xref: /openbmc/qemu/hw/ppc/spapr_pci.c (revision 041c2a31)
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 
26 #include "qemu/osdep.h"
27 #include "qapi/error.h"
28 #include "cpu.h"
29 #include "hw/irq.h"
30 #include "hw/sysbus.h"
31 #include "migration/vmstate.h"
32 #include "hw/pci/pci.h"
33 #include "hw/pci/msi.h"
34 #include "hw/pci/msix.h"
35 #include "hw/pci/pci_host.h"
36 #include "hw/ppc/spapr.h"
37 #include "hw/pci-host/spapr.h"
38 #include "exec/address-spaces.h"
39 #include "exec/ram_addr.h"
40 #include <libfdt.h>
41 #include "trace.h"
42 #include "qemu/error-report.h"
43 #include "qemu/module.h"
44 #include "qapi/qmp/qerror.h"
45 #include "hw/ppc/fdt.h"
46 #include "hw/pci/pci_bridge.h"
47 #include "hw/pci/pci_bus.h"
48 #include "hw/pci/pci_ids.h"
49 #include "hw/ppc/spapr_drc.h"
50 #include "hw/qdev-properties.h"
51 #include "sysemu/device_tree.h"
52 #include "sysemu/kvm.h"
53 #include "sysemu/hostmem.h"
54 #include "sysemu/numa.h"
55 #include "hw/ppc/spapr_numa.h"
56 #include "qemu/log.h"
57 
58 /* Copied from the kernel arch/powerpc/platforms/pseries/msi.c */
59 #define RTAS_QUERY_FN           0
60 #define RTAS_CHANGE_FN          1
61 #define RTAS_RESET_FN           2
62 #define RTAS_CHANGE_MSI_FN      3
63 #define RTAS_CHANGE_MSIX_FN     4
64 
65 /* Interrupt types to return on RTAS_CHANGE_* */
66 #define RTAS_TYPE_MSI           1
67 #define RTAS_TYPE_MSIX          2
68 
69 SpaprPhbState *spapr_pci_find_phb(SpaprMachineState *spapr, uint64_t buid)
70 {
71     SpaprPhbState *sphb;
72 
73     QLIST_FOREACH(sphb, &spapr->phbs, list) {
74         if (sphb->buid != buid) {
75             continue;
76         }
77         return sphb;
78     }
79 
80     return NULL;
81 }
82 
83 PCIDevice *spapr_pci_find_dev(SpaprMachineState *spapr, uint64_t buid,
84                               uint32_t config_addr)
85 {
86     SpaprPhbState *sphb = spapr_pci_find_phb(spapr, buid);
87     PCIHostState *phb = PCI_HOST_BRIDGE(sphb);
88     int bus_num = (config_addr >> 16) & 0xFF;
89     int devfn = (config_addr >> 8) & 0xFF;
90 
91     if (!phb) {
92         return NULL;
93     }
94 
95     return pci_find_device(phb->bus, bus_num, devfn);
96 }
97 
98 static uint32_t rtas_pci_cfgaddr(uint32_t arg)
99 {
100     /* This handles the encoding of extended config space addresses */
101     return ((arg >> 20) & 0xf00) | (arg & 0xff);
102 }
103 
104 static void finish_read_pci_config(SpaprMachineState *spapr, uint64_t buid,
105                                    uint32_t addr, uint32_t size,
106                                    target_ulong rets)
107 {
108     PCIDevice *pci_dev;
109     uint32_t val;
110 
111     if ((size != 1) && (size != 2) && (size != 4)) {
112         /* access must be 1, 2 or 4 bytes */
113         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
114         return;
115     }
116 
117     pci_dev = spapr_pci_find_dev(spapr, buid, addr);
118     addr = rtas_pci_cfgaddr(addr);
119 
120     if (!pci_dev || (addr % size) || (addr >= pci_config_size(pci_dev))) {
121         /* Access must be to a valid device, within bounds and
122          * naturally aligned */
123         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
124         return;
125     }
126 
127     val = pci_host_config_read_common(pci_dev, addr,
128                                       pci_config_size(pci_dev), size);
129 
130     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
131     rtas_st(rets, 1, val);
132 }
133 
134 static void rtas_ibm_read_pci_config(PowerPCCPU *cpu, SpaprMachineState *spapr,
135                                      uint32_t token, uint32_t nargs,
136                                      target_ulong args,
137                                      uint32_t nret, target_ulong rets)
138 {
139     uint64_t buid;
140     uint32_t size, addr;
141 
142     if ((nargs != 4) || (nret != 2)) {
143         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
144         return;
145     }
146 
147     buid = rtas_ldq(args, 1);
148     size = rtas_ld(args, 3);
149     addr = rtas_ld(args, 0);
150 
151     finish_read_pci_config(spapr, buid, addr, size, rets);
152 }
153 
154 static void rtas_read_pci_config(PowerPCCPU *cpu, SpaprMachineState *spapr,
155                                  uint32_t token, uint32_t nargs,
156                                  target_ulong args,
157                                  uint32_t nret, target_ulong rets)
158 {
159     uint32_t size, addr;
160 
161     if ((nargs != 2) || (nret != 2)) {
162         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
163         return;
164     }
165 
166     size = rtas_ld(args, 1);
167     addr = rtas_ld(args, 0);
168 
169     finish_read_pci_config(spapr, 0, addr, size, rets);
170 }
171 
172 static void finish_write_pci_config(SpaprMachineState *spapr, uint64_t buid,
173                                     uint32_t addr, uint32_t size,
174                                     uint32_t val, target_ulong rets)
175 {
176     PCIDevice *pci_dev;
177 
178     if ((size != 1) && (size != 2) && (size != 4)) {
179         /* access must be 1, 2 or 4 bytes */
180         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
181         return;
182     }
183 
184     pci_dev = spapr_pci_find_dev(spapr, buid, addr);
185     addr = rtas_pci_cfgaddr(addr);
186 
187     if (!pci_dev || (addr % size) || (addr >= pci_config_size(pci_dev))) {
188         /* Access must be to a valid device, within bounds and
189          * naturally aligned */
190         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
191         return;
192     }
193 
194     pci_host_config_write_common(pci_dev, addr, pci_config_size(pci_dev),
195                                  val, size);
196 
197     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
198 }
199 
200 static void rtas_ibm_write_pci_config(PowerPCCPU *cpu, SpaprMachineState *spapr,
201                                       uint32_t token, uint32_t nargs,
202                                       target_ulong args,
203                                       uint32_t nret, target_ulong rets)
204 {
205     uint64_t buid;
206     uint32_t val, size, addr;
207 
208     if ((nargs != 5) || (nret != 1)) {
209         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
210         return;
211     }
212 
213     buid = rtas_ldq(args, 1);
214     val = rtas_ld(args, 4);
215     size = rtas_ld(args, 3);
216     addr = rtas_ld(args, 0);
217 
218     finish_write_pci_config(spapr, buid, addr, size, val, rets);
219 }
220 
221 static void rtas_write_pci_config(PowerPCCPU *cpu, SpaprMachineState *spapr,
222                                   uint32_t token, uint32_t nargs,
223                                   target_ulong args,
224                                   uint32_t nret, target_ulong rets)
225 {
226     uint32_t val, size, addr;
227 
228     if ((nargs != 3) || (nret != 1)) {
229         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
230         return;
231     }
232 
233 
234     val = rtas_ld(args, 2);
235     size = rtas_ld(args, 1);
236     addr = rtas_ld(args, 0);
237 
238     finish_write_pci_config(spapr, 0, addr, size, val, rets);
239 }
240 
241 /*
242  * Set MSI/MSIX message data.
243  * This is required for msi_notify()/msix_notify() which
244  * will write at the addresses via spapr_msi_write().
245  *
246  * If hwaddr == 0, all entries will have .data == first_irq i.e.
247  * table will be reset.
248  */
249 static void spapr_msi_setmsg(PCIDevice *pdev, hwaddr addr, bool msix,
250                              unsigned first_irq, unsigned req_num)
251 {
252     unsigned i;
253     MSIMessage msg = { .address = addr, .data = first_irq };
254 
255     if (!msix) {
256         msi_set_message(pdev, msg);
257         trace_spapr_pci_msi_setup(pdev->name, 0, msg.address);
258         return;
259     }
260 
261     for (i = 0; i < req_num; ++i) {
262         msix_set_message(pdev, i, msg);
263         trace_spapr_pci_msi_setup(pdev->name, i, msg.address);
264         if (addr) {
265             ++msg.data;
266         }
267     }
268 }
269 
270 static void rtas_ibm_change_msi(PowerPCCPU *cpu, SpaprMachineState *spapr,
271                                 uint32_t token, uint32_t nargs,
272                                 target_ulong args, uint32_t nret,
273                                 target_ulong rets)
274 {
275     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
276     uint32_t config_addr = rtas_ld(args, 0);
277     uint64_t buid = rtas_ldq(args, 1);
278     unsigned int func = rtas_ld(args, 3);
279     unsigned int req_num = rtas_ld(args, 4); /* 0 == remove all */
280     unsigned int seq_num = rtas_ld(args, 5);
281     unsigned int ret_intr_type;
282     unsigned int irq, max_irqs = 0;
283     SpaprPhbState *phb = NULL;
284     PCIDevice *pdev = NULL;
285     SpaprPciMsi *msi;
286     int *config_addr_key;
287     Error *err = NULL;
288     int i;
289 
290     /* Fins SpaprPhbState */
291     phb = spapr_pci_find_phb(spapr, buid);
292     if (phb) {
293         pdev = spapr_pci_find_dev(spapr, buid, config_addr);
294     }
295     if (!phb || !pdev) {
296         rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
297         return;
298     }
299 
300     switch (func) {
301     case RTAS_CHANGE_FN:
302         if (msi_present(pdev)) {
303             ret_intr_type = RTAS_TYPE_MSI;
304         } else if (msix_present(pdev)) {
305             ret_intr_type = RTAS_TYPE_MSIX;
306         } else {
307             rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
308             return;
309         }
310         break;
311     case RTAS_CHANGE_MSI_FN:
312         if (msi_present(pdev)) {
313             ret_intr_type = RTAS_TYPE_MSI;
314         } else {
315             rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
316             return;
317         }
318         break;
319     case RTAS_CHANGE_MSIX_FN:
320         if (msix_present(pdev)) {
321             ret_intr_type = RTAS_TYPE_MSIX;
322         } else {
323             rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
324             return;
325         }
326         break;
327     default:
328         error_report("rtas_ibm_change_msi(%u) is not implemented", func);
329         rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
330         return;
331     }
332 
333     msi = (SpaprPciMsi *) g_hash_table_lookup(phb->msi, &config_addr);
334 
335     /* Releasing MSIs */
336     if (!req_num) {
337         if (!msi) {
338             trace_spapr_pci_msi("Releasing wrong config", config_addr);
339             rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
340             return;
341         }
342 
343         if (msi_present(pdev)) {
344             spapr_msi_setmsg(pdev, 0, false, 0, 0);
345         }
346         if (msix_present(pdev)) {
347             spapr_msi_setmsg(pdev, 0, true, 0, 0);
348         }
349         g_hash_table_remove(phb->msi, &config_addr);
350 
351         trace_spapr_pci_msi("Released MSIs", config_addr);
352         rtas_st(rets, 0, RTAS_OUT_SUCCESS);
353         rtas_st(rets, 1, 0);
354         return;
355     }
356 
357     /* Enabling MSI */
358 
359     /* Check if the device supports as many IRQs as requested */
360     if (ret_intr_type == RTAS_TYPE_MSI) {
361         max_irqs = msi_nr_vectors_allocated(pdev);
362     } else if (ret_intr_type == RTAS_TYPE_MSIX) {
363         max_irqs = pdev->msix_entries_nr;
364     }
365     if (!max_irqs) {
366         error_report("Requested interrupt type %d is not enabled for device %x",
367                      ret_intr_type, config_addr);
368         rtas_st(rets, 0, -1); /* Hardware error */
369         return;
370     }
371     /* Correct the number if the guest asked for too many */
372     if (req_num > max_irqs) {
373         trace_spapr_pci_msi_retry(config_addr, req_num, max_irqs);
374         req_num = max_irqs;
375         irq = 0; /* to avoid misleading trace */
376         goto out;
377     }
378 
379     /* Allocate MSIs */
380     if (smc->legacy_irq_allocation) {
381         irq = spapr_irq_find(spapr, req_num, ret_intr_type == RTAS_TYPE_MSI,
382                              &err);
383     } else {
384         irq = spapr_irq_msi_alloc(spapr, req_num,
385                                   ret_intr_type == RTAS_TYPE_MSI, &err);
386     }
387     if (err) {
