xref: /openbmc/qemu/hw/ppc/spapr_pci.c (revision b69c3c21)
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     SpaprPciMsi *msi;
284     int *config_addr_key;
285     Error *err = NULL;
286     int i;
287 
288     /* Fins SpaprPhbState */
289     phb = spapr_pci_find_phb(spapr, buid);
290     if (phb) {
291         pdev = spapr_pci_find_dev(spapr, buid, config_addr);
292     }
293     if (!phb || !pdev) {
294         rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
295         return;
296     }
297 
298     switch (func) {
299     case RTAS_CHANGE_FN:
300         if (msi_present(pdev)) {
301             ret_intr_type = RTAS_TYPE_MSI;
302         } else if (msix_present(pdev)) {
303             ret_intr_type = RTAS_TYPE_MSIX;
304         } else {
305             rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
306             return;
307         }
308         break;
309     case RTAS_CHANGE_MSI_FN:
310         if (msi_present(pdev)) {
311             ret_intr_type = RTAS_TYPE_MSI;
312         } else {
313             rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
314             return;
315         }
316         break;
317     case RTAS_CHANGE_MSIX_FN:
318         if (msix_present(pdev)) {
319             ret_intr_type = RTAS_TYPE_MSIX;
320         } else {
321             rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
322             return;
323         }
324         break;
325     default:
326         error_report("rtas_ibm_change_msi(%u) is not implemented", func);
327         rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
328         return;
329     }
330 
331     msi = (SpaprPciMsi *) g_hash_table_lookup(phb->msi, &config_addr);
332 
333     /* Releasing MSIs */
334     if (!req_num) {
335         if (!msi) {
336             trace_spapr_pci_msi("Releasing wrong config", config_addr);
337             rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
338             return;
339         }
340 
341         if (msi_present(pdev)) {
342             spapr_msi_setmsg(pdev, 0, false, 0, 0);
343         }
344         if (msix_present(pdev)) {
345             spapr_msi_setmsg(pdev, 0, true, 0, 0);
346         }
347         g_hash_table_remove(phb->msi, &config_addr);
348 
349         trace_spapr_pci_msi("Released MSIs", config_addr);
350         rtas_st(rets, 0, RTAS_OUT_SUCCESS);
351         rtas_st(rets, 1, 0);
352         return;
353     }
354 
355     /* Enabling MSI */
356 
357     /* Check if the device supports as many IRQs as requested */
358     if (ret_intr_type == RTAS_TYPE_MSI) {
359         max_irqs = msi_nr_vectors_allocated(pdev);
360     } else if (ret_intr_type == RTAS_TYPE_MSIX) {
361         max_irqs = pdev->msix_entries_nr;
362     }
363     if (!max_irqs) {
364         error_report("Requested interrupt type %d is not enabled for device %x",
365                      ret_intr_type, config_addr);
366         rtas_st(rets, 0, -1); /* Hardware error */
367         return;
368     }
369     /* Correct the number if the guest asked for too many */
370     if (req_num > max_irqs) {
371         trace_spapr_pci_msi_retry(config_addr, req_num, max_irqs);
372         req_num = max_irqs;
373         irq = 0; /* to avoid misleading trace */
374         goto out;
375     }
376 
377     /* Allocate MSIs */
378     if (smc->legacy_irq_allocation) {
379         irq = spapr_irq_find(spapr, req_num, ret_intr_type == RTAS_TYPE_MSI,
380                              &err);
381     } else {
382         irq = spapr_irq_msi_alloc(spapr, req_num,
383                                   ret_intr_type == RTAS_TYPE_MSI, &err);
384     }
385     if (err) {
386         error_reportf_err(err, "Can't allocate MSIs for device %x: ",
387                           config_addr);
388         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
389         return;
390     }
391 
392     for (i = 0; i < req_num; i++) {
393         spapr_irq_claim(spapr, irq + i, false, &err);
394         if (err) {
395             if (i) {
396                 spapr_irq_free(spapr, irq, i);
397             }
398             if (!smc->legacy_irq_allocation) {
399                 spapr_irq_msi_free(spapr, irq, req_num);
400             }
401             error_reportf_err(err, "Can't allocate MSIs for device %x: ",
402                               config_addr);
403             rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
404             return;
405         }
406     }
407 
408     /* Release previous MSIs */
409     if (msi) {
410         g_hash_table_remove(phb->msi, &config_addr);
411     }
412 
413     /* Setup MSI/MSIX vectors in the device (via cfgspace or MSIX BAR) */
414     spapr_msi_setmsg(pdev, SPAPR_PCI_MSI_WINDOW, ret_intr_type == RTAS_TYPE_MSIX,
415                      irq, req_num);
416 
417     /* Add MSI device to cache */
418     msi = g_new(SpaprPciMsi, 1);
419     msi->first_irq = irq;
420     msi->num = req_num;
421     config_addr_key = g_new(int, 1);
422     *config_addr_key = config_addr;
423     g_hash_table_insert(phb->msi, config_addr_key, msi);
424 
425 out:
426     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
427     rtas_st(rets, 1, req_num);
428     rtas_st(rets, 2, ++seq_num);
429     if (nret > 3) {
430         rtas_st(rets, 3, ret_intr_type);
431     }
432 
433     trace_spapr_pci_rtas_ibm_change_msi(config_addr, func, req_num, irq);
434 }
435 
436 static void rtas_ibm_query_interrupt_source_number(PowerPCCPU *cpu,
437                                                    SpaprMachineState *spapr,
438                                                    uint32_t token,
439                                                    uint32_t nargs,
440                                                    target_ulong args,
441                                                    uint32_t nret,
442                                                    target_ulong rets)
443 {
444     uint32_t config_addr = rtas_ld(args, 0);
445     uint64_t buid = rtas_ldq(args, 1);
446     unsigned int intr_src_num = -1, ioa_intr_num = rtas_ld(args, 3);
447     SpaprPhbState *phb = NULL;
448     PCIDevice *pdev = NULL;
449     SpaprPciMsi *msi;
450 
451     /* Find SpaprPhbState */
452     phb = spapr_pci_find_phb(spapr, buid);
453     if (phb) {
454         pdev = spapr_pci_find_dev(spapr, buid, config_addr);
455     }
456     if (!phb || !pdev) {
457         rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
458         return;
459     }
460 
461     /* Find device descriptor and start IRQ */
462     msi = (SpaprPciMsi *) g_hash_table_lookup(phb->msi, &config_addr);
463     if (!msi || !msi->first_irq || !msi->num || (ioa_intr_num >= msi->num)) {
464         trace_spapr_pci_msi("Failed to return vector", config_addr);
465         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
466         return;
467     }
468     intr_src_num = msi->first_irq + ioa_intr_num;
469     trace_spapr_pci_rtas_ibm_query_interrupt_source_number(ioa_intr_num,
470                                                            intr_src_num);
471 
472     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
473     rtas_st(rets, 1, intr_src_num);
474     rtas_st(rets, 2, 1);/* 0 == level; 1 == edge */
475 }
476 
477 static void rtas_ibm_set_eeh_option(PowerPCCPU *cpu,
478                                     SpaprMachineState *spapr,
479                                     uint32_t token, uint32_t nargs,
480                                     target_ulong args, uint32_t nret,
481                                     target_ulong rets)
482 {
483     SpaprPhbState *sphb;
484     uint32_t addr, option;
485     uint64_t buid;
486     int ret;
487 
488     if ((nargs != 4) || (nret != 1)) {
489         goto param_error_exit;
490     }
491 
492     buid = rtas_ldq(args, 1);
493     addr = rtas_ld(args, 0);
494     option = rtas_ld(args, 3);
495 
496     sphb = spapr_pci_find_phb(spapr, buid);
497     if (!sphb) {
498         goto param_error_exit;
499     }
500 
501     if (!spapr_phb_eeh_available(sphb)) {
502         goto param_error_exit;
503     }
504 
505     ret = spapr_phb_vfio_eeh_set_option(sphb, addr, option);
506     rtas_st(rets, 0, ret);
507     return;
508 
509 param_error_exit:
510     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
511 }
512 
513 static void rtas_ibm_get_config_addr_info2(PowerPCCPU *cpu,
514                                            SpaprMachineState *spapr,
515                                            uint32_t token, uint32_t nargs,
516                                            target_ulong args, uint32_t nret,
517                                            target_ulong rets)
518 {
519     SpaprPhbState *sphb;
520     PCIDevice *pdev;
521     uint32_t addr, option;
522     uint64_t buid;
523 
524     if ((nargs != 4) || (nret != 2)) {
525         goto param_error_exit;
526     }
527 
528     buid = rtas_ldq(args, 1);
529     sphb = spapr_pci_find_phb(spapr, buid);
530     if (!sphb) {
531         goto param_error_exit;
532     }
533 
534     if (!spapr_phb_eeh_available(sphb)) {
535         goto param_error_exit;
536     }
537 
538     /*
539      * We always have PE address of form "00BB0001". "BB"
540      * represents the bus number of PE's primary bus.
