xref: /openbmc/qemu/hw/ppc/spapr_pci.c (revision f6476697)
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 "hw/irq.h"
29 #include "hw/sysbus.h"
30 #include "migration/vmstate.h"
31 #include "hw/pci/pci.h"
32 #include "hw/pci/msi.h"
33 #include "hw/pci/msix.h"
34 #include "hw/pci/pci_host.h"
35 #include "hw/ppc/spapr.h"
36 #include "hw/pci-host/spapr.h"
37 #include "exec/ram_addr.h"
38 #include <libfdt.h>
39 #include "trace.h"
40 #include "qemu/error-report.h"
41 #include "qemu/module.h"
42 #include "qapi/qmp/qerror.h"
43 #include "hw/ppc/fdt.h"
44 #include "hw/pci/pci_bridge.h"
45 #include "hw/pci/pci_bus.h"
46 #include "hw/pci/pci_ids.h"
47 #include "hw/ppc/spapr_drc.h"
48 #include "hw/qdev-properties.h"
49 #include "sysemu/device_tree.h"
50 #include "sysemu/kvm.h"
51 #include "sysemu/hostmem.h"
52 #include "sysemu/numa.h"
53 #include "hw/ppc/spapr_numa.h"
54 #include "qemu/log.h"
55 
56 /* Copied from the kernel arch/powerpc/platforms/pseries/msi.c */
57 #define RTAS_QUERY_FN           0
58 #define RTAS_CHANGE_FN          1
59 #define RTAS_RESET_FN           2
60 #define RTAS_CHANGE_MSI_FN      3
61 #define RTAS_CHANGE_MSIX_FN     4
62 
63 /* Interrupt types to return on RTAS_CHANGE_* */
64 #define RTAS_TYPE_MSI           1
65 #define RTAS_TYPE_MSIX          2
66 
67 SpaprPhbState *spapr_pci_find_phb(SpaprMachineState *spapr, uint64_t buid)
68 {
69     SpaprPhbState *sphb;
70 
71     QLIST_FOREACH(sphb, &spapr->phbs, list) {
72         if (sphb->buid != buid) {
73             continue;
74         }
75         return sphb;
76     }
77 
78     return NULL;
79 }
80 
81 PCIDevice *spapr_pci_find_dev(SpaprMachineState *spapr, uint64_t buid,
82                               uint32_t config_addr)
83 {
84     SpaprPhbState *sphb = spapr_pci_find_phb(spapr, buid);
85     PCIHostState *phb = PCI_HOST_BRIDGE(sphb);
86     int bus_num = (config_addr >> 16) & 0xFF;
87     int devfn = (config_addr >> 8) & 0xFF;
88 
89     if (!phb) {
90         return NULL;
91     }
92 
93     return pci_find_device(phb->bus, bus_num, devfn);
94 }
95 
96 static uint32_t rtas_pci_cfgaddr(uint32_t arg)
97 {
98     /* This handles the encoding of extended config space addresses */
99     return ((arg >> 20) & 0xf00) | (arg & 0xff);
100 }
101 
102 static void finish_read_pci_config(SpaprMachineState *spapr, uint64_t buid,
103                                    uint32_t addr, uint32_t size,
104                                    target_ulong rets)
105 {
106     PCIDevice *pci_dev;
107     uint32_t val;
108 
109     if ((size != 1) && (size != 2) && (size != 4)) {
110         /* access must be 1, 2 or 4 bytes */
111         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
112         return;
113     }
114 
115     pci_dev = spapr_pci_find_dev(spapr, buid, addr);
116     addr = rtas_pci_cfgaddr(addr);
117 
118     if (!pci_dev || (addr % size) || (addr >= pci_config_size(pci_dev))) {
119         /* Access must be to a valid device, within bounds and
120          * naturally aligned */
121         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
122         return;
123     }
124 
125     val = pci_host_config_read_common(pci_dev, addr,
126                                       pci_config_size(pci_dev), size);
127 
128     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
129     rtas_st(rets, 1, val);
130 }
131 
132 static void rtas_ibm_read_pci_config(PowerPCCPU *cpu, SpaprMachineState *spapr,
133                                      uint32_t token, uint32_t nargs,
134                                      target_ulong args,
135                                      uint32_t nret, target_ulong rets)
136 {
137     uint64_t buid;
138     uint32_t size, addr;
139 
140     if ((nargs != 4) || (nret != 2)) {
141         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
142         return;
143     }
144 
145     buid = rtas_ldq(args, 1);
146     size = rtas_ld(args, 3);
147     addr = rtas_ld(args, 0);
148 
149     finish_read_pci_config(spapr, buid, addr, size, rets);
150 }
151 
152 static void rtas_read_pci_config(PowerPCCPU *cpu, SpaprMachineState *spapr,
153                                  uint32_t token, uint32_t nargs,
154                                  target_ulong args,
155                                  uint32_t nret, target_ulong rets)
156 {
157     uint32_t size, addr;
158 
159     if ((nargs != 2) || (nret != 2)) {
160         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
161         return;
162     }
163 
164     size = rtas_ld(args, 1);
165     addr = rtas_ld(args, 0);
166 
167     finish_read_pci_config(spapr, 0, addr, size, rets);
168 }
169 
170 static void finish_write_pci_config(SpaprMachineState *spapr, uint64_t buid,
171                                     uint32_t addr, uint32_t size,
172                                     uint32_t val, target_ulong rets)
173 {
174     PCIDevice *pci_dev;
175 
176     if ((size != 1) && (size != 2) && (size != 4)) {
177         /* access must be 1, 2 or 4 bytes */
178         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
179         return;
180     }
181 
182     pci_dev = spapr_pci_find_dev(spapr, buid, addr);
183     addr = rtas_pci_cfgaddr(addr);
184 
185     if (!pci_dev || (addr % size) || (addr >= pci_config_size(pci_dev))) {
186         /* Access must be to a valid device, within bounds and
187          * naturally aligned */
188         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
189         return;
190     }
191 
192     pci_host_config_write_common(pci_dev, addr, pci_config_size(pci_dev),
193                                  val, size);
194 
195     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
196 }
197 
198 static void rtas_ibm_write_pci_config(PowerPCCPU *cpu, SpaprMachineState *spapr,
199                                       uint32_t token, uint32_t nargs,
200                                       target_ulong args,
201                                       uint32_t nret, target_ulong rets)
202 {
203     uint64_t buid;
204     uint32_t val, size, addr;
205 
206     if ((nargs != 5) || (nret != 1)) {
207         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
208         return;
209     }
210 
211     buid = rtas_ldq(args, 1);
212     val = rtas_ld(args, 4);
213     size = rtas_ld(args, 3);
214     addr = rtas_ld(args, 0);
215 
216     finish_write_pci_config(spapr, buid, addr, size, val, rets);
217 }
218 
219 static void rtas_write_pci_config(PowerPCCPU *cpu, SpaprMachineState *spapr,
220                                   uint32_t token, uint32_t nargs,
221                                   target_ulong args,
222                                   uint32_t nret, target_ulong rets)
223 {
224     uint32_t val, size, addr;
225 
226     if ((nargs != 3) || (nret != 1)) {
227         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
228         return;
229     }
230 
231 
232     val = rtas_ld(args, 2);
233     size = rtas_ld(args, 1);
234     addr = rtas_ld(args, 0);
235 
236     finish_write_pci_config(spapr, 0, addr, size, val, rets);
237 }
238 
239 /*
240  * Set MSI/MSIX message data.
241  * This is required for msi_notify()/msix_notify() which
242  * will write at the addresses via spapr_msi_write().
243  *
244  * If hwaddr == 0, all entries will have .data == first_irq i.e.
245  * table will be reset.
246  */
247 static void spapr_msi_setmsg(PCIDevice *pdev, hwaddr addr, bool msix,
248                              unsigned first_irq, unsigned req_num)
249 {
250     unsigned i;
251     MSIMessage msg = { .address = addr, .data = first_irq };
252 
253     if (!msix) {
254         msi_set_message(pdev, msg);
255         trace_spapr_pci_msi_setup(pdev->name, 0, msg.address);
256         return;
257     }
258 
259     for (i = 0; i < req_num; ++i) {
260         msix_set_message(pdev, i, msg);
261         trace_spapr_pci_msi_setup(pdev->name, i, msg.address);
262         if (addr) {
263             ++msg.data;
264         }
265     }
266 }
267 
268 static void rtas_ibm_change_msi(PowerPCCPU *cpu, SpaprMachineState *spapr,
269                                 uint32_t token, uint32_t nargs,
270                                 target_ulong args, uint32_t nret,
271                                 target_ulong rets)
272 {
273     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
274     uint32_t config_addr = rtas_ld(args, 0);
275     uint64_t buid = rtas_ldq(args, 1);
276     unsigned int func = rtas_ld(args, 3);
277     unsigned int req_num = rtas_ld(args, 4); /* 0 == remove all */
278     unsigned int seq_num = rtas_ld(args, 5);
279     unsigned int ret_intr_type;
280     unsigned int irq, max_irqs = 0;
281     SpaprPhbState *phb = NULL;
282     PCIDevice *pdev = NULL;
283     SpaprPciMsi *msi;
284     int *config_addr_key;
285     Error *err = NULL;
286     int i;
287 
288     /* Fins SpaprPhbState */
289     phb = spapr_pci_find_phb(spapr, buid);
290     if (phb) {
291         pdev = spapr_pci_find_dev(spapr, buid, config_addr);
292     }
293     if (!phb || !pdev) {
294         rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
295         return;
296     }
297 
298     switch (func) {
299     case RTAS_CHANGE_FN:
300         if (msi_present(pdev)) {
301             ret_intr_type = RTAS_TYPE_MSI;
302         } else if (msix_present(pdev)) {
303             ret_intr_type = RTAS_TYPE_MSIX;
304         } else {
305             rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
306             return;
307         }
308         break;
309     case RTAS_CHANGE_MSI_FN:
310         if (msi_present(pdev)) {
311             ret_intr_type = RTAS_TYPE_MSI;
312         } else {
313             rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
314             return;
315         }
316         break;
317     case RTAS_CHANGE_MSIX_FN:
318         if (msix_present(pdev)) {
319             ret_intr_type = RTAS_TYPE_MSIX;
320         } else {
321             rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
322             return;
323         }
324         break;
325     default:
326         error_report("rtas_ibm_change_msi(%u) is not implemented", func);
327         rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
328         return;
329     }
330 
331     msi = (SpaprPciMsi *) g_hash_table_lookup(phb->msi, &config_addr);
332 
333     /* Releasing MSIs */
334     if (!req_num) {
335         if (!msi) {
336             trace_spapr_pci_msi("Releasing wrong config", config_addr);
337             rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
338             return;
339         }
340 
341         if (msi_present(pdev)) {
342             spapr_msi_setmsg(pdev, 0, false, 0, 0);
343         }
344         if (msix_present(pdev)) {
345             spapr_msi_setmsg(pdev, 0, true, 0, 0);
346         }
347         g_hash_table_remove(phb->msi, &config_addr);
348 
349         trace_spapr_pci_msi("Released MSIs", config_addr);
350         rtas_st(rets, 0, RTAS_OUT_SUCCESS);
351         rtas_st(rets, 1, 0);
352         return;
353     }
354 
355     /* Enabling MSI */
356 
357     /* Check if the device supports as many IRQs as requested */
358     if (ret_intr_type == RTAS_TYPE_MSI) {
359         max_irqs = msi_nr_vectors_allocated(pdev);
360     } else if (ret_intr_type == RTAS_TYPE_MSIX) {
361         max_irqs = pdev->msix_entries_nr;
362     }
363     if (!max_irqs) {
364         error_report("Requested interrupt type %d is not enabled for device %x",
365                      ret_intr_type, config_addr);
366         rtas_st(rets, 0, -1); /* Hardware error */
367         return;
368     }
369     /* Correct the number if the guest asked for too many */
370     if (req_num > max_irqs) {
371         trace_spapr_pci_msi_retry(config_addr, req_num, max_irqs);
372         req_num = max_irqs;
373         irq = 0; /* to avoid misleading trace */
374         goto out;
375     }
376 
377     /* Allocate MSIs */
378     if (smc->legacy_irq_allocation) {
379         irq = spapr_irq_find(spapr, req_num, ret_intr_type == RTAS_TYPE_MSI,
380                              &err);
381     } else {
382         irq = spapr_irq_msi_alloc(spapr, req_num,
383                                   ret_intr_type == RTAS_TYPE_MSI, &err);
384     }
385     if (err) {
386         error_reportf_err(err, "Can't allocate MSIs for device %x: ",
