xref: /openbmc/qemu/hw/ppc/spapr_pci.c (revision 3d558330)
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     g_autofree char *path = NULL;
786     g_autofree char *host = NULL;
787     g_autofree char *devspec = NULL;
788     char *buf = NULL;
789 
790     /* Get the PCI VFIO host id */
791     host = object_property_get_str(OBJECT(pdev), "host", NULL);
792     if (!host) {
793         return NULL;
794     }
795 
796     /* Construct the path of the file that will give us the DT location */
797     path = g_strdup_printf("/sys/bus/pci/devices/%s/devspec", host);
798     if (!g_file_get_contents(path, &devspec, NULL, NULL)) {
799         return NULL;
800     }
801 
802     /* Construct and read from host device tree the loc-code */
803     g_free(path);
804     path = g_strdup_printf("/proc/device-tree%s/ibm,loc-code", devspec);
805     if (!g_file_get_contents(path, &buf, NULL, NULL)) {
806         return NULL;
807     }
808     return buf;
809 }
810 
811 static char *spapr_phb_get_loc_code(SpaprPhbState *sphb, PCIDevice *pdev)
812 {
813     char *buf;
814     const char *devtype = "qemu";
815     uint32_t busnr = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(pdev))));
816 
817     if (object_dynamic_cast(OBJECT(pdev), "vfio-pci")) {
818         buf = spapr_phb_vfio_get_loc_code(sphb, pdev);
819         if (buf) {
820             return buf;
821         }
822         devtype = "vfio";
823     }
824     /*
825      * For emulated devices and VFIO-failure case, make up
826      * the loc-code.
827      */
828     buf = g_strdup_printf("%s_%s:%04x:%02x:%02x.%x",
829                           devtype, pdev->name, sphb->index, busnr,
830                           PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
831     return buf;
832 }
833 
834 /* Macros to operate with address in OF binding to PCI */
835 #define b_x(x, p, l)    (((x) & ((1<<(l))-1)) << (p))
836 #define b_n(x)          b_x((x), 31, 1) /* 0 if relocatable */
837 #define b_p(x)          b_x((x), 30, 1) /* 1 if prefetchable */
838 #define b_t(x)          b_x((x), 29, 1) /* 1 if the address is aliased */
839 #define b_ss(x)         b_x((x), 24, 2) /* the space code */
840 #define b_bbbbbbbb(x)   b_x((x), 16, 8) /* bus number */
841 #define b_ddddd(x)      b_x((x), 11, 5) /* device number */
842 #define b_fff(x)        b_x((x), 8, 3)  /* function number */
843 #define b_rrrrrrrr(x)   b_x((x), 0, 8)  /* register number */
844 
845 /* for 'reg' OF properties */
846 #define RESOURCE_CELLS_SIZE 2
847 #define RESOURCE_CELLS_ADDRESS 3
848 
849 typedef struct ResourceFields {
850     uint32_t phys_hi;
851     uint32_t phys_mid;
852     uint32_t phys_lo;
853     uint32_t size_hi;
854     uint32_t size_lo;
855 } QEMU_PACKED ResourceFields;
856 
857 typedef struct ResourceProps {
858     ResourceFields reg[8];
859     uint32_t reg_len;
860 } ResourceProps;
861 
862 /* fill in the 'reg' OF properties for
863  * a PCI device. 'reg' describes resource requirements for a
864  * device's IO/MEM regions.
865  *
866  * the property is an array of ('phys-addr', 'size') pairs describing
867  * the addressable regions of the PCI device, where 'phys-addr' is a
868  * RESOURCE_CELLS_ADDRESS-tuple of 32-bit integers corresponding to
869  * (phys.hi, phys.mid, phys.lo), and 'size' is a
870  * RESOURCE_CELLS_SIZE-tuple corresponding to (size.hi, size.lo).
871  *
872  * phys.hi = 0xYYXXXXZZ, where:
873  *   0xYY = npt000ss
874  *          |||   |
875  *          |||   +-- space code
876  *          |||               |
877  *          |||               +  00 if configuration space
878  *          |||               +  01 if IO region,
879  *          |||               +  10 if 32-bit MEM region
880  *          |||               +  11 if 64-bit MEM region
881  *          |||
882  *          ||+------ for non-relocatable IO: 1 if aliased
883  *          ||        for relocatable IO: 1 if below 64KB
884  *          ||        for MEM: 1 if below 1MB
885  *          |+------- 1 if region is prefetchable
886  *          +-------- 1 if region is non-relocatable
887  *   0xXXXX = bbbbbbbb dddddfff, encoding bus, slot, and function
888  *            bits respectively
889  *   0xZZ = rrrrrrrr, the register number of the BAR corresponding
890  *          to the region
891  *
892  * phys.mid and phys.lo correspond respectively to the hi/lo portions
893  * of the actual address of the region.
894  *
895  * note also that addresses defined in this property are, at least
896  * for PAPR guests, relative to the PHBs IO/MEM windows, and
897  * correspond directly to the addresses in the BARs.
898  *
899  * in accordance with PCI Bus Binding to Open Firmware,
900  * IEEE Std 1275-1994, section 4.1.1, as implemented by PAPR+ v2.7,
901  * Appendix C.
902  */
903 static void populate_resource_props(PCIDevice *d, ResourceProps *rp)
904 {
905     int bus_num = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(d))));
906     uint32_t dev_id = (b_bbbbbbbb(bus_num) |
907                        b_ddddd(PCI_SLOT(d->devfn)) |
908                        b_fff(PCI_FUNC(d->devfn)));
909     ResourceFields *reg;
910     int i, reg_idx = 0;
911 
912     /* config space region */
913     reg = &rp->reg[reg_idx++];
914     reg->phys_hi = cpu_to_be32(dev_id);
915     reg->phys_mid = 0;
916     reg->phys_lo = 0;
917     reg->size_hi = 0;
918     reg->size_lo = 0;
919 
920     for (i = 0; i < PCI_NUM_REGIONS; i++) {
921         if (!d->io_regions[i].size) {
922             continue;
923         }
924 
925         reg = &rp->reg[reg_idx++];
926 
927         reg->phys_hi = cpu_to_be32(dev_id | b_rrrrrrrr(pci_bar(d, i)));
928         if (d->io_regions[i].type & PCI_BASE_ADDRESS_SPACE_IO) {
929             reg->phys_hi |= cpu_to_be32(b_ss(1));
930         } else if (d->io_regions[i].type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
931             reg->phys_hi |= cpu_to_be32(b_ss(3));
932         } else {
933             reg->phys_hi |= cpu_to_be32(b_ss(2));
934         }
935         reg->phys_mid = 0;
936         reg->phys_lo = 0;
937         reg->size_hi = cpu_to_be32(d->io_regions[i].size >> 32);
938         reg->size_lo = cpu_to_be32(d->io_regions[i].size);
939     }
940 
941     rp->reg_len = reg_idx * sizeof(ResourceFields);
942 }
943 
944 typedef struct PCIClass PCIClass;
945 typedef struct PCISubClass PCISubClass;
946 typedef struct PCIIFace PCIIFace;
947 
948 struct PCIIFace {
949     int iface;
950     const char *name;
951 };
952 
953 struct PCISubClass {
954     int subclass;
955     const char *name;
956     const PCIIFace *iface;
957 };
958 
959 struct PCIClass {
960     const char *name;
961     const PCISubClass *subc;
962 };
963 
964 static const PCISubClass undef_subclass[] = {
965     { PCI_CLASS_NOT_DEFINED_VGA, "display", NULL },
966     { 0xFF, NULL, NULL },
967 };
968 
969 static const PCISubClass mass_subclass[] = {
970     { PCI_CLASS_STORAGE_SCSI, "scsi", NULL },
971     { PCI_CLASS_STORAGE_IDE, "ide", NULL },
972     { PCI_CLASS_STORAGE_FLOPPY, "fdc", NULL },
973     { PCI_CLASS_STORAGE_IPI, "ipi", NULL },
974     { PCI_CLASS_STORAGE_RAID, "raid", NULL },
975     { PCI_CLASS_STORAGE_ATA, "ata", NULL },
976     { PCI_CLASS_STORAGE_SATA, "sata", NULL },
977     { PCI_CLASS_STORAGE_SAS, "sas", NULL },
978     { 0xFF, NULL, NULL },
979 };
980 
981 static const PCISubClass net_subclass[] = {
982     { PCI_CLASS_NETWORK_ETHERNET, "ethernet", NULL },
983     { PCI_CLASS_NETWORK_TOKEN_RING, "token-ring", NULL },
984     { PCI_CLASS_NETWORK_FDDI, "fddi", NULL },
985     { PCI_CLASS_NETWORK_ATM, "atm", NULL },
986     { PCI_CLASS_NETWORK_ISDN, "isdn", NULL },
987     { PCI_CLASS_NETWORK_WORLDFIP, "worldfip", NULL },
988     { PCI_CLASS_NETWORK_PICMG214, "picmg", NULL },
989     { 0xFF, NULL, NULL },
990 };
991 
992 static const PCISubClass displ_subclass[] = {
993     { PCI_CLASS_DISPLAY_VGA, "vga", NULL },
994     { PCI_CLASS_DISPLAY_XGA, "xga", NULL },
995     { PCI_CLASS_DISPLAY_3D, "3d-controller", NULL },
996     { 0xFF, NULL, NULL },
997 };
998 
999 static const PCISubClass media_subclass[] = {
1000     { PCI_CLASS_MULTIMEDIA_VIDEO, "video", NULL },