541      */
542     option = rtas_ld(args, 3);
543     switch (option) {
544     case RTAS_GET_PE_ADDR:
545         addr = rtas_ld(args, 0);
546         pdev = spapr_pci_find_dev(spapr, buid, addr);
547         if (!pdev) {
548             goto param_error_exit;
549         }
550 
551         rtas_st(rets, 1, (pci_bus_num(pci_get_bus(pdev)) << 16) + 1);
552         break;
553     case RTAS_GET_PE_MODE:
554         rtas_st(rets, 1, RTAS_PE_MODE_SHARED);
555         break;
556     default:
557         goto param_error_exit;
558     }
559 
560     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
561     return;
562 
563 param_error_exit:
564     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
565 }
566 
567 static void rtas_ibm_read_slot_reset_state2(PowerPCCPU *cpu,
568                                             SpaprMachineState *spapr,
569                                             uint32_t token, uint32_t nargs,
570                                             target_ulong args, uint32_t nret,
571                                             target_ulong rets)
572 {
573     SpaprPhbState *sphb;
574     uint64_t buid;
575     int state, ret;
576 
577     if ((nargs != 3) || (nret != 4 && nret != 5)) {
578         goto param_error_exit;
579     }
580 
581     buid = rtas_ldq(args, 1);
582     sphb = spapr_pci_find_phb(spapr, buid);
583     if (!sphb) {
584         goto param_error_exit;
585     }
586 
587     if (!spapr_phb_eeh_available(sphb)) {
588         goto param_error_exit;
589     }
590 
591     ret = spapr_phb_vfio_eeh_get_state(sphb, &state);
592     rtas_st(rets, 0, ret);
593     if (ret != RTAS_OUT_SUCCESS) {
594         return;
595     }
596 
597     rtas_st(rets, 1, state);
598     rtas_st(rets, 2, RTAS_EEH_SUPPORT);
599     rtas_st(rets, 3, RTAS_EEH_PE_UNAVAIL_INFO);
600     if (nret >= 5) {
601         rtas_st(rets, 4, RTAS_EEH_PE_RECOVER_INFO);
602     }
603     return;
604 
605 param_error_exit:
606     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
607 }
608 
609 static void rtas_ibm_set_slot_reset(PowerPCCPU *cpu,
610                                     SpaprMachineState *spapr,
611                                     uint32_t token, uint32_t nargs,
612                                     target_ulong args, uint32_t nret,
613                                     target_ulong rets)
614 {
615     SpaprPhbState *sphb;
616     uint32_t option;
617     uint64_t buid;
618     int ret;
619 
620     if ((nargs != 4) || (nret != 1)) {
621         goto param_error_exit;
622     }
623 
624     buid = rtas_ldq(args, 1);
625     option = rtas_ld(args, 3);
626     sphb = spapr_pci_find_phb(spapr, buid);
627     if (!sphb) {
628         goto param_error_exit;
629     }
630 
631     if (!spapr_phb_eeh_available(sphb)) {
632         goto param_error_exit;
633     }
634 
635     ret = spapr_phb_vfio_eeh_reset(sphb, option);
636     rtas_st(rets, 0, ret);
637     return;
638 
639 param_error_exit:
640     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
641 }
642 
643 static void rtas_ibm_configure_pe(PowerPCCPU *cpu,
644                                   SpaprMachineState *spapr,
645                                   uint32_t token, uint32_t nargs,
646                                   target_ulong args, uint32_t nret,
647                                   target_ulong rets)
648 {
649     SpaprPhbState *sphb;
650     uint64_t buid;
651     int ret;
652 
653     if ((nargs != 3) || (nret != 1)) {
654         goto param_error_exit;
655     }
656 
657     buid = rtas_ldq(args, 1);
658     sphb = spapr_pci_find_phb(spapr, buid);
659     if (!sphb) {
660         goto param_error_exit;
661     }
662 
663     if (!spapr_phb_eeh_available(sphb)) {
664         goto param_error_exit;
665     }
666 
667     ret = spapr_phb_vfio_eeh_configure(sphb);
668     rtas_st(rets, 0, ret);
669     return;
670 
671 param_error_exit:
672     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
673 }
674 
675 /* To support it later */
676 static void rtas_ibm_slot_error_detail(PowerPCCPU *cpu,
677                                        SpaprMachineState *spapr,
678                                        uint32_t token, uint32_t nargs,
679                                        target_ulong args, uint32_t nret,
680                                        target_ulong rets)
681 {
682     SpaprPhbState *sphb;
683     int option;
684     uint64_t buid;
685 
686     if ((nargs != 8) || (nret != 1)) {
687         goto param_error_exit;
688     }
689 
690     buid = rtas_ldq(args, 1);
691     sphb = spapr_pci_find_phb(spapr, buid);
692     if (!sphb) {
693         goto param_error_exit;
694     }
695 
696     if (!spapr_phb_eeh_available(sphb)) {
697         goto param_error_exit;
698     }
699 
700     option = rtas_ld(args, 7);
701     switch (option) {
702     case RTAS_SLOT_TEMP_ERR_LOG:
703     case RTAS_SLOT_PERM_ERR_LOG:
704         break;
705     default:
706         goto param_error_exit;
707     }
708 
709     /* We don't have error log yet */
710     rtas_st(rets, 0, RTAS_OUT_NO_ERRORS_FOUND);
711     return;
712 
713 param_error_exit:
714     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
715 }
716 
717 static void pci_spapr_set_irq(void *opaque, int irq_num, int level)
718 {
719     /*
720      * Here we use the number returned by pci_swizzle_map_irq_fn to find a
721      * corresponding qemu_irq.
722      */
723     SpaprPhbState *phb = opaque;
724     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
725 
726     trace_spapr_pci_lsi_set(phb->dtbusname, irq_num, phb->lsi_table[irq_num].irq);
727     qemu_set_irq(spapr_qirq(spapr, phb->lsi_table[irq_num].irq), level);
728 }
729 
730 static PCIINTxRoute spapr_route_intx_pin_to_irq(void *opaque, int pin)
731 {
732     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(opaque);
733     PCIINTxRoute route;
734 
735     route.mode = PCI_INTX_ENABLED;
736     route.irq = sphb->lsi_table[pin].irq;
737 
738     return route;
739 }
740 
741 /*
742  * MSI/MSIX memory region implementation.
743  * The handler handles both MSI and MSIX.
744  * The vector number is encoded in least bits in data.
745  */
746 static void spapr_msi_write(void *opaque, hwaddr addr,
747                             uint64_t data, unsigned size)
748 {
749     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
750     uint32_t irq = data;
751 
752     trace_spapr_pci_msi_write(addr, data, irq);
753 
754     qemu_irq_pulse(spapr_qirq(spapr, irq));
755 }
756 
757 static const MemoryRegionOps spapr_msi_ops = {
758     /* There is no .read as the read result is undefined by PCI spec */
759     .read = NULL,
760     .write = spapr_msi_write,
761     .endianness = DEVICE_LITTLE_ENDIAN
762 };
763 
764 /*
765  * PHB PCI device
766  */
767 static AddressSpace *spapr_pci_dma_iommu(PCIBus *bus, void *opaque, int devfn)
768 {
769     SpaprPhbState *phb = opaque;
770 
771     return &phb->iommu_as;
772 }
773 
774 static char *spapr_phb_vfio_get_loc_code(SpaprPhbState *sphb,  PCIDevice *pdev)
775 {
776     char *path = NULL, *buf = NULL, *host = NULL;
777 
778     /* Get the PCI VFIO host id */
779     host = object_property_get_str(OBJECT(pdev), "host", NULL);
780     if (!host) {
781         goto err_out;
782     }
783 
784     /* Construct the path of the file that will give us the DT location */
785     path = g_strdup_printf("/sys/bus/pci/devices/%s/devspec", host);
786     g_free(host);
787     if (!g_file_get_contents(path, &buf, NULL, NULL)) {
788         goto err_out;
789     }
790     g_free(path);
791 
792     /* Construct and read from host device tree the loc-code */
793     path = g_strdup_printf("/proc/device-tree%s/ibm,loc-code", buf);
794     g_free(buf);
795     if (!g_file_get_contents(path, &buf, NULL, NULL)) {
796         goto err_out;
797     }
798     return buf;
799 
800 err_out:
801     g_free(path);
802     return NULL;
803 }
804 
805 static char *spapr_phb_get_loc_code(SpaprPhbState *sphb, PCIDevice *pdev)
806 {
807     char *buf;
808     const char *devtype = "qemu";
809     uint32_t busnr = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(pdev))));
810 
811     if (object_dynamic_cast(OBJECT(pdev), "vfio-pci")) {
812         buf = spapr_phb_vfio_get_loc_code(sphb, pdev);
813         if (buf) {
814             return buf;
815         }
816         devtype = "vfio";
817     }
818     /*
819      * For emulated devices and VFIO-failure case, make up
820      * the loc-code.
821      */
822     buf = g_strdup_printf("%s_%s:%04x:%02x:%02x.%x",
823                           devtype, pdev->name, sphb->index, busnr,
824                           PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
825     return buf;
826 }
827 
828 /* Macros to operate with address in OF binding to PCI */
829 #define b_x(x, p, l)    (((x) & ((1<<(l))-1)) << (p))
830 #define b_n(x)          b_x((x), 31, 1) /* 0 if relocatable */
831 #define b_p(x)          b_x((x), 30, 1) /* 1 if prefetchable */
832 #define b_t(x)          b_x((x), 29, 1) /* 1 if the address is aliased */
833 #define b_ss(x)         b_x((x), 24, 2) /* the space code */
834 #define b_bbbbbbbb(x)   b_x((x), 16, 8) /* bus number */
835 #define b_ddddd(x)      b_x((x), 11, 5) /* device number */
836 #define b_fff(x)        b_x((x), 8, 3)  /* function number */
837 #define b_rrrrrrrr(x)   b_x((x), 0, 8)  /* register number */
838 
839 /* for 'reg' OF properties */
840 #define RESOURCE_CELLS_SIZE 2
841 #define RESOURCE_CELLS_ADDRESS 3
842 
843 typedef struct ResourceFields {
844     uint32_t phys_hi;
845     uint32_t phys_mid;
846     uint32_t phys_lo;
847     uint32_t size_hi;
848     uint32_t size_lo;
849 } QEMU_PACKED ResourceFields;
850 
851 typedef struct ResourceProps {
852     ResourceFields reg[8];
853     uint32_t reg_len;
854 } ResourceProps;
855 
856 /* fill in the 'reg' OF properties for
857  * a PCI device. 'reg' describes resource requirements for a
858  * device's IO/MEM regions.
859  *
860  * the property is an array of ('phys-addr', 'size') pairs describing
861  * the addressable regions of the PCI device, where 'phys-addr' is a
862  * RESOURCE_CELLS_ADDRESS-tuple of 32-bit integers corresponding to
863  * (phys.hi, phys.mid, phys.lo), and 'size' is a
864  * RESOURCE_CELLS_SIZE-tuple corresponding to (size.hi, size.lo).
865  *
866  * phys.hi = 0xYYXXXXZZ, where:
867  *   0xYY = npt000ss
868  *          |||   |
869  *          |||   +-- space code
870  *          |||               |
871  *          |||               +  00 if configuration space
872  *          |||               +  01 if IO region,
873  *          |||               +  10 if 32-bit MEM region
874  *          |||               +  11 if 64-bit MEM region
875  *          |||
876  *          ||+------ for non-relocatable IO: 1 if aliased
877  *          ||        for relocatable IO: 1 if below 64KB
878  *          ||        for MEM: 1 if below 1MB
879  *          |+------- 1 if region is prefetchable
880  *          +-------- 1 if region is non-relocatable
881  *   0xXXXX = bbbbbbbb dddddfff, encoding bus, slot, and function
882  *            bits respectively
883  *   0xZZ = rrrrrrrr, the register number of the BAR corresponding
884  *          to the region
885  *
886  * phys.mid and phys.lo correspond respectively to the hi/lo portions
887  * of the actual address of the region.
888  *
889  * note also that addresses defined in this property are, at least
890  * for PAPR guests, relative to the PHBs IO/MEM windows, and
891  * correspond directly to the addresses in the BARs.
892  *
893  * in accordance with PCI Bus Binding to Open Firmware,
894  * IEEE Std 1275-1994, section 4.1.1, as implemented by PAPR+ v2.7,
895  * Appendix C.
