo=0
881  * to describe the region as relocatable, with an address-mapping
882  * that corresponds directly to the PHB's address space for the
883  * resource. 'assigned-addresses' always has n=1 set with an absolute
884  * address assigned for the resource. in general, 'assigned-addresses'
885  * won't be populated, since addresses for PCI devices are generally
886  * unmapped initially and left to the guest to assign.
887  *
888  * note also that addresses defined in these properties are, at least
889  * for PAPR guests, relative to the PHBs IO/MEM windows, and
890  * correspond directly to the addresses in the BARs.
891  *
892  * in accordance with PCI Bus Binding to Open Firmware,
893  * IEEE Std 1275-1994, section 4.1.1, as implemented by PAPR+ v2.7,
894  * Appendix C.
895  */
896 static void populate_resource_props(PCIDevice *d, ResourceProps *rp)
897 {
898     int bus_num = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(d))));
899     uint32_t dev_id = (b_bbbbbbbb(bus_num) |
900                        b_ddddd(PCI_SLOT(d->devfn)) |
901                        b_fff(PCI_FUNC(d->devfn)));
902     ResourceFields *reg, *assigned;
903     int i, reg_idx = 0, assigned_idx = 0;
904 
905     /* config space region */
906     reg = &rp->reg[reg_idx++];
907     reg->phys_hi = cpu_to_be32(dev_id);
908     reg->phys_mid = 0;
909     reg->phys_lo = 0;
910     reg->size_hi = 0;
911     reg->size_lo = 0;
912 
913     for (i = 0; i < PCI_NUM_REGIONS; i++) {
914         if (!d->io_regions[i].size) {
915             continue;
916         }
917 
918         reg = &rp->reg[reg_idx++];
919 
920         reg->phys_hi = cpu_to_be32(dev_id | b_rrrrrrrr(pci_bar(d, i)));
921         if (d->io_regions[i].type & PCI_BASE_ADDRESS_SPACE_IO) {
922             reg->phys_hi |= cpu_to_be32(b_ss(1));
923         } else if (d->io_regions[i].type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
924             reg->phys_hi |= cpu_to_be32(b_ss(3));
925         } else {
926             reg->phys_hi |= cpu_to_be32(b_ss(2));
927         }
928         reg->phys_mid = 0;
929         reg->phys_lo = 0;
930         reg->size_hi = cpu_to_be32(d->io_regions[i].size >> 32);
931         reg->size_lo = cpu_to_be32(d->io_regions[i].size);
932 
933         if (d->io_regions[i].addr == PCI_BAR_UNMAPPED) {
934             continue;
935         }
936 
937         assigned = &rp->assigned[assigned_idx++];
938         assigned->phys_hi = cpu_to_be32(reg->phys_hi | b_n(1));
939         assigned->phys_mid = cpu_to_be32(d->io_regions[i].addr >> 32);
940         assigned->phys_lo = cpu_to_be32(d->io_regions[i].addr);
941         assigned->size_hi = reg->size_hi;
942         assigned->size_lo = reg->size_lo;
943     }
944 
945     rp->reg_len = reg_idx * sizeof(ResourceFields);
946     rp->assigned_len = assigned_idx * sizeof(ResourceFields);
947 }
948 
949 static uint32_t spapr_phb_get_pci_drc_index(sPAPRPHBState *phb,
950                                             PCIDevice *pdev);
951 
952 static int spapr_populate_pci_child_dt(PCIDevice *dev, void *fdt, int offset,
953                                        sPAPRPHBState *sphb)
954 {
955     ResourceProps rp;
956     bool is_bridge = false;
957     int pci_status, err;
958     char *buf = NULL;
959     uint32_t drc_index = spapr_phb_get_pci_drc_index(sphb, dev);
960     uint32_t max_msi, max_msix;
961 
962     if (pci_default_read_config(dev, PCI_HEADER_TYPE, 1) ==
963         PCI_HEADER_TYPE_BRIDGE) {
964         is_bridge = true;
965     }
966 
967     /* in accordance with PAPR+ v2.7 13.6.3, Table 181 */
968     _FDT(fdt_setprop_cell(fdt, offset, "vendor-id",
969                           pci_default_read_config(dev, PCI_VENDOR_ID, 2)));
970     _FDT(fdt_setprop_cell(fdt, offset, "device-id",
971                           pci_default_read_config(dev, PCI_DEVICE_ID, 2)));
972     _FDT(fdt_setprop_cell(fdt, offset, "revision-id",
973                           pci_default_read_config(dev, PCI_REVISION_ID, 1)));
974     _FDT(fdt_setprop_cell(fdt, offset, "class-code",
975                           pci_default_read_config(dev, PCI_CLASS_PROG, 3)));
976     if (pci_default_read_config(dev, PCI_INTERRUPT_PIN, 1)) {
977         _FDT(fdt_setprop_cell(fdt, offset, "interrupts",
978                  pci_default_read_config(dev, PCI_INTERRUPT_PIN, 1)));
979     }
980 
981     if (!is_bridge) {
982         _FDT(fdt_setprop_cell(fdt, offset, "min-grant",
983             pci_default_read_config(dev, PCI_MIN_GNT, 1)));
984         _FDT(fdt_setprop_cell(fdt, offset, "max-latency",
985             pci_default_read_config(dev, PCI_MAX_LAT, 1)));
986     }
987 
988     if (pci_default_read_config(dev, PCI_SUBSYSTEM_ID, 2)) {
989         _FDT(fdt_setprop_cell(fdt, offset, "subsystem-id",
990                  pci_default_read_config(dev, PCI_SUBSYSTEM_ID, 2)));
991     }
992 
993     if (pci_default_read_config(dev, PCI_SUBSYSTEM_VENDOR_ID, 2)) {
994         _FDT(fdt_setprop_cell(fdt, offset, "subsystem-vendor-id",