1001     { PCI_CLASS_MULTIMEDIA_AUDIO, "sound", NULL },
1002     { PCI_CLASS_MULTIMEDIA_PHONE, "telephony", NULL },
1003     { 0xFF, NULL, NULL },
1004 };
1005 
1006 static const PCISubClass mem_subclass[] = {
1007     { PCI_CLASS_MEMORY_RAM, "memory", NULL },
1008     { PCI_CLASS_MEMORY_FLASH, "flash", NULL },
1009     { 0xFF, NULL, NULL },
1010 };
1011 
1012 static const PCISubClass bridg_subclass[] = {
1013     { PCI_CLASS_BRIDGE_HOST, "host", NULL },
1014     { PCI_CLASS_BRIDGE_ISA, "isa", NULL },
1015     { PCI_CLASS_BRIDGE_EISA, "eisa", NULL },
1016     { PCI_CLASS_BRIDGE_MC, "mca", NULL },
1017     { PCI_CLASS_BRIDGE_PCI, "pci", NULL },
1018     { PCI_CLASS_BRIDGE_PCMCIA, "pcmcia", NULL },
1019     { PCI_CLASS_BRIDGE_NUBUS, "nubus", NULL },
1020     { PCI_CLASS_BRIDGE_CARDBUS, "cardbus", NULL },
1021     { PCI_CLASS_BRIDGE_RACEWAY, "raceway", NULL },
1022     { PCI_CLASS_BRIDGE_PCI_SEMITP, "semi-transparent-pci", NULL },
1023     { PCI_CLASS_BRIDGE_IB_PCI, "infiniband", NULL },
1024     { 0xFF, NULL, NULL },
1025 };
1026 
1027 static const PCISubClass comm_subclass[] = {
1028     { PCI_CLASS_COMMUNICATION_SERIAL, "serial", NULL },
1029     { PCI_CLASS_COMMUNICATION_PARALLEL, "parallel", NULL },
1030     { PCI_CLASS_COMMUNICATION_MULTISERIAL, "multiport-serial", NULL },
1031     { PCI_CLASS_COMMUNICATION_MODEM, "modem", NULL },
1032     { PCI_CLASS_COMMUNICATION_GPIB, "gpib", NULL },
1033     { PCI_CLASS_COMMUNICATION_SC, "smart-card", NULL },
1034     { 0xFF, NULL, NULL, },
1035 };
1036 
1037 static const PCIIFace pic_iface[] = {
1038     { PCI_CLASS_SYSTEM_PIC_IOAPIC, "io-apic" },
1039     { PCI_CLASS_SYSTEM_PIC_IOXAPIC, "io-xapic" },
1040     { 0xFF, NULL },
1041 };
1042 
1043 static const PCISubClass sys_subclass[] = {
1044     { PCI_CLASS_SYSTEM_PIC, "interrupt-controller", pic_iface },
1045     { PCI_CLASS_SYSTEM_DMA, "dma-controller", NULL },
1046     { PCI_CLASS_SYSTEM_TIMER, "timer", NULL },
1047     { PCI_CLASS_SYSTEM_RTC, "rtc", NULL },
1048     { PCI_CLASS_SYSTEM_PCI_HOTPLUG, "hot-plug-controller", NULL },
1049     { PCI_CLASS_SYSTEM_SDHCI, "sd-host-controller", NULL },
1050     { 0xFF, NULL, NULL },
1051 };
1052 
1053 static const PCISubClass inp_subclass[] = {
1054     { PCI_CLASS_INPUT_KEYBOARD, "keyboard", NULL },
1055     { PCI_CLASS_INPUT_PEN, "pen", NULL },
1056     { PCI_CLASS_INPUT_MOUSE, "mouse", NULL },
1057     { PCI_CLASS_INPUT_SCANNER, "scanner", NULL },
1058     { PCI_CLASS_INPUT_GAMEPORT, "gameport", NULL },
1059     { 0xFF, NULL, NULL },
1060 };
1061 
1062 static const PCISubClass dock_subclass[] = {
1063     { PCI_CLASS_DOCKING_GENERIC, "dock", NULL },
1064     { 0xFF, NULL, NULL },
1065 };
1066 
1067 static const PCISubClass cpu_subclass[] = {
1068     { PCI_CLASS_PROCESSOR_PENTIUM, "pentium", NULL },
1069     { PCI_CLASS_PROCESSOR_POWERPC, "powerpc", NULL },
1070     { PCI_CLASS_PROCESSOR_MIPS, "mips", NULL },
1071     { PCI_CLASS_PROCESSOR_CO, "co-processor", NULL },
1072     { 0xFF, NULL, NULL },
1073 };
1074 
1075 static const PCIIFace usb_iface[] = {
1076     { PCI_CLASS_SERIAL_USB_UHCI, "usb-uhci" },
1077     { PCI_CLASS_SERIAL_USB_OHCI, "usb-ohci", },
1078     { PCI_CLASS_SERIAL_USB_EHCI, "usb-ehci" },
1079     { PCI_CLASS_SERIAL_USB_XHCI, "usb-xhci" },
1080     { PCI_CLASS_SERIAL_USB_UNKNOWN, "usb-unknown" },
1081     { PCI_CLASS_SERIAL_USB_DEVICE, "usb-device" },
1082     { 0xFF, NULL },
1083 };
1084 
1085 static const PCISubClass ser_subclass[] = {
1086     { PCI_CLASS_SERIAL_FIREWIRE, "firewire", NULL },
1087     { PCI_CLASS_SERIAL_ACCESS, "access-bus", NULL },
1088     { PCI_CLASS_SERIAL_SSA, "ssa", NULL },
1089     { PCI_CLASS_SERIAL_USB, "usb", usb_iface },
1090     { PCI_CLASS_SERIAL_FIBER, "fibre-channel", NULL },
1091     { PCI_CLASS_SERIAL_SMBUS, "smb", NULL },
1092     { PCI_CLASS_SERIAL_IB, "infiniband", NULL },
1093     { PCI_CLASS_SERIAL_IPMI, "ipmi", NULL },
1094     { PCI_CLASS_SERIAL_SERCOS, "sercos", NULL },
1095     { PCI_CLASS_SERIAL_CANBUS, "canbus", NULL },
1096     { 0xFF, NULL, NULL },
1097 };
1098 
1099 static const PCISubClass wrl_subclass[] = {
1100     { PCI_CLASS_WIRELESS_IRDA, "irda", NULL },
1101     { PCI_CLASS_WIRELESS_CIR, "consumer-ir", NULL },
1102     { PCI_CLASS_WIRELESS_RF_CONTROLLER, "rf-controller", NULL },
1103     { PCI_CLASS_WIRELESS_BLUETOOTH, "bluetooth", NULL },
1104     { PCI_CLASS_WIRELESS_BROADBAND, "broadband", NULL },
1105     { 0xFF, NULL, NULL },
1106 };
1107 
1108 static const PCISubClass sat_subclass[] = {
1109     { PCI_CLASS_SATELLITE_TV, "satellite-tv", NULL },
1110     { PCI_CLASS_SATELLITE_AUDIO, "satellite-audio", NULL },
1111     { PCI_CLASS_SATELLITE_VOICE, "satellite-voice", NULL },
1112     { PCI_CLASS_SATELLITE_DATA, "satellite-data", NULL },
1113     { 0xFF, NULL, NULL },
1114 };
1115 
1116 static const PCISubClass crypt_subclass[] = {
1117     { PCI_CLASS_CRYPT_NETWORK, "network-encryption", NULL },
1118     { PCI_CLASS_CRYPT_ENTERTAINMENT,
1119       "entertainment-encryption", NULL },
1120     { 0xFF, NULL, NULL },
1121 };
1122 
1123 static const PCISubClass spc_subclass[] = {
1124     { PCI_CLASS_SP_DPIO, "dpio", NULL },
1125     { PCI_CLASS_SP_PERF, "counter", NULL },
1126     { PCI_CLASS_SP_SYNCH, "measurement", NULL },
1127     { PCI_CLASS_SP_MANAGEMENT, "management-card", NULL },
1128     { 0xFF, NULL, NULL },
1129 };
1130 
1131 static const PCIClass pci_classes[] = {
1132     { "legacy-device", undef_subclass },
1133     { "mass-storage",  mass_subclass },
1134     { "network", net_subclass },
1135     { "display", displ_subclass, },
1136     { "multimedia-device", media_subclass },
1137     { "memory-controller", mem_subclass },
1138     { "unknown-bridge", bridg_subclass },
1139     { "communication-controller", comm_subclass},
1140     { "system-peripheral", sys_subclass },
1141     { "input-controller", inp_subclass },
1142     { "docking-station", dock_subclass },
1143     { "cpu", cpu_subclass },
1144     { "serial-bus", ser_subclass },
1145     { "wireless-controller", wrl_subclass },
1146     { "intelligent-io", NULL },
1147     { "satellite-device", sat_subclass },
1148     { "encryption", crypt_subclass },
1149     { "data-processing-controller", spc_subclass },
1150 };
1151 
1152 static const char *dt_name_from_class(uint8_t class, uint8_t subclass,
1153                                       uint8_t iface)
1154 {
1155     const PCIClass *pclass;
1156     const PCISubClass *psubclass;
1157     const PCIIFace *piface;
1158     const char *name;
1159 
1160     if (class >= ARRAY_SIZE(pci_classes)) {
1161         return "pci";
1162     }
1163 
1164     pclass = pci_classes + class;
1165     name = pclass->name;
1166 
1167     if (pclass->subc == NULL) {
1168         return name;
1169     }
1170 
1171     psubclass = pclass->subc;
1172     while ((psubclass->subclass & 0xff) != 0xff) {
1173         if ((psubclass->subclass & 0xff) == subclass) {
1174             name = psubclass->name;
1175             break;
1176         }
1177         psubclass++;
1178     }
1179 
1180     piface = psubclass->iface;
1181     if (piface == NULL) {
1182         return name;
1183     }
1184     while ((piface->iface & 0xff) != 0xff) {
1185         if ((piface->iface & 0xff) == iface) {
1186             name = piface->name;
1187             break;
1188         }
1189         piface++;
1190     }
1191 
1192     return name;
1193 }
1194 
1195 /*
1196  * DRC helper functions
1197  */
1198 
1199 static uint32_t drc_id_from_devfn(SpaprPhbState *phb,
1200                                   uint8_t chassis, int32_t devfn)
1201 {
1202     return (phb->index << 16) | (chassis << 8) | devfn;
1203 }
1204 
1205 static SpaprDrc *drc_from_devfn(SpaprPhbState *phb,
1206                                 uint8_t chassis, int32_t devfn)
1207 {
1208     return spapr_drc_by_id(TYPE_SPAPR_DRC_PCI,
1209                            drc_id_from_devfn(phb, chassis, devfn));
1210 }
1211 
1212 static uint8_t chassis_from_bus(PCIBus *bus)
1213 {
1214     if (pci_bus_is_root(bus)) {
1215         return 0;
1216     } else {
1217         PCIDevice *bridge = pci_bridge_get_device(bus);