896  */
897 static void populate_resource_props(PCIDevice *d, ResourceProps *rp)
898 {
899     int bus_num = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(d))));
900     uint32_t dev_id = (b_bbbbbbbb(bus_num) |
901                        b_ddddd(PCI_SLOT(d->devfn)) |
902                        b_fff(PCI_FUNC(d->devfn)));
903     ResourceFields *reg;
904     int i, reg_idx = 0;
905 
906     /* config space region */
907     reg = &rp->reg[reg_idx++];
908     reg->phys_hi = cpu_to_be32(dev_id);
909     reg->phys_mid = 0;
910     reg->phys_lo = 0;
911     reg->size_hi = 0;
912     reg->size_lo = 0;
913 
914     for (i = 0; i < PCI_NUM_REGIONS; i++) {
915         if (!d->io_regions[i].size) {
916             continue;
917         }
918 
919         reg = &rp->reg[reg_idx++];
920 
921         reg->phys_hi = cpu_to_be32(dev_id | b_rrrrrrrr(pci_bar(d, i)));
922         if (d->io_regions[i].type & PCI_BASE_ADDRESS_SPACE_IO) {
923             reg->phys_hi |= cpu_to_be32(b_ss(1));
924         } else if (d->io_regions[i].type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
925             reg->phys_hi |= cpu_to_be32(b_ss(3));
926         } else {
927             reg->phys_hi |= cpu_to_be32(b_ss(2));
928         }
929         reg->phys_mid = 0;
930         reg->phys_lo = 0;
931         reg->size_hi = cpu_to_be32(d->io_regions[i].size >> 32);
932         reg->size_lo = cpu_to_be32(d->io_regions[i].size);
933     }
934 
935     rp->reg_len = reg_idx * sizeof(ResourceFields);
936 }
937 
938 typedef struct PCIClass PCIClass;
939 typedef struct PCISubClass PCISubClass;
940 typedef struct PCIIFace PCIIFace;
941 
942 struct PCIIFace {
943     int iface;
944     const char *name;
945 };
946 
947 struct PCISubClass {
948     int subclass;
949     const char *name;
950     const PCIIFace *iface;
951 };
952 
953 struct PCIClass {
954     const char *name;
955     const PCISubClass *subc;
956 };
957 
958 static const PCISubClass undef_subclass[] = {
959     { PCI_CLASS_NOT_DEFINED_VGA, "display", NULL },
960     { 0xFF, NULL, NULL },
961 };
962 
963 static const PCISubClass mass_subclass[] = {
964     { PCI_CLASS_STORAGE_SCSI, "scsi", NULL },
965     { PCI_CLASS_STORAGE_IDE, "ide", NULL },
966     { PCI_CLASS_STORAGE_FLOPPY, "fdc", NULL },
967     { PCI_CLASS_STORAGE_IPI, "ipi", NULL },
968     { PCI_CLASS_STORAGE_RAID, "raid", NULL },
969     { PCI_CLASS_STORAGE_ATA, "ata", NULL },
970     { PCI_CLASS_STORAGE_SATA, "sata", NULL },
971     { PCI_CLASS_STORAGE_SAS, "sas", NULL },
972     { 0xFF, NULL, NULL },
973 };
974 
975 static const PCISubClass net_subclass[] = {
976     { PCI_CLASS_NETWORK_ETHERNET, "ethernet", NULL },
977     { PCI_CLASS_NETWORK_TOKEN_RING, "token-ring", NULL },
978     { PCI_CLASS_NETWORK_FDDI, "fddi", NULL },
979     { PCI_CLASS_NETWORK_ATM, "atm", NULL },
980     { PCI_CLASS_NETWORK_ISDN, "isdn", NULL },
981     { PCI_CLASS_NETWORK_WORLDFIP, "worldfip", NULL },
982     { PCI_CLASS_NETWORK_PICMG214, "picmg", NULL },
983     { 0xFF, NULL, NULL },
984 };
985 
986 static const PCISubClass displ_subclass[] = {
987     { PCI_CLASS_DISPLAY_VGA, "vga", NULL },
988     { PCI_CLASS_DISPLAY_XGA, "xga", NULL },
989     { PCI_CLASS_DISPLAY_3D, "3d-controller", NULL },
990     { 0xFF, NULL, NULL },
991 };
992 
993 static const PCISubClass media_subclass[] = {
994     { PCI_CLASS_MULTIMEDIA_VIDEO, "video", NULL },
995     { PCI_CLASS_MULTIMEDIA_AUDIO, "sound", NULL },
996     { PCI_CLASS_MULTIMEDIA_PHONE, "telephony", NULL },
997     { 0xFF, NULL, NULL },
998 };
999 
1000 static const PCISubClass mem_subclass[] = {
1001     { PCI_CLASS_MEMORY_RAM, "memory", NULL },
1002     { PCI_CLASS_MEMORY_FLASH, "flash", NULL },
1003     { 0xFF, NULL, NULL },
1004 };
1005 
1006 static const PCISubClass bridg_subclass[] = {
1007     { PCI_CLASS_BRIDGE_HOST, "host", NULL },
1008     { PCI_CLASS_BRIDGE_ISA, "isa", NULL },
1009     { PCI_CLASS_BRIDGE_EISA, "eisa", NULL },
1010     { PCI_CLASS_BRIDGE_MC, "mca", NULL },
1011     { PCI_CLASS_BRIDGE_PCI, "pci", NULL },
1012     { PCI_CLASS_BRIDGE_PCMCIA, "pcmcia", NULL },
1013     { PCI_CLASS_BRIDGE_NUBUS, "nubus", NULL },
1014     { PCI_CLASS_BRIDGE_CARDBUS, "cardbus", NULL },
1015     { PCI_CLASS_BRIDGE_RACEWAY, "raceway", NULL },
1016     { PCI_CLASS_BRIDGE_PCI_SEMITP, "semi-transparent-pci", NULL },
1017     { PCI_CLASS_BRIDGE_IB_PCI, "infiniband", NULL },
1018     { 0xFF, NULL, NULL },
1019 };
1020 
1021 static const PCISubClass comm_subclass[] = {
1022     { PCI_CLASS_COMMUNICATION_SERIAL, "serial", NULL },
1023     { PCI_CLASS_COMMUNICATION_PARALLEL, "parallel", NULL },
1024     { PCI_CLASS_COMMUNICATION_MULTISERIAL, "multiport-serial", NULL },
1025     { PCI_CLASS_COMMUNICATION_MODEM, "modem", NULL },
1026     { PCI_CLASS_COMMUNICATION_GPIB, "gpib", NULL },
1027     { PCI_CLASS_COMMUNICATION_SC, "smart-card", NULL },
1028     { 0xFF, NULL, NULL, },
1029 };
1030 
1031 static const PCIIFace pic_iface[] = {
1032     { PCI_CLASS_SYSTEM_PIC_IOAPIC, "io-apic" },
1033     { PCI_CLASS_SYSTEM_PIC_IOXAPIC, "io-xapic" },
1034     { 0xFF, NULL },
1035 };
1036 
1037 static const PCISubClass sys_subclass[] = {
1038     { PCI_CLASS_SYSTEM_PIC, "interrupt-controller", pic_iface },
1039     { PCI_CLASS_SYSTEM_DMA, "dma-controller", NULL },
1040     { PCI_CLASS_SYSTEM_TIMER, "timer", NULL },
1041     { PCI_CLASS_SYSTEM_RTC, "rtc", NULL },
1042     { PCI_CLASS_SYSTEM_PCI_HOTPLUG, "hot-plug-controller", NULL },
1043     { PCI_CLASS_SYSTEM_SDHCI, "sd-host-controller", NULL },
1044     { 0xFF, NULL, NULL },
1045 };
1046 
1047 static const PCISubClass inp_subclass[] = {
1048     { PCI_CLASS_INPUT_KEYBOARD, "keyboard", NULL },
1049     { PCI_CLASS_INPUT_PEN, "pen", NULL },
1050     { PCI_CLASS_INPUT_MOUSE, "mouse", NULL },
1051     { PCI_CLASS_INPUT_SCANNER, "scanner", NULL },
1052     { PCI_CLASS_INPUT_GAMEPORT, "gameport", NULL },
1053     { 0xFF, NULL, NULL },
1054 };
1055 
1056 static const PCISubClass dock_subclass[] = {
1057     { PCI_CLASS_DOCKING_GENERIC, "dock", NULL },
1058     { 0xFF, NULL, NULL },
1059 };
1060 
1061 static const PCISubClass cpu_subclass[] = {
1062     { PCI_CLASS_PROCESSOR_PENTIUM, "pentium", NULL },
1063     { PCI_CLASS_PROCESSOR_POWERPC, "powerpc", NULL },
1064     { PCI_CLASS_PROCESSOR_MIPS, "mips", NULL },
1065     { PCI_CLASS_PROCESSOR_CO, "co-processor", NULL },
1066     { 0xFF, NULL, NULL },
1067 };
1068 
1069 static const PCIIFace usb_iface[] = {
1070     { PCI_CLASS_SERIAL_USB_UHCI, "usb-uhci" },
1071     { PCI_CLASS_SERIAL_USB_OHCI, "usb-ohci", },
1072     { PCI_CLASS_SERIAL_USB_EHCI, "usb-ehci" },
1073     { PCI_CLASS_SERIAL_USB_XHCI, "usb-xhci" },
1074     { PCI_CLASS_SERIAL_USB_UNKNOWN, "usb-unknown" },
1075     { PCI_CLASS_SERIAL_USB_DEVICE, "usb-device" },
1076     { 0xFF, NULL },
1077 };
1078 
1079 static const PCISubClass ser_subclass[] = {
1080     { PCI_CLASS_SERIAL_FIREWIRE, "firewire", NULL },
1081     { PCI_CLASS_SERIAL_ACCESS, "access-bus", NULL },
1082     { PCI_CLASS_SERIAL_SSA, "ssa", NULL },
1083     { PCI_CLASS_SERIAL_USB, "usb", usb_iface },
1084     { PCI_CLASS_SERIAL_FIBER, "fibre-channel", NULL },
1085     { PCI_CLASS_SERIAL_SMBUS, "smb", NULL },
1086     { PCI_CLASS_SERIAL_IB, "infiniband", NULL },
1087     { PCI_CLASS_SERIAL_IPMI, "ipmi", NULL },
1088     { PCI_CLASS_SERIAL_SERCOS, "sercos", NULL },
1089     { PCI_CLASS_SERIAL_CANBUS, "canbus", NULL },
1090     { 0xFF, NULL, NULL },
1091 };
1092 
1093 static const PCISubClass wrl_subclass[] = {
1094     { PCI_CLASS_WIRELESS_IRDA, "irda", NULL },
1095     { PCI_CLASS_WIRELESS_CIR, "consumer-ir", NULL },
1096     { PCI_CLASS_WIRELESS_RF_CONTROLLER, "rf-controller", NULL },
1097     { PCI_CLASS_WIRELESS_BLUETOOTH, "bluetooth", NULL },
1098     { PCI_CLASS_WIRELESS_BROADBAND, "broadband", NULL },
1099     { 0xFF, NULL, NULL },
1100 };
1101 
1102 static const PCISubClass sat_subclass[] = {
1103     { PCI_CLASS_SATELLITE_TV, "satellite-tv", NULL },
1104     { PCI_CLASS_SATELLITE_AUDIO, "satellite-audio", NULL },
1105     { PCI_CLASS_SATELLITE_VOICE, "satellite-voice", NULL },
1106     { PCI_CLASS_SATELLITE_DATA, "satellite-data", NULL },
1107     { 0xFF, NULL, NULL },
1108 };
1109 
1110 static const PCISubClass crypt_subclass[] = {
1111     { PCI_CLASS_CRYPT_NETWORK, "network-encryption", NULL },
1112     { PCI_CLASS_CRYPT_ENTERTAINMENT,
1113       "entertainment-encryption", NULL },
1114     { 0xFF, NULL, NULL },
1115 };
1116 
1117 static const PCISubClass spc_subclass[] = {
1118     { PCI_CLASS_SP_DPIO, "dpio", NULL },
1119     { PCI_CLASS_SP_PERF, "counter", NULL },
1120     { PCI_CLASS_SP_SYNCH, "measurement", NULL },
1121     { PCI_CLASS_SP_MANAGEMENT, "management-card", NULL },
1122     { 0xFF, NULL, NULL },
1123 };
1124 
1125 static const PCIClass pci_classes[] = {
1126     { "legacy-device", undef_subclass },
1127     { "mass-storage",  mass_subclass },
1128     { "network", net_subclass },
1129     { "display", displ_subclass, },
1130     { "multimedia-device", media_subclass },
1131     { "memory-controller", mem_subclass },
1132     { "unknown-bridge", bridg_subclass },
1133     { "communication-controller", comm_subclass},
1134     { "system-peripheral", sys_subclass },
1135     { "input-controller", inp_subclass },
1136     { "docking-station", dock_subclass },
1137     { "cpu", cpu_subclass },
1138     { "serial-bus", ser_subclass },
1139     { "wireless-controller", wrl_subclass },
1140     { "intelligent-io", NULL },
1141     { "satellite-device", sat_subclass },
1142     { "encryption", crypt_subclass },
1143     { "data-processing-controller", spc_subclass },
1144 };
1145 
1146 static const char *dt_name_from_class(uint8_t class, uint8_t subclass,
1147                                       uint8_t iface)
1148 {
1149     const PCIClass *pclass;
1150     const PCISubClass *psubclass;
1151     const PCIIFace *piface;
1152     const char *name;
1153 
1154     if (class >= ARRAY_SIZE(pci_classes)) {
1155         return "pci";
1156     }
1157 
1158     pclass = pci_classes + class;
1159     name = pclass->name;
1160 
1161     if (pclass->subc == NULL) {
1162         return name;
1163     }
1164 
1165     psubclass = pclass->subc;
1166     while ((psubclass->subclass & 0xff) != 0xff) {
1167         if ((psubclass->subclass & 0xff) == subclass) {
1168             name = psubclass->name;
1169             break;
1170         }
1171         psubclass++;
1172     }
1173 
1174     piface = psubclass->iface;
1175     if (piface == NULL) {
1176         return name;
1177     }
1178     while ((piface->iface & 0xff) != 0xff) {
1179         if ((piface->iface & 0xff) == iface) {
1180             name = piface->name;
