xref: /openbmc/qemu/hw/ppc/spapr_pci.c (revision 09a274d8)
1 /*
2  * QEMU sPAPR PCI host originated from Uninorth PCI host
3  *
4  * Copyright (c) 2011 Alexey Kardashevskiy, IBM Corporation.
5  * Copyright (C) 2011 David Gibson, IBM Corporation.
6  *
7  * Permission is hereby granted, free of charge, to any person obtaining a copy
8  * of this software and associated documentation files (the "Software"), to deal
9  * in the Software without restriction, including without limitation the rights
10  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11  * copies of the Software, and to permit persons to whom the Software is
12  * furnished to do so, subject to the following conditions:
13  *
14  * The above copyright notice and this permission notice shall be included in
15  * all copies or substantial portions of the Software.
16  *
17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
23  * THE SOFTWARE.
24  */
25 #include "qemu/osdep.h"
26 #include "qapi/error.h"
27 #include "qemu-common.h"
28 #include "cpu.h"
29 #include "hw/hw.h"
30 #include "hw/sysbus.h"
31 #include "hw/pci/pci.h"
32 #include "hw/pci/msi.h"
33 #include "hw/pci/msix.h"
34 #include "hw/pci/pci_host.h"
35 #include "hw/ppc/spapr.h"
36 #include "hw/pci-host/spapr.h"
37 #include "exec/address-spaces.h"
38 #include "exec/ram_addr.h"
39 #include <libfdt.h>
40 #include "trace.h"
41 #include "qemu/error-report.h"
42 #include "qapi/qmp/qerror.h"
43 #include "hw/ppc/fdt.h"
44 #include "hw/pci/pci_bridge.h"
45 #include "hw/pci/pci_bus.h"
46 #include "hw/pci/pci_ids.h"
47 #include "hw/ppc/spapr_drc.h"
48 #include "sysemu/device_tree.h"
49 #include "sysemu/kvm.h"
50 #include "sysemu/hostmem.h"
51 #include "sysemu/numa.h"
52 
53 /* Copied from the kernel arch/powerpc/platforms/pseries/msi.c */
54 #define RTAS_QUERY_FN           0
55 #define RTAS_CHANGE_FN          1
56 #define RTAS_RESET_FN           2
57 #define RTAS_CHANGE_MSI_FN      3
58 #define RTAS_CHANGE_MSIX_FN     4
59 
60 /* Interrupt types to return on RTAS_CHANGE_* */
61 #define RTAS_TYPE_MSI           1
62 #define RTAS_TYPE_MSIX          2
63 
64 sPAPRPHBState *spapr_pci_find_phb(sPAPRMachineState *spapr, uint64_t buid)
65 {
66     sPAPRPHBState *sphb;
67 
68     QLIST_FOREACH(sphb, &spapr->phbs, list) {
69         if (sphb->buid != buid) {
70             continue;
71         }
72         return sphb;
73     }
74 
75     return NULL;
76 }
77 
78 PCIDevice *spapr_pci_find_dev(sPAPRMachineState *spapr, uint64_t buid,
79                               uint32_t config_addr)
80 {
81     sPAPRPHBState *sphb = spapr_pci_find_phb(spapr, buid);
82     PCIHostState *phb = PCI_HOST_BRIDGE(sphb);
83     int bus_num = (config_addr >> 16) & 0xFF;
84     int devfn = (config_addr >> 8) & 0xFF;
85 
86     if (!phb) {
87         return NULL;
88     }
89 
90     return pci_find_device(phb->bus, bus_num, devfn);
91 }
92 
93 static uint32_t rtas_pci_cfgaddr(uint32_t arg)
94 {
95     /* This handles the encoding of extended config space addresses */
96     return ((arg >> 20) & 0xf00) | (arg & 0xff);
97 }
98 
99 static void finish_read_pci_config(sPAPRMachineState *spapr, uint64_t buid,
100                                    uint32_t addr, uint32_t size,
101                                    target_ulong rets)
102 {
103     PCIDevice *pci_dev;
104     uint32_t val;
105 
106     if ((size != 1) && (size != 2) && (size != 4)) {
107         /* access must be 1, 2 or 4 bytes */
108         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
109         return;
110     }
111 
112     pci_dev = spapr_pci_find_dev(spapr, buid, addr);
113     addr = rtas_pci_cfgaddr(addr);
114 
115     if (!pci_dev || (addr % size) || (addr >= pci_config_size(pci_dev))) {
116         /* Access must be to a valid device, within bounds and
117          * naturally aligned */
118         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
119         return;
120     }
121 
122     val = pci_host_config_read_common(pci_dev, addr,
123                                       pci_config_size(pci_dev), size);
124 
125     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
126     rtas_st(rets, 1, val);
127 }
128 
129 static void rtas_ibm_read_pci_config(PowerPCCPU *cpu, sPAPRMachineState *spapr,
130                                      uint32_t token, uint32_t nargs,
131                                      target_ulong args,
132                                      uint32_t nret, target_ulong rets)
133 {
134     uint64_t buid;
135     uint32_t size, addr;
136 
137     if ((nargs != 4) || (nret != 2)) {
138         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
139         return;
140     }
141 
142     buid = rtas_ldq(args, 1);
143     size = rtas_ld(args, 3);
144     addr = rtas_ld(args, 0);
145 
146     finish_read_pci_config(spapr, buid, addr, size, rets);
147 }
148 
149 static void rtas_read_pci_config(PowerPCCPU *cpu, sPAPRMachineState *spapr,
150                                  uint32_t token, uint32_t nargs,
151                                  target_ulong args,
152                                  uint32_t nret, target_ulong rets)
153 {
154     uint32_t size, addr;
155 
156     if ((nargs != 2) || (nret != 2)) {
157         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
158         return;
159     }
160 
161     size = rtas_ld(args, 1);
162     addr = rtas_ld(args, 0);
163 
164     finish_read_pci_config(spapr, 0, addr, size, rets);
165 }
166 
167 static void finish_write_pci_config(sPAPRMachineState *spapr, uint64_t buid,
168                                     uint32_t addr, uint32_t size,
169                                     uint32_t val, target_ulong rets)
170 {
171     PCIDevice *pci_dev;
172 
173     if ((size != 1) && (size != 2) && (size != 4)) {
174         /* access must be 1, 2 or 4 bytes */
175         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
176         return;
177     }
178 
179     pci_dev = spapr_pci_find_dev(spapr, buid, addr);
180     addr = rtas_pci_cfgaddr(addr);
181 
182     if (!pci_dev || (addr % size) || (addr >= pci_config_size(pci_dev))) {
183         /* Access must be to a valid device, within bounds and
184          * naturally aligned */
185         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
186         return;
187     }
188 
189     pci_host_config_write_common(pci_dev, addr, pci_config_size(pci_dev),
190                                  val, size);
191 
192     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
193 }
194 
195 static void rtas_ibm_write_pci_config(PowerPCCPU *cpu, sPAPRMachineState *spapr,
196                                       uint32_t token, uint32_t nargs,
197                                       target_ulong args,
198                                       uint32_t nret, target_ulong rets)
199 {
200     uint64_t buid;
201     uint32_t val, size, addr;
202 
203     if ((nargs != 5) || (nret != 1)) {
204         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
205         return;
206     }
207 
208     buid = rtas_ldq(args, 1);
209     val = rtas_ld(args, 4);
210     size = rtas_ld(args, 3);
211     addr = rtas_ld(args, 0);
212 
213     finish_write_pci_config(spapr, buid, addr, size, val, rets);
214 }
215 
216 static void rtas_write_pci_config(PowerPCCPU *cpu, sPAPRMachineState *spapr,
217                                   uint32_t token, uint32_t nargs,
218                                   target_ulong args,
219                                   uint32_t nret, target_ulong rets)
220 {
221     uint32_t val, size, addr;
222 
223     if ((nargs != 3) || (nret != 1)) {
224         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
225         return;
226     }
227 
228 
229     val = rtas_ld(args, 2);
230     size = rtas_ld(args, 1);
231     addr = rtas_ld(args, 0);
232 
233     finish_write_pci_config(spapr, 0, addr, size, val, rets);
234 }
235 
236 /*
237  * Set MSI/MSIX message data.
238  * This is required for msi_notify()/msix_notify() which
239  * will write at the addresses via spapr_msi_write().
240  *
241  * If hwaddr == 0, all entries will have .data == first_irq i.e.
242  * table will be reset.
243  */
244 static void spapr_msi_setmsg(PCIDevice *pdev, hwaddr addr, bool msix,
245                              unsigned first_irq, unsigned req_num)
246 {
247     unsigned i;
248     MSIMessage msg = { .address = addr, .data = first_irq };
249 
250     if (!msix) {
251         msi_set_message(pdev, msg);
252         trace_spapr_pci_msi_setup(pdev->name, 0, msg.address);
253         return;
254     }
255 
256     for (i = 0; i < req_num; ++i) {
257         msix_set_message(pdev, i, msg);
258         trace_spapr_pci_msi_setup(pdev->name, i, msg.address);
259         if (addr) {
260             ++msg.data;
261         }
262     }
263 }
264 
265 static void rtas_ibm_change_msi(PowerPCCPU *cpu, sPAPRMachineState *spapr,
266                                 uint32_t token, uint32_t nargs,
267                                 target_ulong args, uint32_t nret,
268                                 target_ulong rets)
269 {
270     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
271     uint32_t config_addr = rtas_ld(args, 0);
272     uint64_t buid = rtas_ldq(args, 1);
273     unsigned int func = rtas_ld(args, 3);
274     unsigned int req_num = rtas_ld(args, 4); /* 0 == remove all */
275     unsigned int seq_num = rtas_ld(args, 5);
276     unsigned int ret_intr_type;
277     unsigned int irq, max_irqs = 0;
278     sPAPRPHBState *phb = NULL;
279     PCIDevice *pdev = NULL;
280     spapr_pci_msi *msi;
281     int *config_addr_key;
282     Error *err = NULL;
283     int i;
284 
285     /* Fins sPAPRPHBState */
286     phb = spapr_pci_find_phb(spapr, buid);
287     if (phb) {
288         pdev = spapr_pci_find_dev(spapr, buid, config_addr);
289     }
290     if (!phb || !pdev) {
291         rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
292         return;
293     }
294 
295     switch (func) {
296     case RTAS_CHANGE_FN:
297         if (msi_present(pdev)) {
298             ret_intr_type = RTAS_TYPE_MSI;
299         } else if (msix_present(pdev)) {
300             ret_intr_type = RTAS_TYPE_MSIX;
301         } else {
302             rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
303             return;
304         }
305         break;
306     case RTAS_CHANGE_MSI_FN:
307         if (msi_present(pdev)) {
308             ret_intr_type = RTAS_TYPE_MSI;
309         } else {
310             rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
311             return;
312         }
313         break;
314     case RTAS_CHANGE_MSIX_FN:
315         if (msix_present(pdev)) {
316             ret_intr_type = RTAS_TYPE_MSIX;
317         } else {
318             rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
319             return;
320         }
321         break;
322     default:
323         error_report("rtas_ibm_change_msi(%u) is not implemented", func);
324         rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
325         return;
326     }
327 
328     msi = (spapr_pci_msi *) g_hash_table_lookup(phb->msi, &config_addr);
329 
330     /* Releasing MSIs */
331     if (!req_num) {
332         if (!msi) {
333             trace_spapr_pci_msi("Releasing wrong config", config_addr);
334             rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
335             return;
336         }
337 
338         if (!smc->legacy_irq_allocation) {
339             spapr_irq_msi_free(spapr, msi->first_irq, msi->num);
340         }
341         spapr_irq_free(spapr, msi->first_irq, msi->num);
342         if (msi_present(pdev)) {
343             spapr_msi_setmsg(pdev, 0, false, 0, 0);
344         }
345         if (msix_present(pdev)) {
346             spapr_msi_setmsg(pdev, 0, true, 0, 0);
347         }
348         g_hash_table_remove(phb->msi, &config_addr);
349 
350         trace_spapr_pci_msi("Released MSIs", config_addr);
351         rtas_st(rets, 0, RTAS_OUT_SUCCESS);
352         rtas_st(rets, 1, 0);
353         return;
354     }
355 
356     /* Enabling MSI */
357 
358     /* Check if the device supports as many IRQs as requested */
359     if (ret_intr_type == RTAS_TYPE_MSI) {
360         max_irqs = msi_nr_vectors_allocated(pdev);
361     } else if (ret_intr_type == RTAS_TYPE_MSIX) {
362         max_irqs = pdev->msix_entries_nr;
363     }