1218 
1219         return object_property_get_uint(OBJECT(bridge), "chassis_nr",
1220                                         &error_abort);
1221     }
1222 }
1223 
1224 static SpaprDrc *drc_from_dev(SpaprPhbState *phb, PCIDevice *dev)
1225 {
1226     uint8_t chassis = chassis_from_bus(pci_get_bus(dev));
1227 
1228     return drc_from_devfn(phb, chassis, dev->devfn);
1229 }
1230 
1231 static void add_drcs(SpaprPhbState *phb, PCIBus *bus)
1232 {
1233     Object *owner;
1234     int i;
1235     uint8_t chassis;
1236 
1237     if (!phb->dr_enabled) {
1238         return;
1239     }
1240 
1241     chassis = chassis_from_bus(bus);
1242 
1243     if (pci_bus_is_root(bus)) {
1244         owner = OBJECT(phb);
1245     } else {
1246         owner = OBJECT(pci_bridge_get_device(bus));
1247     }
1248 
1249     for (i = 0; i < PCI_SLOT_MAX * PCI_FUNC_MAX; i++) {
1250         spapr_dr_connector_new(owner, TYPE_SPAPR_DRC_PCI,
1251                                drc_id_from_devfn(phb, chassis, i));
1252     }
1253 }
1254 
1255 static void remove_drcs(SpaprPhbState *phb, PCIBus *bus)
1256 {
1257     int i;
1258     uint8_t chassis;
1259 
1260     if (!phb->dr_enabled) {
1261         return;
1262     }
1263 
1264     chassis = chassis_from_bus(bus);
1265 
1266     for (i = PCI_SLOT_MAX * PCI_FUNC_MAX - 1; i >= 0; i--) {
1267         SpaprDrc *drc = drc_from_devfn(phb, chassis, i);
1268 
1269         if (drc) {
1270             object_unparent(OBJECT(drc));
1271         }
1272     }
1273 }
1274 
1275 typedef struct PciWalkFdt {
1276     void *fdt;
1277     int offset;
1278     SpaprPhbState *sphb;
1279     int err;
1280 } PciWalkFdt;
1281 
1282 static int spapr_dt_pci_device(SpaprPhbState *sphb, PCIDevice *dev,
1283                                void *fdt, int parent_offset);
1284 
1285 static void spapr_dt_pci_device_cb(PCIBus *bus, PCIDevice *pdev,
1286                                    void *opaque)
1287 {
1288     PciWalkFdt *p = opaque;
1289     int err;
1290 
1291     if (p->err) {
1292         /* Something's already broken, don't keep going */
1293         return;
1294     }
1295 
1296     err = spapr_dt_pci_device(p->sphb, pdev, p->fdt, p->offset);
1297     if (err < 0) {
1298         p->err = err;
1299     }
1300 }
1301 
1302 /* Augment PCI device node with bridge specific information */
1303 static int spapr_dt_pci_bus(SpaprPhbState *sphb, PCIBus *bus,
1304                                void *fdt, int offset)
1305 {
1306     Object *owner;
1307     PciWalkFdt cbinfo = {
1308         .fdt = fdt,
1309         .offset = offset,
1310         .sphb = sphb,
1311         .err = 0,
1312     };
1313     int ret;
1314 
1315     _FDT(fdt_setprop_cell(fdt, offset, "#address-cells",
1316                           RESOURCE_CELLS_ADDRESS));
1317     _FDT(fdt_setprop_cell(fdt, offset, "#size-cells",
1318                           RESOURCE_CELLS_SIZE));
1319 
1320     assert(bus);
1321     pci_for_each_device_under_bus_reverse(bus, spapr_dt_pci_device_cb, &cbinfo);
1322     if (cbinfo.err) {
1323         return cbinfo.err;
1324     }
1325 
1326     if (pci_bus_is_root(bus)) {
1327         owner = OBJECT(sphb);
1328     } else {
1329         owner = OBJECT(pci_bridge_get_device(bus));
1330     }
1331 
1332     ret = spapr_dt_drc(fdt, offset, owner,
1333                        SPAPR_DR_CONNECTOR_TYPE_PCI);
1334     if (ret) {
1335         return ret;
1336     }
1337 
1338     return offset;
1339 }
1340 
1341 char *spapr_pci_fw_dev_name(PCIDevice *dev)
1342 {
1343     const gchar *basename;
1344     int slot = PCI_SLOT(dev->devfn);
1345     int func = PCI_FUNC(dev->devfn);
1346     uint32_t ccode = pci_default_read_config(dev, PCI_CLASS_PROG, 3);
1347 
1348     basename = dt_name_from_class((ccode >> 16) & 0xff, (ccode >> 8) & 0xff,
1349                                   ccode & 0xff);
1350 
1351     if (func != 0) {
1352         return g_strdup_printf("%s@%x,%x", basename, slot, func);
1353     } else {
1354         return g_strdup_printf("%s@%x", basename, slot);
1355     }
1356 }
1357 
1358 /* create OF node for pci device and required OF DT properties */
1359 static int spapr_dt_pci_device(SpaprPhbState *sphb, PCIDevice *dev,
1360                                void *fdt, int parent_offset)
1361 {
1362     int offset;
1363     g_autofree gchar *nodename = spapr_pci_fw_dev_name(dev);
1364     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
1365     ResourceProps rp;
1366     SpaprDrc *drc = drc_from_dev(sphb, dev);
1367     uint32_t vendor_id = pci_default_read_config(dev, PCI_VENDOR_ID, 2);
1368     uint32_t device_id = pci_default_read_config(dev, PCI_DEVICE_ID, 2);
1369     uint32_t revision_id = pci_default_read_config(dev, PCI_REVISION_ID, 1);
1370     uint32_t ccode = pci_default_read_config(dev, PCI_CLASS_PROG, 3);
1371     uint32_t irq_pin = pci_default_read_config(dev, PCI_INTERRUPT_PIN, 1);
1372     uint32_t subsystem_id = pci_default_read_config(dev, PCI_SUBSYSTEM_ID, 2);
1373     uint32_t subsystem_vendor_id =
1374         pci_default_read_config(dev, PCI_SUBSYSTEM_VENDOR_ID, 2);
1375     uint32_t cache_line_size =
1376         pci_default_read_config(dev, PCI_CACHE_LINE_SIZE, 1);
1377     uint32_t pci_status = pci_default_read_config(dev, PCI_STATUS, 2);
1378     gchar *loc_code;
1379 
1380     _FDT(offset = fdt_add_subnode(fdt, parent_offset, nodename));
1381 
1382     /* in accordance with PAPR+ v2.7 13.6.3, Table 181 */
1383     _FDT(fdt_setprop_cell(fdt, offset, "vendor-id", vendor_id));
1384     _FDT(fdt_setprop_cell(fdt, offset, "device-id", device_id));
1385     _FDT(fdt_setprop_cell(fdt, offset, "revision-id", revision_id));
1386 
1387     _FDT(fdt_setprop_cell(fdt, offset, "class-code", ccode));
1388     if (irq_pin) {
1389         _FDT(fdt_setprop_cell(fdt, offset, "interrupts", irq_pin));
1390     }
1391 
1392     if (subsystem_id) {
1393         _FDT(fdt_setprop_cell(fdt, offset, "subsystem-id", subsystem_id));
1394     }
1395 
1396     if (subsystem_vendor_id) {
1397         _FDT(fdt_setprop_cell(fdt, offset, "subsystem-vendor-id",
1398                               subsystem_vendor_id));
1399     }
1400 
1401     _FDT(fdt_setprop_cell(fdt, offset, "cache-line-size", cache_line_size));
1402 
1403 
1404     /* the following fdt cells are masked off the pci status register */
1405     _FDT(fdt_setprop_cell(fdt, offset, "devsel-speed",
1406                           PCI_STATUS_DEVSEL_MASK & pci_status));
1407 
1408     if (pci_status & PCI_STATUS_FAST_BACK) {
1409         _FDT(fdt_setprop(fdt, offset, "fast-back-to-back", NULL, 0));
1410     }
1411     if (pci_status & PCI_STATUS_66MHZ) {
1412         _FDT(fdt_setprop(fdt, offset, "66mhz-capable", NULL, 0));
1413     }
1414     if (pci_status & PCI_STATUS_UDF) {
1415         _FDT(fdt_setprop(fdt, offset, "udf-supported", NULL, 0));
1416     }
1417 
1418     loc_code = spapr_phb_get_loc_code(sphb, dev);
1419     _FDT(fdt_setprop_string(fdt, offset, "ibm,loc-code", loc_code));
1420     g_free(loc_code);
1421 
1422     if (drc) {
1423         _FDT(fdt_setprop_cell(fdt, offset, "ibm,my-drc-index",
1424                               spapr_drc_index(drc)));
1425     }
1426 
1427     if (msi_present(dev)) {
1428         uint32_t max_msi = msi_nr_vectors_allocated(dev);
1429         if (max_msi) {
1430             _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi", max_msi));
1431         }
1432     }
1433     if (msix_present(dev)) {
1434         uint32_t max_msix = dev->msix_entries_nr;
1435         if (max_msix) {
1436             _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi-x", max_msix));
1437         }
1438     }
1439 
1440     populate_resource_props(dev, &rp);
1441     _FDT(fdt_setprop(fdt, offset, "reg", (uint8_t *)rp.reg, rp.reg_len));
1442 
1443     if (sphb->pcie_ecs && pci_is_express(dev)) {
1444         _FDT(fdt_setprop_cell(fdt, offset, "ibm,pci-config-space-type", 0x1));
1445     }
1446 
1447     spapr_phb_nvgpu_populate_pcidev_dt(dev, fdt, offset, sphb);
1448 
1449     if (!pc->is_bridge) {
1450         /* Properties only for non-bridges */
