xref: /openbmc/qemu/hw/ppc/spapr_pci.c (revision 135b03cb)
1 /*
2  * QEMU sPAPR PCI host originated from Uninorth PCI host
3  *
4  * Copyright (c) 2011 Alexey Kardashevskiy, IBM Corporation.
5  * Copyright (C) 2011 David Gibson, IBM Corporation.
6  *
7  * Permission is hereby granted, free of charge, to any person obtaining a copy
8  * of this software and associated documentation files (the "Software"), to deal
9  * in the Software without restriction, including without limitation the rights
10  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11  * copies of the Software, and to permit persons to whom the Software is
12  * furnished to do so, subject to the following conditions:
13  *
14  * The above copyright notice and this permission notice shall be included in
15  * all copies or substantial portions of the Software.
16  *
17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
23  * THE SOFTWARE.
24  */
25 
26 #include "qemu/osdep.h"
27 #include "qapi/error.h"
28 #include "cpu.h"
29 #include "hw/irq.h"
30 #include "hw/sysbus.h"
31 #include "migration/vmstate.h"
32 #include "hw/pci/pci.h"
33 #include "hw/pci/msi.h"
34 #include "hw/pci/msix.h"
35 #include "hw/pci/pci_host.h"
36 #include "hw/ppc/spapr.h"
37 #include "hw/pci-host/spapr.h"
38 #include "exec/address-spaces.h"
39 #include "exec/ram_addr.h"
40 #include <libfdt.h>
41 #include "trace.h"
42 #include "qemu/error-report.h"
43 #include "qemu/module.h"
44 #include "qapi/qmp/qerror.h"
45 #include "hw/ppc/fdt.h"
46 #include "hw/pci/pci_bridge.h"
47 #include "hw/pci/pci_bus.h"
48 #include "hw/pci/pci_ids.h"
49 #include "hw/ppc/spapr_drc.h"
50 #include "hw/qdev-properties.h"
51 #include "sysemu/device_tree.h"
52 #include "sysemu/kvm.h"
53 #include "sysemu/hostmem.h"
54 #include "sysemu/numa.h"
55 
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     spapr_pci_msi *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 = (spapr_pci_msi *) 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 (!smc->legacy_irq_allocation) {
342             spapr_irq_msi_free(spapr, msi->first_irq, msi->num);
343         }
344         spapr_irq_free(spapr, msi->first_irq, msi->num);
345         if (msi_present(pdev)) {
346             spapr_msi_setmsg(pdev, 0, false, 0, 0);
347         }
348         if (msix_present(pdev)) {
349             spapr_msi_setmsg(pdev, 0, true, 0, 0);
350         }
351         g_hash_table_remove(phb->msi, &config_addr);
352 
353         trace_spapr_pci_msi("Released MSIs", config_addr);
354         rtas_st(rets, 0, RTAS_OUT_SUCCESS);
355         rtas_st(rets, 1, 0);
356         return;
357     }
358 
359     /* Enabling MSI */
360 
361     /* Check if the device supports as many IRQs as requested */
362     if (ret_intr_type == RTAS_TYPE_MSI) {
363         max_irqs = msi_nr_vectors_allocated(pdev);
364     } else if (ret_intr_type == RTAS_TYPE_MSIX) {
365         max_irqs = pdev->msix_entries_nr;
366     }
367     if (!max_irqs) {
368         error_report("Requested interrupt type %d is not enabled for device %x",
369                      ret_intr_type, config_addr);
370         rtas_st(rets, 0, -1); /* Hardware error */
371         return;
372     }
373     /* Correct the number if the guest asked for too many */
374     if (req_num > max_irqs) {
375         trace_spapr_pci_msi_retry(config_addr, req_num, max_irqs);
376         req_num = max_irqs;
377         irq = 0; /* to avoid misleading trace */
378         goto out;
379     }
380 
381     /* Allocate MSIs */
382     if (smc->legacy_irq_allocation) {
383         irq = spapr_irq_find(spapr, req_num, ret_intr_type == RTAS_TYPE_MSI,
384                              &err);
385     } else {
386         irq = spapr_irq_msi_alloc(spapr, req_num,
387                                   ret_intr_type == RTAS_TYPE_MSI, &err);
388     }
389     if (err) {
390         error_reportf_err(err, "Can't allocate MSIs for device %x: ",
391                           config_addr);
392         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
393         return;
394     }
395 
396     for (i = 0; i < req_num; i++) {
397         spapr_irq_claim(spapr, irq + i, false, &err);
398         if (err) {
399             if (i) {
400                 spapr_irq_free(spapr, irq, i);
401             }
402             if (!smc->legacy_irq_allocation) {
403                 spapr_irq_msi_free(spapr, irq, req_num);
404             }
405             error_reportf_err(err, "Can't allocate MSIs for device %x: ",
406                               config_addr);
407             rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
408             return;
409         }
410     }
411 
412     /* Release previous MSIs */
413     if (msi) {
414         if (!smc->legacy_irq_allocation) {
415             spapr_irq_msi_free(spapr, msi->first_irq, msi->num);
416         }
417         spapr_irq_free(spapr, msi->first_irq, msi->num);
418         g_hash_table_remove(phb->msi, &config_addr);
419     }
420 
421     /* Setup MSI/MSIX vectors in the device (via cfgspace or MSIX BAR) */
422     spapr_msi_setmsg(pdev, SPAPR_PCI_MSI_WINDOW, ret_intr_type == RTAS_TYPE_MSIX,
423                      irq, req_num);
424 
425     /* Add MSI device to cache */
426     msi = g_new(spapr_pci_msi, 1);
427     msi->first_irq = irq;
428     msi->num = req_num;
429     config_addr_key = g_new(int, 1);
430     *config_addr_key = config_addr;
431     g_hash_table_insert(phb->msi, config_addr_key, msi);
432 
433 out:
434     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
435     rtas_st(rets, 1, req_num);
436     rtas_st(rets, 2, ++seq_num);
437     if (nret > 3) {
438         rtas_st(rets, 3, ret_intr_type);
439     }
440 
441     trace_spapr_pci_rtas_ibm_change_msi(config_addr, func, req_num, irq);
442 }
443 
444 static void rtas_ibm_query_interrupt_source_number(PowerPCCPU *cpu,
445                                                    SpaprMachineState *spapr,
446                                                    uint32_t token,
447                                                    uint32_t nargs,
448                                                    target_ulong args,
449                                                    uint32_t nret,
450                                                    target_ulong rets)
451 {
452     uint32_t config_addr = rtas_ld(args, 0);
453     uint64_t buid = rtas_ldq(args, 1);
454     unsigned int intr_src_num = -1, ioa_intr_num = rtas_ld(args, 3);
455     SpaprPhbState *phb = NULL;
456     PCIDevice *pdev = NULL;
457     spapr_pci_msi *msi;
458 
459     /* Find SpaprPhbState */
460     phb = spapr_pci_find_phb(spapr, buid);
461     if (phb) {
462         pdev = spapr_pci_find_dev(spapr, buid, config_addr);
463     }
464     if (!phb || !pdev) {
465         rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
466         return;
467     }
468 
469     /* Find device descriptor and start IRQ */
470     msi = (spapr_pci_msi *) g_hash_table_lookup(phb->msi, &config_addr);
471     if (!msi || !msi->first_irq || !msi->num || (ioa_intr_num >= msi->num)) {
472         trace_spapr_pci_msi("Failed to return vector", config_addr);
473         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
474         return;
475     }
476     intr_src_num = msi->first_irq + ioa_intr_num;
477     trace_spapr_pci_rtas_ibm_query_interrupt_source_number(ioa_intr_num,
478                                                            intr_src_num);
479 
480     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
481     rtas_st(rets, 1, intr_src_num);
482     rtas_st(rets, 2, 1);/* 0 == level; 1 == edge */
483 }
484 
485 static void rtas_ibm_set_eeh_option(PowerPCCPU *cpu,
486                                     SpaprMachineState *spapr,
487                                     uint32_t token, uint32_t nargs,
488                                     target_ulong args, uint32_t nret,
489                                     target_ulong rets)
490 {
491     SpaprPhbState *sphb;
492     uint32_t addr, option;
493     uint64_t buid;
494     int ret;
495 
496     if ((nargs != 4) || (nret != 1)) {
497         goto param_error_exit;
498     }
499 
500     buid = rtas_ldq(args, 1);
501     addr = rtas_ld(args, 0);
502     option = rtas_ld(args, 3);
503 
504     sphb = spapr_pci_find_phb(spapr, buid);
505     if (!sphb) {
506         goto param_error_exit;
507     }
508 
509     if (!spapr_phb_eeh_available(sphb)) {
510         goto param_error_exit;
511     }
512 
513     ret = spapr_phb_vfio_eeh_set_option(sphb, addr, option);
514     rtas_st(rets, 0, ret);
515     return;
516 
517 param_error_exit:
518     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
519 }
520 
521 static void rtas_ibm_get_config_addr_info2(PowerPCCPU *cpu,
522                                            SpaprMachineState *spapr,
523                                            uint32_t token, uint32_t nargs,
524                                            target_ulong args, uint32_t nret,
525                                            target_ulong rets)
526 {
527     SpaprPhbState *sphb;
528     PCIDevice *pdev;
529     uint32_t addr, option;
530     uint64_t buid;
531 
532     if ((nargs != 4) || (nret != 2)) {
533         goto param_error_exit;
534     }
535 
536     buid = rtas_ldq(args, 1);
537     sphb = spapr_pci_find_phb(spapr, buid);
538     if (!sphb) {
539         goto param_error_exit;
540     }
541 
542     if (!spapr_phb_eeh_available(sphb)) {
543         goto param_error_exit;
544     }
545 
546     /*
547      * We always have PE address of form "00BB0001". "BB"
548      * represents the bus number of PE's primary bus.