364     if (!max_irqs) {
365         error_report("Requested interrupt type %d is not enabled for device %x",
366                      ret_intr_type, config_addr);
367         rtas_st(rets, 0, -1); /* Hardware error */
368         return;
369     }
370     /* Correct the number if the guest asked for too many */
371     if (req_num > max_irqs) {
372         trace_spapr_pci_msi_retry(config_addr, req_num, max_irqs);
373         req_num = max_irqs;
374         irq = 0; /* to avoid misleading trace */
375         goto out;
376     }
377 
378     /* Allocate MSIs */
379     if (smc->legacy_irq_allocation) {
380         irq = spapr_irq_find(spapr, req_num, ret_intr_type == RTAS_TYPE_MSI,
381                              &err);
382     } else {
383         irq = spapr_irq_msi_alloc(spapr, req_num,
384                                   ret_intr_type == RTAS_TYPE_MSI, &err);
385     }
386     if (err) {
387         error_reportf_err(err, "Can't allocate MSIs for device %x: ",
388                           config_addr);
389         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
390         return;
391     }
392 
393     for (i = 0; i < req_num; i++) {
394         spapr_irq_claim(spapr, irq + i, false, &err);
395         if (err) {
396             if (i) {
397                 spapr_irq_free(spapr, irq, i);
398             }
399             if (!smc->legacy_irq_allocation) {
400                 spapr_irq_msi_free(spapr, irq, req_num);
401             }
402             error_reportf_err(err, "Can't allocate MSIs for device %x: ",
403                               config_addr);
404             rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
405             return;
406         }
407     }
408 
409     /* Release previous MSIs */
410     if (msi) {
411         if (!smc->legacy_irq_allocation) {
412             spapr_irq_msi_free(spapr, msi->first_irq, msi->num);
413         }
414         spapr_irq_free(spapr, msi->first_irq, msi->num);
415         g_hash_table_remove(phb->msi, &config_addr);
416     }
417 
418     /* Setup MSI/MSIX vectors in the device (via cfgspace or MSIX BAR) */
419     spapr_msi_setmsg(pdev, SPAPR_PCI_MSI_WINDOW, ret_intr_type == RTAS_TYPE_MSIX,
420                      irq, req_num);
421 
422     /* Add MSI device to cache */
423     msi = g_new(spapr_pci_msi, 1);
424     msi->first_irq = irq;
425     msi->num = req_num;
426     config_addr_key = g_new(int, 1);
427     *config_addr_key = config_addr;
428     g_hash_table_insert(phb->msi, config_addr_key, msi);
429 
430 out:
431     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
432     rtas_st(rets, 1, req_num);
433     rtas_st(rets, 2, ++seq_num);
434     if (nret > 3) {
435         rtas_st(rets, 3, ret_intr_type);
436     }
437 
438     trace_spapr_pci_rtas_ibm_change_msi(config_addr, func, req_num, irq);
439 }
440 
441 static void rtas_ibm_query_interrupt_source_number(PowerPCCPU *cpu,
442                                                    sPAPRMachineState *spapr,
443                                                    uint32_t token,
444                                                    uint32_t nargs,
445                                                    target_ulong args,
446                                                    uint32_t nret,
447                                                    target_ulong rets)
448 {
449     uint32_t config_addr = rtas_ld(args, 0);
450     uint64_t buid = rtas_ldq(args, 1);
451     unsigned int intr_src_num = -1, ioa_intr_num = rtas_ld(args, 3);
452     sPAPRPHBState *phb = NULL;
453     PCIDevice *pdev = NULL;
454     spapr_pci_msi *msi;
455 
456     /* Find sPAPRPHBState */
457     phb = spapr_pci_find_phb(spapr, buid);
458     if (phb) {
459         pdev = spapr_pci_find_dev(spapr, buid, config_addr);
460     }
461     if (!phb || !pdev) {
462         rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
463         return;
464     }
465 
466     /* Find device descriptor and start IRQ */
467     msi = (spapr_pci_msi *) g_hash_table_lookup(phb->msi, &config_addr);
468     if (!msi || !msi->first_irq || !msi->num || (ioa_intr_num >= msi->num)) {
469         trace_spapr_pci_msi("Failed to return vector", config_addr);
470         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
471         return;
472     }
473     intr_src_num = msi->first_irq + ioa_intr_num;
474     trace_spapr_pci_rtas_ibm_query_interrupt_source_number(ioa_intr_num,
475                                                            intr_src_num);
476 
477     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
478     rtas_st(rets, 1, intr_src_num);
479     rtas_st(rets, 2, 1);/* 0 == level; 1 == edge */
480 }
481 
482 static void rtas_ibm_set_eeh_option(PowerPCCPU *cpu,
483                                     sPAPRMachineState *spapr,
484                                     uint32_t token, uint32_t nargs,
485                                     target_ulong args, uint32_t nret,
486                                     target_ulong rets)
487 {
488     sPAPRPHBState *sphb;
489     uint32_t addr, option;
490     uint64_t buid;
491     int ret;
492 
493     if ((nargs != 4) || (nret != 1)) {
494         goto param_error_exit;
495     }
496 
497     buid = rtas_ldq(args, 1);
498     addr = rtas_ld(args, 0);
499     option = rtas_ld(args, 3);
500 
501     sphb = spapr_pci_find_phb(spapr, buid);
502     if (!sphb) {
503         goto param_error_exit;
504     }
505 
506     if (!spapr_phb_eeh_available(sphb)) {
507         goto param_error_exit;
508     }
509 
510     ret = spapr_phb_vfio_eeh_set_option(sphb, addr, option);
511     rtas_st(rets, 0, ret);
512     return;
513 
514 param_error_exit:
515     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
516 }
517 
518 static void rtas_ibm_get_config_addr_info2(PowerPCCPU *cpu,
519                                            sPAPRMachineState *spapr,
520                                            uint32_t token, uint32_t nargs,
521                                            target_ulong args, uint32_t nret,
522                                            target_ulong rets)
523 {
524     sPAPRPHBState *sphb;
525     PCIDevice *pdev;
526     uint32_t addr, option;
527     uint64_t buid;
528 
529     if ((nargs != 4) || (nret != 2)) {
530         goto param_error_exit;
531     }
532 
533     buid = rtas_ldq(args, 1);
534     sphb = spapr_pci_find_phb(spapr, buid);
535     if (!sphb) {
536         goto param_error_exit;
537     }
538 
539     if (!spapr_phb_eeh_available(sphb)) {
540         goto param_error_exit;
541     }
542 
543     /*
544      * We always have PE address of form "00BB0001". "BB"
545      * represents the bus number of PE's primary bus.
546      */
547     option = rtas_ld(args, 3);
548     switch (option) {
549     case RTAS_GET_PE_ADDR:
550         addr = rtas_ld(args, 0);
551         pdev = spapr_pci_find_dev(spapr, buid, addr);
552         if (!pdev) {
553             goto param_error_exit;
554         }
555 
556         rtas_st(rets, 1, (pci_bus_num(pci_get_bus(pdev)) << 16) + 1);
557         break;
558     case RTAS_GET_PE_MODE:
559         rtas_st(rets, 1, RTAS_PE_MODE_SHARED);
560         break;
561     default:
562         goto param_error_exit;
563     }
564 
565     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
566     return;
567 
568 param_error_exit:
569     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
570 }
571 
572 static void rtas_ibm_read_slot_reset_state2(PowerPCCPU *cpu,
573                                             sPAPRMachineState *spapr,
574                                             uint32_t token, uint32_t nargs,
575                                             target_ulong args, uint32_t nret,
576                                             target_ulong rets)
577 {
578     sPAPRPHBState *sphb;
579     uint64_t buid;
580     int state, ret;
581 
582     if ((nargs != 3) || (nret != 4 && nret != 5)) {
583         goto param_error_exit;
584     }
585 
586     buid = rtas_ldq(args, 1);
587     sphb = spapr_pci_find_phb(spapr, buid);
588     if (!sphb) {
589         goto param_error_exit;
590     }
591 
592     if (!spapr_phb_eeh_available(sphb)) {
593         goto param_error_exit;
594     }
595 
596     ret = spapr_phb_vfio_eeh_get_state(sphb, &state);
597     rtas_st(rets, 0, ret);
598     if (ret != RTAS_OUT_SUCCESS) {
599         return;
600     }
601 
602     rtas_st(rets, 1, state);
603     rtas_st(rets, 2, RTAS_EEH_SUPPORT);
604     rtas_st(rets, 3, RTAS_EEH_PE_UNAVAIL_INFO);
605     if (nret >= 5) {
606         rtas_st(rets, 4, RTAS_EEH_PE_RECOVER_INFO);
607     }
608     return;
609 
610 param_error_exit:
611     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
612 }
613 
614 static void rtas_ibm_set_slot_reset(PowerPCCPU *cpu,
615                                     sPAPRMachineState *spapr,
616                                     uint32_t token, uint32_t nargs,
617                                     target_ulong args, uint32_t nret,
618                                     target_ulong rets)
619 {
620     sPAPRPHBState *sphb;
621     uint32_t option;
622     uint64_t buid;
623     int ret;
624 
625     if ((nargs != 4) || (nret != 1)) {
626         goto param_error_exit;
627     }
628 
629     buid = rtas_ldq(args, 1);
630     option = rtas_ld(args, 3);
631     sphb = spapr_pci_find_phb(spapr, buid);
632     if (!sphb) {
633         goto param_error_exit;
634     }
635 
636     if (!spapr_phb_eeh_available(sphb)) {
637         goto param_error_exit;
638     }
639 
640     ret = spapr_phb_vfio_eeh_reset(sphb, option);
641     rtas_st(rets, 0, ret);
642     return;
643 
644 param_error_exit:
645     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
646 }
647 
648 static void rtas_ibm_configure_pe(PowerPCCPU *cpu,
649                                   sPAPRMachineState *spapr,
650                                   uint32_t token, uint32_t nargs,
651                                   target_ulong args, uint32_t nret,
652                                   target_ulong rets)
653 {
654     sPAPRPHBState *sphb;
655     uint64_t buid;
656     int ret;
657 
658     if ((nargs != 3) || (nret != 1)) {
659         goto param_error_exit;
660     }
661 
662     buid = rtas_ldq(args, 1);
663     sphb = spapr_pci_find_phb(spapr, buid);
664     if (!sphb) {
665         goto param_error_exit;
666     }
667 
668     if (!spapr_phb_eeh_available(sphb)) {
669         goto param_error_exit;
670     }
671 
672     ret = spapr_phb_vfio_eeh_configure(sphb);
673     rtas_st(rets, 0, ret);
674     return;
675 
676 param_error_exit:
677     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
678 }
679 
680 /* To support it later */
681 static void rtas_ibm_slot_error_detail(PowerPCCPU *cpu,
682                                        sPAPRMachineState *spapr,
683                                        uint32_t token, uint32_t nargs,
684                                        target_ulong args, uint32_t nret,
685                                        target_ulong rets)
686 {
687     sPAPRPHBState *sphb;
688     int option;
689     uint64_t buid;
690 
691     if ((nargs != 8) || (nret != 1)) {
692         goto param_error_exit;
693     }
694 
695     buid = rtas_ldq(args, 1);
696     sphb = spapr_pci_find_phb(spapr, buid);
697     if (!sphb) {
698         goto param_error_exit;
699     }
700 
701     if (!spapr_phb_eeh_available(sphb)) {
702         goto param_error_exit;
703     }
704 
705     option = rtas_ld(args, 7);
706     switch (option) {
707     case RTAS_SLOT_TEMP_ERR_LOG:
708     case RTAS_SLOT_PERM_ERR_LOG:
709         break;
710     default:
711         goto param_error_exit;
712     }
713 
714     /* We don't have error log yet */
715     rtas_st(rets, 0, RTAS_OUT_NO_ERRORS_FOUND);
716     return;
717 
718 param_error_exit:
719     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
720 }
721 
722 static int pci_spapr_swizzle(int slot, int pin)
723 {
724     return (slot + pin) % PCI_NUM_PINS;
725 }
726 
727 static int pci_spapr_map_irq(PCIDevice *pci_dev, int irq_num)
728 {
729     /*
730      * Here we need to convert pci_dev + irq_num to some unique value
731      * which is less than number of IRQs on the specific bus (4).  We
732      * use standard PCI swizzling, that is (slot number + pin number)
733      * % 4.