387                           config_addr);
388         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
389         return;
390     }
391 
392     for (i = 0; i < req_num; i++) {
393         spapr_irq_claim(spapr, irq + i, false, &err);
394         if (err) {
395             if (i) {
396                 spapr_irq_free(spapr, irq, i);
397             }
398             if (!smc->legacy_irq_allocation) {
399                 spapr_irq_msi_free(spapr, irq, req_num);
400             }
401             error_reportf_err(err, "Can't allocate MSIs for device %x: ",
402                               config_addr);
403             rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
404             return;
405         }
406     }
407 
408     /* Release previous MSIs */
409     if (msi) {
410         g_hash_table_remove(phb->msi, &config_addr);
411     }
412 
413     /* Setup MSI/MSIX vectors in the device (via cfgspace or MSIX BAR) */
414     spapr_msi_setmsg(pdev, SPAPR_PCI_MSI_WINDOW, ret_intr_type == RTAS_TYPE_MSIX,
415                      irq, req_num);
416 
417     /* Add MSI device to cache */
418     msi = g_new(SpaprPciMsi, 1);
419     msi->first_irq = irq;
420     msi->num = req_num;
421     config_addr_key = g_new(int, 1);
422     *config_addr_key = config_addr;
423     g_hash_table_insert(phb->msi, config_addr_key, msi);
424 
425 out:
426     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
427     rtas_st(rets, 1, req_num);
428     rtas_st(rets, 2, ++seq_num);
429     if (nret > 3) {
430         rtas_st(rets, 3, ret_intr_type);
431     }
432 
433     trace_spapr_pci_rtas_ibm_change_msi(config_addr, func, req_num, irq);
434 }
435 
436 static void rtas_ibm_query_interrupt_source_number(PowerPCCPU *cpu,
437                                                    SpaprMachineState *spapr,
438                                                    uint32_t token,
439                                                    uint32_t nargs,
440                                                    target_ulong args,
441                                                    uint32_t nret,
442                                                    target_ulong rets)
443 {
444     uint32_t config_addr = rtas_ld(args, 0);
445     uint64_t buid = rtas_ldq(args, 1);
446     unsigned int intr_src_num = -1, ioa_intr_num = rtas_ld(args, 3);
447     SpaprPhbState *phb = NULL;
448     PCIDevice *pdev = NULL;
449     SpaprPciMsi *msi;
450 
451     /* Find SpaprPhbState */
452     phb = spapr_pci_find_phb(spapr, buid);
453     if (phb) {
454         pdev = spapr_pci_find_dev(spapr, buid, config_addr);
455     }
456     if (!phb || !pdev) {
457         rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
458         return;
459     }
460 
461     /* Find device descriptor and start IRQ */
462     msi = (SpaprPciMsi *) g_hash_table_lookup(phb->msi, &config_addr);
463     if (!msi || !msi->first_irq || !msi->num || (ioa_intr_num >= msi->num)) {
464         trace_spapr_pci_msi("Failed to return vector", config_addr);
465         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
466         return;
467     }
468     intr_src_num = msi->first_irq + ioa_intr_num;
469     trace_spapr_pci_rtas_ibm_query_interrupt_source_number(ioa_intr_num,
470                                                            intr_src_num);
471 
472     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
473     rtas_st(rets, 1, intr_src_num);
474     rtas_st(rets, 2, 1);/* 0 == level; 1 == edge */
475 }
476 
477 static void rtas_ibm_set_eeh_option(PowerPCCPU *cpu,
478                                     SpaprMachineState *spapr,
479                                     uint32_t token, uint32_t nargs,
480                                     target_ulong args, uint32_t nret,
481                                     target_ulong rets)
482 {
483     SpaprPhbState *sphb;
484     uint32_t addr, option;
485     uint64_t buid;
486     int ret;
487 
488     if ((nargs != 4) || (nret != 1)) {
489         goto param_error_exit;
490     }
491 
492     buid = rtas_ldq(args, 1);
493     addr = rtas_ld(args, 0);
494     option = rtas_ld(args, 3);
495 
496     sphb = spapr_pci_find_phb(spapr, buid);
497     if (!sphb) {
498         goto param_error_exit;
499     }
500 
501     if (!spapr_phb_eeh_available(sphb)) {
502         goto param_error_exit;
503     }
504 
505     ret = spapr_phb_vfio_eeh_set_option(sphb, addr, option);
506     rtas_st(rets, 0, ret);
507     return;
508 
509 param_error_exit:
510     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
511 }
512 
513 static void rtas_ibm_get_config_addr_info2(PowerPCCPU *cpu,
514                                            SpaprMachineState *spapr,
515                                            uint32_t token, uint32_t nargs,
516                                            target_ulong args, uint32_t nret,
517                                            target_ulong rets)
518 {
519     SpaprPhbState *sphb;
520     PCIDevice *pdev;
521     uint32_t addr, option;
522     uint64_t buid;
523 
524     if ((nargs != 4) || (nret != 2)) {
525         goto param_error_exit;
526     }
527 
528     buid = rtas_ldq(args, 1);
529     sphb = spapr_pci_find_phb(spapr, buid);
530     if (!sphb) {
531         goto param_error_exit;
532     }
533 
534     if (!spapr_phb_eeh_available(sphb)) {
535         goto param_error_exit;
536     }
537 
538     /*
539      * We always have PE address of form "00BB0001". "BB"
540      * represents the bus number of PE's primary bus.
541      */
542     option = rtas_ld(args, 3);
543     switch (option) {
544     case RTAS_GET_PE_ADDR:
545         addr = rtas_ld(args, 0);
546         pdev = spapr_pci_find_dev(spapr, buid, addr);
547         if (!pdev) {
548             goto param_error_exit;
549         }
550 
551         rtas_st(rets, 1, (pci_bus_num(pci_get_bus(pdev)) << 16) + 1);
552         break;
553     case RTAS_GET_PE_MODE:
554         rtas_st(rets, 1, RTAS_PE_MODE_SHARED);
555         break;
556     default:
557         goto param_error_exit;
558     }
559 
560     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
561     return;
562 
563 param_error_exit:
564     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
565 }
566 
567 static void rtas_ibm_read_slot_reset_state2(PowerPCCPU *cpu,
568                                             SpaprMachineState *spapr,
569                                             uint32_t token, uint32_t nargs,
570                                             target_ulong args, uint32_t nret,
571                                             target_ulong rets)
572 {
573     SpaprPhbState *sphb;
574     uint64_t buid;
575     int state, ret;
576 
577     if ((nargs != 3) || (nret != 4 && nret != 5)) {
578         goto param_error_exit;
579     }
580 
581     buid = rtas_ldq(args, 1);
582     sphb = spapr_pci_find_phb(spapr, buid);
583     if (!sphb) {
584         goto param_error_exit;
585     }
586 
587     if (!spapr_phb_eeh_available(sphb)) {
588         goto param_error_exit;
589     }
590 
591     ret = spapr_phb_vfio_eeh_get_state(sphb, &state);
592     rtas_st(rets, 0, ret);
593     if (ret != RTAS_OUT_SUCCESS) {
594         return;
595     }
596 
597     rtas_st(rets, 1, state);
598     rtas_st(rets, 2, RTAS_EEH_SUPPORT);
599     rtas_st(rets, 3, RTAS_EEH_PE_UNAVAIL_INFO);
600     if (nret >= 5) {
601         rtas_st(rets, 4, RTAS_EEH_PE_RECOVER_INFO);
602     }
603     return;
604 
605 param_error_exit:
606     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
607 }
608 
609 static void rtas_ibm_set_slot_reset(PowerPCCPU *cpu,
610                                     SpaprMachineState *spapr,
611                                     uint32_t token, uint32_t nargs,
612                                     target_ulong args, uint32_t nret,
613                                     target_ulong rets)
614 {
615     SpaprPhbState *sphb;
616     uint32_t option;
617     uint64_t buid;
618     int ret;
619 
620     if ((nargs != 4) || (nret != 1)) {
621         goto param_error_exit;
622     }
623 
624     buid = rtas_ldq(args, 1);
625     option = rtas_ld(args, 3);
626     sphb = spapr_pci_find_phb(spapr, buid);
627     if (!sphb) {
628         goto param_error_exit;
629     }
630 
631     if (!spapr_phb_eeh_available(sphb)) {
632         goto param_error_exit;
633     }
634 
635     ret = spapr_phb_vfio_eeh_reset(sphb, option);
636     rtas_st(rets, 0, ret);
637     return;
638 
639 param_error_exit:
640     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
641 }
642 
643 static void rtas_ibm_configure_pe(PowerPCCPU *cpu,
644                                   SpaprMachineState *spapr,
645                                   uint32_t token, uint32_t nargs,
646                                   target_ulong args, uint32_t nret,
647                                   target_ulong rets)
648 {
649     SpaprPhbState *sphb;
650     uint64_t buid;
651     int ret;
652 
653     if ((nargs != 3) || (nret != 1)) {
654         goto param_error_exit;
655     }
656 
657     buid = rtas_ldq(args, 1);
658     sphb = spapr_pci_find_phb(spapr, buid);
659     if (!sphb) {
660         goto param_error_exit;
661     }
662 
663     if (!spapr_phb_eeh_available(sphb)) {
664         goto param_error_exit;
665     }
666 
667     ret = spapr_phb_vfio_eeh_configure(sphb);
668     rtas_st(rets, 0, ret);
669     return;
670 
671 param_error_exit:
672     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
673 }
674 
675 /* To support it later */
676 static void rtas_ibm_slot_error_detail(PowerPCCPU *cpu,
677                                        SpaprMachineState *spapr,
678                                        uint32_t token, uint32_t nargs,
679                                        target_ulong args, uint32_t nret,
680                                        target_ulong rets)
681 {
682     SpaprPhbState *sphb;
683     int option;
684     uint64_t buid;
685 
686     if ((nargs != 8) || (nret != 1)) {
687         goto param_error_exit;
688     }
689 
690     buid = rtas_ldq(args, 1);
691     sphb = spapr_pci_find_phb(spapr, buid);
692     if (!sphb) {
693         goto param_error_exit;
694     }
695 
696     if (!spapr_phb_eeh_available(sphb)) {
697         goto param_error_exit;
698     }
699 
700     option = rtas_ld(args, 7);
701     switch (option) {
702     case RTAS_SLOT_TEMP_ERR_LOG:
703     case RTAS_SLOT_PERM_ERR_LOG:
704         break;
705     default:
706         goto param_error_exit;
707     }
708 
709     /* We don't have error log yet */
710     rtas_st(rets, 0, RTAS_OUT_NO_ERRORS_FOUND);
711     return;
712 
713 param_error_exit:
714     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
715 }
716 
717 static void pci_spapr_set_irq(void *opaque, int irq_num, int level)
718 {
719     /*
720      * Here we use the number returned by pci_swizzle_map_irq_fn to find a
721      * corresponding qemu_irq.
722      */
723     SpaprPhbState *phb = opaque;
724     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
725 
726     trace_spapr_pci_lsi_set(phb->dtbusname, irq_num, phb->lsi_table[irq_num].irq);
727     qemu_set_irq(spapr_qirq(spapr, phb->lsi_table[irq_num].irq), level);
728 }
729 
730 static PCIINTxRoute spapr_route_intx_pin_to_irq(void *opaque, int pin)
731 {
732     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(opaque);
733     PCIINTxRoute route;
734 
735     route.mode = PCI_INTX_ENABLED;
736     route.irq = sphb->lsi_table[pin].irq;
737 
738     return route;
739 }
740 
741 static uint64_t spapr_msi_read(void *opaque, hwaddr addr, unsigned size)
742 {
743     qemu_log_mask(LOG_GUEST_ERROR, "%s: invalid access\n", __func__);
744     return 0;
745 }
746 
747 /*
748  * MSI/MSIX memory region implementation.
749  * The handler handles both MSI and MSIX.
750  * The vector number is encoded in least bits in data.
751  */
752 static void spapr_msi_write(void *opaque, hwaddr addr,
753                             uint64_t data, unsigned size)
754 {
755     SpaprMachineState *spapr = opaque;
756     uint32_t irq = data;
757 
758     trace_spapr_pci_msi_write(addr, data, irq);
759 
760     qemu_irq_pulse(spapr_qirq(spapr, irq));
761 }
762 
763 static const MemoryRegionOps spapr_msi_ops = {
764     /*
765      * .read result is undefined by PCI spec.