549      */
550     option = rtas_ld(args, 3);
551     switch (option) {
552     case RTAS_GET_PE_ADDR:
553         addr = rtas_ld(args, 0);
554         pdev = spapr_pci_find_dev(spapr, buid, addr);
555         if (!pdev) {
556             goto param_error_exit;
557         }
558 
559         rtas_st(rets, 1, (pci_bus_num(pci_get_bus(pdev)) << 16) + 1);
560         break;
561     case RTAS_GET_PE_MODE:
562         rtas_st(rets, 1, RTAS_PE_MODE_SHARED);
563         break;
564     default:
565         goto param_error_exit;
566     }
567 
568     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
569     return;
570 
571 param_error_exit:
572     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
573 }
574 
575 static void rtas_ibm_read_slot_reset_state2(PowerPCCPU *cpu,
576                                             SpaprMachineState *spapr,
577                                             uint32_t token, uint32_t nargs,
578                                             target_ulong args, uint32_t nret,
579                                             target_ulong rets)
580 {
581     SpaprPhbState *sphb;
582     uint64_t buid;
583     int state, ret;
584 
585     if ((nargs != 3) || (nret != 4 && nret != 5)) {
586         goto param_error_exit;
587     }
588 
589     buid = rtas_ldq(args, 1);
590     sphb = spapr_pci_find_phb(spapr, buid);
591     if (!sphb) {
592         goto param_error_exit;
593     }
594 
595     if (!spapr_phb_eeh_available(sphb)) {
596         goto param_error_exit;
597     }
598 
599     ret = spapr_phb_vfio_eeh_get_state(sphb, &state);
600     rtas_st(rets, 0, ret);
601     if (ret != RTAS_OUT_SUCCESS) {
602         return;
603     }
604 
605     rtas_st(rets, 1, state);
606     rtas_st(rets, 2, RTAS_EEH_SUPPORT);
607     rtas_st(rets, 3, RTAS_EEH_PE_UNAVAIL_INFO);
608     if (nret >= 5) {
609         rtas_st(rets, 4, RTAS_EEH_PE_RECOVER_INFO);
610     }
611     return;
612 
613 param_error_exit:
614     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
615 }
616 
617 static void rtas_ibm_set_slot_reset(PowerPCCPU *cpu,
618                                     SpaprMachineState *spapr,
619                                     uint32_t token, uint32_t nargs,
620                                     target_ulong args, uint32_t nret,
621                                     target_ulong rets)
622 {
623     SpaprPhbState *sphb;
624     uint32_t option;
625     uint64_t buid;
626     int ret;
627 
628     if ((nargs != 4) || (nret != 1)) {
629         goto param_error_exit;
630     }
631 
632     buid = rtas_ldq(args, 1);
633     option = rtas_ld(args, 3);
634     sphb = spapr_pci_find_phb(spapr, buid);
635     if (!sphb) {
636         goto param_error_exit;
637     }
638 
639     if (!spapr_phb_eeh_available(sphb)) {
640         goto param_error_exit;
641     }
642 
643     ret = spapr_phb_vfio_eeh_reset(sphb, option);
644     rtas_st(rets, 0, ret);
645     return;
646 
647 param_error_exit:
648     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
649 }
650 
651 static void rtas_ibm_configure_pe(PowerPCCPU *cpu,
652                                   SpaprMachineState *spapr,
653                                   uint32_t token, uint32_t nargs,
654                                   target_ulong args, uint32_t nret,
655                                   target_ulong rets)
656 {
657     SpaprPhbState *sphb;
658     uint64_t buid;
659     int ret;
660 
661     if ((nargs != 3) || (nret != 1)) {
662         goto param_error_exit;
663     }
664 
665     buid = rtas_ldq(args, 1);
666     sphb = spapr_pci_find_phb(spapr, buid);
667     if (!sphb) {
668         goto param_error_exit;
669     }
670 
671     if (!spapr_phb_eeh_available(sphb)) {
672         goto param_error_exit;
673     }
674 
675     ret = spapr_phb_vfio_eeh_configure(sphb);
676     rtas_st(rets, 0, ret);
677     return;
678 
679 param_error_exit:
680     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
681 }
682 
683 /* To support it later */
684 static void rtas_ibm_slot_error_detail(PowerPCCPU *cpu,
685                                        SpaprMachineState *spapr,
686                                        uint32_t token, uint32_t nargs,
687                                        target_ulong args, uint32_t nret,
688                                        target_ulong rets)
689 {
690     SpaprPhbState *sphb;
691     int option;
692     uint64_t buid;
693 
694     if ((nargs != 8) || (nret != 1)) {
695         goto param_error_exit;
696     }
697 
698     buid = rtas_ldq(args, 1);
699     sphb = spapr_pci_find_phb(spapr, buid);
700     if (!sphb) {
701         goto param_error_exit;
702     }
703 
704     if (!spapr_phb_eeh_available(sphb)) {
705         goto param_error_exit;
706     }
707 
708     option = rtas_ld(args, 7);
709     switch (option) {
710     case RTAS_SLOT_TEMP_ERR_LOG:
711     case RTAS_SLOT_PERM_ERR_LOG:
712         break;
713     default:
714         goto param_error_exit;
715     }
716 
717     /* We don't have error log yet */
718     rtas_st(rets, 0, RTAS_OUT_NO_ERRORS_FOUND);
719     return;
720 
721 param_error_exit:
722     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
723 }
724 
725 static void pci_spapr_set_irq(void *opaque, int irq_num, int level)
726 {
727     /*
728      * Here we use the number returned by pci_swizzle_map_irq_fn to find a
729      * corresponding qemu_irq.
730      */
731     SpaprPhbState *phb = opaque;
732 
733     trace_spapr_pci_lsi_set(phb->dtbusname, irq_num, phb->lsi_table[irq_num].irq);
734     qemu_set_irq(spapr_phb_lsi_qirq(phb, irq_num), level);
735 }
736 
737 static PCIINTxRoute spapr_route_intx_pin_to_irq(void *opaque, int pin)
738 {
739     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(opaque);
740     PCIINTxRoute route;
741 
742     route.mode = PCI_INTX_ENABLED;
743     route.irq = sphb->lsi_table[pin].irq;
744 
745     return route;
746 }
747 
748 /*
749  * MSI/MSIX memory region implementation.
750  * The handler handles both MSI and MSIX.
751  * The vector number is encoded in least bits in data.
752  */
753 static void spapr_msi_write(void *opaque, hwaddr addr,
754                             uint64_t data, unsigned size)
755 {
756     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
757     uint32_t irq = data;
758 
759     trace_spapr_pci_msi_write(addr, data, irq);
760 
761     qemu_irq_pulse(spapr_qirq(spapr, irq));
762 }
763 
764 static const MemoryRegionOps spapr_msi_ops = {
765     /* There is no .read as the read result is undefined by PCI spec */
766     .read = NULL,
767     .write = spapr_msi_write,
768     .endianness = DEVICE_LITTLE_ENDIAN
769 };
770 
771 /*
772  * PHB PCI device
773  */
774 static AddressSpace *spapr_pci_dma_iommu(PCIBus *bus, void *opaque, int devfn)
775 {
776     SpaprPhbState *phb = opaque;
777 
778     return &phb->iommu_as;
779 }
780 
781 static char *spapr_phb_vfio_get_loc_code(SpaprPhbState *sphb,  PCIDevice *pdev)
782 {
783     char *path = NULL, *buf = NULL, *host = NULL;
784 
785     /* Get the PCI VFIO host id */
786     host = object_property_get_str(OBJECT(pdev), "host", NULL);
787     if (!host) {
788         goto err_out;
789     }
790 
791     /* Construct the path of the file that will give us the DT location */
792     path = g_strdup_printf("/sys/bus/pci/devices/%s/devspec", host);
793     g_free(host);
794     if (!g_file_get_contents(path, &buf, NULL, NULL)) {
795         goto err_out;
796     }
797     g_free(path);
798 
799     /* Construct and read from host device tree the loc-code */
800     path = g_strdup_printf("/proc/device-tree%s/ibm,loc-code", buf);
801     g_free(buf);
802     if (!g_file_get_contents(path, &buf, NULL, NULL)) {
803         goto err_out;
804     }
805     return buf;
806 
807 err_out:
808     g_free(path);
809     return NULL;
810 }
811 
812 static char *spapr_phb_get_loc_code(SpaprPhbState *sphb, PCIDevice *pdev)
813 {
814     char *buf;
815     const char *devtype = "qemu";
816     uint32_t busnr = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(pdev))));
817 
818     if (object_dynamic_cast(OBJECT(pdev), "vfio-pci")) {
819         buf = spapr_phb_vfio_get_loc_code(sphb, pdev);
820         if (buf) {
821             return buf;
822         }
823         devtype = "vfio";
824     }
825     /*
826      * For emulated devices and VFIO-failure case, make up
827      * the loc-code.
828      */
829     buf = g_strdup_printf("%s_%s:%04x:%02x:%02x.%x",
830                           devtype, pdev->name, sphb->index, busnr,
831                           PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
832     return buf;
833 }
834 
835 /* Macros to operate with address in OF binding to PCI */
836 #define b_x(x, p, l)    (((x) & ((1<<(l))-1)) << (p))
837 #define b_n(x)          b_x((x), 31, 1) /* 0 if relocatable */
838 #define b_p(x)          b_x((x), 30, 1) /* 1 if prefetchable */
839 #define b_t(x)          b_x((x), 29, 1) /* 1 if the address is aliased */
840 #define b_ss(x)         b_x((x), 24, 2) /* the space code */
841 #define b_bbbbbbbb(x)   b_x((x), 16, 8) /* bus number */
842 #define b_ddddd(x)      b_x((x), 11, 5) /* device number */
843 #define b_fff(x)        b_x((x), 8, 3)  /* function number */
844 #define b_rrrrrrrr(x)   b_x((x), 0, 8)  /* register number */
845 
846 /* for 'reg'/'assigned-addresses' OF properties */
847 #define RESOURCE_CELLS_SIZE 2
848 #define RESOURCE_CELLS_ADDRESS 3
849 
850 typedef struct ResourceFields {
851     uint32_t phys_hi;
852     uint32_t phys_mid;
853     uint32_t phys_lo;
854     uint32_t size_hi;
855     uint32_t size_lo;
856 } QEMU_PACKED ResourceFields;
857 
858 typedef struct ResourceProps {
859     ResourceFields reg[8];
860     ResourceFields assigned[7];
861     uint32_t reg_len;
862     uint32_t assigned_len;
863 } ResourceProps;
864 
865 /* fill in the 'reg'/'assigned-resources' OF properties for
866  * a PCI device. 'reg' describes resource requirements for a
867  * device's IO/MEM regions, 'assigned-addresses' describes the
868  * actual resource assignments.
869  *
870  * the properties are arrays of ('phys-addr', 'size') pairs describing
871  * the addressable regions of the PCI device, where 'phys-addr' is a
872  * RESOURCE_CELLS_ADDRESS-tuple of 32-bit integers corresponding to
873  * (phys.hi, phys.mid, phys.lo), and 'size' is a
874  * RESOURCE_CELLS_SIZE-tuple corresponding to (size.hi, size.lo).
875  *
876  * phys.hi = 0xYYXXXXZZ, where:
877  *   0xYY = npt000ss
878  *          |||   |
879  *          |||   +-- space code
880  *          |||               |
881  *          |||               +  00 if configuration space
882  *          |||               +  01 if IO region,
883  *          |||               +  10 if 32-bit MEM region
884  *          |||               +  11 if 64-bit MEM region
885  *          |||
886  *          ||+------ for non-relocatable IO: 1 if aliased
887  *          ||        for relocatable IO: 1 if below 64KB
888  *          ||        for MEM: 1 if below 1MB
889  *          |+------- 1 if region is prefetchable
890  *          +-------- 1 if region is non-relocatable
891  *   0xXXXX = bbbbbbbb dddddfff, encoding bus, slot, and function
892  *            bits respectively
893  *   0xZZ = rrrrrrrr, the register number of the BAR corresponding
894  *          to the region
895  *
896  * phys.mid and phys.lo correspond respectively to the hi/lo portions
897  * of the actual address of the region.
898  *
899  * how the phys-addr/size values are used differ slightly between
900  * 'reg' and 'assigned-addresses' properties. namely, 'reg' has
901  * an additional description for the config space region of the
902  * device, and in the case of QEMU has n=0 and phys.mid=phys.lo=0
903  * to describe the region as relocatable, with an address-mapping
904  * that corresponds directly to the PHB's address space for the
905  * resource. 'assigned-addresses' always has n=1 set with an absolute
906  * address assigned for the resource. in general, 'assigned-addresses'
907  * won't be populated, since addresses for PCI devices are generally
908  * unmapped initially and left to the guest to assign.
909  *
910  * note also that addresses defined in these properties are, at least
911  * for PAPR guests, relative to the PHBs IO/MEM windows, and
912  * correspond directly to the addresses in the BARs.
913  *
914  * in accordance with PCI Bus Binding to Open Firmware,
915  * IEEE Std 1275-1994, section 4.1.1, as implemented by PAPR+ v2.7,
916  * Appendix C.
