xref: /qemu/hw/ppc/spapr.c (revision 7265bc3e5429a630a8de2459d510d5d66a072af9)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  */
26 
27 #include "qemu/osdep.h"
28 #include "qemu-common.h"
29 #include "qemu/datadir.h"
30 #include "qapi/error.h"
31 #include "qapi/visitor.h"
32 #include "sysemu/sysemu.h"
33 #include "sysemu/hostmem.h"
34 #include "sysemu/numa.h"
35 #include "sysemu/qtest.h"
36 #include "sysemu/reset.h"
37 #include "sysemu/runstate.h"
38 #include "qemu/log.h"
39 #include "hw/fw-path-provider.h"
40 #include "elf.h"
41 #include "net/net.h"
42 #include "sysemu/device_tree.h"
43 #include "sysemu/cpus.h"
44 #include "sysemu/hw_accel.h"
45 #include "kvm_ppc.h"
46 #include "migration/misc.h"
47 #include "migration/qemu-file-types.h"
48 #include "migration/global_state.h"
49 #include "migration/register.h"
50 #include "migration/blocker.h"
51 #include "mmu-hash64.h"
52 #include "mmu-book3s-v3.h"
53 #include "cpu-models.h"
54 #include "hw/core/cpu.h"
55 
56 #include "hw/boards.h"
57 #include "hw/ppc/ppc.h"
58 #include "hw/loader.h"
59 
60 #include "hw/ppc/fdt.h"
61 #include "hw/ppc/spapr.h"
62 #include "hw/ppc/spapr_vio.h"
63 #include "hw/qdev-properties.h"
64 #include "hw/pci-host/spapr.h"
65 #include "hw/pci/msi.h"
66 
67 #include "hw/pci/pci.h"
68 #include "hw/scsi/scsi.h"
69 #include "hw/virtio/virtio-scsi.h"
70 #include "hw/virtio/vhost-scsi-common.h"
71 
72 #include "exec/address-spaces.h"
73 #include "exec/ram_addr.h"
74 #include "hw/usb.h"
75 #include "qemu/config-file.h"
76 #include "qemu/error-report.h"
77 #include "trace.h"
78 #include "hw/nmi.h"
79 #include "hw/intc/intc.h"
80 
81 #include "hw/ppc/spapr_cpu_core.h"
82 #include "hw/mem/memory-device.h"
83 #include "hw/ppc/spapr_tpm_proxy.h"
84 #include "hw/ppc/spapr_nvdimm.h"
85 #include "hw/ppc/spapr_numa.h"
86 #include "hw/ppc/pef.h"
87 
88 #include "monitor/monitor.h"
89 
90 #include <libfdt.h>
91 
92 /* SLOF memory layout:
93  *
94  * SLOF raw image loaded at 0, copies its romfs right below the flat
95  * device-tree, then position SLOF itself 31M below that
96  *
97  * So we set FW_OVERHEAD to 40MB which should account for all of that
98  * and more
99  *
100  * We load our kernel at 4M, leaving space for SLOF initial image
101  */
102 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
103 #define FW_MAX_SIZE             0x400000
104 #define FW_FILE_NAME            "slof.bin"
105 #define FW_OVERHEAD             0x2800000
106 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
107 
108 #define MIN_RMA_SLOF            (128 * MiB)
109 
110 #define PHANDLE_INTC            0x00001111
111 
112 /* These two functions implement the VCPU id numbering: one to compute them
113  * all and one to identify thread 0 of a VCORE. Any change to the first one
114  * is likely to have an impact on the second one, so let's keep them close.
115  */
116 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
117 {
118     MachineState *ms = MACHINE(spapr);
119     unsigned int smp_threads = ms->smp.threads;
120 
121     assert(spapr->vsmt);
122     return
123         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
124 }
125 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
126                                       PowerPCCPU *cpu)
127 {
128     assert(spapr->vsmt);
129     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
130 }
131 
132 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
133 {
134     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
135      * and newer QEMUs don't even have them. In both cases, we don't want
136      * to send anything on the wire.
