xref: /openbmc/qemu/hw/ppc/spapr_pci.c (revision 646c5478)
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 #include "qemu/osdep.h"
26 #include "qapi/error.h"
27 #include "qemu-common.h"
28 #include "cpu.h"
29 #include "hw/hw.h"
30 #include "hw/sysbus.h"
31 #include "hw/pci/pci.h"
32 #include "hw/pci/msi.h"
33 #include "hw/pci/msix.h"
34 #include "hw/pci/pci_host.h"
35 #include "hw/ppc/spapr.h"
36 #include "hw/pci-host/spapr.h"
37 #include "exec/address-spaces.h"
38 #include "exec/ram_addr.h"
39 #include <libfdt.h>
40 #include "trace.h"
41 #include "qemu/error-report.h"
42 #include "qapi/qmp/qerror.h"
43 
44 #include "hw/pci/pci_bridge.h"
45 #include "hw/pci/pci_bus.h"
46 #include "hw/ppc/spapr_drc.h"
47 #include "sysemu/device_tree.h"
48 #include "sysemu/kvm.h"
49 #include "sysemu/hostmem.h"
50 #include "sysemu/numa.h"
51 
52 #include "hw/vfio/vfio.h"
53 
54 /* Copied from the kernel arch/powerpc/platforms/pseries/msi.c */
55 #define RTAS_QUERY_FN           0
56 #define RTAS_CHANGE_FN          1
57 #define RTAS_RESET_FN           2
58 #define RTAS_CHANGE_MSI_FN      3
59 #define RTAS_CHANGE_MSIX_FN     4
60 
61 /* Interrupt types to return on RTAS_CHANGE_* */
62 #define RTAS_TYPE_MSI           1
63 #define RTAS_TYPE_MSIX          2
64 
65 #define FDT_NAME_MAX          128
66 
67 #define _FDT(exp) \
68     do { \
69         int ret = (exp);                                           \
70         if (ret < 0) {                                             \
71             return ret;                                            \
72         }                                                          \
73     } while (0)
74 
75 sPAPRPHBState *spapr_pci_find_phb(sPAPRMachineState *spapr, uint64_t buid)
76 {
77     sPAPRPHBState *sphb;
78 
79     QLIST_FOREACH(sphb, &spapr->phbs, list) {
80         if (sphb->buid != buid) {
81             continue;
82         }
83         return sphb;
84     }
85 
86     return NULL;
87 }
88 
89 PCIDevice *spapr_pci_find_dev(sPAPRMachineState *spapr, uint64_t buid,
90                               uint32_t config_addr)
91 {
92     sPAPRPHBState *sphb = spapr_pci_find_phb(spapr, buid);
93     PCIHostState *phb = PCI_HOST_BRIDGE(sphb);
94     int bus_num = (config_addr >> 16) & 0xFF;
95     int devfn = (config_addr >> 8) & 0xFF;
96 
97     if (!phb) {
98         return NULL;
99     }
100 
101     return pci_find_device(phb->bus, bus_num, devfn);
102 }
103 
104 static uint32_t rtas_pci_cfgaddr(uint32_t arg)
105 {
106     /* This handles the encoding of extended config space addresses */
107     return ((arg >> 20) & 0xf00) | (arg & 0xff);
108 }
109 
110 static void finish_read_pci_config(sPAPRMachineState *spapr, uint64_t buid,
111                                    uint32_t addr, uint32_t size,
112                                    target_ulong rets)
113 {
114     PCIDevice *pci_dev;
115     uint32_t val;
116 
117     if ((size != 1) && (size != 2) && (size != 4)) {
118         /* access must be 1, 2 or 4 bytes */
119         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
120         return;
121     }
122 
123     pci_dev = spapr_pci_find_dev(spapr, buid, addr);
124     addr = rtas_pci_cfgaddr(addr);
125 
126     if (!pci_dev || (addr % size) || (addr >= pci_config_size(pci_dev))) {
127         /* Access must be to a valid device, within bounds and
128          * naturally aligned */
129         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
130         return;
131     }
132 
133     val = pci_host_config_read_common(pci_dev, addr,
134                                       pci_config_size(pci_dev), size);
135 
136     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
137     rtas_st(rets, 1, val);
138 }
139 
140 static void rtas_ibm_read_pci_config(PowerPCCPU *cpu, sPAPRMachineState *spapr,
141                                      uint32_t token, uint32_t nargs,
142                                      target_ulong args,
143                                      uint32_t nret, target_ulong rets)
144 {
145     uint64_t buid;
146     uint32_t size, addr;
147 
148     if ((nargs != 4) || (nret != 2)) {
149         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
150         return;
151     }
152 
153     buid = rtas_ldq(args, 1);
154     size = rtas_ld(args, 3);
155     addr = rtas_ld(args, 0);
156 
157     finish_read_pci_config(spapr, buid, addr, size, rets);
158 }
159 
160 static void rtas_read_pci_config(PowerPCCPU *cpu, sPAPRMachineState *spapr,
161                                  uint32_t token, uint32_t nargs,
162                                  target_ulong args,
163                                  uint32_t nret, target_ulong rets)
164 {
165     uint32_t size, addr;
166 
167     if ((nargs != 2) || (nret != 2)) {
168         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
169         return;
170     }
171 
172     size = rtas_ld(args, 1);
173     addr = rtas_ld(args, 0);
174 
175     finish_read_pci_config(spapr, 0, addr, size, rets);
176 }
177 
178 static void finish_write_pci_config(sPAPRMachineState *spapr, uint64_t buid,
179                                     uint32_t addr, uint32_t size,
180                                     uint32_t val, target_ulong rets)
181 {
182     PCIDevice *pci_dev;
183 
184     if ((size != 1) && (size != 2) && (size != 4)) {
185         /* access must be 1, 2 or 4 bytes */
186         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
187         return;
188     }
189 
190     pci_dev = spapr_pci_find_dev(spapr, buid, addr);
191     addr = rtas_pci_cfgaddr(addr);
192 
193     if (!pci_dev || (addr % size) || (addr >= pci_config_size(pci_dev))) {
194         /* Access must be to a valid device, within bounds and
195          * naturally aligned */
196         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
197         return;
198     }
199 
200     pci_host_config_write_common(pci_dev, addr, pci_config_size(pci_dev),
201                                  val, size);
202 
203     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
204 }
205 
206 static void rtas_ibm_write_pci_config(PowerPCCPU *cpu, sPAPRMachineState *spapr,
207                                       uint32_t token, uint32_t nargs,
208                                       target_ulong args,
209                                       uint32_t nret, target_ulong rets)
210 {
211     uint64_t buid;
212     uint32_t val, size, addr;
213 
214     if ((nargs != 5) || (nret != 1)) {
215         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
216         return;
217     }
218 
219     buid = rtas_ldq(args, 1);
220     val = rtas_ld(args, 4);
221     size = rtas_ld(args, 3);
222     addr = rtas_ld(args, 0);
223 
224     finish_write_pci_config(spapr, buid, addr, size, val, rets);
225 }
226 
227 static void rtas_write_pci_config(PowerPCCPU *cpu, sPAPRMachineState *spapr,
228                                   uint32_t token, uint32_t nargs,
229                                   target_ulong args,
230                                   uint32_t nret, target_ulong rets)
231 {
232     uint32_t val, size, addr;
233 
234     if ((nargs != 3) || (nret != 1)) {
235         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
236         return;
237     }
238 
239 
240     val = rtas_ld(args, 2);
241     size = rtas_ld(args, 1);
242     addr = rtas_ld(args, 0);
243 
244     finish_write_pci_config(spapr, 0, addr, size, val, rets);
245 }
246 
247 /*
248  * Set MSI/MSIX message data.
249  * This is required for msi_notify()/msix_notify() which
250  * will write at the addresses via spapr_msi_write().
251  *
252  * If hwaddr == 0, all entries will have .data == first_irq i.e.
253  * table will be reset.