917  */
918 static void populate_resource_props(PCIDevice *d, ResourceProps *rp)
919 {
920     int bus_num = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(d))));
921     uint32_t dev_id = (b_bbbbbbbb(bus_num) |
922                        b_ddddd(PCI_SLOT(d->devfn)) |
923                        b_fff(PCI_FUNC(d->devfn)));
924     ResourceFields *reg, *assigned;
925     int i, reg_idx = 0, assigned_idx = 0;
926 
927     /* config space region */
928     reg = &rp->reg[reg_idx++];
929     reg->phys_hi = cpu_to_be32(dev_id);
930     reg->phys_mid = 0;
931     reg->phys_lo = 0;
932     reg->size_hi = 0;
933     reg->size_lo = 0;
934 
935     for (i = 0; i < PCI_NUM_REGIONS; i++) {
936         if (!d->io_regions[i].size) {
937             continue;
938         }
939 
940         reg = &rp->reg[reg_idx++];
941 
942         reg->phys_hi = cpu_to_be32(dev_id | b_rrrrrrrr(pci_bar(d, i)));
943         if (d->io_regions[i].type & PCI_BASE_ADDRESS_SPACE_IO) {
944             reg->phys_hi |= cpu_to_be32(b_ss(1));
945         } else if (d->io_regions[i].type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
946             reg->phys_hi |= cpu_to_be32(b_ss(3));
947         } else {
948             reg->phys_hi |= cpu_to_be32(b_ss(2));
949         }
950         reg->phys_mid = 0;
951         reg->phys_lo = 0;
952         reg->size_hi = cpu_to_be32(d->io_regions[i].size >> 32);
953         reg->size_lo = cpu_to_be32(d->io_regions[i].size);
954 
955         if (d->io_regions[i].addr == PCI_BAR_UNMAPPED) {
956             continue;
957         }
958 
959         assigned = &rp->assigned[assigned_idx++];
960         assigned->phys_hi = cpu_to_be32(be32_to_cpu(reg->phys_hi) | b_n(1));
961         assigned->phys_mid = cpu_to_be32(d->io_regions[i].addr >> 32);
962         assigned->phys_lo = cpu_to_be32(d->io_regions[i].addr);
963         assigned->size_hi = reg->size_hi;
964         assigned->size_lo = reg->size_lo;
965     }
966 
967     rp->reg_len = reg_idx * sizeof(ResourceFields);
968     rp->assigned_len = assigned_idx * sizeof(ResourceFields);
969 }
970 
971 typedef struct PCIClass PCIClass;
972 typedef struct PCISubClass PCISubClass;
973 typedef struct PCIIFace PCIIFace;
974 
975 struct PCIIFace {
976     int iface;
977     const char *name;
978 };
979 
980 struct PCISubClass {
981     int subclass;
982     const char *name;
983     const PCIIFace *iface;
984 };
985 
986 struct PCIClass {
987     const char *name;
988     const PCISubClass *subc;
989 };
990 
991 static const PCISubClass undef_subclass[] = {
992     { PCI_CLASS_NOT_DEFINED_VGA, "display", NULL },
993     { 0xFF, NULL, NULL },
994 };
995 
996 static const PCISubClass mass_subclass[] = {
997     { PCI_CLASS_STORAGE_SCSI, "scsi", NULL },
998     { PCI_CLASS_STORAGE_IDE, "ide", NULL },
999     { PCI_CLASS_STORAGE_FLOPPY, "fdc", NULL },
1000     { PCI_CLASS_STORAGE_IPI, "ipi", NULL },
1001     { PCI_CLASS_STORAGE_RAID, "raid", NULL },
1002     { PCI_CLASS_STORAGE_ATA, "ata", NULL },
1003     { PCI_CLASS_STORAGE_SATA, "sata", NULL },
1004     { PCI_CLASS_STORAGE_SAS, "sas", NULL },
1005     { 0xFF, NULL, NULL },
1006 };
1007 
1008 static const PCISubClass net_subclass[] = {
1009     { PCI_CLASS_NETWORK_ETHERNET, "ethernet", NULL },
1010     { PCI_CLASS_NETWORK_TOKEN_RING, "token-ring", NULL },
1011     { PCI_CLASS_NETWORK_FDDI, "fddi", NULL },
1012     { PCI_CLASS_NETWORK_ATM, "atm", NULL },
1013     { PCI_CLASS_NETWORK_ISDN, "isdn", NULL },
1014     { PCI_CLASS_NETWORK_WORLDFIP, "worldfip", NULL },
1015     { PCI_CLASS_NETWORK_PICMG214, "picmg", NULL },
1016     { 0xFF, NULL, NULL },
1017 };
1018 
1019 static const PCISubClass displ_subclass[] = {
1020     { PCI_CLASS_DISPLAY_VGA, "vga", NULL },
1021     { PCI_CLASS_DISPLAY_XGA, "xga", NULL },
1022     { PCI_CLASS_DISPLAY_3D, "3d-controller", NULL },
1023     { 0xFF, NULL, NULL },
1024 };
1025 
1026 static const PCISubClass media_subclass[] = {
1027     { PCI_CLASS_MULTIMEDIA_VIDEO, "video", NULL },
1028     { PCI_CLASS_MULTIMEDIA_AUDIO, "sound", NULL },
1029     { PCI_CLASS_MULTIMEDIA_PHONE, "telephony", NULL },
1030     { 0xFF, NULL, NULL },
1031 };
1032 
1033 static const PCISubClass mem_subclass[] = {
1034     { PCI_CLASS_MEMORY_RAM, "memory", NULL },
1035     { PCI_CLASS_MEMORY_FLASH, "flash", NULL },
1036     { 0xFF, NULL, NULL },
1037 };
1038 
1039 static const PCISubClass bridg_subclass[] = {
1040     { PCI_CLASS_BRIDGE_HOST, "host", NULL },
1041     { PCI_CLASS_BRIDGE_ISA, "isa", NULL },
1042     { PCI_CLASS_BRIDGE_EISA, "eisa", NULL },
1043     { PCI_CLASS_BRIDGE_MC, "mca", NULL },
1044     { PCI_CLASS_BRIDGE_PCI, "pci", NULL },
1045     { PCI_CLASS_BRIDGE_PCMCIA, "pcmcia", NULL },
1046     { PCI_CLASS_BRIDGE_NUBUS, "nubus", NULL },
1047     { PCI_CLASS_BRIDGE_CARDBUS, "cardbus", NULL },
1048     { PCI_CLASS_BRIDGE_RACEWAY, "raceway", NULL },
1049     { PCI_CLASS_BRIDGE_PCI_SEMITP, "semi-transparent-pci", NULL },
1050     { PCI_CLASS_BRIDGE_IB_PCI, "infiniband", NULL },
1051     { 0xFF, NULL, NULL },
1052 };
1053 
1054 static const PCISubClass comm_subclass[] = {
1055     { PCI_CLASS_COMMUNICATION_SERIAL, "serial", NULL },
1056     { PCI_CLASS_COMMUNICATION_PARALLEL, "parallel", NULL },
1057     { PCI_CLASS_COMMUNICATION_MULTISERIAL, "multiport-serial", NULL },
1058     { PCI_CLASS_COMMUNICATION_MODEM, "modem", NULL },
1059     { PCI_CLASS_COMMUNICATION_GPIB, "gpib", NULL },
1060     { PCI_CLASS_COMMUNICATION_SC, "smart-card", NULL },
1061     { 0xFF, NULL, NULL, },
1062 };
1063 
1064 static const PCIIFace pic_iface[] = {
1065     { PCI_CLASS_SYSTEM_PIC_IOAPIC, "io-apic" },
1066     { PCI_CLASS_SYSTEM_PIC_IOXAPIC, "io-xapic" },
1067     { 0xFF, NULL },
1068 };
1069 
1070 static const PCISubClass sys_subclass[] = {
1071     { PCI_CLASS_SYSTEM_PIC, "interrupt-controller", pic_iface },
1072     { PCI_CLASS_SYSTEM_DMA, "dma-controller", NULL },
1073     { PCI_CLASS_SYSTEM_TIMER, "timer", NULL },
1074     { PCI_CLASS_SYSTEM_RTC, "rtc", NULL },
1075     { PCI_CLASS_SYSTEM_PCI_HOTPLUG, "hot-plug-controller", NULL },
1076     { PCI_CLASS_SYSTEM_SDHCI, "sd-host-controller", NULL },
1077     { 0xFF, NULL, NULL },
1078 };
1079 
1080 static const PCISubClass inp_subclass[] = {
1081     { PCI_CLASS_INPUT_KEYBOARD, "keyboard", NULL },
1082     { PCI_CLASS_INPUT_PEN, "pen", NULL },
1083     { PCI_CLASS_INPUT_MOUSE, "mouse", NULL },
1084     { PCI_CLASS_INPUT_SCANNER, "scanner", NULL },
1085     { PCI_CLASS_INPUT_GAMEPORT, "gameport", NULL },
1086     { 0xFF, NULL, NULL },
1087 };
1088 
1089 static const PCISubClass dock_subclass[] = {
1090     { PCI_CLASS_DOCKING_GENERIC, "dock", NULL },
1091     { 0xFF, NULL, NULL },
1092 };
1093 
1094 static const PCISubClass cpu_subclass[] = {
1095     { PCI_CLASS_PROCESSOR_PENTIUM, "pentium", NULL },
1096     { PCI_CLASS_PROCESSOR_POWERPC, "powerpc", NULL },
1097     { PCI_CLASS_PROCESSOR_MIPS, "mips", NULL },
1098     { PCI_CLASS_PROCESSOR_CO, "co-processor", NULL },
1099     { 0xFF, NULL, NULL },
1100 };
1101 
1102 static const PCIIFace usb_iface[] = {
1103     { PCI_CLASS_SERIAL_USB_UHCI, "usb-uhci" },
1104     { PCI_CLASS_SERIAL_USB_OHCI, "usb-ohci", },
1105     { PCI_CLASS_SERIAL_USB_EHCI, "usb-ehci" },
1106     { PCI_CLASS_SERIAL_USB_XHCI, "usb-xhci" },
1107     { PCI_CLASS_SERIAL_USB_UNKNOWN, "usb-unknown" },
1108     { PCI_CLASS_SERIAL_USB_DEVICE, "usb-device" },
1109     { 0xFF, NULL },
1110 };
1111 
1112 static const PCISubClass ser_subclass[] = {
1113     { PCI_CLASS_SERIAL_FIREWIRE, "firewire", NULL },
1114     { PCI_CLASS_SERIAL_ACCESS, "access-bus", NULL },
1115     { PCI_CLASS_SERIAL_SSA, "ssa", NULL },
1116     { PCI_CLASS_SERIAL_USB, "usb", usb_iface },
1117     { PCI_CLASS_SERIAL_FIBER, "fibre-channel", NULL },
1118     { PCI_CLASS_SERIAL_SMBUS, "smb", NULL },
1119     { PCI_CLASS_SERIAL_IB, "infiniband", NULL },
1120     { PCI_CLASS_SERIAL_IPMI, "ipmi", NULL },
1121     { PCI_CLASS_SERIAL_SERCOS, "sercos", NULL },
1122     { PCI_CLASS_SERIAL_CANBUS, "canbus", NULL },
1123     { 0xFF, NULL, NULL },
1124 };
1125 
1126 static const PCISubClass wrl_subclass[] = {
1127     { PCI_CLASS_WIRELESS_IRDA, "irda", NULL },
1128     { PCI_CLASS_WIRELESS_CIR, "consumer-ir", NULL },
1129     { PCI_CLASS_WIRELESS_RF_CONTROLLER, "rf-controller", NULL },
1130     { PCI_CLASS_WIRELESS_BLUETOOTH, "bluetooth", NULL },
1131     { PCI_CLASS_WIRELESS_BROADBAND, "broadband", NULL },
1132     { 0xFF, NULL, NULL },
1133 };
1134 
1135 static const PCISubClass sat_subclass[] = {
1136     { PCI_CLASS_SATELLITE_TV, "satellite-tv", NULL },
1137     { PCI_CLASS_SATELLITE_AUDIO, "satellite-audio", NULL },
1138     { PCI_CLASS_SATELLITE_VOICE, "satellite-voice", NULL },
1139     { PCI_CLASS_SATELLITE_DATA, "satellite-data", NULL },
1140     { 0xFF, NULL, NULL },
1141 };
1142 
1143 static const PCISubClass crypt_subclass[] = {
1144     { PCI_CLASS_CRYPT_NETWORK, "network-encryption", NULL },
1145     { PCI_CLASS_CRYPT_ENTERTAINMENT,
1146       "entertainment-encryption", NULL },
1147     { 0xFF, NULL, NULL },
1148 };
1149 
1150 static const PCISubClass spc_subclass[] = {
1151     { PCI_CLASS_SP_DPIO, "dpio", NULL },
1152     { PCI_CLASS_SP_PERF, "counter", NULL },
1153     { PCI_CLASS_SP_SYNCH, "measurement", NULL },
1154     { PCI_CLASS_SP_MANAGEMENT, "management-card", NULL },
1155     { 0xFF, NULL, NULL },
1156 };
1157 
1158 static const PCIClass pci_classes[] = {
1159     { "legacy-device", undef_subclass },
1160     { "mass-storage",  mass_subclass },
1161     { "network", net_subclass },
1162     { "display", displ_subclass, },
1163     { "multimedia-device", media_subclass },
1164     { "memory-controller", mem_subclass },
1165     { "unknown-bridge", bridg_subclass },
1166     { "communication-controller", comm_subclass},
1167     { "system-peripheral", sys_subclass },
1168     { "input-controller", inp_subclass },
1169     { "docking-station", dock_subclass },
1170     { "cpu", cpu_subclass },
1171     { "serial-bus", ser_subclass },
1172     { "wireless-controller", wrl_subclass },
1173     { "intelligent-io", NULL },
1174     { "satellite-device", sat_subclass },
1175     { "encryption", crypt_subclass },
1176     { "data-processing-controller", spc_subclass },
1177 };
1178 
1179 static const char *dt_name_from_class(uint8_t class, uint8_t subclass,
1180                                       uint8_t iface)
1181 {
1182     const PCIClass *pclass;
1183     const PCISubClass *psubclass;
1184     const PCIIFace *piface;
1185     const char *name;
1186 
1187     if (class >= ARRAY_SIZE(pci_classes)) {
1188         return "pci";
1189     }
1190 
1191     pclass = pci_classes + class;
1192     name = pclass->name;
1193 
1194     if (pclass->subc == NULL) {
1195         return name;
1196     }
1197 
1198     psubclass = pclass->subc;
1199     while ((psubclass->subclass & 0xff) != 0xff) {