1451         uint32_t min_grant = pci_default_read_config(dev, PCI_MIN_GNT, 1);
1452         uint32_t max_latency = pci_default_read_config(dev, PCI_MAX_LAT, 1);
1453         _FDT(fdt_setprop_cell(fdt, offset, "min-grant", min_grant));
1454         _FDT(fdt_setprop_cell(fdt, offset, "max-latency", max_latency));
1455         return offset;
1456     } else {
1457         PCIBus *sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(dev));
1458 
1459         return spapr_dt_pci_bus(sphb, sec_bus, fdt, offset);
1460     }
1461 }
1462 
1463 /* Callback to be called during DRC release. */
1464 void spapr_phb_remove_pci_device_cb(DeviceState *dev)
1465 {
1466     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
1467 
1468     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
1469     object_unparent(OBJECT(dev));
1470 }
1471 
1472 int spapr_pci_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
1473                           void *fdt, int *fdt_start_offset, Error **errp)
1474 {
1475     HotplugHandler *plug_handler = qdev_get_hotplug_handler(drc->dev);
1476     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(plug_handler);
1477     PCIDevice *pdev = PCI_DEVICE(drc->dev);
1478 
1479     *fdt_start_offset = spapr_dt_pci_device(sphb, pdev, fdt, 0);
1480     return 0;
1481 }
1482 
1483 static void spapr_pci_bridge_plug(SpaprPhbState *phb,
1484                                   PCIBridge *bridge)
1485 {
1486     PCIBus *bus = pci_bridge_get_sec_bus(bridge);
1487 
1488     add_drcs(phb, bus);
1489 }
1490 
1491 /* Returns non-zero if the value of "chassis_nr" is already in use */
1492 static int check_chassis_nr(Object *obj, void *opaque)
1493 {
1494     int new_chassis_nr =
1495         object_property_get_uint(opaque, "chassis_nr", &error_abort);
1496     int chassis_nr =
1497         object_property_get_uint(obj, "chassis_nr", NULL);
1498 
1499     if (!object_dynamic_cast(obj, TYPE_PCI_BRIDGE)) {
1500         return 0;
1501     }
1502 
1503     /* Skip unsupported bridge types */
1504     if (!chassis_nr) {
1505         return 0;
1506     }
1507 
1508     /* Skip self */
1509     if (obj == opaque) {
1510         return 0;
1511     }
1512 
1513     return chassis_nr == new_chassis_nr;
1514 }
1515 
1516 static bool bridge_has_valid_chassis_nr(Object *bridge, Error **errp)
1517 {
1518     int chassis_nr =
1519         object_property_get_uint(bridge, "chassis_nr", NULL);
1520 
1521     /*
1522      * slotid_cap_init() already ensures that "chassis_nr" isn't null for
1523      * standard PCI bridges, so this really tells if "chassis_nr" is present
1524      * or not.
1525      */
1526     if (!chassis_nr) {
1527         error_setg(errp, "PCI Bridge lacks a \"chassis_nr\" property");
1528         error_append_hint(errp, "Try -device pci-bridge instead.\n");
1529         return false;
1530     }
1531 
1532     /* We want unique values for "chassis_nr" */
1533     if (object_child_foreach_recursive(object_get_root(), check_chassis_nr,
1534                                        bridge)) {
1535         error_setg(errp, "Bridge chassis %d already in use", chassis_nr);
1536         return false;
1537     }
1538 
1539     return true;
1540 }
1541 
1542 static void spapr_pci_pre_plug(HotplugHandler *plug_handler,
1543                                DeviceState *plugged_dev, Error **errp)
1544 {
1545     SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1546     PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1547     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(plugged_dev);
1548     SpaprDrc *drc = drc_from_dev(phb, pdev);
1549     PCIBus *bus = PCI_BUS(qdev_get_parent_bus(DEVICE(pdev)));
1550     uint32_t slotnr = PCI_SLOT(pdev->devfn);
1551 
1552     if (!phb->dr_enabled) {
1553         /* if this is a hotplug operation initiated by the user
1554          * we need to let them know it's not enabled
1555          */
1556         if (plugged_dev->hotplugged) {
1557             error_setg(errp, QERR_BUS_NO_HOTPLUG,
1558                        object_get_typename(OBJECT(phb)));
1559             return;
1560         }
1561     }
1562 
1563     if (pc->is_bridge) {
1564         if (!bridge_has_valid_chassis_nr(OBJECT(plugged_dev), errp)) {
1565             return;
1566         }
1567     }
1568 
1569     /* Following the QEMU convention used for PCIe multifunction
1570      * hotplug, we do not allow functions to be hotplugged to a
1571      * slot that already has function 0 present
1572      */
1573     if (plugged_dev->hotplugged && bus->devices[PCI_DEVFN(slotnr, 0)] &&
1574         PCI_FUNC(pdev->devfn) != 0) {
1575         error_setg(errp, "PCI: slot %d function 0 already occupied by %s,"
1576                    " additional functions can no longer be exposed to guest.",
1577                    slotnr, bus->devices[PCI_DEVFN(slotnr, 0)]->name);
1578     }
1579 
1580     if (drc && drc->dev) {
1581         error_setg(errp, "PCI: slot %d already occupied by %s", slotnr,
1582                    pci_get_function_0(PCI_DEVICE(drc->dev))->name);
1583         return;
1584     }
1585 }
1586 
1587 static void spapr_pci_plug(HotplugHandler *plug_handler,
1588                            DeviceState *plugged_dev, Error **errp)
1589 {
1590     SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1591     PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1592     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(plugged_dev);
1593     SpaprDrc *drc = drc_from_dev(phb, pdev);
1594     uint32_t slotnr = PCI_SLOT(pdev->devfn);
1595 
1596     /*
1597      * If DR is disabled we don't need to do anything in the case of
1598      * hotplug or coldplug callbacks.
1599      */
1600     if (!phb->dr_enabled) {
1601         return;
1602     }
1603 
1604     g_assert(drc);
1605 
1606     if (pc->is_bridge) {
1607         spapr_pci_bridge_plug(phb, PCI_BRIDGE(plugged_dev));
1608     }
1609 
1610     /* spapr_pci_pre_plug() already checked the DRC is attachable */
1611     spapr_drc_attach(drc, DEVICE(pdev));
1612 
1613     /* If this is function 0, signal hotplug for all the device functions.
1614      * Otherwise defer sending the hotplug event.
1615      */
1616     if (!spapr_drc_hotplugged(plugged_dev)) {
1617         spapr_drc_reset(drc);
1618     } else if (PCI_FUNC(pdev->devfn) == 0) {
1619         int i;
1620         uint8_t chassis = chassis_from_bus(pci_get_bus(pdev));
1621 
1622         for (i = 0; i < 8; i++) {
1623             SpaprDrc *func_drc;
1624             SpaprDrcClass *func_drck;
1625             SpaprDREntitySense state;
1626 
1627             func_drc = drc_from_devfn(phb, chassis, PCI_DEVFN(slotnr, i));
1628             func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1629             state = func_drck->dr_entity_sense(func_drc);
1630 
1631             if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) {
1632                 spapr_hotplug_req_add_by_index(func_drc);
1633             }
1634         }
1635     }
1636 }
1637 
1638 static void spapr_pci_bridge_unplug(SpaprPhbState *phb,
1639                                     PCIBridge *bridge)
1640 {
1641     PCIBus *bus = pci_bridge_get_sec_bus(bridge);
1642 
1643     remove_drcs(phb, bus);
1644 }
1645 
1646 static void spapr_pci_unplug(HotplugHandler *plug_handler,
1647                              DeviceState *plugged_dev, Error **errp)
1648 {
1649     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(plugged_dev);
1650     SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1651 
1652     /* some version guests do not wait for completion of a device
1653      * cleanup (generally done asynchronously by the kernel) before
1654      * signaling to QEMU that the device is safe, but instead sleep
1655      * for some 'safe' period of time. unfortunately on a busy host
1656      * this sleep isn't guaranteed to be long enough, resulting in
1657      * bad things like IRQ lines being left asserted during final
1658      * device removal. to deal with this we call reset just prior
1659      * to finalizing the device, which will put the device back into
1660      * an 'idle' state, as the device cleanup code expects.