388         error_reportf_err(err, "Can't allocate MSIs for device %x: ",
389                           config_addr);
390         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
391         return;
392     }
393 
394     for (i = 0; i < req_num; i++) {
395         spapr_irq_claim(spapr, irq + i, false, &err);
396         if (err) {
397             if (i) {
398                 spapr_irq_free(spapr, irq, i);
399             }
400             if (!smc->legacy_irq_allocation) {
401                 spapr_irq_msi_free(spapr, irq, req_num);
402             }
403             error_reportf_err(err, "Can't allocate MSIs for device %x: ",
404                               config_addr);
405             rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
406             return;
407         }
408     }
409 
410     /* Release previous MSIs */
411     if (msi) {
412         g_hash_table_remove(phb->msi, &config_addr);
413     }
414 
415     /* Setup MSI/MSIX vectors in the device (via cfgspace or MSIX BAR) */
416     spapr_msi_setmsg(pdev, SPAPR_PCI_MSI_WINDOW, ret_intr_type == RTAS_TYPE_MSIX,
417                      irq, req_num);
418 
419     /* Add MSI device to cache */
420     msi = g_new(SpaprPciMsi, 1);
421     msi->first_irq = irq;
422     msi->num = req_num;
423     config_addr_key = g_new(int, 1);
424     *config_addr_key = config_addr;
425     g_hash_table_insert(phb->msi, config_addr_key, msi);
426 
427 out:
428     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
429     rtas_st(rets, 1, req_num);
430     rtas_st(rets, 2, ++seq_num);
431     if (nret > 3) {
432         rtas_st(rets, 3, ret_intr_type);
433     }
434 
435     trace_spapr_pci_rtas_ibm_change_msi(config_addr, func, req_num, irq);
436 }
437 
438 static void rtas_ibm_query_interrupt_source_number(PowerPCCPU *cpu,
439                                                    SpaprMachineState *spapr,
440                                                    uint32_t token,
441                                                    uint32_t nargs,
442                                                    target_ulong args,
443                                                    uint32_t nret,
444                                                    target_ulong rets)
445 {
446     uint32_t config_addr = rtas_ld(args, 0);
447     uint64_t buid = rtas_ldq(args, 1);
448     unsigned int intr_src_num = -1, ioa_intr_num = rtas_ld(args, 3);
449     SpaprPhbState *phb = NULL;
450     PCIDevice *pdev = NULL;
451     SpaprPciMsi *msi;
452 
453     /* Find SpaprPhbState */
454     phb = spapr_pci_find_phb(spapr, buid);
455     if (phb) {
456         pdev = spapr_pci_find_dev(spapr, buid, config_addr);
457     }
458     if (!phb || !pdev) {
459         rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
460         return;
461     }
462 
463     /* Find device descriptor and start IRQ */
464     msi = (SpaprPciMsi *) g_hash_table_lookup(phb->msi, &config_addr);
465     if (!msi || !msi->first_irq || !msi->num || (ioa_intr_num >= msi->num)) {
466         trace_spapr_pci_msi("Failed to return vector", config_addr);
467         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
468         return;
469     }
470     intr_src_num = msi->first_irq + ioa_intr_num;
471     trace_spapr_pci_rtas_ibm_query_interrupt_source_number(ioa_intr_num,
472                                                            intr_src_num);
473 
474     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
475     rtas_st(rets, 1, intr_src_num);
476     rtas_st(rets, 2, 1);/* 0 == level; 1 == edge */
477 }
478 
479 static void rtas_ibm_set_eeh_option(PowerPCCPU *cpu,
480                                     SpaprMachineState *spapr,
481                                     uint32_t token, uint32_t nargs,
482                                     target_ulong args, uint32_t nret,
483                                     target_ulong rets)
484 {
485     SpaprPhbState *sphb;
486     uint32_t addr, option;
487     uint64_t buid;
488     int ret;
489 
490     if ((nargs != 4) || (nret != 1)) {
491         goto param_error_exit;
492     }
493 
494     buid = rtas_ldq(args, 1);
495     addr = rtas_ld(args, 0);
496     option = rtas_ld(args, 3);
497 
498     sphb = spapr_pci_find_phb(spapr, buid);
499     if (!sphb) {
500         goto param_error_exit;
501     }
502 
503     if (!spapr_phb_eeh_available(sphb)) {
504         goto param_error_exit;
505     }
506 
507     ret = spapr_phb_vfio_eeh_set_option(sphb, addr, option);
508     rtas_st(rets, 0, ret);
509     return;
510 
511 param_error_exit:
512     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
513 }
514 
515 static void rtas_ibm_get_config_addr_info2(PowerPCCPU *cpu,
516                                            SpaprMachineState *spapr,
517                                            uint32_t token, uint32_t nargs,
518                                            target_ulong args, uint32_t nret,
519                                            target_ulong rets)
520 {
521     SpaprPhbState *sphb;
522     PCIDevice *pdev;
523     uint32_t addr, option;
524     uint64_t buid;
525 
526     if ((nargs != 4) || (nret != 2)) {
527         goto param_error_exit;
528     }
529 
530     buid = rtas_ldq(args, 1);
531     sphb = spapr_pci_find_phb(spapr, buid);
532     if (!sphb) {
533         goto param_error_exit;
534     }
535 
536     if (!spapr_phb_eeh_available(sphb)) {
537         goto param_error_exit;
538     }
539 
540     /*
541      * We always have PE address of form "00BB0001". "BB"
542      * represents the bus number of PE's primary bus.
543      */
544     option = rtas_ld(args, 3);
545     switch (option) {
546     case RTAS_GET_PE_ADDR:
547         addr = rtas_ld(args, 0);
548         pdev = spapr_pci_find_dev(spapr, buid, addr);
549         if (!pdev) {
550             goto param_error_exit;
551         }
552 
553         rtas_st(rets, 1, (pci_bus_num(pci_get_bus(pdev)) << 16) + 1);
554         break;
555     case RTAS_GET_PE_MODE:
556         rtas_st(rets, 1, RTAS_PE_MODE_SHARED);
557         break;
558     default:
559         goto param_error_exit;
560     }
561 
562     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
563     return;
564 
565 param_error_exit:
566     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
567 }
568 
569 static void rtas_ibm_read_slot_reset_state2(PowerPCCPU *cpu,
570                                             SpaprMachineState *spapr,
571                                             uint32_t token, uint32_t nargs,
572                                             target_ulong args, uint32_t nret,
573                                             target_ulong rets)
574 {
575     SpaprPhbState *sphb;
576     uint64_t buid;
577     int state, ret;
578 
579     if ((nargs != 3) || (nret != 4 && nret != 5)) {
580         goto param_error_exit;
581     }
582 
583     buid = rtas_ldq(args, 1);
584     sphb = spapr_pci_find_phb(spapr, buid);
585     if (!sphb) {
586         goto param_error_exit;
587     }
588 
589     if (!spapr_phb_eeh_available(sphb)) {
590         goto param_error_exit;
591     }
592 
593     ret = spapr_phb_vfio_eeh_get_state(sphb, &state);
594     rtas_st(rets, 0, ret);
595     if (ret != RTAS_OUT_SUCCESS) {
596         return;
597     }
598 
599     rtas_st(rets, 1, state);
600     rtas_st(rets, 2, RTAS_EEH_SUPPORT);
601     rtas_st(rets, 3, RTAS_EEH_PE_UNAVAIL_INFO);
602     if (nret >= 5) {
603         rtas_st(rets, 4, RTAS_EEH_PE_RECOVER_INFO);
604     }
605     return;
606 
607 param_error_exit:
608     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
609 }
610 
611 static void rtas_ibm_set_slot_reset(PowerPCCPU *cpu,
612                                     SpaprMachineState *spapr,
613                                     uint32_t token, uint32_t nargs,
614                                     target_ulong args, uint32_t nret,
615                                     target_ulong rets)
616 {
617     SpaprPhbState *sphb;
618     uint32_t option;
619     uint64_t buid;
620     int ret;
621 
622     if ((nargs != 4) || (nret != 1)) {
623         goto param_error_exit;
624     }
625 
626     buid = rtas_ldq(args, 1);
627     option = rtas_ld(args, 3);
628     sphb = spapr_pci_find_phb(spapr, buid);
629     if (!sphb) {
630         goto param_error_exit;
631     }
632 
633     if (!spapr_phb_eeh_available(sphb)) {
634         goto param_error_exit;
635     }
636 
637     ret = spapr_phb_vfio_eeh_reset(sphb, option);
638     rtas_st(rets, 0, ret);
639     return;
640 
641 param_error_exit:
642     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
643 }
644 
645 static void rtas_ibm_configure_pe(PowerPCCPU *cpu,
646                                   SpaprMachineState *spapr,
647                                   uint32_t token, uint32_t nargs,
648                                   target_ulong args, uint32_t nret,
649                                   target_ulong rets)
650 {
651     SpaprPhbState *sphb;
652     uint64_t buid;
653     int ret;
654 
655     if ((nargs != 3) || (nret != 1)) {
656         goto param_error_exit;
657     }
658 
659     buid = rtas_ldq(args, 1);
660     sphb = spapr_pci_find_phb(spapr, buid);
661     if (!sphb) {
662         goto param_error_exit;
663     }
664 
665     if (!spapr_phb_eeh_available(sphb)) {
666         goto param_error_exit;
667     }
668 
669     ret = spapr_phb_vfio_eeh_configure(sphb);
670     rtas_st(rets, 0, ret);
671     return;
672 
673 param_error_exit:
674     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
675 }
676 
677 /* To support it later */
678 static void rtas_ibm_slot_error_detail(PowerPCCPU *cpu,
679                                        SpaprMachineState *spapr,
680                                        uint32_t token, uint32_t nargs,
681                                        target_ulong args, uint32_t nret,
682                                        target_ulong rets)
683 {
684     SpaprPhbState *sphb;
685     int option;
686     uint64_t buid;
687 
688     if ((nargs != 8) || (nret != 1)) {
689         goto param_error_exit;
690     }
691 
692     buid = rtas_ldq(args, 1);
693     sphb = spapr_pci_find_phb(spapr, buid);
694     if (!sphb) {
695         goto param_error_exit;
696     }
697 
698     if (!spapr_phb_eeh_available(sphb)) {
699         goto param_error_exit;
700     }
701 
702     option = rtas_ld(args, 7);
703     switch (option) {
704     case RTAS_SLOT_TEMP_ERR_LOG:
705     case RTAS_SLOT_PERM_ERR_LOG:
706         break;
707     default:
708         goto param_error_exit;
709     }
710 
711     /* We don't have error log yet */
712     rtas_st(rets, 0, RTAS_OUT_NO_ERRORS_FOUND);
713     return;
714 
715 param_error_exit:
716     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
717 }
718 
719 static void pci_spapr_set_irq(void *opaque, int irq_num, int level)
720 {
721     /*
722      * Here we use the number returned by pci_swizzle_map_irq_fn to find a
723      * corresponding qemu_irq.
724      */
725     SpaprPhbState *phb = opaque;
726     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
727 
728     trace_spapr_pci_lsi_set(phb->dtbusname, irq_num, phb->lsi_table[irq_num].irq);
729     qemu_set_irq(spapr_qirq(spapr, phb->lsi_table[irq_num].irq), level);
730 }
731 
732 static PCIINTxRoute spapr_route_intx_pin_to_irq(void *opaque, int pin)
733 {
734     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(opaque);
735     PCIINTxRoute route;
736 
737     route.mode = PCI_INTX_ENABLED;
738     route.irq = sphb->lsi_table[pin].irq;
739 
740     return route;
741 }
742 
743 static uint64_t spapr_msi_read(void *opaque, hwaddr addr, unsigned size)
744 {
745     qemu_log_mask(LOG_GUEST_ERROR, "%s: invalid access\n", __func__);
746     return 0;
747 }
748 
749 /*
750  * MSI/MSIX memory region implementation.
751  * The handler handles both MSI and MSIX.
752  * The vector number is encoded in least bits in data.
753  */
754 static void spapr_msi_write(void *opaque, hwaddr addr,
755                             uint64_t data, unsigned size)
756 {
757     SpaprMachineState *spapr = opaque;
758     uint32_t irq = data;
759 
760     trace_spapr_pci_msi_write(addr, data, irq);
761 
762     qemu_irq_pulse(spapr_qirq(spapr, irq));
763 }
764 
765 static const MemoryRegionOps spapr_msi_ops = {
766     /*
767      * .read result is undefined by PCI spec.