1200         if ((psubclass->subclass & 0xff) == subclass) {
1201             name = psubclass->name;
1202             break;
1203         }
1204         psubclass++;
1205     }
1206 
1207     piface = psubclass->iface;
1208     if (piface == NULL) {
1209         return name;
1210     }
1211     while ((piface->iface & 0xff) != 0xff) {
1212         if ((piface->iface & 0xff) == iface) {
1213             name = piface->name;
1214             break;
1215         }
1216         piface++;
1217     }
1218 
1219     return name;
1220 }
1221 
1222 /*
1223  * DRC helper functions
1224  */
1225 
1226 static uint32_t drc_id_from_devfn(SpaprPhbState *phb,
1227                                   uint8_t chassis, int32_t devfn)
1228 {
1229     return (phb->index << 16) | (chassis << 8) | devfn;
1230 }
1231 
1232 static SpaprDrc *drc_from_devfn(SpaprPhbState *phb,
1233                                 uint8_t chassis, int32_t devfn)
1234 {
1235     return spapr_drc_by_id(TYPE_SPAPR_DRC_PCI,
1236                            drc_id_from_devfn(phb, chassis, devfn));
1237 }
1238 
1239 static uint8_t chassis_from_bus(PCIBus *bus, Error **errp)
1240 {
1241     if (pci_bus_is_root(bus)) {
1242         return 0;
1243     } else {
1244         PCIDevice *bridge = pci_bridge_get_device(bus);
1245 
1246         return object_property_get_uint(OBJECT(bridge), "chassis_nr", errp);
1247     }
1248 }
1249 
1250 static SpaprDrc *drc_from_dev(SpaprPhbState *phb, PCIDevice *dev)
1251 {
1252     Error *local_err = NULL;
1253     uint8_t chassis = chassis_from_bus(pci_get_bus(dev), &local_err);
1254 
1255     if (local_err) {
1256         error_report_err(local_err);
1257         return NULL;
1258     }
1259 
1260     return drc_from_devfn(phb, chassis, dev->devfn);
1261 }
1262 
1263 static void add_drcs(SpaprPhbState *phb, PCIBus *bus, Error **errp)
1264 {
1265     Object *owner;
1266     int i;
1267     uint8_t chassis;
1268     Error *local_err = NULL;
1269 
1270     if (!phb->dr_enabled) {
1271         return;
1272     }
1273 
1274     chassis = chassis_from_bus(bus, &local_err);
1275     if (local_err) {
1276         error_propagate(errp, local_err);
1277         return;
1278     }
1279 
1280     if (pci_bus_is_root(bus)) {
1281         owner = OBJECT(phb);
1282     } else {
1283         owner = OBJECT(pci_bridge_get_device(bus));
1284     }
1285 
1286     for (i = 0; i < PCI_SLOT_MAX * PCI_FUNC_MAX; i++) {
1287         spapr_dr_connector_new(owner, TYPE_SPAPR_DRC_PCI,
1288                                drc_id_from_devfn(phb, chassis, i));
1289     }
1290 }
1291 
1292 static void remove_drcs(SpaprPhbState *phb, PCIBus *bus, Error **errp)
1293 {
1294     int i;
1295     uint8_t chassis;
1296     Error *local_err = NULL;
1297 
1298     if (!phb->dr_enabled) {
1299         return;
1300     }
1301 
1302     chassis = chassis_from_bus(bus, &local_err);
1303     if (local_err) {
1304         error_propagate(errp, local_err);
1305         return;
1306     }
1307 
1308     for (i = PCI_SLOT_MAX * PCI_FUNC_MAX - 1; i >= 0; i--) {
1309         SpaprDrc *drc = drc_from_devfn(phb, chassis, i);
1310 
1311         if (drc) {
1312             object_unparent(OBJECT(drc));
1313         }
1314     }
1315 }
1316 
1317 typedef struct PciWalkFdt {
1318     void *fdt;
1319     int offset;
1320     SpaprPhbState *sphb;
1321     int err;
1322 } PciWalkFdt;
1323 
1324 static int spapr_dt_pci_device(SpaprPhbState *sphb, PCIDevice *dev,
1325                                void *fdt, int parent_offset);
1326 
1327 static void spapr_dt_pci_device_cb(PCIBus *bus, PCIDevice *pdev,
1328                                    void *opaque)
1329 {
1330     PciWalkFdt *p = opaque;
1331     int err;
1332 
1333     if (p->err) {
1334         /* Something's already broken, don't keep going */
1335         return;
1336     }
1337 
1338     err = spapr_dt_pci_device(p->sphb, pdev, p->fdt, p->offset);
1339     if (err < 0) {
1340         p->err = err;
1341     }
1342 }
1343 
1344 /* Augment PCI device node with bridge specific information */
1345 static int spapr_dt_pci_bus(SpaprPhbState *sphb, PCIBus *bus,
1346                                void *fdt, int offset)
1347 {
1348     Object *owner;
1349     PciWalkFdt cbinfo = {
1350         .fdt = fdt,
1351         .offset = offset,
1352         .sphb = sphb,
1353         .err = 0,
1354     };
1355     int ret;
1356 
1357     _FDT(fdt_setprop_cell(fdt, offset, "#address-cells",
1358                           RESOURCE_CELLS_ADDRESS));
1359     _FDT(fdt_setprop_cell(fdt, offset, "#size-cells",
1360                           RESOURCE_CELLS_SIZE));
1361 
1362     assert(bus);
1363     pci_for_each_device_reverse(bus, pci_bus_num(bus),
1364                                 spapr_dt_pci_device_cb, &cbinfo);
1365     if (cbinfo.err) {
1366         return cbinfo.err;
1367     }
1368 
1369     if (pci_bus_is_root(bus)) {
1370         owner = OBJECT(sphb);
1371     } else {
1372         owner = OBJECT(pci_bridge_get_device(bus));
1373     }
1374 
1375     ret = spapr_dt_drc(fdt, offset, owner,
1376                        SPAPR_DR_CONNECTOR_TYPE_PCI);
1377     if (ret) {
1378         return ret;
1379     }
1380 
1381     return offset;
1382 }
1383 
1384 /* create OF node for pci device and required OF DT properties */
1385 static int spapr_dt_pci_device(SpaprPhbState *sphb, PCIDevice *dev,
1386                                void *fdt, int parent_offset)
1387 {
1388     int offset;
1389     const gchar *basename;
1390     gchar *nodename;
1391     int slot = PCI_SLOT(dev->devfn);
1392     int func = PCI_FUNC(dev->devfn);
1393     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
1394     ResourceProps rp;
1395     SpaprDrc *drc = drc_from_dev(sphb, dev);
1396     uint32_t vendor_id = pci_default_read_config(dev, PCI_VENDOR_ID, 2);
1397     uint32_t device_id = pci_default_read_config(dev, PCI_DEVICE_ID, 2);
1398     uint32_t revision_id = pci_default_read_config(dev, PCI_REVISION_ID, 1);
1399     uint32_t ccode = pci_default_read_config(dev, PCI_CLASS_PROG, 3);
1400     uint32_t irq_pin = pci_default_read_config(dev, PCI_INTERRUPT_PIN, 1);
1401     uint32_t subsystem_id = pci_default_read_config(dev, PCI_SUBSYSTEM_ID, 2);
1402     uint32_t subsystem_vendor_id =
1403         pci_default_read_config(dev, PCI_SUBSYSTEM_VENDOR_ID, 2);
1404     uint32_t cache_line_size =
1405         pci_default_read_config(dev, PCI_CACHE_LINE_SIZE, 1);
1406     uint32_t pci_status = pci_default_read_config(dev, PCI_STATUS, 2);
1407     gchar *loc_code;
1408 
1409     basename = dt_name_from_class((ccode >> 16) & 0xff, (ccode >> 8) & 0xff,
1410                                   ccode & 0xff);
1411 
1412     if (func != 0) {
1413         nodename = g_strdup_printf("%s@%x,%x", basename, slot, func);
1414     } else {
1415         nodename = g_strdup_printf("%s@%x", basename, slot);
1416     }
1417 
1418     _FDT(offset = fdt_add_subnode(fdt, parent_offset, nodename));
1419 
1420     g_free(nodename);
1421 
1422     /* in accordance with PAPR+ v2.7 13.6.3, Table 181 */
1423     _FDT(fdt_setprop_cell(fdt, offset, "vendor-id", vendor_id));
1424     _FDT(fdt_setprop_cell(fdt, offset, "device-id", device_id));
1425     _FDT(fdt_setprop_cell(fdt, offset, "revision-id", revision_id));
1426 
1427     _FDT(fdt_setprop_cell(fdt, offset, "class-code", ccode));
1428     if (irq_pin) {
1429         _FDT(fdt_setprop_cell(fdt, offset, "interrupts", irq_pin));
1430     }
1431 
1432     if (subsystem_id) {
1433         _FDT(fdt_setprop_cell(fdt, offset, "subsystem-id", subsystem_id));
1434     }
1435 
1436     if (subsystem_vendor_id) {
1437         _FDT(fdt_setprop_cell(fdt, offset, "subsystem-vendor-id",
1438                               subsystem_vendor_id));
1439     }
1440 
1441     _FDT(fdt_setprop_cell(fdt, offset, "cache-line-size", cache_line_size));
1442 
1443 
1444     /* the following fdt cells are masked off the pci status register */
1445     _FDT(fdt_setprop_cell(fdt, offset, "devsel-speed",
1446                           PCI_STATUS_DEVSEL_MASK & pci_status));
1447 
1448     if (pci_status & PCI_STATUS_FAST_BACK) {
1449         _FDT(fdt_setprop(fdt, offset, "fast-back-to-back", NULL, 0));
1450     }
1451     if (pci_status & PCI_STATUS_66MHZ) {
1452         _FDT(fdt_setprop(fdt, offset, "66mhz-capable", NULL, 0));
1453     }
1454     if (pci_status & PCI_STATUS_UDF) {
1455         _FDT(fdt_setprop(fdt, offset, "udf-supported", NULL, 0));
1456     }
1457 
1458     loc_code = spapr_phb_get_loc_code(sphb, dev);
1459     _FDT(fdt_setprop_string(fdt, offset, "ibm,loc-code", loc_code));
1460     g_free(loc_code);
1461 
1462     if (drc) {
1463         _FDT(fdt_setprop_cell(fdt, offset, "ibm,my-drc-index",
1464                               spapr_drc_index(drc)));
1465     }
1466 
1467     if (msi_present(dev)) {
1468         uint32_t max_msi = msi_nr_vectors_allocated(dev);
1469         if (max_msi) {
1470             _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi", max_msi));
1471         }
1472     }
1473     if (msix_present(dev)) {
1474         uint32_t max_msix = dev->msix_entries_nr;
1475         if (max_msix) {
1476             _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi-x", max_msix));
1477         }
1478     }
1479 
1480     populate_resource_props(dev, &rp);
1481     _FDT(fdt_setprop(fdt, offset, "reg", (uint8_t *)rp.reg, rp.reg_len));
1482     _FDT(fdt_setprop(fdt, offset, "assigned-addresses",
1483                      (uint8_t *)rp.assigned, rp.assigned_len));
1484 
1485     if (sphb->pcie_ecs && pci_is_express(dev)) {
1486         _FDT(fdt_setprop_cell(fdt, offset, "ibm,pci-config-space-type", 0x1));
1487     }
1488 
1489     spapr_phb_nvgpu_populate_pcidev_dt(dev, fdt, offset, sphb);
1490 
1491     if (!pc->is_bridge) {
1492         /* Properties only for non-bridges */
1493         uint32_t min_grant = pci_default_read_config(dev, PCI_MIN_GNT, 1);
1494         uint32_t max_latency = pci_default_read_config(dev, PCI_MAX_LAT, 1);
1495         _FDT(fdt_setprop_cell(fdt, offset, "min-grant", min_grant));
1496         _FDT(fdt_setprop_cell(fdt, offset, "max-latency", max_latency));
1497         return offset;
1498     } else {
1499         PCIBus *sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(dev));
1500 
1501         return spapr_dt_pci_bus(sphb, sec_bus, fdt, offset);
1502     }
1503 }
1504 
1505 /* Callback to be called during DRC release. */
1506 void spapr_phb_remove_pci_device_cb(DeviceState *dev)
1507 {
1508     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
1509 
1510     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
1511     object_unparent(OBJECT(dev));
1512 }
1513 
1514 int spapr_pci_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