734      */
735     return pci_spapr_swizzle(PCI_SLOT(pci_dev->devfn), irq_num);
736 }
737 
738 static void pci_spapr_set_irq(void *opaque, int irq_num, int level)
739 {
740     /*
741      * Here we use the number returned by pci_spapr_map_irq to find a
742      * corresponding qemu_irq.
743      */
744     sPAPRPHBState *phb = opaque;
745 
746     trace_spapr_pci_lsi_set(phb->dtbusname, irq_num, phb->lsi_table[irq_num].irq);
747     qemu_set_irq(spapr_phb_lsi_qirq(phb, irq_num), level);
748 }
749 
750 static PCIINTxRoute spapr_route_intx_pin_to_irq(void *opaque, int pin)
751 {
752     sPAPRPHBState *sphb = SPAPR_PCI_HOST_BRIDGE(opaque);
753     PCIINTxRoute route;
754 
755     route.mode = PCI_INTX_ENABLED;
756     route.irq = sphb->lsi_table[pin].irq;
757 
758     return route;
759 }
760 
761 /*
762  * MSI/MSIX memory region implementation.
763  * The handler handles both MSI and MSIX.
764  * The vector number is encoded in least bits in data.
765  */
766 static void spapr_msi_write(void *opaque, hwaddr addr,
767                             uint64_t data, unsigned size)
768 {
769     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
770     uint32_t irq = data;
771 
772     trace_spapr_pci_msi_write(addr, data, irq);
773 
774     qemu_irq_pulse(spapr_qirq(spapr, irq));
775 }
776 
777 static const MemoryRegionOps spapr_msi_ops = {
778     /* There is no .read as the read result is undefined by PCI spec */
779     .read = NULL,
780     .write = spapr_msi_write,
781     .endianness = DEVICE_LITTLE_ENDIAN
782 };
783 
784 /*
785  * PHB PCI device
786  */
787 static AddressSpace *spapr_pci_dma_iommu(PCIBus *bus, void *opaque, int devfn)
788 {
789     sPAPRPHBState *phb = opaque;
790 
791     return &phb->iommu_as;
792 }
793 
794 static char *spapr_phb_vfio_get_loc_code(sPAPRPHBState *sphb,  PCIDevice *pdev)
795 {
796     char *path = NULL, *buf = NULL, *host = NULL;
797 
798     /* Get the PCI VFIO host id */
799     host = object_property_get_str(OBJECT(pdev), "host", NULL);
800     if (!host) {
801         goto err_out;
802     }
803 
804     /* Construct the path of the file that will give us the DT location */
805     path = g_strdup_printf("/sys/bus/pci/devices/%s/devspec", host);
806     g_free(host);
807     if (!g_file_get_contents(path, &buf, NULL, NULL)) {
808         goto err_out;
809     }
810     g_free(path);
811 
812     /* Construct and read from host device tree the loc-code */
813     path = g_strdup_printf("/proc/device-tree%s/ibm,loc-code", buf);
814     g_free(buf);
815     if (!g_file_get_contents(path, &buf, NULL, NULL)) {
816         goto err_out;
817     }
818     return buf;
819 
820 err_out:
821     g_free(path);
822     return NULL;
823 }
824 
825 static char *spapr_phb_get_loc_code(sPAPRPHBState *sphb, PCIDevice *pdev)
826 {
827     char *buf;
828     const char *devtype = "qemu";
829     uint32_t busnr = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(pdev))));
830 
831     if (object_dynamic_cast(OBJECT(pdev), "vfio-pci")) {
832         buf = spapr_phb_vfio_get_loc_code(sphb, pdev);
833         if (buf) {
834             return buf;
835         }
836         devtype = "vfio";
837     }
838     /*
839      * For emulated devices and VFIO-failure case, make up
840      * the loc-code.
841      */
842     buf = g_strdup_printf("%s_%s:%04x:%02x:%02x.%x",
843                           devtype, pdev->name, sphb->index, busnr,
844                           PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
845     return buf;
846 }
847 
848 /* Macros to operate with address in OF binding to PCI */
849 #define b_x(x, p, l)    (((x) & ((1<<(l))-1)) << (p))
850 #define b_n(x)          b_x((x), 31, 1) /* 0 if relocatable */
851 #define b_p(x)          b_x((x), 30, 1) /* 1 if prefetchable */
852 #define b_t(x)          b_x((x), 29, 1) /* 1 if the address is aliased */
853 #define b_ss(x)         b_x((x), 24, 2) /* the space code */
854 #define b_bbbbbbbb(x)   b_x((x), 16, 8) /* bus number */
855 #define b_ddddd(x)      b_x((x), 11, 5) /* device number */
856 #define b_fff(x)        b_x((x), 8, 3)  /* function number */
857 #define b_rrrrrrrr(x)   b_x((x), 0, 8)  /* register number */
858 
859 /* for 'reg'/'assigned-addresses' OF properties */
860 #define RESOURCE_CELLS_SIZE 2
861 #define RESOURCE_CELLS_ADDRESS 3
862 
863 typedef struct ResourceFields {
864     uint32_t phys_hi;
865     uint32_t phys_mid;
866     uint32_t phys_lo;
867     uint32_t size_hi;
868     uint32_t size_lo;
869 } QEMU_PACKED ResourceFields;
870 
871 typedef struct ResourceProps {
872     ResourceFields reg[8];
873     ResourceFields assigned[7];
874     uint32_t reg_len;
875     uint32_t assigned_len;
876 } ResourceProps;
877 
878 /* fill in the 'reg'/'assigned-resources' OF properties for
879  * a PCI device. 'reg' describes resource requirements for a
880  * device's IO/MEM regions, 'assigned-addresses' describes the
881  * actual resource assignments.
882  *
883  * the properties are arrays of ('phys-addr', 'size') pairs describing
884  * the addressable regions of the PCI device, where 'phys-addr' is a
885  * RESOURCE_CELLS_ADDRESS-tuple of 32-bit integers corresponding to
886  * (phys.hi, phys.mid, phys.lo), and 'size' is a
887  * RESOURCE_CELLS_SIZE-tuple corresponding to (size.hi, size.lo).
888  *
889  * phys.hi = 0xYYXXXXZZ, where:
890  *   0xYY = npt000ss
891  *          |||   |
892  *          |||   +-- space code
893  *          |||               |
894  *          |||               +  00 if configuration space
895  *          |||               +  01 if IO region,
896  *          |||               +  10 if 32-bit MEM region
897  *          |||               +  11 if 64-bit MEM region
898  *          |||
899  *          ||+------ for non-relocatable IO: 1 if aliased
900  *          ||        for relocatable IO: 1 if below 64KB
901  *          ||        for MEM: 1 if below 1MB
902  *          |+------- 1 if region is prefetchable
903  *          +-------- 1 if region is non-relocatable
904  *   0xXXXX = bbbbbbbb dddddfff, encoding bus, slot, and function
905  *            bits respectively
906  *   0xZZ = rrrrrrrr, the register number of the BAR corresponding
907  *          to the region
908  *
909  * phys.mid and phys.lo correspond respectively to the hi/lo portions
910  * of the actual address of the region.
911  *
912  * how the phys-addr/size values are used differ slightly between
913  * 'reg' and 'assigned-addresses' properties. namely, 'reg' has
914  * an additional description for the config space region of the
915  * device, and in the case of QEMU has n=0 and phys.mid=phys.lo=0
916  * to describe the region as relocatable, with an address-mapping
917  * that corresponds directly to the PHB's address space for the
918  * resource. 'assigned-addresses' always has n=1 set with an absolute
919  * address assigned for the resource. in general, 'assigned-addresses'
920  * won't be populated, since addresses for PCI devices are generally
921  * unmapped initially and left to the guest to assign.
922  *
923  * note also that addresses defined in these properties are, at least
924  * for PAPR guests, relative to the PHBs IO/MEM windows, and
925  * correspond directly to the addresses in the BARs.
926  *
927  * in accordance with PCI Bus Binding to Open Firmware,
928  * IEEE Std 1275-1994, section 4.1.1, as implemented by PAPR+ v2.7,
929  * Appendix C.