766      * define .read method to avoid assert failure in memory_region_init_io
767      */
768     .read = spapr_msi_read,
769     .write = spapr_msi_write,
770     .endianness = DEVICE_LITTLE_ENDIAN
771 };
772 
773 /*
774  * PHB PCI device
775  */
776 static AddressSpace *spapr_pci_dma_iommu(PCIBus *bus, void *opaque, int devfn)
777 {
778     SpaprPhbState *phb = opaque;
779 
780     return &phb->iommu_as;
781 }
782 
783 static char *spapr_phb_vfio_get_loc_code(SpaprPhbState *sphb,  PCIDevice *pdev)
784 {
785     char *path = NULL, *buf = NULL, *host = NULL;
786 
787     /* Get the PCI VFIO host id */
788     host = object_property_get_str(OBJECT(pdev), "host", NULL);
789     if (!host) {
790         goto err_out;
791     }
792 
793     /* Construct the path of the file that will give us the DT location */
794     path = g_strdup_printf("/sys/bus/pci/devices/%s/devspec", host);
795     g_free(host);
796     if (!g_file_get_contents(path, &buf, NULL, NULL)) {
797         goto err_out;
798     }
799     g_free(path);
800 
801     /* Construct and read from host device tree the loc-code */
802     path = g_strdup_printf("/proc/device-tree%s/ibm,loc-code", buf);
803     g_free(buf);
804     if (!g_file_get_contents(path, &buf, NULL, NULL)) {
805         goto err_out;
806     }
807     return buf;
808 
809 err_out:
810     g_free(path);
811     return NULL;
812 }
813 
814 static char *spapr_phb_get_loc_code(SpaprPhbState *sphb, PCIDevice *pdev)
815 {
816     char *buf;
817     const char *devtype = "qemu";
818     uint32_t busnr = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(pdev))));
819 
820     if (object_dynamic_cast(OBJECT(pdev), "vfio-pci")) {
821         buf = spapr_phb_vfio_get_loc_code(sphb, pdev);
822         if (buf) {
823             return buf;
824         }
825         devtype = "vfio";
826     }
827     /*
828      * For emulated devices and VFIO-failure case, make up
829      * the loc-code.
830      */
831     buf = g_strdup_printf("%s_%s:%04x:%02x:%02x.%x",
832                           devtype, pdev->name, sphb->index, busnr,
833                           PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
834     return buf;
835 }
836 
837 /* Macros to operate with address in OF binding to PCI */
838 #define b_x(x, p, l)    (((x) & ((1<<(l))-1)) << (p))
839 #define b_n(x)          b_x((x), 31, 1) /* 0 if relocatable */
840 #define b_p(x)          b_x((x), 30, 1) /* 1 if prefetchable */
841 #define b_t(x)          b_x((x), 29, 1) /* 1 if the address is aliased */
842 #define b_ss(x)         b_x((x), 24, 2) /* the space code */
843 #define b_bbbbbbbb(x)   b_x((x), 16, 8) /* bus number */
844 #define b_ddddd(x)      b_x((x), 11, 5) /* device number */
845 #define b_fff(x)        b_x((x), 8, 3)  /* function number */
846 #define b_rrrrrrrr(x)   b_x((x), 0, 8)  /* register number */
847 
848 /* for 'reg' OF properties */
849 #define RESOURCE_CELLS_SIZE 2
850 #define RESOURCE_CELLS_ADDRESS 3
851 
852 typedef struct ResourceFields {
853     uint32_t phys_hi;
854     uint32_t phys_mid;
855     uint32_t phys_lo;
856     uint32_t size_hi;
857     uint32_t size_lo;
858 } QEMU_PACKED ResourceFields;
859 
860 typedef struct ResourceProps {
861     ResourceFields reg[8];
862     uint32_t reg_len;
863 } ResourceProps;
864 
865 /* fill in the 'reg' OF properties for
866  * a PCI device. 'reg' describes resource requirements for a
867  * device's IO/MEM regions.
868  *
869  * the property is an array of ('phys-addr', 'size') pairs describing
870  * the addressable regions of the PCI device, where 'phys-addr' is a
871  * RESOURCE_CELLS_ADDRESS-tuple of 32-bit integers corresponding to
872  * (phys.hi, phys.mid, phys.lo), and 'size' is a
873  * RESOURCE_CELLS_SIZE-tuple corresponding to (size.hi, size.lo).
874  *
875  * phys.hi = 0xYYXXXXZZ, where:
876  *   0xYY = npt000ss
877  *          |||   |
878  *          |||   +-- space code
879  *          |||               |
880  *          |||               +  00 if configuration space
881  *          |||               +  01 if IO region,
882  *          |||               +  10 if 32-bit MEM region
883  *          |||               +  11 if 64-bit MEM region
884  *          |||
885  *          ||+------ for non-relocatable IO: 1 if aliased
886  *          ||        for relocatable IO: 1 if below 64KB
887  *          ||        for MEM: 1 if below 1MB
888  *          |+------- 1 if region is prefetchable
889  *          +-------- 1 if region is non-relocatable
890  *   0xXXXX = bbbbbbbb dddddfff, encoding bus, slot, and function
891  *            bits respectively
892  *   0xZZ = rrrrrrrr, the register number of the BAR corresponding
893  *          to the region
894  *
895  * phys.mid and phys.lo correspond respectively to the hi/lo portions
896  * of the actual address of the region.
897  *
898  * note also that addresses defined in this property are, at least
899  * for PAPR guests, relative to the PHBs IO/MEM windows, and
900  * correspond directly to the addresses in the BARs.
901  *
902  * in accordance with PCI Bus Binding to Open Firmware,
903  * IEEE Std 1275-1994, section 4.1.1, as implemented by PAPR+ v2.7,
904  * Appendix C.
905  */
906 static void populate_resource_props(PCIDevice *d, ResourceProps *rp)
907 {
908     int bus_num = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(d))));
909     uint32_t dev_id = (b_bbbbbbbb(bus_num) |
910                        b_ddddd(PCI_SLOT(d->devfn)) |
911                        b_fff(PCI_FUNC(d->devfn)));
912     ResourceFields *reg;
913     int i, reg_idx = 0;
914 
915     /* config space region */
916     reg = &rp->reg[reg_idx++];
917     reg->phys_hi = cpu_to_be32(dev_id);
918     reg->phys_mid = 0;
919     reg->phys_lo = 0;
920     reg->size_hi = 0;
921     reg->size_lo = 0;
922 
923     for (i = 0; i < PCI_NUM_REGIONS; i++) {
924         if (!d->io_regions[i].size) {
925             continue;
926         }
927 
928         reg = &rp->reg[reg_idx++];
929 
930         reg->phys_hi = cpu_to_be32(dev_id | b_rrrrrrrr(pci_bar(d, i)));
931         if (d->io_regions[i].type & PCI_BASE_ADDRESS_SPACE_IO) {
932             reg->phys_hi |= cpu_to_be32(b_ss(1));
933         } else if (d->io_regions[i].type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
934             reg->phys_hi |= cpu_to_be32(b_ss(3));
935         } else {
936             reg->phys_hi |= cpu_to_be32(b_ss(2));
937         }
938         reg->phys_mid = 0;
939         reg->phys_lo = 0;
940         reg->size_hi = cpu_to_be32(d->io_regions[i].size >> 32);
941         reg->size_lo = cpu_to_be32(d->io_regions[i].size);
942     }
943 
944     rp->reg_len = reg_idx * sizeof(ResourceFields);
945 }
946 
947 typedef struct PCIClass PCIClass;
948 typedef struct PCISubClass PCISubClass;
949 typedef struct PCIIFace PCIIFace;
950 
951 struct PCIIFace {
952     int iface;
953     const char *name;
954 };
955 
956 struct PCISubClass {
957     int subclass;
958     const char *name;
959     const PCIIFace *iface;
960 };
961 
962 struct PCIClass {
963     const char *name;
964     const PCISubClass *subc;
965 };
966 
967 static const PCISubClass undef_subclass[] = {
968     { PCI_CLASS_NOT_DEFINED_VGA, "display", NULL },
969     { 0xFF, NULL, NULL },
970 };
971 
972 static const PCISubClass mass_subclass[] = {
973     { PCI_CLASS_STORAGE_SCSI, "scsi", NULL },
974     { PCI_CLASS_STORAGE_IDE, "ide", NULL },
975     { PCI_CLASS_STORAGE_FLOPPY, "fdc", NULL },
976     { PCI_CLASS_STORAGE_IPI, "ipi", NULL },
977     { PCI_CLASS_STORAGE_RAID, "raid", NULL },
978     { PCI_CLASS_STORAGE_ATA, "ata", NULL },
979     { PCI_CLASS_STORAGE_SATA, "sata", NULL },
980     { PCI_CLASS_STORAGE_SAS, "sas", NULL },
981     { 0xFF, NULL, NULL },
982 };
983 
984 static const PCISubClass net_subclass[] = {
985     { PCI_CLASS_NETWORK_ETHERNET, "ethernet", NULL },
986     { PCI_CLASS_NETWORK_TOKEN_RING, "token-ring", NULL },
987     { PCI_CLASS_NETWORK_FDDI, "fddi", NULL },
988     { PCI_CLASS_NETWORK_ATM, "atm", NULL },
989     { PCI_CLASS_NETWORK_ISDN, "isdn", NULL },
990     { PCI_CLASS_NETWORK_WORLDFIP, "worldfip", NULL },
991     { PCI_CLASS_NETWORK_PICMG214, "picmg", NULL },
992     { 0xFF, NULL, NULL },
993 };
994 
995 static const PCISubClass displ_subclass[] = {
996     { PCI_CLASS_DISPLAY_VGA, "vga", NULL },
997     { PCI_CLASS_DISPLAY_XGA, "xga", NULL },
998     { PCI_CLASS_DISPLAY_3D, "3d-controller", NULL },
999     { 0xFF, NULL, NULL },
1000 };
1001 
1002 static const PCISubClass media_subclass[] = {
1003     { PCI_CLASS_MULTIMEDIA_VIDEO, "video", NULL },
1004     { PCI_CLASS_MULTIMEDIA_AUDIO, "sound", NULL },
1005     { PCI_CLASS_MULTIMEDIA_PHONE, "telephony", NULL },
1006     { 0xFF, NULL, NULL },
1007 };
1008 
1009 static const PCISubClass mem_subclass[] = {
1010     { PCI_CLASS_MEMORY_RAM, "memory", NULL },
1011     { PCI_CLASS_MEMORY_FLASH, "flash", NULL },
1012     { 0xFF, NULL, NULL },
1013 };
1014 
1015 static const PCISubClass bridg_subclass[] = {
1016     { PCI_CLASS_BRIDGE_HOST, "host", NULL },
1017     { PCI_CLASS_BRIDGE_ISA, "isa", NULL },
1018     { PCI_CLASS_BRIDGE_EISA, "eisa", NULL },
1019     { PCI_CLASS_BRIDGE_MC, "mca", NULL },
1020     { PCI_CLASS_BRIDGE_PCI, "pci", NULL },
1021     { PCI_CLASS_BRIDGE_PCMCIA, "pcmcia", NULL },
1022     { PCI_CLASS_BRIDGE_NUBUS, "nubus", NULL },
1023     { PCI_CLASS_BRIDGE_CARDBUS, "cardbus", NULL },
1024     { PCI_CLASS_BRIDGE_RACEWAY, "raceway", NULL },
1025     { PCI_CLASS_BRIDGE_PCI_SEMITP, "semi-transparent-pci", NULL },
1026     { PCI_CLASS_BRIDGE_IB_PCI, "infiniband", NULL },
1027     { 0xFF, NULL, NULL },
1028 };
1029 
1030 static const PCISubClass comm_subclass[] = {
1031     { PCI_CLASS_COMMUNICATION_SERIAL, "serial", NULL },
1032     { PCI_CLASS_COMMUNICATION_PARALLEL, "parallel", NULL },
1033     { PCI_CLASS_COMMUNICATION_MULTISERIAL, "multiport-serial", NULL },
1034     { PCI_CLASS_COMMUNICATION_MODEM, "modem", NULL },
1035     { PCI_CLASS_COMMUNICATION_GPIB, "gpib", NULL },
1036     { PCI_CLASS_COMMUNICATION_SC, "smart-card", NULL },
1037     { 0xFF, NULL, NULL, },
1038 };
1039 
1040 static const PCIIFace pic_iface[] = {
1041     { PCI_CLASS_SYSTEM_PIC_IOAPIC, "io-apic" },
1042     { PCI_CLASS_SYSTEM_PIC_IOXAPIC, "io-xapic" },
1043     { 0xFF, NULL },
1044 };
1045 
1046 static const PCISubClass sys_subclass[] = {
1047     { PCI_CLASS_SYSTEM_PIC, "interrupt-controller", pic_iface },
1048     { PCI_CLASS_SYSTEM_DMA, "dma-controller", NULL },
1049     { PCI_CLASS_SYSTEM_TIMER, "timer", NULL },
1050     { PCI_CLASS_SYSTEM_RTC, "rtc", NULL },
1051     { PCI_CLASS_SYSTEM_PCI_HOTPLUG, "hot-plug-controller", NULL },
1052     { PCI_CLASS_SYSTEM_SDHCI, "sd-host-controller", NULL },
1053     { 0xFF, NULL, NULL },
1054 };
1055 
1056 static const PCISubClass inp_subclass[] = {
1057     { PCI_CLASS_INPUT_KEYBOARD, "keyboard", NULL },
1058     { PCI_CLASS_INPUT_PEN, "pen", NULL },
1059     { PCI_CLASS_INPUT_MOUSE, "mouse", NULL },
1060     { PCI_CLASS_INPUT_SCANNER, "scanner", NULL },