1515                           void *fdt, int *fdt_start_offset, Error **errp)
1516 {
1517     HotplugHandler *plug_handler = qdev_get_hotplug_handler(drc->dev);
1518     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(plug_handler);
1519     PCIDevice *pdev = PCI_DEVICE(drc->dev);
1520 
1521     *fdt_start_offset = spapr_dt_pci_device(sphb, pdev, fdt, 0);
1522     return 0;
1523 }
1524 
1525 static void spapr_pci_bridge_plug(SpaprPhbState *phb,
1526                                   PCIBridge *bridge,
1527                                   Error **errp)
1528 {
1529     Error *local_err = NULL;
1530     PCIBus *bus = pci_bridge_get_sec_bus(bridge);
1531 
1532     add_drcs(phb, bus, &local_err);
1533     if (local_err) {
1534         error_propagate(errp, local_err);
1535         return;
1536     }
1537 }
1538 
1539 static void spapr_pci_plug(HotplugHandler *plug_handler,
1540                            DeviceState *plugged_dev, Error **errp)
1541 {
1542     SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1543     PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1544     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(plugged_dev);
1545     SpaprDrc *drc = drc_from_dev(phb, pdev);
1546     Error *local_err = NULL;
1547     PCIBus *bus = PCI_BUS(qdev_get_parent_bus(DEVICE(pdev)));
1548     uint32_t slotnr = PCI_SLOT(pdev->devfn);
1549 
1550     /* if DR is disabled we don't need to do anything in the case of
1551      * hotplug or coldplug callbacks
1552      */
1553     if (!phb->dr_enabled) {
1554         /* if this is a hotplug operation initiated by the user
1555          * we need to let them know it's not enabled
1556          */
1557         if (plugged_dev->hotplugged) {
1558             error_setg(&local_err, QERR_BUS_NO_HOTPLUG,
1559                        object_get_typename(OBJECT(phb)));
1560         }
1561         goto out;
1562     }
1563 
1564     g_assert(drc);
1565 
1566     if (pc->is_bridge) {
1567         spapr_pci_bridge_plug(phb, PCI_BRIDGE(plugged_dev), &local_err);
1568         if (local_err) {
1569             error_propagate(errp, local_err);
1570             return;
1571         }
1572     }
1573 
1574     /* Following the QEMU convention used for PCIe multifunction
1575      * hotplug, we do not allow functions to be hotplugged to a
1576      * slot that already has function 0 present
1577      */
1578     if (plugged_dev->hotplugged && bus->devices[PCI_DEVFN(slotnr, 0)] &&
1579         PCI_FUNC(pdev->devfn) != 0) {
1580         error_setg(&local_err, "PCI: slot %d function 0 already ocuppied by %s,"
1581                    " additional functions can no longer be exposed to guest.",
1582                    slotnr, bus->devices[PCI_DEVFN(slotnr, 0)]->name);
1583         goto out;
1584     }
1585 
1586     spapr_drc_attach(drc, DEVICE(pdev), &local_err);
1587     if (local_err) {
1588         goto out;
1589     }
1590 
1591     /* If this is function 0, signal hotplug for all the device functions.
1592      * Otherwise defer sending the hotplug event.
1593      */
1594     if (!spapr_drc_hotplugged(plugged_dev)) {
1595         spapr_drc_reset(drc);
1596     } else if (PCI_FUNC(pdev->devfn) == 0) {
1597         int i;
1598         uint8_t chassis = chassis_from_bus(pci_get_bus(pdev), &local_err);
1599 
1600         if (local_err) {
1601             error_propagate(errp, local_err);
1602             return;
1603         }
1604 
1605         for (i = 0; i < 8; i++) {
1606             SpaprDrc *func_drc;
1607             SpaprDrcClass *func_drck;
1608             SpaprDREntitySense state;
1609 
1610             func_drc = drc_from_devfn(phb, chassis, PCI_DEVFN(slotnr, i));
1611             func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1612             state = func_drck->dr_entity_sense(func_drc);
1613 
1614             if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) {
1615                 spapr_hotplug_req_add_by_index(func_drc);
1616             }
1617         }
1618     }
1619 
1620 out:
1621     error_propagate(errp, local_err);
1622 }
1623 
1624 static void spapr_pci_bridge_unplug(SpaprPhbState *phb,
1625                                     PCIBridge *bridge,
1626                                     Error **errp)
1627 {
1628     Error *local_err = NULL;
1629     PCIBus *bus = pci_bridge_get_sec_bus(bridge);
1630 
1631     remove_drcs(phb, bus, &local_err);
1632     if (local_err) {
1633         error_propagate(errp, local_err);
1634         return;
1635     }
1636 }
1637 
1638 static void spapr_pci_unplug(HotplugHandler *plug_handler,
1639                              DeviceState *plugged_dev, Error **errp)
1640 {
1641     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(plugged_dev);
1642     SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1643 
1644     /* some version guests do not wait for completion of a device
1645      * cleanup (generally done asynchronously by the kernel) before
1646      * signaling to QEMU that the device is safe, but instead sleep
1647      * for some 'safe' period of time. unfortunately on a busy host
1648      * this sleep isn't guaranteed to be long enough, resulting in
1649      * bad things like IRQ lines being left asserted during final
1650      * device removal. to deal with this we call reset just prior
1651      * to finalizing the device, which will put the device back into
1652      * an 'idle' state, as the device cleanup code expects.
1653      */
1654     pci_device_reset(PCI_DEVICE(plugged_dev));
1655 
1656     if (pc->is_bridge) {
1657         Error *local_err = NULL;
1658         spapr_pci_bridge_unplug(phb, PCI_BRIDGE(plugged_dev), &local_err);
1659         if (local_err) {
1660             error_propagate(errp, local_err);
1661         }
1662         return;
1663     }
1664 
1665     object_property_set_bool(OBJECT(plugged_dev), false, "realized", NULL);
1666 }
1667 
1668 static void spapr_pci_unplug_request(HotplugHandler *plug_handler,
1669                                      DeviceState *plugged_dev, Error **errp)
1670 {
1671     SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1672     PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1673     SpaprDrc *drc = drc_from_dev(phb, pdev);
1674 
1675     if (!phb->dr_enabled) {
1676         error_setg(errp, QERR_BUS_NO_HOTPLUG,
1677                    object_get_typename(OBJECT(phb)));
1678         return;
1679     }
1680 
1681     g_assert(drc);
1682     g_assert(drc->dev == plugged_dev);
1683 
1684     if (!spapr_drc_unplug_requested(drc)) {
1685         PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(plugged_dev);
1686         uint32_t slotnr = PCI_SLOT(pdev->devfn);
1687         SpaprDrc *func_drc;
1688         SpaprDrcClass *func_drck;
1689         SpaprDREntitySense state;
1690         int i;
1691         Error *local_err = NULL;
1692         uint8_t chassis = chassis_from_bus(pci_get_bus(pdev), &local_err);
1693 
1694         if (local_err) {
1695             error_propagate(errp, local_err);
1696             return;
1697         }
1698 
1699         if (pc->is_bridge) {
1700             error_setg(errp, "PCI: Hot unplug of PCI bridges not supported");
1701         }
1702 
1703         /* ensure any other present functions are pending unplug */
1704         if (PCI_FUNC(pdev->devfn) == 0) {
1705             for (i = 1; i < 8; i++) {
1706                 func_drc = drc_from_devfn(phb, chassis, PCI_DEVFN(slotnr, i));
1707                 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1708                 state = func_drck->dr_entity_sense(func_drc);
1709                 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT
1710                     && !spapr_drc_unplug_requested(func_drc)) {
1711                     error_setg(errp,
1712                                "PCI: slot %d, function %d still present. "
1713                                "Must unplug all non-0 functions first.",
1714                                slotnr, i);
1715                     return;
1716                 }
1717             }
1718         }
1719 
1720         spapr_drc_detach(drc);
1721 
1722         /* if this isn't func 0, defer unplug event. otherwise signal removal
1723          * for all present functions
1724          */
1725         if (PCI_FUNC(pdev->devfn) == 0) {
1726             for (i = 7; i >= 0; i--) {
1727                 func_drc = drc_from_devfn(phb, chassis, PCI_DEVFN(slotnr, i));
1728                 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1729                 state = func_drck->dr_entity_sense(func_drc);
1730                 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) {
1731                     spapr_hotplug_req_remove_by_index(func_drc);
1732                 }
1733             }
1734         }
1735     }
1736 }
1737 
1738 static void spapr_phb_finalizefn(Object *obj)
1739 {
1740     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(obj);
1741 
1742     g_free(sphb->dtbusname);
1743     sphb->dtbusname = NULL;
1744 }
1745 
1746 static void spapr_phb_unrealize(DeviceState *dev, Error **errp)
1747 {
1748     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
1749     SysBusDevice *s = SYS_BUS_DEVICE(dev);
1750     PCIHostState *phb = PCI_HOST_BRIDGE(s);
1751     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(phb);
1752     SpaprTceTable *tcet;
1753     int i;
1754     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
1755     Error *local_err = NULL;
1756 
1757     spapr_phb_nvgpu_free(sphb);
1758 
1759     if (sphb->msi) {
1760         g_hash_table_unref(sphb->msi);
1761         sphb->msi = NULL;
1762     }
1763 
1764     /*
1765      * Remove IO/MMIO subregions and aliases, rest should get cleaned
1766      * via PHB's unrealize->object_finalize
1767      */
1768     for (i = windows_supported - 1; i >= 0; i--) {
1769         tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[i]);
1770         if (tcet) {
1771             memory_region_del_subregion(&sphb->iommu_root,
1772                                         spapr_tce_get_iommu(tcet));
1773         }
1774     }
1775 
1776     remove_drcs(sphb, phb->bus, &local_err);
1777     if (local_err) {
1778         error_propagate(errp, local_err);
1779         return;
1780     }
1781 
1782     for (i = PCI_NUM_PINS - 1; i >= 0; i--) {
1783         if (sphb->lsi_table[i].irq) {
1784             spapr_irq_free(spapr, sphb->lsi_table[i].irq, 1);
1785             sphb->lsi_table[i].irq = 0;
1786         }
1787     }
1788 
1789     QLIST_REMOVE(sphb, list);
1790 
1791     memory_region_del_subregion(&sphb->iommu_root, &sphb->msiwindow);
1792 
1793     /*
1794      * An attached PCI device may have memory listeners, eg. VFIO PCI. We have
1795      * unmapped all sections. Remove the listeners now, before destroying the
1796      * address space.