768      * define .read method to avoid assert failure in memory_region_init_io
769      */
770     .read = spapr_msi_read,
771     .write = spapr_msi_write,
772     .endianness = DEVICE_LITTLE_ENDIAN
773 };
774 
775 /*
776  * PHB PCI device
777  */
778 static AddressSpace *spapr_pci_dma_iommu(PCIBus *bus, void *opaque, int devfn)
779 {
780     SpaprPhbState *phb = opaque;
781 
782     return &phb->iommu_as;
783 }
784 
785 static char *spapr_phb_vfio_get_loc_code(SpaprPhbState *sphb,  PCIDevice *pdev)
786 {
787     char *path = NULL, *buf = NULL, *host = NULL;
788 
789     /* Get the PCI VFIO host id */
790     host = object_property_get_str(OBJECT(pdev), "host", NULL);
791     if (!host) {
792         goto err_out;
793     }
794 
795     /* Construct the path of the file that will give us the DT location */
796     path = g_strdup_printf("/sys/bus/pci/devices/%s/devspec", host);
797     g_free(host);
798     if (!g_file_get_contents(path, &buf, NULL, NULL)) {
799         goto err_out;
800     }
801     g_free(path);
802 
803     /* Construct and read from host device tree the loc-code */
804     path = g_strdup_printf("/proc/device-tree%s/ibm,loc-code", buf);
805     g_free(buf);
806     if (!g_file_get_contents(path, &buf, NULL, NULL)) {
807         goto err_out;
808     }
809     return buf;
810 
811 err_out:
812     g_free(path);
813     return NULL;
814 }
815 
816 static char *spapr_phb_get_loc_code(SpaprPhbState *sphb, PCIDevice *pdev)
817 {
818     char *buf;
819     const char *devtype = "qemu";
820     uint32_t busnr = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(pdev))));
821 
822     if (object_dynamic_cast(OBJECT(pdev), "vfio-pci")) {
823         buf = spapr_phb_vfio_get_loc_code(sphb, pdev);
824         if (buf) {
825             return buf;
826         }
827         devtype = "vfio";
828     }
829     /*
830      * For emulated devices and VFIO-failure case, make up
831      * the loc-code.
832      */
833     buf = g_strdup_printf("%s_%s:%04x:%02x:%02x.%x",
834                           devtype, pdev->name, sphb->index, busnr,
835                           PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
836     return buf;
837 }
838 
839 /* Macros to operate with address in OF binding to PCI */
840 #define b_x(x, p, l)    (((x) & ((1<<(l))-1)) << (p))
841 #define b_n(x)          b_x((x), 31, 1) /* 0 if relocatable */
842 #define b_p(x)          b_x((x), 30, 1) /* 1 if prefetchable */
843 #define b_t(x)          b_x((x), 29, 1) /* 1 if the address is aliased */
844 #define b_ss(x)         b_x((x), 24, 2) /* the space code */
845 #define b_bbbbbbbb(x)   b_x((x), 16, 8) /* bus number */
846 #define b_ddddd(x)      b_x((x), 11, 5) /* device number */
847 #define b_fff(x)        b_x((x), 8, 3)  /* function number */
848 #define b_rrrrrrrr(x)   b_x((x), 0, 8)  /* register number */
849 
850 /* for 'reg' OF properties */
851 #define RESOURCE_CELLS_SIZE 2
852 #define RESOURCE_CELLS_ADDRESS 3
853 
854 typedef struct ResourceFields {
855     uint32_t phys_hi;
856     uint32_t phys_mid;
857     uint32_t phys_lo;
858     uint32_t size_hi;
859     uint32_t size_lo;
860 } QEMU_PACKED ResourceFields;
861 
862 typedef struct ResourceProps {
863     ResourceFields reg[8];
864     uint32_t reg_len;
865 } ResourceProps;
866 
867 /* fill in the 'reg' OF properties for
868  * a PCI device. 'reg' describes resource requirements for a
869  * device's IO/MEM regions.
870  *
871  * the property is an array of ('phys-addr', 'size') pairs describing
872  * the addressable regions of the PCI device, where 'phys-addr' is a
873  * RESOURCE_CELLS_ADDRESS-tuple of 32-bit integers corresponding to
874  * (phys.hi, phys.mid, phys.lo), and 'size' is a
875  * RESOURCE_CELLS_SIZE-tuple corresponding to (size.hi, size.lo).
876  *
877  * phys.hi = 0xYYXXXXZZ, where:
878  *   0xYY = npt000ss
879  *          |||   |
880  *          |||   +-- space code
881  *          |||               |
882  *          |||               +  00 if configuration space
883  *          |||               +  01 if IO region,
884  *          |||               +  10 if 32-bit MEM region
885  *          |||               +  11 if 64-bit MEM region
886  *          |||
887  *          ||+------ for non-relocatable IO: 1 if aliased
888  *          ||        for relocatable IO: 1 if below 64KB
889  *          ||        for MEM: 1 if below 1MB
890  *          |+------- 1 if region is prefetchable
891  *          +-------- 1 if region is non-relocatable
892  *   0xXXXX = bbbbbbbb dddddfff, encoding bus, slot, and function
893  *            bits respectively
894  *   0xZZ = rrrrrrrr, the register number of the BAR corresponding
895  *          to the region
896  *
897  * phys.mid and phys.lo correspond respectively to the hi/lo portions
898  * of the actual address of the region.
899  *
900  * note also that addresses defined in this property are, at least
901  * for PAPR guests, relative to the PHBs IO/MEM windows, and
902  * correspond directly to the addresses in the BARs.
903  *
904  * in accordance with PCI Bus Binding to Open Firmware,
905  * IEEE Std 1275-1994, section 4.1.1, as implemented by PAPR+ v2.7,
906  * Appendix C.
907  */
908 static void populate_resource_props(PCIDevice *d, ResourceProps *rp)
909 {
910     int bus_num = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(d))));
911     uint32_t dev_id = (b_bbbbbbbb(bus_num) |
912                        b_ddddd(PCI_SLOT(d->devfn)) |
913                        b_fff(PCI_FUNC(d->devfn)));
914     ResourceFields *reg;
915     int i, reg_idx = 0;
916 
917     /* config space region */
918     reg = &rp->reg[reg_idx++];
919     reg->phys_hi = cpu_to_be32(dev_id);
920     reg->phys_mid = 0;
921     reg->phys_lo = 0;
922     reg->size_hi = 0;
923     reg->size_lo = 0;
924 
925     for (i = 0; i < PCI_NUM_REGIONS; i++) {
926         if (!d->io_regions[i].size) {
927             continue;
928         }
929 
930         reg = &rp->reg[reg_idx++];
931 
932         reg->phys_hi = cpu_to_be32(dev_id | b_rrrrrrrr(pci_bar(d, i)));
933         if (d->io_regions[i].type & PCI_BASE_ADDRESS_SPACE_IO) {
934             reg->phys_hi |= cpu_to_be32(b_ss(1));
935         } else if (d->io_regions[i].type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
936             reg->phys_hi |= cpu_to_be32(b_ss(3));
937         } else {
938             reg->phys_hi |= cpu_to_be32(b_ss(2));
939         }
940         reg->phys_mid = 0;
941         reg->phys_lo = 0;
942         reg->size_hi = cpu_to_be32(d->io_regions[i].size >> 32);
943         reg->size_lo = cpu_to_be32(d->io_regions[i].size);
944     }
945 
946     rp->reg_len = reg_idx * sizeof(ResourceFields);
947 }
948 
949 typedef struct PCIClass PCIClass;
950 typedef struct PCISubClass PCISubClass;
951 typedef struct PCIIFace PCIIFace;
952 
953 struct PCIIFace {
954     int iface;
955     const char *name;
956 };
957 
958 struct PCISubClass {
959     int subclass;
960     const char *name;
961     const PCIIFace *iface;
962 };
963 
964 struct PCIClass {
965     const char *name;
966     const PCISubClass *subc;
967 };
968 
969 static const PCISubClass undef_subclass[] = {
970     { PCI_CLASS_NOT_DEFINED_VGA, "display", NULL },
971     { 0xFF, NULL, NULL },
972 };
973 
974 static const PCISubClass mass_subclass[] = {
975     { PCI_CLASS_STORAGE_SCSI, "scsi", NULL },
976     { PCI_CLASS_STORAGE_IDE, "ide", NULL },
977     { PCI_CLASS_STORAGE_FLOPPY, "fdc", NULL },
978     { PCI_CLASS_STORAGE_IPI, "ipi", NULL },
979     { PCI_CLASS_STORAGE_RAID, "raid", NULL },
980     { PCI_CLASS_STORAGE_ATA, "ata", NULL },
981     { PCI_CLASS_STORAGE_SATA, "sata", NULL },
982     { PCI_CLASS_STORAGE_SAS, "sas", NULL },
983     { 0xFF, NULL, NULL },
984 };
985 
986 static const PCISubClass net_subclass[] = {
987     { PCI_CLASS_NETWORK_ETHERNET, "ethernet", NULL },
988     { PCI_CLASS_NETWORK_TOKEN_RING, "token-ring", NULL },
989     { PCI_CLASS_NETWORK_FDDI, "fddi", NULL },
990     { PCI_CLASS_NETWORK_ATM, "atm", NULL },
991     { PCI_CLASS_NETWORK_ISDN, "isdn", NULL },
992     { PCI_CLASS_NETWORK_WORLDFIP, "worldfip", NULL },
993     { PCI_CLASS_NETWORK_PICMG214, "picmg", NULL },
994     { 0xFF, NULL, NULL },
995 };
996 
997 static const PCISubClass displ_subclass[] = {
998     { PCI_CLASS_DISPLAY_VGA, "vga", NULL },
999     { PCI_CLASS_DISPLAY_XGA, "xga", NULL },
1000     { PCI_CLASS_DISPLAY_3D, "3d-controller", NULL },
1001     { 0xFF, NULL, NULL },
1002 };
1003 
1004 static const PCISubClass media_subclass[] = {
1005     { PCI_CLASS_MULTIMEDIA_VIDEO, "video", NULL },
1006     { PCI_CLASS_MULTIMEDIA_AUDIO, "sound", NULL },
1007     { PCI_CLASS_MULTIMEDIA_PHONE, "telephony", NULL },
1008     { 0xFF, NULL, NULL },
1009 };
1010 
1011 static const PCISubClass mem_subclass[] = {
1012     { PCI_CLASS_MEMORY_RAM, "memory", NULL },
1013     { PCI_CLASS_MEMORY_FLASH, "flash", NULL },
1014     { 0xFF, NULL, NULL },
1015 };
1016 
1017 static const PCISubClass bridg_subclass[] = {
1018     { PCI_CLASS_BRIDGE_HOST, "host", NULL },
1019     { PCI_CLASS_BRIDGE_ISA, "isa", NULL },
1020     { PCI_CLASS_BRIDGE_EISA, "eisa", NULL },
1021     { PCI_CLASS_BRIDGE_MC, "mca", NULL },
1022     { PCI_CLASS_BRIDGE_PCI, "pci", NULL },
1023     { PCI_CLASS_BRIDGE_PCMCIA, "pcmcia", NULL },
1024     { PCI_CLASS_BRIDGE_NUBUS, "nubus", NULL },
1025     { PCI_CLASS_BRIDGE_CARDBUS, "cardbus", NULL },
1026     { PCI_CLASS_BRIDGE_RACEWAY, "raceway", NULL },
1027     { PCI_CLASS_BRIDGE_PCI_SEMITP, "semi-transparent-pci", NULL },
1028     { PCI_CLASS_BRIDGE_IB_PCI, "infiniband", NULL },
1029     { 0xFF, NULL, NULL },
1030 };
1031 
1032 static const PCISubClass comm_subclass[] = {
1033     { PCI_CLASS_COMMUNICATION_SERIAL, "serial", NULL },
1034     { PCI_CLASS_COMMUNICATION_PARALLEL, "parallel", NULL },
1035     { PCI_CLASS_COMMUNICATION_MULTISERIAL, "multiport-serial", NULL },
1036     { PCI_CLASS_COMMUNICATION_MODEM, "modem", NULL },
1037     { PCI_CLASS_COMMUNICATION_GPIB, "gpib", NULL },
1038     { PCI_CLASS_COMMUNICATION_SC, "smart-card", NULL },
1039     { 0xFF, NULL, NULL, },
1040 };
1041 
1042 static const PCIIFace pic_iface[] = {
1043     { PCI_CLASS_SYSTEM_PIC_IOAPIC, "io-apic" },
1044     { PCI_CLASS_SYSTEM_PIC_IOXAPIC, "io-xapic" },
1045     { 0xFF, NULL },
1046 };
1047 
1048 static const PCISubClass sys_subclass[] = {
1049     { PCI_CLASS_SYSTEM_PIC, "interrupt-controller", pic_iface },
1050     { PCI_CLASS_SYSTEM_DMA, "dma-controller", NULL },
1051     { PCI_CLASS_SYSTEM_TIMER, "timer", NULL },
1052     { PCI_CLASS_SYSTEM_RTC, "rtc", NULL },
1053     { PCI_CLASS_SYSTEM_PCI_HOTPLUG, "hot-plug-controller", NULL },
1054     { PCI_CLASS_SYSTEM_SDHCI, "sd-host-controller", NULL },
1055     { 0xFF, NULL, NULL },
1056 };
1057 
1058 static const PCISubClass inp_subclass[] = {
1059     { PCI_CLASS_INPUT_KEYBOARD, "keyboard", NULL },
1060     { PCI_CLASS_INPUT_PEN, "pen", NULL },
1061     { PCI_CLASS_INPUT_MOUSE, "mouse", NULL },