1061     { PCI_CLASS_INPUT_GAMEPORT, "gameport", NULL },
1062     { 0xFF, NULL, NULL },
1063 };
1064 
1065 static const PCISubClass dock_subclass[] = {
1066     { PCI_CLASS_DOCKING_GENERIC, "dock", NULL },
1067     { 0xFF, NULL, NULL },
1068 };
1069 
1070 static const PCISubClass cpu_subclass[] = {
1071     { PCI_CLASS_PROCESSOR_PENTIUM, "pentium", NULL },
1072     { PCI_CLASS_PROCESSOR_POWERPC, "powerpc", NULL },
1073     { PCI_CLASS_PROCESSOR_MIPS, "mips", NULL },
1074     { PCI_CLASS_PROCESSOR_CO, "co-processor", NULL },
1075     { 0xFF, NULL, NULL },
1076 };
1077 
1078 static const PCIIFace usb_iface[] = {
1079     { PCI_CLASS_SERIAL_USB_UHCI, "usb-uhci" },
1080     { PCI_CLASS_SERIAL_USB_OHCI, "usb-ohci", },
1081     { PCI_CLASS_SERIAL_USB_EHCI, "usb-ehci" },
1082     { PCI_CLASS_SERIAL_USB_XHCI, "usb-xhci" },
1083     { PCI_CLASS_SERIAL_USB_UNKNOWN, "usb-unknown" },
1084     { PCI_CLASS_SERIAL_USB_DEVICE, "usb-device" },
1085     { 0xFF, NULL },
1086 };
1087 
1088 static const PCISubClass ser_subclass[] = {
1089     { PCI_CLASS_SERIAL_FIREWIRE, "firewire", NULL },
1090     { PCI_CLASS_SERIAL_ACCESS, "access-bus", NULL },
1091     { PCI_CLASS_SERIAL_SSA, "ssa", NULL },
1092     { PCI_CLASS_SERIAL_USB, "usb", usb_iface },
1093     { PCI_CLASS_SERIAL_FIBER, "fibre-channel", NULL },
1094     { PCI_CLASS_SERIAL_SMBUS, "smb", NULL },
1095     { PCI_CLASS_SERIAL_IB, "infiniband", NULL },
1096     { PCI_CLASS_SERIAL_IPMI, "ipmi", NULL },
1097     { PCI_CLASS_SERIAL_SERCOS, "sercos", NULL },
1098     { PCI_CLASS_SERIAL_CANBUS, "canbus", NULL },
1099     { 0xFF, NULL, NULL },
1100 };
1101 
1102 static const PCISubClass wrl_subclass[] = {
1103     { PCI_CLASS_WIRELESS_IRDA, "irda", NULL },
1104     { PCI_CLASS_WIRELESS_CIR, "consumer-ir", NULL },
1105     { PCI_CLASS_WIRELESS_RF_CONTROLLER, "rf-controller", NULL },
1106     { PCI_CLASS_WIRELESS_BLUETOOTH, "bluetooth", NULL },
1107     { PCI_CLASS_WIRELESS_BROADBAND, "broadband", NULL },
1108     { 0xFF, NULL, NULL },
1109 };
1110 
1111 static const PCISubClass sat_subclass[] = {
1112     { PCI_CLASS_SATELLITE_TV, "satellite-tv", NULL },
1113     { PCI_CLASS_SATELLITE_AUDIO, "satellite-audio", NULL },
1114     { PCI_CLASS_SATELLITE_VOICE, "satellite-voice", NULL },
1115     { PCI_CLASS_SATELLITE_DATA, "satellite-data", NULL },
1116     { 0xFF, NULL, NULL },
1117 };
1118 
1119 static const PCISubClass crypt_subclass[] = {
1120     { PCI_CLASS_CRYPT_NETWORK, "network-encryption", NULL },
1121     { PCI_CLASS_CRYPT_ENTERTAINMENT,
1122       "entertainment-encryption", NULL },
1123     { 0xFF, NULL, NULL },
1124 };
1125 
1126 static const PCISubClass spc_subclass[] = {
1127     { PCI_CLASS_SP_DPIO, "dpio", NULL },
1128     { PCI_CLASS_SP_PERF, "counter", NULL },
1129     { PCI_CLASS_SP_SYNCH, "measurement", NULL },
1130     { PCI_CLASS_SP_MANAGEMENT, "management-card", NULL },
1131     { 0xFF, NULL, NULL },
1132 };
1133 
1134 static const PCIClass pci_classes[] = {
1135     { "legacy-device", undef_subclass },
1136     { "mass-storage",  mass_subclass },
1137     { "network", net_subclass },
1138     { "display", displ_subclass, },
1139     { "multimedia-device", media_subclass },
1140     { "memory-controller", mem_subclass },
1141     { "unknown-bridge", bridg_subclass },
1142     { "communication-controller", comm_subclass},
1143     { "system-peripheral", sys_subclass },
1144     { "input-controller", inp_subclass },
1145     { "docking-station", dock_subclass },
1146     { "cpu", cpu_subclass },
1147     { "serial-bus", ser_subclass },
1148     { "wireless-controller", wrl_subclass },
1149     { "intelligent-io", NULL },
1150     { "satellite-device", sat_subclass },
1151     { "encryption", crypt_subclass },
1152     { "data-processing-controller", spc_subclass },
1153 };
1154 
1155 static const char *dt_name_from_class(uint8_t class, uint8_t subclass,
1156                                       uint8_t iface)
1157 {
1158     const PCIClass *pclass;
1159     const PCISubClass *psubclass;
1160     const PCIIFace *piface;
1161     const char *name;
1162 
1163     if (class >= ARRAY_SIZE(pci_classes)) {
1164         return "pci";
1165     }
1166 
1167     pclass = pci_classes + class;
1168     name = pclass->name;
1169 
1170     if (pclass->subc == NULL) {
1171         return name;
1172     }
1173 
1174     psubclass = pclass->subc;
1175     while ((psubclass->subclass & 0xff) != 0xff) {
1176         if ((psubclass->subclass & 0xff) == subclass) {
1177             name = psubclass->name;
1178             break;
1179         }
1180         psubclass++;
1181     }
1182 
1183     piface = psubclass->iface;
1184     if (piface == NULL) {
1185         return name;
1186     }
1187     while ((piface->iface & 0xff) != 0xff) {
1188         if ((piface->iface & 0xff) == iface) {
1189             name = piface->name;
1190             break;
1191         }
1192         piface++;
1193     }
1194 
1195     return name;
1196 }
1197 
1198 /*
1199  * DRC helper functions
1200  */
1201 
1202 static uint32_t drc_id_from_devfn(SpaprPhbState *phb,
1203                                   uint8_t chassis, int32_t devfn)
1204 {
1205     return (phb->index << 16) | (chassis << 8) | devfn;
1206 }
1207 
1208 static SpaprDrc *drc_from_devfn(SpaprPhbState *phb,
1209                                 uint8_t chassis, int32_t devfn)
1210 {
1211     return spapr_drc_by_id(TYPE_SPAPR_DRC_PCI,
1212                            drc_id_from_devfn(phb, chassis, devfn));
1213 }
1214 
1215 static uint8_t chassis_from_bus(PCIBus *bus)
1216 {
1217     if (pci_bus_is_root(bus)) {
1218         return 0;
1219     } else {
1220         PCIDevice *bridge = pci_bridge_get_device(bus);
1221 
1222         return object_property_get_uint(OBJECT(bridge), "chassis_nr",
1223                                         &error_abort);
1224     }
1225 }
1226 
1227 static SpaprDrc *drc_from_dev(SpaprPhbState *phb, PCIDevice *dev)
1228 {
1229     uint8_t chassis = chassis_from_bus(pci_get_bus(dev));
1230 
1231     return drc_from_devfn(phb, chassis, dev->devfn);
1232 }
1233 
1234 static void add_drcs(SpaprPhbState *phb, PCIBus *bus)
1235 {
1236     Object *owner;
1237     int i;
1238     uint8_t chassis;
1239 
1240     if (!phb->dr_enabled) {
1241         return;
1242     }
1243 
1244     chassis = chassis_from_bus(bus);
1245 
1246     if (pci_bus_is_root(bus)) {
1247         owner = OBJECT(phb);
1248     } else {
1249         owner = OBJECT(pci_bridge_get_device(bus));
1250     }
1251 
1252     for (i = 0; i < PCI_SLOT_MAX * PCI_FUNC_MAX; i++) {
1253         spapr_dr_connector_new(owner, TYPE_SPAPR_DRC_PCI,
1254                                drc_id_from_devfn(phb, chassis, i));
1255     }
1256 }
1257 
1258 static void remove_drcs(SpaprPhbState *phb, PCIBus *bus)
1259 {
1260     int i;
1261     uint8_t chassis;
1262 
1263     if (!phb->dr_enabled) {
1264         return;
1265     }
1266 
1267     chassis = chassis_from_bus(bus);
1268 
1269     for (i = PCI_SLOT_MAX * PCI_FUNC_MAX - 1; i >= 0; i--) {
1270         SpaprDrc *drc = drc_from_devfn(phb, chassis, i);
1271 
1272         if (drc) {
1273             object_unparent(OBJECT(drc));
1274         }
1275     }
1276 }
1277 
1278 typedef struct PciWalkFdt {
1279     void *fdt;
1280     int offset;
1281     SpaprPhbState *sphb;
1282     int err;
1283 } PciWalkFdt;
1284 
1285 static int spapr_dt_pci_device(SpaprPhbState *sphb, PCIDevice *dev,
1286                                void *fdt, int parent_offset);
1287 
1288 static void spapr_dt_pci_device_cb(PCIBus *bus, PCIDevice *pdev,
1289                                    void *opaque)
1290 {
1291     PciWalkFdt *p = opaque;
1292     int err;
1293 
1294     if (p->err) {
1295         /* Something's already broken, don't keep going */
1296         return;
1297     }
1298 
1299     err = spapr_dt_pci_device(p->sphb, pdev, p->fdt, p->offset);
1300     if (err < 0) {
1301         p->err = err;
1302     }
1303 }
1304 
1305 /* Augment PCI device node with bridge specific information */
1306 static int spapr_dt_pci_bus(SpaprPhbState *sphb, PCIBus *bus,
1307                                void *fdt, int offset)
1308 {
1309     Object *owner;
1310     PciWalkFdt cbinfo = {
1311         .fdt = fdt,
1312         .offset = offset,
1313         .sphb = sphb,
1314         .err = 0,
1315     };
1316     int ret;
1317 
1318     _FDT(fdt_setprop_cell(fdt, offset, "#address-cells",
1319                           RESOURCE_CELLS_ADDRESS));
1320     _FDT(fdt_setprop_cell(fdt, offset, "#size-cells",
1321                           RESOURCE_CELLS_SIZE));
1322 
1323     assert(bus);
1324     pci_for_each_device_reverse(bus, pci_bus_num(bus),
1325                                 spapr_dt_pci_device_cb, &cbinfo);
1326     if (cbinfo.err) {
1327         return cbinfo.err;
1328     }
1329 
1330     if (pci_bus_is_root(bus)) {
1331         owner = OBJECT(sphb);
1332     } else {
1333         owner = OBJECT(pci_bridge_get_device(bus));
1334     }
1335 
1336     ret = spapr_dt_drc(fdt, offset, owner,
1337                        SPAPR_DR_CONNECTOR_TYPE_PCI);
1338     if (ret) {
1339         return ret;
1340     }
1341 
1342     return offset;
1343 }
1344 
1345 char *spapr_pci_fw_dev_name(PCIDevice *dev)
1346 {
1347     const gchar *basename;
1348     int slot = PCI_SLOT(dev->devfn);
1349     int func = PCI_FUNC(dev->devfn);
1350     uint32_t ccode = pci_default_read_config(dev, PCI_CLASS_PROG, 3);
1351 
1352     basename = dt_name_from_class((ccode >> 16) & 0xff, (ccode >> 8) & 0xff,
1353                                   ccode & 0xff);
1354 
1355     if (func != 0) {
1356         return g_strdup_printf("%s@%x,%x", basename, slot, func);
1357     } else {
1358         return g_strdup_printf("%s@%x", basename, slot);
1359     }
1360 }
1361 
1362 /* create OF node for pci device and required OF DT properties */
1363 static int spapr_dt_pci_device(SpaprPhbState *sphb, PCIDevice *dev,
1364                                void *fdt, int parent_offset)
1365 {
1366     int offset;
1367     g_autofree gchar *nodename = spapr_pci_fw_dev_name(dev);
1368     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
1369     ResourceProps rp;
1370     SpaprDrc *drc = drc_from_dev(sphb, dev);
1371     uint32_t vendor_id = pci_default_read_config(dev, PCI_VENDOR_ID, 2);
1372     uint32_t device_id = pci_default_read_config(dev, PCI_DEVICE_ID, 2);
1373     uint32_t revision_id = pci_default_read_config(dev, PCI_REVISION_ID, 1);
1374     uint32_t ccode = pci_default_read_config(dev, PCI_CLASS_PROG, 3);
1375     uint32_t irq_pin = pci_default_read_config(dev, PCI_INTERRUPT_PIN, 1);
1376     uint32_t subsystem_id = pci_default_read_config(dev, PCI_SUBSYSTEM_ID, 2);
1377     uint32_t subsystem_vendor_id =
1378         pci_default_read_config(dev, PCI_SUBSYSTEM_VENDOR_ID, 2);
1379     uint32_t cache_line_size =
1380         pci_default_read_config(dev, PCI_CACHE_LINE_SIZE, 1);
1381     uint32_t pci_status = pci_default_read_config(dev, PCI_STATUS, 2);