995                  pci_default_read_config(dev, PCI_SUBSYSTEM_VENDOR_ID, 2)));
996     }
997 
998     _FDT(fdt_setprop_cell(fdt, offset, "cache-line-size",
999         pci_default_read_config(dev, PCI_CACHE_LINE_SIZE, 1)));
1000 
1001     /* the following fdt cells are masked off the pci status register */
1002     pci_status = pci_default_read_config(dev, PCI_STATUS, 2);
1003     _FDT(fdt_setprop_cell(fdt, offset, "devsel-speed",
1004                           PCI_STATUS_DEVSEL_MASK & pci_status));
1005 
1006     if (pci_status & PCI_STATUS_FAST_BACK) {
1007         _FDT(fdt_setprop(fdt, offset, "fast-back-to-back", NULL, 0));
1008     }
1009     if (pci_status & PCI_STATUS_66MHZ) {
1010         _FDT(fdt_setprop(fdt, offset, "66mhz-capable", NULL, 0));
1011     }
1012     if (pci_status & PCI_STATUS_UDF) {
1013         _FDT(fdt_setprop(fdt, offset, "udf-supported", NULL, 0));
1014     }
1015 
1016     /* NOTE: this is normally generated by firmware via path/unit name,
1017      * but in our case we must set it manually since it does not get
1018      * processed by OF beforehand
1019      */
1020     _FDT(fdt_setprop_string(fdt, offset, "name", "pci"));
1021     buf = spapr_phb_get_loc_code(sphb, dev);
1022     if (!buf) {
1023         error_report("Failed setting the ibm,loc-code");
1024         return -1;
1025     }
1026 
1027     err = fdt_setprop_string(fdt, offset, "ibm,loc-code", buf);
1028     g_free(buf);
1029     if (err < 0) {
1030         return err;
1031     }
1032 
1033     if (drc_index) {
1034         _FDT(fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index));
1035     }
1036 
1037     _FDT(fdt_setprop_cell(fdt, offset, "#address-cells",
1038                           RESOURCE_CELLS_ADDRESS));
1039     _FDT(fdt_setprop_cell(fdt, offset, "#size-cells",
1040                           RESOURCE_CELLS_SIZE));
1041 
1042     max_msi = msi_nr_vectors_allocated(dev);
1043     if (max_msi) {
1044         _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi", max_msi));
1045     }
1046     max_msix = dev->msix_entries_nr;
1047     if (max_msix) {
1048         _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi-x", max_msix));
1049     }
1050 
1051     populate_resource_props(dev, &rp);
1052     _FDT(fdt_setprop(fdt, offset, "reg", (uint8_t *)rp.reg, rp.reg_len));
1053     _FDT(fdt_setprop(fdt, offset, "assigned-addresses",
1054                      (uint8_t *)rp.assigned, rp.assigned_len));
1055 
1056     return 0;
1057 }
1058 
1059 /* create OF node for pci device and required OF DT properties */
1060 static int spapr_create_pci_child_dt(sPAPRPHBState *phb, PCIDevice *dev,
1061                                      void *fdt, int node_offset)
1062 {
1063     int offset, ret;
1064     int slot = PCI_SLOT(dev->devfn);
1065     int func = PCI_FUNC(dev->devfn);
1066     char nodename[FDT_NAME_MAX];
1067 
1068     if (func != 0) {
1069         snprintf(nodename, FDT_NAME_MAX, "pci@%x,%x", slot, func);
1070     } else {
1071         snprintf(nodename, FDT_NAME_MAX, "pci@%x", slot);
1072     }
1073     offset = fdt_add_subnode(fdt, node_offset, nodename);
1074     ret = spapr_populate_pci_child_dt(dev, fdt, offset, phb);
1075 
1076     g_assert(!ret);
1077     if (ret) {
1078         return 0;
1079     }
1080     return offset;
1081 }
1082 
1083 static void spapr_phb_add_pci_device(sPAPRDRConnector *drc,
1084                                      sPAPRPHBState *phb,
1085                                      PCIDevice *pdev,
1086                                      Error **errp)
1087 {
1088     sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
1089     DeviceState *dev = DEVICE(pdev);
1090     void *fdt = NULL;
1091     int fdt_start_offset = 0, fdt_size;
1092 
1093     if (dev->hotplugged) {
1094         fdt = create_device_tree(&fdt_size);
1095         fdt_start_offset = spapr_create_pci_child_dt(phb, pdev, fdt, 0);
1096         if (!fdt_start_offset) {
1097             error_setg(errp, "Failed to create pci child device tree node");
1098             goto out;
1099         }
1100     }
1101 
1102     drck->attach(drc, DEVICE(pdev),
1103                  fdt, fdt_start_offset, !dev->hotplugged, errp);
1104 out:
1105     if (*errp) {
1106         g_free(fdt);
1107     }
1108 }
1109 
1110 static void spapr_phb_remove_pci_device_cb(DeviceState *dev, void *opaque)
1111 {
1112     /* some version guests do not wait for completion of a device
1113      * cleanup (generally done asynchronously by the kernel) before
1114      * signaling to QEMU that the device is safe, but instead sleep
1115      * for some 'safe' period of time. unfortunately on a busy host
1116      * this sleep isn't guaranteed to be long enough, resulting in
1117      * bad things like IRQ lines being left asserted during final
1118      * device removal. to deal with this we call reset just prior
1119      * to finalizing the device, which will put the device back into
1120      * an 'idle' state, as the device cleanup code expects.