254  */
255 static void spapr_msi_setmsg(PCIDevice *pdev, hwaddr addr, bool msix,
256                              unsigned first_irq, unsigned req_num)
257 {
258     unsigned i;
259     MSIMessage msg = { .address = addr, .data = first_irq };
260 
261     if (!msix) {
262         msi_set_message(pdev, msg);
263         trace_spapr_pci_msi_setup(pdev->name, 0, msg.address);
264         return;
265     }
266 
267     for (i = 0; i < req_num; ++i) {
268         msix_set_message(pdev, i, msg);
269         trace_spapr_pci_msi_setup(pdev->name, i, msg.address);
270         if (addr) {
271             ++msg.data;
272         }
273     }
274 }
275 
276 static void rtas_ibm_change_msi(PowerPCCPU *cpu, sPAPRMachineState *spapr,
277                                 uint32_t token, uint32_t nargs,
278                                 target_ulong args, uint32_t nret,
279                                 target_ulong rets)
280 {
281     uint32_t config_addr = rtas_ld(args, 0);
282     uint64_t buid = rtas_ldq(args, 1);
283     unsigned int func = rtas_ld(args, 3);
284     unsigned int req_num = rtas_ld(args, 4); /* 0 == remove all */
285     unsigned int seq_num = rtas_ld(args, 5);
286     unsigned int ret_intr_type;
287     unsigned int irq, max_irqs = 0;
288     sPAPRPHBState *phb = NULL;
289     PCIDevice *pdev = NULL;
290     spapr_pci_msi *msi;
291     int *config_addr_key;
292     Error *err = NULL;
293 
294     switch (func) {
295     case RTAS_CHANGE_MSI_FN:
296     case RTAS_CHANGE_FN:
297         ret_intr_type = RTAS_TYPE_MSI;
298         break;
299     case RTAS_CHANGE_MSIX_FN:
300         ret_intr_type = RTAS_TYPE_MSIX;
301         break;
302     default:
303         error_report("rtas_ibm_change_msi(%u) is not implemented", func);
304         rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
305         return;
306     }
307 
308     /* Fins sPAPRPHBState */
309     phb = spapr_pci_find_phb(spapr, buid);
310     if (phb) {
311         pdev = spapr_pci_find_dev(spapr, buid, config_addr);
312     }
313     if (!phb || !pdev) {
314         rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
315         return;
316     }
317 
318     msi = (spapr_pci_msi *) g_hash_table_lookup(phb->msi, &config_addr);
319 
320     /* Releasing MSIs */
321     if (!req_num) {
322         if (!msi) {
323             trace_spapr_pci_msi("Releasing wrong config", config_addr);
324             rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
325             return;
326         }
327 
328         xics_spapr_free(spapr->xics, msi->first_irq, msi->num);
329         if (msi_present(pdev)) {
330             spapr_msi_setmsg(pdev, 0, false, 0, 0);
331         }
332         if (msix_present(pdev)) {
333             spapr_msi_setmsg(pdev, 0, true, 0, 0);
334         }
335         g_hash_table_remove(phb->msi, &config_addr);
336 
337         trace_spapr_pci_msi("Released MSIs", config_addr);
338         rtas_st(rets, 0, RTAS_OUT_SUCCESS);
339         rtas_st(rets, 1, 0);
340         return;
341     }
342 
343     /* Enabling MSI */
344 
345     /* Check if the device supports as many IRQs as requested */
346     if (ret_intr_type == RTAS_TYPE_MSI) {
347         max_irqs = msi_nr_vectors_allocated(pdev);
348     } else if (ret_intr_type == RTAS_TYPE_MSIX) {
349         max_irqs = pdev->msix_entries_nr;
350     }
351     if (!max_irqs) {
352         error_report("Requested interrupt type %d is not enabled for device %x",
353                      ret_intr_type, config_addr);
354         rtas_st(rets, 0, -1); /* Hardware error */
355         return;
356     }
357     /* Correct the number if the guest asked for too many */
358     if (req_num > max_irqs) {
359         trace_spapr_pci_msi_retry(config_addr, req_num, max_irqs);
360         req_num = max_irqs;
361         irq = 0; /* to avoid misleading trace */
362         goto out;
363     }
364 
365     /* Allocate MSIs */
366     irq = xics_spapr_alloc_block(spapr->xics, req_num, false,
367                            ret_intr_type == RTAS_TYPE_MSI, &err);
368     if (err) {
369         error_reportf_err(err, "Can't allocate MSIs for device %x: ",
370                           config_addr);
371         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
372         return;
373     }
374 
375     /* Release previous MSIs */
376     if (msi) {
377         xics_spapr_free(spapr->xics, msi->first_irq, msi->num);
378         g_hash_table_remove(phb->msi, &config_addr);
379     }
380 
381     /* Setup MSI/MSIX vectors in the device (via cfgspace or MSIX BAR) */
382     spapr_msi_setmsg(pdev, SPAPR_PCI_MSI_WINDOW, ret_intr_type == RTAS_TYPE_MSIX,
383                      irq, req_num);
384 
385     /* Add MSI device to cache */
386     msi = g_new(spapr_pci_msi, 1);
387     msi->first_irq = irq;
388     msi->num = req_num;
389     config_addr_key = g_new(int, 1);
390     *config_addr_key = config_addr;
391     g_hash_table_insert(phb->msi, config_addr_key, msi);
392 
393 out:
394     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
395     rtas_st(rets, 1, req_num);
396     rtas_st(rets, 2, ++seq_num);
397     if (nret > 3) {
398         rtas_st(rets, 3, ret_intr_type);
399     }
400 
401     trace_spapr_pci_rtas_ibm_change_msi(config_addr, func, req_num, irq);
402 }
403 
404 static void rtas_ibm_query_interrupt_source_number(PowerPCCPU *cpu,
405                                                    sPAPRMachineState *spapr,
406                                                    uint32_t token,
407                                                    uint32_t nargs,
408                                                    target_ulong args,
409                                                    uint32_t nret,
410                                                    target_ulong rets)
411 {
412     uint32_t config_addr = rtas_ld(args, 0);
413     uint64_t buid = rtas_ldq(args, 1);
414     unsigned int intr_src_num = -1, ioa_intr_num = rtas_ld(args, 3);
415     sPAPRPHBState *phb = NULL;
416     PCIDevice *pdev = NULL;
417     spapr_pci_msi *msi;
418 
419     /* Find sPAPRPHBState */
420     phb = spapr_pci_find_phb(spapr, buid);
421     if (phb) {
422         pdev = spapr_pci_find_dev(spapr, buid, config_addr);
423     }
424     if (!phb || !pdev) {
425         rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
426         return;
427     }
428 
429     /* Find device descriptor and start IRQ */
430     msi = (spapr_pci_msi *) g_hash_table_lookup(phb->msi, &config_addr);
431     if (!msi || !msi->first_irq || !msi->num || (ioa_intr_num >= msi->num)) {
432         trace_spapr_pci_msi("Failed to return vector", config_addr);
433         rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
434         return;
435     }
436     intr_src_num = msi->first_irq + ioa_intr_num;
437     trace_spapr_pci_rtas_ibm_query_interrupt_source_number(ioa_intr_num,
438                                                            intr_src_num);
439 
440     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
441     rtas_st(rets, 1, intr_src_num);
442     rtas_st(rets, 2, 1);/* 0 == level; 1 == edge */
443 }
444 
445 static void rtas_ibm_set_eeh_option(PowerPCCPU *cpu,
446                                     sPAPRMachineState *spapr,
447                                     uint32_t token, uint32_t nargs,
448                                     target_ulong args, uint32_t nret,
449                                     target_ulong rets)
450 {
451     sPAPRPHBState *sphb;
452     uint32_t addr, option;
453     uint64_t buid;
454     int ret;
455 
456     if ((nargs != 4) || (nret != 1)) {
457         goto param_error_exit;
458     }
459 
460     buid = rtas_ldq(args, 1);
461     addr = rtas_ld(args, 0);
462     option = rtas_ld(args, 3);
463 
464     sphb = spapr_pci_find_phb(spapr, buid);
465     if (!sphb) {
466         goto param_error_exit;
467     }
468 
469     if (!spapr_phb_eeh_available(sphb)) {
470         goto param_error_exit;
471     }
472 
473     ret = spapr_phb_vfio_eeh_set_option(sphb, addr, option);
474     rtas_st(rets, 0, ret);
475     return;
476 
477 param_error_exit:
478     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
479 }
480 
481 static void rtas_ibm_get_config_addr_info2(PowerPCCPU *cpu,
482                                            sPAPRMachineState *spapr,
483                                            uint32_t token, uint32_t nargs,
484                                            target_ulong args, uint32_t nret,
485                                            target_ulong rets)
486 {
487     sPAPRPHBState *sphb;
488     PCIDevice *pdev;
489     uint32_t addr, option;
490     uint64_t buid;
491 
492     if ((nargs != 4) || (nret != 2)) {
493         goto param_error_exit;
494     }
495 
496     buid = rtas_ldq(args, 1);
497     sphb = spapr_pci_find_phb(spapr, buid);
498     if (!sphb) {
499         goto param_error_exit;
500     }
501 
502     if (!spapr_phb_eeh_available(sphb)) {
503         goto param_error_exit;
504     }
505 
506     /*
507      * We always have PE address of form "00BB0001". "BB"
508      * represents the bus number of PE's primary bus.
509      */
510     option = rtas_ld(args, 3);
511     switch (option) {
512     case RTAS_GET_PE_ADDR:
513         addr = rtas_ld(args, 0);
514         pdev = spapr_pci_find_dev(spapr, buid, addr);
515         if (!pdev) {
516             goto param_error_exit;
517         }
518 
519         rtas_st(rets, 1, (pci_bus_num(pdev->bus) << 16) + 1);
520         break;
521     case RTAS_GET_PE_MODE:
522         rtas_st(rets, 1, RTAS_PE_MODE_SHARED);
523         break;
524     default:
525         goto param_error_exit;
526     }
527 
528     rtas_st(rets, 0, RTAS_OUT_SUCCESS);
529     return;
530 
531 param_error_exit:
532     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
533 }
534 
535 static void rtas_ibm_read_slot_reset_state2(PowerPCCPU *cpu,
536                                             sPAPRMachineState *spapr,
537                                             uint32_t token, uint32_t nargs,
538                                             target_ulong args, uint32_t nret,
539                                             target_ulong rets)
540 {
541     sPAPRPHBState *sphb;
542     uint64_t buid;
543     int state, ret;
544 
545     if ((nargs != 3) || (nret != 4 && nret != 5)) {
546         goto param_error_exit;
547     }
548 
549     buid = rtas_ldq(args, 1);
550     sphb = spapr_pci_find_phb(spapr, buid);
551     if (!sphb) {
552         goto param_error_exit;
553     }
554 
555     if (!spapr_phb_eeh_available(sphb)) {
556         goto param_error_exit;
557     }
558 
559     ret = spapr_phb_vfio_eeh_get_state(sphb, &state);
560     rtas_st(rets, 0, ret);
561     if (ret != RTAS_OUT_SUCCESS) {
562         return;
563     }
564 
565     rtas_st(rets, 1, state);
566     rtas_st(rets, 2, RTAS_EEH_SUPPORT);
567     rtas_st(rets, 3, RTAS_EEH_PE_UNAVAIL_INFO);
568     if (nret >= 5) {
569         rtas_st(rets, 4, RTAS_EEH_PE_RECOVER_INFO);
570     }
571     return;
572 
573 param_error_exit:
574     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
575 }
576 
577 static void rtas_ibm_set_slot_reset(PowerPCCPU *cpu,
578                                     sPAPRMachineState *spapr,
579                                     uint32_t token, uint32_t nargs,
580                                     target_ulong args, uint32_t nret,
581                                     target_ulong rets)
582 {
583     sPAPRPHBState *sphb;
584     uint32_t option;
585     uint64_t buid;
586     int ret;
587 
588     if ((nargs != 4) || (nret != 1)) {
589         goto param_error_exit;
590     }
591 
592     buid = rtas_ldq(args, 1);
593     option = rtas_ld(args, 3);
594     sphb = spapr_pci_find_phb(spapr, buid);
595     if (!sphb) {
596         goto param_error_exit;
597     }
598 
599     if (!spapr_phb_eeh_available(sphb)) {
600         goto param_error_exit;
601     }
602 
603     ret = spapr_phb_vfio_eeh_reset(sphb, option);
604     rtas_st(rets, 0, ret);
605     return;
606 
607 param_error_exit:
608     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
609 }
610 
611 static void rtas_ibm_configure_pe(PowerPCCPU *cpu,
612                                   sPAPRMachineState *spapr,
613                                   uint32_t token, uint32_t nargs,
614                                   target_ulong args, uint32_t nret,
615                                   target_ulong rets)
616 {
617     sPAPRPHBState *sphb;
618     uint64_t buid;
619     int ret;
620 
621     if ((nargs != 3) || (nret != 1)) {
622         goto param_error_exit;
623     }
624 
625     buid = rtas_ldq(args, 1);
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_configure(sphb);
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 /* To support it later */
644 static void rtas_ibm_slot_error_detail(PowerPCCPU *cpu,
645                                        sPAPRMachineState *spapr,
646                                        uint32_t token, uint32_t nargs,
647                                        target_ulong args, uint32_t nret,
648                                        target_ulong rets)
649 {
650     sPAPRPHBState *sphb;
651     int option;
652     uint64_t buid;
653 
654     if ((nargs != 8) || (nret != 1)) {
655         goto param_error_exit;
656     }
657 
658     buid = rtas_ldq(args, 1);
659     sphb = spapr_pci_find_phb(spapr, buid);
660     if (!sphb) {
661         goto param_error_exit;
662     }
663 
664     if (!spapr_phb_eeh_available(sphb)) {
665         goto param_error_exit;
666     }
667 
668     option = rtas_ld(args, 7);
669     switch (option) {
670     case RTAS_SLOT_TEMP_ERR_LOG:
671     case RTAS_SLOT_PERM_ERR_LOG:
672         break;
673     default:
674         goto param_error_exit;
675     }
676 
677     /* We don't have error log yet */
678     rtas_st(rets, 0, RTAS_OUT_NO_ERRORS_FOUND);
679     return;
680 
681 param_error_exit:
682     rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
683 }
684 
685 static int pci_spapr_swizzle(int slot, int pin)
686 {
687     return (slot + pin) % PCI_NUM_PINS;
688 }
689 
690 static int pci_spapr_map_irq(PCIDevice *pci_dev, int irq_num)
691 {
692     /*
693      * Here we need to convert pci_dev + irq_num to some unique value
694      * which is less than number of IRQs on the specific bus (4).  We
695      * use standard PCI swizzling, that is (slot number + pin number)
696      * % 4.