137      */
138     return false;
139 }
140 
141 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
142     .name = "icp/server",
143     .version_id = 1,
144     .minimum_version_id = 1,
145     .needed = pre_2_10_vmstate_dummy_icp_needed,
146     .fields = (VMStateField[]) {
147         VMSTATE_UNUSED(4), /* uint32_t xirr */
148         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
149         VMSTATE_UNUSED(1), /* uint8_t mfrr */
150         VMSTATE_END_OF_LIST()
151     },
152 };
153 
154 static void pre_2_10_vmstate_register_dummy_icp(int i)
155 {
156     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
157                      (void *)(uintptr_t) i);
158 }
159 
160 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
161 {
162     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
163                        (void *)(uintptr_t) i);
164 }
165 
166 int spapr_max_server_number(SpaprMachineState *spapr)
167 {
168     MachineState *ms = MACHINE(spapr);
169 
170     assert(spapr->vsmt);
171     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
172 }
173 
174 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
175                                   int smt_threads)
176 {
177     int i, ret = 0;
178     uint32_t servers_prop[smt_threads];
179     uint32_t gservers_prop[smt_threads * 2];
180     int index = spapr_get_vcpu_id(cpu);
181 
182     if (cpu->compat_pvr) {
183         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
184         if (ret < 0) {
185             return ret;
186         }
187     }
188 
189     /* Build interrupt servers and gservers properties */
190     for (i = 0; i < smt_threads; i++) {
191         servers_prop[i] = cpu_to_be32(index + i);
192         /* Hack, direct the group queues back to cpu 0 */
193         gservers_prop[i*2] = cpu_to_be32(index + i);
194         gservers_prop[i*2 + 1] = 0;
195     }
196     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
197                       servers_prop, sizeof(servers_prop));
198     if (ret < 0) {
199         return ret;
200     }
201     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
202                       gservers_prop, sizeof(gservers_prop));
203 
204     return ret;
205 }
206 
207 static void spapr_dt_pa_features(SpaprMachineState *spapr,
208                                  PowerPCCPU *cpu,
209                                  void *fdt, int offset)
210 {
211     uint8_t pa_features_206[] = { 6, 0,
212         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
213     uint8_t pa_features_207[] = { 24, 0,
214         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
215         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
216         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
217         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
218     uint8_t pa_features_300[] = { 66, 0,
219         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
220         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
221         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
222         /* 6: DS207 */
223         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
224         /* 16: Vector */
225         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
226         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
227         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
228         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
229         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
230         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
231         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
232         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
233         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
234         /* 42: PM, 44: PC RA, 46: SC vec'd */
235         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
236         /* 48: SIMD, 50: QP BFP, 52: String */
237         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
238         /* 54: DecFP, 56: DecI, 58: SHA */
239         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
240         /* 60: NM atomic, 62: RNG */
241         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
242     };
243     uint8_t *pa_features = NULL;
244     size_t pa_size;
245 
246     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
247         pa_features = pa_features_206;
248         pa_size = sizeof(pa_features_206);
249     }
250     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
251         pa_features = pa_features_207;
252         pa_size = sizeof(pa_features_207);
253     }
254     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
255         pa_features = pa_features_300;
256         pa_size = sizeof(pa_features_300);
257     }
258     if (!pa_features) {
259         return;
260     }
261 
262     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
263         /*
264          * Note: we keep CI large pages off by default because a 64K capable
265          * guest provisioned with large pages might otherwise try to map a qemu
266          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
267          * even if that qemu runs on a 4k host.
268          * We dd this bit back here if we are confident this is not an issue
269          */
270         pa_features[3] |= 0x20;
271     }
272     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
273         pa_features[24] |= 0x80;    /* Transactional memory support */
274     }
275     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
276         /* Workaround for broken kernels that attempt (guest) radix
277          * mode when they can't handle it, if they see the radix bit set
278          * in pa-features. So hide it from them. */
279         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
280     }
281 
282     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
283 }
284 
285 static hwaddr spapr_node0_size(MachineState *machine)
286 {
287     if (machine->numa_state->num_nodes) {
288         int i;
289         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
290             if (machine->numa_state->nodes[i].node_mem) {
291                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
292                            machine->ram_size);
293             }
294         }
295     }
296     return machine->ram_size;
297 }
298 
299 bool spapr_machine_using_legacy_numa(SpaprMachineState *spapr)
300 {
301     MachineState *machine = MACHINE(spapr);
302     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
303 
304     return smc->pre_5_2_numa_associativity ||
305            machine->numa_state->num_nodes <= 1;
306 }
307 
308 static void add_str(GString *s, const gchar *s1)
309 {
310     g_string_append_len(s, s1, strlen(s1) + 1);
311 }
312 
313 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid,
314                                 hwaddr start, hwaddr size)
315 {
316     char mem_name[32];
317     uint64_t mem_reg_property[2];
318     int off;
319 
320     mem_reg_property[0] = cpu_to_be64(start);
321     mem_reg_property[1] = cpu_to_be64(size);
322 
323     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
324     off = fdt_add_subnode(fdt, 0, mem_name);
325     _FDT(off);
326     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
327     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
328                       sizeof(mem_reg_property))));