1181             break;
1182         }
1183         piface++;
1184     }
1185 
1186     return name;
1187 }
1188 
1189 /*
1190  * DRC helper functions
1191  */
1192 
1193 static uint32_t drc_id_from_devfn(SpaprPhbState *phb,
1194                                   uint8_t chassis, int32_t devfn)
1195 {
1196     return (phb->index << 16) | (chassis << 8) | devfn;
1197 }
1198 
1199 static SpaprDrc *drc_from_devfn(SpaprPhbState *phb,
1200                                 uint8_t chassis, int32_t devfn)
1201 {
1202     return spapr_drc_by_id(TYPE_SPAPR_DRC_PCI,
1203                            drc_id_from_devfn(phb, chassis, devfn));
1204 }
1205 
1206 static uint8_t chassis_from_bus(PCIBus *bus, Error **errp)
1207 {
1208     if (pci_bus_is_root(bus)) {
1209         return 0;
1210     } else {
1211         PCIDevice *bridge = pci_bridge_get_device(bus);
1212 
1213         return object_property_get_uint(OBJECT(bridge), "chassis_nr", errp);
1214     }
1215 }
1216 
1217 static SpaprDrc *drc_from_dev(SpaprPhbState *phb, PCIDevice *dev)
1218 {
1219     Error *local_err = NULL;
1220     uint8_t chassis = chassis_from_bus(pci_get_bus(dev), &local_err);
1221 
1222     if (local_err) {
1223         error_report_err(local_err);
1224         return NULL;
1225     }
1226 
1227     return drc_from_devfn(phb, chassis, dev->devfn);
1228 }
1229 
1230 static void add_drcs(SpaprPhbState *phb, PCIBus *bus, Error **errp)
1231 {
1232     Object *owner;
1233     int i;
1234     uint8_t chassis;
1235     Error *local_err = NULL;
1236 
1237     if (!phb->dr_enabled) {
1238         return;
1239     }
1240 
1241     chassis = chassis_from_bus(bus, &local_err);
1242     if (local_err) {
1243         error_propagate(errp, local_err);
1244         return;
1245     }
1246 
1247     if (pci_bus_is_root(bus)) {
1248         owner = OBJECT(phb);
1249     } else {
1250         owner = OBJECT(pci_bridge_get_device(bus));
1251     }
1252 
1253     for (i = 0; i < PCI_SLOT_MAX * PCI_FUNC_MAX; i++) {
1254         spapr_dr_connector_new(owner, TYPE_SPAPR_DRC_PCI,
1255                                drc_id_from_devfn(phb, chassis, i));
1256     }
1257 }
1258 
1259 static void remove_drcs(SpaprPhbState *phb, PCIBus *bus, Error **errp)
1260 {
1261     int i;
1262     uint8_t chassis;
1263     Error *local_err = NULL;
1264 
1265     if (!phb->dr_enabled) {
1266         return;
1267     }
1268 
1269     chassis = chassis_from_bus(bus, &local_err);
1270     if (local_err) {
1271         error_propagate(errp, local_err);
1272         return;
1273     }
1274 
1275     for (i = PCI_SLOT_MAX * PCI_FUNC_MAX - 1; i >= 0; i--) {
1276         SpaprDrc *drc = drc_from_devfn(phb, chassis, i);
1277 
1278         if (drc) {
1279             object_unparent(OBJECT(drc));
1280         }
1281     }
1282 }
1283 
1284 typedef struct PciWalkFdt {
1285     void *fdt;
1286     int offset;
1287     SpaprPhbState *sphb;
1288     int err;
1289 } PciWalkFdt;
1290 
1291 static int spapr_dt_pci_device(SpaprPhbState *sphb, PCIDevice *dev,
1292                                void *fdt, int parent_offset);
1293 
1294 static void spapr_dt_pci_device_cb(PCIBus *bus, PCIDevice *pdev,
1295                                    void *opaque)
1296 {
1297     PciWalkFdt *p = opaque;
1298     int err;
1299 
1300     if (p->err) {
1301         /* Something's already broken, don't keep going */
1302         return;
1303     }
1304 
1305     err = spapr_dt_pci_device(p->sphb, pdev, p->fdt, p->offset);
1306     if (err < 0) {
1307         p->err = err;
1308     }
1309 }
1310 
1311 /* Augment PCI device node with bridge specific information */
1312 static int spapr_dt_pci_bus(SpaprPhbState *sphb, PCIBus *bus,
1313                                void *fdt, int offset)
1314 {
1315     Object *owner;
1316     PciWalkFdt cbinfo = {
1317         .fdt = fdt,
1318         .offset = offset,
1319         .sphb = sphb,
1320         .err = 0,
1321     };
1322     int ret;
1323 
1324     _FDT(fdt_setprop_cell(fdt, offset, "#address-cells",
1325                           RESOURCE_CELLS_ADDRESS));
1326     _FDT(fdt_setprop_cell(fdt, offset, "#size-cells",
1327                           RESOURCE_CELLS_SIZE));
1328 
1329     assert(bus);
1330     pci_for_each_device_reverse(bus, pci_bus_num(bus),
1331                                 spapr_dt_pci_device_cb, &cbinfo);
1332     if (cbinfo.err) {
1333         return cbinfo.err;
1334     }
1335 
1336     if (pci_bus_is_root(bus)) {
1337         owner = OBJECT(sphb);
1338     } else {
1339         owner = OBJECT(pci_bridge_get_device(bus));
1340     }
1341 
1342     ret = spapr_dt_drc(fdt, offset, owner,
1343                        SPAPR_DR_CONNECTOR_TYPE_PCI);
1344     if (ret) {
1345         return ret;
1346     }
1347 
1348     return offset;
1349 }
1350 
1351 /* create OF node for pci device and required OF DT properties */
1352 static int spapr_dt_pci_device(SpaprPhbState *sphb, PCIDevice *dev,
1353                                void *fdt, int parent_offset)
1354 {
1355     int offset;
1356     const gchar *basename;
1357     gchar *nodename;
1358     int slot = PCI_SLOT(dev->devfn);
1359     int func = PCI_FUNC(dev->devfn);
1360     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
1361     ResourceProps rp;
1362     SpaprDrc *drc = drc_from_dev(sphb, dev);
1363     uint32_t vendor_id = pci_default_read_config(dev, PCI_VENDOR_ID, 2);
1364     uint32_t device_id = pci_default_read_config(dev, PCI_DEVICE_ID, 2);
1365     uint32_t revision_id = pci_default_read_config(dev, PCI_REVISION_ID, 1);
1366     uint32_t ccode = pci_default_read_config(dev, PCI_CLASS_PROG, 3);
1367     uint32_t irq_pin = pci_default_read_config(dev, PCI_INTERRUPT_PIN, 1);
1368     uint32_t subsystem_id = pci_default_read_config(dev, PCI_SUBSYSTEM_ID, 2);
1369     uint32_t subsystem_vendor_id =
1370         pci_default_read_config(dev, PCI_SUBSYSTEM_VENDOR_ID, 2);
1371     uint32_t cache_line_size =
1372         pci_default_read_config(dev, PCI_CACHE_LINE_SIZE, 1);
1373     uint32_t pci_status = pci_default_read_config(dev, PCI_STATUS, 2);
1374     gchar *loc_code;
1375 
1376     basename = dt_name_from_class((ccode >> 16) & 0xff, (ccode >> 8) & 0xff,
1377                                   ccode & 0xff);
1378 
1379     if (func != 0) {
1380         nodename = g_strdup_printf("%s@%x,%x", basename, slot, func);
1381     } else {
1382         nodename = g_strdup_printf("%s@%x", basename, slot);
1383     }
1384 
1385     _FDT(offset = fdt_add_subnode(fdt, parent_offset, nodename));
1386 
1387     g_free(nodename);
1388 
1389     /* in accordance with PAPR+ v2.7 13.6.3, Table 181 */
1390     _FDT(fdt_setprop_cell(fdt, offset, "vendor-id", vendor_id));
1391     _FDT(fdt_setprop_cell(fdt, offset, "device-id", device_id));
1392     _FDT(fdt_setprop_cell(fdt, offset, "revision-id", revision_id));
1393 
1394     _FDT(fdt_setprop_cell(fdt, offset, "class-code", ccode));
1395     if (irq_pin) {
1396         _FDT(fdt_setprop_cell(fdt, offset, "interrupts", irq_pin));
1397     }
1398 
1399     if (subsystem_id) {
1400         _FDT(fdt_setprop_cell(fdt, offset, "subsystem-id", subsystem_id));
1401     }
1402 
1403     if (subsystem_vendor_id) {
1404         _FDT(fdt_setprop_cell(fdt, offset, "subsystem-vendor-id",
1405                               subsystem_vendor_id));
1406     }
1407 
1408     _FDT(fdt_setprop_cell(fdt, offset, "cache-line-size", cache_line_size));
1409 
1410 
1411     /* the following fdt cells are masked off the pci status register */
1412     _FDT(fdt_setprop_cell(fdt, offset, "devsel-speed",
1413                           PCI_STATUS_DEVSEL_MASK & pci_status));
1414 
1415     if (pci_status & PCI_STATUS_FAST_BACK) {
1416         _FDT(fdt_setprop(fdt, offset, "fast-back-to-back", NULL, 0));
1417     }
1418     if (pci_status & PCI_STATUS_66MHZ) {
1419         _FDT(fdt_setprop(fdt, offset, "66mhz-capable", NULL, 0));
1420     }
1421     if (pci_status & PCI_STATUS_UDF) {
1422         _FDT(fdt_setprop(fdt, offset, "udf-supported", NULL, 0));
1423     }
1424 
1425     loc_code = spapr_phb_get_loc_code(sphb, dev);
1426     _FDT(fdt_setprop_string(fdt, offset, "ibm,loc-code", loc_code));
1427     g_free(loc_code);
1428 
1429     if (drc) {
1430         _FDT(fdt_setprop_cell(fdt, offset, "ibm,my-drc-index",
1431                               spapr_drc_index(drc)));
1432     }
1433 
1434     if (msi_present(dev)) {
1435         uint32_t max_msi = msi_nr_vectors_allocated(dev);
1436         if (max_msi) {
1437             _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi", max_msi));
1438         }
1439     }
1440     if (msix_present(dev)) {
1441         uint32_t max_msix = dev->msix_entries_nr;
1442         if (max_msix) {
1443             _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi-x", max_msix));
1444         }
1445     }
1446 
1447     populate_resource_props(dev, &rp);
1448     _FDT(fdt_setprop(fdt, offset, "reg", (uint8_t *)rp.reg, rp.reg_len));
1449 
1450     if (sphb->pcie_ecs && pci_is_express(dev)) {
1451         _FDT(fdt_setprop_cell(fdt, offset, "ibm,pci-config-space-type", 0x1));
1452     }
1453 
1454     spapr_phb_nvgpu_populate_pcidev_dt(dev, fdt, offset, sphb);
1455 
1456     if (!pc->is_bridge) {
1457         /* Properties only for non-bridges */
1458         uint32_t min_grant = pci_default_read_config(dev, PCI_MIN_GNT, 1);
1459         uint32_t max_latency = pci_default_read_config(dev, PCI_MAX_LAT, 1);
1460         _FDT(fdt_setprop_cell(fdt, offset, "min-grant", min_grant));
1461         _FDT(fdt_setprop_cell(fdt, offset, "max-latency", max_latency));
1462         return offset;
1463     } else {
1464         PCIBus *sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(dev));
1465 
1466         return spapr_dt_pci_bus(sphb, sec_bus, fdt, offset);
1467     }
1468 }
1469 
1470 /* Callback to be called during DRC release. */
1471 void spapr_phb_remove_pci_device_cb(DeviceState *dev)
1472 {
1473     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
1474 
1475     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
1476     object_unparent(OBJECT(dev));
1477 }
1478 
1479 int spapr_pci_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
1480                           void *fdt, int *fdt_start_offset, Error **errp)
1481 {
1482     HotplugHandler *plug_handler = qdev_get_hotplug_handler(drc->dev);
1483     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(plug_handler);
1484     PCIDevice *pdev = PCI_DEVICE(drc->dev);
1485 
1486     *fdt_start_offset = spapr_dt_pci_device(sphb, pdev, fdt, 0);
1487     return 0;
1488 }
1489 