697      */
698     return pci_spapr_swizzle(PCI_SLOT(pci_dev->devfn), irq_num);
699 }
700 
701 static void pci_spapr_set_irq(void *opaque, int irq_num, int level)
702 {
703     /*
704      * Here we use the number returned by pci_spapr_map_irq to find a
705      * corresponding qemu_irq.
706      */
707     sPAPRPHBState *phb = opaque;
708 
709     trace_spapr_pci_lsi_set(phb->dtbusname, irq_num, phb->lsi_table[irq_num].irq);
710     qemu_set_irq(spapr_phb_lsi_qirq(phb, irq_num), level);
711 }
712 
713 static PCIINTxRoute spapr_route_intx_pin_to_irq(void *opaque, int pin)
714 {
715     sPAPRPHBState *sphb = SPAPR_PCI_HOST_BRIDGE(opaque);
716     PCIINTxRoute route;
717 
718     route.mode = PCI_INTX_ENABLED;
719     route.irq = sphb->lsi_table[pin].irq;
720 
721     return route;
722 }
723 
724 /*
725  * MSI/MSIX memory region implementation.
726  * The handler handles both MSI and MSIX.
727  * For MSI-X, the vector number is encoded as a part of the address,
728  * data is set to 0.
729  * For MSI, the vector number is encoded in least bits in data.
730  */
731 static void spapr_msi_write(void *opaque, hwaddr addr,
732                             uint64_t data, unsigned size)
733 {
734     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
735     uint32_t irq = data;
736 
737     trace_spapr_pci_msi_write(addr, data, irq);
738 
739     qemu_irq_pulse(xics_get_qirq(spapr->xics, irq));
740 }
741 
742 static const MemoryRegionOps spapr_msi_ops = {
743     /* There is no .read as the read result is undefined by PCI spec */
744     .read = NULL,
745     .write = spapr_msi_write,
746     .endianness = DEVICE_LITTLE_ENDIAN
747 };
748 
749 /*
750  * PHB PCI device
751  */
752 static AddressSpace *spapr_pci_dma_iommu(PCIBus *bus, void *opaque, int devfn)
753 {
754     sPAPRPHBState *phb = opaque;
755 
756     return &phb->iommu_as;
757 }
758 
759 static char *spapr_phb_vfio_get_loc_code(sPAPRPHBState *sphb,  PCIDevice *pdev)
760 {
761     char *path = NULL, *buf = NULL, *host = NULL;
762 
763     /* Get the PCI VFIO host id */
764     host = object_property_get_str(OBJECT(pdev), "host", NULL);
765     if (!host) {
766         goto err_out;
767     }
768 
769     /* Construct the path of the file that will give us the DT location */
770     path = g_strdup_printf("/sys/bus/pci/devices/%s/devspec", host);
771     g_free(host);
772     if (!path || !g_file_get_contents(path, &buf, NULL, NULL)) {
773         goto err_out;
774     }
775     g_free(path);
776 
777     /* Construct and read from host device tree the loc-code */
778     path = g_strdup_printf("/proc/device-tree%s/ibm,loc-code", buf);
779     g_free(buf);
780     if (!path || !g_file_get_contents(path, &buf, NULL, NULL)) {
781         goto err_out;
782     }
783     return buf;
784 
785 err_out:
786     g_free(path);
787     return NULL;
788 }
789 
790 static char *spapr_phb_get_loc_code(sPAPRPHBState *sphb, PCIDevice *pdev)
791 {
792     char *buf;
793     const char *devtype = "qemu";
794     uint32_t busnr = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(pdev))));
795 
796     if (object_dynamic_cast(OBJECT(pdev), "vfio-pci")) {
797         buf = spapr_phb_vfio_get_loc_code(sphb, pdev);
798         if (buf) {
799             return buf;
800         }
801         devtype = "vfio";
802     }
803     /*
804      * For emulated devices and VFIO-failure case, make up
805      * the loc-code.
806      */
807     buf = g_strdup_printf("%s_%s:%04x:%02x:%02x.%x",
808                           devtype, pdev->name, sphb->index, busnr,
809                           PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
810     return buf;
811 }
812 
813 /* Macros to operate with address in OF binding to PCI */
814 #define b_x(x, p, l)    (((x) & ((1<<(l))-1)) << (p))
815 #define b_n(x)          b_x((x), 31, 1) /* 0 if relocatable */
816 #define b_p(x)          b_x((x), 30, 1) /* 1 if prefetchable */
817 #define b_t(x)          b_x((x), 29, 1) /* 1 if the address is aliased */
818 #define b_ss(x)         b_x((x), 24, 2) /* the space code */
819 #define b_bbbbbbbb(x)   b_x((x), 16, 8) /* bus number */
820 #define b_ddddd(x)      b_x((x), 11, 5) /* device number */
821 #define b_fff(x)        b_x((x), 8, 3)  /* function number */
822 #define b_rrrrrrrr(x)   b_x((x), 0, 8)  /* register number */
823 
824 /* for 'reg'/'assigned-addresses' OF properties */
825 #define RESOURCE_CELLS_SIZE 2
826 #define RESOURCE_CELLS_ADDRESS 3
827 
828 typedef struct ResourceFields {
829     uint32_t phys_hi;
830     uint32_t phys_mid;
831     uint32_t phys_lo;
832     uint32_t size_hi;
833     uint32_t size_lo;
834 } QEMU_PACKED ResourceFields;
835 
836 typedef struct ResourceProps {
837     ResourceFields reg[8];
838     ResourceFields assigned[7];
839     uint32_t reg_len;
840     uint32_t assigned_len;
841 } ResourceProps;
842 
843 /* fill in the 'reg'/'assigned-resources' OF properties for
844  * a PCI device. 'reg' describes resource requirements for a
845  * device's IO/MEM regions, 'assigned-addresses' describes the
846  * actual resource assignments.
847  *
848  * the properties are arrays of ('phys-addr', 'size') pairs describing
849  * the addressable regions of the PCI device, where 'phys-addr' is a
850  * RESOURCE_CELLS_ADDRESS-tuple of 32-bit integers corresponding to
851  * (phys.hi, phys.mid, phys.lo), and 'size' is a
852  * RESOURCE_CELLS_SIZE-tuple corresponding to (size.hi, size.lo).
853  *
854  * phys.hi = 0xYYXXXXZZ, where:
855  *   0xYY = npt000ss
856  *          |||   |
857  *          |||   +-- space code
858  *          |||               |
859  *          |||               +  00 if configuration space
860  *          |||               +  01 if IO region,
861  *          |||               +  10 if 32-bit MEM region
862  *          |||               +  11 if 64-bit MEM region
863  *          |||
864  *          ||+------ for non-relocatable IO: 1 if aliased
865  *          ||        for relocatable IO: 1 if below 64KB
866  *          ||        for MEM: 1 if below 1MB
867  *          |+------- 1 if region is prefetchable
868  *          +-------- 1 if region is non-relocatable
869  *   0xXXXX = bbbbbbbb dddddfff, encoding bus, slot, and function
870  *            bits respectively
871  *   0xZZ = rrrrrrrr, the register number of the BAR corresponding
872  *          to the region
873  *
874  * phys.mid and phys.lo correspond respectively to the hi/lo portions
875  * of the actual address of the region.
876  *
877  * how the phys-addr/size values are used differ slightly between
878  * 'reg' and 'assigned-addresses' properties. namely, 'reg' has
879  * an additional description for the config space region of the
880  * device, and in the case of QEMU has n=0 and phys.mid=phys.lo=0
881  * to describe the region as relocatable, with an address-mapping
882  * that corresponds directly to the PHB's address space for the
883  * resource. 'assigned-addresses' always has n=1 set with an absolute
884  * address assigned for the resource. in general, 'assigned-addresses'
885  * won't be populated, since addresses for PCI devices are generally
886  * unmapped initially and left to the guest to assign.
887  *
888  * note also that addresses defined in these properties are, at least
889  * for PAPR guests, relative to the PHBs IO/MEM windows, and
890  * correspond directly to the addresses in the BARs.
891  *
892  * in accordance with PCI Bus Binding to Open Firmware,
893  * IEEE Std 1275-1994, section 4.1.1, as implemented by PAPR+ v2.7,
894  * Appendix C.