1062     { PCI_CLASS_INPUT_SCANNER, "scanner", NULL },
1063     { PCI_CLASS_INPUT_GAMEPORT, "gameport", NULL },
1064     { 0xFF, NULL, NULL },
1065 };
1066 
1067 static const PCISubClass dock_subclass[] = {
1068     { PCI_CLASS_DOCKING_GENERIC, "dock", NULL },
1069     { 0xFF, NULL, NULL },
1070 };
1071 
1072 static const PCISubClass cpu_subclass[] = {
1073     { PCI_CLASS_PROCESSOR_PENTIUM, "pentium", NULL },
1074     { PCI_CLASS_PROCESSOR_POWERPC, "powerpc", NULL },
1075     { PCI_CLASS_PROCESSOR_MIPS, "mips", NULL },
1076     { PCI_CLASS_PROCESSOR_CO, "co-processor", NULL },
1077     { 0xFF, NULL, NULL },
1078 };
1079 
1080 static const PCIIFace usb_iface[] = {
1081     { PCI_CLASS_SERIAL_USB_UHCI, "usb-uhci" },
1082     { PCI_CLASS_SERIAL_USB_OHCI, "usb-ohci", },
1083     { PCI_CLASS_SERIAL_USB_EHCI, "usb-ehci" },
1084     { PCI_CLASS_SERIAL_USB_XHCI, "usb-xhci" },
1085     { PCI_CLASS_SERIAL_USB_UNKNOWN, "usb-unknown" },
1086     { PCI_CLASS_SERIAL_USB_DEVICE, "usb-device" },
1087     { 0xFF, NULL },
1088 };
1089 
1090 static const PCISubClass ser_subclass[] = {
1091     { PCI_CLASS_SERIAL_FIREWIRE, "firewire", NULL },
1092     { PCI_CLASS_SERIAL_ACCESS, "access-bus", NULL },
1093     { PCI_CLASS_SERIAL_SSA, "ssa", NULL },
1094     { PCI_CLASS_SERIAL_USB, "usb", usb_iface },
1095     { PCI_CLASS_SERIAL_FIBER, "fibre-channel", NULL },
1096     { PCI_CLASS_SERIAL_SMBUS, "smb", NULL },
1097     { PCI_CLASS_SERIAL_IB, "infiniband", NULL },
1098     { PCI_CLASS_SERIAL_IPMI, "ipmi", NULL },
1099     { PCI_CLASS_SERIAL_SERCOS, "sercos", NULL },
1100     { PCI_CLASS_SERIAL_CANBUS, "canbus", NULL },
1101     { 0xFF, NULL, NULL },
1102 };
1103 
1104 static const PCISubClass wrl_subclass[] = {
1105     { PCI_CLASS_WIRELESS_IRDA, "irda", NULL },
1106     { PCI_CLASS_WIRELESS_CIR, "consumer-ir", NULL },
1107     { PCI_CLASS_WIRELESS_RF_CONTROLLER, "rf-controller", NULL },
1108     { PCI_CLASS_WIRELESS_BLUETOOTH, "bluetooth", NULL },
1109     { PCI_CLASS_WIRELESS_BROADBAND, "broadband", NULL },
1110     { 0xFF, NULL, NULL },
1111 };
1112 
1113 static const PCISubClass sat_subclass[] = {
1114     { PCI_CLASS_SATELLITE_TV, "satellite-tv", NULL },
1115     { PCI_CLASS_SATELLITE_AUDIO, "satellite-audio", NULL },
1116     { PCI_CLASS_SATELLITE_VOICE, "satellite-voice", NULL },
1117     { PCI_CLASS_SATELLITE_DATA, "satellite-data", NULL },
1118     { 0xFF, NULL, NULL },
1119 };
1120 
1121 static const PCISubClass crypt_subclass[] = {
1122     { PCI_CLASS_CRYPT_NETWORK, "network-encryption", NULL },
1123     { PCI_CLASS_CRYPT_ENTERTAINMENT,
1124       "entertainment-encryption", NULL },
1125     { 0xFF, NULL, NULL },
1126 };
1127 
1128 static const PCISubClass spc_subclass[] = {
1129     { PCI_CLASS_SP_DPIO, "dpio", NULL },
1130     { PCI_CLASS_SP_PERF, "counter", NULL },
1131     { PCI_CLASS_SP_SYNCH, "measurement", NULL },
1132     { PCI_CLASS_SP_MANAGEMENT, "management-card", NULL },
1133     { 0xFF, NULL, NULL },
1134 };
1135 
1136 static const PCIClass pci_classes[] = {
1137     { "legacy-device", undef_subclass },
1138     { "mass-storage",  mass_subclass },
1139     { "network", net_subclass },
1140     { "display", displ_subclass, },
1141     { "multimedia-device", media_subclass },
1142     { "memory-controller", mem_subclass },
1143     { "unknown-bridge", bridg_subclass },
1144     { "communication-controller", comm_subclass},
1145     { "system-peripheral", sys_subclass },
1146     { "input-controller", inp_subclass },
1147     { "docking-station", dock_subclass },
1148     { "cpu", cpu_subclass },
1149     { "serial-bus", ser_subclass },
1150     { "wireless-controller", wrl_subclass },
1151     { "intelligent-io", NULL },
1152     { "satellite-device", sat_subclass },
1153     { "encryption", crypt_subclass },
1154     { "data-processing-controller", spc_subclass },
1155 };
1156 
1157 static const char *dt_name_from_class(uint8_t class, uint8_t subclass,
1158                                       uint8_t iface)
1159 {
1160     const PCIClass *pclass;
1161     const PCISubClass *psubclass;
1162     const PCIIFace *piface;
1163     const char *name;
1164 
1165     if (class >= ARRAY_SIZE(pci_classes)) {
1166         return "pci";
1167     }
1168 
1169     pclass = pci_classes + class;
1170     name = pclass->name;
1171 
1172     if (pclass->subc == NULL) {
1173         return name;
1174     }
1175 
1176     psubclass = pclass->subc;
1177     while ((psubclass->subclass & 0xff) != 0xff) {
1178         if ((psubclass->subclass & 0xff) == subclass) {
1179             name = psubclass->name;
1180             break;
1181         }
1182         psubclass++;
1183     }
1184 
1185     piface = psubclass->iface;
1186     if (piface == NULL) {
1187         return name;
1188     }
1189     while ((piface->iface & 0xff) != 0xff) {
1190         if ((piface->iface & 0xff) == iface) {
1191             name = piface->name;
1192             break;
1193         }
1194         piface++;
1195     }
1196 
1197     return name;
1198 }
1199 
1200 /*
1201  * DRC helper functions
1202  */
1203 
1204 static uint32_t drc_id_from_devfn(SpaprPhbState *phb,
1205                                   uint8_t chassis, int32_t devfn)
1206 {
1207     return (phb->index << 16) | (chassis << 8) | devfn;
1208 }
1209 
1210 static SpaprDrc *drc_from_devfn(SpaprPhbState *phb,
1211                                 uint8_t chassis, int32_t devfn)
1212 {
1213     return spapr_drc_by_id(TYPE_SPAPR_DRC_PCI,
1214                            drc_id_from_devfn(phb, chassis, devfn));
1215 }
1216 
1217 static uint8_t chassis_from_bus(PCIBus *bus)
1218 {
1219     if (pci_bus_is_root(bus)) {
1220         return 0;
1221     } else {
1222         PCIDevice *bridge = pci_bridge_get_device(bus);
1223 
1224         return object_property_get_uint(OBJECT(bridge), "chassis_nr",
1225                                         &error_abort);
1226     }
1227 }
1228 
1229 static SpaprDrc *drc_from_dev(SpaprPhbState *phb, PCIDevice *dev)
1230 {
1231     uint8_t chassis = chassis_from_bus(pci_get_bus(dev));
1232 
1233     return drc_from_devfn(phb, chassis, dev->devfn);
1234 }
1235 
1236 static void add_drcs(SpaprPhbState *phb, PCIBus *bus)
1237 {
1238     Object *owner;
1239     int i;
1240     uint8_t chassis;
1241 
1242     if (!phb->dr_enabled) {
1243         return;
1244     }
1245 
1246     chassis = chassis_from_bus(bus);
1247 
1248     if (pci_bus_is_root(bus)) {
1249         owner = OBJECT(phb);
1250     } else {
1251         owner = OBJECT(pci_bridge_get_device(bus));
1252     }
1253 
1254     for (i = 0; i < PCI_SLOT_MAX * PCI_FUNC_MAX; i++) {
1255         spapr_dr_connector_new(owner, TYPE_SPAPR_DRC_PCI,
1256                                drc_id_from_devfn(phb, chassis, i));
1257     }
1258 }
1259 
1260 static void remove_drcs(SpaprPhbState *phb, PCIBus *bus)
1261 {
1262     int i;
1263     uint8_t chassis;
1264 
1265     if (!phb->dr_enabled) {
1266         return;
1267     }
1268 
1269     chassis = chassis_from_bus(bus);
1270 
1271     for (i = PCI_SLOT_MAX * PCI_FUNC_MAX - 1; i >= 0; i--) {
1272         SpaprDrc *drc = drc_from_devfn(phb, chassis, i);
1273 
1274         if (drc) {
1275             object_unparent(OBJECT(drc));
1276         }
1277     }
1278 }
1279 
1280 typedef struct PciWalkFdt {
1281     void *fdt;
1282     int offset;
1283     SpaprPhbState *sphb;
1284     int err;
1285 } PciWalkFdt;
1286 
1287 static int spapr_dt_pci_device(SpaprPhbState *sphb, PCIDevice *dev,
1288                                void *fdt, int parent_offset);
1289 
1290 static void spapr_dt_pci_device_cb(PCIBus *bus, PCIDevice *pdev,
1291                                    void *opaque)
1292 {
1293     PciWalkFdt *p = opaque;
1294     int err;
1295 
1296     if (p->err) {
1297         /* Something's already broken, don't keep going */
1298         return;
1299     }
1300 
1301     err = spapr_dt_pci_device(p->sphb, pdev, p->fdt, p->offset);
1302     if (err < 0) {
1303         p->err = err;
1304     }
1305 }
1306 
1307 /* Augment PCI device node with bridge specific information */
1308 static int spapr_dt_pci_bus(SpaprPhbState *sphb, PCIBus *bus,
1309                                void *fdt, int offset)
1310 {
1311     Object *owner;
1312     PciWalkFdt cbinfo = {
1313         .fdt = fdt,
1314         .offset = offset,
1315         .sphb = sphb,
1316         .err = 0,
1317     };
1318     int ret;
1319 
1320     _FDT(fdt_setprop_cell(fdt, offset, "#address-cells",
1321                           RESOURCE_CELLS_ADDRESS));
1322     _FDT(fdt_setprop_cell(fdt, offset, "#size-cells",
1323                           RESOURCE_CELLS_SIZE));
1324 
1325     assert(bus);
1326     pci_for_each_device_reverse(bus, pci_bus_num(bus),
1327                                 spapr_dt_pci_device_cb, &cbinfo);
1328     if (cbinfo.err) {
1329         return cbinfo.err;
1330     }
1331 
1332     if (pci_bus_is_root(bus)) {
1333         owner = OBJECT(sphb);
1334     } else {
1335         owner = OBJECT(pci_bridge_get_device(bus));
1336     }
1337 
1338     ret = spapr_dt_drc(fdt, offset, owner,
1339                        SPAPR_DR_CONNECTOR_TYPE_PCI);
1340     if (ret) {
1341         return ret;
1342     }
1343 
1344     return offset;
1345 }
1346 
1347 /* create OF node for pci device and required OF DT properties */
1348 static int spapr_dt_pci_device(SpaprPhbState *sphb, PCIDevice *dev,
1349                                void *fdt, int parent_offset)
1350 {
1351     int offset;
1352     const gchar *basename;
1353     gchar *nodename;
1354     int slot = PCI_SLOT(dev->devfn);
1355     int func = PCI_FUNC(dev->devfn);
1356     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
1357     ResourceProps rp;
1358     SpaprDrc *drc = drc_from_dev(sphb, dev);
1359     uint32_t vendor_id = pci_default_read_config(dev, PCI_VENDOR_ID, 2);
1360     uint32_t device_id = pci_default_read_config(dev, PCI_DEVICE_ID, 2);
1361     uint32_t revision_id = pci_default_read_config(dev, PCI_REVISION_ID, 1);
1362     uint32_t ccode = pci_default_read_config(dev, PCI_CLASS_PROG, 3);
1363     uint32_t irq_pin = pci_default_read_config(dev, PCI_INTERRUPT_PIN, 1);
1364     uint32_t subsystem_id = pci_default_read_config(dev, PCI_SUBSYSTEM_ID, 2);
1365     uint32_t subsystem_vendor_id =
1366         pci_default_read_config(dev, PCI_SUBSYSTEM_VENDOR_ID, 2);
1367     uint32_t cache_line_size =
1368         pci_default_read_config(dev, PCI_CACHE_LINE_SIZE, 1);
1369     uint32_t pci_status = pci_default_read_config(dev, PCI_STATUS, 2);
1370     gchar *loc_code;
1371 
1372     basename = dt_name_from_class((ccode >> 16) & 0xff, (ccode >> 8) & 0xff,
1373                                   ccode & 0xff);
1374 
1375     if (func != 0) {
1376         nodename = g_strdup_printf("%s@%x,%x", basename, slot, func);
1377     } else {
1378         nodename = g_strdup_printf("%s@%x", basename, slot);
1379     }
1380 
1381     _FDT(offset = fdt_add_subnode(fdt, parent_offset, nodename));