1121      */
1122     pci_device_reset(PCI_DEVICE(dev));
1123     object_unparent(OBJECT(dev));
1124 }
1125 
1126 static void spapr_phb_remove_pci_device(sPAPRDRConnector *drc,
1127                                         sPAPRPHBState *phb,
1128                                         PCIDevice *pdev,
1129                                         Error **errp)
1130 {
1131     sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
1132 
1133     drck->detach(drc, DEVICE(pdev), spapr_phb_remove_pci_device_cb, phb, errp);
1134 }
1135 
1136 static sPAPRDRConnector *spapr_phb_get_pci_drc(sPAPRPHBState *phb,
1137                                                PCIDevice *pdev)
1138 {
1139     uint32_t busnr = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(pdev))));
1140     return spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_PCI,
1141                                     (phb->index << 16) |
1142                                     (busnr << 8) |
1143                                     pdev->devfn);
1144 }
1145 
1146 static uint32_t spapr_phb_get_pci_drc_index(sPAPRPHBState *phb,
1147                                             PCIDevice *pdev)
1148 {
1149     sPAPRDRConnector *drc = spapr_phb_get_pci_drc(phb, pdev);
1150     sPAPRDRConnectorClass *drck;
1151 
1152     if (!drc) {
1153         return 0;
1154     }
1155 
1156     drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
1157     return drck->get_index(drc);
1158 }
1159 
1160 static void spapr_phb_hot_plug_child(HotplugHandler *plug_handler,
1161                                      DeviceState *plugged_dev, Error **errp)
1162 {
1163     sPAPRPHBState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1164     PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1165     sPAPRDRConnector *drc = spapr_phb_get_pci_drc(phb, pdev);
1166     Error *local_err = NULL;
1167 
1168     /* if DR is disabled we don't need to do anything in the case of
1169      * hotplug or coldplug callbacks
1170      */
1171     if (!phb->dr_enabled) {
1172         /* if this is a hotplug operation initiated by the user
1173          * we need to let them know it's not enabled
1174          */
1175         if (plugged_dev->hotplugged) {
1176             error_setg(errp, QERR_BUS_NO_HOTPLUG,
1177                        object_get_typename(OBJECT(phb)));
1178         }
1179         return;
1180     }
1181 
1182     g_assert(drc);
1183 
1184     spapr_phb_add_pci_device(drc, phb, pdev, &local_err);
1185     if (local_err) {
1186         error_propagate(errp, local_err);
1187         return;
1188     }
1189     if (plugged_dev->hotplugged) {
1190         spapr_hotplug_req_add_by_index(drc);
1191     }
1192 }
1193 
1194 static void spapr_phb_hot_unplug_child(HotplugHandler *plug_handler,
1195                                        DeviceState *plugged_dev, Error **errp)
1196 {
1197     sPAPRPHBState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1198     PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1199     sPAPRDRConnectorClass *drck;
1200     sPAPRDRConnector *drc = spapr_phb_get_pci_drc(phb, pdev);
1201     Error *local_err = NULL;
1202 
1203     if (!phb->dr_enabled) {
1204         error_setg(errp, QERR_BUS_NO_HOTPLUG,
1205                    object_get_typename(OBJECT(phb)));
1206         return;
1207     }
1208 
1209     g_assert(drc);
1210 
1211     drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
1212     if (!drck->release_pending(drc)) {
1213         spapr_phb_remove_pci_device(drc, phb, pdev, &local_err);
1214         if (local_err) {
1215             error_propagate(errp, local_err);
1216             return;
1217         }
1218         spapr_hotplug_req_remove_by_index(drc);
1219     }
1220 }
1221 
1222 static void spapr_phb_realize(DeviceState *dev, Error **errp)
1223 {
1224     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
1225     SysBusDevice *s = SYS_BUS_DEVICE(dev);
1226     sPAPRPHBState *sphb = SPAPR_PCI_HOST_BRIDGE(s);
1227     PCIHostState *phb = PCI_HOST_BRIDGE(s);
1228     sPAPRPHBClass *info = SPAPR_PCI_HOST_BRIDGE_GET_CLASS(s);
1229     char *namebuf;
1230     int i;
1231     PCIBus *bus;
1232     uint64_t msi_window_size = 4096;
1233 
1234     if (sphb->index != (uint32_t)-1) {
1235         hwaddr windows_base;
1236 
1237         if ((sphb->buid != (uint64_t)-1) || (sphb->dma_liobn != (uint32_t)-1)
1238             || (sphb->mem_win_addr != (hwaddr)-1)