895  */
896 static void populate_resource_props(PCIDevice *d, ResourceProps *rp)
897 {
898     int bus_num = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(d))));
899     uint32_t dev_id = (b_bbbbbbbb(bus_num) |
900                        b_ddddd(PCI_SLOT(d->devfn)) |
901                        b_fff(PCI_FUNC(d->devfn)));
902     ResourceFields *reg, *assigned;
903     int i, reg_idx = 0, assigned_idx = 0;
904 
905     /* config space region */
906     reg = &rp->reg[reg_idx++];
907     reg->phys_hi = cpu_to_be32(dev_id);
908     reg->phys_mid = 0;
909     reg->phys_lo = 0;
910     reg->size_hi = 0;
911     reg->size_lo = 0;
912 
913     for (i = 0; i < PCI_NUM_REGIONS; i++) {
914         if (!d->io_regions[i].size) {
915             continue;
916         }
917 
918         reg = &rp->reg[reg_idx++];
919 
920         reg->phys_hi = cpu_to_be32(dev_id | b_rrrrrrrr(pci_bar(d, i)));
921         if (d->io_regions[i].type & PCI_BASE_ADDRESS_SPACE_IO) {
922             reg->phys_hi |= cpu_to_be32(b_ss(1));
923         } else if (d->io_regions[i].type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
924             reg->phys_hi |= cpu_to_be32(b_ss(3));
925         } else {
926             reg->phys_hi |= cpu_to_be32(b_ss(2));
927         }
928         reg->phys_mid = 0;
929         reg->phys_lo = 0;
930         reg->size_hi = cpu_to_be32(d->io_regions[i].size >> 32);
931         reg->size_lo = cpu_to_be32(d->io_regions[i].size);
932 
933         if (d->io_regions[i].addr == PCI_BAR_UNMAPPED) {
934             continue;
935         }
936 
937         assigned = &rp->assigned[assigned_idx++];
938         assigned->phys_hi = cpu_to_be32(reg->phys_hi | b_n(1));
939         assigned->phys_mid = cpu_to_be32(d->io_regions[i].addr >> 32);
940         assigned->phys_lo = cpu_to_be32(d->io_regions[i].addr);
941         assigned->size_hi = reg->size_hi;
942         assigned->size_lo = reg->size_lo;
943     }
944 
945     rp->reg_len = reg_idx * sizeof(ResourceFields);
946     rp->assigned_len = assigned_idx * sizeof(ResourceFields);
947 }
948 
949 static uint32_t spapr_phb_get_pci_drc_index(sPAPRPHBState *phb,
950                                             PCIDevice *pdev);
951 
952 static int spapr_populate_pci_child_dt(PCIDevice *dev, void *fdt, int offset,
953                                        sPAPRPHBState *sphb)
954 {
955     ResourceProps rp;
956     bool is_bridge = false;
957     int pci_status, err;
958     char *buf = NULL;
959     uint32_t drc_index = spapr_phb_get_pci_drc_index(sphb, dev);
960     uint32_t max_msi, max_msix;
961 
962     if (pci_default_read_config(dev, PCI_HEADER_TYPE, 1) ==
963         PCI_HEADER_TYPE_BRIDGE) {
964         is_bridge = true;
965     }
966 
967     /* in accordance with PAPR+ v2.7 13.6.3, Table 181 */
968     _FDT(fdt_setprop_cell(fdt, offset, "vendor-id",
969                           pci_default_read_config(dev, PCI_VENDOR_ID, 2)));
970     _FDT(fdt_setprop_cell(fdt, offset, "device-id",
971                           pci_default_read_config(dev, PCI_DEVICE_ID, 2)));
972     _FDT(fdt_setprop_cell(fdt, offset, "revision-id",
973                           pci_default_read_config(dev, PCI_REVISION_ID, 1)));
974     _FDT(fdt_setprop_cell(fdt, offset, "class-code",
975                           pci_default_read_config(dev, PCI_CLASS_PROG, 3)));
976     if (pci_default_read_config(dev, PCI_INTERRUPT_PIN, 1)) {
977         _FDT(fdt_setprop_cell(fdt, offset, "interrupts",
978                  pci_default_read_config(dev, PCI_INTERRUPT_PIN, 1)));
979     }
980 
981     if (!is_bridge) {
982         _FDT(fdt_setprop_cell(fdt, offset, "min-grant",
983             pci_default_read_config(dev, PCI_MIN_GNT, 1)));
984         _FDT(fdt_setprop_cell(fdt, offset, "max-latency",
985             pci_default_read_config(dev, PCI_MAX_LAT, 1)));
986     }
987 
988     if (pci_default_read_config(dev, PCI_SUBSYSTEM_ID, 2)) {
989         _FDT(fdt_setprop_cell(fdt, offset, "subsystem-id",
990                  pci_default_read_config(dev, PCI_SUBSYSTEM_ID, 2)));
991     }
992 
993     if (pci_default_read_config(dev, PCI_SUBSYSTEM_VENDOR_ID, 2)) {
994         _FDT(fdt_setprop_cell(fdt, offset, "subsystem-vendor-id",
995                  pci_default_read_config(dev, PCI_SUBSYSTEM_VENDOR_ID, 2)));
996     }
997 
998     _FDT(fdt_setprop_cell(fdt, offset, "cache-line-size",
999         pci_default_read_config(dev, PCI_CACHE_LINE_SIZE, 1)));
1000 
1001     /* the following fdt cells are masked off the pci status register */
1002     pci_status = pci_default_read_config(dev, PCI_STATUS, 2);
1003     _FDT(fdt_setprop_cell(fdt, offset, "devsel-speed",
1004                           PCI_STATUS_DEVSEL_MASK & pci_status));
1005 
1006     if (pci_status & PCI_STATUS_FAST_BACK) {
1007         _FDT(fdt_setprop(fdt, offset, "fast-back-to-back", NULL, 0));
1008     }
1009     if (pci_status & PCI_STATUS_66MHZ) {
1010         _FDT(fdt_setprop(fdt, offset, "66mhz-capable", NULL, 0));
1011     }
1012     if (pci_status & PCI_STATUS_UDF) {
1013         _FDT(fdt_setprop(fdt, offset, "udf-supported", NULL, 0));
1014     }
1015 
1016     /* NOTE: this is normally generated by firmware via path/unit name,
1017      * but in our case we must set it manually since it does not get
1018      * processed by OF beforehand
1019      */
1020     _FDT(fdt_setprop_string(fdt, offset, "name", "pci"));
1021     buf = spapr_phb_get_loc_code(sphb, dev);
1022     if (!buf) {
1023         error_report("Failed setting the ibm,loc-code");
1024         return -1;
1025     }
1026 
1027     err = fdt_setprop_string(fdt, offset, "ibm,loc-code", buf);
1028     g_free(buf);
1029     if (err < 0) {
1030         return err;
1031     }
1032 
1033     if (drc_index) {
1034         _FDT(fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index));
1035     }
1036 
1037     _FDT(fdt_setprop_cell(fdt, offset, "#address-cells",
1038                           RESOURCE_CELLS_ADDRESS));
1039     _FDT(fdt_setprop_cell(fdt, offset, "#size-cells",
1040                           RESOURCE_CELLS_SIZE));
1041 
1042     max_msi = msi_nr_vectors_allocated(dev);
1043     if (max_msi) {
1044         _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi", max_msi));
1045     }
1046     max_msix = dev->msix_entries_nr;
1047     if (max_msix) {
1048         _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi-x", max_msix));
1049     }
1050 
1051     populate_resource_props(dev, &rp);
1052     _FDT(fdt_setprop(fdt, offset, "reg", (uint8_t *)rp.reg, rp.reg_len));
1053     _FDT(fdt_setprop(fdt, offset, "assigned-addresses",
1054                      (uint8_t *)rp.assigned, rp.assigned_len));
1055 
1056     return 0;
1057 }
1058 
1059 /* create OF node for pci device and required OF DT properties */
1060 static int spapr_create_pci_child_dt(sPAPRPHBState *phb, PCIDevice *dev,
1061                                      void *fdt, int node_offset)
1062 {
1063     int offset, ret;
1064     int slot = PCI_SLOT(dev->devfn);
1065     int func = PCI_FUNC(dev->devfn);
1066     char nodename[FDT_NAME_MAX];
1067 
1068     if (func != 0) {
1069         snprintf(nodename, FDT_NAME_MAX, "pci@%x,%x", slot, func);
1070     } else {
1071         snprintf(nodename, FDT_NAME_MAX, "pci@%x", slot);
1072     }
1073     offset = fdt_add_subnode(fdt, node_offset, nodename);
1074     ret = spapr_populate_pci_child_dt(dev, fdt, offset, phb);
1075 
1076     g_assert(!ret);
1077     if (ret) {
1078         return 0;
1079     }
1080     return offset;
1081 }
1082 
1083 static void spapr_phb_add_pci_device(sPAPRDRConnector *drc,
1084                                      sPAPRPHBState *phb,
1085                                      PCIDevice *pdev,
1086                                      Error **errp)
1087 {
1088     sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
1089     DeviceState *dev = DEVICE(pdev);
1090     void *fdt = NULL;
1091     int fdt_start_offset = 0, fdt_size;
1092 
1093     fdt = create_device_tree(&fdt_size);
1094     fdt_start_offset = spapr_create_pci_child_dt(phb, pdev, fdt, 0);
1095     if (!fdt_start_offset) {
1096         error_setg(errp, "Failed to create pci child device tree node");
1097         goto out;
1098     }
1099 
1100     drck->attach(drc, DEVICE(pdev),
1101                  fdt, fdt_start_offset, !dev->hotplugged, errp);
1102 out:
1103     if (*errp) {
1104         g_free(fdt);
1105     }
1106 }
1107 
1108 static void spapr_phb_remove_pci_device_cb(DeviceState *dev, void *opaque)
1109 {
1110     /* some version guests do not wait for completion of a device
1111      * cleanup (generally done asynchronously by the kernel) before
1112      * signaling to QEMU that the device is safe, but instead sleep
1113      * for some 'safe' period of time. unfortunately on a busy host
1114      * this sleep isn't guaranteed to be long enough, resulting in
1115      * bad things like IRQ lines being left asserted during final
1116      * device removal. to deal with this we call reset just prior
1117      * to finalizing the device, which will put the device back into
1118      * an 'idle' state, as the device cleanup code expects.