329     spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid);
330     return off;
331 }
332 
333 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
334 {
335     MemoryDeviceInfoList *info;
336 
337     for (info = list; info; info = info->next) {
338         MemoryDeviceInfo *value = info->value;
339 
340         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
341             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
342 
343             if (addr >= pcdimm_info->addr &&
344                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
345                 return pcdimm_info->node;
346             }
347         }
348     }
349 
350     return -1;
351 }
352 
353 struct sPAPRDrconfCellV2 {
354      uint32_t seq_lmbs;
355      uint64_t base_addr;
356      uint32_t drc_index;
357      uint32_t aa_index;
358      uint32_t flags;
359 } QEMU_PACKED;
360 
361 typedef struct DrconfCellQueue {
362     struct sPAPRDrconfCellV2 cell;
363     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
364 } DrconfCellQueue;
365 
366 static DrconfCellQueue *
367 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
368                       uint32_t drc_index, uint32_t aa_index,
369                       uint32_t flags)
370 {
371     DrconfCellQueue *elem;
372 
373     elem = g_malloc0(sizeof(*elem));
374     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
375     elem->cell.base_addr = cpu_to_be64(base_addr);
376     elem->cell.drc_index = cpu_to_be32(drc_index);
377     elem->cell.aa_index = cpu_to_be32(aa_index);
378     elem->cell.flags = cpu_to_be32(flags);
379 
380     return elem;
381 }
382 
383 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
384                                       int offset, MemoryDeviceInfoList *dimms)
385 {
386     MachineState *machine = MACHINE(spapr);
387     uint8_t *int_buf, *cur_index;
388     int ret;
389     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
390     uint64_t addr, cur_addr, size;
391     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
392     uint64_t mem_end = machine->device_memory->base +
393                        memory_region_size(&machine->device_memory->mr);
394     uint32_t node, buf_len, nr_entries = 0;
395     SpaprDrc *drc;
396     DrconfCellQueue *elem, *next;
397     MemoryDeviceInfoList *info;
398     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
399         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
400 
401     /* Entry to cover RAM and the gap area */
402     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
403                                  SPAPR_LMB_FLAGS_RESERVED |
404                                  SPAPR_LMB_FLAGS_DRC_INVALID);
405     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
406     nr_entries++;
407 
408     cur_addr = machine->device_memory->base;
409     for (info = dimms; info; info = info->next) {
410         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
411 
412         addr = di->addr;
413         size = di->size;
414         node = di->node;
415 
416         /*
417          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
418          * area is marked hotpluggable in the next iteration for the bigger
419          * chunk including the NVDIMM occupied area.
420          */
421         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
422             continue;
423 
424         /* Entry for hot-pluggable area */
425         if (cur_addr < addr) {
426             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
427             g_assert(drc);
428             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
429                                          cur_addr, spapr_drc_index(drc), -1, 0);
430             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
431             nr_entries++;
432         }
433 
434         /* Entry for DIMM */
435         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
436         g_assert(drc);
437         elem = spapr_get_drconf_cell(size / lmb_size, addr,
438                                      spapr_drc_index(drc), node,
439                                      (SPAPR_LMB_FLAGS_ASSIGNED |
440                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
441         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
442         nr_entries++;
443         cur_addr = addr + size;
444     }
445 
446     /* Entry for remaining hotpluggable area */
447     if (cur_addr < mem_end) {
448         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
449         g_assert(drc);
450         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
451                                      cur_addr, spapr_drc_index(drc), -1, 0);
452         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
453         nr_entries++;
454     }
455 
456     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
457     int_buf = cur_index = g_malloc0(buf_len);
458     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
459     cur_index += sizeof(nr_entries);
460 
461     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
462         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
463         cur_index += sizeof(elem->cell);
464         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
465         g_free(elem);
466     }
467 
468     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
469     g_free(int_buf);
470     if (ret < 0) {
471         return -1;
472     }
473     return 0;
474 }
475 
476 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
477                                    int offset, MemoryDeviceInfoList *dimms)
478 {
479     MachineState *machine = MACHINE(spapr);
480     int i, ret;
481     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
482     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
483     uint32_t nr_lmbs = (machine->device_memory->base +
484                        memory_region_size(&machine->device_memory->mr)) /
485                        lmb_size;
486     uint32_t *int_buf, *cur_index, buf_len;
487 
488     /*
489      * Allocate enough buffer size to fit in ibm,dynamic-memory
490      */
491     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
492     cur_index = int_buf = g_malloc0(buf_len);
493     int_buf[0] = cpu_to_be32(nr_lmbs);
494     cur_index++;
495     for (i = 0; i < nr_lmbs; i++) {
496         uint64_t addr = i * lmb_size;
497         uint32_t *dynamic_memory = cur_index;
498 
499         if (i >= device_lmb_start) {
500             SpaprDrc *drc;
501 
502             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
503             g_assert(drc);
504 
505             dynamic_memory[0] = cpu_to_be32(addr >> 32);
506             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
507             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
508             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
509             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
510             if (memory_region_present(get_system_memory(), addr)) {
511                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
512             } else {
513                 dynamic_memory[5] = cpu_to_be32(0);
514             }
515         } else {
516             /*
517              * LMB information for RMA, boot time RAM and gap b/n RAM and
518              * device memory region -- all these are marked as reserved
519              * and as having no valid DRC.