1239             || (sphb->io_win_addr != (hwaddr)-1)) {
1240             error_setg(errp, "Either \"index\" or other parameters must"
1241                        " be specified for PAPR PHB, not both");
1242             return;
1243         }
1244 
1245         if (sphb->index > SPAPR_PCI_MAX_INDEX) {
1246             error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
1247                        SPAPR_PCI_MAX_INDEX);
1248             return;
1249         }
1250 
1251         sphb->buid = SPAPR_PCI_BASE_BUID + sphb->index;
1252         sphb->dma_liobn = SPAPR_PCI_LIOBN(sphb->index, 0);
1253 
1254         windows_base = SPAPR_PCI_WINDOW_BASE
1255             + sphb->index * SPAPR_PCI_WINDOW_SPACING;
1256         sphb->mem_win_addr = windows_base + SPAPR_PCI_MMIO_WIN_OFF;
1257         sphb->io_win_addr = windows_base + SPAPR_PCI_IO_WIN_OFF;
1258     }
1259 
1260     if (sphb->buid == (uint64_t)-1) {
1261         error_setg(errp, "BUID not specified for PHB");
1262         return;
1263     }
1264 
1265     if (sphb->dma_liobn == (uint32_t)-1) {
1266         error_setg(errp, "LIOBN not specified for PHB");
1267         return;
1268     }
1269 
1270     if (sphb->mem_win_addr == (hwaddr)-1) {
1271         error_setg(errp, "Memory window address not specified for PHB");
1272         return;
1273     }
1274 
1275     if (sphb->io_win_addr == (hwaddr)-1) {
1276         error_setg(errp, "IO window address not specified for PHB");
1277         return;
1278     }
1279 
1280     if (spapr_pci_find_phb(spapr, sphb->buid)) {
1281         error_setg(errp, "PCI host bridges must have unique BUIDs");
1282         return;
1283     }
1284 
1285     sphb->dtbusname = g_strdup_printf("pci@%" PRIx64, sphb->buid);
1286 
1287     namebuf = alloca(strlen(sphb->dtbusname) + 32);
1288 
1289     /* Initialize memory regions */
1290     sprintf(namebuf, "%s.mmio", sphb->dtbusname);
1291     memory_region_init(&sphb->memspace, OBJECT(sphb), namebuf, UINT64_MAX);
1292 
1293     sprintf(namebuf, "%s.mmio-alias", sphb->dtbusname);
1294     memory_region_init_alias(&sphb->memwindow, OBJECT(sphb),
1295                              namebuf, &sphb->memspace,
1296                              SPAPR_PCI_MEM_WIN_BUS_OFFSET, sphb->mem_win_size);
1297     memory_region_add_subregion(get_system_memory(), sphb->mem_win_addr,
1298                                 &sphb->memwindow);
1299 
1300     /* Initialize IO regions */
1301     sprintf(namebuf, "%s.io", sphb->dtbusname);
1302     memory_region_init(&sphb->iospace, OBJECT(sphb),
1303                        namebuf, SPAPR_PCI_IO_WIN_SIZE);
1304 
1305     sprintf(namebuf, "%s.io-alias", sphb->dtbusname);
1306     memory_region_init_alias(&sphb->iowindow, OBJECT(sphb), namebuf,
1307                              &sphb->iospace, 0, SPAPR_PCI_IO_WIN_SIZE);
1308     memory_region_add_subregion(get_system_memory(), sphb->io_win_addr,
1309                                 &sphb->iowindow);
1310 
1311     bus = pci_register_bus(dev, NULL,
1312                            pci_spapr_set_irq, pci_spapr_map_irq, sphb,
1313                            &sphb->memspace, &sphb->iospace,
1314                            PCI_DEVFN(0, 0), PCI_NUM_PINS, TYPE_PCI_BUS);
1315     phb->bus = bus;
1316     qbus_set_hotplug_handler(BUS(phb->bus), DEVICE(sphb), NULL);
1317 
1318     /*
1319      * Initialize PHB address space.
1320      * By default there will be at least one subregion for default
1321      * 32bit DMA window.
1322      * Later the guest might want to create another DMA window
1323      * which will become another memory subregion.
1324      */
1325     sprintf(namebuf, "%s.iommu-root", sphb->dtbusname);
1326 
1327     memory_region_init(&sphb->iommu_root, OBJECT(sphb),
1328                        namebuf, UINT64_MAX);
1329     address_space_init(&sphb->iommu_as, &sphb->iommu_root,
1330                        sphb->dtbusname);
1331 
1332     /*
1333      * As MSI/MSIX interrupts trigger by writing at MSI/MSIX vectors,
1334      * we need to allocate some memory to catch those writes coming
1335      * from msi_notify()/msix_notify().
1336      * As MSIMessage:addr is going to be the same and MSIMessage:data
1337      * is going to be a VIRQ number, 4 bytes of the MSI MR will only
1338      * be used.
1339      *
1340      * For KVM we want to ensure that this memory is a full page so that
1341      * our memory slot is of page size granularity.