1382 
1383     g_free(nodename);
1384 
1385     /* in accordance with PAPR+ v2.7 13.6.3, Table 181 */
1386     _FDT(fdt_setprop_cell(fdt, offset, "vendor-id", vendor_id));
1387     _FDT(fdt_setprop_cell(fdt, offset, "device-id", device_id));
1388     _FDT(fdt_setprop_cell(fdt, offset, "revision-id", revision_id));
1389 
1390     _FDT(fdt_setprop_cell(fdt, offset, "class-code", ccode));
1391     if (irq_pin) {
1392         _FDT(fdt_setprop_cell(fdt, offset, "interrupts", irq_pin));
1393     }
1394 
1395     if (subsystem_id) {
1396         _FDT(fdt_setprop_cell(fdt, offset, "subsystem-id", subsystem_id));
1397     }
1398 
1399     if (subsystem_vendor_id) {
1400         _FDT(fdt_setprop_cell(fdt, offset, "subsystem-vendor-id",
1401                               subsystem_vendor_id));
1402     }
1403 
1404     _FDT(fdt_setprop_cell(fdt, offset, "cache-line-size", cache_line_size));
1405 
1406 
1407     /* the following fdt cells are masked off the pci status register */
1408     _FDT(fdt_setprop_cell(fdt, offset, "devsel-speed",
1409                           PCI_STATUS_DEVSEL_MASK & pci_status));
1410 
1411     if (pci_status & PCI_STATUS_FAST_BACK) {
1412         _FDT(fdt_setprop(fdt, offset, "fast-back-to-back", NULL, 0));
1413     }
1414     if (pci_status & PCI_STATUS_66MHZ) {
1415         _FDT(fdt_setprop(fdt, offset, "66mhz-capable", NULL, 0));
1416     }
1417     if (pci_status & PCI_STATUS_UDF) {
1418         _FDT(fdt_setprop(fdt, offset, "udf-supported", NULL, 0));
1419     }
1420 
1421     loc_code = spapr_phb_get_loc_code(sphb, dev);
1422     _FDT(fdt_setprop_string(fdt, offset, "ibm,loc-code", loc_code));
1423     g_free(loc_code);
1424 
1425     if (drc) {
1426         _FDT(fdt_setprop_cell(fdt, offset, "ibm,my-drc-index",
1427                               spapr_drc_index(drc)));
1428     }
1429 
1430     if (msi_present(dev)) {
1431         uint32_t max_msi = msi_nr_vectors_allocated(dev);
1432         if (max_msi) {
1433             _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi", max_msi));
1434         }
1435     }
1436     if (msix_present(dev)) {
1437         uint32_t max_msix = dev->msix_entries_nr;
1438         if (max_msix) {
1439             _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi-x", max_msix));
1440         }
1441     }
1442 
1443     populate_resource_props(dev, &rp);
1444     _FDT(fdt_setprop(fdt, offset, "reg", (uint8_t *)rp.reg, rp.reg_len));
1445 
1446     if (sphb->pcie_ecs && pci_is_express(dev)) {
1447         _FDT(fdt_setprop_cell(fdt, offset, "ibm,pci-config-space-type", 0x1));
1448     }
1449 
1450     spapr_phb_nvgpu_populate_pcidev_dt(dev, fdt, offset, sphb);
1451 
1452     if (!pc->is_bridge) {
1453         /* Properties only for non-bridges */
1454         uint32_t min_grant = pci_default_read_config(dev, PCI_MIN_GNT, 1);
1455         uint32_t max_latency = pci_default_read_config(dev, PCI_MAX_LAT, 1);
1456         _FDT(fdt_setprop_cell(fdt, offset, "min-grant", min_grant));
1457         _FDT(fdt_setprop_cell(fdt, offset, "max-latency", max_latency));
1458         return offset;
1459     } else {
1460         PCIBus *sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(dev));
1461 
1462         return spapr_dt_pci_bus(sphb, sec_bus, fdt, offset);
1463     }
1464 }
1465 
1466 /* Callback to be called during DRC release. */
1467 void spapr_phb_remove_pci_device_cb(DeviceState *dev)
1468 {
1469     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
1470 
1471     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
1472     object_unparent(OBJECT(dev));
1473 }
1474 
1475 int spapr_pci_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
1476                           void *fdt, int *fdt_start_offset, Error **errp)
1477 {
1478     HotplugHandler *plug_handler = qdev_get_hotplug_handler(drc->dev);
1479     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(plug_handler);
1480     PCIDevice *pdev = PCI_DEVICE(drc->dev);
1481 
1482     *fdt_start_offset = spapr_dt_pci_device(sphb, pdev, fdt, 0);
1483     return 0;
1484 }
1485 
1486 static void spapr_pci_bridge_plug(SpaprPhbState *phb,
1487                                   PCIBridge *bridge)
1488 {
1489     PCIBus *bus = pci_bridge_get_sec_bus(bridge);
1490 
1491     add_drcs(phb, bus);
1492 }
1493 
1494 /* Returns non-zero if the value of "chassis_nr" is already in use */
1495 static int check_chassis_nr(Object *obj, void *opaque)
1496 {
1497     int new_chassis_nr =
1498         object_property_get_uint(opaque, "chassis_nr", &error_abort);
1499     int chassis_nr =
1500         object_property_get_uint(obj, "chassis_nr", NULL);
1501 
1502     if (!object_dynamic_cast(obj, TYPE_PCI_BRIDGE)) {
1503         return 0;
1504     }
1505 
1506     /* Skip unsupported bridge types */
1507     if (!chassis_nr) {
1508         return 0;
1509     }
1510 
1511     /* Skip self */
1512     if (obj == opaque) {
1513         return 0;
1514     }
1515 
1516     return chassis_nr == new_chassis_nr;
1517 }
1518 
1519 static bool bridge_has_valid_chassis_nr(Object *bridge, Error **errp)
1520 {
1521     int chassis_nr =
1522         object_property_get_uint(bridge, "chassis_nr", NULL);
1523 
1524     /*
1525      * slotid_cap_init() already ensures that "chassis_nr" isn't null for
1526      * standard PCI bridges, so this really tells if "chassis_nr" is present
1527      * or not.
1528      */
1529     if (!chassis_nr) {
1530         error_setg(errp, "PCI Bridge lacks a \"chassis_nr\" property");
1531         error_append_hint(errp, "Try -device pci-bridge instead.\n");
1532         return false;
1533     }
1534 
1535     /* We want unique values for "chassis_nr" */
1536     if (object_child_foreach_recursive(object_get_root(), check_chassis_nr,
1537                                        bridge)) {
1538         error_setg(errp, "Bridge chassis %d already in use", chassis_nr);
1539         return false;
1540     }
1541 
1542     return true;
1543 }
1544 
1545 static void spapr_pci_pre_plug(HotplugHandler *plug_handler,
1546                                DeviceState *plugged_dev, Error **errp)
1547 {
1548     SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1549     PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1550     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(plugged_dev);
1551     SpaprDrc *drc = drc_from_dev(phb, pdev);
1552     PCIBus *bus = PCI_BUS(qdev_get_parent_bus(DEVICE(pdev)));
1553     uint32_t slotnr = PCI_SLOT(pdev->devfn);
1554 
1555     if (!phb->dr_enabled) {
1556         /* if this is a hotplug operation initiated by the user
1557          * we need to let them know it's not enabled
1558          */
1559         if (plugged_dev->hotplugged) {
1560             error_setg(errp, QERR_BUS_NO_HOTPLUG,
1561                        object_get_typename(OBJECT(phb)));
1562             return;
1563         }
1564     }
1565 
1566     if (pc->is_bridge) {
1567         if (!bridge_has_valid_chassis_nr(OBJECT(plugged_dev), errp)) {
1568             return;
1569         }
1570     }
1571 
1572     /* Following the QEMU convention used for PCIe multifunction
1573      * hotplug, we do not allow functions to be hotplugged to a
1574      * slot that already has function 0 present
1575      */
1576     if (plugged_dev->hotplugged && bus->devices[PCI_DEVFN(slotnr, 0)] &&
1577         PCI_FUNC(pdev->devfn) != 0) {
1578         error_setg(errp, "PCI: slot %d function 0 already occupied by %s,"
1579                    " additional functions can no longer be exposed to guest.",
1580                    slotnr, bus->devices[PCI_DEVFN(slotnr, 0)]->name);
1581     }
1582 
1583     if (drc && drc->dev) {
1584         error_setg(errp, "PCI: slot %d already occupied by %s", slotnr,
1585                    pci_get_function_0(PCI_DEVICE(drc->dev))->name);
1586         return;
1587     }
1588 }
1589 
1590 static void spapr_pci_plug(HotplugHandler *plug_handler,
1591                            DeviceState *plugged_dev, Error **errp)
1592 {
1593     SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1594     PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1595     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(plugged_dev);
1596     SpaprDrc *drc = drc_from_dev(phb, pdev);
1597     uint32_t slotnr = PCI_SLOT(pdev->devfn);
1598 
1599     /*
1600      * If DR is disabled we don't need to do anything in the case of
1601      * hotplug or coldplug callbacks.
1602      */
1603     if (!phb->dr_enabled) {
1604         return;
1605     }
1606 
1607     g_assert(drc);
1608 
1609     if (pc->is_bridge) {
1610         spapr_pci_bridge_plug(phb, PCI_BRIDGE(plugged_dev));
1611     }
1612 
1613     /* spapr_pci_pre_plug() already checked the DRC is attachable */
1614     spapr_drc_attach(drc, DEVICE(pdev));
1615 
1616     /* If this is function 0, signal hotplug for all the device functions.
1617      * Otherwise defer sending the hotplug event.
1618      */
1619     if (!spapr_drc_hotplugged(plugged_dev)) {
1620         spapr_drc_reset(drc);
1621     } else if (PCI_FUNC(pdev->devfn) == 0) {
1622         int i;
1623         uint8_t chassis = chassis_from_bus(pci_get_bus(pdev));
1624 
1625         for (i = 0; i < 8; i++) {
1626             SpaprDrc *func_drc;
1627             SpaprDrcClass *func_drck;
1628             SpaprDREntitySense state;
1629 
1630             func_drc = drc_from_devfn(phb, chassis, PCI_DEVFN(slotnr, i));
1631             func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1632             state = func_drck->dr_entity_sense(func_drc);
1633 
1634             if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) {
1635                 spapr_hotplug_req_add_by_index(func_drc);
1636             }
1637         }
1638     }
1639 }
1640 
1641 static void spapr_pci_bridge_unplug(SpaprPhbState *phb,
1642                                     PCIBridge *bridge)
1643 {
1644     PCIBus *bus = pci_bridge_get_sec_bus(bridge);
1645 
1646     remove_drcs(phb, bus);
1647 }
1648 
1649 static void spapr_pci_unplug(HotplugHandler *plug_handler,
1650                              DeviceState *plugged_dev, Error **errp)
1651 {
1652     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(plugged_dev);
1653     SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1654 
1655     /* some version guests do not wait for completion of a device
1656      * cleanup (generally done asynchronously by the kernel) before
1657      * signaling to QEMU that the device is safe, but instead sleep
1658      * for some 'safe' period of time. unfortunately on a busy host
1659      * this sleep isn't guaranteed to be long enough, resulting in
1660      * bad things like IRQ lines being left asserted during final
1661      * device removal. to deal with this we call reset just prior
1662      * to finalizing the device, which will put the device back into
1663      * an 'idle' state, as the device cleanup code expects.