520              */
521             dynamic_memory[0] = cpu_to_be32(addr >> 32);
522             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
523             dynamic_memory[2] = cpu_to_be32(0);
524             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
525             dynamic_memory[4] = cpu_to_be32(-1);
526             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
527                                             SPAPR_LMB_FLAGS_DRC_INVALID);
528         }
529 
530         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
531     }
532     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
533     g_free(int_buf);
534     if (ret < 0) {
535         return -1;
536     }
537     return 0;
538 }
539 
540 /*
541  * Adds ibm,dynamic-reconfiguration-memory node.
542  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
543  * of this device tree node.
544  */
545 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
546                                                    void *fdt)
547 {
548     MachineState *machine = MACHINE(spapr);
549     int ret, offset;
550     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
551     uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
552                                 cpu_to_be32(lmb_size & 0xffffffff)};
553     MemoryDeviceInfoList *dimms = NULL;
554 
555     /*
556      * Don't create the node if there is no device memory
557      */
558     if (machine->ram_size == machine->maxram_size) {
559         return 0;
560     }
561 
562     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
563 
564     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
565                     sizeof(prop_lmb_size));
566     if (ret < 0) {
567         return ret;
568     }
569 
570     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
571     if (ret < 0) {
572         return ret;
573     }
574 
575     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
576     if (ret < 0) {
577         return ret;
578     }
579 
580     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
581     dimms = qmp_memory_device_list();
582     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
583         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
584     } else {
585         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
586     }
587     qapi_free_MemoryDeviceInfoList(dimms);
588 
589     if (ret < 0) {
590         return ret;
591     }
592 
593     ret = spapr_numa_write_assoc_lookup_arrays(spapr, fdt, offset);
594 
595     return ret;
596 }
597 
598 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
599 {
600     MachineState *machine = MACHINE(spapr);
601     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
602     hwaddr mem_start, node_size;
603     int i, nb_nodes = machine->numa_state->num_nodes;
604     NodeInfo *nodes = machine->numa_state->nodes;
605 
606     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
607         if (!nodes[i].node_mem) {
608             continue;
609         }
610         if (mem_start >= machine->ram_size) {
611             node_size = 0;
612         } else {
613             node_size = nodes[i].node_mem;
614             if (node_size > machine->ram_size - mem_start) {
615                 node_size = machine->ram_size - mem_start;
616             }
617         }
618         if (!mem_start) {
619             /* spapr_machine_init() checks for rma_size <= node0_size
620              * already */
621             spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size);
622             mem_start += spapr->rma_size;
623             node_size -= spapr->rma_size;
624         }
625         for ( ; node_size; ) {
626             hwaddr sizetmp = pow2floor(node_size);
627 
628             /* mem_start != 0 here */
629             if (ctzl(mem_start) < ctzl(sizetmp)) {
630                 sizetmp = 1ULL << ctzl(mem_start);
631             }
632 
633             spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp);
634             node_size -= sizetmp;
635             mem_start += sizetmp;
636         }
637     }
638 
639     /* Generate ibm,dynamic-reconfiguration-memory node if required */
640     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
641         int ret;
642 
643         g_assert(smc->dr_lmb_enabled);
644         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
645         if (ret) {
646             return ret;
647         }
648     }
649 
650     return 0;
651 }
652 
653 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
654                          SpaprMachineState *spapr)
655 {
656     MachineState *ms = MACHINE(spapr);
657     PowerPCCPU *cpu = POWERPC_CPU(cs);
658     CPUPPCState *env = &cpu->env;
659     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
660     int index = spapr_get_vcpu_id(cpu);
661     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
662                        0xffffffff, 0xffffffff};
663     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
664         : SPAPR_TIMEBASE_FREQ;
665     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
666     uint32_t page_sizes_prop[64];
667     size_t page_sizes_prop_size;
668     unsigned int smp_threads = ms->smp.threads;
669     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
670     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
671     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
672     SpaprDrc *drc;
673     int drc_index;
674     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
675     int i;
676 
677     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
678     if (drc) {
679         drc_index = spapr_drc_index(drc);
680         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
681     }
682 
683     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
684     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
685 
686     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
687     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
688                            env->dcache_line_size)));
689     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
690                            env->dcache_line_size)));