1382     gchar *loc_code;
1383 
1384     _FDT(offset = fdt_add_subnode(fdt, parent_offset, nodename));
1385 
1386     /* in accordance with PAPR+ v2.7 13.6.3, Table 181 */
1387     _FDT(fdt_setprop_cell(fdt, offset, "vendor-id", vendor_id));
1388     _FDT(fdt_setprop_cell(fdt, offset, "device-id", device_id));
1389     _FDT(fdt_setprop_cell(fdt, offset, "revision-id", revision_id));
1390 
1391     _FDT(fdt_setprop_cell(fdt, offset, "class-code", ccode));
1392     if (irq_pin) {
1393         _FDT(fdt_setprop_cell(fdt, offset, "interrupts", irq_pin));
1394     }
1395 
1396     if (subsystem_id) {
1397         _FDT(fdt_setprop_cell(fdt, offset, "subsystem-id", subsystem_id));
1398     }
1399 
1400     if (subsystem_vendor_id) {
1401         _FDT(fdt_setprop_cell(fdt, offset, "subsystem-vendor-id",
1402                               subsystem_vendor_id));
1403     }
1404 
1405     _FDT(fdt_setprop_cell(fdt, offset, "cache-line-size", cache_line_size));
1406 
1407 
1408     /* the following fdt cells are masked off the pci status register */
1409     _FDT(fdt_setprop_cell(fdt, offset, "devsel-speed",
1410                           PCI_STATUS_DEVSEL_MASK & pci_status));
1411 
1412     if (pci_status & PCI_STATUS_FAST_BACK) {
1413         _FDT(fdt_setprop(fdt, offset, "fast-back-to-back", NULL, 0));
1414     }
1415     if (pci_status & PCI_STATUS_66MHZ) {
1416         _FDT(fdt_setprop(fdt, offset, "66mhz-capable", NULL, 0));
1417     }
1418     if (pci_status & PCI_STATUS_UDF) {
1419         _FDT(fdt_setprop(fdt, offset, "udf-supported", NULL, 0));
1420     }
1421 
1422     loc_code = spapr_phb_get_loc_code(sphb, dev);
1423     _FDT(fdt_setprop_string(fdt, offset, "ibm,loc-code", loc_code));
1424     g_free(loc_code);
1425 
1426     if (drc) {
1427         _FDT(fdt_setprop_cell(fdt, offset, "ibm,my-drc-index",
1428                               spapr_drc_index(drc)));
1429     }
1430 
1431     if (msi_present(dev)) {
1432         uint32_t max_msi = msi_nr_vectors_allocated(dev);
1433         if (max_msi) {
1434             _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi", max_msi));
1435         }
1436     }
1437     if (msix_present(dev)) {
1438         uint32_t max_msix = dev->msix_entries_nr;
1439         if (max_msix) {
1440             _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi-x", max_msix));
1441         }
1442     }
1443 
1444     populate_resource_props(dev, &rp);
1445     _FDT(fdt_setprop(fdt, offset, "reg", (uint8_t *)rp.reg, rp.reg_len));
1446 
1447     if (sphb->pcie_ecs && pci_is_express(dev)) {
1448         _FDT(fdt_setprop_cell(fdt, offset, "ibm,pci-config-space-type", 0x1));
1449     }
1450 
1451     spapr_phb_nvgpu_populate_pcidev_dt(dev, fdt, offset, sphb);
1452 
1453     if (!pc->is_bridge) {
1454         /* Properties only for non-bridges */
1455         uint32_t min_grant = pci_default_read_config(dev, PCI_MIN_GNT, 1);
1456         uint32_t max_latency = pci_default_read_config(dev, PCI_MAX_LAT, 1);
1457         _FDT(fdt_setprop_cell(fdt, offset, "min-grant", min_grant));
1458         _FDT(fdt_setprop_cell(fdt, offset, "max-latency", max_latency));
1459         return offset;
1460     } else {
1461         PCIBus *sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(dev));
1462 
1463         return spapr_dt_pci_bus(sphb, sec_bus, fdt, offset);
1464     }
1465 }
1466 
1467 /* Callback to be called during DRC release. */
1468 void spapr_phb_remove_pci_device_cb(DeviceState *dev)
1469 {
1470     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
1471 
1472     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
1473     object_unparent(OBJECT(dev));
1474 }
1475 
1476 int spapr_pci_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
1477                           void *fdt, int *fdt_start_offset, Error **errp)
1478 {
1479     HotplugHandler *plug_handler = qdev_get_hotplug_handler(drc->dev);
1480     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(plug_handler);
1481     PCIDevice *pdev = PCI_DEVICE(drc->dev);
1482 
1483     *fdt_start_offset = spapr_dt_pci_device(sphb, pdev, fdt, 0);
1484     return 0;
1485 }
1486 
1487 static void spapr_pci_bridge_plug(SpaprPhbState *phb,
1488                                   PCIBridge *bridge)
1489 {
1490     PCIBus *bus = pci_bridge_get_sec_bus(bridge);
1491 
1492     add_drcs(phb, bus);
1493 }
1494 
1495 /* Returns non-zero if the value of "chassis_nr" is already in use */
1496 static int check_chassis_nr(Object *obj, void *opaque)
1497 {
1498     int new_chassis_nr =
1499         object_property_get_uint(opaque, "chassis_nr", &error_abort);
1500     int chassis_nr =
1501         object_property_get_uint(obj, "chassis_nr", NULL);
1502 
1503     if (!object_dynamic_cast(obj, TYPE_PCI_BRIDGE)) {
1504         return 0;
1505     }
1506 
1507     /* Skip unsupported bridge types */
1508     if (!chassis_nr) {
1509         return 0;
1510     }
1511 
1512     /* Skip self */
1513     if (obj == opaque) {
1514         return 0;
1515     }
1516 
1517     return chassis_nr == new_chassis_nr;
1518 }
1519 
1520 static bool bridge_has_valid_chassis_nr(Object *bridge, Error **errp)
1521 {
1522     int chassis_nr =
1523         object_property_get_uint(bridge, "chassis_nr", NULL);
1524 
1525     /*
1526      * slotid_cap_init() already ensures that "chassis_nr" isn't null for
1527      * standard PCI bridges, so this really tells if "chassis_nr" is present
1528      * or not.
1529      */
1530     if (!chassis_nr) {
1531         error_setg(errp, "PCI Bridge lacks a \"chassis_nr\" property");
1532         error_append_hint(errp, "Try -device pci-bridge instead.\n");
1533         return false;
1534     }
1535 
1536     /* We want unique values for "chassis_nr" */
1537     if (object_child_foreach_recursive(object_get_root(), check_chassis_nr,
1538                                        bridge)) {
1539         error_setg(errp, "Bridge chassis %d already in use", chassis_nr);
1540         return false;
1541     }
1542 
1543     return true;
1544 }
1545 
1546 static void spapr_pci_pre_plug(HotplugHandler *plug_handler,
1547                                DeviceState *plugged_dev, Error **errp)
1548 {
1549     SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1550     PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1551     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(plugged_dev);
1552     SpaprDrc *drc = drc_from_dev(phb, pdev);
1553     PCIBus *bus = PCI_BUS(qdev_get_parent_bus(DEVICE(pdev)));
1554     uint32_t slotnr = PCI_SLOT(pdev->devfn);
1555 
1556     if (!phb->dr_enabled) {
1557         /* if this is a hotplug operation initiated by the user
1558          * we need to let them know it's not enabled
1559          */
1560         if (plugged_dev->hotplugged) {
1561             error_setg(errp, QERR_BUS_NO_HOTPLUG,
1562                        object_get_typename(OBJECT(phb)));
1563             return;
1564         }
1565     }
1566 
1567     if (pc->is_bridge) {
1568         if (!bridge_has_valid_chassis_nr(OBJECT(plugged_dev), errp)) {
1569             return;
1570         }
1571     }
1572 
1573     /* Following the QEMU convention used for PCIe multifunction
1574      * hotplug, we do not allow functions to be hotplugged to a
1575      * slot that already has function 0 present
1576      */
1577     if (plugged_dev->hotplugged && bus->devices[PCI_DEVFN(slotnr, 0)] &&
1578         PCI_FUNC(pdev->devfn) != 0) {
1579         error_setg(errp, "PCI: slot %d function 0 already occupied by %s,"
1580                    " additional functions can no longer be exposed to guest.",
1581                    slotnr, bus->devices[PCI_DEVFN(slotnr, 0)]->name);
1582     }
1583 
1584     if (drc && drc->dev) {
1585         error_setg(errp, "PCI: slot %d already occupied by %s", slotnr,
1586                    pci_get_function_0(PCI_DEVICE(drc->dev))->name);
1587         return;
1588     }
1589 }
1590 
1591 static void spapr_pci_plug(HotplugHandler *plug_handler,
1592                            DeviceState *plugged_dev, Error **errp)
1593 {
1594     SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1595     PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1596     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(plugged_dev);
1597     SpaprDrc *drc = drc_from_dev(phb, pdev);
1598     uint32_t slotnr = PCI_SLOT(pdev->devfn);
1599 
1600     /*
1601      * If DR is disabled we don't need to do anything in the case of
1602      * hotplug or coldplug callbacks.
1603      */
1604     if (!phb->dr_enabled) {
1605         return;
1606     }
1607 
1608     g_assert(drc);
1609 
1610     if (pc->is_bridge) {
1611         spapr_pci_bridge_plug(phb, PCI_BRIDGE(plugged_dev));
1612     }
1613 
1614     /* spapr_pci_pre_plug() already checked the DRC is attachable */
1615     spapr_drc_attach(drc, DEVICE(pdev));
1616 
1617     /* If this is function 0, signal hotplug for all the device functions.
1618      * Otherwise defer sending the hotplug event.
1619      */
1620     if (!spapr_drc_hotplugged(plugged_dev)) {
1621         spapr_drc_reset(drc);
1622     } else if (PCI_FUNC(pdev->devfn) == 0) {
1623         int i;
1624         uint8_t chassis = chassis_from_bus(pci_get_bus(pdev));
1625 
1626         for (i = 0; i < 8; i++) {
1627             SpaprDrc *func_drc;
1628             SpaprDrcClass *func_drck;
1629             SpaprDREntitySense state;
1630 
1631             func_drc = drc_from_devfn(phb, chassis, PCI_DEVFN(slotnr, i));
1632             func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1633             state = func_drck->dr_entity_sense(func_drc);
1634 
1635             if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) {
1636                 spapr_hotplug_req_add_by_index(func_drc);
1637             }
1638         }
1639     }
1640 }
1641 
1642 static void spapr_pci_bridge_unplug(SpaprPhbState *phb,
1643                                     PCIBridge *bridge)
1644 {
1645     PCIBus *bus = pci_bridge_get_sec_bus(bridge);
1646 
1647     remove_drcs(phb, bus);
1648 }
1649 
1650 static void spapr_pci_unplug(HotplugHandler *plug_handler,
1651                              DeviceState *plugged_dev, Error **errp)
1652 {
1653     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(plugged_dev);
1654     SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1655 
1656     /* some version guests do not wait for completion of a device
1657      * cleanup (generally done asynchronously by the kernel) before
1658      * signaling to QEMU that the device is safe, but instead sleep
1659      * for some 'safe' period of time. unfortunately on a busy host
1660      * this sleep isn't guaranteed to be long enough, resulting in
1661      * bad things like IRQ lines being left asserted during final
1662      * device removal. to deal with this we call reset just prior
1663      * to finalizing the device, which will put the device back into
1664      * an 'idle' state, as the device cleanup code expects.