1797      */
1798     address_space_remove_listeners(&sphb->iommu_as);
1799     address_space_destroy(&sphb->iommu_as);
1800 
1801     qbus_set_hotplug_handler(BUS(phb->bus), NULL, &error_abort);
1802     pci_unregister_root_bus(phb->bus);
1803 
1804     memory_region_del_subregion(get_system_memory(), &sphb->iowindow);
1805     if (sphb->mem64_win_pciaddr != (hwaddr)-1) {
1806         memory_region_del_subregion(get_system_memory(), &sphb->mem64window);
1807     }
1808     memory_region_del_subregion(get_system_memory(), &sphb->mem32window);
1809 }
1810 
1811 static void spapr_phb_realize(DeviceState *dev, Error **errp)
1812 {
1813     /* We don't use SPAPR_MACHINE() in order to exit gracefully if the user
1814      * tries to add a sPAPR PHB to a non-pseries machine.
1815      */
1816     SpaprMachineState *spapr =
1817         (SpaprMachineState *) object_dynamic_cast(qdev_get_machine(),
1818                                                   TYPE_SPAPR_MACHINE);
1819     SpaprMachineClass *smc = spapr ? SPAPR_MACHINE_GET_CLASS(spapr) : NULL;
1820     SysBusDevice *s = SYS_BUS_DEVICE(dev);
1821     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(s);
1822     PCIHostState *phb = PCI_HOST_BRIDGE(s);
1823     char *namebuf;
1824     int i;
1825     PCIBus *bus;
1826     uint64_t msi_window_size = 4096;
1827     SpaprTceTable *tcet;
1828     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
1829     Error *local_err = NULL;
1830 
1831     if (!spapr) {
1832         error_setg(errp, TYPE_SPAPR_PCI_HOST_BRIDGE " needs a pseries machine");
1833         return;
1834     }
1835 
1836     assert(sphb->index != (uint32_t)-1); /* checked in spapr_phb_pre_plug() */
1837 
1838     if (sphb->mem64_win_size != 0) {
1839         if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) {
1840             error_setg(errp, "32-bit memory window of size 0x%"HWADDR_PRIx
1841                        " (max 2 GiB)", sphb->mem_win_size);
1842             return;
1843         }
1844 
1845         /* 64-bit window defaults to identity mapping */
1846         sphb->mem64_win_pciaddr = sphb->mem64_win_addr;
1847     } else if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) {
1848         /*
1849          * For compatibility with old configuration, if no 64-bit MMIO
1850          * window is specified, but the ordinary (32-bit) memory
1851          * window is specified as > 2GiB, we treat it as a 2GiB 32-bit
1852          * window, with a 64-bit MMIO window following on immediately
1853          * afterwards
1854          */
1855         sphb->mem64_win_size = sphb->mem_win_size - SPAPR_PCI_MEM32_WIN_SIZE;
1856         sphb->mem64_win_addr = sphb->mem_win_addr + SPAPR_PCI_MEM32_WIN_SIZE;
1857         sphb->mem64_win_pciaddr =
1858             SPAPR_PCI_MEM_WIN_BUS_OFFSET + SPAPR_PCI_MEM32_WIN_SIZE;
1859         sphb->mem_win_size = SPAPR_PCI_MEM32_WIN_SIZE;
1860     }
1861 
1862     if (spapr_pci_find_phb(spapr, sphb->buid)) {
1863         SpaprPhbState *s;
1864 
1865         error_setg(errp, "PCI host bridges must have unique indexes");
1866         error_append_hint(errp, "The following indexes are already in use:");
1867         QLIST_FOREACH(s, &spapr->phbs, list) {
1868             error_append_hint(errp, " %d", s->index);
1869         }
1870         error_append_hint(errp, "\nTry another value for the index property\n");
1871         return;
1872     }
1873 
1874     if (sphb->numa_node != -1 &&
1875         (sphb->numa_node >= MAX_NODES || !numa_info[sphb->numa_node].present)) {
1876         error_setg(errp, "Invalid NUMA node ID for PCI host bridge");
1877         return;
1878     }
1879 
1880     sphb->dtbusname = g_strdup_printf("pci@%" PRIx64, sphb->buid);
1881 
1882     /* Initialize memory regions */
1883     namebuf = g_strdup_printf("%s.mmio", sphb->dtbusname);
1884     memory_region_init(&sphb->memspace, OBJECT(sphb), namebuf, UINT64_MAX);
1885     g_free(namebuf);
1886 
1887     namebuf = g_strdup_printf("%s.mmio32-alias", sphb->dtbusname);
1888     memory_region_init_alias(&sphb->mem32window, OBJECT(sphb),
1889                              namebuf, &sphb->memspace,
1890                              SPAPR_PCI_MEM_WIN_BUS_OFFSET, sphb->mem_win_size);
1891     g_free(namebuf);
1892     memory_region_add_subregion(get_system_memory(), sphb->mem_win_addr,
1893                                 &sphb->mem32window);
1894 
1895     if (sphb->mem64_win_size != 0) {
1896         namebuf = g_strdup_printf("%s.mmio64-alias", sphb->dtbusname);
1897         memory_region_init_alias(&sphb->mem64window, OBJECT(sphb),
1898                                  namebuf, &sphb->memspace,
1899                                  sphb->mem64_win_pciaddr, sphb->mem64_win_size);
1900         g_free(namebuf);
1901 
1902         memory_region_add_subregion(get_system_memory(),
1903                                     sphb->mem64_win_addr,
1904                                     &sphb->mem64window);
1905     }
1906 
1907     /* Initialize IO regions */
1908     namebuf = g_strdup_printf("%s.io", sphb->dtbusname);
1909     memory_region_init(&sphb->iospace, OBJECT(sphb),
1910                        namebuf, SPAPR_PCI_IO_WIN_SIZE);
1911     g_free(namebuf);
1912 
1913     namebuf = g_strdup_printf("%s.io-alias", sphb->dtbusname);
1914     memory_region_init_alias(&sphb->iowindow, OBJECT(sphb), namebuf,
1915                              &sphb->iospace, 0, SPAPR_PCI_IO_WIN_SIZE);
1916     g_free(namebuf);
1917     memory_region_add_subregion(get_system_memory(), sphb->io_win_addr,
1918                                 &sphb->iowindow);
1919 
1920     bus = pci_register_root_bus(dev, NULL,
1921                                 pci_spapr_set_irq, pci_swizzle_map_irq_fn, sphb,
1922                                 &sphb->memspace, &sphb->iospace,
1923                                 PCI_DEVFN(0, 0), PCI_NUM_PINS,
1924                                 TYPE_PCI_BUS);
1925 
1926     /*
1927      * Despite resembling a vanilla PCI bus in most ways, the PAPR
1928      * para-virtualized PCI bus *does* permit PCI-E extended config
1929      * space access
1930      */
1931     if (sphb->pcie_ecs) {
1932         bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
1933     }
1934     phb->bus = bus;
1935     qbus_set_hotplug_handler(BUS(phb->bus), OBJECT(sphb), NULL);
1936 
1937     /*
1938      * Initialize PHB address space.
1939      * By default there will be at least one subregion for default
1940      * 32bit DMA window.
1941      * Later the guest might want to create another DMA window
1942      * which will become another memory subregion.
1943      */
1944     namebuf = g_strdup_printf("%s.iommu-root", sphb->dtbusname);
1945     memory_region_init(&sphb->iommu_root, OBJECT(sphb),
1946                        namebuf, UINT64_MAX);
1947     g_free(namebuf);
1948     address_space_init(&sphb->iommu_as, &sphb->iommu_root,
1949                        sphb->dtbusname);
1950 
1951     /*
1952      * As MSI/MSIX interrupts trigger by writing at MSI/MSIX vectors,
1953      * we need to allocate some memory to catch those writes coming
1954      * from msi_notify()/msix_notify().
1955      * As MSIMessage:addr is going to be the same and MSIMessage:data
1956      * is going to be a VIRQ number, 4 bytes of the MSI MR will only
1957      * be used.
1958      *
1959      * For KVM we want to ensure that this memory is a full page so that
1960      * our memory slot is of page size granularity.
1961      */
1962     if (kvm_enabled()) {
1963         msi_window_size = getpagesize();
1964     }
1965 
1966     memory_region_init_io(&sphb->msiwindow, OBJECT(sphb), &spapr_msi_ops, spapr,
1967                           "msi", msi_window_size);
1968     memory_region_add_subregion(&sphb->iommu_root, SPAPR_PCI_MSI_WINDOW,
1969                                 &sphb->msiwindow);
1970 
1971     pci_setup_iommu(bus, spapr_pci_dma_iommu, sphb);
1972 
1973     pci_bus_set_route_irq_fn(bus, spapr_route_intx_pin_to_irq);
1974 
1975     QLIST_INSERT_HEAD(&spapr->phbs, sphb, list);
1976 
1977     /* Initialize the LSI table */
1978     for (i = 0; i < PCI_NUM_PINS; i++) {
1979         uint32_t irq = SPAPR_IRQ_PCI_LSI + sphb->index * PCI_NUM_PINS + i;
1980 
1981         if (smc->legacy_irq_allocation) {
1982             irq = spapr_irq_findone(spapr, &local_err);
1983             if (local_err) {
1984                 error_propagate_prepend(errp, local_err,
1985                                         "can't allocate LSIs: ");
1986                 /*
1987                  * Older machines will never support PHB hotplug, ie, this is an
1988                  * init only path and QEMU will terminate. No need to rollback.