691     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
692                            env->icache_line_size)));
693     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
694                            env->icache_line_size)));
695 
696     if (pcc->l1_dcache_size) {
697         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
698                                pcc->l1_dcache_size)));
699     } else {
700         warn_report("Unknown L1 dcache size for cpu");
701     }
702     if (pcc->l1_icache_size) {
703         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
704                                pcc->l1_icache_size)));
705     } else {
706         warn_report("Unknown L1 icache size for cpu");
707     }
708 
709     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
710     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
711     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
712     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
713     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
714     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
715 
716     if (env->spr_cb[SPR_PURR].oea_read) {
717         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
718     }
719     if (env->spr_cb[SPR_SPURR].oea_read) {
720         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
721     }
722 
723     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
724         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
725                           segs, sizeof(segs))));
726     }
727 
728     /* Advertise VSX (vector extensions) if available
729      *   1               == VMX / Altivec available
730      *   2               == VSX available
731      *
732      * Only CPUs for which we create core types in spapr_cpu_core.c
733      * are possible, and all of those have VMX */
734     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
735         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
736     } else {
737         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
738     }
739 
740     /* Advertise DFP (Decimal Floating Point) if available
741      *   0 / no property == no DFP
742      *   1               == DFP available */
743     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
744         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
745     }
746 
747     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
748                                                       sizeof(page_sizes_prop));
749     if (page_sizes_prop_size) {
750         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
751                           page_sizes_prop, page_sizes_prop_size)));
752     }
753 
754     spapr_dt_pa_features(spapr, cpu, fdt, offset);
755 
756     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
757                            cs->cpu_index / vcpus_per_socket)));
758 
759     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
760                       pft_size_prop, sizeof(pft_size_prop))));
761 
762     if (ms->numa_state->num_nodes > 1) {
763         _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
764     }
765 
766     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
767 
768     if (pcc->radix_page_info) {
769         for (i = 0; i < pcc->radix_page_info->count; i++) {
770             radix_AP_encodings[i] =
771                 cpu_to_be32(pcc->radix_page_info->entries[i]);
772         }
773         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
774                           radix_AP_encodings,
775                           pcc->radix_page_info->count *
776                           sizeof(radix_AP_encodings[0]))));
777     }
778 
779     /*
780      * We set this property to let the guest know that it can use the large
781      * decrementer and its width in bits.
782      */
783     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
784         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
785                               pcc->lrg_decr_bits)));
786 }
787 
788 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
789 {
790     CPUState **rev;
791     CPUState *cs;
792     int n_cpus;
793     int cpus_offset;
794     int i;
795 
796     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
797     _FDT(cpus_offset);
798     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
799     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
800 
801     /*
802      * We walk the CPUs in reverse order to ensure that CPU DT nodes
803      * created by fdt_add_subnode() end up in the right order in FDT
804      * for the guest kernel the enumerate the CPUs correctly.
805      *
806      * The CPU list cannot be traversed in reverse order, so we need
807      * to do extra work.
808      */
809     n_cpus = 0;
810     rev = NULL;
811     CPU_FOREACH(cs) {
812         rev = g_renew(CPUState *, rev, n_cpus + 1);
813         rev[n_cpus++] = cs;
814     }
815 
816     for (i = n_cpus - 1; i >= 0; i--) {
817         CPUState *cs = rev[i];
818         PowerPCCPU *cpu = POWERPC_CPU(cs);
819         int index = spapr_get_vcpu_id(cpu);
820         DeviceClass *dc = DEVICE_GET_CLASS(cs);
821         g_autofree char *nodename = NULL;
822         int offset;
823 
824         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
825             continue;
826         }
827 
828         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
829         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
830         _FDT(offset);
831         spapr_dt_cpu(cs, fdt, offset, spapr);
832     }
833 
834     g_free(rev);
835 }
836 
837 static int spapr_dt_rng(void *fdt)
838 {
839     int node;
840     int ret;
841 
842     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
843     if (node <= 0) {
844         return -1;
845     }
846     ret = fdt_setprop_string(fdt, node, "device_type",
847                              "ibm,platform-facilities");
848     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
849     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
850 
851     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