930  */
931 static void populate_resource_props(PCIDevice *d, ResourceProps *rp)
932 {
933     int bus_num = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(d))));
934     uint32_t dev_id = (b_bbbbbbbb(bus_num) |
935                        b_ddddd(PCI_SLOT(d->devfn)) |
936                        b_fff(PCI_FUNC(d->devfn)));
937     ResourceFields *reg, *assigned;
938     int i, reg_idx = 0, assigned_idx = 0;
939 
940     /* config space region */
941     reg = &rp->reg[reg_idx++];
942     reg->phys_hi = cpu_to_be32(dev_id);
943     reg->phys_mid = 0;
944     reg->phys_lo = 0;
945     reg->size_hi = 0;
946     reg->size_lo = 0;
947 
948     for (i = 0; i < PCI_NUM_REGIONS; i++) {
949         if (!d->io_regions[i].size) {
950             continue;
951         }
952 
953         reg = &rp->reg[reg_idx++];
954 
955         reg->phys_hi = cpu_to_be32(dev_id | b_rrrrrrrr(pci_bar(d, i)));
956         if (d->io_regions[i].type & PCI_BASE_ADDRESS_SPACE_IO) {
957             reg->phys_hi |= cpu_to_be32(b_ss(1));
958         } else if (d->io_regions[i].type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
959             reg->phys_hi |= cpu_to_be32(b_ss(3));
960         } else {
961             reg->phys_hi |= cpu_to_be32(b_ss(2));
962         }
963         reg->phys_mid = 0;
964         reg->phys_lo = 0;
965         reg->size_hi = cpu_to_be32(d->io_regions[i].size >> 32);
966         reg->size_lo = cpu_to_be32(d->io_regions[i].size);
967 
968         if (d->io_regions[i].addr == PCI_BAR_UNMAPPED) {
969             continue;
970         }
971 
972         assigned = &rp->assigned[assigned_idx++];
973         assigned->phys_hi = cpu_to_be32(be32_to_cpu(reg->phys_hi) | b_n(1));
974         assigned->phys_mid = cpu_to_be32(d->io_regions[i].addr >> 32);
975         assigned->phys_lo = cpu_to_be32(d->io_regions[i].addr);
976         assigned->size_hi = reg->size_hi;
977         assigned->size_lo = reg->size_lo;
978     }
979 
980     rp->reg_len = reg_idx * sizeof(ResourceFields);
981     rp->assigned_len = assigned_idx * sizeof(ResourceFields);
982 }
983 
984 typedef struct PCIClass PCIClass;
985 typedef struct PCISubClass PCISubClass;
986 typedef struct PCIIFace PCIIFace;
987 
988 struct PCIIFace {
989     int iface;
990     const char *name;
991 };
992 
993 struct PCISubClass {
994     int subclass;
995     const char *name;
996     const PCIIFace *iface;
997 };
998 
999 struct PCIClass {
1000     const char *name;
1001     const PCISubClass *subc;
1002 };
1003 
1004 static const PCISubClass undef_subclass[] = {
1005     { PCI_CLASS_NOT_DEFINED_VGA, "display", NULL },
1006     { 0xFF, NULL, NULL },
1007 };
1008 
1009 static const PCISubClass mass_subclass[] = {
1010     { PCI_CLASS_STORAGE_SCSI, "scsi", NULL },
1011     { PCI_CLASS_STORAGE_IDE, "ide", NULL },
1012     { PCI_CLASS_STORAGE_FLOPPY, "fdc", NULL },
1013     { PCI_CLASS_STORAGE_IPI, "ipi", NULL },
1014     { PCI_CLASS_STORAGE_RAID, "raid", NULL },
1015     { PCI_CLASS_STORAGE_ATA, "ata", NULL },
1016     { PCI_CLASS_STORAGE_SATA, "sata", NULL },
1017     { PCI_CLASS_STORAGE_SAS, "sas", NULL },
1018     { 0xFF, NULL, NULL },
1019 };
1020 
1021 static const PCISubClass net_subclass[] = {
1022     { PCI_CLASS_NETWORK_ETHERNET, "ethernet", NULL },
1023     { PCI_CLASS_NETWORK_TOKEN_RING, "token-ring", NULL },
1024     { PCI_CLASS_NETWORK_FDDI, "fddi", NULL },
1025     { PCI_CLASS_NETWORK_ATM, "atm", NULL },
1026     { PCI_CLASS_NETWORK_ISDN, "isdn", NULL },
1027     { PCI_CLASS_NETWORK_WORLDFIP, "worldfip", NULL },
1028     { PCI_CLASS_NETWORK_PICMG214, "picmg", NULL },
1029     { 0xFF, NULL, NULL },
1030 };
1031 
1032 static const PCISubClass displ_subclass[] = {
1033     { PCI_CLASS_DISPLAY_VGA, "vga", NULL },
1034     { PCI_CLASS_DISPLAY_XGA, "xga", NULL },
1035     { PCI_CLASS_DISPLAY_3D, "3d-controller", NULL },
1036     { 0xFF, NULL, NULL },
1037 };
1038 
1039 static const PCISubClass media_subclass[] = {
1040     { PCI_CLASS_MULTIMEDIA_VIDEO, "video", NULL },
1041     { PCI_CLASS_MULTIMEDIA_AUDIO, "sound", NULL },
1042     { PCI_CLASS_MULTIMEDIA_PHONE, "telephony", NULL },
1043     { 0xFF, NULL, NULL },
1044 };
1045 
1046 static const PCISubClass mem_subclass[] = {
1047     { PCI_CLASS_MEMORY_RAM, "memory", NULL },
1048     { PCI_CLASS_MEMORY_FLASH, "flash", NULL },
1049     { 0xFF, NULL, NULL },
1050 };
1051 
1052 static const PCISubClass bridg_subclass[] = {
1053     { PCI_CLASS_BRIDGE_HOST, "host", NULL },
1054     { PCI_CLASS_BRIDGE_ISA, "isa", NULL },
1055     { PCI_CLASS_BRIDGE_EISA, "eisa", NULL },
1056     { PCI_CLASS_BRIDGE_MC, "mca", NULL },
1057     { PCI_CLASS_BRIDGE_PCI, "pci", NULL },
1058     { PCI_CLASS_BRIDGE_PCMCIA, "pcmcia", NULL },
1059     { PCI_CLASS_BRIDGE_NUBUS, "nubus", NULL },
1060     { PCI_CLASS_BRIDGE_CARDBUS, "cardbus", NULL },
1061     { PCI_CLASS_BRIDGE_RACEWAY, "raceway", NULL },
1062     { PCI_CLASS_BRIDGE_PCI_SEMITP, "semi-transparent-pci", NULL },
1063     { PCI_CLASS_BRIDGE_IB_PCI, "infiniband", NULL },
1064     { 0xFF, NULL, NULL },
1065 };
1066 
1067 static const PCISubClass comm_subclass[] = {
1068     { PCI_CLASS_COMMUNICATION_SERIAL, "serial", NULL },
1069     { PCI_CLASS_COMMUNICATION_PARALLEL, "parallel", NULL },
1070     { PCI_CLASS_COMMUNICATION_MULTISERIAL, "multiport-serial", NULL },
1071     { PCI_CLASS_COMMUNICATION_MODEM, "modem", NULL },
1072     { PCI_CLASS_COMMUNICATION_GPIB, "gpib", NULL },
1073     { PCI_CLASS_COMMUNICATION_SC, "smart-card", NULL },
1074     { 0xFF, NULL, NULL, },
1075 };
1076 
1077 static const PCIIFace pic_iface[] = {
1078     { PCI_CLASS_SYSTEM_PIC_IOAPIC, "io-apic" },
1079     { PCI_CLASS_SYSTEM_PIC_IOXAPIC, "io-xapic" },
1080     { 0xFF, NULL },
1081 };
1082 
1083 static const PCISubClass sys_subclass[] = {
1084     { PCI_CLASS_SYSTEM_PIC, "interrupt-controller", pic_iface },
1085     { PCI_CLASS_SYSTEM_DMA, "dma-controller", NULL },
1086     { PCI_CLASS_SYSTEM_TIMER, "timer", NULL },
1087     { PCI_CLASS_SYSTEM_RTC, "rtc", NULL },
1088     { PCI_CLASS_SYSTEM_PCI_HOTPLUG, "hot-plug-controller", NULL },
1089     { PCI_CLASS_SYSTEM_SDHCI, "sd-host-controller", NULL },
1090     { 0xFF, NULL, NULL },
1091 };
1092 
1093 static const PCISubClass inp_subclass[] = {
1094     { PCI_CLASS_INPUT_KEYBOARD, "keyboard", NULL },
1095     { PCI_CLASS_INPUT_PEN, "pen", NULL },
1096     { PCI_CLASS_INPUT_MOUSE, "mouse", NULL },
1097     { PCI_CLASS_INPUT_SCANNER, "scanner", NULL },
1098     { PCI_CLASS_INPUT_GAMEPORT, "gameport", NULL },
1099     { 0xFF, NULL, NULL },
1100 };
1101 
1102 static const PCISubClass dock_subclass[] = {
1103     { PCI_CLASS_DOCKING_GENERIC, "dock", NULL },
1104     { 0xFF, NULL, NULL },
1105 };
1106 
1107 static const PCISubClass cpu_subclass[] = {
1108     { PCI_CLASS_PROCESSOR_PENTIUM, "pentium", NULL },
1109     { PCI_CLASS_PROCESSOR_POWERPC, "powerpc", NULL },
1110     { PCI_CLASS_PROCESSOR_MIPS, "mips", NULL },
1111     { PCI_CLASS_PROCESSOR_CO, "co-processor", NULL },
1112     { 0xFF, NULL, NULL },
1113 };
1114 
1115 static const PCIIFace usb_iface[] = {
1116     { PCI_CLASS_SERIAL_USB_UHCI, "usb-uhci" },
1117     { PCI_CLASS_SERIAL_USB_OHCI, "usb-ohci", },
1118     { PCI_CLASS_SERIAL_USB_EHCI, "usb-ehci" },
1119     { PCI_CLASS_SERIAL_USB_XHCI, "usb-xhci" },
1120     { PCI_CLASS_SERIAL_USB_UNKNOWN, "usb-unknown" },
1121     { PCI_CLASS_SERIAL_USB_DEVICE, "usb-device" },
1122     { 0xFF, NULL },
1123 };
1124 
1125 static const PCISubClass ser_subclass[] = {
1126     { PCI_CLASS_SERIAL_FIREWIRE, "firewire", NULL },
1127     { PCI_CLASS_SERIAL_ACCESS, "access-bus", NULL },
1128     { PCI_CLASS_SERIAL_SSA, "ssa", NULL },
1129     { PCI_CLASS_SERIAL_USB, "usb", usb_iface },
1130     { PCI_CLASS_SERIAL_FIBER, "fibre-channel", NULL },
1131     { PCI_CLASS_SERIAL_SMBUS, "smb", NULL },
1132     { PCI_CLASS_SERIAL_IB, "infiniband", NULL },
1133     { PCI_CLASS_SERIAL_IPMI, "ipmi", NULL },
1134     { PCI_CLASS_SERIAL_SERCOS, "sercos", NULL },
1135     { PCI_CLASS_SERIAL_CANBUS, "canbus", NULL },
1136     { 0xFF, NULL, NULL },
1137 };
1138 
1139 static const PCISubClass wrl_subclass[] = {
1140     { PCI_CLASS_WIRELESS_IRDA, "irda", NULL },
1141     { PCI_CLASS_WIRELESS_CIR, "consumer-ir", NULL },
1142     { PCI_CLASS_WIRELESS_RF_CONTROLLER, "rf-controller", NULL },
1143     { PCI_CLASS_WIRELESS_BLUETOOTH, "bluetooth", NULL },
1144     { PCI_CLASS_WIRELESS_BROADBAND, "broadband", NULL },
1145     { 0xFF, NULL, NULL },
1146 };
1147 
1148 static const PCISubClass sat_subclass[] = {
1149     { PCI_CLASS_SATELLITE_TV, "satellite-tv", NULL },
1150     { PCI_CLASS_SATELLITE_AUDIO, "satellite-audio", NULL },
1151     { PCI_CLASS_SATELLITE_VOICE, "satellite-voice", NULL },
1152     { PCI_CLASS_SATELLITE_DATA, "satellite-data", NULL },
1153     { 0xFF, NULL, NULL },
1154 };
1155 
1156 static const PCISubClass crypt_subclass[] = {
1157     { PCI_CLASS_CRYPT_NETWORK, "network-encryption", NULL },
1158     { PCI_CLASS_CRYPT_ENTERTAINMENT,
1159       "entertainment-encryption", NULL },
1160     { 0xFF, NULL, NULL },
1161 };
1162 
1163 static const PCISubClass spc_subclass[] = {
1164     { PCI_CLASS_SP_DPIO, "dpio", NULL },
1165     { PCI_CLASS_SP_PERF, "counter", NULL },
1166     { PCI_CLASS_SP_SYNCH, "measurement", NULL },
1167     { PCI_CLASS_SP_MANAGEMENT, "management-card", NULL },
1168     { 0xFF, NULL, NULL },
1169 };
1170 
1171 static const PCIClass pci_classes[] = {
1172     { "legacy-device", undef_subclass },
1173     { "mass-storage",  mass_subclass },
1174     { "network", net_subclass },
1175     { "display", displ_subclass, },
1176     { "multimedia-device", media_subclass },
1177     { "memory-controller", mem_subclass },
1178     { "unknown-bridge", bridg_subclass },
1179     { "communication-controller", comm_subclass},
1180     { "system-peripheral", sys_subclass },
1181     { "input-controller", inp_subclass },
1182     { "docking-station", dock_subclass },
1183     { "cpu", cpu_subclass },
1184     { "serial-bus", ser_subclass },
1185     { "wireless-controller", wrl_subclass },
1186     { "intelligent-io", NULL },
1187     { "satellite-device", sat_subclass },
1188     { "encryption", crypt_subclass },
1189     { "data-processing-controller", spc_subclass },
1190 };
1191 
1192 static const char *pci_find_device_name(uint8_t class, uint8_t subclass,
1193                                         uint8_t iface)
1194 {
1195     const PCIClass *pclass;
1196     const PCISubClass *psubclass;
1197     const PCIIFace *piface;
1198     const char *name;
1199 
1200     if (class >= ARRAY_SIZE(pci_classes)) {
1201         return "pci";
1202     }
1203 
1204     pclass = pci_classes + class;
1205     name = pclass->name;
1206 
1207     if (pclass->subc == NULL) {
1208         return name;
1209     }
1210 
1211     psubclass = pclass->subc;
1212     while ((psubclass->subclass & 0xff) != 0xff) {
1213         if ((psubclass->subclass & 0xff) == subclass) {
1214             name = psubclass->name;
1215             break;
1216         }
1217         psubclass++;
1218     }
1219 
1220     piface = psubclass->iface;
1221     if (piface == NULL) {
1222         return name;
1223     }
1224     while ((piface->iface & 0xff) != 0xff) {
1225         if ((piface->iface & 0xff) == iface) {