1989                  */
1990                 return;
1991             }
1992         }
1993 
1994         spapr_irq_claim(spapr, irq, true, &local_err);
1995         if (local_err) {
1996             error_propagate_prepend(errp, local_err, "can't allocate LSIs: ");
1997             goto unrealize;
1998         }
1999 
2000         sphb->lsi_table[i].irq = irq;
2001     }
2002 
2003     /* allocate connectors for child PCI devices */
2004     add_drcs(sphb, phb->bus, &local_err);
2005     if (local_err) {
2006         error_propagate(errp, local_err);
2007         goto unrealize;
2008     }
2009 
2010     /* DMA setup */
2011     for (i = 0; i < windows_supported; ++i) {
2012         tcet = spapr_tce_new_table(DEVICE(sphb), sphb->dma_liobn[i]);
2013         if (!tcet) {
2014             error_setg(errp, "Creating window#%d failed for %s",
2015                        i, sphb->dtbusname);
2016             goto unrealize;
2017         }
2018         memory_region_add_subregion(&sphb->iommu_root, 0,
2019                                     spapr_tce_get_iommu(tcet));
2020     }
2021 
2022     sphb->msi = g_hash_table_new_full(g_int_hash, g_int_equal, g_free, g_free);
2023     return;
2024 
2025 unrealize:
2026     spapr_phb_unrealize(dev, NULL);
2027 }
2028 
2029 static int spapr_phb_children_reset(Object *child, void *opaque)
2030 {
2031     DeviceState *dev = (DeviceState *) object_dynamic_cast(child, TYPE_DEVICE);
2032 
2033     if (dev) {
2034         device_reset(dev);
2035     }
2036 
2037     return 0;
2038 }
2039 
2040 void spapr_phb_dma_reset(SpaprPhbState *sphb)
2041 {
2042     int i;
2043     SpaprTceTable *tcet;
2044 
2045     for (i = 0; i < SPAPR_PCI_DMA_MAX_WINDOWS; ++i) {
2046         tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[i]);
2047 
2048         if (tcet && tcet->nb_table) {
2049             spapr_tce_table_disable(tcet);
2050         }
2051     }
2052 
2053     /* Register default 32bit DMA window */
2054     tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[0]);
2055     spapr_tce_table_enable(tcet, SPAPR_TCE_PAGE_SHIFT, sphb->dma_win_addr,
2056                            sphb->dma_win_size >> SPAPR_TCE_PAGE_SHIFT);
2057 }
2058 
2059 static void spapr_phb_reset(DeviceState *qdev)
2060 {
2061     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(qdev);
2062     Error *errp = NULL;
2063 
2064     spapr_phb_dma_reset(sphb);
2065     spapr_phb_nvgpu_free(sphb);
2066     spapr_phb_nvgpu_setup(sphb, &errp);
2067     if (errp) {
2068         error_report_err(errp);
2069     }
2070 
2071     /* Reset the IOMMU state */
2072     object_child_foreach(OBJECT(qdev), spapr_phb_children_reset, NULL);
2073 
2074     if (spapr_phb_eeh_available(SPAPR_PCI_HOST_BRIDGE(qdev))) {
2075         spapr_phb_vfio_reset(qdev);
2076     }
2077 }
2078 
2079 static Property spapr_phb_properties[] = {
2080     DEFINE_PROP_UINT32("index", SpaprPhbState, index, -1),
2081     DEFINE_PROP_UINT64("mem_win_size", SpaprPhbState, mem_win_size,
2082                        SPAPR_PCI_MEM32_WIN_SIZE),
2083     DEFINE_PROP_UINT64("mem64_win_size", SpaprPhbState, mem64_win_size,
2084                        SPAPR_PCI_MEM64_WIN_SIZE),
2085     DEFINE_PROP_UINT64("io_win_size", SpaprPhbState, io_win_size,
2086                        SPAPR_PCI_IO_WIN_SIZE),
2087     DEFINE_PROP_BOOL("dynamic-reconfiguration", SpaprPhbState, dr_enabled,
2088                      true),
2089     /* Default DMA window is 0..1GB */
2090     DEFINE_PROP_UINT64("dma_win_addr", SpaprPhbState, dma_win_addr, 0),
2091     DEFINE_PROP_UINT64("dma_win_size", SpaprPhbState, dma_win_size, 0x40000000),
2092     DEFINE_PROP_UINT64("dma64_win_addr", SpaprPhbState, dma64_win_addr,
2093                        0x800000000000000ULL),
2094     DEFINE_PROP_BOOL("ddw", SpaprPhbState, ddw_enabled, true),
2095     DEFINE_PROP_UINT64("pgsz", SpaprPhbState, page_size_mask,
2096                        (1ULL << 12) | (1ULL << 16)),
2097     DEFINE_PROP_UINT32("numa_node", SpaprPhbState, numa_node, -1),
2098     DEFINE_PROP_BOOL("pre-2.8-migration", SpaprPhbState,
2099                      pre_2_8_migration, false),
2100     DEFINE_PROP_BOOL("pcie-extended-configuration-space", SpaprPhbState,
2101                      pcie_ecs, true),
2102     DEFINE_PROP_UINT64("gpa", SpaprPhbState, nv2_gpa_win_addr, 0),
2103     DEFINE_PROP_UINT64("atsd", SpaprPhbState, nv2_atsd_win_addr, 0),
2104     DEFINE_PROP_END_OF_LIST(),
2105 };
2106 
2107 static const VMStateDescription vmstate_spapr_pci_lsi = {
2108     .name = "spapr_pci/lsi",
2109     .version_id = 1,
2110     .minimum_version_id = 1,
2111     .fields = (VMStateField[]) {
2112         VMSTATE_UINT32_EQUAL(irq, struct spapr_pci_lsi, NULL),
2113 
2114         VMSTATE_END_OF_LIST()
2115     },
2116 };
2117 
2118 static const VMStateDescription vmstate_spapr_pci_msi = {
2119     .name = "spapr_pci/msi",
2120     .version_id = 1,
2121     .minimum_version_id = 1,
2122     .fields = (VMStateField []) {
2123         VMSTATE_UINT32(key, spapr_pci_msi_mig),
2124         VMSTATE_UINT32(value.first_irq, spapr_pci_msi_mig),
2125         VMSTATE_UINT32(value.num, spapr_pci_msi_mig),
2126         VMSTATE_END_OF_LIST()
2127     },
2128 };
2129 
2130 static int spapr_pci_pre_save(void *opaque)
2131 {
2132     SpaprPhbState *sphb = opaque;
2133     GHashTableIter iter;
2134     gpointer key, value;
2135     int i;
2136 
2137     if (sphb->pre_2_8_migration) {
2138         sphb->mig_liobn = sphb->dma_liobn[0];
2139         sphb->mig_mem_win_addr = sphb->mem_win_addr;
2140         sphb->mig_mem_win_size = sphb->mem_win_size;
2141         sphb->mig_io_win_addr = sphb->io_win_addr;
2142         sphb->mig_io_win_size = sphb->io_win_size;
2143 
2144         if ((sphb->mem64_win_size != 0)
2145             && (sphb->mem64_win_addr
2146                 == (sphb->mem_win_addr + sphb->mem_win_size))) {
2147             sphb->mig_mem_win_size += sphb->mem64_win_size;
2148         }
2149     }
2150 
2151     g_free(sphb->msi_devs);
2152     sphb->msi_devs = NULL;
2153     sphb->msi_devs_num = g_hash_table_size(sphb->msi);
2154     if (!sphb->msi_devs_num) {
2155         return 0;
2156     }
2157     sphb->msi_devs = g_new(spapr_pci_msi_mig, sphb->msi_devs_num);
2158 
2159     g_hash_table_iter_init(&iter, sphb->msi);
2160     for (i = 0; g_hash_table_iter_next(&iter, &key, &value); ++i) {
2161         sphb->msi_devs[i].key = *(uint32_t *) key;
2162         sphb->msi_devs[i].value = *(spapr_pci_msi *) value;
2163     }
2164 
2165     return 0;
2166 }
2167 
2168 static int spapr_pci_post_load(void *opaque, int version_id)
2169 {
2170     SpaprPhbState *sphb = opaque;
2171     gpointer key, value;
2172     int i;
2173 
2174     for (i = 0; i < sphb->msi_devs_num; ++i) {
2175         key = g_memdup(&sphb->msi_devs[i].key,
2176                        sizeof(sphb->msi_devs[i].key));
2177         value = g_memdup(&sphb->msi_devs[i].value,
2178                          sizeof(sphb->msi_devs[i].value));
2179         g_hash_table_insert(sphb->msi, key, value);
2180     }
2181     g_free(sphb->msi_devs);
2182     sphb->msi_devs = NULL;
2183     sphb->msi_devs_num = 0;
2184 
2185     return 0;
2186 }
2187 
2188 static bool pre_2_8_migration(void *opaque, int version_id)
2189 {
2190     SpaprPhbState *sphb = opaque;
2191 
2192     return sphb->pre_2_8_migration;
2193 }
2194 
2195 static const VMStateDescription vmstate_spapr_pci = {
2196     .name = "spapr_pci",
2197     .version_id = 2,
2198     .minimum_version_id = 2,
2199     .pre_save = spapr_pci_pre_save,
2200     .post_load = spapr_pci_post_load,
2201     .fields = (VMStateField[]) {
2202         VMSTATE_UINT64_EQUAL(buid, SpaprPhbState, NULL),
2203         VMSTATE_UINT32_TEST(mig_liobn, SpaprPhbState, pre_2_8_migration),
2204         VMSTATE_UINT64_TEST(mig_mem_win_addr, SpaprPhbState, pre_2_8_migration),
2205         VMSTATE_UINT64_TEST(mig_mem_win_size, SpaprPhbState, pre_2_8_migration),
2206         VMSTATE_UINT64_TEST(mig_io_win_addr, SpaprPhbState, pre_2_8_migration),
2207         VMSTATE_UINT64_TEST(mig_io_win_size, SpaprPhbState, pre_2_8_migration),
2208         VMSTATE_STRUCT_ARRAY(lsi_table, SpaprPhbState, PCI_NUM_PINS, 0,
2209                              vmstate_spapr_pci_lsi, struct spapr_pci_lsi),
2210         VMSTATE_INT32(msi_devs_num, SpaprPhbState),
2211         VMSTATE_STRUCT_VARRAY_ALLOC(msi_devs, SpaprPhbState, msi_devs_num, 0,
2212                                     vmstate_spapr_pci_msi, spapr_pci_msi_mig),
2213         VMSTATE_END_OF_LIST()
2214     },
2215 };
2216 
2217 static const char *spapr_phb_root_bus_path(PCIHostState *host_bridge,
2218                                            PCIBus *rootbus)
2219 {
2220     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(host_bridge);
2221 
2222     return sphb->dtbusname;
2223 }
2224 
2225 static void spapr_phb_class_init(ObjectClass *klass, void *data)
2226 {
2227     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(klass);
2228     DeviceClass *dc = DEVICE_CLASS(klass);
2229     HotplugHandlerClass *hp = HOTPLUG_HANDLER_CLASS(klass);
2230 
2231     hc->root_bus_path = spapr_phb_root_bus_path;
2232     dc->realize = spapr_phb_realize;
2233     dc->unrealize = spapr_phb_unrealize;
2234     dc->props = spapr_phb_properties;
2235     dc->reset = spapr_phb_reset;
2236     dc->vmsd = &vmstate_spapr_pci;
2237     /* Supported by TYPE_SPAPR_MACHINE */
2238     dc->user_creatable = true;
2239     set_bit(DEVICE_CATEGORY_BRIDGE, dc->categories);
2240     hp->plug = spapr_pci_plug;
2241     hp->unplug = spapr_pci_unplug;
2242     hp->unplug_request = spapr_pci_unplug_request;
2243 }
2244 
2245 static const TypeInfo spapr_phb_info = {
2246     .name          = TYPE_SPAPR_PCI_HOST_BRIDGE,
2247     .parent        = TYPE_PCI_HOST_BRIDGE,
2248     .instance_size = sizeof(SpaprPhbState),
2249     .instance_finalize = spapr_phb_finalizefn,
2250     .class_init    = spapr_phb_class_init,
2251     .interfaces    = (InterfaceInfo[]) {
2252         { TYPE_HOTPLUG_HANDLER },
2253         { }
2254     }
2255 };
2256 
2257 static void spapr_phb_pci_enumerate_bridge(PCIBus *bus, PCIDevice *pdev,
2258                                            void *opaque)