852     if (node <= 0) {
853         return -1;
854     }
855     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
856 
857     return ret ? -1 : 0;
858 }
859 
860 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
861 {
862     MachineState *ms = MACHINE(spapr);
863     int rtas;
864     GString *hypertas = g_string_sized_new(256);
865     GString *qemu_hypertas = g_string_sized_new(256);
866     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
867         memory_region_size(&MACHINE(spapr)->device_memory->mr);
868     uint32_t lrdr_capacity[] = {
869         cpu_to_be32(max_device_addr >> 32),
870         cpu_to_be32(max_device_addr & 0xffffffff),
871         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
872         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
873         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
874     };
875 
876     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
877 
878     /* hypertas */
879     add_str(hypertas, "hcall-pft");
880     add_str(hypertas, "hcall-term");
881     add_str(hypertas, "hcall-dabr");
882     add_str(hypertas, "hcall-interrupt");
883     add_str(hypertas, "hcall-tce");
884     add_str(hypertas, "hcall-vio");
885     add_str(hypertas, "hcall-splpar");
886     add_str(hypertas, "hcall-join");
887     add_str(hypertas, "hcall-bulk");
888     add_str(hypertas, "hcall-set-mode");
889     add_str(hypertas, "hcall-sprg0");
890     add_str(hypertas, "hcall-copy");
891     add_str(hypertas, "hcall-debug");
892     add_str(hypertas, "hcall-vphn");
893     add_str(qemu_hypertas, "hcall-memop1");
894 
895     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
896         add_str(hypertas, "hcall-multi-tce");
897     }
898 
899     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
900         add_str(hypertas, "hcall-hpt-resize");
901     }
902 
903     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
904                      hypertas->str, hypertas->len));
905     g_string_free(hypertas, TRUE);
906     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
907                      qemu_hypertas->str, qemu_hypertas->len));
908     g_string_free(qemu_hypertas, TRUE);
909 
910     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
911 
912     /*
913      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
914      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
915      *
916      * The system reset requirements are driven by existing Linux and PowerVM
917      * implementation which (contrary to PAPR) saves r3 in the error log
918      * structure like machine check, so Linux expects to find the saved r3
919      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
920      * does not look at the error value).
921      *
922      * System reset interrupts are not subject to interlock like machine
923      * check, so this memory area could be corrupted if the sreset is
924      * interrupted by a machine check (or vice versa) if it was shared. To
925      * prevent this, system reset uses per-CPU areas for the sreset save
926      * area. A system reset that interrupts a system reset handler could
927      * still overwrite this area, but Linux doesn't try to recover in that
928      * case anyway.
929      *
930      * The extra 8 bytes is required because Linux's FWNMI error log check
931      * is off-by-one.
932      */
933     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_ERROR_LOG_MAX +
934 			  ms->smp.max_cpus * sizeof(uint64_t)*2 + sizeof(uint64_t)));
935     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
936                           RTAS_ERROR_LOG_MAX));
937     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
938                           RTAS_EVENT_SCAN_RATE));
939 
940     g_assert(msi_nonbroken);
941     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
942 
943     /*
944      * According to PAPR, rtas ibm,os-term does not guarantee a return
945      * back to the guest cpu.
946      *
947      * While an additional ibm,extended-os-term property indicates
948      * that rtas call return will always occur. Set this property.
949      */
950     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
951 
952     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
953                      lrdr_capacity, sizeof(lrdr_capacity)));
954 
955     spapr_dt_rtas_tokens(fdt, rtas);
956 }
957 
958 /*
959  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
960  * and the XIVE features that the guest may request and thus the valid
961  * values for bytes 23..26 of option vector 5:
962  */
963 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
964                                           int chosen)
965 {
966     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
967 
968     char val[2 * 4] = {
969         23, 0x00, /* XICS / XIVE mode */
970         24, 0x00, /* Hash/Radix, filled in below. */
971         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
972         26, 0x40, /* Radix options: GTSE == yes. */
973     };
974 
975     if (spapr->irq->xics && spapr->irq->xive) {
976         val[1] = SPAPR_OV5_XIVE_BOTH;
977     } else if (spapr->irq->xive) {
978         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
979     } else {
980         assert(spapr->irq->xics);
981         val[1] = SPAPR_OV5_XIVE_LEGACY;
982     }
983 
984     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
985                           first_ppc_cpu->compat_pvr)) {
986         /*
987          * If we're in a pre POWER9 compat mode then the guest should
988          * do hash and use the legacy interrupt mode
989          */
990         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
991         val[3] = 0x00; /* Hash */
992     } else if (kvm_enabled()) {
993         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
994             val[3] = 0x80; /* OV5_MMU_BOTH */
995         } else if (kvmppc_has_cap_mmu_radix()) {
996             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
997         } else {
998             val[3] = 0x00; /* Hash */
999         }
1000     } else {