1661      */
1662     pci_device_reset(PCI_DEVICE(plugged_dev));
1663 
1664     if (pc->is_bridge) {
1665         spapr_pci_bridge_unplug(phb, PCI_BRIDGE(plugged_dev));
1666         return;
1667     }
1668 
1669     qdev_unrealize(plugged_dev);
1670 }
1671 
1672 static void spapr_pci_unplug_request(HotplugHandler *plug_handler,
1673                                      DeviceState *plugged_dev, Error **errp)
1674 {
1675     SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1676     PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1677     SpaprDrc *drc = drc_from_dev(phb, pdev);
1678 
1679     if (!phb->dr_enabled) {
1680         error_setg(errp, QERR_BUS_NO_HOTPLUG,
1681                    object_get_typename(OBJECT(phb)));
1682         return;
1683     }
1684 
1685     g_assert(drc);
1686     g_assert(drc->dev == plugged_dev);
1687 
1688     if (!spapr_drc_unplug_requested(drc)) {
1689         PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(plugged_dev);
1690         uint32_t slotnr = PCI_SLOT(pdev->devfn);
1691         SpaprDrc *func_drc;
1692         SpaprDrcClass *func_drck;
1693         SpaprDREntitySense state;
1694         int i;
1695         uint8_t chassis = chassis_from_bus(pci_get_bus(pdev));
1696 
1697         if (pc->is_bridge) {
1698             error_setg(errp, "PCI: Hot unplug of PCI bridges not supported");
1699             return;
1700         }
1701         if (object_property_get_uint(OBJECT(pdev), "nvlink2-tgt", NULL)) {
1702             error_setg(errp, "PCI: Cannot unplug NVLink2 devices");
1703             return;
1704         }
1705 
1706         /* ensure any other present functions are pending unplug */
1707         if (PCI_FUNC(pdev->devfn) == 0) {
1708             for (i = 1; i < 8; i++) {
1709                 func_drc = drc_from_devfn(phb, chassis, PCI_DEVFN(slotnr, i));
1710                 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1711                 state = func_drck->dr_entity_sense(func_drc);
1712                 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT
1713                     && !spapr_drc_unplug_requested(func_drc)) {
1714                     /*
1715                      * Attempting to remove function 0 of a multifunction
1716                      * device will will cascade into removing all child
1717                      * functions, even if their unplug weren't requested
1718                      * beforehand.
1719                      */
1720                     spapr_drc_unplug_request(func_drc);
1721                 }
1722             }
1723         }
1724 
1725         spapr_drc_unplug_request(drc);
1726 
1727         /* if this isn't func 0, defer unplug event. otherwise signal removal
1728          * for all present functions
1729          */
1730         if (PCI_FUNC(pdev->devfn) == 0) {
1731             for (i = 7; i >= 0; i--) {
1732                 func_drc = drc_from_devfn(phb, chassis, PCI_DEVFN(slotnr, i));
1733                 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1734                 state = func_drck->dr_entity_sense(func_drc);
1735                 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) {
1736                     spapr_hotplug_req_remove_by_index(func_drc);
1737                 }
1738             }
1739         }
1740     } else {
1741         error_setg(errp,
1742                    "PCI device unplug already in progress for device %s",
1743                    drc->dev->id);
1744     }
1745 }
1746 
1747 static void spapr_phb_finalizefn(Object *obj)
1748 {
1749     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(obj);
1750 
1751     g_free(sphb->dtbusname);
1752     sphb->dtbusname = NULL;
1753 }
1754 
1755 static void spapr_phb_unrealize(DeviceState *dev)
1756 {
1757     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
1758     SysBusDevice *s = SYS_BUS_DEVICE(dev);
1759     PCIHostState *phb = PCI_HOST_BRIDGE(s);
1760     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(phb);
1761     SpaprTceTable *tcet;
1762     int i;
1763     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
1764 
1765     spapr_phb_nvgpu_free(sphb);
1766 
1767     if (sphb->msi) {
1768         g_hash_table_unref(sphb->msi);
1769         sphb->msi = NULL;
1770     }
1771 
1772     /*
1773      * Remove IO/MMIO subregions and aliases, rest should get cleaned
1774      * via PHB's unrealize->object_finalize
1775      */
1776     for (i = windows_supported - 1; i >= 0; i--) {
1777         tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[i]);
1778         if (tcet) {
1779             memory_region_del_subregion(&sphb->iommu_root,
1780                                         spapr_tce_get_iommu(tcet));
1781         }
1782     }
1783 
1784     remove_drcs(sphb, phb->bus);
1785 
1786     for (i = PCI_NUM_PINS - 1; i >= 0; i--) {
1787         if (sphb->lsi_table[i].irq) {
1788             spapr_irq_free(spapr, sphb->lsi_table[i].irq, 1);
1789             sphb->lsi_table[i].irq = 0;
1790         }
1791     }
1792 
1793     QLIST_REMOVE(sphb, list);
1794 
1795     memory_region_del_subregion(&sphb->iommu_root, &sphb->msiwindow);
1796 
1797     /*
1798      * An attached PCI device may have memory listeners, eg. VFIO PCI. We have
1799      * unmapped all sections. Remove the listeners now, before destroying the
1800      * address space.
1801      */
1802     address_space_remove_listeners(&sphb->iommu_as);
1803     address_space_destroy(&sphb->iommu_as);
1804 
1805     qbus_set_hotplug_handler(BUS(phb->bus), NULL);
1806     pci_unregister_root_bus(phb->bus);
1807 
1808     memory_region_del_subregion(get_system_memory(), &sphb->iowindow);
1809     if (sphb->mem64_win_pciaddr != (hwaddr)-1) {
1810         memory_region_del_subregion(get_system_memory(), &sphb->mem64window);
1811     }
1812     memory_region_del_subregion(get_system_memory(), &sphb->mem32window);
1813 }
1814 
1815 static void spapr_phb_destroy_msi(gpointer opaque)
1816 {
1817     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
1818     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
1819     SpaprPciMsi *msi = opaque;
1820 
1821     if (!smc->legacy_irq_allocation) {
1822         spapr_irq_msi_free(spapr, msi->first_irq, msi->num);
1823     }
1824     spapr_irq_free(spapr, msi->first_irq, msi->num);
1825     g_free(msi);
1826 }
1827 
1828 static void spapr_phb_realize(DeviceState *dev, Error **errp)
1829 {
1830     ERRP_GUARD();
1831     /* We don't use SPAPR_MACHINE() in order to exit gracefully if the user
1832      * tries to add a sPAPR PHB to a non-pseries machine.
1833      */
1834     SpaprMachineState *spapr =
1835         (SpaprMachineState *) object_dynamic_cast(qdev_get_machine(),
1836                                                   TYPE_SPAPR_MACHINE);
1837     SpaprMachineClass *smc = spapr ? SPAPR_MACHINE_GET_CLASS(spapr) : NULL;
1838     SysBusDevice *s = SYS_BUS_DEVICE(dev);
1839     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(s);
1840     PCIHostState *phb = PCI_HOST_BRIDGE(s);
1841     MachineState *ms = MACHINE(spapr);
1842     char *namebuf;
1843     int i;
1844     PCIBus *bus;
1845     uint64_t msi_window_size = 4096;
1846     SpaprTceTable *tcet;
1847     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
1848 
1849     if (!spapr) {
1850         error_setg(errp, TYPE_SPAPR_PCI_HOST_BRIDGE " needs a pseries machine");
1851         return;
1852     }
1853 
1854     assert(sphb->index != (uint32_t)-1); /* checked in spapr_phb_pre_plug() */
1855 
1856     if (sphb->mem64_win_size != 0) {
1857         if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) {
1858             error_setg(errp, "32-bit memory window of size 0x%"HWADDR_PRIx
1859                        " (max 2 GiB)", sphb->mem_win_size);
1860             return;
1861         }
1862 
1863         /* 64-bit window defaults to identity mapping */
1864         sphb->mem64_win_pciaddr = sphb->mem64_win_addr;
1865     } else if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) {
1866         /*
1867          * For compatibility with old configuration, if no 64-bit MMIO
1868          * window is specified, but the ordinary (32-bit) memory
1869          * window is specified as > 2GiB, we treat it as a 2GiB 32-bit
1870          * window, with a 64-bit MMIO window following on immediately
1871          * afterwards
1872          */
1873         sphb->mem64_win_size = sphb->mem_win_size - SPAPR_PCI_MEM32_WIN_SIZE;
1874         sphb->mem64_win_addr = sphb->mem_win_addr + SPAPR_PCI_MEM32_WIN_SIZE;
1875         sphb->mem64_win_pciaddr =
1876             SPAPR_PCI_MEM_WIN_BUS_OFFSET + SPAPR_PCI_MEM32_WIN_SIZE;
1877         sphb->mem_win_size = SPAPR_PCI_MEM32_WIN_SIZE;
1878     }
1879 
1880     if (spapr_pci_find_phb(spapr, sphb->buid)) {
1881         SpaprPhbState *s;
1882 
1883         error_setg(errp, "PCI host bridges must have unique indexes");
1884         error_append_hint(errp, "The following indexes are already in use:");
1885         QLIST_FOREACH(s, &spapr->phbs, list) {
1886             error_append_hint(errp, " %d", s->index);
1887         }
1888         error_append_hint(errp, "\nTry another value for the index property\n");
1889         return;
1890     }
1891 
1892     if (sphb->numa_node != -1 &&
1893         (sphb->numa_node >= MAX_NODES ||
1894          !ms->numa_state->nodes[sphb->numa_node].present)) {
1895         error_setg(errp, "Invalid NUMA node ID for PCI host bridge");
1896         return;
1897     }
1898 
1899     sphb->dtbusname = g_strdup_printf("pci@%" PRIx64, sphb->buid);
1900 
1901     /* Initialize memory regions */
1902     namebuf = g_strdup_printf("%s.mmio", sphb->dtbusname);
1903     memory_region_init(&sphb->memspace, OBJECT(sphb), namebuf, UINT64_MAX);
1904     g_free(namebuf);
1905 
1906     namebuf = g_strdup_printf("%s.mmio32-alias", sphb->dtbusname);
1907     memory_region_init_alias(&sphb->mem32window, OBJECT(sphb),
1908                              namebuf, &sphb->memspace,
1909                              SPAPR_PCI_MEM_WIN_BUS_OFFSET, sphb->mem_win_size);
1910     g_free(namebuf);
1911     memory_region_add_subregion(get_system_memory(), sphb->mem_win_addr,
1912                                 &sphb->mem32window);
1913 
1914     if (sphb->mem64_win_size != 0) {
1915         namebuf = g_strdup_printf("%s.mmio64-alias", sphb->dtbusname);
1916         memory_region_init_alias(&sphb->mem64window, OBJECT(sphb),
1917                                  namebuf, &sphb->memspace,
1918                                  sphb->mem64_win_pciaddr, sphb->mem64_win_size);
1919         g_free(namebuf);
1920 
1921         memory_region_add_subregion(get_system_memory(),
1922                                     sphb->mem64_win_addr,
1923                                     &sphb->mem64window);
1924     }
1925 
1926     /* Initialize IO regions */
1927     namebuf = g_strdup_printf("%s.io", sphb->dtbusname);
1928     memory_region_init(&sphb->iospace, OBJECT(sphb),
1929                        namebuf, SPAPR_PCI_IO_WIN_SIZE);
1930     g_free(namebuf);
1931 
1932     namebuf = g_strdup_printf("%s.io-alias", sphb->dtbusname);
1933     memory_region_init_alias(&sphb->iowindow, OBJECT(sphb), namebuf,
1934                              &sphb->iospace, 0, SPAPR_PCI_IO_WIN_SIZE);
1935     g_free(namebuf);
1936     memory_region_add_subregion(get_system_memory(), sphb->io_win_addr,
1937                                 &sphb->iowindow);
1938 
1939     bus = pci_register_root_bus(dev, NULL,
1940                                 pci_spapr_set_irq, pci_swizzle_map_irq_fn, sphb,
1941                                 &sphb->memspace, &sphb->iospace,
1942                                 PCI_DEVFN(0, 0), PCI_NUM_PINS,
1943                                 TYPE_PCI_BUS);
1944 
1945     /*
1946      * Despite resembling a vanilla PCI bus in most ways, the PAPR
1947      * para-virtualized PCI bus *does* permit PCI-E extended config
1948      * space access
1949      */
1950     if (sphb->pcie_ecs) {
1951         bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
1952     }
1953     phb->bus = bus;
1954     qbus_set_hotplug_handler(BUS(phb->bus), OBJECT(sphb));
1955 
1956     /*
1957      * Initialize PHB address space.