1664      */
1665     pci_device_reset(PCI_DEVICE(plugged_dev));
1666 
1667     if (pc->is_bridge) {
1668         spapr_pci_bridge_unplug(phb, PCI_BRIDGE(plugged_dev));
1669         return;
1670     }
1671 
1672     qdev_unrealize(plugged_dev);
1673 }
1674 
1675 static void spapr_pci_unplug_request(HotplugHandler *plug_handler,
1676                                      DeviceState *plugged_dev, Error **errp)
1677 {
1678     SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1679     PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1680     SpaprDrc *drc = drc_from_dev(phb, pdev);
1681 
1682     if (!phb->dr_enabled) {
1683         error_setg(errp, QERR_BUS_NO_HOTPLUG,
1684                    object_get_typename(OBJECT(phb)));
1685         return;
1686     }
1687 
1688     g_assert(drc);
1689     g_assert(drc->dev == plugged_dev);
1690 
1691     if (!spapr_drc_unplug_requested(drc)) {
1692         PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(plugged_dev);
1693         uint32_t slotnr = PCI_SLOT(pdev->devfn);
1694         SpaprDrc *func_drc;
1695         SpaprDrcClass *func_drck;
1696         SpaprDREntitySense state;
1697         int i;
1698         uint8_t chassis = chassis_from_bus(pci_get_bus(pdev));
1699 
1700         if (pc->is_bridge) {
1701             error_setg(errp, "PCI: Hot unplug of PCI bridges not supported");
1702             return;
1703         }
1704         if (object_property_get_uint(OBJECT(pdev), "nvlink2-tgt", NULL)) {
1705             error_setg(errp, "PCI: Cannot unplug NVLink2 devices");
1706             return;
1707         }
1708 
1709         /* ensure any other present functions are pending unplug */
1710         if (PCI_FUNC(pdev->devfn) == 0) {
1711             for (i = 1; i < 8; i++) {
1712                 func_drc = drc_from_devfn(phb, chassis, PCI_DEVFN(slotnr, i));
1713                 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1714                 state = func_drck->dr_entity_sense(func_drc);
1715                 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT
1716                     && !spapr_drc_unplug_requested(func_drc)) {
1717                     /*
1718                      * Attempting to remove function 0 of a multifunction
1719                      * device will will cascade into removing all child
1720                      * functions, even if their unplug weren't requested
1721                      * beforehand.
1722                      */
1723                     spapr_drc_detach(func_drc);
1724                 }
1725             }
1726         }
1727 
1728         spapr_drc_detach(drc);
1729 
1730         /* if this isn't func 0, defer unplug event. otherwise signal removal
1731          * for all present functions
1732          */
1733         if (PCI_FUNC(pdev->devfn) == 0) {
1734             for (i = 7; i >= 0; i--) {
1735                 func_drc = drc_from_devfn(phb, chassis, PCI_DEVFN(slotnr, i));
1736                 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1737                 state = func_drck->dr_entity_sense(func_drc);
1738                 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) {
1739                     spapr_hotplug_req_remove_by_index(func_drc);
1740                 }
1741             }
1742         }
1743     }
1744 }
1745 
1746 static void spapr_phb_finalizefn(Object *obj)
1747 {
1748     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(obj);
1749 
1750     g_free(sphb->dtbusname);
1751     sphb->dtbusname = NULL;
1752 }
1753 
1754 static void spapr_phb_unrealize(DeviceState *dev)
1755 {
1756     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
1757     SysBusDevice *s = SYS_BUS_DEVICE(dev);
1758     PCIHostState *phb = PCI_HOST_BRIDGE(s);
1759     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(phb);
1760     SpaprTceTable *tcet;
1761     int i;
1762     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
1763 
1764     spapr_phb_nvgpu_free(sphb);
1765 
1766     if (sphb->msi) {
1767         g_hash_table_unref(sphb->msi);
1768         sphb->msi = NULL;
1769     }
1770 
1771     /*
1772      * Remove IO/MMIO subregions and aliases, rest should get cleaned
1773      * via PHB's unrealize->object_finalize
1774      */
1775     for (i = windows_supported - 1; i >= 0; i--) {
1776         tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[i]);
1777         if (tcet) {
1778             memory_region_del_subregion(&sphb->iommu_root,
1779                                         spapr_tce_get_iommu(tcet));
1780         }
1781     }
1782 
1783     remove_drcs(sphb, phb->bus);
1784 
1785     for (i = PCI_NUM_PINS - 1; i >= 0; i--) {
1786         if (sphb->lsi_table[i].irq) {
1787             spapr_irq_free(spapr, sphb->lsi_table[i].irq, 1);
1788             sphb->lsi_table[i].irq = 0;
1789         }
1790     }
1791 
1792     QLIST_REMOVE(sphb, list);
1793 
1794     memory_region_del_subregion(&sphb->iommu_root, &sphb->msiwindow);
1795 
1796     /*
1797      * An attached PCI device may have memory listeners, eg. VFIO PCI. We have
1798      * unmapped all sections. Remove the listeners now, before destroying the
1799      * address space.
1800      */
1801     address_space_remove_listeners(&sphb->iommu_as);
1802     address_space_destroy(&sphb->iommu_as);
1803 
1804     qbus_set_hotplug_handler(BUS(phb->bus), NULL);
1805     pci_unregister_root_bus(phb->bus);
1806 
1807     memory_region_del_subregion(get_system_memory(), &sphb->iowindow);
1808     if (sphb->mem64_win_pciaddr != (hwaddr)-1) {
1809         memory_region_del_subregion(get_system_memory(), &sphb->mem64window);
1810     }
1811     memory_region_del_subregion(get_system_memory(), &sphb->mem32window);
1812 }
1813 
1814 static void spapr_phb_destroy_msi(gpointer opaque)
1815 {
1816     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
1817     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
1818     SpaprPciMsi *msi = opaque;
1819 
1820     if (!smc->legacy_irq_allocation) {
1821         spapr_irq_msi_free(spapr, msi->first_irq, msi->num);
1822     }
1823     spapr_irq_free(spapr, msi->first_irq, msi->num);
1824     g_free(msi);
1825 }
1826 
1827 static void spapr_phb_realize(DeviceState *dev, Error **errp)
1828 {
1829     ERRP_GUARD();
1830     /* We don't use SPAPR_MACHINE() in order to exit gracefully if the user
1831      * tries to add a sPAPR PHB to a non-pseries machine.
1832      */
1833     SpaprMachineState *spapr =
1834         (SpaprMachineState *) object_dynamic_cast(qdev_get_machine(),
1835                                                   TYPE_SPAPR_MACHINE);
1836     SpaprMachineClass *smc = spapr ? SPAPR_MACHINE_GET_CLASS(spapr) : NULL;
1837     SysBusDevice *s = SYS_BUS_DEVICE(dev);
1838     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(s);
1839     PCIHostState *phb = PCI_HOST_BRIDGE(s);
1840     MachineState *ms = MACHINE(spapr);
1841     char *namebuf;
1842     int i;
1843     PCIBus *bus;
1844     uint64_t msi_window_size = 4096;
1845     SpaprTceTable *tcet;
1846     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
1847 
1848     if (!spapr) {
1849         error_setg(errp, TYPE_SPAPR_PCI_HOST_BRIDGE " needs a pseries machine");
1850         return;
1851     }
1852 
1853     assert(sphb->index != (uint32_t)-1); /* checked in spapr_phb_pre_plug() */
1854 
1855     if (sphb->mem64_win_size != 0) {
1856         if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) {
1857             error_setg(errp, "32-bit memory window of size 0x%"HWADDR_PRIx
1858                        " (max 2 GiB)", sphb->mem_win_size);
1859             return;
1860         }
1861 
1862         /* 64-bit window defaults to identity mapping */
1863         sphb->mem64_win_pciaddr = sphb->mem64_win_addr;
1864     } else if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) {
1865         /*
1866          * For compatibility with old configuration, if no 64-bit MMIO
1867          * window is specified, but the ordinary (32-bit) memory
1868          * window is specified as > 2GiB, we treat it as a 2GiB 32-bit
1869          * window, with a 64-bit MMIO window following on immediately
1870          * afterwards
1871          */
1872         sphb->mem64_win_size = sphb->mem_win_size - SPAPR_PCI_MEM32_WIN_SIZE;
1873         sphb->mem64_win_addr = sphb->mem_win_addr + SPAPR_PCI_MEM32_WIN_SIZE;
1874         sphb->mem64_win_pciaddr =
1875             SPAPR_PCI_MEM_WIN_BUS_OFFSET + SPAPR_PCI_MEM32_WIN_SIZE;
1876         sphb->mem_win_size = SPAPR_PCI_MEM32_WIN_SIZE;
1877     }
1878 
1879     if (spapr_pci_find_phb(spapr, sphb->buid)) {
1880         SpaprPhbState *s;
1881 
1882         error_setg(errp, "PCI host bridges must have unique indexes");
1883         error_append_hint(errp, "The following indexes are already in use:");
1884         QLIST_FOREACH(s, &spapr->phbs, list) {
1885             error_append_hint(errp, " %d", s->index);
1886         }
1887         error_append_hint(errp, "\nTry another value for the index property\n");
1888         return;
1889     }
1890 
1891     if (sphb->numa_node != -1 &&
1892         (sphb->numa_node >= MAX_NODES ||
1893          !ms->numa_state->nodes[sphb->numa_node].present)) {
1894         error_setg(errp, "Invalid NUMA node ID for PCI host bridge");
1895         return;
1896     }
1897 
1898     sphb->dtbusname = g_strdup_printf("pci@%" PRIx64, sphb->buid);
1899 
1900     /* Initialize memory regions */
1901     namebuf = g_strdup_printf("%s.mmio", sphb->dtbusname);
1902     memory_region_init(&sphb->memspace, OBJECT(sphb), namebuf, UINT64_MAX);
1903     g_free(namebuf);
1904 
1905     namebuf = g_strdup_printf("%s.mmio32-alias", sphb->dtbusname);
1906     memory_region_init_alias(&sphb->mem32window, OBJECT(sphb),
1907                              namebuf, &sphb->memspace,
1908                              SPAPR_PCI_MEM_WIN_BUS_OFFSET, sphb->mem_win_size);
1909     g_free(namebuf);
1910     memory_region_add_subregion(get_system_memory(), sphb->mem_win_addr,
1911                                 &sphb->mem32window);
1912 
1913     if (sphb->mem64_win_size != 0) {
1914         namebuf = g_strdup_printf("%s.mmio64-alias", sphb->dtbusname);
1915         memory_region_init_alias(&sphb->mem64window, OBJECT(sphb),
1916                                  namebuf, &sphb->memspace,
1917                                  sphb->mem64_win_pciaddr, sphb->mem64_win_size);
1918         g_free(namebuf);
1919 
1920         memory_region_add_subregion(get_system_memory(),
1921                                     sphb->mem64_win_addr,
1922                                     &sphb->mem64window);
1923     }
1924 
1925     /* Initialize IO regions */
1926     namebuf = g_strdup_printf("%s.io", sphb->dtbusname);
1927     memory_region_init(&sphb->iospace, OBJECT(sphb),
1928                        namebuf, SPAPR_PCI_IO_WIN_SIZE);
1929     g_free(namebuf);
1930 
1931     namebuf = g_strdup_printf("%s.io-alias", sphb->dtbusname);
1932     memory_region_init_alias(&sphb->iowindow, OBJECT(sphb), namebuf,
1933                              &sphb->iospace, 0, SPAPR_PCI_IO_WIN_SIZE);
1934     g_free(namebuf);
1935     memory_region_add_subregion(get_system_memory(), sphb->io_win_addr,
1936                                 &sphb->iowindow);
1937 
1938     bus = pci_register_root_bus(dev, NULL,
1939                                 pci_spapr_set_irq, pci_swizzle_map_irq_fn, sphb,
1940                                 &sphb->memspace, &sphb->iospace,
1941                                 PCI_DEVFN(0, 0), PCI_NUM_PINS,
1942                                 TYPE_PCI_BUS);
1943 
1944     /*
1945      * Despite resembling a vanilla PCI bus in most ways, the PAPR
1946      * para-virtualized PCI bus *does* permit PCI-E extended config
1947      * space access
1948      */
1949     if (sphb->pcie_ecs) {
1950         bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
1951     }
1952     phb->bus = bus;
1953     qbus_set_hotplug_handler(BUS(phb->bus), OBJECT(sphb));
1954 
1955     /*
1956      * Initialize PHB address space.
1957      * By default there will be at least one subregion for default
1958      * 32bit DMA window.
1959      * Later the guest might want to create another DMA window
1960      * which will become another memory subregion.
1961      */
1962     namebuf = g_strdup_printf("%s.iommu-root", sphb->dtbusname);
1963     memory_region_init(&sphb->iommu_root, OBJECT(sphb),
1964                        namebuf, UINT64_MAX);
1965     g_free(namebuf);
1966     address_space_init(&sphb->iommu_as, &sphb->iommu_root,
1967                        sphb->dtbusname);
1968 
1969     /*
1970      * As MSI/MSIX interrupts trigger by writing at MSI/MSIX vectors,
1971      * we need to allocate some memory to catch those writes coming
1972      * from msi_notify()/msix_notify().
1973      * As MSIMessage:addr is going to be the same and MSIMessage:data
1974      * is going to be a VIRQ number, 4 bytes of the MSI MR will only
1975      * be used.
1976      *
1977      * For KVM we want to ensure that this memory is a full page so that
1978      * our memory slot is of page size granularity.