1665      */
1666     pci_device_reset(PCI_DEVICE(plugged_dev));
1667 
1668     if (pc->is_bridge) {
1669         spapr_pci_bridge_unplug(phb, PCI_BRIDGE(plugged_dev));
1670         return;
1671     }
1672 
1673     qdev_unrealize(plugged_dev);
1674 }
1675 
1676 static void spapr_pci_unplug_request(HotplugHandler *plug_handler,
1677                                      DeviceState *plugged_dev, Error **errp)
1678 {
1679     SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1680     PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1681     SpaprDrc *drc = drc_from_dev(phb, pdev);
1682 
1683     if (!phb->dr_enabled) {
1684         error_setg(errp, QERR_BUS_NO_HOTPLUG,
1685                    object_get_typename(OBJECT(phb)));
1686         return;
1687     }
1688 
1689     g_assert(drc);
1690     g_assert(drc->dev == plugged_dev);
1691 
1692     if (!spapr_drc_unplug_requested(drc)) {
1693         PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(plugged_dev);
1694         uint32_t slotnr = PCI_SLOT(pdev->devfn);
1695         SpaprDrc *func_drc;
1696         SpaprDrcClass *func_drck;
1697         SpaprDREntitySense state;
1698         int i;
1699         uint8_t chassis = chassis_from_bus(pci_get_bus(pdev));
1700 
1701         if (pc->is_bridge) {
1702             error_setg(errp, "PCI: Hot unplug of PCI bridges not supported");
1703             return;
1704         }
1705         if (object_property_get_uint(OBJECT(pdev), "nvlink2-tgt", NULL)) {
1706             error_setg(errp, "PCI: Cannot unplug NVLink2 devices");
1707             return;
1708         }
1709 
1710         /* ensure any other present functions are pending unplug */
1711         if (PCI_FUNC(pdev->devfn) == 0) {
1712             for (i = 1; i < 8; i++) {
1713                 func_drc = drc_from_devfn(phb, chassis, PCI_DEVFN(slotnr, i));
1714                 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1715                 state = func_drck->dr_entity_sense(func_drc);
1716                 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT
1717                     && !spapr_drc_unplug_requested(func_drc)) {
1718                     /*
1719                      * Attempting to remove function 0 of a multifunction
1720                      * device will will cascade into removing all child
1721                      * functions, even if their unplug weren't requested
1722                      * beforehand.
1723                      */
1724                     spapr_drc_unplug_request(func_drc);
1725                 }
1726             }
1727         }
1728 
1729         spapr_drc_unplug_request(drc);
1730 
1731         /* if this isn't func 0, defer unplug event. otherwise signal removal
1732          * for all present functions
1733          */
1734         if (PCI_FUNC(pdev->devfn) == 0) {
1735             for (i = 7; i >= 0; i--) {
1736                 func_drc = drc_from_devfn(phb, chassis, PCI_DEVFN(slotnr, i));
1737                 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1738                 state = func_drck->dr_entity_sense(func_drc);
1739                 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) {
1740                     spapr_hotplug_req_remove_by_index(func_drc);
1741                 }
1742             }
1743         }
1744     } else {
1745         error_setg(errp,
1746                    "PCI device unplug already in progress for device %s",
1747                    drc->dev->id);
1748     }
1749 }
1750 
1751 static void spapr_phb_finalizefn(Object *obj)
1752 {
1753     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(obj);
1754 
1755     g_free(sphb->dtbusname);
1756     sphb->dtbusname = NULL;
1757 }
1758 
1759 static void spapr_phb_unrealize(DeviceState *dev)
1760 {
1761     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
1762     SysBusDevice *s = SYS_BUS_DEVICE(dev);
1763     PCIHostState *phb = PCI_HOST_BRIDGE(s);
1764     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(phb);
1765     SpaprTceTable *tcet;
1766     int i;
1767     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
1768 
1769     spapr_phb_nvgpu_free(sphb);
1770 
1771     if (sphb->msi) {
1772         g_hash_table_unref(sphb->msi);
1773         sphb->msi = NULL;
1774     }
1775 
1776     /*
1777      * Remove IO/MMIO subregions and aliases, rest should get cleaned
1778      * via PHB's unrealize->object_finalize
1779      */
1780     for (i = windows_supported - 1; i >= 0; i--) {
1781         tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[i]);
1782         if (tcet) {
1783             memory_region_del_subregion(&sphb->iommu_root,
1784                                         spapr_tce_get_iommu(tcet));
1785         }
1786     }
1787 
1788     remove_drcs(sphb, phb->bus);
1789 
1790     for (i = PCI_NUM_PINS - 1; i >= 0; i--) {
1791         if (sphb->lsi_table[i].irq) {
1792             spapr_irq_free(spapr, sphb->lsi_table[i].irq, 1);
1793             sphb->lsi_table[i].irq = 0;
1794         }
1795     }
1796 
1797     QLIST_REMOVE(sphb, list);
1798 
1799     memory_region_del_subregion(&sphb->iommu_root, &sphb->msiwindow);
1800 
1801     /*
1802      * An attached PCI device may have memory listeners, eg. VFIO PCI. We have
1803      * unmapped all sections. Remove the listeners now, before destroying the
1804      * address space.
1805      */
1806     address_space_remove_listeners(&sphb->iommu_as);
1807     address_space_destroy(&sphb->iommu_as);
1808 
1809     qbus_set_hotplug_handler(BUS(phb->bus), NULL);
1810     pci_unregister_root_bus(phb->bus);
1811 
1812     memory_region_del_subregion(get_system_memory(), &sphb->iowindow);
1813     if (sphb->mem64_win_pciaddr != (hwaddr)-1) {
1814         memory_region_del_subregion(get_system_memory(), &sphb->mem64window);
1815     }
1816     memory_region_del_subregion(get_system_memory(), &sphb->mem32window);
1817 }
1818 
1819 static void spapr_phb_destroy_msi(gpointer opaque)
1820 {
1821     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
1822     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
1823     SpaprPciMsi *msi = opaque;
1824 
1825     if (!smc->legacy_irq_allocation) {
1826         spapr_irq_msi_free(spapr, msi->first_irq, msi->num);
1827     }
1828     spapr_irq_free(spapr, msi->first_irq, msi->num);
1829     g_free(msi);
1830 }
1831 
1832 static void spapr_phb_realize(DeviceState *dev, Error **errp)
1833 {
1834     ERRP_GUARD();
1835     /* We don't use SPAPR_MACHINE() in order to exit gracefully if the user
1836      * tries to add a sPAPR PHB to a non-pseries machine.
1837      */
1838     SpaprMachineState *spapr =
1839         (SpaprMachineState *) object_dynamic_cast(qdev_get_machine(),
1840                                                   TYPE_SPAPR_MACHINE);
1841     SpaprMachineClass *smc = spapr ? SPAPR_MACHINE_GET_CLASS(spapr) : NULL;
1842     SysBusDevice *s = SYS_BUS_DEVICE(dev);
1843     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(s);
1844     PCIHostState *phb = PCI_HOST_BRIDGE(s);
1845     MachineState *ms = MACHINE(spapr);
1846     char *namebuf;
1847     int i;
1848     PCIBus *bus;
1849     uint64_t msi_window_size = 4096;
1850     SpaprTceTable *tcet;
1851     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
1852 
1853     if (!spapr) {
1854         error_setg(errp, TYPE_SPAPR_PCI_HOST_BRIDGE " needs a pseries machine");
1855         return;
1856     }
1857 
1858     assert(sphb->index != (uint32_t)-1); /* checked in spapr_phb_pre_plug() */
1859 
1860     if (sphb->mem64_win_size != 0) {
1861         if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) {
1862             error_setg(errp, "32-bit memory window of size 0x%"HWADDR_PRIx
1863                        " (max 2 GiB)", sphb->mem_win_size);
1864             return;
1865         }
1866 
1867         /* 64-bit window defaults to identity mapping */
1868         sphb->mem64_win_pciaddr = sphb->mem64_win_addr;
1869     } else if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) {
1870         /*
1871          * For compatibility with old configuration, if no 64-bit MMIO
1872          * window is specified, but the ordinary (32-bit) memory
1873          * window is specified as > 2GiB, we treat it as a 2GiB 32-bit
1874          * window, with a 64-bit MMIO window following on immediately
1875          * afterwards
1876          */
1877         sphb->mem64_win_size = sphb->mem_win_size - SPAPR_PCI_MEM32_WIN_SIZE;
1878         sphb->mem64_win_addr = sphb->mem_win_addr + SPAPR_PCI_MEM32_WIN_SIZE;
1879         sphb->mem64_win_pciaddr =
1880             SPAPR_PCI_MEM_WIN_BUS_OFFSET + SPAPR_PCI_MEM32_WIN_SIZE;
1881         sphb->mem_win_size = SPAPR_PCI_MEM32_WIN_SIZE;
1882     }
1883 
1884     if (spapr_pci_find_phb(spapr, sphb->buid)) {
1885         SpaprPhbState *s;
1886 
1887         error_setg(errp, "PCI host bridges must have unique indexes");
1888         error_append_hint(errp, "The following indexes are already in use:");
1889         QLIST_FOREACH(s, &spapr->phbs, list) {
1890             error_append_hint(errp, " %d", s->index);
1891         }
1892         error_append_hint(errp, "\nTry another value for the index property\n");
1893         return;
1894     }
1895 
1896     if (sphb->numa_node != -1 &&
1897         (sphb->numa_node >= MAX_NODES ||
1898          !ms->numa_state->nodes[sphb->numa_node].present)) {
1899         error_setg(errp, "Invalid NUMA node ID for PCI host bridge");
1900         return;
1901     }
1902 
1903     sphb->dtbusname = g_strdup_printf("pci@%" PRIx64, sphb->buid);
1904 
1905     /* Initialize memory regions */
1906     namebuf = g_strdup_printf("%s.mmio", sphb->dtbusname);
1907     memory_region_init(&sphb->memspace, OBJECT(sphb), namebuf, UINT64_MAX);
1908     g_free(namebuf);
1909 
1910     namebuf = g_strdup_printf("%s.mmio32-alias", sphb->dtbusname);
1911     memory_region_init_alias(&sphb->mem32window, OBJECT(sphb),
1912                              namebuf, &sphb->memspace,
1913                              SPAPR_PCI_MEM_WIN_BUS_OFFSET, sphb->mem_win_size);
1914     g_free(namebuf);
1915     memory_region_add_subregion(get_system_memory(), sphb->mem_win_addr,
1916                                 &sphb->mem32window);
1917 
1918     if (sphb->mem64_win_size != 0) {
1919         namebuf = g_strdup_printf("%s.mmio64-alias", sphb->dtbusname);
1920         memory_region_init_alias(&sphb->mem64window, OBJECT(sphb),
1921                                  namebuf, &sphb->memspace,
1922                                  sphb->mem64_win_pciaddr, sphb->mem64_win_size);
1923         g_free(namebuf);
1924 
1925         memory_region_add_subregion(get_system_memory(),
1926                                     sphb->mem64_win_addr,
1927                                     &sphb->mem64window);
1928     }
1929 
1930     /* Initialize IO regions */
1931     namebuf = g_strdup_printf("%s.io", sphb->dtbusname);
1932     memory_region_init(&sphb->iospace, OBJECT(sphb),
1933                        namebuf, SPAPR_PCI_IO_WIN_SIZE);
1934     g_free(namebuf);
1935 
1936     namebuf = g_strdup_printf("%s.io-alias", sphb->dtbusname);
1937     memory_region_init_alias(&sphb->iowindow, OBJECT(sphb), namebuf,
1938                              &sphb->iospace, 0, SPAPR_PCI_IO_WIN_SIZE);
1939     g_free(namebuf);
1940     memory_region_add_subregion(get_system_memory(), sphb->io_win_addr,
1941                                 &sphb->iowindow);
1942 
1943     bus = pci_register_root_bus(dev, NULL,
1944                                 pci_spapr_set_irq, pci_swizzle_map_irq_fn, sphb,
1945                                 &sphb->memspace, &sphb->iospace,
1946                                 PCI_DEVFN(0, 0), PCI_NUM_PINS,
1947                                 TYPE_PCI_BUS);
1948 
1949     /*
1950      * Despite resembling a vanilla PCI bus in most ways, the PAPR
1951      * para-virtualized PCI bus *does* permit PCI-E extended config
1952      * space access
1953      */
1954     if (sphb->pcie_ecs) {
1955         bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
1956     }
1957     phb->bus = bus;
1958     qbus_set_hotplug_handler(BUS(phb->bus), OBJECT(sphb));
1959 
1960     /*
1961      * Initialize PHB address space.
1962      * By default there will be at least one subregion for default
1963      * 32bit DMA window.
1964      * Later the guest might want to create another DMA window
1965      * which will become another memory subregion.
1966      */
1967     namebuf = g_strdup_printf("%s.iommu-root", sphb->dtbusname);
1968     memory_region_init(&sphb->iommu_root, OBJECT(sphb),
1969                        namebuf, UINT64_MAX);
1970     g_free(namebuf);
1971     address_space_init(&sphb->iommu_as, &sphb->iommu_root,
1972                        sphb->dtbusname);
1973 
1974     /*
1975      * As MSI/MSIX interrupts trigger by writing at MSI/MSIX vectors,
1976      * we need to allocate some memory to catch those writes coming
1977      * from msi_notify()/msix_notify().
1978      * As MSIMessage:addr is going to be the same and MSIMessage:data
1979      * is going to be a VIRQ number, 4 bytes of the MSI MR will only
1980      * be used.
1981      *
1982      * For KVM we want to ensure that this memory is a full page so that
1983      * our memory slot is of page size granularity.