1001         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1002         val[3] = 0xC0;
1003     }
1004     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1005                      val, sizeof(val)));
1006 }
1007 
1008 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1009 {
1010     MachineState *machine = MACHINE(spapr);
1011     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1012     int chosen;
1013 
1014     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1015 
1016     if (reset) {
1017         const char *boot_device = machine->boot_order;
1018         char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1019         size_t cb = 0;
1020         char *bootlist = get_boot_devices_list(&cb);
1021 
1022         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1023             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1024                                     machine->kernel_cmdline));
1025         }
1026 
1027         if (spapr->initrd_size) {
1028             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1029                                   spapr->initrd_base));
1030             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1031                                   spapr->initrd_base + spapr->initrd_size));
1032         }
1033 
1034         if (spapr->kernel_size) {
1035             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1036                                   cpu_to_be64(spapr->kernel_size) };
1037 
1038             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1039                          &kprop, sizeof(kprop)));
1040             if (spapr->kernel_le) {
1041                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1042             }
1043         }
1044         if (boot_menu) {
1045             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1046         }
1047         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1048         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1049         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1050 
1051         if (cb && bootlist) {
1052             int i;
1053 
1054             for (i = 0; i < cb; i++) {
1055                 if (bootlist[i] == '\n') {
1056                     bootlist[i] = ' ';
1057                 }
1058             }
1059             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1060         }
1061 
1062         if (boot_device && strlen(boot_device)) {
1063             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1064         }
1065 
1066         if (!spapr->has_graphics && stdout_path) {
1067             /*
1068              * "linux,stdout-path" and "stdout" properties are
1069              * deprecated by linux kernel. New platforms should only
1070              * use the "stdout-path" property. Set the new property
1071              * and continue using older property to remain compatible
1072              * with the existing firmware.
1073              */
1074             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1075             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1076         }
1077 
1078         /*
1079          * We can deal with BAR reallocation just fine, advertise it
1080          * to the guest
1081          */
1082         if (smc->linux_pci_probe) {
1083             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1084         }
1085 
1086         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1087 
1088         g_free(stdout_path);
1089         g_free(bootlist);
1090     }
1091 
1092     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1093 }
1094 
1095 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1096 {
1097     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1098      * KVM to work under pHyp with some guest co-operation */
1099     int hypervisor;
1100     uint8_t hypercall[16];
1101 
1102     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1103     /* indicate KVM hypercall interface */
1104     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1105     if (kvmppc_has_cap_fixup_hcalls()) {
1106         /*
1107          * Older KVM versions with older guest kernels were broken
1108          * with the magic page, don't allow the guest to map it.
1109          */
1110         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1111                                   sizeof(hypercall))) {
1112             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1113                              hypercall, sizeof(hypercall)));
1114         }
1115     }
1116 }
1117 
1118 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1119 {
1120     MachineState *machine = MACHINE(spapr);
1121     MachineClass *mc = MACHINE_GET_CLASS(machine);
1122     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1123     uint32_t root_drc_type_mask = 0;
1124     int ret;
1125     void *fdt;
1126     SpaprPhbState *phb;
1127     char *buf;
1128 
1129     fdt = g_malloc0(space);
1130     _FDT((fdt_create_empty_tree(fdt, space)));
1131 
1132     /* Root node */
1133     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1134     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1135     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1136 
1137     /* Guest UUID & Name*/
1138     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1139     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1140     if (qemu_uuid_set) {
1141         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1142     }
1143     g_free(buf);
1144 
1145     if (qemu_get_vm_name()) {
1146         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1147                                 qemu_get_vm_name()));
1148     }
1149 
1150     /* Host Model & Serial Number */
1151     if (spapr->host_model) {
1152         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1153     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1154         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1155         g_free(buf);
1156     }
1157 
1158     if (spapr->host_serial) {