1490 static void spapr_pci_bridge_plug(SpaprPhbState *phb,
1491                                   PCIBridge *bridge,
1492                                   Error **errp)
1493 {
1494     Error *local_err = NULL;
1495     PCIBus *bus = pci_bridge_get_sec_bus(bridge);
1496 
1497     add_drcs(phb, bus, &local_err);
1498     if (local_err) {
1499         error_propagate(errp, local_err);
1500         return;
1501     }
1502 }
1503 
1504 static void spapr_pci_plug(HotplugHandler *plug_handler,
1505                            DeviceState *plugged_dev, Error **errp)
1506 {
1507     SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1508     PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1509     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(plugged_dev);
1510     SpaprDrc *drc = drc_from_dev(phb, pdev);
1511     Error *local_err = NULL;
1512     PCIBus *bus = PCI_BUS(qdev_get_parent_bus(DEVICE(pdev)));
1513     uint32_t slotnr = PCI_SLOT(pdev->devfn);
1514 
1515     /* if DR is disabled we don't need to do anything in the case of
1516      * hotplug or coldplug callbacks
1517      */
1518     if (!phb->dr_enabled) {
1519         /* if this is a hotplug operation initiated by the user
1520          * we need to let them know it's not enabled
1521          */
1522         if (plugged_dev->hotplugged) {
1523             error_setg(&local_err, QERR_BUS_NO_HOTPLUG,
1524                        object_get_typename(OBJECT(phb)));
1525         }
1526         goto out;
1527     }
1528 
1529     g_assert(drc);
1530 
1531     if (pc->is_bridge) {
1532         spapr_pci_bridge_plug(phb, PCI_BRIDGE(plugged_dev), &local_err);
1533         if (local_err) {
1534             error_propagate(errp, local_err);
1535             return;
1536         }
1537     }
1538 
1539     /* Following the QEMU convention used for PCIe multifunction
1540      * hotplug, we do not allow functions to be hotplugged to a
1541      * slot that already has function 0 present
1542      */
1543     if (plugged_dev->hotplugged && bus->devices[PCI_DEVFN(slotnr, 0)] &&
1544         PCI_FUNC(pdev->devfn) != 0) {
1545         error_setg(&local_err, "PCI: slot %d function 0 already ocuppied by %s,"
1546                    " additional functions can no longer be exposed to guest.",
1547                    slotnr, bus->devices[PCI_DEVFN(slotnr, 0)]->name);
1548         goto out;
1549     }
1550 
1551     spapr_drc_attach(drc, DEVICE(pdev), &local_err);
1552     if (local_err) {
1553         goto out;
1554     }
1555 
1556     /* If this is function 0, signal hotplug for all the device functions.
1557      * Otherwise defer sending the hotplug event.
1558      */
1559     if (!spapr_drc_hotplugged(plugged_dev)) {
1560         spapr_drc_reset(drc);
1561     } else if (PCI_FUNC(pdev->devfn) == 0) {
1562         int i;
1563         uint8_t chassis = chassis_from_bus(pci_get_bus(pdev), &local_err);
1564 
1565         if (local_err) {
1566             error_propagate(errp, local_err);
1567             return;
1568         }
1569 
1570         for (i = 0; i < 8; i++) {
1571             SpaprDrc *func_drc;
1572             SpaprDrcClass *func_drck;
1573             SpaprDREntitySense state;
1574 
1575             func_drc = drc_from_devfn(phb, chassis, PCI_DEVFN(slotnr, i));
1576             func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1577             state = func_drck->dr_entity_sense(func_drc);
1578 
1579             if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) {
1580                 spapr_hotplug_req_add_by_index(func_drc);
1581             }
1582         }
1583     }
1584 
1585 out:
1586     error_propagate(errp, local_err);
1587 }
1588 
1589 static void spapr_pci_bridge_unplug(SpaprPhbState *phb,
1590                                     PCIBridge *bridge,
1591                                     Error **errp)
1592 {
1593     Error *local_err = NULL;
1594     PCIBus *bus = pci_bridge_get_sec_bus(bridge);
1595 
1596     remove_drcs(phb, bus, &local_err);
1597     if (local_err) {
1598         error_propagate(errp, local_err);
1599         return;
1600     }
1601 }
1602 
1603 static void spapr_pci_unplug(HotplugHandler *plug_handler,
1604                              DeviceState *plugged_dev, Error **errp)
1605 {
1606     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(plugged_dev);
1607     SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1608 
1609     /* some version guests do not wait for completion of a device
1610      * cleanup (generally done asynchronously by the kernel) before
1611      * signaling to QEMU that the device is safe, but instead sleep
1612      * for some 'safe' period of time. unfortunately on a busy host
1613      * this sleep isn't guaranteed to be long enough, resulting in
1614      * bad things like IRQ lines being left asserted during final
1615      * device removal. to deal with this we call reset just prior
1616      * to finalizing the device, which will put the device back into
1617      * an 'idle' state, as the device cleanup code expects.
1618      */
1619     pci_device_reset(PCI_DEVICE(plugged_dev));
1620 
1621     if (pc->is_bridge) {
1622         Error *local_err = NULL;
1623         spapr_pci_bridge_unplug(phb, PCI_BRIDGE(plugged_dev), &local_err);
1624         if (local_err) {
1625             error_propagate(errp, local_err);
1626         }
1627         return;
1628     }
1629 
1630     object_property_set_bool(OBJECT(plugged_dev), false, "realized",
1631                              &error_abort);
1632 }
1633 
1634 static void spapr_pci_unplug_request(HotplugHandler *plug_handler,
1635                                      DeviceState *plugged_dev, Error **errp)
1636 {
1637     SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1638     PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1639     SpaprDrc *drc = drc_from_dev(phb, pdev);
1640 
1641     if (!phb->dr_enabled) {
1642         error_setg(errp, QERR_BUS_NO_HOTPLUG,
1643                    object_get_typename(OBJECT(phb)));
1644         return;
1645     }
1646 
1647     g_assert(drc);
1648     g_assert(drc->dev == plugged_dev);
1649 
1650     if (!spapr_drc_unplug_requested(drc)) {
1651         PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(plugged_dev);
1652         uint32_t slotnr = PCI_SLOT(pdev->devfn);
1653         SpaprDrc *func_drc;
1654         SpaprDrcClass *func_drck;
1655         SpaprDREntitySense state;
1656         int i;
1657         Error *local_err = NULL;
1658         uint8_t chassis = chassis_from_bus(pci_get_bus(pdev), &local_err);
1659 
1660         if (local_err) {
1661             error_propagate(errp, local_err);
1662             return;
1663         }
1664 
1665         if (pc->is_bridge) {
1666             error_setg(errp, "PCI: Hot unplug of PCI bridges not supported");
1667             return;
1668         }
1669         if (object_property_get_uint(OBJECT(pdev), "nvlink2-tgt", NULL)) {
1670             error_setg(errp, "PCI: Cannot unplug NVLink2 devices");
1671             return;
1672         }
1673 
1674         /* ensure any other present functions are pending unplug */
1675         if (PCI_FUNC(pdev->devfn) == 0) {
1676             for (i = 1; i < 8; i++) {
1677                 func_drc = drc_from_devfn(phb, chassis, PCI_DEVFN(slotnr, i));
1678                 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1679                 state = func_drck->dr_entity_sense(func_drc);
1680                 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT
1681                     && !spapr_drc_unplug_requested(func_drc)) {
1682                     /*
1683                      * Attempting to remove function 0 of a multifunction
1684                      * device will will cascade into removing all child
1685                      * functions, even if their unplug weren't requested
1686                      * beforehand.
1687                      */
1688                     spapr_drc_detach(func_drc);
1689                 }
1690             }
1691         }
1692 
1693         spapr_drc_detach(drc);
1694 
1695         /* if this isn't func 0, defer unplug event. otherwise signal removal
1696          * for all present functions
1697          */
1698         if (PCI_FUNC(pdev->devfn) == 0) {
1699             for (i = 7; i >= 0; i--) {
1700                 func_drc = drc_from_devfn(phb, chassis, PCI_DEVFN(slotnr, i));
1701                 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1702                 state = func_drck->dr_entity_sense(func_drc);
1703                 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) {
1704                     spapr_hotplug_req_remove_by_index(func_drc);
1705                 }
1706             }
1707         }
1708     }
1709 }
1710 
1711 static void spapr_phb_finalizefn(Object *obj)
1712 {
1713     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(obj);
1714 
1715     g_free(sphb->dtbusname);
1716     sphb->dtbusname = NULL;
1717 }
1718 
1719 static void spapr_phb_unrealize(DeviceState *dev)
1720 {
1721     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
1722     SysBusDevice *s = SYS_BUS_DEVICE(dev);
1723     PCIHostState *phb = PCI_HOST_BRIDGE(s);
1724     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(phb);
1725     SpaprTceTable *tcet;
1726     int i;
1727     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
1728 
1729     spapr_phb_nvgpu_free(sphb);
1730 
1731     if (sphb->msi) {
1732         g_hash_table_unref(sphb->msi);
1733         sphb->msi = NULL;
1734     }
1735 
1736     /*
1737      * Remove IO/MMIO subregions and aliases, rest should get cleaned
1738      * via PHB's unrealize->object_finalize
1739      */
1740     for (i = windows_supported - 1; i >= 0; i--) {
1741         tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[i]);
1742         if (tcet) {
1743             memory_region_del_subregion(&sphb->iommu_root,
1744                                         spapr_tce_get_iommu(tcet));
1745         }
1746     }
1747 
1748     remove_drcs(sphb, phb->bus, &error_abort);
1749 
1750     for (i = PCI_NUM_PINS - 1; i >= 0; i--) {
1751         if (sphb->lsi_table[i].irq) {
1752             spapr_irq_free(spapr, sphb->lsi_table[i].irq, 1);
1753             sphb->lsi_table[i].irq = 0;
1754         }
1755     }
1756 
1757     QLIST_REMOVE(sphb, list);
1758 
1759     memory_region_del_subregion(&sphb->iommu_root, &sphb->msiwindow);
1760 
1761     /*
1762      * An attached PCI device may have memory listeners, eg. VFIO PCI. We have
1763      * unmapped all sections. Remove the listeners now, before destroying the
1764      * address space.