1342      */
1343 #ifdef CONFIG_KVM
1344     if (kvm_enabled()) {
1345         msi_window_size = getpagesize();
1346     }
1347 #endif
1348 
1349     memory_region_init_io(&sphb->msiwindow, NULL, &spapr_msi_ops, spapr,
1350                           "msi", msi_window_size);
1351     memory_region_add_subregion(&sphb->iommu_root, SPAPR_PCI_MSI_WINDOW,
1352                                 &sphb->msiwindow);
1353 
1354     pci_setup_iommu(bus, spapr_pci_dma_iommu, sphb);
1355 
1356     pci_bus_set_route_irq_fn(bus, spapr_route_intx_pin_to_irq);
1357 
1358     QLIST_INSERT_HEAD(&spapr->phbs, sphb, list);
1359 
1360     /* Initialize the LSI table */
1361     for (i = 0; i < PCI_NUM_PINS; i++) {
1362         uint32_t irq;
1363 
1364         irq = xics_alloc_block(spapr->icp, 0, 1, true, false);
1365         if (!irq) {
1366             error_setg(errp, "spapr_allocate_lsi failed");
1367             return;
1368         }
1369 
1370         sphb->lsi_table[i].irq = irq;
1371     }
1372 
1373     /* allocate connectors for child PCI devices */
1374     if (sphb->dr_enabled) {
1375         for (i = 0; i < PCI_SLOT_MAX * 8; i++) {
1376             spapr_dr_connector_new(OBJECT(phb),
1377                                    SPAPR_DR_CONNECTOR_TYPE_PCI,
1378                                    (sphb->index << 16) | i);
1379         }
1380     }
1381 
1382     if (!info->finish_realize) {
1383         error_setg(errp, "finish_realize not defined");
1384         return;
1385     }
1386 
1387     info->finish_realize(sphb, errp);
1388 
1389     sphb->msi = g_hash_table_new_full(g_int_hash, g_int_equal, g_free, g_free);
1390 }
1391 
1392 static void spapr_phb_finish_realize(sPAPRPHBState *sphb, Error **errp)
1393 {
1394     sPAPRTCETable *tcet;
1395     uint32_t nb_table;
1396 
1397     nb_table = SPAPR_PCI_DMA32_SIZE >> SPAPR_TCE_PAGE_SHIFT;
1398     tcet = spapr_tce_new_table(DEVICE(sphb), sphb->dma_liobn,
1399                                0, SPAPR_TCE_PAGE_SHIFT, nb_table, false);
1400     if (!tcet) {
1401         error_setg(errp, "Unable to create TCE table for %s",
1402                    sphb->dtbusname);
1403         return ;
1404     }
1405 
1406     /* Register default 32bit DMA window */
1407     memory_region_add_subregion(&sphb->iommu_root, 0,
1408                                 spapr_tce_get_iommu(tcet));
1409 }
1410 
1411 static int spapr_phb_children_reset(Object *child, void *opaque)
1412 {
1413     DeviceState *dev = (DeviceState *) object_dynamic_cast(child, TYPE_DEVICE);
1414 
1415     if (dev) {
1416         device_reset(dev);
1417     }
1418 
1419     return 0;
1420 }
1421 
1422 static void spapr_phb_reset(DeviceState *qdev)
1423 {
1424     /* Reset the IOMMU state */
1425     object_child_foreach(OBJECT(qdev), spapr_phb_children_reset, NULL);
1426 }
1427 
1428 static Property spapr_phb_properties[] = {
1429     DEFINE_PROP_UINT32("index", sPAPRPHBState, index, -1),
1430     DEFINE_PROP_UINT64("buid", sPAPRPHBState, buid, -1),
1431     DEFINE_PROP_UINT32("liobn", sPAPRPHBState, dma_liobn, -1),
1432     DEFINE_PROP_UINT64("mem_win_addr", sPAPRPHBState, mem_win_addr, -1),
1433     DEFINE_PROP_UINT64("mem_win_size", sPAPRPHBState, mem_win_size,
1434                        SPAPR_PCI_MMIO_WIN_SIZE),
1435     DEFINE_PROP_UINT64("io_win_addr", sPAPRPHBState, io_win_addr, -1),
1436     DEFINE_PROP_UINT64("io_win_size", sPAPRPHBState, io_win_size,
1437                        SPAPR_PCI_IO_WIN_SIZE),
1438     DEFINE_PROP_BOOL("dynamic-reconfiguration", sPAPRPHBState, dr_enabled,
1439                      true),
1440     DEFINE_PROP_END_OF_LIST(),
1441 };
1442 
1443 static const VMStateDescription vmstate_spapr_pci_lsi = {
1444     .name = "spapr_pci/lsi",
1445     .version_id = 1,
1446     .minimum_version_id = 1,
1447     .fields = (VMStateField[]) {
1448         VMSTATE_UINT32_EQUAL(irq, struct spapr_pci_lsi),
1449 
1450         VMSTATE_END_OF_LIST()
1451     },
1452 };
1453 
1454 static const VMStateDescription vmstate_spapr_pci_msi = {
1455     .name = "spapr_pci/msi",
1456     .version_id = 1,
1457     .minimum_version_id = 1,
1458     .fields = (VMStateField []) {
1459         VMSTATE_UINT32(key, spapr_pci_msi_mig),
1460         VMSTATE_UINT32(value.first_irq, spapr_pci_msi_mig),
1461         VMSTATE_UINT32(value.num, spapr_pci_msi_mig),
1462         VMSTATE_END_OF_LIST()
1463     },
1464 };
1465 
1466 static void spapr_pci_pre_save(void *opaque)
1467 {
1468     sPAPRPHBState *sphb = opaque;
1469     GHashTableIter iter;
1470     gpointer key, value;
1471     int i;
1472 
1473     g_free(sphb->msi_devs);
1474     sphb->msi_devs = NULL;
1475     sphb->msi_devs_num = g_hash_table_size(sphb->msi);
1476     if (!sphb->msi_devs_num) {
1477         return;
1478     }
1479     sphb->msi_devs = g_malloc(sphb->msi_devs_num * sizeof(spapr_pci_msi_mig));