1119      */
1120     pci_device_reset(PCI_DEVICE(dev));
1121     object_unparent(OBJECT(dev));
1122 }
1123 
1124 static void spapr_phb_remove_pci_device(sPAPRDRConnector *drc,
1125                                         sPAPRPHBState *phb,
1126                                         PCIDevice *pdev,
1127                                         Error **errp)
1128 {
1129     sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
1130 
1131     drck->detach(drc, DEVICE(pdev), spapr_phb_remove_pci_device_cb, phb, errp);
1132 }
1133 
1134 static sPAPRDRConnector *spapr_phb_get_pci_func_drc(sPAPRPHBState *phb,
1135                                                     uint32_t busnr,
1136                                                     int32_t devfn)
1137 {
1138     return spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_PCI,
1139                                     (phb->index << 16) |
1140                                     (busnr << 8) |
1141                                     devfn);
1142 }
1143 
1144 static sPAPRDRConnector *spapr_phb_get_pci_drc(sPAPRPHBState *phb,
1145                                                PCIDevice *pdev)
1146 {
1147     uint32_t busnr = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(pdev))));
1148     return spapr_phb_get_pci_func_drc(phb, busnr, pdev->devfn);
1149 }
1150 
1151 static uint32_t spapr_phb_get_pci_drc_index(sPAPRPHBState *phb,
1152                                             PCIDevice *pdev)
1153 {
1154     sPAPRDRConnector *drc = spapr_phb_get_pci_drc(phb, pdev);
1155     sPAPRDRConnectorClass *drck;
1156 
1157     if (!drc) {
1158         return 0;
1159     }
1160 
1161     drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
1162     return drck->get_index(drc);
1163 }
1164 
1165 static void spapr_phb_hot_plug_child(HotplugHandler *plug_handler,
1166                                      DeviceState *plugged_dev, Error **errp)
1167 {
1168     sPAPRPHBState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1169     PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1170     sPAPRDRConnector *drc = spapr_phb_get_pci_drc(phb, pdev);
1171     Error *local_err = NULL;
1172     PCIBus *bus = PCI_BUS(qdev_get_parent_bus(DEVICE(pdev)));
1173     uint32_t slotnr = PCI_SLOT(pdev->devfn);
1174 
1175     /* if DR is disabled we don't need to do anything in the case of
1176      * hotplug or coldplug callbacks
1177      */
1178     if (!phb->dr_enabled) {
1179         /* if this is a hotplug operation initiated by the user
1180          * we need to let them know it's not enabled
1181          */
1182         if (plugged_dev->hotplugged) {
1183             error_setg(errp, QERR_BUS_NO_HOTPLUG,
1184                        object_get_typename(OBJECT(phb)));
1185         }
1186         return;
1187     }
1188 
1189     g_assert(drc);
1190 
1191     /* Following the QEMU convention used for PCIe multifunction
1192      * hotplug, we do not allow functions to be hotplugged to a
1193      * slot that already has function 0 present
1194      */
1195     if (plugged_dev->hotplugged && bus->devices[PCI_DEVFN(slotnr, 0)] &&
1196         PCI_FUNC(pdev->devfn) != 0) {
1197         error_setg(errp, "PCI: slot %d function 0 already ocuppied by %s,"
1198                    " additional functions can no longer be exposed to guest.",
1199                    slotnr, bus->devices[PCI_DEVFN(slotnr, 0)]->name);
1200         return;
1201     }
1202 
1203     spapr_phb_add_pci_device(drc, phb, pdev, &local_err);
1204     if (local_err) {
1205         error_propagate(errp, local_err);
1206         return;
1207     }
1208 
1209     /* If this is function 0, signal hotplug for all the device functions.
1210      * Otherwise defer sending the hotplug event.
1211      */
1212     if (plugged_dev->hotplugged && PCI_FUNC(pdev->devfn) == 0) {
1213         int i;
1214 
1215         for (i = 0; i < 8; i++) {
1216             sPAPRDRConnector *func_drc;
1217             sPAPRDRConnectorClass *func_drck;
1218             sPAPRDREntitySense state;
1219 
1220             func_drc = spapr_phb_get_pci_func_drc(phb, pci_bus_num(bus),
1221                                                   PCI_DEVFN(slotnr, i));
1222             func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1223             func_drck->entity_sense(func_drc, &state);
1224 
1225             if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) {
1226                 spapr_hotplug_req_add_by_index(func_drc);
1227             }
1228         }
1229     }
1230 }
1231 
1232 static void spapr_phb_hot_unplug_child(HotplugHandler *plug_handler,
1233                                        DeviceState *plugged_dev, Error **errp)
1234 {
1235     sPAPRPHBState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1236     PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1237     sPAPRDRConnectorClass *drck;
1238     sPAPRDRConnector *drc = spapr_phb_get_pci_drc(phb, pdev);
1239     Error *local_err = NULL;
1240 
1241     if (!phb->dr_enabled) {
1242         error_setg(errp, QERR_BUS_NO_HOTPLUG,
1243                    object_get_typename(OBJECT(phb)));
1244         return;
1245     }
1246 
1247     g_assert(drc);
1248 
1249     drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
1250     if (!drck->release_pending(drc)) {
1251         PCIBus *bus = PCI_BUS(qdev_get_parent_bus(DEVICE(pdev)));
1252         uint32_t slotnr = PCI_SLOT(pdev->devfn);
1253         sPAPRDRConnector *func_drc;
1254         sPAPRDRConnectorClass *func_drck;
1255         sPAPRDREntitySense state;
1256         int i;
1257 
1258         /* ensure any other present functions are pending unplug */
1259         if (PCI_FUNC(pdev->devfn) == 0) {
1260             for (i = 1; i < 8; i++) {
1261                 func_drc = spapr_phb_get_pci_func_drc(phb, pci_bus_num(bus),
1262                                                       PCI_DEVFN(slotnr, i));
1263                 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1264                 func_drck->entity_sense(func_drc, &state);
1265                 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT
1266                     && !func_drck->release_pending(func_drc)) {
1267                     error_setg(errp,
1268                                "PCI: slot %d, function %d still present. "
1269                                "Must unplug all non-0 functions first.",
1270                                slotnr, i);
1271                     return;
1272                 }
1273             }
1274         }
1275 
1276         spapr_phb_remove_pci_device(drc, phb, pdev, &local_err);
1277         if (local_err) {
1278             error_propagate(errp, local_err);
1279             return;
1280         }
1281 
1282         /* if this isn't func 0, defer unplug event. otherwise signal removal
1283          * for all present functions
1284          */
1285         if (PCI_FUNC(pdev->devfn) == 0) {
1286             for (i = 7; i >= 0; i--) {
1287                 func_drc = spapr_phb_get_pci_func_drc(phb, pci_bus_num(bus),
1288                                                       PCI_DEVFN(slotnr, i));
1289                 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1290                 func_drck->entity_sense(func_drc, &state);
1291                 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) {
1292                     spapr_hotplug_req_remove_by_index(func_drc);
1293                 }
1294             }
1295         }
1296     }
1297 }
1298 
1299 static void spapr_phb_realize(DeviceState *dev, Error **errp)
1300 {
1301     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
1302     SysBusDevice *s = SYS_BUS_DEVICE(dev);
1303     sPAPRPHBState *sphb = SPAPR_PCI_HOST_BRIDGE(s);
1304     PCIHostState *phb = PCI_HOST_BRIDGE(s);
1305     char *namebuf;
1306     int i;
1307     PCIBus *bus;
1308     uint64_t msi_window_size = 4096;
1309     sPAPRTCETable *tcet;
1310     const unsigned windows_supported =
1311         sphb->ddw_enabled ? SPAPR_PCI_DMA_MAX_WINDOWS : 1;
1312 
1313     if (sphb->index != (uint32_t)-1) {
1314         sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
1315         Error *local_err = NULL;
1316 
1317         if ((sphb->buid != (uint64_t)-1) || (sphb->dma_liobn[0] != (uint32_t)-1)
1318             || (sphb->dma_liobn[1] != (uint32_t)-1 && windows_supported == 2)
1319             || (sphb->mem_win_addr != (hwaddr)-1)
1320             || (sphb->mem64_win_addr != (hwaddr)-1)
1321             || (sphb->io_win_addr != (hwaddr)-1)) {
1322             error_setg(errp, "Either \"index\" or other parameters must"
1323                        " be specified for PAPR PHB, not both");
1324             return;
1325         }
1326 
1327         smc->phb_placement(spapr, sphb->index,
1328                            &sphb->buid, &sphb->io_win_addr,
1329                            &sphb->mem_win_addr, &sphb->mem64_win_addr,
1330                            windows_supported, sphb->dma_liobn, &local_err);
1331         if (local_err) {
1332             error_propagate(errp, local_err);
1333             return;
1334         }
1335     }
1336 
1337     if (sphb->buid == (uint64_t)-1) {
1338         error_setg(errp, "BUID not specified for PHB");
1339         return;
1340     }
1341 
1342     if ((sphb->dma_liobn[0] == (uint32_t)-1) ||
1343         ((sphb->dma_liobn[1] == (uint32_t)-1) && (windows_supported > 1))) {
1344         error_setg(errp, "LIOBN(s) not specified for PHB");
1345         return;
1346     }
1347 
1348     if (sphb->mem_win_addr == (hwaddr)-1) {
1349         error_setg(errp, "Memory window address not specified for PHB");
1350         return;
1351     }
1352 
1353     if (sphb->io_win_addr == (hwaddr)-1) {
1354         error_setg(errp, "IO window address not specified for PHB");
1355         return;
1356     }
1357 
1358     if (sphb->mem64_win_size != 0) {
1359         if (sphb->mem64_win_addr == (hwaddr)-1) {
1360             error_setg(errp,
1361                        "64-bit memory window address not specified for PHB");
1362             return;
1363         }
1364 
1365         if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) {
1366             error_setg(errp, "32-bit memory window of size 0x%"HWADDR_PRIx
1367                        " (max 2 GiB)", sphb->mem_win_size);
1368             return;
1369         }
1370 
1371         if (sphb->mem64_win_pciaddr == (hwaddr)-1) {
1372             /* 64-bit window defaults to identity mapping */
1373             sphb->mem64_win_pciaddr = sphb->mem64_win_addr;
1374         }
1375     } else if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) {
1376         /*
1377          * For compatibility with old configuration, if no 64-bit MMIO
1378          * window is specified, but the ordinary (32-bit) memory
1379          * window is specified as > 2GiB, we treat it as a 2GiB 32-bit
1380          * window, with a 64-bit MMIO window following on immediately
1381          * afterwards
1382          */
1383         sphb->mem64_win_size = sphb->mem_win_size - SPAPR_PCI_MEM32_WIN_SIZE;
1384         sphb->mem64_win_addr = sphb->mem_win_addr + SPAPR_PCI_MEM32_WIN_SIZE;
1385         sphb->mem64_win_pciaddr =
1386             SPAPR_PCI_MEM_WIN_BUS_OFFSET + SPAPR_PCI_MEM32_WIN_SIZE;
1387         sphb->mem_win_size = SPAPR_PCI_MEM32_WIN_SIZE;
1388     }
1389 
1390     if (spapr_pci_find_phb(spapr, sphb->buid)) {
1391         error_setg(errp, "PCI host bridges must have unique BUIDs");
1392         return;
1393     }
1394 
1395     if (sphb->numa_node != -1 &&
1396         (sphb->numa_node >= MAX_NODES || !numa_info[sphb->numa_node].present)) {
1397         error_setg(errp, "Invalid NUMA node ID for PCI host bridge");
1398         return;
1399     }
1400 
1401     sphb->dtbusname = g_strdup_printf("pci@%" PRIx64, sphb->buid);
1402 
1403     namebuf = alloca(strlen(sphb->dtbusname) + 32);
1404 
1405     /* Initialize memory regions */
1406     sprintf(namebuf, "%s.mmio", sphb->dtbusname);
1407     memory_region_init(&sphb->memspace, OBJECT(sphb), namebuf, UINT64_MAX);
1408 
1409     sprintf(namebuf, "%s.mmio32-alias", sphb->dtbusname);
1410     memory_region_init_alias(&sphb->mem32window, OBJECT(sphb),
1411                              namebuf, &sphb->memspace,
1412                              SPAPR_PCI_MEM_WIN_BUS_OFFSET, sphb->mem_win_size);
1413     memory_region_add_subregion(get_system_memory(), sphb->mem_win_addr,
1414                                 &sphb->mem32window);
1415 
1416     sprintf(namebuf, "%s.mmio64-alias", sphb->dtbusname);
1417     memory_region_init_alias(&sphb->mem64window, OBJECT(sphb),
1418                              namebuf, &sphb->memspace,
1419                              sphb->mem64_win_pciaddr, sphb->mem64_win_size);
1420     memory_region_add_subregion(get_system_memory(), sphb->mem64_win_addr,
1421                                 &sphb->mem64window);
1422 
1423     /* Initialize IO regions */
1424     sprintf(namebuf, "%s.io", sphb->dtbusname);
1425     memory_region_init(&sphb->iospace, OBJECT(sphb),
1426                        namebuf, SPAPR_PCI_IO_WIN_SIZE);
1427 
1428     sprintf(namebuf, "%s.io-alias", sphb->dtbusname);
1429     memory_region_init_alias(&sphb->iowindow, OBJECT(sphb), namebuf,
1430                              &sphb->iospace, 0, SPAPR_PCI_IO_WIN_SIZE);
1431     memory_region_add_subregion(get_system_memory(), sphb->io_win_addr,
1432                                 &sphb->iowindow);
1433 
1434     bus = pci_register_bus(dev, NULL,
1435                            pci_spapr_set_irq, pci_spapr_map_irq, sphb,
1436                            &sphb->memspace, &sphb->iospace,
1437                            PCI_DEVFN(0, 0), PCI_NUM_PINS, TYPE_PCI_BUS);
1438     phb->bus = bus;
1439     qbus_set_hotplug_handler(BUS(phb->bus), DEVICE(sphb), NULL);
1440 
1441     /*
1442      * Initialize PHB address space.