1979      */
1980     if (kvm_enabled()) {
1981         msi_window_size = qemu_real_host_page_size;
1982     }
1983 
1984     memory_region_init_io(&sphb->msiwindow, OBJECT(sphb), &spapr_msi_ops, spapr,
1985                           "msi", msi_window_size);
1986     memory_region_add_subregion(&sphb->iommu_root, SPAPR_PCI_MSI_WINDOW,
1987                                 &sphb->msiwindow);
1988 
1989     pci_setup_iommu(bus, spapr_pci_dma_iommu, sphb);
1990 
1991     pci_bus_set_route_irq_fn(bus, spapr_route_intx_pin_to_irq);
1992 
1993     QLIST_INSERT_HEAD(&spapr->phbs, sphb, list);
1994 
1995     /* Initialize the LSI table */
1996     for (i = 0; i < PCI_NUM_PINS; i++) {
1997         int irq = SPAPR_IRQ_PCI_LSI + sphb->index * PCI_NUM_PINS + i;
1998 
1999         if (smc->legacy_irq_allocation) {
2000             irq = spapr_irq_findone(spapr, errp);
2001             if (irq < 0) {
2002                 error_prepend(errp, "can't allocate LSIs: ");
2003                 /*
2004                  * Older machines will never support PHB hotplug, ie, this is an
2005                  * init only path and QEMU will terminate. No need to rollback.
2006                  */
2007                 return;
2008             }
2009         }
2010 
2011         if (spapr_irq_claim(spapr, irq, true, errp) < 0) {
2012             error_prepend(errp, "can't allocate LSIs: ");
2013             goto unrealize;
2014         }
2015 
2016         sphb->lsi_table[i].irq = irq;
2017     }
2018 
2019     /* allocate connectors for child PCI devices */
2020     add_drcs(sphb, phb->bus);
2021 
2022     /* DMA setup */
2023     for (i = 0; i < windows_supported; ++i) {
2024         tcet = spapr_tce_new_table(DEVICE(sphb), sphb->dma_liobn[i]);
2025         if (!tcet) {
2026             error_setg(errp, "Creating window#%d failed for %s",
2027                        i, sphb->dtbusname);
2028             goto unrealize;
2029         }
2030         memory_region_add_subregion(&sphb->iommu_root, 0,
2031                                     spapr_tce_get_iommu(tcet));
2032     }
2033 
2034     sphb->msi = g_hash_table_new_full(g_int_hash, g_int_equal, g_free,
2035                                       spapr_phb_destroy_msi);
2036     return;
2037 
2038 unrealize:
2039     spapr_phb_unrealize(dev);
2040 }
2041 
2042 static int spapr_phb_children_reset(Object *child, void *opaque)
2043 {
2044     DeviceState *dev = (DeviceState *) object_dynamic_cast(child, TYPE_DEVICE);
2045 
2046     if (dev) {
2047         device_legacy_reset(dev);
2048     }
2049 
2050     return 0;
2051 }
2052 
2053 void spapr_phb_dma_reset(SpaprPhbState *sphb)
2054 {
2055     int i;
2056     SpaprTceTable *tcet;
2057 
2058     for (i = 0; i < SPAPR_PCI_DMA_MAX_WINDOWS; ++i) {
2059         tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[i]);
2060 
2061         if (tcet && tcet->nb_table) {
2062             spapr_tce_table_disable(tcet);
2063         }
2064     }
2065 
2066     /* Register default 32bit DMA window */
2067     tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[0]);
2068     spapr_tce_table_enable(tcet, SPAPR_TCE_PAGE_SHIFT, sphb->dma_win_addr,
2069                            sphb->dma_win_size >> SPAPR_TCE_PAGE_SHIFT);
2070 }
2071 
2072 static void spapr_phb_reset(DeviceState *qdev)
2073 {
2074     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(qdev);
2075     Error *err = NULL;
2076 
2077     spapr_phb_dma_reset(sphb);
2078     spapr_phb_nvgpu_free(sphb);
2079     spapr_phb_nvgpu_setup(sphb, &err);
2080     if (err) {
2081         error_report_err(err);
2082     }
2083 
2084     /* Reset the IOMMU state */
2085     object_child_foreach(OBJECT(qdev), spapr_phb_children_reset, NULL);
2086 
2087     if (spapr_phb_eeh_available(SPAPR_PCI_HOST_BRIDGE(qdev))) {
2088         spapr_phb_vfio_reset(qdev);
2089     }
2090 
2091     g_hash_table_remove_all(sphb->msi);
2092 }
2093 
2094 static Property spapr_phb_properties[] = {
2095     DEFINE_PROP_UINT32("index", SpaprPhbState, index, -1),
2096     DEFINE_PROP_UINT64("mem_win_size", SpaprPhbState, mem_win_size,
2097                        SPAPR_PCI_MEM32_WIN_SIZE),
2098     DEFINE_PROP_UINT64("mem64_win_size", SpaprPhbState, mem64_win_size,
2099                        SPAPR_PCI_MEM64_WIN_SIZE),
2100     DEFINE_PROP_UINT64("io_win_size", SpaprPhbState, io_win_size,
2101                        SPAPR_PCI_IO_WIN_SIZE),
2102     DEFINE_PROP_BOOL("dynamic-reconfiguration", SpaprPhbState, dr_enabled,
2103                      true),
2104     /* Default DMA window is 0..1GB */
2105     DEFINE_PROP_UINT64("dma_win_addr", SpaprPhbState, dma_win_addr, 0),
2106     DEFINE_PROP_UINT64("dma_win_size", SpaprPhbState, dma_win_size, 0x40000000),
2107     DEFINE_PROP_UINT64("dma64_win_addr", SpaprPhbState, dma64_win_addr,
2108                        0x800000000000000ULL),
2109     DEFINE_PROP_BOOL("ddw", SpaprPhbState, ddw_enabled, true),
2110     DEFINE_PROP_UINT64("pgsz", SpaprPhbState, page_size_mask,
2111                        (1ULL << 12) | (1ULL << 16)
2112                        | (1ULL << 21) | (1ULL << 24)),
2113     DEFINE_PROP_UINT32("numa_node", SpaprPhbState, numa_node, -1),
2114     DEFINE_PROP_BOOL("pre-2.8-migration", SpaprPhbState,
2115                      pre_2_8_migration, false),
2116     DEFINE_PROP_BOOL("pcie-extended-configuration-space", SpaprPhbState,
2117                      pcie_ecs, true),
2118     DEFINE_PROP_UINT64("gpa", SpaprPhbState, nv2_gpa_win_addr, 0),
2119     DEFINE_PROP_UINT64("atsd", SpaprPhbState, nv2_atsd_win_addr, 0),
2120     DEFINE_PROP_BOOL("pre-5.1-associativity", SpaprPhbState,
2121                      pre_5_1_assoc, false),
2122     DEFINE_PROP_END_OF_LIST(),
2123 };
2124 
2125 static const VMStateDescription vmstate_spapr_pci_lsi = {
2126     .name = "spapr_pci/lsi",
2127     .version_id = 1,
2128     .minimum_version_id = 1,
2129     .fields = (VMStateField[]) {
2130         VMSTATE_UINT32_EQUAL(irq, SpaprPciLsi, NULL),
2131 
2132         VMSTATE_END_OF_LIST()
2133     },
2134 };
2135 
2136 static const VMStateDescription vmstate_spapr_pci_msi = {
2137     .name = "spapr_pci/msi",
2138     .version_id = 1,
2139     .minimum_version_id = 1,
2140     .fields = (VMStateField []) {
2141         VMSTATE_UINT32(key, SpaprPciMsiMig),
2142         VMSTATE_UINT32(value.first_irq, SpaprPciMsiMig),
2143         VMSTATE_UINT32(value.num, SpaprPciMsiMig),
2144         VMSTATE_END_OF_LIST()
2145     },
2146 };
2147 
2148 static int spapr_pci_pre_save(void *opaque)
2149 {
2150     SpaprPhbState *sphb = opaque;
2151     GHashTableIter iter;
2152     gpointer key, value;
2153     int i;
2154 
2155     if (sphb->pre_2_8_migration) {
2156         sphb->mig_liobn = sphb->dma_liobn[0];
2157         sphb->mig_mem_win_addr = sphb->mem_win_addr;
2158         sphb->mig_mem_win_size = sphb->mem_win_size;
2159         sphb->mig_io_win_addr = sphb->io_win_addr;
2160         sphb->mig_io_win_size = sphb->io_win_size;
2161 
2162         if ((sphb->mem64_win_size != 0)
2163             && (sphb->mem64_win_addr
2164                 == (sphb->mem_win_addr + sphb->mem_win_size))) {
2165             sphb->mig_mem_win_size += sphb->mem64_win_size;
2166         }
2167     }
2168 
2169     g_free(sphb->msi_devs);
2170     sphb->msi_devs = NULL;
2171     sphb->msi_devs_num = g_hash_table_size(sphb->msi);
2172     if (!sphb->msi_devs_num) {
2173         return 0;
2174     }
2175     sphb->msi_devs = g_new(SpaprPciMsiMig, sphb->msi_devs_num);
2176 
2177     g_hash_table_iter_init(&iter, sphb->msi);
2178     for (i = 0; g_hash_table_iter_next(&iter, &key, &value); ++i) {
2179         sphb->msi_devs[i].key = *(uint32_t *) key;
2180         sphb->msi_devs[i].value = *(SpaprPciMsi *) value;
2181     }
2182 
2183     return 0;
2184 }
2185 
2186 static int spapr_pci_post_save(void *opaque)
2187 {
2188     SpaprPhbState *sphb = opaque;
2189 
2190     g_free(sphb->msi_devs);
2191     sphb->msi_devs = NULL;
2192     sphb->msi_devs_num = 0;
2193     return 0;
2194 }
2195 
2196 static int spapr_pci_post_load(void *opaque, int version_id)
2197 {
2198     SpaprPhbState *sphb = opaque;
2199     gpointer key, value;
2200     int i;
2201 
2202     for (i = 0; i < sphb->msi_devs_num; ++i) {
2203         key = g_memdup(&sphb->msi_devs[i].key,
2204                        sizeof(sphb->msi_devs[i].key));
2205         value = g_memdup(&sphb->msi_devs[i].value,
2206                          sizeof(sphb->msi_devs[i].value));
2207         g_hash_table_insert(sphb->msi, key, value);
2208     }
2209     g_free(sphb->msi_devs);
2210     sphb->msi_devs = NULL;
2211     sphb->msi_devs_num = 0;
2212 
2213     return 0;
2214 }
2215 
2216 static bool pre_2_8_migration(void *opaque, int version_id)
2217 {
2218     SpaprPhbState *sphb = opaque;
2219 
2220     return sphb->pre_2_8_migration;
2221 }
2222 
2223 static const VMStateDescription vmstate_spapr_pci = {
2224     .name = "spapr_pci",
2225     .version_id = 2,
2226     .minimum_version_id = 2,
2227     .pre_save = spapr_pci_pre_save,
2228     .post_save = spapr_pci_post_save,
2229     .post_load = spapr_pci_post_load,
2230     .fields = (VMStateField[]) {
2231         VMSTATE_UINT64_EQUAL(buid, SpaprPhbState, NULL),
2232         VMSTATE_UINT32_TEST(mig_liobn, SpaprPhbState, pre_2_8_migration),
2233         VMSTATE_UINT64_TEST(mig_mem_win_addr, SpaprPhbState, pre_2_8_migration),
2234         VMSTATE_UINT64_TEST(mig_mem_win_size, SpaprPhbState, pre_2_8_migration),
2235         VMSTATE_UINT64_TEST(mig_io_win_addr, SpaprPhbState, pre_2_8_migration),
2236         VMSTATE_UINT64_TEST(mig_io_win_size, SpaprPhbState, pre_2_8_migration),
2237         VMSTATE_STRUCT_ARRAY(lsi_table, SpaprPhbState, PCI_NUM_PINS, 0,
2238                              vmstate_spapr_pci_lsi, SpaprPciLsi),
2239         VMSTATE_INT32(msi_devs_num, SpaprPhbState),
2240         VMSTATE_STRUCT_VARRAY_ALLOC(msi_devs, SpaprPhbState, msi_devs_num, 0,
2241                                     vmstate_spapr_pci_msi, SpaprPciMsiMig),
2242         VMSTATE_END_OF_LIST()
2243     },
2244 };
2245 
2246 static const char *spapr_phb_root_bus_path(PCIHostState *host_bridge,
2247                                            PCIBus *rootbus)
2248 {
2249     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(host_bridge);
2250 
2251     return sphb->dtbusname;
2252 }
2253 
2254 static void spapr_phb_class_init(ObjectClass *klass, void *data)
2255 {
2256     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(klass);
2257     DeviceClass *dc = DEVICE_CLASS(klass);
2258     HotplugHandlerClass *hp = HOTPLUG_HANDLER_CLASS(klass);
2259 
2260     hc->root_bus_path = spapr_phb_root_bus_path;
2261     dc->realize = spapr_phb_realize;
2262     dc->unrealize = spapr_phb_unrealize;
2263     device_class_set_props(dc, spapr_phb_properties);
2264     dc->reset = spapr_phb_reset;
2265     dc->vmsd = &vmstate_spapr_pci;