1226             name = piface->name;
1227             break;
1228         }
1229         piface++;
1230     }
1231 
1232     return name;
1233 }
1234 
1235 static gchar *pci_get_node_name(PCIDevice *dev)
1236 {
1237     int slot = PCI_SLOT(dev->devfn);
1238     int func = PCI_FUNC(dev->devfn);
1239     uint32_t ccode = pci_default_read_config(dev, PCI_CLASS_PROG, 3);
1240     const char *name;
1241 
1242     name = pci_find_device_name((ccode >> 16) & 0xff, (ccode >> 8) & 0xff,
1243                                 ccode & 0xff);
1244 
1245     if (func != 0) {
1246         return g_strdup_printf("%s@%x,%x", name, slot, func);
1247     } else {
1248         return g_strdup_printf("%s@%x", name, slot);
1249     }
1250 }
1251 
1252 static uint32_t spapr_phb_get_pci_drc_index(sPAPRPHBState *phb,
1253                                             PCIDevice *pdev);
1254 
1255 static void spapr_populate_pci_child_dt(PCIDevice *dev, void *fdt, int offset,
1256                                        sPAPRPHBState *sphb)
1257 {
1258     ResourceProps rp;
1259     bool is_bridge = false;
1260     int pci_status;
1261     char *buf = NULL;
1262     uint32_t drc_index = spapr_phb_get_pci_drc_index(sphb, dev);
1263     uint32_t ccode = pci_default_read_config(dev, PCI_CLASS_PROG, 3);
1264     uint32_t max_msi, max_msix;
1265 
1266     if (pci_default_read_config(dev, PCI_HEADER_TYPE, 1) ==
1267         PCI_HEADER_TYPE_BRIDGE) {
1268         is_bridge = true;
1269     }
1270 
1271     /* in accordance with PAPR+ v2.7 13.6.3, Table 181 */
1272     _FDT(fdt_setprop_cell(fdt, offset, "vendor-id",
1273                           pci_default_read_config(dev, PCI_VENDOR_ID, 2)));
1274     _FDT(fdt_setprop_cell(fdt, offset, "device-id",
1275                           pci_default_read_config(dev, PCI_DEVICE_ID, 2)));
1276     _FDT(fdt_setprop_cell(fdt, offset, "revision-id",
1277                           pci_default_read_config(dev, PCI_REVISION_ID, 1)));
1278     _FDT(fdt_setprop_cell(fdt, offset, "class-code", ccode));
1279     if (pci_default_read_config(dev, PCI_INTERRUPT_PIN, 1)) {
1280         _FDT(fdt_setprop_cell(fdt, offset, "interrupts",
1281                  pci_default_read_config(dev, PCI_INTERRUPT_PIN, 1)));
1282     }
1283 
1284     if (!is_bridge) {
1285         _FDT(fdt_setprop_cell(fdt, offset, "min-grant",
1286             pci_default_read_config(dev, PCI_MIN_GNT, 1)));
1287         _FDT(fdt_setprop_cell(fdt, offset, "max-latency",
1288             pci_default_read_config(dev, PCI_MAX_LAT, 1)));
1289     }
1290 
1291     if (pci_default_read_config(dev, PCI_SUBSYSTEM_ID, 2)) {
1292         _FDT(fdt_setprop_cell(fdt, offset, "subsystem-id",
1293                  pci_default_read_config(dev, PCI_SUBSYSTEM_ID, 2)));
1294     }
1295 
1296     if (pci_default_read_config(dev, PCI_SUBSYSTEM_VENDOR_ID, 2)) {
1297         _FDT(fdt_setprop_cell(fdt, offset, "subsystem-vendor-id",
1298                  pci_default_read_config(dev, PCI_SUBSYSTEM_VENDOR_ID, 2)));
1299     }
1300 
1301     _FDT(fdt_setprop_cell(fdt, offset, "cache-line-size",
1302         pci_default_read_config(dev, PCI_CACHE_LINE_SIZE, 1)));
1303 
1304     /* the following fdt cells are masked off the pci status register */
1305     pci_status = pci_default_read_config(dev, PCI_STATUS, 2);
1306     _FDT(fdt_setprop_cell(fdt, offset, "devsel-speed",
1307                           PCI_STATUS_DEVSEL_MASK & pci_status));
1308 
1309     if (pci_status & PCI_STATUS_FAST_BACK) {
1310         _FDT(fdt_setprop(fdt, offset, "fast-back-to-back", NULL, 0));
1311     }
1312     if (pci_status & PCI_STATUS_66MHZ) {
1313         _FDT(fdt_setprop(fdt, offset, "66mhz-capable", NULL, 0));
1314     }
1315     if (pci_status & PCI_STATUS_UDF) {
1316         _FDT(fdt_setprop(fdt, offset, "udf-supported", NULL, 0));
1317     }
1318 
1319     _FDT(fdt_setprop_string(fdt, offset, "name",
1320                             pci_find_device_name((ccode >> 16) & 0xff,
1321                                                  (ccode >> 8) & 0xff,
1322                                                  ccode & 0xff)));
1323 
1324     buf = spapr_phb_get_loc_code(sphb, dev);
1325     _FDT(fdt_setprop_string(fdt, offset, "ibm,loc-code", buf));
1326     g_free(buf);
1327 
1328     if (drc_index) {
1329         _FDT(fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index));
1330     }
1331 
1332     _FDT(fdt_setprop_cell(fdt, offset, "#address-cells",
1333                           RESOURCE_CELLS_ADDRESS));
1334     _FDT(fdt_setprop_cell(fdt, offset, "#size-cells",
1335                           RESOURCE_CELLS_SIZE));
1336 
1337     if (msi_present(dev)) {
1338         max_msi = msi_nr_vectors_allocated(dev);
1339         if (max_msi) {
1340             _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi", max_msi));
1341         }
1342     }
1343     if (msix_present(dev)) {
1344         max_msix = dev->msix_entries_nr;
1345         if (max_msix) {
1346             _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi-x", max_msix));
1347         }
1348     }
1349 
1350     populate_resource_props(dev, &rp);
1351     _FDT(fdt_setprop(fdt, offset, "reg", (uint8_t *)rp.reg, rp.reg_len));
1352     _FDT(fdt_setprop(fdt, offset, "assigned-addresses",
1353                      (uint8_t *)rp.assigned, rp.assigned_len));
1354 
1355     if (sphb->pcie_ecs && pci_is_express(dev)) {
1356         _FDT(fdt_setprop_cell(fdt, offset, "ibm,pci-config-space-type", 0x1));
1357     }
1358 }
1359 
1360 /* create OF node for pci device and required OF DT properties */
1361 static int spapr_create_pci_child_dt(sPAPRPHBState *phb, PCIDevice *dev,
1362                                      void *fdt, int node_offset)
1363 {
1364     int offset;
1365     gchar *nodename;
1366 
1367     nodename = pci_get_node_name(dev);
1368     _FDT(offset = fdt_add_subnode(fdt, node_offset, nodename));
1369     g_free(nodename);
1370 
1371     spapr_populate_pci_child_dt(dev, fdt, offset, phb);
1372 
1373     return offset;
1374 }
1375 
1376 /* Callback to be called during DRC release. */
1377 void spapr_phb_remove_pci_device_cb(DeviceState *dev)
1378 {
1379     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
1380 
1381     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
1382 }
1383 
1384 static sPAPRDRConnector *spapr_phb_get_pci_func_drc(sPAPRPHBState *phb,
1385                                                     uint32_t busnr,
1386                                                     int32_t devfn)
1387 {
1388     return spapr_drc_by_id(TYPE_SPAPR_DRC_PCI,
1389                            (phb->index << 16) | (busnr << 8) | devfn);
1390 }
1391 
1392 static sPAPRDRConnector *spapr_phb_get_pci_drc(sPAPRPHBState *phb,
1393                                                PCIDevice *pdev)
1394 {
1395     uint32_t busnr = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(pdev))));
1396     return spapr_phb_get_pci_func_drc(phb, busnr, pdev->devfn);
1397 }
1398 
1399 static uint32_t spapr_phb_get_pci_drc_index(sPAPRPHBState *phb,
1400                                             PCIDevice *pdev)
1401 {
1402     sPAPRDRConnector *drc = spapr_phb_get_pci_drc(phb, pdev);
1403 
1404     if (!drc) {
1405         return 0;
1406     }
1407 
1408     return spapr_drc_index(drc);
1409 }
1410 
1411 static void spapr_pci_plug(HotplugHandler *plug_handler,
1412                            DeviceState *plugged_dev, Error **errp)
1413 {
1414     sPAPRPHBState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1415     PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1416     sPAPRDRConnector *drc = spapr_phb_get_pci_drc(phb, pdev);
1417     Error *local_err = NULL;
1418     PCIBus *bus = PCI_BUS(qdev_get_parent_bus(DEVICE(pdev)));
1419     uint32_t slotnr = PCI_SLOT(pdev->devfn);
1420     void *fdt = NULL;
1421     int fdt_start_offset, fdt_size;
1422 
1423     /* if DR is disabled we don't need to do anything in the case of
1424      * hotplug or coldplug callbacks
1425      */
1426     if (!phb->dr_enabled) {
1427         /* if this is a hotplug operation initiated by the user
1428          * we need to let them know it's not enabled
1429          */
1430         if (plugged_dev->hotplugged) {
1431             error_setg(&local_err, QERR_BUS_NO_HOTPLUG,
1432                        object_get_typename(OBJECT(phb)));
1433         }
1434         goto out;
1435     }
1436 
1437     g_assert(drc);
1438 
1439     /* Following the QEMU convention used for PCIe multifunction
1440      * hotplug, we do not allow functions to be hotplugged to a
1441      * slot that already has function 0 present
1442      */
1443     if (plugged_dev->hotplugged && bus->devices[PCI_DEVFN(slotnr, 0)] &&
1444         PCI_FUNC(pdev->devfn) != 0) {
1445         error_setg(&local_err, "PCI: slot %d function 0 already ocuppied by %s,"
1446                    " additional functions can no longer be exposed to guest.",
1447                    slotnr, bus->devices[PCI_DEVFN(slotnr, 0)]->name);
1448         goto out;
1449     }
1450 
1451     fdt = create_device_tree(&fdt_size);
1452     fdt_start_offset = spapr_create_pci_child_dt(phb, pdev, fdt, 0);
1453 
1454     spapr_drc_attach(drc, DEVICE(pdev), fdt, fdt_start_offset, &local_err);
1455     if (local_err) {
1456         goto out;
1457     }
1458 
1459     /* If this is function 0, signal hotplug for all the device functions.
1460      * Otherwise defer sending the hotplug event.
1461      */
1462     if (!spapr_drc_hotplugged(plugged_dev)) {
1463         spapr_drc_reset(drc);
1464     } else if (PCI_FUNC(pdev->devfn) == 0) {
1465         int i;
1466 
1467         for (i = 0; i < 8; i++) {
1468             sPAPRDRConnector *func_drc;
1469             sPAPRDRConnectorClass *func_drck;
1470             sPAPRDREntitySense state;
1471 
1472             func_drc = spapr_phb_get_pci_func_drc(phb, pci_bus_num(bus),
1473                                                   PCI_DEVFN(slotnr, i));
1474             func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1475             state = func_drck->dr_entity_sense(func_drc);
1476 
1477             if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) {
1478                 spapr_hotplug_req_add_by_index(func_drc);
1479             }
1480         }
1481     }
1482 
1483 out:
1484     if (local_err) {
1485         error_propagate(errp, local_err);
1486         g_free(fdt);
1487     }
1488 }
1489 
1490 static void spapr_pci_unplug(HotplugHandler *plug_handler,
1491                              DeviceState *plugged_dev, Error **errp)
1492 {
1493     /* some version guests do not wait for completion of a device
1494      * cleanup (generally done asynchronously by the kernel) before
1495      * signaling to QEMU that the device is safe, but instead sleep
1496      * for some 'safe' period of time. unfortunately on a busy host
1497      * this sleep isn't guaranteed to be long enough, resulting in
1498      * bad things like IRQ lines being left asserted during final
1499      * device removal. to deal with this we call reset just prior
1500      * to finalizing the device, which will put the device back into
1501      * an 'idle' state, as the device cleanup code expects.