1443      * By default there will be at least one subregion for default
1444      * 32bit DMA window.
1445      * Later the guest might want to create another DMA window
1446      * which will become another memory subregion.
1447      */
1448     sprintf(namebuf, "%s.iommu-root", sphb->dtbusname);
1449 
1450     memory_region_init(&sphb->iommu_root, OBJECT(sphb),
1451                        namebuf, UINT64_MAX);
1452     address_space_init(&sphb->iommu_as, &sphb->iommu_root,
1453                        sphb->dtbusname);
1454 
1455     /*
1456      * As MSI/MSIX interrupts trigger by writing at MSI/MSIX vectors,
1457      * we need to allocate some memory to catch those writes coming
1458      * from msi_notify()/msix_notify().
1459      * As MSIMessage:addr is going to be the same and MSIMessage:data
1460      * is going to be a VIRQ number, 4 bytes of the MSI MR will only
1461      * be used.
1462      *
1463      * For KVM we want to ensure that this memory is a full page so that
1464      * our memory slot is of page size granularity.
1465      */
1466 #ifdef CONFIG_KVM
1467     if (kvm_enabled()) {
1468         msi_window_size = getpagesize();
1469     }
1470 #endif
1471 
1472     memory_region_init_io(&sphb->msiwindow, NULL, &spapr_msi_ops, spapr,
1473                           "msi", msi_window_size);
1474     memory_region_add_subregion(&sphb->iommu_root, SPAPR_PCI_MSI_WINDOW,
1475                                 &sphb->msiwindow);
1476 
1477     pci_setup_iommu(bus, spapr_pci_dma_iommu, sphb);
1478 
1479     pci_bus_set_route_irq_fn(bus, spapr_route_intx_pin_to_irq);
1480 
1481     QLIST_INSERT_HEAD(&spapr->phbs, sphb, list);
1482 
1483     /* Initialize the LSI table */
1484     for (i = 0; i < PCI_NUM_PINS; i++) {
1485         uint32_t irq;
1486         Error *local_err = NULL;
1487 
1488         irq = xics_spapr_alloc_block(spapr->xics, 1, true, false, &local_err);
1489         if (local_err) {
1490             error_propagate(errp, local_err);
1491             error_prepend(errp, "can't allocate LSIs: ");
1492             return;
1493         }
1494 
1495         sphb->lsi_table[i].irq = irq;
1496     }
1497 
1498     /* allocate connectors for child PCI devices */
1499     if (sphb->dr_enabled) {
1500         for (i = 0; i < PCI_SLOT_MAX * 8; i++) {
1501             spapr_dr_connector_new(OBJECT(phb),
1502                                    SPAPR_DR_CONNECTOR_TYPE_PCI,
1503                                    (sphb->index << 16) | i);
1504         }
1505     }
1506 
1507     /* DMA setup */
1508     for (i = 0; i < windows_supported; ++i) {
1509         tcet = spapr_tce_new_table(DEVICE(sphb), sphb->dma_liobn[i]);
1510         if (!tcet) {
1511             error_setg(errp, "Creating window#%d failed for %s",
1512                        i, sphb->dtbusname);
1513             return;
1514         }
1515         memory_region_add_subregion_overlap(&sphb->iommu_root, 0,
1516                                             spapr_tce_get_iommu(tcet), 0);
1517     }
1518 
1519     sphb->msi = g_hash_table_new_full(g_int_hash, g_int_equal, g_free, g_free);
1520 }
1521 
1522 static int spapr_phb_children_reset(Object *child, void *opaque)
1523 {
1524     DeviceState *dev = (DeviceState *) object_dynamic_cast(child, TYPE_DEVICE);
1525 
1526     if (dev) {
1527         device_reset(dev);
1528     }
1529 
1530     return 0;
1531 }
1532 
1533 void spapr_phb_dma_reset(sPAPRPHBState *sphb)
1534 {
1535     int i;
1536     sPAPRTCETable *tcet;
1537 
1538     for (i = 0; i < SPAPR_PCI_DMA_MAX_WINDOWS; ++i) {
1539         tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[i]);
1540 
1541         if (tcet && tcet->nb_table) {
1542             spapr_tce_table_disable(tcet);
1543         }
1544     }
1545 
1546     /* Register default 32bit DMA window */
1547     tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[0]);
1548     spapr_tce_table_enable(tcet, SPAPR_TCE_PAGE_SHIFT, sphb->dma_win_addr,
1549                            sphb->dma_win_size >> SPAPR_TCE_PAGE_SHIFT);
1550 }
1551 
1552 static void spapr_phb_reset(DeviceState *qdev)
1553 {
1554     sPAPRPHBState *sphb = SPAPR_PCI_HOST_BRIDGE(qdev);
1555 
1556     spapr_phb_dma_reset(sphb);
1557 
1558     /* Reset the IOMMU state */
1559     object_child_foreach(OBJECT(qdev), spapr_phb_children_reset, NULL);
1560 
1561     if (spapr_phb_eeh_available(SPAPR_PCI_HOST_BRIDGE(qdev))) {
1562         spapr_phb_vfio_reset(qdev);
1563     }
1564 }
1565 
1566 static Property spapr_phb_properties[] = {
1567     DEFINE_PROP_UINT32("index", sPAPRPHBState, index, -1),
1568     DEFINE_PROP_UINT64("buid", sPAPRPHBState, buid, -1),
1569     DEFINE_PROP_UINT32("liobn", sPAPRPHBState, dma_liobn[0], -1),
1570     DEFINE_PROP_UINT32("liobn64", sPAPRPHBState, dma_liobn[1], -1),
1571     DEFINE_PROP_UINT64("mem_win_addr", sPAPRPHBState, mem_win_addr, -1),
1572     DEFINE_PROP_UINT64("mem_win_size", sPAPRPHBState, mem_win_size,
1573                        SPAPR_PCI_MEM32_WIN_SIZE),
1574     DEFINE_PROP_UINT64("mem64_win_addr", sPAPRPHBState, mem64_win_addr, -1),
1575     DEFINE_PROP_UINT64("mem64_win_size", sPAPRPHBState, mem64_win_size,
1576                        SPAPR_PCI_MEM64_WIN_SIZE),
1577     DEFINE_PROP_UINT64("mem64_win_pciaddr", sPAPRPHBState, mem64_win_pciaddr,
1578                        -1),
1579     DEFINE_PROP_UINT64("io_win_addr", sPAPRPHBState, io_win_addr, -1),
1580     DEFINE_PROP_UINT64("io_win_size", sPAPRPHBState, io_win_size,
1581                        SPAPR_PCI_IO_WIN_SIZE),
1582     DEFINE_PROP_BOOL("dynamic-reconfiguration", sPAPRPHBState, dr_enabled,
1583                      true),
1584     /* Default DMA window is 0..1GB */
1585     DEFINE_PROP_UINT64("dma_win_addr", sPAPRPHBState, dma_win_addr, 0),
1586     DEFINE_PROP_UINT64("dma_win_size", sPAPRPHBState, dma_win_size, 0x40000000),
1587     DEFINE_PROP_UINT64("dma64_win_addr", sPAPRPHBState, dma64_win_addr,
1588                        0x800000000000000ULL),
1589     DEFINE_PROP_BOOL("ddw", sPAPRPHBState, ddw_enabled, true),
1590     DEFINE_PROP_UINT64("pgsz", sPAPRPHBState, page_size_mask,
1591                        (1ULL << 12) | (1ULL << 16)),
1592     DEFINE_PROP_UINT32("numa_node", sPAPRPHBState, numa_node, -1),
1593     DEFINE_PROP_BOOL("pre-2.8-migration", sPAPRPHBState,
1594                      pre_2_8_migration, false),
1595     DEFINE_PROP_END_OF_LIST(),
1596 };
1597 
1598 static const VMStateDescription vmstate_spapr_pci_lsi = {
1599     .name = "spapr_pci/lsi",
1600     .version_id = 1,
1601     .minimum_version_id = 1,
1602     .fields = (VMStateField[]) {
1603         VMSTATE_UINT32_EQUAL(irq, struct spapr_pci_lsi),
1604 
1605         VMSTATE_END_OF_LIST()
1606     },
1607 };
1608 
1609 static const VMStateDescription vmstate_spapr_pci_msi = {
1610     .name = "spapr_pci/msi",
1611     .version_id = 1,
1612     .minimum_version_id = 1,
1613     .fields = (VMStateField []) {
1614         VMSTATE_UINT32(key, spapr_pci_msi_mig),
1615         VMSTATE_UINT32(value.first_irq, spapr_pci_msi_mig),
1616         VMSTATE_UINT32(value.num, spapr_pci_msi_mig),
1617         VMSTATE_END_OF_LIST()
1618     },
1619 };
1620 
1621 static void spapr_pci_pre_save(void *opaque)
1622 {
1623     sPAPRPHBState *sphb = opaque;
1624     GHashTableIter iter;
1625     gpointer key, value;
1626     int i;
1627 
1628     g_free(sphb->msi_devs);
1629     sphb->msi_devs = NULL;
1630     sphb->msi_devs_num = g_hash_table_size(sphb->msi);
1631     if (!sphb->msi_devs_num) {
1632         return;
1633     }
1634     sphb->msi_devs = g_malloc(sphb->msi_devs_num * sizeof(spapr_pci_msi_mig));
1635 
1636     g_hash_table_iter_init(&iter, sphb->msi);
1637     for (i = 0; g_hash_table_iter_next(&iter, &key, &value); ++i) {
1638         sphb->msi_devs[i].key = *(uint32_t *) key;
1639         sphb->msi_devs[i].value = *(spapr_pci_msi *) value;
1640     }
1641 
1642     if (sphb->pre_2_8_migration) {