2266     /* Supported by TYPE_SPAPR_MACHINE */
2267     dc->user_creatable = true;
2268     set_bit(DEVICE_CATEGORY_BRIDGE, dc->categories);
2269     hp->pre_plug = spapr_pci_pre_plug;
2270     hp->plug = spapr_pci_plug;
2271     hp->unplug = spapr_pci_unplug;
2272     hp->unplug_request = spapr_pci_unplug_request;
2273 }
2274 
2275 static const TypeInfo spapr_phb_info = {
2276     .name          = TYPE_SPAPR_PCI_HOST_BRIDGE,
2277     .parent        = TYPE_PCI_HOST_BRIDGE,
2278     .instance_size = sizeof(SpaprPhbState),
2279     .instance_finalize = spapr_phb_finalizefn,
2280     .class_init    = spapr_phb_class_init,
2281     .interfaces    = (InterfaceInfo[]) {
2282         { TYPE_HOTPLUG_HANDLER },
2283         { }
2284     }
2285 };
2286 
2287 static void spapr_phb_pci_enumerate_bridge(PCIBus *bus, PCIDevice *pdev,
2288                                            void *opaque)
2289 {
2290     unsigned int *bus_no = opaque;
2291     PCIBus *sec_bus = NULL;
2292 
2293     if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) !=
2294          PCI_HEADER_TYPE_BRIDGE)) {
2295         return;
2296     }
2297 
2298     (*bus_no)++;
2299     pci_default_write_config(pdev, PCI_PRIMARY_BUS, pci_dev_bus_num(pdev), 1);
2300     pci_default_write_config(pdev, PCI_SECONDARY_BUS, *bus_no, 1);
2301     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
2302 
2303     sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev));
2304     if (!sec_bus) {
2305         return;
2306     }
2307 
2308     pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
2309                         spapr_phb_pci_enumerate_bridge, bus_no);
2310     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
2311 }
2312 
2313 static void spapr_phb_pci_enumerate(SpaprPhbState *phb)
2314 {
2315     PCIBus *bus = PCI_HOST_BRIDGE(phb)->bus;
2316     unsigned int bus_no = 0;
2317 
2318     pci_for_each_device(bus, pci_bus_num(bus),
2319                         spapr_phb_pci_enumerate_bridge,
2320                         &bus_no);
2321 
2322 }
2323 
2324 int spapr_dt_phb(SpaprMachineState *spapr, SpaprPhbState *phb,
2325                  uint32_t intc_phandle, void *fdt, int *node_offset)
2326 {
2327     int bus_off, i, j, ret;
2328     uint32_t bus_range[] = { cpu_to_be32(0), cpu_to_be32(0xff) };
2329     struct {
2330         uint32_t hi;
2331         uint64_t child;
2332         uint64_t parent;
2333         uint64_t size;
2334     } QEMU_PACKED ranges[] = {
2335         {
2336             cpu_to_be32(b_ss(1)), cpu_to_be64(0),
2337             cpu_to_be64(phb->io_win_addr),
2338             cpu_to_be64(memory_region_size(&phb->iospace)),
2339         },
2340         {
2341             cpu_to_be32(b_ss(2)), cpu_to_be64(SPAPR_PCI_MEM_WIN_BUS_OFFSET),
2342             cpu_to_be64(phb->mem_win_addr),
2343             cpu_to_be64(phb->mem_win_size),
2344         },
2345         {
2346             cpu_to_be32(b_ss(3)), cpu_to_be64(phb->mem64_win_pciaddr),
2347             cpu_to_be64(phb->mem64_win_addr),
2348             cpu_to_be64(phb->mem64_win_size),
2349         },
2350     };
2351     const unsigned sizeof_ranges =
2352         (phb->mem64_win_size ? 3 : 2) * sizeof(ranges[0]);
2353     uint64_t bus_reg[] = { cpu_to_be64(phb->buid), 0 };
2354     uint32_t interrupt_map_mask[] = {
2355         cpu_to_be32(b_ddddd(-1)|b_fff(0)), 0x0, 0x0, cpu_to_be32(-1)};
2356     uint32_t interrupt_map[PCI_SLOT_MAX * PCI_NUM_PINS][7];
2357     uint32_t ddw_applicable[] = {
2358         cpu_to_be32(RTAS_IBM_QUERY_PE_DMA_WINDOW),
2359         cpu_to_be32(RTAS_IBM_CREATE_PE_DMA_WINDOW),
2360         cpu_to_be32(RTAS_IBM_REMOVE_PE_DMA_WINDOW)
2361     };
2362     uint32_t ddw_extensions[] = {
2363         cpu_to_be32(1),
2364         cpu_to_be32(RTAS_IBM_RESET_PE_DMA_WINDOW)
2365     };
2366     SpaprTceTable *tcet;
2367     SpaprDrc *drc;
2368     Error *err = NULL;
2369 
2370     /* Start populating the FDT */
2371     _FDT(bus_off = fdt_add_subnode(fdt, 0, phb->dtbusname));
2372     if (node_offset) {
2373         *node_offset = bus_off;
2374     }
2375 
2376     /* Write PHB properties */
2377     _FDT(fdt_setprop_string(fdt, bus_off, "device_type", "pci"));
2378     _FDT(fdt_setprop_string(fdt, bus_off, "compatible", "IBM,Logical_PHB"));
2379     _FDT(fdt_setprop_cell(fdt, bus_off, "#interrupt-cells", 0x1));
2380     _FDT(fdt_setprop(fdt, bus_off, "used-by-rtas", NULL, 0));
2381     _FDT(fdt_setprop(fdt, bus_off, "bus-range", &bus_range, sizeof(bus_range)));
2382     _FDT(fdt_setprop(fdt, bus_off, "ranges", &ranges, sizeof_ranges));
2383     _FDT(fdt_setprop(fdt, bus_off, "reg", &bus_reg, sizeof(bus_reg)));
2384     _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pci-config-space-type", 0x1));
2385     _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pe-total-#msi",
2386                           spapr_irq_nr_msis(spapr)));
2387 
2388     /* Dynamic DMA window */
2389     if (phb->ddw_enabled) {
2390         _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-applicable", &ddw_applicable,
2391                          sizeof(ddw_applicable)));
2392         _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-extensions",
2393                          &ddw_extensions, sizeof(ddw_extensions)));
2394     }
2395 
2396     /* Advertise NUMA via ibm,associativity */
2397     if (phb->numa_node != -1) {
2398         spapr_numa_write_associativity_dt(spapr, fdt, bus_off, phb->numa_node);
2399     }
2400 
2401     /* Build the interrupt-map, this must matches what is done
2402      * in pci_swizzle_map_irq_fn
2403      */
2404     _FDT(fdt_setprop(fdt, bus_off, "interrupt-map-mask",
2405                      &interrupt_map_mask, sizeof(interrupt_map_mask)));
2406     for (i = 0; i < PCI_SLOT_MAX; i++) {
2407         for (j = 0; j < PCI_NUM_PINS; j++) {
2408             uint32_t *irqmap = interrupt_map[i*PCI_NUM_PINS + j];
2409             int lsi_num = pci_swizzle(i, j);
2410 
2411             irqmap[0] = cpu_to_be32(b_ddddd(i)|b_fff(0));
2412             irqmap[1] = 0;
2413             irqmap[2] = 0;
2414             irqmap[3] = cpu_to_be32(j+1);
2415             irqmap[4] = cpu_to_be32(intc_phandle);
2416             spapr_dt_irq(&irqmap[5], phb->lsi_table[lsi_num].irq, true);
2417         }
2418     }
2419     /* Write interrupt map */
2420     _FDT(fdt_setprop(fdt, bus_off, "interrupt-map", &interrupt_map,
2421                      sizeof(interrupt_map)));
2422 
2423     tcet = spapr_tce_find_by_liobn(phb->dma_liobn[0]);
2424     if (!tcet) {
2425         return -1;
2426     }
2427     spapr_dma_dt(fdt, bus_off, "ibm,dma-window",
2428                  tcet->liobn, tcet->bus_offset,
2429                  tcet->nb_table << tcet->page_shift);
2430 
2431     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, phb->index);
2432     if (drc) {
2433         uint32_t drc_index = cpu_to_be32(spapr_drc_index(drc));
2434 
2435         _FDT(fdt_setprop(fdt, bus_off, "ibm,my-drc-index", &drc_index,
2436                          sizeof(drc_index)));
2437     }
2438 
2439     /* Walk the bridges and program the bus numbers*/
2440     spapr_phb_pci_enumerate(phb);
2441     _FDT(fdt_setprop_cell(fdt, bus_off, "qemu,phb-enumerated", 0x1));
2442 
2443     /* Walk the bridge and subordinate buses */
2444     ret = spapr_dt_pci_bus(phb, PCI_HOST_BRIDGE(phb)->bus, fdt, bus_off);
2445     if (ret < 0) {
2446         return ret;
2447     }
2448 
2449     spapr_phb_nvgpu_populate_dt(phb, fdt, bus_off, &err);
2450     if (err) {
2451         error_report_err(err);
2452     }
2453     spapr_phb_nvgpu_ram_populate_dt(phb, fdt);
2454 
2455     return 0;
2456 }
2457 
2458 void spapr_pci_rtas_init(void)
2459 {
2460     spapr_rtas_register(RTAS_READ_PCI_CONFIG, "read-pci-config",
2461                         rtas_read_pci_config);
2462     spapr_rtas_register(RTAS_WRITE_PCI_CONFIG, "write-pci-config",
2463                         rtas_write_pci_config);
2464     spapr_rtas_register(RTAS_IBM_READ_PCI_CONFIG, "ibm,read-pci-config",
2465                         rtas_ibm_read_pci_config);
2466     spapr_rtas_register(RTAS_IBM_WRITE_PCI_CONFIG, "ibm,write-pci-config",
2467                         rtas_ibm_write_pci_config);
2468     if (msi_nonbroken) {
2469         spapr_rtas_register(RTAS_IBM_QUERY_INTERRUPT_SOURCE_NUMBER,
2470                             "ibm,query-interrupt-source-number",
2471                             rtas_ibm_query_interrupt_source_number);
2472         spapr_rtas_register(RTAS_IBM_CHANGE_MSI, "ibm,change-msi",
2473                             rtas_ibm_change_msi);
2474     }
2475 
2476     spapr_rtas_register(RTAS_IBM_SET_EEH_OPTION,
2477                         "ibm,set-eeh-option",
2478                         rtas_ibm_set_eeh_option);
2479     spapr_rtas_register(RTAS_IBM_GET_CONFIG_ADDR_INFO2,
2480                         "ibm,get-config-addr-info2",
2481                         rtas_ibm_get_config_addr_info2);
2482     spapr_rtas_register(RTAS_IBM_READ_SLOT_RESET_STATE2,
2483                         "ibm,read-slot-reset-state2",
2484                         rtas_ibm_read_slot_reset_state2);
2485     spapr_rtas_register(RTAS_IBM_SET_SLOT_RESET,
2486                         "ibm,set-slot-reset",
2487                         rtas_ibm_set_slot_reset);
2488     spapr_rtas_register(RTAS_IBM_CONFIGURE_PE,
2489                         "ibm,configure-pe",
2490                         rtas_ibm_configure_pe);
2491     spapr_rtas_register(RTAS_IBM_SLOT_ERROR_DETAIL,
2492                         "ibm,slot-error-detail",
2493                         rtas_ibm_slot_error_detail);
2494 }
2495 
2496 static void spapr_pci_register_types(void)
2497 {
2498     type_register_static(&spapr_phb_info);
2499 }
2500 
2501 type_init(spapr_pci_register_types)
2502 
2503 static int spapr_switch_one_vga(DeviceState *dev, void *opaque)
2504 {
2505     bool be = *(bool *)opaque;
2506 
2507     if (object_dynamic_cast(OBJECT(dev), "VGA")
2508         || object_dynamic_cast(OBJECT(dev), "secondary-vga")
2509         || object_dynamic_cast(OBJECT(dev), "bochs-display")
2510         || object_dynamic_cast(OBJECT(dev), "virtio-vga")) {
2511         object_property_set_bool(OBJECT(dev), "big-endian-framebuffer", be,
2512                                  &error_abort);
2513     }
2514     return 0;
2515 }
2516 
2517 void spapr_pci_switch_vga(SpaprMachineState *spapr, bool big_endian)
2518 {
2519     SpaprPhbState *sphb;
2520 
2521     /*
2522      * For backward compatibility with existing guests, we switch
2523      * the endianness of the VGA controller when changing the guest
2524      * interrupt mode
2525      */
2526     QLIST_FOREACH(sphb, &spapr->phbs, list) {
2527         BusState *bus = &PCI_HOST_BRIDGE(sphb)->bus->qbus;
2528         qbus_walk_children(bus, spapr_switch_one_vga, NULL, NULL, NULL,
2529                            &big_endian);
2530     }
2531 }
2532