1502      */
1503     pci_device_reset(PCI_DEVICE(plugged_dev));
1504     object_unparent(OBJECT(plugged_dev));
1505 }
1506 
1507 static void spapr_pci_unplug_request(HotplugHandler *plug_handler,
1508                                      DeviceState *plugged_dev, Error **errp)
1509 {
1510     sPAPRPHBState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1511     PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1512     sPAPRDRConnector *drc = spapr_phb_get_pci_drc(phb, pdev);
1513 
1514     if (!phb->dr_enabled) {
1515         error_setg(errp, QERR_BUS_NO_HOTPLUG,
1516                    object_get_typename(OBJECT(phb)));
1517         return;
1518     }
1519 
1520     g_assert(drc);
1521     g_assert(drc->dev == plugged_dev);
1522 
1523     if (!spapr_drc_unplug_requested(drc)) {
1524         PCIBus *bus = PCI_BUS(qdev_get_parent_bus(DEVICE(pdev)));
1525         uint32_t slotnr = PCI_SLOT(pdev->devfn);
1526         sPAPRDRConnector *func_drc;
1527         sPAPRDRConnectorClass *func_drck;
1528         sPAPRDREntitySense state;
1529         int i;
1530 
1531         /* ensure any other present functions are pending unplug */
1532         if (PCI_FUNC(pdev->devfn) == 0) {
1533             for (i = 1; i < 8; i++) {
1534                 func_drc = spapr_phb_get_pci_func_drc(phb, pci_bus_num(bus),
1535                                                       PCI_DEVFN(slotnr, i));
1536                 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1537                 state = func_drck->dr_entity_sense(func_drc);
1538                 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT
1539                     && !spapr_drc_unplug_requested(func_drc)) {
1540                     error_setg(errp,
1541                                "PCI: slot %d, function %d still present. "
1542                                "Must unplug all non-0 functions first.",
1543                                slotnr, i);
1544                     return;
1545                 }
1546             }
1547         }
1548 
1549         spapr_drc_detach(drc);
1550 
1551         /* if this isn't func 0, defer unplug event. otherwise signal removal
1552          * for all present functions
1553          */
1554         if (PCI_FUNC(pdev->devfn) == 0) {
1555             for (i = 7; i >= 0; i--) {
1556                 func_drc = spapr_phb_get_pci_func_drc(phb, pci_bus_num(bus),
1557                                                       PCI_DEVFN(slotnr, i));
1558                 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1559                 state = func_drck->dr_entity_sense(func_drc);
1560                 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) {
1561                     spapr_hotplug_req_remove_by_index(func_drc);
1562                 }
1563             }
1564         }
1565     }
1566 }
1567 
1568 static void spapr_phb_realize(DeviceState *dev, Error **errp)
1569 {
1570     /* We don't use SPAPR_MACHINE() in order to exit gracefully if the user
1571      * tries to add a sPAPR PHB to a non-pseries machine.
1572      */
1573     sPAPRMachineState *spapr =
1574         (sPAPRMachineState *) object_dynamic_cast(qdev_get_machine(),
1575                                                   TYPE_SPAPR_MACHINE);
1576     sPAPRMachineClass *smc = spapr ? SPAPR_MACHINE_GET_CLASS(spapr) : NULL;
1577     SysBusDevice *s = SYS_BUS_DEVICE(dev);
1578     sPAPRPHBState *sphb = SPAPR_PCI_HOST_BRIDGE(s);
1579     PCIHostState *phb = PCI_HOST_BRIDGE(s);
1580     char *namebuf;
1581     int i;
1582     PCIBus *bus;
1583     uint64_t msi_window_size = 4096;
1584     sPAPRTCETable *tcet;
1585     const unsigned windows_supported =
1586         sphb->ddw_enabled ? SPAPR_PCI_DMA_MAX_WINDOWS : 1;
1587 
1588     if (!spapr) {
1589         error_setg(errp, TYPE_SPAPR_PCI_HOST_BRIDGE " needs a pseries machine");
1590         return;
1591     }
1592 
1593     if (sphb->index != (uint32_t)-1) {
1594         Error *local_err = NULL;
1595 
1596         smc->phb_placement(spapr, sphb->index,
1597                            &sphb->buid, &sphb->io_win_addr,
1598                            &sphb->mem_win_addr, &sphb->mem64_win_addr,
1599                            windows_supported, sphb->dma_liobn, &local_err);
1600         if (local_err) {
1601             error_propagate(errp, local_err);
1602             return;
1603         }
1604     } else {
1605         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
1606         return;
1607     }
1608 
1609     if (sphb->mem64_win_size != 0) {
1610         if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) {
1611             error_setg(errp, "32-bit memory window of size 0x%"HWADDR_PRIx
1612                        " (max 2 GiB)", sphb->mem_win_size);
1613             return;
1614         }
1615 
1616         /* 64-bit window defaults to identity mapping */
1617         sphb->mem64_win_pciaddr = sphb->mem64_win_addr;
1618     } else if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) {
1619         /*
1620          * For compatibility with old configuration, if no 64-bit MMIO
1621          * window is specified, but the ordinary (32-bit) memory
1622          * window is specified as > 2GiB, we treat it as a 2GiB 32-bit
1623          * window, with a 64-bit MMIO window following on immediately
1624          * afterwards
1625          */
1626         sphb->mem64_win_size = sphb->mem_win_size - SPAPR_PCI_MEM32_WIN_SIZE;
1627         sphb->mem64_win_addr = sphb->mem_win_addr + SPAPR_PCI_MEM32_WIN_SIZE;
1628         sphb->mem64_win_pciaddr =
1629             SPAPR_PCI_MEM_WIN_BUS_OFFSET + SPAPR_PCI_MEM32_WIN_SIZE;
1630         sphb->mem_win_size = SPAPR_PCI_MEM32_WIN_SIZE;
1631     }
1632 
1633     if (spapr_pci_find_phb(spapr, sphb->buid)) {
1634         error_setg(errp, "PCI host bridges must have unique BUIDs");
1635         return;
1636     }
1637 
1638     if (sphb->numa_node != -1 &&
1639         (sphb->numa_node >= MAX_NODES || !numa_info[sphb->numa_node].present)) {
1640         error_setg(errp, "Invalid NUMA node ID for PCI host bridge");
1641         return;
1642     }
1643 
1644     sphb->dtbusname = g_strdup_printf("pci@%" PRIx64, sphb->buid);
1645 
1646     /* Initialize memory regions */
1647     namebuf = g_strdup_printf("%s.mmio", sphb->dtbusname);
1648     memory_region_init(&sphb->memspace, OBJECT(sphb), namebuf, UINT64_MAX);
1649     g_free(namebuf);
1650 
1651     namebuf = g_strdup_printf("%s.mmio32-alias", sphb->dtbusname);
1652     memory_region_init_alias(&sphb->mem32window, OBJECT(sphb),
1653                              namebuf, &sphb->memspace,
1654                              SPAPR_PCI_MEM_WIN_BUS_OFFSET, sphb->mem_win_size);
1655     g_free(namebuf);
1656     memory_region_add_subregion(get_system_memory(), sphb->mem_win_addr,
1657                                 &sphb->mem32window);
1658 
1659     if (sphb->mem64_win_size != 0) {
1660         namebuf = g_strdup_printf("%s.mmio64-alias", sphb->dtbusname);
1661         memory_region_init_alias(&sphb->mem64window, OBJECT(sphb),
1662                                  namebuf, &sphb->memspace,
1663                                  sphb->mem64_win_pciaddr, sphb->mem64_win_size);
1664         g_free(namebuf);
1665 
1666         memory_region_add_subregion(get_system_memory(),
1667                                     sphb->mem64_win_addr,
1668                                     &sphb->mem64window);
1669     }
1670 
1671     /* Initialize IO regions */
1672     namebuf = g_strdup_printf("%s.io", sphb->dtbusname);
1673     memory_region_init(&sphb->iospace, OBJECT(sphb),
1674                        namebuf, SPAPR_PCI_IO_WIN_SIZE);
1675     g_free(namebuf);
1676 
1677     namebuf = g_strdup_printf("%s.io-alias", sphb->dtbusname);
1678     memory_region_init_alias(&sphb->iowindow, OBJECT(sphb), namebuf,
1679                              &sphb->iospace, 0, SPAPR_PCI_IO_WIN_SIZE);
1680     g_free(namebuf);
1681     memory_region_add_subregion(get_system_memory(), sphb->io_win_addr,
1682                                 &sphb->iowindow);
1683 
1684     bus = pci_register_root_bus(dev, NULL,
1685                                 pci_spapr_set_irq, pci_spapr_map_irq, sphb,
1686                                 &sphb->memspace, &sphb->iospace,
1687                                 PCI_DEVFN(0, 0), PCI_NUM_PINS, TYPE_PCI_BUS);
1688     phb->bus = bus;
1689     qbus_set_hotplug_handler(BUS(phb->bus), OBJECT(sphb), NULL);
1690 
1691     /*
1692      * Initialize PHB address space.
1693      * By default there will be at least one subregion for default
1694      * 32bit DMA window.
1695      * Later the guest might want to create another DMA window
1696      * which will become another memory subregion.
1697      */
1698     namebuf = g_strdup_printf("%s.iommu-root", sphb->dtbusname);
1699     memory_region_init(&sphb->iommu_root, OBJECT(sphb),
1700                        namebuf, UINT64_MAX);
1701     g_free(namebuf);
1702     address_space_init(&sphb->iommu_as, &sphb->iommu_root,
1703                        sphb->dtbusname);
1704 
1705     /*
1706      * As MSI/MSIX interrupts trigger by writing at MSI/MSIX vectors,
1707      * we need to allocate some memory to catch those writes coming
1708      * from msi_notify()/msix_notify().
1709      * As MSIMessage:addr is going to be the same and MSIMessage:data
1710      * is going to be a VIRQ number, 4 bytes of the MSI MR will only
1711      * be used.
1712      *
1713      * For KVM we want to ensure that this memory is a full page so that
1714      * our memory slot is of page size granularity.