1765      */
1766     address_space_remove_listeners(&sphb->iommu_as);
1767     address_space_destroy(&sphb->iommu_as);
1768 
1769     qbus_set_hotplug_handler(BUS(phb->bus), NULL, &error_abort);
1770     pci_unregister_root_bus(phb->bus);
1771 
1772     memory_region_del_subregion(get_system_memory(), &sphb->iowindow);
1773     if (sphb->mem64_win_pciaddr != (hwaddr)-1) {
1774         memory_region_del_subregion(get_system_memory(), &sphb->mem64window);
1775     }
1776     memory_region_del_subregion(get_system_memory(), &sphb->mem32window);
1777 }
1778 
1779 static void spapr_phb_destroy_msi(gpointer opaque)
1780 {
1781     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
1782     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
1783     SpaprPciMsi *msi = opaque;
1784 
1785     if (!smc->legacy_irq_allocation) {
1786         spapr_irq_msi_free(spapr, msi->first_irq, msi->num);
1787     }
1788     spapr_irq_free(spapr, msi->first_irq, msi->num);
1789     g_free(msi);
1790 }
1791 
1792 static void spapr_phb_realize(DeviceState *dev, Error **errp)
1793 {
1794     /* We don't use SPAPR_MACHINE() in order to exit gracefully if the user
1795      * tries to add a sPAPR PHB to a non-pseries machine.
1796      */
1797     SpaprMachineState *spapr =
1798         (SpaprMachineState *) object_dynamic_cast(qdev_get_machine(),
1799                                                   TYPE_SPAPR_MACHINE);
1800     SpaprMachineClass *smc = spapr ? SPAPR_MACHINE_GET_CLASS(spapr) : NULL;
1801     SysBusDevice *s = SYS_BUS_DEVICE(dev);
1802     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(s);
1803     PCIHostState *phb = PCI_HOST_BRIDGE(s);
1804     MachineState *ms = MACHINE(spapr);
1805     char *namebuf;
1806     int i;
1807     PCIBus *bus;
1808     uint64_t msi_window_size = 4096;
1809     SpaprTceTable *tcet;
1810     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
1811     Error *local_err = NULL;
1812 
1813     if (!spapr) {
1814         error_setg(errp, TYPE_SPAPR_PCI_HOST_BRIDGE " needs a pseries machine");
1815         return;
1816     }
1817 
1818     assert(sphb->index != (uint32_t)-1); /* checked in spapr_phb_pre_plug() */
1819 
1820     if (sphb->mem64_win_size != 0) {
1821         if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) {
1822             error_setg(errp, "32-bit memory window of size 0x%"HWADDR_PRIx
1823                        " (max 2 GiB)", sphb->mem_win_size);
1824             return;
1825         }
1826 
1827         /* 64-bit window defaults to identity mapping */
1828         sphb->mem64_win_pciaddr = sphb->mem64_win_addr;
1829     } else if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) {
1830         /*
1831          * For compatibility with old configuration, if no 64-bit MMIO
1832          * window is specified, but the ordinary (32-bit) memory
1833          * window is specified as > 2GiB, we treat it as a 2GiB 32-bit
1834          * window, with a 64-bit MMIO window following on immediately
1835          * afterwards
1836          */
1837         sphb->mem64_win_size = sphb->mem_win_size - SPAPR_PCI_MEM32_WIN_SIZE;
1838         sphb->mem64_win_addr = sphb->mem_win_addr + SPAPR_PCI_MEM32_WIN_SIZE;
1839         sphb->mem64_win_pciaddr =
1840             SPAPR_PCI_MEM_WIN_BUS_OFFSET + SPAPR_PCI_MEM32_WIN_SIZE;
1841         sphb->mem_win_size = SPAPR_PCI_MEM32_WIN_SIZE;
1842     }
1843 
1844     if (spapr_pci_find_phb(spapr, sphb->buid)) {
1845         SpaprPhbState *s;
1846 
1847         error_setg(errp, "PCI host bridges must have unique indexes");
1848         error_append_hint(errp, "The following indexes are already in use:");
1849         QLIST_FOREACH(s, &spapr->phbs, list) {
1850             error_append_hint(errp, " %d", s->index);
1851         }
1852         error_append_hint(errp, "\nTry another value for the index property\n");
1853         return;
1854     }
1855 
1856     if (sphb->numa_node != -1 &&
1857         (sphb->numa_node >= MAX_NODES ||
1858          !ms->numa_state->nodes[sphb->numa_node].present)) {
1859         error_setg(errp, "Invalid NUMA node ID for PCI host bridge");
1860         return;
1861     }
1862 
1863     sphb->dtbusname = g_strdup_printf("pci@%" PRIx64, sphb->buid);
1864 
1865     /* Initialize memory regions */
1866     namebuf = g_strdup_printf("%s.mmio", sphb->dtbusname);
1867     memory_region_init(&sphb->memspace, OBJECT(sphb), namebuf, UINT64_MAX);
1868     g_free(namebuf);
1869 
1870     namebuf = g_strdup_printf("%s.mmio32-alias", sphb->dtbusname);
1871     memory_region_init_alias(&sphb->mem32window, OBJECT(sphb),
1872                              namebuf, &sphb->memspace,
1873                              SPAPR_PCI_MEM_WIN_BUS_OFFSET, sphb->mem_win_size);
1874     g_free(namebuf);
1875     memory_region_add_subregion(get_system_memory(), sphb->mem_win_addr,
1876                                 &sphb->mem32window);
1877 
1878     if (sphb->mem64_win_size != 0) {
1879         namebuf = g_strdup_printf("%s.mmio64-alias", sphb->dtbusname);
1880         memory_region_init_alias(&sphb->mem64window, OBJECT(sphb),
1881                                  namebuf, &sphb->memspace,
1882                                  sphb->mem64_win_pciaddr, sphb->mem64_win_size);
1883         g_free(namebuf);
1884 
1885         memory_region_add_subregion(get_system_memory(),
1886                                     sphb->mem64_win_addr,
1887                                     &sphb->mem64window);
1888     }
1889 
1890     /* Initialize IO regions */
1891     namebuf = g_strdup_printf("%s.io", sphb->dtbusname);
1892     memory_region_init(&sphb->iospace, OBJECT(sphb),
1893                        namebuf, SPAPR_PCI_IO_WIN_SIZE);
1894     g_free(namebuf);
1895 
1896     namebuf = g_strdup_printf("%s.io-alias", sphb->dtbusname);
1897     memory_region_init_alias(&sphb->iowindow, OBJECT(sphb), namebuf,
1898                              &sphb->iospace, 0, SPAPR_PCI_IO_WIN_SIZE);
1899     g_free(namebuf);
1900     memory_region_add_subregion(get_system_memory(), sphb->io_win_addr,
1901                                 &sphb->iowindow);
1902 
1903     bus = pci_register_root_bus(dev, NULL,
1904                                 pci_spapr_set_irq, pci_swizzle_map_irq_fn, sphb,
1905                                 &sphb->memspace, &sphb->iospace,
1906                                 PCI_DEVFN(0, 0), PCI_NUM_PINS,
1907                                 TYPE_PCI_BUS);
1908 
1909     /*
1910      * Despite resembling a vanilla PCI bus in most ways, the PAPR
1911      * para-virtualized PCI bus *does* permit PCI-E extended config
1912      * space access
1913      */
1914     if (sphb->pcie_ecs) {
1915         bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
1916     }
1917     phb->bus = bus;
1918     qbus_set_hotplug_handler(BUS(phb->bus), OBJECT(sphb), NULL);
1919 
1920     /*
1921      * Initialize PHB address space.
1922      * By default there will be at least one subregion for default
1923      * 32bit DMA window.
1924      * Later the guest might want to create another DMA window
1925      * which will become another memory subregion.
1926      */
1927     namebuf = g_strdup_printf("%s.iommu-root", sphb->dtbusname);
1928     memory_region_init(&sphb->iommu_root, OBJECT(sphb),
1929                        namebuf, UINT64_MAX);
1930     g_free(namebuf);
1931     address_space_init(&sphb->iommu_as, &sphb->iommu_root,
1932                        sphb->dtbusname);
1933 
1934     /*
1935      * As MSI/MSIX interrupts trigger by writing at MSI/MSIX vectors,
1936      * we need to allocate some memory to catch those writes coming
1937      * from msi_notify()/msix_notify().
1938      * As MSIMessage:addr is going to be the same and MSIMessage:data
1939      * is going to be a VIRQ number, 4 bytes of the MSI MR will only
1940      * be used.
1941      *
1942      * For KVM we want to ensure that this memory is a full page so that
1943      * our memory slot is of page size granularity.
1944      */
1945     if (kvm_enabled()) {
1946         msi_window_size = qemu_real_host_page_size;
1947     }
1948 
1949     memory_region_init_io(&sphb->msiwindow, OBJECT(sphb), &spapr_msi_ops, spapr,
1950                           "msi", msi_window_size);
1951     memory_region_add_subregion(&sphb->iommu_root, SPAPR_PCI_MSI_WINDOW,
1952                                 &sphb->msiwindow);
1953 
1954     pci_setup_iommu(bus, spapr_pci_dma_iommu, sphb);
1955 
1956     pci_bus_set_route_irq_fn(bus, spapr_route_intx_pin_to_irq);
1957 
1958     QLIST_INSERT_HEAD(&spapr->phbs, sphb, list);
1959 
1960     /* Initialize the LSI table */
1961     for (i = 0; i < PCI_NUM_PINS; i++) {
1962         uint32_t irq = SPAPR_IRQ_PCI_LSI + sphb->index * PCI_NUM_PINS + i;
1963 
1964         if (smc->legacy_irq_allocation) {
1965             irq = spapr_irq_findone(spapr, &local_err);
1966             if (local_err) {
1967                 error_propagate_prepend(errp, local_err,
1968                                         "can't allocate LSIs: ");
1969                 /*
1970                  * Older machines will never support PHB hotplug, ie, this is an
1971                  * init only path and QEMU will terminate. No need to rollback.