1643         sphb->mig_liobn = sphb->dma_liobn[0];
1644         sphb->mig_mem_win_addr = sphb->mem_win_addr;
1645         sphb->mig_mem_win_size = sphb->mem_win_size;
1646         sphb->mig_io_win_addr = sphb->io_win_addr;
1647         sphb->mig_io_win_size = sphb->io_win_size;
1648 
1649         if ((sphb->mem64_win_size != 0)
1650             && (sphb->mem64_win_addr
1651                 == (sphb->mem_win_addr + sphb->mem_win_size))) {
1652             sphb->mig_mem_win_size += sphb->mem64_win_size;
1653         }
1654     }
1655 }
1656 
1657 static int spapr_pci_post_load(void *opaque, int version_id)
1658 {
1659     sPAPRPHBState *sphb = opaque;
1660     gpointer key, value;
1661     int i;
1662 
1663     for (i = 0; i < sphb->msi_devs_num; ++i) {
1664         key = g_memdup(&sphb->msi_devs[i].key,
1665                        sizeof(sphb->msi_devs[i].key));
1666         value = g_memdup(&sphb->msi_devs[i].value,
1667                          sizeof(sphb->msi_devs[i].value));
1668         g_hash_table_insert(sphb->msi, key, value);
1669     }
1670     g_free(sphb->msi_devs);
1671     sphb->msi_devs = NULL;
1672     sphb->msi_devs_num = 0;
1673 
1674     return 0;
1675 }
1676 
1677 static bool pre_2_8_migration(void *opaque, int version_id)
1678 {
1679     sPAPRPHBState *sphb = opaque;
1680 
1681     return sphb->pre_2_8_migration;
1682 }
1683 
1684 static const VMStateDescription vmstate_spapr_pci = {
1685     .name = "spapr_pci",
1686     .version_id = 2,
1687     .minimum_version_id = 2,
1688     .pre_save = spapr_pci_pre_save,
1689     .post_load = spapr_pci_post_load,
1690     .fields = (VMStateField[]) {
1691         VMSTATE_UINT64_EQUAL(buid, sPAPRPHBState),
1692         VMSTATE_UINT32_TEST(mig_liobn, sPAPRPHBState, pre_2_8_migration),
1693         VMSTATE_UINT64_TEST(mig_mem_win_addr, sPAPRPHBState, pre_2_8_migration),
1694         VMSTATE_UINT64_TEST(mig_mem_win_size, sPAPRPHBState, pre_2_8_migration),
1695         VMSTATE_UINT64_TEST(mig_io_win_addr, sPAPRPHBState, pre_2_8_migration),
1696         VMSTATE_UINT64_TEST(mig_io_win_size, sPAPRPHBState, pre_2_8_migration),
1697         VMSTATE_STRUCT_ARRAY(lsi_table, sPAPRPHBState, PCI_NUM_PINS, 0,
1698                              vmstate_spapr_pci_lsi, struct spapr_pci_lsi),
1699         VMSTATE_INT32(msi_devs_num, sPAPRPHBState),
1700         VMSTATE_STRUCT_VARRAY_ALLOC(msi_devs, sPAPRPHBState, msi_devs_num, 0,
1701                                     vmstate_spapr_pci_msi, spapr_pci_msi_mig),
1702         VMSTATE_END_OF_LIST()
1703     },
1704 };
1705 
1706 static const char *spapr_phb_root_bus_path(PCIHostState *host_bridge,
1707                                            PCIBus *rootbus)
1708 {
1709     sPAPRPHBState *sphb = SPAPR_PCI_HOST_BRIDGE(host_bridge);
1710 
1711     return sphb->dtbusname;
1712 }
1713 
1714 static void spapr_phb_class_init(ObjectClass *klass, void *data)
1715 {
1716     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(klass);
1717     DeviceClass *dc = DEVICE_CLASS(klass);
1718     HotplugHandlerClass *hp = HOTPLUG_HANDLER_CLASS(klass);
1719 
1720     hc->root_bus_path = spapr_phb_root_bus_path;
1721     dc->realize = spapr_phb_realize;
1722     dc->props = spapr_phb_properties;
1723     dc->reset = spapr_phb_reset;
1724     dc->vmsd = &vmstate_spapr_pci;
1725     set_bit(DEVICE_CATEGORY_BRIDGE, dc->categories);
1726     hp->plug = spapr_phb_hot_plug_child;
1727     hp->unplug = spapr_phb_hot_unplug_child;
1728 }
1729 
1730 static const TypeInfo spapr_phb_info = {
1731     .name          = TYPE_SPAPR_PCI_HOST_BRIDGE,
1732     .parent        = TYPE_PCI_HOST_BRIDGE,
1733     .instance_size = sizeof(sPAPRPHBState),
1734     .class_init    = spapr_phb_class_init,
1735     .interfaces    = (InterfaceInfo[]) {
1736         { TYPE_HOTPLUG_HANDLER },
1737         { }
1738     }
1739 };
1740 
1741 PCIHostState *spapr_create_phb(sPAPRMachineState *spapr, int index)
1742 {
1743     DeviceState *dev;
1744 
1745     dev = qdev_create(NULL, TYPE_SPAPR_PCI_HOST_BRIDGE);
1746     qdev_prop_set_uint32(dev, "index", index);
1747     qdev_init_nofail(dev);
1748 
1749     return PCI_HOST_BRIDGE(dev);
1750 }
1751 
1752 typedef struct sPAPRFDT {
1753     void *fdt;
1754     int node_off;
1755     sPAPRPHBState *sphb;
1756 } sPAPRFDT;
1757 
1758 static void spapr_populate_pci_devices_dt(PCIBus *bus, PCIDevice *pdev,
1759                                           void *opaque)
1760 {
1761     PCIBus *sec_bus;
1762     sPAPRFDT *p = opaque;
1763     int offset;
1764     sPAPRFDT s_fdt;
1765 
1766     offset = spapr_create_pci_child_dt(p->sphb, pdev, p->fdt, p->node_off);
1767     if (!offset) {
1768         error_report("Failed to create pci child device tree node");
1769         return;
1770     }
1771 
1772     if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) !=
1773          PCI_HEADER_TYPE_BRIDGE)) {
1774         return;
1775     }
1776 
1777     sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev));
1778     if (!sec_bus) {
1779         return;
1780     }
1781 
1782     s_fdt.fdt = p->fdt;
1783     s_fdt.node_off = offset;
1784     s_fdt.sphb = p->sphb;
1785     pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
1786                         spapr_populate_pci_devices_dt,
1787                         &s_fdt);
1788 }
1789 
1790 static void spapr_phb_pci_enumerate_bridge(PCIBus *bus, PCIDevice *pdev,
1791                                            void *opaque)
1792 {
1793     unsigned int *bus_no = opaque;
1794     unsigned int primary = *bus_no;
1795     unsigned int subordinate = 0xff;
1796     PCIBus *sec_bus = NULL;
1797 
1798     if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) !=
1799          PCI_HEADER_TYPE_BRIDGE)) {
1800         return;
1801     }
1802 
1803     (*bus_no)++;
1804     pci_default_write_config(pdev, PCI_PRIMARY_BUS, primary, 1);
1805     pci_default_write_config(pdev, PCI_SECONDARY_BUS, *bus_no, 1);
1806     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
1807 
1808     sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev));
1809     if (!sec_bus) {
1810         return;
1811     }
1812 
1813     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, subordinate, 1);
1814     pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
1815                         spapr_phb_pci_enumerate_bridge, bus_no);
1816     pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
1817 }
1818 
1819 static void spapr_phb_pci_enumerate(sPAPRPHBState *phb)
1820 {
1821     PCIBus *bus = PCI_HOST_BRIDGE(phb)->bus;
1822     unsigned int bus_no = 0;
1823 
1824     pci_for_each_device(bus, pci_bus_num(bus),
1825                         spapr_phb_pci_enumerate_bridge,
1826                         &bus_no);
1827 
1828 }
1829 
1830 int spapr_populate_pci_dt(sPAPRPHBState *phb,
1831                           uint32_t xics_phandle,
1832                           void *fdt)
1833 {
1834     int bus_off, i, j, ret;
1835     char nodename[FDT_NAME_MAX];
1836     uint32_t bus_range[] = { cpu_to_be32(0), cpu_to_be32(0xff) };
1837     struct {
1838         uint32_t hi;
1839         uint64_t child;
1840         uint64_t parent;
1841         uint64_t size;
1842     } QEMU_PACKED ranges[] = {
1843         {
1844             cpu_to_be32(b_ss(1)), cpu_to_be64(0),
1845             cpu_to_be64(phb->io_win_addr),
1846             cpu_to_be64(memory_region_size(&phb->iospace)),
1847         },
1848         {
1849             cpu_to_be32(b_ss(2)), cpu_to_be64(SPAPR_PCI_MEM_WIN_BUS_OFFSET),
1850             cpu_to_be64(phb->mem_win_addr),
1851             cpu_to_be64(phb->mem_win_size),
1852         },
1853         {
1854             cpu_to_be32(b_ss(3)), cpu_to_be64(phb->mem64_win_pciaddr),
1855             cpu_to_be64(phb->mem64_win_addr),
1856             cpu_to_be64(phb->mem64_win_size),
1857         },
1858     };
1859     const unsigned sizeof_ranges =
1860         (phb->mem64_win_size ? 3 : 2) * sizeof(ranges[0]);