1480 
1481     g_hash_table_iter_init(&iter, sphb->msi);
1482     for (i = 0; g_hash_table_iter_next(&iter, &key, &value); ++i) {
1483         sphb->msi_devs[i].key = *(uint32_t *) key;
1484         sphb->msi_devs[i].value = *(spapr_pci_msi *) value;
1485     }
1486 }
1487 
1488 static int spapr_pci_post_load(void *opaque, int version_id)
1489 {
1490     sPAPRPHBState *sphb = opaque;
1491     gpointer key, value;
1492     int i;
1493 
1494     for (i = 0; i < sphb->msi_devs_num; ++i) {
1495         key = g_memdup(&sphb->msi_devs[i].key,
1496                        sizeof(sphb->msi_devs[i].key));
1497         value = g_memdup(&sphb->msi_devs[i].value,
1498                          sizeof(sphb->msi_devs[i].value));
1499         g_hash_table_insert(sphb->msi, key, value);
1500     }
1501     g_free(sphb->msi_devs);
1502     sphb->msi_devs = NULL;
1503     sphb->msi_devs_num = 0;
1504 
1505     return 0;
1506 }
1507 
1508 static const VMStateDescription vmstate_spapr_pci = {
1509     .name = "spapr_pci",
1510     .version_id = 2,
1511     .minimum_version_id = 2,
1512     .pre_save = spapr_pci_pre_save,
1513     .post_load = spapr_pci_post_load,
1514     .fields = (VMStateField[]) {
1515         VMSTATE_UINT64_EQUAL(buid, sPAPRPHBState),
1516         VMSTATE_UINT32_EQUAL(dma_liobn, sPAPRPHBState),
1517         VMSTATE_UINT64_EQUAL(mem_win_addr, sPAPRPHBState),
1518         VMSTATE_UINT64_EQUAL(mem_win_size, sPAPRPHBState),
1519         VMSTATE_UINT64_EQUAL(io_win_addr, sPAPRPHBState),
1520         VMSTATE_UINT64_EQUAL(io_win_size, sPAPRPHBState),
1521         VMSTATE_STRUCT_ARRAY(lsi_table, sPAPRPHBState, PCI_NUM_PINS, 0,
1522                              vmstate_spapr_pci_lsi, struct spapr_pci_lsi),
1523         VMSTATE_INT32(msi_devs_num, sPAPRPHBState),
1524         VMSTATE_STRUCT_VARRAY_ALLOC(msi_devs, sPAPRPHBState, msi_devs_num, 0,
1525                                     vmstate_spapr_pci_msi, spapr_pci_msi_mig),
1526         VMSTATE_END_OF_LIST()
1527     },
1528 };
1529 
1530 static const char *spapr_phb_root_bus_path(PCIHostState *host_bridge,
1531                                            PCIBus *rootbus)
1532 {
1533     sPAPRPHBState *sphb = SPAPR_PCI_HOST_BRIDGE(host_bridge);
1534 
1535     return sphb->dtbusname;
1536 }
1537 
1538 static void spapr_phb_class_init(ObjectClass *klass, void *data)
1539 {
1540     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(klass);
1541     DeviceClass *dc = DEVICE_CLASS(klass);
1542     sPAPRPHBClass *spc = SPAPR_PCI_HOST_BRIDGE_CLASS(klass);
1543     HotplugHandlerClass *hp = HOTPLUG_HANDLER_CLASS(klass);
1544 
1545     hc->root_bus_path = spapr_phb_root_bus_path;
1546     dc->realize = spapr_phb_realize;
1547     dc->props = spapr_phb_properties;
1548     dc->reset = spapr_phb_reset;
1549     dc->vmsd = &vmstate_spapr_pci;
1550     set_bit(DEVICE_CATEGORY_BRIDGE, dc->categories);
1551     dc->cannot_instantiate_with_device_add_yet = false;
1552     spc->finish_realize = spapr_phb_finish_realize;
1553     hp->plug = spapr_phb_hot_plug_child;
1554     hp->unplug = spapr_phb_hot_unplug_child;
1555 }
1556 
1557 static const TypeInfo spapr_phb_info = {
1558     .name          = TYPE_SPAPR_PCI_HOST_BRIDGE,
1559     .parent        = TYPE_PCI_HOST_BRIDGE,
1560     .instance_size = sizeof(sPAPRPHBState),
1561     .class_init    = spapr_phb_class_init,
1562     .class_size    = sizeof(sPAPRPHBClass),
1563     .interfaces    = (InterfaceInfo[]) {
1564         { TYPE_HOTPLUG_HANDLER },
1565         { }
1566     }
1567 };
1568 
1569 PCIHostState *spapr_create_phb(sPAPRMachineState *spapr, int index)
1570 {
1571     DeviceState *dev;
1572 
1573     dev = qdev_create(NULL, TYPE_SPAPR_PCI_HOST_BRIDGE);
1574     qdev_prop_set_uint32(dev, "index", index);
1575     qdev_init_nofail(dev);
1576 
1577     return PCI_HOST_BRIDGE(dev);
1578 }
1579 
1580 typedef struct sPAPRFDT {
1581     void *fdt;
1582     int node_off;
1583     sPAPRPHBState *sphb;
1584 } sPAPRFDT;
1585 
1586 static void spapr_populate_pci_devices_dt(PCIBus *bus, PCIDevice *pdev,
1587                                           void *opaque)
1588 {
1589     PCIBus *sec_bus;
1590     sPAPRFDT *p = opaque;
1591     int offset;
1592     sPAPRFDT s_fdt;
1593 
1594     offset = spapr_create_pci_child_dt(p->sphb, pdev, p->fdt, p->node_off);
1595     if (!offset) {
1596         error_report("Failed to create pci child device tree node");
1597         return;
1598     }
1599 
1600     if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) !=
1601          PCI_HEADER_TYPE_BRIDGE)) {
1602         return;
1603     }
1604 
1605     sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev));
1606     if (!sec_bus) {
1607         return;
1608     }
1609 
1610     s_fdt.fdt = p->fdt;
1611     s_fdt.node_off = offset;
1612     s_fdt.sphb = p->sphb;