1972                  */
1973                 return;
1974             }
1975         }
1976 
1977         spapr_irq_claim(spapr, irq, true, &local_err);
1978         if (local_err) {
1979             error_propagate_prepend(errp, local_err, "can't allocate LSIs: ");
1980             goto unrealize;
1981         }
1982 
1983         sphb->lsi_table[i].irq = irq;
1984     }
1985 
1986     /* allocate connectors for child PCI devices */
1987     add_drcs(sphb, phb->bus, &local_err);
1988     if (local_err) {
1989         error_propagate(errp, local_err);
1990         goto unrealize;
1991     }
1992 
1993     /* DMA setup */
1994     for (i = 0; i < windows_supported; ++i) {
1995         tcet = spapr_tce_new_table(DEVICE(sphb), sphb->dma_liobn[i]);
1996         if (!tcet) {
1997             error_setg(errp, "Creating window#%d failed for %s",
1998                        i, sphb->dtbusname);
1999             goto unrealize;
2000         }
2001         memory_region_add_subregion(&sphb->iommu_root, 0,
2002                                     spapr_tce_get_iommu(tcet));
2003     }
2004 
2005     sphb->msi = g_hash_table_new_full(g_int_hash, g_int_equal, g_free,
2006                                       spapr_phb_destroy_msi);
2007     return;
2008 
2009 unrealize:
2010     spapr_phb_unrealize(dev);
2011 }
2012 
2013 static int spapr_phb_children_reset(Object *child, void *opaque)
2014 {
2015     DeviceState *dev = (DeviceState *) object_dynamic_cast(child, TYPE_DEVICE);
2016 
2017     if (dev) {
2018         device_legacy_reset(dev);
2019     }
2020 
2021     return 0;
2022 }
2023 
2024 void spapr_phb_dma_reset(SpaprPhbState *sphb)
2025 {
2026     int i;
2027     SpaprTceTable *tcet;
2028 
2029     for (i = 0; i < SPAPR_PCI_DMA_MAX_WINDOWS; ++i) {
2030         tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[i]);
2031 
2032         if (tcet && tcet->nb_table) {
2033             spapr_tce_table_disable(tcet);
2034         }
2035     }
2036 
2037     /* Register default 32bit DMA window */
2038     tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[0]);
2039     spapr_tce_table_enable(tcet, SPAPR_TCE_PAGE_SHIFT, sphb->dma_win_addr,
2040                            sphb->dma_win_size >> SPAPR_TCE_PAGE_SHIFT);
2041 }
2042 
2043 static void spapr_phb_reset(DeviceState *qdev)
2044 {
2045     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(qdev);
2046     Error *err = NULL;
2047 
2048     spapr_phb_dma_reset(sphb);
2049     spapr_phb_nvgpu_free(sphb);
2050     spapr_phb_nvgpu_setup(sphb, &err);
2051     if (err) {
2052         error_report_err(err);
2053     }
2054 
2055     /* Reset the IOMMU state */
2056     object_child_foreach(OBJECT(qdev), spapr_phb_children_reset, NULL);
2057 
2058     if (spapr_phb_eeh_available(SPAPR_PCI_HOST_BRIDGE(qdev))) {
2059         spapr_phb_vfio_reset(qdev);
2060     }
2061 
2062     g_hash_table_remove_all(sphb->msi);
2063 }
2064 
2065 static Property spapr_phb_properties[] = {
2066     DEFINE_PROP_UINT32("index", SpaprPhbState, index, -1),
2067     DEFINE_PROP_UINT64("mem_win_size", SpaprPhbState, mem_win_size,
2068                        SPAPR_PCI_MEM32_WIN_SIZE),
2069     DEFINE_PROP_UINT64("mem64_win_size", SpaprPhbState, mem64_win_size,
2070                        SPAPR_PCI_MEM64_WIN_SIZE),
2071     DEFINE_PROP_UINT64("io_win_size", SpaprPhbState, io_win_size,
2072                        SPAPR_PCI_IO_WIN_SIZE),
2073     DEFINE_PROP_BOOL("dynamic-reconfiguration", SpaprPhbState, dr_enabled,
2074                      true),
2075     /* Default DMA window is 0..1GB */
2076     DEFINE_PROP_UINT64("dma_win_addr", SpaprPhbState, dma_win_addr, 0),
2077     DEFINE_PROP_UINT64("dma_win_size", SpaprPhbState, dma_win_size, 0x40000000),
2078     DEFINE_PROP_UINT64("dma64_win_addr", SpaprPhbState, dma64_win_addr,
2079                        0x800000000000000ULL),
2080     DEFINE_PROP_BOOL("ddw", SpaprPhbState, ddw_enabled, true),
2081     DEFINE_PROP_UINT64("pgsz", SpaprPhbState, page_size_mask,
2082                        (1ULL << 12) | (1ULL << 16)
2083                        | (1ULL << 21) | (1ULL << 24)),
2084     DEFINE_PROP_UINT32("numa_node", SpaprPhbState, numa_node, -1),
2085     DEFINE_PROP_BOOL("pre-2.8-migration", SpaprPhbState,
2086                      pre_2_8_migration, false),
2087     DEFINE_PROP_BOOL("pcie-extended-configuration-space", SpaprPhbState,
2088                      pcie_ecs, true),
2089     DEFINE_PROP_UINT64("gpa", SpaprPhbState, nv2_gpa_win_addr, 0),
2090     DEFINE_PROP_UINT64("atsd", SpaprPhbState, nv2_atsd_win_addr, 0),
2091     DEFINE_PROP_END_OF_LIST(),
2092 };
2093 
2094 static const VMStateDescription vmstate_spapr_pci_lsi = {
2095     .name = "spapr_pci/lsi",
2096     .version_id = 1,
2097     .minimum_version_id = 1,
2098     .fields = (VMStateField[]) {
2099         VMSTATE_UINT32_EQUAL(irq, SpaprPciLsi, NULL),
2100 
2101         VMSTATE_END_OF_LIST()
2102     },
2103 };
2104 
2105 static const VMStateDescription vmstate_spapr_pci_msi = {
2106     .name = "spapr_pci/msi",
2107     .version_id = 1,
2108     .minimum_version_id = 1,
2109     .fields = (VMStateField []) {
2110         VMSTATE_UINT32(key, SpaprPciMsiMig),
2111         VMSTATE_UINT32(value.first_irq, SpaprPciMsiMig),
2112         VMSTATE_UINT32(value.num, SpaprPciMsiMig),
2113         VMSTATE_END_OF_LIST()
2114     },
2115 };
2116 
2117 static int spapr_pci_pre_save(void *opaque)
2118 {
2119     SpaprPhbState *sphb = opaque;
2120     GHashTableIter iter;
2121     gpointer key, value;
2122     int i;
2123 
2124     if (sphb->pre_2_8_migration) {
2125         sphb->mig_liobn = sphb->dma_liobn[0];
2126         sphb->mig_mem_win_addr = sphb->mem_win_addr;
2127         sphb->mig_mem_win_size = sphb->mem_win_size;
2128         sphb->mig_io_win_addr = sphb->io_win_addr;
2129         sphb->mig_io_win_size = sphb->io_win_size;
2130 
2131         if ((sphb->mem64_win_size != 0)
2132             && (sphb->mem64_win_addr
2133                 == (sphb->mem_win_addr + sphb->mem_win_size))) {
2134             sphb->mig_mem_win_size += sphb->mem64_win_size;
2135         }
2136     }
2137 
2138     g_free(sphb->msi_devs);
2139     sphb->msi_devs = NULL;
2140     sphb->msi_devs_num = g_hash_table_size(sphb->msi);
2141     if (!sphb->msi_devs_num) {
2142         return 0;
2143     }
2144     sphb->msi_devs = g_new(SpaprPciMsiMig, sphb->msi_devs_num);
2145 
2146     g_hash_table_iter_init(&iter, sphb->msi);
2147     for (i = 0; g_hash_table_iter_next(&iter, &key, &value); ++i) {
2148         sphb->msi_devs[i].key = *(uint32_t *) key;
2149         sphb->msi_devs[i].value = *(SpaprPciMsi *) value;
2150     }
2151 
2152     return 0;
2153 }
2154 
2155 static int spapr_pci_post_load(void *opaque, int version_id)
2156 {
2157     SpaprPhbState *sphb = opaque;
2158     gpointer key, value;
2159     int i;
2160 
2161     for (i = 0; i < sphb->msi_devs_num; ++i) {
2162         key = g_memdup(&sphb->msi_devs[i].key,
2163                        sizeof(sphb->msi_devs[i].key));
2164         value = g_memdup(&sphb->msi_devs[i].value,
2165                          sizeof(sphb->msi_devs[i].value));
2166         g_hash_table_insert(sphb->msi, key, value);
2167     }
2168     g_free(sphb->msi_devs);
2169     sphb->msi_devs = NULL;
2170     sphb->msi_devs_num = 0;
2171 
2172     return 0;
2173 }
2174 
2175 static bool pre_2_8_migration(void *opaque, int version_id)
2176 {
2177     SpaprPhbState *sphb = opaque;
2178 
2179     return sphb->pre_2_8_migration;
2180 }
2181 
2182 static const VMStateDescription vmstate_spapr_pci = {
2183     .name = "spapr_pci",
2184     .version_id = 2,
2185     .minimum_version_id = 2,
2186     .pre_save = spapr_pci_pre_save,
2187     .post_load = spapr_pci_post_load,
2188     .fields = (VMStateField[]) {
2189         VMSTATE_UINT64_EQUAL(buid, SpaprPhbState, NULL),
2190         VMSTATE_UINT32_TEST(mig_liobn, SpaprPhbState, pre_2_8_migration),
2191         VMSTATE_UINT64_TEST(mig_mem_win_addr, SpaprPhbState, pre_2_8_migration),
2192         VMSTATE_UINT64_TEST(mig_mem_win_size, SpaprPhbState, pre_2_8_migration),
2193         VMSTATE_UINT64_TEST(mig_io_win_addr, SpaprPhbState, pre_2_8_migration),
2194         VMSTATE_UINT64_TEST(mig_io_win_size, SpaprPhbState, pre_2_8_migration),
2195         VMSTATE_STRUCT_ARRAY(lsi_table, SpaprPhbState, PCI_NUM_PINS, 0,
2196                              vmstate_spapr_pci_lsi, SpaprPciLsi),
2197         VMSTATE_INT32(msi_devs_num, SpaprPhbState),
2198         VMSTATE_STRUCT_VARRAY_ALLOC(msi_devs, SpaprPhbState, msi_devs_num, 0,
2199                                     vmstate_spapr_pci_msi, SpaprPciMsiMig),
2200         VMSTATE_END_OF_LIST()
2201     },
2202 };
2203 
2204 static const char *spapr_phb_root_bus_path(PCIHostState *host_bridge,
2205                                            PCIBus *rootbus)
2206 {
2207     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(host_bridge);
2208 
2209     return sphb->dtbusname;
2210 }
2211 
2212 static void spapr_phb_class_init(ObjectClass *klass, void *data)
2213 {
2214     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(klass);
2215     DeviceClass *dc = DEVICE_CLASS(klass);
2216     HotplugHandlerClass *hp = HOTPLUG_HANDLER_CLASS(klass);
2217 
2218     hc->root_bus_path = spapr_phb_root_bus_path;
2219     dc->realize = spapr_phb_realize;
2220     dc->unrealize = spapr_phb_unrealize;
2221     device_class_set_props(dc, spapr_phb_properties);
2222     dc->reset = spapr_phb_reset;
2223     dc->vmsd = &vmstate_spapr_pci;
2224     /* Supported by TYPE_SPAPR_MACHINE */
2225     dc->user_creatable = true;
2226     set_bit(DEVICE_CATEGORY_BRIDGE, dc->categories);
2227     hp->plug = spapr_pci_plug;
2228     hp->unplug = spapr_pci_unplug;
2229     hp->unplug_request = spapr_pci_unplug_request;
2230 }
2231 
2232 static const TypeInfo spapr_phb_info = {
2233     .name          = TYPE_SPAPR_PCI_HOST_BRIDGE,
2234     .parent        = TYPE_PCI_HOST_BRIDGE,
2235     .instance_size = sizeof(SpaprPhbState),
2236     .instance_finalize = spapr_phb_finalizefn,
2237     .class_init    = spapr_phb_class_init,
2238     .interfaces    = (InterfaceInfo[]) {
2239         { TYPE_HOTPLUG_HANDLER },
2240         { }
2241     }
2242 };
2243 
2244 static void spapr_phb_pci_enumerate_bridge(PCIBus *bus, PCIDevice *pdev,