1715      */
1716 #ifdef CONFIG_KVM
1717     if (kvm_enabled()) {
1718         msi_window_size = getpagesize();
1719     }
1720 #endif
1721 
1722     memory_region_init_io(&sphb->msiwindow, OBJECT(sphb), &spapr_msi_ops, spapr,
1723                           "msi", msi_window_size);
1724     memory_region_add_subregion(&sphb->iommu_root, SPAPR_PCI_MSI_WINDOW,
1725                                 &sphb->msiwindow);
1726 
1727     pci_setup_iommu(bus, spapr_pci_dma_iommu, sphb);
1728 
1729     pci_bus_set_route_irq_fn(bus, spapr_route_intx_pin_to_irq);
1730 
1731     QLIST_INSERT_HEAD(&spapr->phbs, sphb, list);
1732 
1733     /* Initialize the LSI table */
1734     for (i = 0; i < PCI_NUM_PINS; i++) {
1735         uint32_t irq = SPAPR_IRQ_PCI_LSI + sphb->index * PCI_NUM_PINS + i;
1736         Error *local_err = NULL;
1737 
1738         if (smc->legacy_irq_allocation) {
1739             irq = spapr_irq_findone(spapr, &local_err);
1740             if (local_err) {
1741                 error_propagate_prepend(errp, local_err,
1742                                         "can't allocate LSIs: ");
1743                 return;
1744             }
1745         }
1746 
1747         spapr_irq_claim(spapr, irq, true, &local_err);
1748         if (local_err) {
1749             error_propagate_prepend(errp, local_err, "can't allocate LSIs: ");
1750             return;
1751         }
1752 
1753         sphb->lsi_table[i].irq = irq;
1754     }
1755 
1756     /* allocate connectors for child PCI devices */
1757     if (sphb->dr_enabled) {
1758         for (i = 0; i < PCI_SLOT_MAX * 8; i++) {
1759             spapr_dr_connector_new(OBJECT(phb), TYPE_SPAPR_DRC_PCI,
1760                                    (sphb->index << 16) | i);
1761         }
1762     }
1763 
1764     /* DMA setup */
1765     for (i = 0; i < windows_supported; ++i) {
1766         tcet = spapr_tce_new_table(DEVICE(sphb), sphb->dma_liobn[i]);
1767         if (!tcet) {
1768             error_setg(errp, "Creating window#%d failed for %s",
1769                        i, sphb->dtbusname);
1770             return;
1771         }
1772         memory_region_add_subregion(&sphb->iommu_root, 0,
1773                                     spapr_tce_get_iommu(tcet));
1774     }
1775 
1776     sphb->msi = g_hash_table_new_full(g_int_hash, g_int_equal, g_free, g_free);
1777 }
1778 
1779 static int spapr_phb_children_reset(Object *child, void *opaque)
1780 {
1781     DeviceState *dev = (DeviceState *) object_dynamic_cast(child, TYPE_DEVICE);
1782 
1783     if (dev) {
1784         device_reset(dev);
1785     }
1786 
1787     return 0;
1788 }
1789 
1790 void spapr_phb_dma_reset(sPAPRPHBState *sphb)
1791 {
1792     int i;
1793     sPAPRTCETable *tcet;
1794 
1795     for (i = 0; i < SPAPR_PCI_DMA_MAX_WINDOWS; ++i) {
1796         tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[i]);
1797 
1798         if (tcet && tcet->nb_table) {
1799             spapr_tce_table_disable(tcet);
1800         }
1801     }
1802 
1803     /* Register default 32bit DMA window */
1804     tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[0]);
1805     spapr_tce_table_enable(tcet, SPAPR_TCE_PAGE_SHIFT, sphb->dma_win_addr,
1806                            sphb->dma_win_size >> SPAPR_TCE_PAGE_SHIFT);
1807 }
1808 
1809 static void spapr_phb_reset(DeviceState *qdev)
1810 {
1811     sPAPRPHBState *sphb = SPAPR_PCI_HOST_BRIDGE(qdev);
1812 
1813     spapr_phb_dma_reset(sphb);
1814 
1815     /* Reset the IOMMU state */
1816     object_child_foreach(OBJECT(qdev), spapr_phb_children_reset, NULL);
1817 
1818     if (spapr_phb_eeh_available(SPAPR_PCI_HOST_BRIDGE(qdev))) {
1819         spapr_phb_vfio_reset(qdev);
1820     }
1821 }
1822 
1823 static Property spapr_phb_properties[] = {
1824     DEFINE_PROP_UINT32("index", sPAPRPHBState, index, -1),
1825     DEFINE_PROP_UINT64("mem_win_size", sPAPRPHBState, mem_win_size,
1826                        SPAPR_PCI_MEM32_WIN_SIZE),
1827     DEFINE_PROP_UINT64("mem64_win_size", sPAPRPHBState, mem64_win_size,
1828                        SPAPR_PCI_MEM64_WIN_SIZE),
1829     DEFINE_PROP_UINT64("io_win_size", sPAPRPHBState, io_win_size,
1830                        SPAPR_PCI_IO_WIN_SIZE),
1831     DEFINE_PROP_BOOL("dynamic-reconfiguration", sPAPRPHBState, dr_enabled,
1832                      true),
1833     /* Default DMA window is 0..1GB */
1834     DEFINE_PROP_UINT64("dma_win_addr", sPAPRPHBState, dma_win_addr, 0),
1835     DEFINE_PROP_UINT64("dma_win_size", sPAPRPHBState, dma_win_size, 0x40000000),
1836     DEFINE_PROP_UINT64("dma64_win_addr", sPAPRPHBState, dma64_win_addr,
1837                        0x800000000000000ULL),
1838     DEFINE_PROP_BOOL("ddw", sPAPRPHBState, ddw_enabled, true),
1839     DEFINE_PROP_UINT64("pgsz", sPAPRPHBState, page_size_mask,
1840                        (1ULL << 12) | (1ULL << 16)),
1841     DEFINE_PROP_UINT32("numa_node", sPAPRPHBState, numa_node, -1),
1842     DEFINE_PROP_BOOL("pre-2.8-migration", sPAPRPHBState,
1843                      pre_2_8_migration, false),
1844     DEFINE_PROP_BOOL("pcie-extended-configuration-space", sPAPRPHBState,
1845                      pcie_ecs, true),
1846     DEFINE_PROP_END_OF_LIST(),
1847 };
1848 
1849 static const VMStateDescription vmstate_spapr_pci_lsi = {
1850     .name = "spapr_pci/lsi",
1851     .version_id = 1,
1852     .minimum_version_id = 1,
1853     .fields = (VMStateField[]) {
1854         VMSTATE_UINT32_EQUAL(irq, struct spapr_pci_lsi, NULL),
1855 
1856         VMSTATE_END_OF_LIST()
1857     },
1858 };
1859 
1860 static const VMStateDescription vmstate_spapr_pci_msi = {
1861     .name = "spapr_pci/msi",
1862     .version_id = 1,
1863     .minimum_version_id = 1,
1864     .fields = (VMStateField []) {
1865         VMSTATE_UINT32(key, spapr_pci_msi_mig),
1866         VMSTATE_UINT32(value.first_irq, spapr_pci_msi_mig),
1867         VMSTATE_UINT32(value.num, spapr_pci_msi_mig),
1868         VMSTATE_END_OF_LIST()
1869     },
1870 };
1871 
1872 static int spapr_pci_pre_save(void *opaque)
1873 {
1874     sPAPRPHBState *sphb = opaque;
1875     GHashTableIter iter;
1876     gpointer key, value;
1877     int i;
1878 
1879     if (sphb->pre_2_8_migration) {
1880         sphb->mig_liobn = sphb->dma_liobn[0];
1881         sphb->mig_mem_win_addr = sphb->mem_win_addr;
1882         sphb->mig_mem_win_size = sphb->mem_win_size;
1883         sphb->mig_io_win_addr = sphb->io_win_addr;
1884         sphb->mig_io_win_size = sphb->io_win_size;
1885 
1886         if ((sphb->mem64_win_size != 0)
1887             && (sphb->mem64_win_addr
1888                 == (sphb->mem_win_addr + sphb->mem_win_size))) {
1889             sphb->mig_mem_win_size += sphb->mem64_win_size;
1890         }
1891     }
1892 
1893     g_free(sphb->msi_devs);
1894     sphb->msi_devs = NULL;
1895     sphb->msi_devs_num = g_hash_table_size(sphb->msi);
1896     if (!sphb->msi_devs_num) {
1897         return 0;
1898     }
1899     sphb->msi_devs = g_new(spapr_pci_msi_mig, sphb->msi_devs_num);
1900 
1901     g_hash_table_iter_init(&iter, sphb->msi);
1902     for (i = 0; g_hash_table_iter_next(&iter, &key, &value); ++i) {
1903         sphb->msi_devs[i].key = *(uint32_t *) key;
1904         sphb->msi_devs[i].value = *(spapr_pci_msi *) value;
1905     }
1906 
1907     return 0;
1908 }
1909 
1910 static int spapr_pci_post_load(void *opaque, int version_id)
1911 {
1912     sPAPRPHBState *sphb = opaque;
1913     gpointer key, value;
1914     int i;
1915 
1916     for (i = 0; i < sphb->msi_devs_num; ++i) {
1917         key = g_memdup(&sphb->msi_devs[i].key,
1918                        sizeof(sphb->msi_devs[i].key));
1919         value = g_memdup(&sphb->msi_devs[i].value,
1920                          sizeof(sphb->msi_devs[i].value));
1921         g_hash_table_insert(sphb->msi, key, value);
1922     }
1923     g_free(sphb->msi_devs);
1924     sphb->msi_devs = NULL;
1925     sphb->msi_devs_num = 0;
1926 
1927     return 0;
1928 }
1929 
1930 static bool pre_2_8_migration(void *opaque, int version_id)
1931 {
1932     sPAPRPHBState *sphb = opaque;
1933 
1934     return sphb->pre_2_8_migration;
1935 }
1936 
1937 static const VMStateDescription vmstate_spapr_pci = {
1938     .name = "spapr_pci",
1939     .version_id = 2,
1940     .minimum_version_id = 2,
1941     .pre_save = spapr_pci_pre_save,
1942     .post_load = spapr_pci_post_load,
1943     .fields = (VMStateField[]) {
1944         VMSTATE_UINT64_EQUAL(buid, sPAPRPHBState, NULL),
1945         VMSTATE_UINT32_TEST(mig_liobn, sPAPRPHBState, pre_2_8_migration),
1946         VMSTATE_UINT64_TEST(mig_mem_win_addr, sPAPRPHBState, pre_2_8_migration),
1947         VMSTATE_UINT64_TEST(mig_mem_win_size, sPAPRPHBState, pre_2_8_migration),
1948         VMSTATE_UINT64_TEST(mig_io_win_addr, sPAPRPHBState, pre_2_8_migration),
1949         VMSTATE_UINT64_TEST(mig_io_win_size, sPAPRPHBState, pre_2_8_migration),
1950         VMSTATE_STRUCT_ARRAY(lsi_table, sPAPRPHBState, PCI_NUM_PINS, 0,
1951                              vmstate_spapr_pci_lsi, struct spapr_pci_lsi),
1952         VMSTATE_INT32(msi_devs_num, sPAPRPHBState),
1953         VMSTATE_STRUCT_VARRAY_ALLOC(msi_devs, sPAPRPHBState, msi_devs_num, 0,
1954                                     vmstate_spapr_pci_msi, spapr_pci_msi_mig),
1955         VMSTATE_END_OF_LIST()
1956     },
1957 };
1958 
1959 static const char *spapr_phb_root_bus_path(PCIHostState *host_bridge,
1960                                            PCIBus *rootbus)
1961 {
1962     sPAPRPHBState *sphb = SPAPR_PCI_HOST_BRIDGE(host_bridge);
1963 
1964     return sphb->dtbusname;
1965 }
1966 
1967 static void spapr_phb_class_init(ObjectClass *klass, void *data)
1968 {
1969     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(klass);
1970     DeviceClass *dc = DEVICE_CLASS(klass);
1971     HotplugHandlerClass *hp = HOTPLUG_HANDLER_CLASS(klass);
1972 
1973     hc->root_bus_path = spapr_phb_root_bus_path;
1974     dc->realize = spapr_phb_realize;
1975     dc->props = spapr_phb_properties;
1976     dc->reset = spapr_phb_reset;
1977     dc->vmsd = &vmstate_spapr_pci;
1978     /* Supported by TYPE_SPAPR_MACHINE */
1979     dc->user_creatable = true;
1980     set_bit(DEVICE_CATEGORY_BRIDGE, dc->categories);
1981     hp->plug = spapr_pci_plug;
1982     hp->unplug = spapr_pci_unplug;
1983     hp->unplug_request = spapr_pci_unplug_request;
1984 }
1985 
1986 static const TypeInfo spapr_phb_info = {
1987     .name          = TYPE_SPAPR_PCI_HOST_BRIDGE,
1988     .parent        = TYPE_PCI_HOST_BRIDGE,
1989     .instance_size = sizeof(sPAPRPHBState),
1990     .class_init    = spapr_phb_class_init,
1991     .interfaces    = (InterfaceInfo[]) {
1992         { TYPE_HOTPLUG_HANDLER },
1993         { }
1994     }
1995 };
1996 
1997 typedef struct sPAPRFDT {
1998     void *fdt;
1999     int node_off;
2000     sPAPRPHBState *sphb;
2001 } sPAPRFDT;
2002 
2003 static void spapr_populate_pci_devices_dt(PCIBus *bus, PCIDevice *pdev,
2004                                           void *opaque)
2005 {
2006     PCIBus *sec_bus;
2007     sPAPRFDT *p = opaque;
2008     int offset;
2009     sPAPRFDT s_fdt;
2010 