1984      */
1985     if (kvm_enabled()) {
1986         msi_window_size = qemu_real_host_page_size;
1987     }
1988 
1989     memory_region_init_io(&sphb->msiwindow, OBJECT(sphb), &spapr_msi_ops, spapr,
1990                           "msi", msi_window_size);
1991     memory_region_add_subregion(&sphb->iommu_root, SPAPR_PCI_MSI_WINDOW,
1992                                 &sphb->msiwindow);
1993 
1994     pci_setup_iommu(bus, spapr_pci_dma_iommu, sphb);
1995 
1996     pci_bus_set_route_irq_fn(bus, spapr_route_intx_pin_to_irq);
1997 
1998     QLIST_INSERT_HEAD(&spapr->phbs, sphb, list);
1999 
2000     /* Initialize the LSI table */
2001     for (i = 0; i < PCI_NUM_PINS; i++) {
2002         int irq = SPAPR_IRQ_PCI_LSI + sphb->index * PCI_NUM_PINS + i;
2003 
2004         if (smc->legacy_irq_allocation) {
2005             irq = spapr_irq_findone(spapr, errp);
2006             if (irq < 0) {
2007                 error_prepend(errp, "can't allocate LSIs: ");
2008                 /*
2009                  * Older machines will never support PHB hotplug, ie, this is an
2010                  * init only path and QEMU will terminate. No need to rollback.
2011                  */
2012                 return;
2013             }
2014         }
2015 
2016         if (spapr_irq_claim(spapr, irq, true, errp) < 0) {
2017             error_prepend(errp, "can't allocate LSIs: ");
2018             goto unrealize;
2019         }
2020 
2021         sphb->lsi_table[i].irq = irq;
2022     }
2023 
2024     /* allocate connectors for child PCI devices */
2025     add_drcs(sphb, phb->bus);
2026 
2027     /* DMA setup */
2028     for (i = 0; i < windows_supported; ++i) {
2029         tcet = spapr_tce_new_table(DEVICE(sphb), sphb->dma_liobn[i]);
2030         if (!tcet) {
2031             error_setg(errp, "Creating window#%d failed for %s",
2032                        i, sphb->dtbusname);
2033             goto unrealize;
2034         }
2035         memory_region_add_subregion(&sphb->iommu_root, 0,
2036                                     spapr_tce_get_iommu(tcet));
2037     }
2038 
2039     sphb->msi = g_hash_table_new_full(g_int_hash, g_int_equal, g_free,
2040                                       spapr_phb_destroy_msi);
2041     return;
2042 
2043 unrealize:
2044     spapr_phb_unrealize(dev);
2045 }
2046 
2047 static int spapr_phb_children_reset(Object *child, void *opaque)
2048 {
2049     DeviceState *dev = (DeviceState *) object_dynamic_cast(child, TYPE_DEVICE);
2050 
2051     if (dev) {
2052         device_legacy_reset(dev);
2053     }
2054 
2055     return 0;
2056 }
2057 
2058 void spapr_phb_dma_reset(SpaprPhbState *sphb)
2059 {
2060     int i;
2061     SpaprTceTable *tcet;
2062 
2063     for (i = 0; i < SPAPR_PCI_DMA_MAX_WINDOWS; ++i) {
2064         tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[i]);
2065 
2066         if (tcet && tcet->nb_table) {
2067             spapr_tce_table_disable(tcet);
2068         }
2069     }
2070 
2071     /* Register default 32bit DMA window */
2072     tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[0]);
2073     spapr_tce_table_enable(tcet, SPAPR_TCE_PAGE_SHIFT, sphb->dma_win_addr,
2074                            sphb->dma_win_size >> SPAPR_TCE_PAGE_SHIFT);
2075 }
2076 
2077 static void spapr_phb_reset(DeviceState *qdev)
2078 {
2079     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(qdev);
2080     Error *err = NULL;
2081 
2082     spapr_phb_dma_reset(sphb);
2083     spapr_phb_nvgpu_free(sphb);
2084     spapr_phb_nvgpu_setup(sphb, &err);
2085     if (err) {
2086         error_report_err(err);
2087     }
2088 
2089     /* Reset the IOMMU state */
2090     object_child_foreach(OBJECT(qdev), spapr_phb_children_reset, NULL);
2091 
2092     if (spapr_phb_eeh_available(SPAPR_PCI_HOST_BRIDGE(qdev))) {
2093         spapr_phb_vfio_reset(qdev);
2094     }
2095 
2096     g_hash_table_remove_all(sphb->msi);
2097 }
2098 
2099 static Property spapr_phb_properties[] = {
2100     DEFINE_PROP_UINT32("index", SpaprPhbState, index, -1),
2101     DEFINE_PROP_UINT64("mem_win_size", SpaprPhbState, mem_win_size,
2102                        SPAPR_PCI_MEM32_WIN_SIZE),
2103     DEFINE_PROP_UINT64("mem64_win_size", SpaprPhbState, mem64_win_size,
2104                        SPAPR_PCI_MEM64_WIN_SIZE),
2105     DEFINE_PROP_UINT64("io_win_size", SpaprPhbState, io_win_size,
2106                        SPAPR_PCI_IO_WIN_SIZE),
2107     DEFINE_PROP_BOOL("dynamic-reconfiguration", SpaprPhbState, dr_enabled,
2108                      true),
2109     /* Default DMA window is 0..1GB */
2110     DEFINE_PROP_UINT64("dma_win_addr", SpaprPhbState, dma_win_addr, 0),
2111     DEFINE_PROP_UINT64("dma_win_size", SpaprPhbState, dma_win_size, 0x40000000),
2112     DEFINE_PROP_UINT64("dma64_win_addr", SpaprPhbState, dma64_win_addr,
2113                        0x800000000000000ULL),
2114     DEFINE_PROP_BOOL("ddw", SpaprPhbState, ddw_enabled, true),
2115     DEFINE_PROP_UINT64("pgsz", SpaprPhbState, page_size_mask,
2116                        (1ULL << 12) | (1ULL << 16)
2117                        | (1ULL << 21) | (1ULL << 24)),
2118     DEFINE_PROP_UINT32("numa_node", SpaprPhbState, numa_node, -1),
2119     DEFINE_PROP_BOOL("pre-2.8-migration", SpaprPhbState,
2120                      pre_2_8_migration, false),
2121     DEFINE_PROP_BOOL("pcie-extended-configuration-space", SpaprPhbState,
2122                      pcie_ecs, true),
2123     DEFINE_PROP_UINT64("gpa", SpaprPhbState, nv2_gpa_win_addr, 0),
2124     DEFINE_PROP_UINT64("atsd", SpaprPhbState, nv2_atsd_win_addr, 0),
2125     DEFINE_PROP_BOOL("pre-5.1-associativity", SpaprPhbState,
2126                      pre_5_1_assoc, false),
2127     DEFINE_PROP_END_OF_LIST(),
2128 };
2129 
2130 static const VMStateDescription vmstate_spapr_pci_lsi = {
2131     .name = "spapr_pci/lsi",
2132     .version_id = 1,
2133     .minimum_version_id = 1,
2134     .fields = (VMStateField[]) {
2135         VMSTATE_UINT32_EQUAL(irq, SpaprPciLsi, NULL),
2136 
2137         VMSTATE_END_OF_LIST()
2138     },
2139 };
2140 
2141 static const VMStateDescription vmstate_spapr_pci_msi = {
2142     .name = "spapr_pci/msi",
2143     .version_id = 1,
2144     .minimum_version_id = 1,
2145     .fields = (VMStateField []) {
2146         VMSTATE_UINT32(key, SpaprPciMsiMig),
2147         VMSTATE_UINT32(value.first_irq, SpaprPciMsiMig),
2148         VMSTATE_UINT32(value.num, SpaprPciMsiMig),
2149         VMSTATE_END_OF_LIST()
2150     },
2151 };
2152 
2153 static int spapr_pci_pre_save(void *opaque)
2154 {
2155     SpaprPhbState *sphb = opaque;
2156     GHashTableIter iter;
2157     gpointer key, value;
2158     int i;
2159 
2160     if (sphb->pre_2_8_migration) {
2161         sphb->mig_liobn = sphb->dma_liobn[0];
2162         sphb->mig_mem_win_addr = sphb->mem_win_addr;
2163         sphb->mig_mem_win_size = sphb->mem_win_size;
2164         sphb->mig_io_win_addr = sphb->io_win_addr;
2165         sphb->mig_io_win_size = sphb->io_win_size;
2166 
2167         if ((sphb->mem64_win_size != 0)
2168             && (sphb->mem64_win_addr
2169                 == (sphb->mem_win_addr + sphb->mem_win_size))) {
2170             sphb->mig_mem_win_size += sphb->mem64_win_size;
2171         }
2172     }
2173 
2174     g_free(sphb->msi_devs);
2175     sphb->msi_devs = NULL;
2176     sphb->msi_devs_num = g_hash_table_size(sphb->msi);
2177     if (!sphb->msi_devs_num) {
2178         return 0;
2179     }
2180     sphb->msi_devs = g_new(SpaprPciMsiMig, sphb->msi_devs_num);
2181 
2182     g_hash_table_iter_init(&iter, sphb->msi);
2183     for (i = 0; g_hash_table_iter_next(&iter, &key, &value); ++i) {
2184         sphb->msi_devs[i].key = *(uint32_t *) key;
2185         sphb->msi_devs[i].value = *(SpaprPciMsi *) value;
2186     }
2187 
2188     return 0;
2189 }
2190 
2191 static int spapr_pci_post_save(void *opaque)
2192 {
2193     SpaprPhbState *sphb = opaque;
2194 
2195     g_free(sphb->msi_devs);
2196     sphb->msi_devs = NULL;
2197     sphb->msi_devs_num = 0;
2198     return 0;
2199 }
2200 
2201 static int spapr_pci_post_load(void *opaque, int version_id)
2202 {
2203     SpaprPhbState *sphb = opaque;
2204     gpointer key, value;
2205     int i;
2206 
2207     for (i = 0; i < sphb->msi_devs_num; ++i) {
2208         key = g_memdup(&sphb->msi_devs[i].key,
2209                        sizeof(sphb->msi_devs[i].key));
2210         value = g_memdup(&sphb->msi_devs[i].value,
2211                          sizeof(sphb->msi_devs[i].value));
2212         g_hash_table_insert(sphb->msi, key, value);
2213     }
2214     g_free(sphb->msi_devs);
2215     sphb->msi_devs = NULL;
2216     sphb->msi_devs_num = 0;
2217 
2218     return 0;
2219 }
2220 
2221 static bool pre_2_8_migration(void *opaque, int version_id)
2222 {
2223     SpaprPhbState *sphb = opaque;
2224 
2225     return sphb->pre_2_8_migration;
2226 }
2227 
2228 static const VMStateDescription vmstate_spapr_pci = {
2229     .name = "spapr_pci",
2230     .version_id = 2,
2231     .minimum_version_id = 2,
2232     .pre_save = spapr_pci_pre_save,
2233     .post_save = spapr_pci_post_save,
2234     .post_load = spapr_pci_post_load,
2235     .fields = (VMStateField[]) {
2236         VMSTATE_UINT64_EQUAL(buid, SpaprPhbState, NULL),
2237         VMSTATE_UINT32_TEST(mig_liobn, SpaprPhbState, pre_2_8_migration),
2238         VMSTATE_UINT64_TEST(mig_mem_win_addr, SpaprPhbState, pre_2_8_migration),
2239         VMSTATE_UINT64_TEST(mig_mem_win_size, SpaprPhbState, pre_2_8_migration),
2240         VMSTATE_UINT64_TEST(mig_io_win_addr, SpaprPhbState, pre_2_8_migration),
2241         VMSTATE_UINT64_TEST(mig_io_win_size, SpaprPhbState, pre_2_8_migration),
2242         VMSTATE_STRUCT_ARRAY(lsi_table, SpaprPhbState, PCI_NUM_PINS, 0,
2243                              vmstate_spapr_pci_lsi, SpaprPciLsi),
2244         VMSTATE_INT32(msi_devs_num, SpaprPhbState),
2245         VMSTATE_STRUCT_VARRAY_ALLOC(msi_devs, SpaprPhbState, msi_devs_num, 0,
2246                                     vmstate_spapr_pci_msi, SpaprPciMsiMig),
2247         VMSTATE_END_OF_LIST()
2248     },
2249 };
2250 
2251 static const char *spapr_phb_root_bus_path(PCIHostState *host_bridge,
2252                                            PCIBus *rootbus)
2253 {
2254     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(host_bridge);
2255 
2256     return sphb->dtbusname;
2257 }
2258 
2259 static void spapr_phb_class_init(ObjectClass *klass, void *data)
2260 {
2261     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(klass);
2262     DeviceClass *dc = DEVICE_CLASS(klass);
2263     HotplugHandlerClass *hp = HOTPLUG_HANDLER_CLASS(klass);
2264 
2265     hc->root_bus_path = spapr_phb_root_bus_path;
2266     dc->realize = spapr_phb_realize;
2267     dc->unrealize = spapr_phb_unrealize;
2268     device_class_set_props(dc, spapr_phb_properties);
2269     dc->reset = spapr_phb_reset;
2270     dc->vmsd = &vmstate_spapr_pci;