2245                                            void *opaque)
2246 {
2247     unsigned int *bus_no = opaque;
2248     PCIBus *sec_bus = NULL;
2249 
2250     if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) !=
2251          PCI_HEADER_TYPE_BRIDGE)) {
2252         return;
2253     }
2254 
2255     (*bus_no)++;
2256     pci_default_write_config(pdev, PCI_PRIMARY_BUS, pci_dev_bus_num(pdev), 1);
2257     pci_default_write_config(pdev, PCI_SECONDARY_BUS, *bus_no, 1);
2258     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
2259 
2260     sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev));
2261     if (!sec_bus) {
2262         return;
2263     }
2264 
2265     pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
2266                         spapr_phb_pci_enumerate_bridge, bus_no);
2267     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
2268 }
2269 
2270 static void spapr_phb_pci_enumerate(SpaprPhbState *phb)
2271 {
2272     PCIBus *bus = PCI_HOST_BRIDGE(phb)->bus;
2273     unsigned int bus_no = 0;
2274 
2275     pci_for_each_device(bus, pci_bus_num(bus),
2276                         spapr_phb_pci_enumerate_bridge,
2277                         &bus_no);
2278 
2279 }
2280 
2281 int spapr_dt_phb(SpaprMachineState *spapr, SpaprPhbState *phb,
2282                  uint32_t intc_phandle, void *fdt, int *node_offset)
2283 {
2284     int bus_off, i, j, ret;
2285     uint32_t bus_range[] = { cpu_to_be32(0), cpu_to_be32(0xff) };
2286     struct {
2287         uint32_t hi;
2288         uint64_t child;
2289         uint64_t parent;
2290         uint64_t size;
2291     } QEMU_PACKED ranges[] = {
2292         {
2293             cpu_to_be32(b_ss(1)), cpu_to_be64(0),
2294             cpu_to_be64(phb->io_win_addr),
2295             cpu_to_be64(memory_region_size(&phb->iospace)),
2296         },
2297         {
2298             cpu_to_be32(b_ss(2)), cpu_to_be64(SPAPR_PCI_MEM_WIN_BUS_OFFSET),
2299             cpu_to_be64(phb->mem_win_addr),
2300             cpu_to_be64(phb->mem_win_size),
2301         },
2302         {
2303             cpu_to_be32(b_ss(3)), cpu_to_be64(phb->mem64_win_pciaddr),
2304             cpu_to_be64(phb->mem64_win_addr),
2305             cpu_to_be64(phb->mem64_win_size),
2306         },
2307     };
2308     const unsigned sizeof_ranges =
2309         (phb->mem64_win_size ? 3 : 2) * sizeof(ranges[0]);
2310     uint64_t bus_reg[] = { cpu_to_be64(phb->buid), 0 };
2311     uint32_t interrupt_map_mask[] = {
2312         cpu_to_be32(b_ddddd(-1)|b_fff(0)), 0x0, 0x0, cpu_to_be32(-1)};
2313     uint32_t interrupt_map[PCI_SLOT_MAX * PCI_NUM_PINS][7];
2314     uint32_t ddw_applicable[] = {
2315         cpu_to_be32(RTAS_IBM_QUERY_PE_DMA_WINDOW),
2316         cpu_to_be32(RTAS_IBM_CREATE_PE_DMA_WINDOW),
2317         cpu_to_be32(RTAS_IBM_REMOVE_PE_DMA_WINDOW)
2318     };
2319     uint32_t ddw_extensions[] = {
2320         cpu_to_be32(1),
2321         cpu_to_be32(RTAS_IBM_RESET_PE_DMA_WINDOW)
2322     };
2323     uint32_t associativity[] = {cpu_to_be32(0x4),
2324                                 cpu_to_be32(0x0),
2325                                 cpu_to_be32(0x0),
2326                                 cpu_to_be32(0x0),
2327                                 cpu_to_be32(phb->numa_node)};
2328     SpaprTceTable *tcet;
2329     SpaprDrc *drc;
2330     Error *err = NULL;
2331 
2332     /* Start populating the FDT */
2333     _FDT(bus_off = fdt_add_subnode(fdt, 0, phb->dtbusname));
2334     if (node_offset) {
2335         *node_offset = bus_off;
2336     }
2337 
2338     /* Write PHB properties */
2339     _FDT(fdt_setprop_string(fdt, bus_off, "device_type", "pci"));
2340     _FDT(fdt_setprop_string(fdt, bus_off, "compatible", "IBM,Logical_PHB"));
2341     _FDT(fdt_setprop_cell(fdt, bus_off, "#interrupt-cells", 0x1));
2342     _FDT(fdt_setprop(fdt, bus_off, "used-by-rtas", NULL, 0));
2343     _FDT(fdt_setprop(fdt, bus_off, "bus-range", &bus_range, sizeof(bus_range)));
2344     _FDT(fdt_setprop(fdt, bus_off, "ranges", &ranges, sizeof_ranges));
2345     _FDT(fdt_setprop(fdt, bus_off, "reg", &bus_reg, sizeof(bus_reg)));
2346     _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pci-config-space-type", 0x1));
2347     _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pe-total-#msi",
2348                           spapr_irq_nr_msis(spapr)));
2349 
2350     /* Dynamic DMA window */
2351     if (phb->ddw_enabled) {
2352         _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-applicable", &ddw_applicable,
2353                          sizeof(ddw_applicable)));
2354         _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-extensions",
2355                          &ddw_extensions, sizeof(ddw_extensions)));
2356     }
2357 
2358     /* Advertise NUMA via ibm,associativity */
2359     if (phb->numa_node != -1) {
2360         _FDT(fdt_setprop(fdt, bus_off, "ibm,associativity", associativity,
2361                          sizeof(associativity)));
2362     }
2363 
2364     /* Build the interrupt-map, this must matches what is done
2365      * in pci_swizzle_map_irq_fn
2366      */
2367     _FDT(fdt_setprop(fdt, bus_off, "interrupt-map-mask",
2368                      &interrupt_map_mask, sizeof(interrupt_map_mask)));
2369     for (i = 0; i < PCI_SLOT_MAX; i++) {
2370         for (j = 0; j < PCI_NUM_PINS; j++) {
2371             uint32_t *irqmap = interrupt_map[i*PCI_NUM_PINS + j];
2372             int lsi_num = pci_swizzle(i, j);
2373 
2374             irqmap[0] = cpu_to_be32(b_ddddd(i)|b_fff(0));
2375             irqmap[1] = 0;
2376             irqmap[2] = 0;
2377             irqmap[3] = cpu_to_be32(j+1);
2378             irqmap[4] = cpu_to_be32(intc_phandle);
2379             spapr_dt_irq(&irqmap[5], phb->lsi_table[lsi_num].irq, true);
2380         }
2381     }
2382     /* Write interrupt map */
2383     _FDT(fdt_setprop(fdt, bus_off, "interrupt-map", &interrupt_map,
2384                      sizeof(interrupt_map)));
2385 
2386     tcet = spapr_tce_find_by_liobn(phb->dma_liobn[0]);
2387     if (!tcet) {
2388         return -1;
2389     }
2390     spapr_dma_dt(fdt, bus_off, "ibm,dma-window",
2391                  tcet->liobn, tcet->bus_offset,
2392                  tcet->nb_table << tcet->page_shift);
2393 
2394     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, phb->index);
2395     if (drc) {
2396         uint32_t drc_index = cpu_to_be32(spapr_drc_index(drc));
2397 
2398         _FDT(fdt_setprop(fdt, bus_off, "ibm,my-drc-index", &drc_index,
2399                          sizeof(drc_index)));
2400     }
2401 
2402     /* Walk the bridges and program the bus numbers*/
2403     spapr_phb_pci_enumerate(phb);
2404     _FDT(fdt_setprop_cell(fdt, bus_off, "qemu,phb-enumerated", 0x1));
2405 
2406     /* Walk the bridge and subordinate buses */
2407     ret = spapr_dt_pci_bus(phb, PCI_HOST_BRIDGE(phb)->bus, fdt, bus_off);
2408     if (ret < 0) {
2409         return ret;
2410     }
2411 
2412     spapr_phb_nvgpu_populate_dt(phb, fdt, bus_off, &err);
2413     if (err) {
2414         error_report_err(err);
2415     }
2416     spapr_phb_nvgpu_ram_populate_dt(phb, fdt);
2417 
2418     return 0;
2419 }
2420 
2421 void spapr_pci_rtas_init(void)
2422 {
2423     spapr_rtas_register(RTAS_READ_PCI_CONFIG, "read-pci-config",
2424                         rtas_read_pci_config);
2425     spapr_rtas_register(RTAS_WRITE_PCI_CONFIG, "write-pci-config",
2426                         rtas_write_pci_config);
2427     spapr_rtas_register(RTAS_IBM_READ_PCI_CONFIG, "ibm,read-pci-config",
2428                         rtas_ibm_read_pci_config);
2429     spapr_rtas_register(RTAS_IBM_WRITE_PCI_CONFIG, "ibm,write-pci-config",
2430                         rtas_ibm_write_pci_config);
2431     if (msi_nonbroken) {
2432         spapr_rtas_register(RTAS_IBM_QUERY_INTERRUPT_SOURCE_NUMBER,
2433                             "ibm,query-interrupt-source-number",
2434                             rtas_ibm_query_interrupt_source_number);
2435         spapr_rtas_register(RTAS_IBM_CHANGE_MSI, "ibm,change-msi",
2436                             rtas_ibm_change_msi);
2437     }
2438 
2439     spapr_rtas_register(RTAS_IBM_SET_EEH_OPTION,
2440                         "ibm,set-eeh-option",
2441                         rtas_ibm_set_eeh_option);
2442     spapr_rtas_register(RTAS_IBM_GET_CONFIG_ADDR_INFO2,
2443                         "ibm,get-config-addr-info2",
2444                         rtas_ibm_get_config_addr_info2);
2445     spapr_rtas_register(RTAS_IBM_READ_SLOT_RESET_STATE2,
2446                         "ibm,read-slot-reset-state2",
2447                         rtas_ibm_read_slot_reset_state2);
2448     spapr_rtas_register(RTAS_IBM_SET_SLOT_RESET,
2449                         "ibm,set-slot-reset",
2450                         rtas_ibm_set_slot_reset);
2451     spapr_rtas_register(RTAS_IBM_CONFIGURE_PE,
2452                         "ibm,configure-pe",
2453                         rtas_ibm_configure_pe);
2454     spapr_rtas_register(RTAS_IBM_SLOT_ERROR_DETAIL,
2455                         "ibm,slot-error-detail",
2456                         rtas_ibm_slot_error_detail);
2457 }
2458 
2459 static void spapr_pci_register_types(void)
2460 {
2461     type_register_static(&spapr_phb_info);
2462 }
2463 
2464 type_init(spapr_pci_register_types)
2465 
2466 static int spapr_switch_one_vga(DeviceState *dev, void *opaque)
2467 {
2468     bool be = *(bool *)opaque;
2469 
2470     if (object_dynamic_cast(OBJECT(dev), "VGA")
2471         || object_dynamic_cast(OBJECT(dev), "secondary-vga")) {
2472         object_property_set_bool(OBJECT(dev), be, "big-endian-framebuffer",
2473                                  &error_abort);
2474     }
2475     return 0;
2476 }
2477 
2478 void spapr_pci_switch_vga(bool big_endian)
2479 {
2480     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
2481     SpaprPhbState *sphb;
2482 
2483     /*
2484      * For backward compatibility with existing guests, we switch
2485      * the endianness of the VGA controller when changing the guest
2486      * interrupt mode
2487      */
2488     QLIST_FOREACH(sphb, &spapr->phbs, list) {
2489         BusState *bus = &PCI_HOST_BRIDGE(sphb)->bus->qbus;
2490         qbus_walk_children(bus, spapr_switch_one_vga, NULL, NULL, NULL,
2491                            &big_endian);
2492     }
2493 }
2494