2259 {
2260     unsigned int *bus_no = opaque;
2261     PCIBus *sec_bus = NULL;
2262 
2263     if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) !=
2264          PCI_HEADER_TYPE_BRIDGE)) {
2265         return;
2266     }
2267 
2268     (*bus_no)++;
2269     pci_default_write_config(pdev, PCI_PRIMARY_BUS, pci_dev_bus_num(pdev), 1);
2270     pci_default_write_config(pdev, PCI_SECONDARY_BUS, *bus_no, 1);
2271     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
2272 
2273     sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev));
2274     if (!sec_bus) {
2275         return;
2276     }
2277 
2278     pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
2279                         spapr_phb_pci_enumerate_bridge, bus_no);
2280     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
2281 }
2282 
2283 static void spapr_phb_pci_enumerate(SpaprPhbState *phb)
2284 {
2285     PCIBus *bus = PCI_HOST_BRIDGE(phb)->bus;
2286     unsigned int bus_no = 0;
2287 
2288     pci_for_each_device(bus, pci_bus_num(bus),
2289                         spapr_phb_pci_enumerate_bridge,
2290                         &bus_no);
2291 
2292 }
2293 
2294 int spapr_dt_phb(SpaprPhbState *phb, uint32_t intc_phandle, void *fdt,
2295                  uint32_t nr_msis, int *node_offset)
2296 {
2297     int bus_off, i, j, ret;
2298     uint32_t bus_range[] = { cpu_to_be32(0), cpu_to_be32(0xff) };
2299     struct {
2300         uint32_t hi;
2301         uint64_t child;
2302         uint64_t parent;
2303         uint64_t size;
2304     } QEMU_PACKED ranges[] = {
2305         {
2306             cpu_to_be32(b_ss(1)), cpu_to_be64(0),
2307             cpu_to_be64(phb->io_win_addr),
2308             cpu_to_be64(memory_region_size(&phb->iospace)),
2309         },
2310         {
2311             cpu_to_be32(b_ss(2)), cpu_to_be64(SPAPR_PCI_MEM_WIN_BUS_OFFSET),
2312             cpu_to_be64(phb->mem_win_addr),
2313             cpu_to_be64(phb->mem_win_size),
2314         },
2315         {
2316             cpu_to_be32(b_ss(3)), cpu_to_be64(phb->mem64_win_pciaddr),
2317             cpu_to_be64(phb->mem64_win_addr),
2318             cpu_to_be64(phb->mem64_win_size),
2319         },
2320     };
2321     const unsigned sizeof_ranges =
2322         (phb->mem64_win_size ? 3 : 2) * sizeof(ranges[0]);
2323     uint64_t bus_reg[] = { cpu_to_be64(phb->buid), 0 };
2324     uint32_t interrupt_map_mask[] = {
2325         cpu_to_be32(b_ddddd(-1)|b_fff(0)), 0x0, 0x0, cpu_to_be32(-1)};
2326     uint32_t interrupt_map[PCI_SLOT_MAX * PCI_NUM_PINS][7];
2327     uint32_t ddw_applicable[] = {
2328         cpu_to_be32(RTAS_IBM_QUERY_PE_DMA_WINDOW),
2329         cpu_to_be32(RTAS_IBM_CREATE_PE_DMA_WINDOW),
2330         cpu_to_be32(RTAS_IBM_REMOVE_PE_DMA_WINDOW)
2331     };
2332     uint32_t ddw_extensions[] = {
2333         cpu_to_be32(1),
2334         cpu_to_be32(RTAS_IBM_RESET_PE_DMA_WINDOW)
2335     };
2336     uint32_t associativity[] = {cpu_to_be32(0x4),
2337                                 cpu_to_be32(0x0),
2338                                 cpu_to_be32(0x0),
2339                                 cpu_to_be32(0x0),
2340                                 cpu_to_be32(phb->numa_node)};
2341     SpaprTceTable *tcet;
2342     SpaprDrc *drc;
2343     Error *errp = NULL;
2344 
2345     /* Start populating the FDT */
2346     _FDT(bus_off = fdt_add_subnode(fdt, 0, phb->dtbusname));
2347     if (node_offset) {
2348         *node_offset = bus_off;
2349     }
2350 
2351     /* Write PHB properties */
2352     _FDT(fdt_setprop_string(fdt, bus_off, "device_type", "pci"));
2353     _FDT(fdt_setprop_string(fdt, bus_off, "compatible", "IBM,Logical_PHB"));
2354     _FDT(fdt_setprop_cell(fdt, bus_off, "#interrupt-cells", 0x1));
2355     _FDT(fdt_setprop(fdt, bus_off, "used-by-rtas", NULL, 0));
2356     _FDT(fdt_setprop(fdt, bus_off, "bus-range", &bus_range, sizeof(bus_range)));
2357     _FDT(fdt_setprop(fdt, bus_off, "ranges", &ranges, sizeof_ranges));
2358     _FDT(fdt_setprop(fdt, bus_off, "reg", &bus_reg, sizeof(bus_reg)));
2359     _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pci-config-space-type", 0x1));
2360     _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pe-total-#msi", nr_msis));
2361 
2362     /* Dynamic DMA window */
2363     if (phb->ddw_enabled) {
2364         _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-applicable", &ddw_applicable,
2365                          sizeof(ddw_applicable)));
2366         _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-extensions",
2367                          &ddw_extensions, sizeof(ddw_extensions)));
2368     }
2369 
2370     /* Advertise NUMA via ibm,associativity */
2371     if (phb->numa_node != -1) {
2372         _FDT(fdt_setprop(fdt, bus_off, "ibm,associativity", associativity,
2373                          sizeof(associativity)));
2374     }
2375 
2376     /* Build the interrupt-map, this must matches what is done
2377      * in pci_swizzle_map_irq_fn
2378      */
2379     _FDT(fdt_setprop(fdt, bus_off, "interrupt-map-mask",
2380                      &interrupt_map_mask, sizeof(interrupt_map_mask)));
2381     for (i = 0; i < PCI_SLOT_MAX; i++) {
2382         for (j = 0; j < PCI_NUM_PINS; j++) {
2383             uint32_t *irqmap = interrupt_map[i*PCI_NUM_PINS + j];
2384             int lsi_num = pci_swizzle(i, j);
2385 
2386             irqmap[0] = cpu_to_be32(b_ddddd(i)|b_fff(0));
2387             irqmap[1] = 0;
2388             irqmap[2] = 0;
2389             irqmap[3] = cpu_to_be32(j+1);
2390             irqmap[4] = cpu_to_be32(intc_phandle);
2391             spapr_dt_irq(&irqmap[5], phb->lsi_table[lsi_num].irq, true);
2392         }
2393     }
2394     /* Write interrupt map */
2395     _FDT(fdt_setprop(fdt, bus_off, "interrupt-map", &interrupt_map,
2396                      sizeof(interrupt_map)));
2397 
2398     tcet = spapr_tce_find_by_liobn(phb->dma_liobn[0]);
2399     if (!tcet) {
2400         return -1;
2401     }
2402     spapr_dma_dt(fdt, bus_off, "ibm,dma-window",
2403                  tcet->liobn, tcet->bus_offset,
2404                  tcet->nb_table << tcet->page_shift);
2405 
2406     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, phb->index);
2407     if (drc) {
2408         uint32_t drc_index = cpu_to_be32(spapr_drc_index(drc));
2409 
2410         _FDT(fdt_setprop(fdt, bus_off, "ibm,my-drc-index", &drc_index,
2411                          sizeof(drc_index)));
2412     }
2413 
2414     /* Walk the bridges and program the bus numbers*/
2415     spapr_phb_pci_enumerate(phb);
2416     _FDT(fdt_setprop_cell(fdt, bus_off, "qemu,phb-enumerated", 0x1));
2417 
2418     /* Walk the bridge and subordinate buses */
2419     ret = spapr_dt_pci_bus(phb, PCI_HOST_BRIDGE(phb)->bus, fdt, bus_off);
2420     if (ret < 0) {
2421         return ret;
2422     }
2423 
2424     spapr_phb_nvgpu_populate_dt(phb, fdt, bus_off, &errp);
2425     if (errp) {
2426         error_report_err(errp);
2427     }
2428     spapr_phb_nvgpu_ram_populate_dt(phb, fdt);
2429 
2430     return 0;
2431 }
2432 
2433 void spapr_pci_rtas_init(void)
2434 {
2435     spapr_rtas_register(RTAS_READ_PCI_CONFIG, "read-pci-config",
2436                         rtas_read_pci_config);
2437     spapr_rtas_register(RTAS_WRITE_PCI_CONFIG, "write-pci-config",
2438                         rtas_write_pci_config);
2439     spapr_rtas_register(RTAS_IBM_READ_PCI_CONFIG, "ibm,read-pci-config",
2440                         rtas_ibm_read_pci_config);
2441     spapr_rtas_register(RTAS_IBM_WRITE_PCI_CONFIG, "ibm,write-pci-config",
2442                         rtas_ibm_write_pci_config);
2443     if (msi_nonbroken) {
2444         spapr_rtas_register(RTAS_IBM_QUERY_INTERRUPT_SOURCE_NUMBER,
2445                             "ibm,query-interrupt-source-number",
2446                             rtas_ibm_query_interrupt_source_number);
2447         spapr_rtas_register(RTAS_IBM_CHANGE_MSI, "ibm,change-msi",
2448                             rtas_ibm_change_msi);
2449     }
2450 
2451     spapr_rtas_register(RTAS_IBM_SET_EEH_OPTION,
2452                         "ibm,set-eeh-option",
2453                         rtas_ibm_set_eeh_option);
2454     spapr_rtas_register(RTAS_IBM_GET_CONFIG_ADDR_INFO2,
2455                         "ibm,get-config-addr-info2",
2456                         rtas_ibm_get_config_addr_info2);
2457     spapr_rtas_register(RTAS_IBM_READ_SLOT_RESET_STATE2,
2458                         "ibm,read-slot-reset-state2",
2459                         rtas_ibm_read_slot_reset_state2);
2460     spapr_rtas_register(RTAS_IBM_SET_SLOT_RESET,
2461                         "ibm,set-slot-reset",
2462                         rtas_ibm_set_slot_reset);
2463     spapr_rtas_register(RTAS_IBM_CONFIGURE_PE,
2464                         "ibm,configure-pe",
2465                         rtas_ibm_configure_pe);
2466     spapr_rtas_register(RTAS_IBM_SLOT_ERROR_DETAIL,
2467                         "ibm,slot-error-detail",
2468                         rtas_ibm_slot_error_detail);
2469 }
2470 
2471 static void spapr_pci_register_types(void)
2472 {
2473     type_register_static(&spapr_phb_info);
2474 }
2475 
2476 type_init(spapr_pci_register_types)
2477 
2478 static int spapr_switch_one_vga(DeviceState *dev, void *opaque)
2479 {
2480     bool be = *(bool *)opaque;
2481 
2482     if (object_dynamic_cast(OBJECT(dev), "VGA")
2483         || object_dynamic_cast(OBJECT(dev), "secondary-vga")) {
2484         object_property_set_bool(OBJECT(dev), be, "big-endian-framebuffer",
2485                                  &error_abort);
2486     }
2487     return 0;
2488 }
2489 
2490 void spapr_pci_switch_vga(bool big_endian)
2491 {
2492     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
2493     SpaprPhbState *sphb;
2494 
2495     /*
2496      * For backward compatibility with existing guests, we switch
2497      * the endianness of the VGA controller when changing the guest
2498      * interrupt mode
2499      */
2500     QLIST_FOREACH(sphb, &spapr->phbs, list) {
2501         BusState *bus = &PCI_HOST_BRIDGE(sphb)->bus->qbus;
2502         qbus_walk_children(bus, spapr_switch_one_vga, NULL, NULL, NULL,
2503                            &big_endian);
2504     }
2505 }
2506