1958      * By default there will be at least one subregion for default
1959      * 32bit DMA window.
1960      * Later the guest might want to create another DMA window
1961      * which will become another memory subregion.
1962      */
1963     namebuf = g_strdup_printf("%s.iommu-root", sphb->dtbusname);
1964     memory_region_init(&sphb->iommu_root, OBJECT(sphb),
1965                        namebuf, UINT64_MAX);
1966     g_free(namebuf);
1967     address_space_init(&sphb->iommu_as, &sphb->iommu_root,
1968                        sphb->dtbusname);
1969 
1970     /*
1971      * As MSI/MSIX interrupts trigger by writing at MSI/MSIX vectors,
1972      * we need to allocate some memory to catch those writes coming
1973      * from msi_notify()/msix_notify().
1974      * As MSIMessage:addr is going to be the same and MSIMessage:data
1975      * is going to be a VIRQ number, 4 bytes of the MSI MR will only
1976      * be used.
1977      *
1978      * For KVM we want to ensure that this memory is a full page so that
1979      * our memory slot is of page size granularity.
1980      */
1981     if (kvm_enabled()) {
1982         msi_window_size = qemu_real_host_page_size();
1983     }
1984 
1985     memory_region_init_io(&sphb->msiwindow, OBJECT(sphb), &spapr_msi_ops, spapr,
1986                           "msi", msi_window_size);
1987     memory_region_add_subregion(&sphb->iommu_root, SPAPR_PCI_MSI_WINDOW,
1988                                 &sphb->msiwindow);
1989 
1990     pci_setup_iommu(bus, spapr_pci_dma_iommu, sphb);
1991 
1992     pci_bus_set_route_irq_fn(bus, spapr_route_intx_pin_to_irq);
1993 
1994     QLIST_INSERT_HEAD(&spapr->phbs, sphb, list);
1995 
1996     /* Initialize the LSI table */
1997     for (i = 0; i < PCI_NUM_PINS; i++) {
1998         int irq = SPAPR_IRQ_PCI_LSI + sphb->index * PCI_NUM_PINS + i;
1999 
2000         if (smc->legacy_irq_allocation) {
2001             irq = spapr_irq_findone(spapr, errp);
2002             if (irq < 0) {
2003                 error_prepend(errp, "can't allocate LSIs: ");
2004                 /*
2005                  * Older machines will never support PHB hotplug, ie, this is an
2006                  * init only path and QEMU will terminate. No need to rollback.
2007                  */
2008                 return;
2009             }
2010         }
2011 
2012         if (spapr_irq_claim(spapr, irq, true, errp) < 0) {
2013             error_prepend(errp, "can't allocate LSIs: ");
2014             goto unrealize;
2015         }
2016 
2017         sphb->lsi_table[i].irq = irq;
2018     }
2019 
2020     /* allocate connectors for child PCI devices */
2021     add_drcs(sphb, phb->bus);
2022 
2023     /* DMA setup */
2024     for (i = 0; i < windows_supported; ++i) {
2025         tcet = spapr_tce_new_table(DEVICE(sphb), sphb->dma_liobn[i]);
2026         if (!tcet) {
2027             error_setg(errp, "Creating window#%d failed for %s",
2028                        i, sphb->dtbusname);
2029             goto unrealize;
2030         }
2031         memory_region_add_subregion(&sphb->iommu_root, 0,
2032                                     spapr_tce_get_iommu(tcet));
2033     }
2034 
2035     sphb->msi = g_hash_table_new_full(g_int_hash, g_int_equal, g_free,
2036                                       spapr_phb_destroy_msi);
2037     return;
2038 
2039 unrealize:
2040     spapr_phb_unrealize(dev);
2041 }
2042 
2043 static int spapr_phb_children_reset(Object *child, void *opaque)
2044 {
2045     DeviceState *dev = (DeviceState *) object_dynamic_cast(child, TYPE_DEVICE);
2046 
2047     if (dev) {
2048         device_cold_reset(dev);
2049     }
2050 
2051     return 0;
2052 }
2053 
2054 void spapr_phb_dma_reset(SpaprPhbState *sphb)
2055 {
2056     int i;
2057     SpaprTceTable *tcet;
2058 
2059     for (i = 0; i < SPAPR_PCI_DMA_MAX_WINDOWS; ++i) {
2060         tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[i]);
2061 
2062         if (tcet && tcet->nb_table) {
2063             spapr_tce_table_disable(tcet);
2064         }
2065     }
2066 
2067     /* Register default 32bit DMA window */
2068     tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[0]);
2069     spapr_tce_table_enable(tcet, SPAPR_TCE_PAGE_SHIFT, sphb->dma_win_addr,
2070                            sphb->dma_win_size >> SPAPR_TCE_PAGE_SHIFT);
2071     tcet->def_win = true;
2072 }
2073 
2074 static void spapr_phb_reset(DeviceState *qdev)
2075 {
2076     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(qdev);
2077     Error *err = NULL;
2078 
2079     spapr_phb_dma_reset(sphb);
2080     spapr_phb_nvgpu_free(sphb);
2081     spapr_phb_nvgpu_setup(sphb, &err);
2082     if (err) {
2083         error_report_err(err);
2084     }
2085 
2086     /* Reset the IOMMU state */
2087     object_child_foreach(OBJECT(qdev), spapr_phb_children_reset, NULL);
2088 
2089     if (spapr_phb_eeh_available(SPAPR_PCI_HOST_BRIDGE(qdev))) {
2090         spapr_phb_vfio_reset(qdev);
2091     }
2092 
2093     g_hash_table_remove_all(sphb->msi);
2094 }
2095 
2096 static Property spapr_phb_properties[] = {
2097     DEFINE_PROP_UINT32("index", SpaprPhbState, index, -1),
2098     DEFINE_PROP_UINT64("mem_win_size", SpaprPhbState, mem_win_size,
2099                        SPAPR_PCI_MEM32_WIN_SIZE),
2100     DEFINE_PROP_UINT64("mem64_win_size", SpaprPhbState, mem64_win_size,
2101                        SPAPR_PCI_MEM64_WIN_SIZE),
2102     DEFINE_PROP_UINT64("io_win_size", SpaprPhbState, io_win_size,
2103                        SPAPR_PCI_IO_WIN_SIZE),
2104     DEFINE_PROP_BOOL("dynamic-reconfiguration", SpaprPhbState, dr_enabled,
2105                      true),
2106     /* Default DMA window is 0..1GB */
2107     DEFINE_PROP_UINT64("dma_win_addr", SpaprPhbState, dma_win_addr, 0),
2108     DEFINE_PROP_UINT64("dma_win_size", SpaprPhbState, dma_win_size, 0x40000000),
2109     DEFINE_PROP_UINT64("dma64_win_addr", SpaprPhbState, dma64_win_addr,
2110                        0x800000000000000ULL),
2111     DEFINE_PROP_BOOL("ddw", SpaprPhbState, ddw_enabled, true),
2112     DEFINE_PROP_UINT64("pgsz", SpaprPhbState, page_size_mask,
2113                        (1ULL << 12) | (1ULL << 16)
2114                        | (1ULL << 21) | (1ULL << 24)),
2115     DEFINE_PROP_UINT32("numa_node", SpaprPhbState, numa_node, -1),
2116     DEFINE_PROP_BOOL("pre-2.8-migration", SpaprPhbState,
2117                      pre_2_8_migration, false),
2118     DEFINE_PROP_BOOL("pcie-extended-configuration-space", SpaprPhbState,
2119                      pcie_ecs, true),
2120     DEFINE_PROP_UINT64("gpa", SpaprPhbState, nv2_gpa_win_addr, 0),
2121     DEFINE_PROP_UINT64("atsd", SpaprPhbState, nv2_atsd_win_addr, 0),
2122     DEFINE_PROP_BOOL("pre-5.1-associativity", SpaprPhbState,
2123                      pre_5_1_assoc, false),
2124     DEFINE_PROP_END_OF_LIST(),
2125 };
2126 
2127 static const VMStateDescription vmstate_spapr_pci_lsi = {
2128     .name = "spapr_pci/lsi",
2129     .version_id = 1,
2130     .minimum_version_id = 1,
2131     .fields = (VMStateField[]) {
2132         VMSTATE_UINT32_EQUAL(irq, SpaprPciLsi, NULL),
2133 
2134         VMSTATE_END_OF_LIST()
2135     },
2136 };
2137 
2138 static const VMStateDescription vmstate_spapr_pci_msi = {
2139     .name = "spapr_pci/msi",
2140     .version_id = 1,
2141     .minimum_version_id = 1,
2142     .fields = (VMStateField []) {
2143         VMSTATE_UINT32(key, SpaprPciMsiMig),
2144         VMSTATE_UINT32(value.first_irq, SpaprPciMsiMig),
2145         VMSTATE_UINT32(value.num, SpaprPciMsiMig),
2146         VMSTATE_END_OF_LIST()
2147     },
2148 };
2149 
2150 static int spapr_pci_pre_save(void *opaque)
2151 {