1613     pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
1614                         spapr_populate_pci_devices_dt,
1615                         &s_fdt);
1616 }
1617 
1618 static void spapr_phb_pci_enumerate_bridge(PCIBus *bus, PCIDevice *pdev,
1619                                            void *opaque)
1620 {
1621     unsigned int *bus_no = opaque;
1622     unsigned int primary = *bus_no;
1623     unsigned int subordinate = 0xff;
1624     PCIBus *sec_bus = NULL;
1625 
1626     if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) !=
1627          PCI_HEADER_TYPE_BRIDGE)) {
1628         return;
1629     }
1630 
1631     (*bus_no)++;
1632     pci_default_write_config(pdev, PCI_PRIMARY_BUS, primary, 1);
1633     pci_default_write_config(pdev, PCI_SECONDARY_BUS, *bus_no, 1);
1634     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
1635 
1636     sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev));
1637     if (!sec_bus) {
1638         return;
1639     }
1640 
1641     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, subordinate, 1);
1642     pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
1643                         spapr_phb_pci_enumerate_bridge, bus_no);
1644     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
1645 }
1646 
1647 static void spapr_phb_pci_enumerate(sPAPRPHBState *phb)
1648 {
1649     PCIBus *bus = PCI_HOST_BRIDGE(phb)->bus;
1650     unsigned int bus_no = 0;
1651 
1652     pci_for_each_device(bus, pci_bus_num(bus),
1653                         spapr_phb_pci_enumerate_bridge,
1654                         &bus_no);
1655 
1656 }
1657 
1658 int spapr_populate_pci_dt(sPAPRPHBState *phb,
1659                           uint32_t xics_phandle,
1660                           void *fdt)
1661 {
1662     int bus_off, i, j, ret;
1663     char nodename[FDT_NAME_MAX];
1664     uint32_t bus_range[] = { cpu_to_be32(0), cpu_to_be32(0xff) };
1665     const uint64_t mmiosize = memory_region_size(&phb->memwindow);
1666     const uint64_t w32max = (1ULL << 32) - SPAPR_PCI_MEM_WIN_BUS_OFFSET;
1667     const uint64_t w32size = MIN(w32max, mmiosize);
1668     const uint64_t w64size = (mmiosize > w32size) ? (mmiosize - w32size) : 0;
1669     struct {
1670         uint32_t hi;
1671         uint64_t child;
1672         uint64_t parent;
1673         uint64_t size;
1674     } QEMU_PACKED ranges[] = {
1675         {
1676             cpu_to_be32(b_ss(1)), cpu_to_be64(0),
1677             cpu_to_be64(phb->io_win_addr),
1678             cpu_to_be64(memory_region_size(&phb->iospace)),
1679         },
1680         {
1681             cpu_to_be32(b_ss(2)), cpu_to_be64(SPAPR_PCI_MEM_WIN_BUS_OFFSET),
1682             cpu_to_be64(phb->mem_win_addr),
1683             cpu_to_be64(w32size),
1684         },
1685         {
1686             cpu_to_be32(b_ss(3)), cpu_to_be64(1ULL << 32),
1687             cpu_to_be64(phb->mem_win_addr + w32size),
1688             cpu_to_be64(w64size)
1689         },
1690     };
1691     const unsigned sizeof_ranges = (w64size ? 3 : 2) * sizeof(ranges[0]);
1692     uint64_t bus_reg[] = { cpu_to_be64(phb->buid), 0 };
1693     uint32_t interrupt_map_mask[] = {
1694         cpu_to_be32(b_ddddd(-1)|b_fff(0)), 0x0, 0x0, cpu_to_be32(-1)};
1695     uint32_t interrupt_map[PCI_SLOT_MAX * PCI_NUM_PINS][7];
1696     sPAPRTCETable *tcet;
1697     PCIBus *bus = PCI_HOST_BRIDGE(phb)->bus;
1698     sPAPRFDT s_fdt;
1699 
1700     /* Start populating the FDT */
1701     snprintf(nodename, FDT_NAME_MAX, "pci@%" PRIx64, phb->buid);
1702     bus_off = fdt_add_subnode(fdt, 0, nodename);
1703     if (bus_off < 0) {
1704         return bus_off;
1705     }
1706 
1707     /* Write PHB properties */
1708     _FDT(fdt_setprop_string(fdt, bus_off, "device_type", "pci"));
1709     _FDT(fdt_setprop_string(fdt, bus_off, "compatible", "IBM,Logical_PHB"));
1710     _FDT(fdt_setprop_cell(fdt, bus_off, "#address-cells", 0x3));
1711     _FDT(fdt_setprop_cell(fdt, bus_off, "#size-cells", 0x2));
1712     _FDT(fdt_setprop_cell(fdt, bus_off, "#interrupt-cells", 0x1));
1713     _FDT(fdt_setprop(fdt, bus_off, "used-by-rtas", NULL, 0));
1714     _FDT(fdt_setprop(fdt, bus_off, "bus-range", &bus_range, sizeof(bus_range)));
1715     _FDT(fdt_setprop(fdt, bus_off, "ranges", &ranges, sizeof_ranges));
1716     _FDT(fdt_setprop(fdt, bus_off, "reg", &bus_reg, sizeof(bus_reg)));
1717     _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pci-config-space-type", 0x1));
1718     _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pe-total-#msi", XICS_IRQS));
1719 
1720     /* Build the interrupt-map, this must matches what is done
1721      * in pci_spapr_map_irq
1722      */
1723     _FDT(fdt_setprop(fdt, bus_off, "interrupt-map-mask",
1724                      &interrupt_map_mask, sizeof(interrupt_map_mask)));
1725     for (i = 0; i < PCI_SLOT_MAX; i++) {