2271     /* Supported by TYPE_SPAPR_MACHINE */
2272     dc->user_creatable = true;
2273     set_bit(DEVICE_CATEGORY_BRIDGE, dc->categories);
2274     hp->pre_plug = spapr_pci_pre_plug;
2275     hp->plug = spapr_pci_plug;
2276     hp->unplug = spapr_pci_unplug;
2277     hp->unplug_request = spapr_pci_unplug_request;
2278 }
2279 
2280 static const TypeInfo spapr_phb_info = {
2281     .name          = TYPE_SPAPR_PCI_HOST_BRIDGE,
2282     .parent        = TYPE_PCI_HOST_BRIDGE,
2283     .instance_size = sizeof(SpaprPhbState),
2284     .instance_finalize = spapr_phb_finalizefn,
2285     .class_init    = spapr_phb_class_init,
2286     .interfaces    = (InterfaceInfo[]) {
2287         { TYPE_HOTPLUG_HANDLER },
2288         { }
2289     }
2290 };
2291 
2292 static void spapr_phb_pci_enumerate_bridge(PCIBus *bus, PCIDevice *pdev,
2293                                            void *opaque)
2294 {
2295     unsigned int *bus_no = opaque;
2296     PCIBus *sec_bus = NULL;
2297 
2298     if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) !=
2299          PCI_HEADER_TYPE_BRIDGE)) {
2300         return;
2301     }
2302 
2303     (*bus_no)++;
2304     pci_default_write_config(pdev, PCI_PRIMARY_BUS, pci_dev_bus_num(pdev), 1);
2305     pci_default_write_config(pdev, PCI_SECONDARY_BUS, *bus_no, 1);
2306     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
2307 
2308     sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev));
2309     if (!sec_bus) {
2310         return;
2311     }
2312 
2313     pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
2314                         spapr_phb_pci_enumerate_bridge, bus_no);
2315     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
2316 }
2317 
2318 static void spapr_phb_pci_enumerate(SpaprPhbState *phb)
2319 {
2320     PCIBus *bus = PCI_HOST_BRIDGE(phb)->bus;
2321     unsigned int bus_no = 0;
2322 
2323     pci_for_each_device(bus, pci_bus_num(bus),
2324                         spapr_phb_pci_enumerate_bridge,
2325                         &bus_no);
2326 
2327 }
2328 
2329 int spapr_dt_phb(SpaprMachineState *spapr, SpaprPhbState *phb,
2330                  uint32_t intc_phandle, void *fdt, int *node_offset)
2331 {
2332     int bus_off, i, j, ret;
2333     uint32_t bus_range[] = { cpu_to_be32(0), cpu_to_be32(0xff) };
2334     struct {
2335         uint32_t hi;
2336         uint64_t child;
2337         uint64_t parent;
2338         uint64_t size;
2339     } QEMU_PACKED ranges[] = {
2340         {
2341             cpu_to_be32(b_ss(1)), cpu_to_be64(0),
2342             cpu_to_be64(phb->io_win_addr),
2343             cpu_to_be64(memory_region_size(&phb->iospace)),
2344         },
2345         {
2346             cpu_to_be32(b_ss(2)), cpu_to_be64(SPAPR_PCI_MEM_WIN_BUS_OFFSET),
2347             cpu_to_be64(phb->mem_win_addr),
2348             cpu_to_be64(phb->mem_win_size),
2349         },
2350         {
2351             cpu_to_be32(b_ss(3)), cpu_to_be64(phb->mem64_win_pciaddr),
2352             cpu_to_be64(phb->mem64_win_addr),
2353             cpu_to_be64(phb->mem64_win_size),
2354         },
2355     };
2356     const unsigned sizeof_ranges =
2357         (phb->mem64_win_size ? 3 : 2) * sizeof(ranges[0]);
2358     uint64_t bus_reg[] = { cpu_to_be64(phb->buid), 0 };
2359     uint32_t interrupt_map_mask[] = {
2360         cpu_to_be32(b_ddddd(-1)|b_fff(0)), 0x0, 0x0, cpu_to_be32(-1)};
2361     uint32_t interrupt_map[PCI_SLOT_MAX * PCI_NUM_PINS][7];
2362     uint32_t ddw_applicable[] = {
2363         cpu_to_be32(RTAS_IBM_QUERY_PE_DMA_WINDOW),
2364         cpu_to_be32(RTAS_IBM_CREATE_PE_DMA_WINDOW),
2365         cpu_to_be32(RTAS_IBM_REMOVE_PE_DMA_WINDOW)
2366     };
2367     uint32_t ddw_extensions[] = {
2368         cpu_to_be32(1),
2369         cpu_to_be32(RTAS_IBM_RESET_PE_DMA_WINDOW)
2370     };
2371     SpaprTceTable *tcet;
2372     SpaprDrc *drc;
2373     Error *err = NULL;
2374 
2375     /* Start populating the FDT */
2376     _FDT(bus_off = fdt_add_subnode(fdt, 0, phb->dtbusname));
2377     if (node_offset) {
2378         *node_offset = bus_off;
2379     }
2380 
2381     /* Write PHB properties */
2382     _FDT(fdt_setprop_string(fdt, bus_off, "device_type", "pci"));
2383     _FDT(fdt_setprop_string(fdt, bus_off, "compatible", "IBM,Logical_PHB"));
2384     _FDT(fdt_setprop_cell(fdt, bus_off, "#interrupt-cells", 0x1));
2385     _FDT(fdt_setprop(fdt, bus_off, "used-by-rtas", NULL, 0));
2386     _FDT(fdt_setprop(fdt, bus_off, "bus-range", &bus_range, sizeof(bus_range)));
2387     _FDT(fdt_setprop(fdt, bus_off, "ranges", &ranges, sizeof_ranges));
2388     _FDT(fdt_setprop(fdt, bus_off, "reg", &bus_reg, sizeof(bus_reg)));
2389     _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pci-config-space-type", 0x1));
2390     _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pe-total-#msi",
2391                           spapr_irq_nr_msis(spapr)));
2392 
2393     /* Dynamic DMA window */
2394     if (phb->ddw_enabled) {
2395         _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-applicable", &ddw_applicable,
2396                          sizeof(ddw_applicable)));
2397         _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-extensions",
2398                          &ddw_extensions, sizeof(ddw_extensions)));
2399     }
2400 
2401     /* Advertise NUMA via ibm,associativity */
2402     if (phb->numa_node != -1) {
2403         spapr_numa_write_associativity_dt(spapr, fdt, bus_off, phb->numa_node);
2404     }
2405 
2406     /* Build the interrupt-map, this must matches what is done
2407      * in pci_swizzle_map_irq_fn
2408      */
2409     _FDT(fdt_setprop(fdt, bus_off, "interrupt-map-mask",
2410                      &interrupt_map_mask, sizeof(interrupt_map_mask)));
2411     for (i = 0; i < PCI_SLOT_MAX; i++) {
2412         for (j = 0; j < PCI_NUM_PINS; j++) {
2413             uint32_t *irqmap = interrupt_map[i*PCI_NUM_PINS + j];
2414             int lsi_num = pci_swizzle(i, j);
2415 
2416             irqmap[0] = cpu_to_be32(b_ddddd(i)|b_fff(0));
2417             irqmap[1] = 0;
2418             irqmap[2] = 0;
2419             irqmap[3] = cpu_to_be32(j+1);
2420             irqmap[4] = cpu_to_be32(intc_phandle);
2421             spapr_dt_irq(&irqmap[5], phb->lsi_table[lsi_num].irq, true);
2422         }
2423     }
2424     /* Write interrupt map */
2425     _FDT(fdt_setprop(fdt, bus_off, "interrupt-map", &interrupt_map,
2426                      sizeof(interrupt_map)));
2427 
2428     tcet = spapr_tce_find_by_liobn(phb->dma_liobn[0]);
2429     if (!tcet) {
2430         return -1;
2431     }
2432     spapr_dma_dt(fdt, bus_off, "ibm,dma-window",
2433                  tcet->liobn, tcet->bus_offset,
2434                  tcet->nb_table << tcet->page_shift);
2435 
2436     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, phb->index);
2437     if (drc) {
2438         uint32_t drc_index = cpu_to_be32(spapr_drc_index(drc));
2439 
2440         _FDT(fdt_setprop(fdt, bus_off, "ibm,my-drc-index", &drc_index,
2441                          sizeof(drc_index)));
2442     }
2443 
2444     /* Walk the bridges and program the bus numbers*/
2445     spapr_phb_pci_enumerate(phb);
2446     _FDT(fdt_setprop_cell(fdt, bus_off, "qemu,phb-enumerated", 0x1));
2447 
2448     /* Walk the bridge and subordinate buses */
2449     ret = spapr_dt_pci_bus(phb, PCI_HOST_BRIDGE(phb)->bus, fdt, bus_off);
2450     if (ret < 0) {
2451         return ret;
2452     }
2453 
2454     spapr_phb_nvgpu_populate_dt(phb, fdt, bus_off, &err);
2455     if (err) {
2456         error_report_err(err);
2457     }
2458     spapr_phb_nvgpu_ram_populate_dt(phb, fdt);
2459 
2460     return 0;
2461 }
2462 
2463 void spapr_pci_rtas_init(void)
2464 {
2465     spapr_rtas_register(RTAS_READ_PCI_CONFIG, "read-pci-config",
2466                         rtas_read_pci_config);
2467     spapr_rtas_register(RTAS_WRITE_PCI_CONFIG, "write-pci-config",
2468                         rtas_write_pci_config);
2469     spapr_rtas_register(RTAS_IBM_READ_PCI_CONFIG, "ibm,read-pci-config",
2470                         rtas_ibm_read_pci_config);
2471     spapr_rtas_register(RTAS_IBM_WRITE_PCI_CONFIG, "ibm,write-pci-config",
2472                         rtas_ibm_write_pci_config);
2473     if (msi_nonbroken) {
2474         spapr_rtas_register(RTAS_IBM_QUERY_INTERRUPT_SOURCE_NUMBER,
2475                             "ibm,query-interrupt-source-number",
2476                             rtas_ibm_query_interrupt_source_number);
2477         spapr_rtas_register(RTAS_IBM_CHANGE_MSI, "ibm,change-msi",
2478                             rtas_ibm_change_msi);
2479     }
2480 
2481     spapr_rtas_register(RTAS_IBM_SET_EEH_OPTION,
2482                         "ibm,set-eeh-option",
2483                         rtas_ibm_set_eeh_option);
2484     spapr_rtas_register(RTAS_IBM_GET_CONFIG_ADDR_INFO2,
2485                         "ibm,get-config-addr-info2",
2486                         rtas_ibm_get_config_addr_info2);
2487     spapr_rtas_register(RTAS_IBM_READ_SLOT_RESET_STATE2,
2488                         "ibm,read-slot-reset-state2",
2489                         rtas_ibm_read_slot_reset_state2);
2490     spapr_rtas_register(RTAS_IBM_SET_SLOT_RESET,
2491                         "ibm,set-slot-reset",
2492                         rtas_ibm_set_slot_reset);
2493     spapr_rtas_register(RTAS_IBM_CONFIGURE_PE,
2494                         "ibm,configure-pe",
2495                         rtas_ibm_configure_pe);
2496     spapr_rtas_register(RTAS_IBM_SLOT_ERROR_DETAIL,
2497                         "ibm,slot-error-detail",
2498                         rtas_ibm_slot_error_detail);
2499 }
2500 
2501 static void spapr_pci_register_types(void)
2502 {
2503     type_register_static(&spapr_phb_info);
2504 }
2505 
2506 type_init(spapr_pci_register_types)
2507 
2508 static int spapr_switch_one_vga(DeviceState *dev, void *opaque)
2509 {
2510     bool be = *(bool *)opaque;
2511 
2512     if (object_dynamic_cast(OBJECT(dev), "VGA")
2513         || object_dynamic_cast(OBJECT(dev), "secondary-vga")
2514         || object_dynamic_cast(OBJECT(dev), "bochs-display")
2515         || object_dynamic_cast(OBJECT(dev), "virtio-vga")) {
2516         object_property_set_bool(OBJECT(dev), "big-endian-framebuffer", be,
2517                                  &error_abort);
2518     }
2519     return 0;
2520 }
2521 
2522 void spapr_pci_switch_vga(SpaprMachineState *spapr, bool big_endian)
2523 {
2524     SpaprPhbState *sphb;
2525 
2526     /*
2527      * For backward compatibility with existing guests, we switch
2528      * the endianness of the VGA controller when changing the guest
2529      * interrupt mode
2530      */
2531     QLIST_FOREACH(sphb, &spapr->phbs, list) {
2532         BusState *bus = &PCI_HOST_BRIDGE(sphb)->bus->qbus;
2533         qbus_walk_children(bus, spapr_switch_one_vga, NULL, NULL, NULL,
2534                            &big_endian);
2535     }
2536 }
2537