2152     SpaprPhbState *sphb = opaque;
2153     GHashTableIter iter;
2154     gpointer key, value;
2155     int i;
2156 
2157     if (sphb->pre_2_8_migration) {
2158         sphb->mig_liobn = sphb->dma_liobn[0];
2159         sphb->mig_mem_win_addr = sphb->mem_win_addr;
2160         sphb->mig_mem_win_size = sphb->mem_win_size;
2161         sphb->mig_io_win_addr = sphb->io_win_addr;
2162         sphb->mig_io_win_size = sphb->io_win_size;
2163 
2164         if ((sphb->mem64_win_size != 0)
2165             && (sphb->mem64_win_addr
2166                 == (sphb->mem_win_addr + sphb->mem_win_size))) {
2167             sphb->mig_mem_win_size += sphb->mem64_win_size;
2168         }
2169     }
2170 
2171     g_free(sphb->msi_devs);
2172     sphb->msi_devs = NULL;
2173     sphb->msi_devs_num = g_hash_table_size(sphb->msi);
2174     if (!sphb->msi_devs_num) {
2175         return 0;
2176     }
2177     sphb->msi_devs = g_new(SpaprPciMsiMig, sphb->msi_devs_num);
2178 
2179     g_hash_table_iter_init(&iter, sphb->msi);
2180     for (i = 0; g_hash_table_iter_next(&iter, &key, &value); ++i) {
2181         sphb->msi_devs[i].key = *(uint32_t *) key;
2182         sphb->msi_devs[i].value = *(SpaprPciMsi *) value;
2183     }
2184 
2185     return 0;
2186 }
2187 
2188 static int spapr_pci_post_save(void *opaque)
2189 {
2190     SpaprPhbState *sphb = opaque;
2191 
2192     g_free(sphb->msi_devs);
2193     sphb->msi_devs = NULL;
2194     sphb->msi_devs_num = 0;
2195     return 0;
2196 }
2197 
2198 static int spapr_pci_post_load(void *opaque, int version_id)
2199 {
2200     SpaprPhbState *sphb = opaque;
2201     gpointer key, value;
2202     int i;
2203 
2204     for (i = 0; i < sphb->msi_devs_num; ++i) {
2205         key = g_memdup(&sphb->msi_devs[i].key,
2206                        sizeof(sphb->msi_devs[i].key));
2207         value = g_memdup(&sphb->msi_devs[i].value,
2208                          sizeof(sphb->msi_devs[i].value));
2209         g_hash_table_insert(sphb->msi, key, value);
2210     }
2211     g_free(sphb->msi_devs);
2212     sphb->msi_devs = NULL;
2213     sphb->msi_devs_num = 0;
2214 
2215     return 0;
2216 }
2217 
2218 static bool pre_2_8_migration(void *opaque, int version_id)
2219 {
2220     SpaprPhbState *sphb = opaque;
2221 
2222     return sphb->pre_2_8_migration;
2223 }
2224 
2225 static const VMStateDescription vmstate_spapr_pci = {
2226     .name = "spapr_pci",
2227     .version_id = 2,
2228     .minimum_version_id = 2,
2229     .pre_save = spapr_pci_pre_save,
2230     .post_save = spapr_pci_post_save,
2231     .post_load = spapr_pci_post_load,
2232     .fields = (VMStateField[]) {
2233         VMSTATE_UINT64_EQUAL(buid, SpaprPhbState, NULL),
2234         VMSTATE_UINT32_TEST(mig_liobn, SpaprPhbState, pre_2_8_migration),
2235         VMSTATE_UINT64_TEST(mig_mem_win_addr, SpaprPhbState, pre_2_8_migration),
2236         VMSTATE_UINT64_TEST(mig_mem_win_size, SpaprPhbState, pre_2_8_migration),
2237         VMSTATE_UINT64_TEST(mig_io_win_addr, SpaprPhbState, pre_2_8_migration),
2238         VMSTATE_UINT64_TEST(mig_io_win_size, SpaprPhbState, pre_2_8_migration),
2239         VMSTATE_STRUCT_ARRAY(lsi_table, SpaprPhbState, PCI_NUM_PINS, 0,
2240                              vmstate_spapr_pci_lsi, SpaprPciLsi),
2241         VMSTATE_INT32(msi_devs_num, SpaprPhbState),
2242         VMSTATE_STRUCT_VARRAY_ALLOC(msi_devs, SpaprPhbState, msi_devs_num, 0,
2243                                     vmstate_spapr_pci_msi, SpaprPciMsiMig),
2244         VMSTATE_END_OF_LIST()
2245     },
2246 };
2247 
2248 static const char *spapr_phb_root_bus_path(PCIHostState *host_bridge,
2249                                            PCIBus *rootbus)
2250 {
2251     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(host_bridge);
2252 
2253     return sphb->dtbusname;
2254 }
2255 
2256 static void spapr_phb_class_init(ObjectClass *klass, void *data)
2257 {
2258     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(klass);
2259     DeviceClass *dc = DEVICE_CLASS(klass);
2260     HotplugHandlerClass *hp = HOTPLUG_HANDLER_CLASS(klass);
2261 
2262     hc->root_bus_path = spapr_phb_root_bus_path;
2263     dc->realize = spapr_phb_realize;
2264     dc->unrealize = spapr_phb_unrealize;
2265     device_class_set_props(dc, spapr_phb_properties);
2266     dc->reset = spapr_phb_reset;
2267     dc->vmsd = &vmstate_spapr_pci;
2268     /* Supported by TYPE_SPAPR_MACHINE */
2269     dc->user_creatable = true;
2270     set_bit(DEVICE_CATEGORY_BRIDGE, dc->categories);
2271     hp->pre_plug = spapr_pci_pre_plug;
2272     hp->plug = spapr_pci_plug;
2273     hp->unplug = spapr_pci_unplug;
2274     hp->unplug_request = spapr_pci_unplug_request;
2275 }
2276 
2277 static const TypeInfo spapr_phb_info = {
2278     .name          = TYPE_SPAPR_PCI_HOST_BRIDGE,
2279     .parent        = TYPE_PCI_HOST_BRIDGE,
2280     .instance_size = sizeof(SpaprPhbState),
2281     .instance_finalize = spapr_phb_finalizefn,
2282     .class_init    = spapr_phb_class_init,
2283     .interfaces    = (InterfaceInfo[]) {
2284         { TYPE_HOTPLUG_HANDLER },
2285         { }
2286     }
2287 };
2288 
2289 static void spapr_phb_pci_enumerate_bridge(PCIBus *bus, PCIDevice *pdev,
2290                                            void *opaque)
2291 {
2292     unsigned int *bus_no = opaque;
2293     PCIBus *sec_bus = NULL;
2294 
2295     if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) !=
2296          PCI_HEADER_TYPE_BRIDGE)) {
2297         return;
2298     }
2299 
2300     (*bus_no)++;
2301     pci_default_write_config(pdev, PCI_PRIMARY_BUS, pci_dev_bus_num(pdev), 1);
2302     pci_default_write_config(pdev, PCI_SECONDARY_BUS, *bus_no, 1);
2303     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
2304 
2305     sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev));
2306     if (!sec_bus) {
2307         return;
2308     }
2309 
2310     pci_for_each_device_under_bus(sec_bus, spapr_phb_pci_enumerate_bridge,
2311                                   bus_no);
2312     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
2313 }
2314 
2315 static void spapr_phb_pci_enumerate(SpaprPhbState *phb)
2316 {
2317     PCIBus *bus = PCI_HOST_BRIDGE(phb)->bus;
2318     unsigned int bus_no = 0;
2319 
2320     pci_for_each_device_under_bus(bus, spapr_phb_pci_enumerate_bridge,
2321                                   &bus_no);
2322 
2323 }
2324 
2325 int spapr_dt_phb(SpaprMachineState *spapr, SpaprPhbState *phb,
2326                  uint32_t intc_phandle, void *fdt, int *node_offset)
2327 {
2328     int bus_off, i, j, ret;
2329     uint32_t bus_range[] = { cpu_to_be32(0), cpu_to_be32(0xff) };
2330     struct {
2331         uint32_t hi;
2332         uint64_t child;
2333         uint64_t parent;
2334         uint64_t size;
2335     } QEMU_PACKED ranges[] = {
2336         {
2337             cpu_to_be32(b_ss(1)), cpu_to_be64(0),
2338             cpu_to_be64(phb->io_win_addr),
2339             cpu_to_be64(memory_region_size(&phb->iospace)),
2340         },
2341         {
2342             cpu_to_be32(b_ss(2)), cpu_to_be64(SPAPR_PCI_MEM_WIN_BUS_OFFSET),
2343             cpu_to_be64(phb->mem_win_addr),
2344             cpu_to_be64(phb->mem_win_size),
2345         },
2346         {
2347             cpu_to_be32(b_ss(3)), cpu_to_be64(phb->mem64_win_pciaddr),
2348             cpu_to_be64(phb->mem64_win_addr),
2349             cpu_to_be64(phb->mem64_win_size),
2350         },
2351     };
2352     const unsigned sizeof_ranges =
2353         (phb->mem64_win_size ? 3 : 2) * sizeof(ranges[0]);
2354     uint64_t bus_reg[] = { cpu_to_be64(phb->buid), 0 };
2355     uint32_t interrupt_map_mask[] = {