1726         for (j = 0; j < PCI_NUM_PINS; j++) {
1727             uint32_t *irqmap = interrupt_map[i*PCI_NUM_PINS + j];
1728             int lsi_num = pci_spapr_swizzle(i, j);
1729 
1730             irqmap[0] = cpu_to_be32(b_ddddd(i)|b_fff(0));
1731             irqmap[1] = 0;
1732             irqmap[2] = 0;
1733             irqmap[3] = cpu_to_be32(j+1);
1734             irqmap[4] = cpu_to_be32(xics_phandle);
1735             irqmap[5] = cpu_to_be32(phb->lsi_table[lsi_num].irq);
1736             irqmap[6] = cpu_to_be32(0x8);
1737         }
1738     }
1739     /* Write interrupt map */
1740     _FDT(fdt_setprop(fdt, bus_off, "interrupt-map", &interrupt_map,
1741                      sizeof(interrupt_map)));
1742 
1743     tcet = spapr_tce_find_by_liobn(SPAPR_PCI_LIOBN(phb->index, 0));
1744     spapr_dma_dt(fdt, bus_off, "ibm,dma-window",
1745                  tcet->liobn, tcet->bus_offset,
1746                  tcet->nb_table << tcet->page_shift);
1747 
1748     /* Walk the bridges and program the bus numbers*/
1749     spapr_phb_pci_enumerate(phb);
1750     _FDT(fdt_setprop_cell(fdt, bus_off, "qemu,phb-enumerated", 0x1));
1751 
1752     /* Populate tree nodes with PCI devices attached */
1753     s_fdt.fdt = fdt;
1754     s_fdt.node_off = bus_off;
1755     s_fdt.sphb = phb;
1756     pci_for_each_device(bus, pci_bus_num(bus),
1757                         spapr_populate_pci_devices_dt,
1758                         &s_fdt);
1759 
1760     ret = spapr_drc_populate_dt(fdt, bus_off, OBJECT(phb),
1761                                 SPAPR_DR_CONNECTOR_TYPE_PCI);
1762     if (ret) {
1763         return ret;
1764     }
1765 
1766     return 0;
1767 }
1768 
1769 void spapr_pci_rtas_init(void)
1770 {
1771     spapr_rtas_register(RTAS_READ_PCI_CONFIG, "read-pci-config",
1772                         rtas_read_pci_config);
1773     spapr_rtas_register(RTAS_WRITE_PCI_CONFIG, "write-pci-config",
1774                         rtas_write_pci_config);
1775     spapr_rtas_register(RTAS_IBM_READ_PCI_CONFIG, "ibm,read-pci-config",
1776                         rtas_ibm_read_pci_config);
1777     spapr_rtas_register(RTAS_IBM_WRITE_PCI_CONFIG, "ibm,write-pci-config",
1778                         rtas_ibm_write_pci_config);
1779     if (msi_supported) {
1780         spapr_rtas_register(RTAS_IBM_QUERY_INTERRUPT_SOURCE_NUMBER,
1781                             "ibm,query-interrupt-source-number",
1782                             rtas_ibm_query_interrupt_source_number);
1783         spapr_rtas_register(RTAS_IBM_CHANGE_MSI, "ibm,change-msi",
1784                             rtas_ibm_change_msi);
1785     }
1786 
1787     spapr_rtas_register(RTAS_IBM_SET_EEH_OPTION,
1788                         "ibm,set-eeh-option",
1789                         rtas_ibm_set_eeh_option);
1790     spapr_rtas_register(RTAS_IBM_GET_CONFIG_ADDR_INFO2,
1791                         "ibm,get-config-addr-info2",
1792                         rtas_ibm_get_config_addr_info2);
1793     spapr_rtas_register(RTAS_IBM_READ_SLOT_RESET_STATE2,
1794                         "ibm,read-slot-reset-state2",
1795                         rtas_ibm_read_slot_reset_state2);
1796     spapr_rtas_register(RTAS_IBM_SET_SLOT_RESET,
1797                         "ibm,set-slot-reset",
1798                         rtas_ibm_set_slot_reset);
1799     spapr_rtas_register(RTAS_IBM_CONFIGURE_PE,
1800                         "ibm,configure-pe",
1801                         rtas_ibm_configure_pe);
1802     spapr_rtas_register(RTAS_IBM_SLOT_ERROR_DETAIL,
1803                         "ibm,slot-error-detail",
1804                         rtas_ibm_slot_error_detail);
1805 }
1806 
1807 static void spapr_pci_register_types(void)
1808 {
1809     type_register_static(&spapr_phb_info);
1810 }
1811 
1812 type_init(spapr_pci_register_types)
1813 
1814 static int spapr_switch_one_vga(DeviceState *dev, void *opaque)
1815 {
1816     bool be = *(bool *)opaque;
1817 
1818     if (object_dynamic_cast(OBJECT(dev), "VGA")
1819         || object_dynamic_cast(OBJECT(dev), "secondary-vga")) {
1820         object_property_set_bool(OBJECT(dev), be, "big-endian-framebuffer",
1821                                  &error_abort);
1822     }
1823     return 0;
1824 }
1825 
1826 void spapr_pci_switch_vga(bool big_endian)
1827 {
1828     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
1829     sPAPRPHBState *sphb;
1830 
1831     /*
1832      * For backward compatibility with existing guests, we switch
1833      * the endianness of the VGA controller when changing the guest
1834      * interrupt mode
1835      */
1836     QLIST_FOREACH(sphb, &spapr->phbs, list) {
1837         BusState *bus = &PCI_HOST_BRIDGE(sphb)->bus->qbus;
1838         qbus_walk_children(bus, spapr_switch_one_vga, NULL, NULL, NULL,
1839                            &big_endian);
1840     }
1841 }
1842