1861     uint64_t bus_reg[] = { cpu_to_be64(phb->buid), 0 };
1862     uint32_t interrupt_map_mask[] = {
1863         cpu_to_be32(b_ddddd(-1)|b_fff(0)), 0x0, 0x0, cpu_to_be32(-1)};
1864     uint32_t interrupt_map[PCI_SLOT_MAX * PCI_NUM_PINS][7];
1865     uint32_t ddw_applicable[] = {
1866         cpu_to_be32(RTAS_IBM_QUERY_PE_DMA_WINDOW),
1867         cpu_to_be32(RTAS_IBM_CREATE_PE_DMA_WINDOW),
1868         cpu_to_be32(RTAS_IBM_REMOVE_PE_DMA_WINDOW)
1869     };
1870     uint32_t ddw_extensions[] = {
1871         cpu_to_be32(1),
1872         cpu_to_be32(RTAS_IBM_RESET_PE_DMA_WINDOW)
1873     };
1874     uint32_t associativity[] = {cpu_to_be32(0x4),
1875                                 cpu_to_be32(0x0),
1876                                 cpu_to_be32(0x0),
1877                                 cpu_to_be32(0x0),
1878                                 cpu_to_be32(phb->numa_node)};
1879     sPAPRTCETable *tcet;
1880     PCIBus *bus = PCI_HOST_BRIDGE(phb)->bus;
1881     sPAPRFDT s_fdt;
1882 
1883     /* Start populating the FDT */
1884     snprintf(nodename, FDT_NAME_MAX, "pci@%" PRIx64, phb->buid);
1885     bus_off = fdt_add_subnode(fdt, 0, nodename);
1886     if (bus_off < 0) {
1887         return bus_off;
1888     }
1889 
1890     /* Write PHB properties */
1891     _FDT(fdt_setprop_string(fdt, bus_off, "device_type", "pci"));
1892     _FDT(fdt_setprop_string(fdt, bus_off, "compatible", "IBM,Logical_PHB"));
1893     _FDT(fdt_setprop_cell(fdt, bus_off, "#address-cells", 0x3));
1894     _FDT(fdt_setprop_cell(fdt, bus_off, "#size-cells", 0x2));
1895     _FDT(fdt_setprop_cell(fdt, bus_off, "#interrupt-cells", 0x1));
1896     _FDT(fdt_setprop(fdt, bus_off, "used-by-rtas", NULL, 0));
1897     _FDT(fdt_setprop(fdt, bus_off, "bus-range", &bus_range, sizeof(bus_range)));
1898     _FDT(fdt_setprop(fdt, bus_off, "ranges", &ranges, sizeof_ranges));
1899     _FDT(fdt_setprop(fdt, bus_off, "reg", &bus_reg, sizeof(bus_reg)));
1900     _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pci-config-space-type", 0x1));
1901     _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pe-total-#msi", XICS_IRQS_SPAPR));
1902 
1903     /* Dynamic DMA window */
1904     if (phb->ddw_enabled) {
1905         _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-applicable", &ddw_applicable,
1906                          sizeof(ddw_applicable)));
1907         _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-extensions",
1908                          &ddw_extensions, sizeof(ddw_extensions)));
1909     }
1910 
1911     /* Advertise NUMA via ibm,associativity */
1912     if (phb->numa_node != -1) {
1913         _FDT(fdt_setprop(fdt, bus_off, "ibm,associativity", associativity,
1914                          sizeof(associativity)));
1915     }
1916 
1917     /* Build the interrupt-map, this must matches what is done
1918      * in pci_spapr_map_irq
1919      */
1920     _FDT(fdt_setprop(fdt, bus_off, "interrupt-map-mask",
1921                      &interrupt_map_mask, sizeof(interrupt_map_mask)));
1922     for (i = 0; i < PCI_SLOT_MAX; i++) {
1923         for (j = 0; j < PCI_NUM_PINS; j++) {
1924             uint32_t *irqmap = interrupt_map[i*PCI_NUM_PINS + j];
1925             int lsi_num = pci_spapr_swizzle(i, j);
1926 
1927             irqmap[0] = cpu_to_be32(b_ddddd(i)|b_fff(0));
1928             irqmap[1] = 0;
1929             irqmap[2] = 0;
1930             irqmap[3] = cpu_to_be32(j+1);
1931             irqmap[4] = cpu_to_be32(xics_phandle);
1932             irqmap[5] = cpu_to_be32(phb->lsi_table[lsi_num].irq);
1933             irqmap[6] = cpu_to_be32(0x8);
1934         }
1935     }
1936     /* Write interrupt map */
1937     _FDT(fdt_setprop(fdt, bus_off, "interrupt-map", &interrupt_map,
1938                      sizeof(interrupt_map)));
1939 
1940     tcet = spapr_tce_find_by_liobn(phb->dma_liobn[0]);
1941     if (!tcet) {
1942         return -1;
1943     }
1944     spapr_dma_dt(fdt, bus_off, "ibm,dma-window",
1945                  tcet->liobn, tcet->bus_offset,
1946                  tcet->nb_table << tcet->page_shift);
1947 
1948     /* Walk the bridges and program the bus numbers*/
1949     spapr_phb_pci_enumerate(phb);
1950     _FDT(fdt_setprop_cell(fdt, bus_off, "qemu,phb-enumerated", 0x1));
1951 
1952     /* Populate tree nodes with PCI devices attached */
1953     s_fdt.fdt = fdt;
1954     s_fdt.node_off = bus_off;
1955     s_fdt.sphb = phb;
1956     pci_for_each_device(bus, pci_bus_num(bus),
1957                         spapr_populate_pci_devices_dt,
1958                         &s_fdt);
1959 
1960     ret = spapr_drc_populate_dt(fdt, bus_off, OBJECT(phb),
1961                                 SPAPR_DR_CONNECTOR_TYPE_PCI);
1962     if (ret) {
1963         return ret;
1964     }
1965 
1966     return 0;
1967 }
1968 
1969 void spapr_pci_rtas_init(void)
1970 {
1971     spapr_rtas_register(RTAS_READ_PCI_CONFIG, "read-pci-config",
1972                         rtas_read_pci_config);
1973     spapr_rtas_register(RTAS_WRITE_PCI_CONFIG, "write-pci-config",
1974                         rtas_write_pci_config);
1975     spapr_rtas_register(RTAS_IBM_READ_PCI_CONFIG, "ibm,read-pci-config",
1976                         rtas_ibm_read_pci_config);
1977     spapr_rtas_register(RTAS_IBM_WRITE_PCI_CONFIG, "ibm,write-pci-config",
1978                         rtas_ibm_write_pci_config);
1979     if (msi_nonbroken) {
1980         spapr_rtas_register(RTAS_IBM_QUERY_INTERRUPT_SOURCE_NUMBER,
1981                             "ibm,query-interrupt-source-number",
1982                             rtas_ibm_query_interrupt_source_number);
1983         spapr_rtas_register(RTAS_IBM_CHANGE_MSI, "ibm,change-msi",
1984                             rtas_ibm_change_msi);
1985     }
1986 
1987     spapr_rtas_register(RTAS_IBM_SET_EEH_OPTION,
1988                         "ibm,set-eeh-option",
1989                         rtas_ibm_set_eeh_option);
1990     spapr_rtas_register(RTAS_IBM_GET_CONFIG_ADDR_INFO2,
1991                         "ibm,get-config-addr-info2",
1992                         rtas_ibm_get_config_addr_info2);
1993     spapr_rtas_register(RTAS_IBM_READ_SLOT_RESET_STATE2,
1994                         "ibm,read-slot-reset-state2",
1995                         rtas_ibm_read_slot_reset_state2);
1996     spapr_rtas_register(RTAS_IBM_SET_SLOT_RESET,
1997                         "ibm,set-slot-reset",
1998                         rtas_ibm_set_slot_reset);
1999     spapr_rtas_register(RTAS_IBM_CONFIGURE_PE,
2000                         "ibm,configure-pe",
2001                         rtas_ibm_configure_pe);
2002     spapr_rtas_register(RTAS_IBM_SLOT_ERROR_DETAIL,
2003                         "ibm,slot-error-detail",
2004                         rtas_ibm_slot_error_detail);
2005 }
2006 
2007 static void spapr_pci_register_types(void)
2008 {
2009     type_register_static(&spapr_phb_info);
2010 }
2011 
2012 type_init(spapr_pci_register_types)
2013 
2014 static int spapr_switch_one_vga(DeviceState *dev, void *opaque)
2015 {
2016     bool be = *(bool *)opaque;
2017 
2018     if (object_dynamic_cast(OBJECT(dev), "VGA")
2019         || object_dynamic_cast(OBJECT(dev), "secondary-vga")) {
2020         object_property_set_bool(OBJECT(dev), be, "big-endian-framebuffer",
2021                                  &error_abort);
2022     }
2023     return 0;
2024 }
2025 
2026 void spapr_pci_switch_vga(bool big_endian)
2027 {
2028     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
2029     sPAPRPHBState *sphb;
2030 
2031     /*
2032      * For backward compatibility with existing guests, we switch
2033      * the endianness of the VGA controller when changing the guest
2034      * interrupt mode
2035      */
2036     QLIST_FOREACH(sphb, &spapr->phbs, list) {
2037         BusState *bus = &PCI_HOST_BRIDGE(sphb)->bus->qbus;
2038         qbus_walk_children(bus, spapr_switch_one_vga, NULL, NULL, NULL,
2039                            &big_endian);
2040     }
2041 }
2042