2356         cpu_to_be32(b_ddddd(-1)|b_fff(0)), 0x0, 0x0, cpu_to_be32(-1)};
2357     uint32_t interrupt_map[PCI_SLOT_MAX * PCI_NUM_PINS][7];
2358     uint32_t ddw_applicable[] = {
2359         cpu_to_be32(RTAS_IBM_QUERY_PE_DMA_WINDOW),
2360         cpu_to_be32(RTAS_IBM_CREATE_PE_DMA_WINDOW),
2361         cpu_to_be32(RTAS_IBM_REMOVE_PE_DMA_WINDOW)
2362     };
2363     uint32_t ddw_extensions[] = {
2364         cpu_to_be32(2),
2365         cpu_to_be32(RTAS_IBM_RESET_PE_DMA_WINDOW),
2366         cpu_to_be32(1), /* 1: ibm,query-pe-dma-window 6 outputs, PAPR 2.8 */
2367     };
2368     SpaprTceTable *tcet;
2369     SpaprDrc *drc;
2370     Error *err = NULL;
2371 
2372     /* Start populating the FDT */
2373     _FDT(bus_off = fdt_add_subnode(fdt, 0, phb->dtbusname));
2374     if (node_offset) {
2375         *node_offset = bus_off;
2376     }
2377 
2378     /* Write PHB properties */
2379     _FDT(fdt_setprop_string(fdt, bus_off, "device_type", "pci"));
2380     _FDT(fdt_setprop_string(fdt, bus_off, "compatible", "IBM,Logical_PHB"));
2381     _FDT(fdt_setprop_cell(fdt, bus_off, "#interrupt-cells", 0x1));
2382     _FDT(fdt_setprop(fdt, bus_off, "used-by-rtas", NULL, 0));
2383     _FDT(fdt_setprop(fdt, bus_off, "bus-range", &bus_range, sizeof(bus_range)));
2384     _FDT(fdt_setprop(fdt, bus_off, "ranges", &ranges, sizeof_ranges));
2385     _FDT(fdt_setprop(fdt, bus_off, "reg", &bus_reg, sizeof(bus_reg)));
2386     _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pci-config-space-type", 0x1));
2387     _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pe-total-#msi",
2388                           spapr_irq_nr_msis(spapr)));
2389 
2390     /* Dynamic DMA window */
2391     if (phb->ddw_enabled) {
2392         _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-applicable", &ddw_applicable,
2393                          sizeof(ddw_applicable)));
2394         _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-extensions",
2395                          &ddw_extensions, sizeof(ddw_extensions)));
2396     }
2397 
2398     /* Advertise NUMA via ibm,associativity */
2399     if (phb->numa_node != -1) {
2400         spapr_numa_write_associativity_dt(spapr, fdt, bus_off, phb->numa_node);
2401     }
2402 
2403     /* Build the interrupt-map, this must matches what is done
2404      * in pci_swizzle_map_irq_fn
2405      */
2406     _FDT(fdt_setprop(fdt, bus_off, "interrupt-map-mask",
2407                      &interrupt_map_mask, sizeof(interrupt_map_mask)));
2408     for (i = 0; i < PCI_SLOT_MAX; i++) {
2409         for (j = 0; j < PCI_NUM_PINS; j++) {
2410             uint32_t *irqmap = interrupt_map[i*PCI_NUM_PINS + j];
2411             int lsi_num = pci_swizzle(i, j);
2412 
2413             irqmap[0] = cpu_to_be32(b_ddddd(i)|b_fff(0));
2414             irqmap[1] = 0;
2415             irqmap[2] = 0;
2416             irqmap[3] = cpu_to_be32(j+1);
2417             irqmap[4] = cpu_to_be32(intc_phandle);
2418             spapr_dt_irq(&irqmap[5], phb->lsi_table[lsi_num].irq, true);
2419         }
2420     }
2421     /* Write interrupt map */
2422     _FDT(fdt_setprop(fdt, bus_off, "interrupt-map", &interrupt_map,
2423                      sizeof(interrupt_map)));
2424 
2425     tcet = spapr_tce_find_by_liobn(phb->dma_liobn[0]);
2426     if (!tcet) {
2427         return -1;
2428     }
2429     spapr_dma_dt(fdt, bus_off, "ibm,dma-window",
2430                  tcet->liobn, tcet->bus_offset,
2431                  tcet->nb_table << tcet->page_shift);
2432 
2433     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, phb->index);
2434     if (drc) {
2435         uint32_t drc_index = cpu_to_be32(spapr_drc_index(drc));
2436 
2437         _FDT(fdt_setprop(fdt, bus_off, "ibm,my-drc-index", &drc_index,
2438                          sizeof(drc_index)));
2439     }
2440 
2441     /* Walk the bridges and program the bus numbers*/
2442     spapr_phb_pci_enumerate(phb);
2443     _FDT(fdt_setprop_cell(fdt, bus_off, "qemu,phb-enumerated", 0x1));
2444 
2445     /* Walk the bridge and subordinate buses */
2446     ret = spapr_dt_pci_bus(phb, PCI_HOST_BRIDGE(phb)->bus, fdt, bus_off);
2447     if (ret < 0) {
2448         return ret;
2449     }
2450 
2451     spapr_phb_nvgpu_populate_dt(phb, fdt, bus_off, &err);
2452     if (err) {
2453         error_report_err(err);
2454     }
2455     spapr_phb_nvgpu_ram_populate_dt(phb, fdt);
2456 
2457     return 0;
2458 }
2459 
2460 void spapr_pci_rtas_init(void)
2461 {
2462     spapr_rtas_register(RTAS_READ_PCI_CONFIG, "read-pci-config",
2463                         rtas_read_pci_config);
2464     spapr_rtas_register(RTAS_WRITE_PCI_CONFIG, "write-pci-config",
2465                         rtas_write_pci_config);
2466     spapr_rtas_register(RTAS_IBM_READ_PCI_CONFIG, "ibm,read-pci-config",
2467                         rtas_ibm_read_pci_config);
2468     spapr_rtas_register(RTAS_IBM_WRITE_PCI_CONFIG, "ibm,write-pci-config",
2469                         rtas_ibm_write_pci_config);
2470     if (msi_nonbroken) {
2471         spapr_rtas_register(RTAS_IBM_QUERY_INTERRUPT_SOURCE_NUMBER,
2472                             "ibm,query-interrupt-source-number",
2473                             rtas_ibm_query_interrupt_source_number);
2474         spapr_rtas_register(RTAS_IBM_CHANGE_MSI, "ibm,change-msi",
2475                             rtas_ibm_change_msi);
2476     }
2477 
2478     spapr_rtas_register(RTAS_IBM_SET_EEH_OPTION,
2479                         "ibm,set-eeh-option",
2480                         rtas_ibm_set_eeh_option);
2481     spapr_rtas_register(RTAS_IBM_GET_CONFIG_ADDR_INFO2,
2482                         "ibm,get-config-addr-info2",
2483                         rtas_ibm_get_config_addr_info2);
2484     spapr_rtas_register(RTAS_IBM_READ_SLOT_RESET_STATE2,
2485                         "ibm,read-slot-reset-state2",
2486                         rtas_ibm_read_slot_reset_state2);
2487     spapr_rtas_register(RTAS_IBM_SET_SLOT_RESET,
2488                         "ibm,set-slot-reset",
2489                         rtas_ibm_set_slot_reset);
2490     spapr_rtas_register(RTAS_IBM_CONFIGURE_PE,
2491                         "ibm,configure-pe",
2492                         rtas_ibm_configure_pe);
2493     spapr_rtas_register(RTAS_IBM_SLOT_ERROR_DETAIL,
2494                         "ibm,slot-error-detail",
2495                         rtas_ibm_slot_error_detail);
2496 }
2497 
2498 static void spapr_pci_register_types(void)
2499 {
2500     type_register_static(&spapr_phb_info);
2501 }
2502 
2503 type_init(spapr_pci_register_types)
2504 
2505 static int spapr_switch_one_vga(DeviceState *dev, void *opaque)
2506 {
2507     bool be = *(bool *)opaque;
2508 
2509     if (object_dynamic_cast(OBJECT(dev), "VGA")
2510         || object_dynamic_cast(OBJECT(dev), "secondary-vga")
2511         || object_dynamic_cast(OBJECT(dev), "bochs-display")
2512         || object_dynamic_cast(OBJECT(dev), "virtio-vga")) {
2513         object_property_set_bool(OBJECT(dev), "big-endian-framebuffer", be,
2514                                  &error_abort);
2515     }
2516     return 0;
2517 }
2518 
2519 void spapr_pci_switch_vga(SpaprMachineState *spapr, bool big_endian)
2520 {
2521     SpaprPhbState *sphb;
2522 
2523     /*
2524      * For backward compatibility with existing guests, we switch
2525      * the endianness of the VGA controller when changing the guest
2526      * interrupt mode
2527      */
2528     QLIST_FOREACH(sphb, &spapr->phbs, list) {
2529         BusState *bus = &PCI_HOST_BRIDGE(sphb)->bus->qbus;
2530         qbus_walk_children(bus, spapr_switch_one_vga, NULL, NULL, NULL,
2531                            &big_endian);
2532     }
2533 }
2534