2011     offset = spapr_create_pci_child_dt(p->sphb, pdev, p->fdt, p->node_off);
2012     if (!offset) {
2013         error_report("Failed to create pci child device tree node");
2014         return;
2015     }
2016 
2017     if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) !=
2018          PCI_HEADER_TYPE_BRIDGE)) {
2019         return;
2020     }
2021 
2022     sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev));
2023     if (!sec_bus) {
2024         return;
2025     }
2026 
2027     s_fdt.fdt = p->fdt;
2028     s_fdt.node_off = offset;
2029     s_fdt.sphb = p->sphb;
2030     pci_for_each_device_reverse(sec_bus, pci_bus_num(sec_bus),
2031                                 spapr_populate_pci_devices_dt,
2032                                 &s_fdt);
2033 }
2034 
2035 static void spapr_phb_pci_enumerate_bridge(PCIBus *bus, PCIDevice *pdev,
2036                                            void *opaque)
2037 {
2038     unsigned int *bus_no = opaque;
2039     PCIBus *sec_bus = NULL;
2040 
2041     if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) !=
2042          PCI_HEADER_TYPE_BRIDGE)) {
2043         return;
2044     }
2045 
2046     (*bus_no)++;
2047     pci_default_write_config(pdev, PCI_PRIMARY_BUS, pci_dev_bus_num(pdev), 1);
2048     pci_default_write_config(pdev, PCI_SECONDARY_BUS, *bus_no, 1);
2049     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
2050 
2051     sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev));
2052     if (!sec_bus) {
2053         return;
2054     }
2055 
2056     pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
2057                         spapr_phb_pci_enumerate_bridge, bus_no);
2058     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
2059 }
2060 
2061 static void spapr_phb_pci_enumerate(sPAPRPHBState *phb)
2062 {
2063     PCIBus *bus = PCI_HOST_BRIDGE(phb)->bus;
2064     unsigned int bus_no = 0;
2065 
2066     pci_for_each_device(bus, pci_bus_num(bus),
2067                         spapr_phb_pci_enumerate_bridge,
2068                         &bus_no);
2069 
2070 }
2071 
2072 int spapr_populate_pci_dt(sPAPRPHBState *phb, uint32_t intc_phandle, void *fdt,
2073                           uint32_t nr_msis)
2074 {
2075     int bus_off, i, j, ret;
2076     gchar *nodename;
2077     uint32_t bus_range[] = { cpu_to_be32(0), cpu_to_be32(0xff) };
2078     struct {
2079         uint32_t hi;
2080         uint64_t child;
2081         uint64_t parent;
2082         uint64_t size;
2083     } QEMU_PACKED ranges[] = {
2084         {
2085             cpu_to_be32(b_ss(1)), cpu_to_be64(0),
2086             cpu_to_be64(phb->io_win_addr),
2087             cpu_to_be64(memory_region_size(&phb->iospace)),
2088         },
2089         {
2090             cpu_to_be32(b_ss(2)), cpu_to_be64(SPAPR_PCI_MEM_WIN_BUS_OFFSET),
2091             cpu_to_be64(phb->mem_win_addr),
2092             cpu_to_be64(phb->mem_win_size),
2093         },
2094         {
2095             cpu_to_be32(b_ss(3)), cpu_to_be64(phb->mem64_win_pciaddr),
2096             cpu_to_be64(phb->mem64_win_addr),
2097             cpu_to_be64(phb->mem64_win_size),
2098         },
2099     };
2100     const unsigned sizeof_ranges =
2101         (phb->mem64_win_size ? 3 : 2) * sizeof(ranges[0]);
2102     uint64_t bus_reg[] = { cpu_to_be64(phb->buid), 0 };
2103     uint32_t interrupt_map_mask[] = {
2104         cpu_to_be32(b_ddddd(-1)|b_fff(0)), 0x0, 0x0, cpu_to_be32(-1)};
2105     uint32_t interrupt_map[PCI_SLOT_MAX * PCI_NUM_PINS][7];
2106     uint32_t ddw_applicable[] = {
2107         cpu_to_be32(RTAS_IBM_QUERY_PE_DMA_WINDOW),
2108         cpu_to_be32(RTAS_IBM_CREATE_PE_DMA_WINDOW),
2109         cpu_to_be32(RTAS_IBM_REMOVE_PE_DMA_WINDOW)
2110     };
2111     uint32_t ddw_extensions[] = {
2112         cpu_to_be32(1),
2113         cpu_to_be32(RTAS_IBM_RESET_PE_DMA_WINDOW)
2114     };
2115     uint32_t associativity[] = {cpu_to_be32(0x4),
2116                                 cpu_to_be32(0x0),
2117                                 cpu_to_be32(0x0),
2118                                 cpu_to_be32(0x0),
2119                                 cpu_to_be32(phb->numa_node)};
2120     sPAPRTCETable *tcet;
2121     PCIBus *bus = PCI_HOST_BRIDGE(phb)->bus;
2122     sPAPRFDT s_fdt;
2123 
2124     /* Start populating the FDT */
2125     nodename = g_strdup_printf("pci@%" PRIx64, phb->buid);
2126     _FDT(bus_off = fdt_add_subnode(fdt, 0, nodename));
2127     g_free(nodename);
2128 
2129     /* Write PHB properties */
2130     _FDT(fdt_setprop_string(fdt, bus_off, "device_type", "pci"));
2131     _FDT(fdt_setprop_string(fdt, bus_off, "compatible", "IBM,Logical_PHB"));
2132     _FDT(fdt_setprop_cell(fdt, bus_off, "#address-cells", 0x3));
2133     _FDT(fdt_setprop_cell(fdt, bus_off, "#size-cells", 0x2));
2134     _FDT(fdt_setprop_cell(fdt, bus_off, "#interrupt-cells", 0x1));
2135     _FDT(fdt_setprop(fdt, bus_off, "used-by-rtas", NULL, 0));
2136     _FDT(fdt_setprop(fdt, bus_off, "bus-range", &bus_range, sizeof(bus_range)));
2137     _FDT(fdt_setprop(fdt, bus_off, "ranges", &ranges, sizeof_ranges));
2138     _FDT(fdt_setprop(fdt, bus_off, "reg", &bus_reg, sizeof(bus_reg)));
2139     _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pci-config-space-type", 0x1));
2140     _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pe-total-#msi", nr_msis));
2141 
2142     /* Dynamic DMA window */
2143     if (phb->ddw_enabled) {
2144         _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-applicable", &ddw_applicable,
2145                          sizeof(ddw_applicable)));
2146         _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-extensions",
2147                          &ddw_extensions, sizeof(ddw_extensions)));
2148     }
2149 
2150     /* Advertise NUMA via ibm,associativity */
2151     if (phb->numa_node != -1) {
2152         _FDT(fdt_setprop(fdt, bus_off, "ibm,associativity", associativity,
2153                          sizeof(associativity)));
2154     }
2155 
2156     /* Build the interrupt-map, this must matches what is done
2157      * in pci_spapr_map_irq
2158      */
2159     _FDT(fdt_setprop(fdt, bus_off, "interrupt-map-mask",
2160                      &interrupt_map_mask, sizeof(interrupt_map_mask)));
2161     for (i = 0; i < PCI_SLOT_MAX; i++) {
2162         for (j = 0; j < PCI_NUM_PINS; j++) {
2163             uint32_t *irqmap = interrupt_map[i*PCI_NUM_PINS + j];
2164             int lsi_num = pci_spapr_swizzle(i, j);
2165 
2166             irqmap[0] = cpu_to_be32(b_ddddd(i)|b_fff(0));
2167             irqmap[1] = 0;
2168             irqmap[2] = 0;
2169             irqmap[3] = cpu_to_be32(j+1);
2170             irqmap[4] = cpu_to_be32(intc_phandle);
2171             spapr_dt_irq(&irqmap[5], phb->lsi_table[lsi_num].irq, true);
2172         }
2173     }
2174     /* Write interrupt map */
2175     _FDT(fdt_setprop(fdt, bus_off, "interrupt-map", &interrupt_map,
2176                      sizeof(interrupt_map)));
2177 
2178     tcet = spapr_tce_find_by_liobn(phb->dma_liobn[0]);
2179     if (!tcet) {
2180         return -1;
2181     }
2182     spapr_dma_dt(fdt, bus_off, "ibm,dma-window",
2183                  tcet->liobn, tcet->bus_offset,
2184                  tcet->nb_table << tcet->page_shift);
2185 
2186     /* Walk the bridges and program the bus numbers*/
2187     spapr_phb_pci_enumerate(phb);
2188     _FDT(fdt_setprop_cell(fdt, bus_off, "qemu,phb-enumerated", 0x1));
2189 
2190     /* Populate tree nodes with PCI devices attached */
2191     s_fdt.fdt = fdt;
2192     s_fdt.node_off = bus_off;
2193     s_fdt.sphb = phb;
2194     pci_for_each_device_reverse(bus, pci_bus_num(bus),
2195                                 spapr_populate_pci_devices_dt,
2196                                 &s_fdt);
2197 
2198     ret = spapr_drc_populate_dt(fdt, bus_off, OBJECT(phb),
2199                                 SPAPR_DR_CONNECTOR_TYPE_PCI);
2200     if (ret) {
2201         return ret;
2202     }
2203 
2204     return 0;
2205 }
2206 
2207 void spapr_pci_rtas_init(void)
2208 {
2209     spapr_rtas_register(RTAS_READ_PCI_CONFIG, "read-pci-config",
2210                         rtas_read_pci_config);
2211     spapr_rtas_register(RTAS_WRITE_PCI_CONFIG, "write-pci-config",
2212                         rtas_write_pci_config);
2213     spapr_rtas_register(RTAS_IBM_READ_PCI_CONFIG, "ibm,read-pci-config",
2214                         rtas_ibm_read_pci_config);
2215     spapr_rtas_register(RTAS_IBM_WRITE_PCI_CONFIG, "ibm,write-pci-config",
2216                         rtas_ibm_write_pci_config);
2217     if (msi_nonbroken) {
2218         spapr_rtas_register(RTAS_IBM_QUERY_INTERRUPT_SOURCE_NUMBER,
2219                             "ibm,query-interrupt-source-number",
2220                             rtas_ibm_query_interrupt_source_number);
2221         spapr_rtas_register(RTAS_IBM_CHANGE_MSI, "ibm,change-msi",
2222                             rtas_ibm_change_msi);
2223     }
2224 
2225     spapr_rtas_register(RTAS_IBM_SET_EEH_OPTION,
2226                         "ibm,set-eeh-option",
2227                         rtas_ibm_set_eeh_option);
2228     spapr_rtas_register(RTAS_IBM_GET_CONFIG_ADDR_INFO2,
2229                         "ibm,get-config-addr-info2",
2230                         rtas_ibm_get_config_addr_info2);
2231     spapr_rtas_register(RTAS_IBM_READ_SLOT_RESET_STATE2,
2232                         "ibm,read-slot-reset-state2",
2233                         rtas_ibm_read_slot_reset_state2);
2234     spapr_rtas_register(RTAS_IBM_SET_SLOT_RESET,
2235                         "ibm,set-slot-reset",
2236                         rtas_ibm_set_slot_reset);
2237     spapr_rtas_register(RTAS_IBM_CONFIGURE_PE,
2238                         "ibm,configure-pe",
2239                         rtas_ibm_configure_pe);
2240     spapr_rtas_register(RTAS_IBM_SLOT_ERROR_DETAIL,
2241                         "ibm,slot-error-detail",
2242                         rtas_ibm_slot_error_detail);
2243 }
2244 
2245 static void spapr_pci_register_types(void)
2246 {
2247     type_register_static(&spapr_phb_info);
2248 }
2249 
2250 type_init(spapr_pci_register_types)
2251 
2252 static int spapr_switch_one_vga(DeviceState *dev, void *opaque)
2253 {
2254     bool be = *(bool *)opaque;
2255 
2256     if (object_dynamic_cast(OBJECT(dev), "VGA")
2257         || object_dynamic_cast(OBJECT(dev), "secondary-vga")) {
2258         object_property_set_bool(OBJECT(dev), be, "big-endian-framebuffer",
2259                                  &error_abort);
2260     }
2261     return 0;
2262 }
2263 
2264 void spapr_pci_switch_vga(bool big_endian)
2265 {
2266     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
2267     sPAPRPHBState *sphb;
2268 
2269     /*
2270      * For backward compatibility with existing guests, we switch
2271      * the endianness of the VGA controller when changing the guest
2272      * interrupt mode
2273      */
2274     QLIST_FOREACH(sphb, &spapr->phbs, list) {
2275         BusState *bus = &PCI_HOST_BRIDGE(sphb)->bus->qbus;
2276         qbus_walk_children(bus, spapr_switch_one_vga, NULL, NULL, NULL,
2277                            &big_endian);
2278     }
2279 }
2280