1159         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1160     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1161         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1162         g_free(buf);
1163     }
1164 
1165     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1166     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1167 
1168     /* /interrupt controller */
1169     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1170 
1171     ret = spapr_dt_memory(spapr, fdt);
1172     if (ret < 0) {
1173         error_report("couldn't setup memory nodes in fdt");
1174         exit(1);
1175     }
1176 
1177     /* /vdevice */
1178     spapr_dt_vdevice(spapr->vio_bus, fdt);
1179 
1180     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1181         ret = spapr_dt_rng(fdt);
1182         if (ret < 0) {
1183             error_report("could not set up rng device in the fdt");
1184             exit(1);
1185         }
1186     }
1187 
1188     QLIST_FOREACH(phb, &spapr->phbs, list) {
1189         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1190         if (ret < 0) {
1191             error_report("couldn't setup PCI devices in fdt");
1192             exit(1);
1193         }
1194     }
1195 
1196     spapr_dt_cpus(fdt, spapr);
1197 
1198     /* ibm,drc-indexes and friends */
1199     if (smc->dr_lmb_enabled) {
1200         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_LMB;
1201     }
1202     if (smc->dr_phb_enabled) {
1203         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PHB;
1204     }
1205     if (mc->nvdimm_supported) {
1206         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PMEM;
1207     }
1208     if (root_drc_type_mask) {
1209         _FDT(spapr_dt_drc(fdt, 0, NULL, root_drc_type_mask));
1210     }
1211 
1212     if (mc->has_hotpluggable_cpus) {
1213         int offset = fdt_path_offset(fdt, "/cpus");
1214         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1215         if (ret < 0) {
1216             error_report("Couldn't set up CPU DR device tree properties");
1217             exit(1);
1218         }
1219     }
1220 
1221     /* /event-sources */
1222     spapr_dt_events(spapr, fdt);
1223 
1224     /* /rtas */
1225     spapr_dt_rtas(spapr, fdt);
1226 
1227     /* /chosen */
1228     spapr_dt_chosen(spapr, fdt, reset);
1229 
1230     /* /hypervisor */
1231     if (kvm_enabled()) {
1232         spapr_dt_hypervisor(spapr, fdt);
1233     }
1234 
1235     /* Build memory reserve map */
1236     if (reset) {
1237         if (spapr->kernel_size) {
1238             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1239                                   spapr->kernel_size)));
1240         }
1241         if (spapr->initrd_size) {
1242             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1243                                   spapr->initrd_size)));
1244         }
1245     }
1246 
1247     /* NVDIMM devices */
1248     if (mc->nvdimm_supported) {
1249         spapr_dt_persistent_memory(spapr, fdt);
1250     }
1251 
1252     return fdt;
1253 }
1254 
1255 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1256 {
1257     SpaprMachineState *spapr = opaque;
1258 
1259     return (addr & 0x0fffffff) + spapr->kernel_addr;
1260 }
1261 
1262 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1263                                     PowerPCCPU *cpu)
1264 {
1265     CPUPPCState *env = &cpu->env;
1266 
1267     /* The TCG path should also be holding the BQL at this point */
1268     g_assert(qemu_mutex_iothread_locked());
1269 
1270     if (msr_pr) {
1271         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1272         env->gpr[3] = H_PRIVILEGE;
1273     } else {
1274         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1275     }
1276 }
1277 
1278 struct LPCRSyncState {
1279     target_ulong value;
1280     target_ulong mask;
1281 };
1282 
1283 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1284 {
1285     struct LPCRSyncState *s = arg.host_ptr;
1286     PowerPCCPU *cpu = POWERPC_CPU(cs);
1287     CPUPPCState *env = &cpu->env;
1288     target_ulong lpcr;
1289 
1290     cpu_synchronize_state(cs);
1291     lpcr = env->spr[SPR_LPCR];
1292     lpcr &= ~s->mask;
1293     lpcr |= s->value;
1294     ppc_store_lpcr(cpu, lpcr);
1295 }
1296 
1297 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1298 {
1299     CPUState *cs;
1300     struct LPCRSyncState s = {
1301         .value = value,
1302         .mask = mask
1303     };
1304     CPU_FOREACH(cs) {
1305         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1306     }
1307 }
1308 
1309 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1310 {
1311     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1312 
1313     /* Copy PATE1:GR into PATE0:HR */
1314     entry->dw0 = spapr->patb_entry & PATE0_HR;
1315     entry->dw1 = spapr->patb_entry;
1316 }
1317 
1318 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1319 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1320 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1321 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1322 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1323 
1324 /*
1325  * Get the fd to access the kernel htab, re-opening it if necessary
1326  */
1327 static int get_htab_fd(SpaprMachineState *spapr)
1328 {
1329     Error *local_err = NULL;
1330 
1331     if (spapr->htab_fd >= 0) {
1332         return spapr->htab_fd;
1333     }
1334 
1335     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1336     if (spapr->htab_fd < 0) {
1337         error_report_err(local_err);
1338     }
1339 
1340     return spapr->htab_fd;
1341 }
1342 
1343 void close_htab_fd(SpaprMachineState *spapr)
1344 {
1345     if (spapr->htab_fd >= 0) {
1346         close(spapr->htab_fd);
1347     }
1348     spapr->htab_fd = -1;
1349 }
1350 
1351 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1352 {
1353     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1354 
1355     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1356 }
1357 
1358 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1359 {
1360     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1361 
1362     assert(kvm_enabled());
1363 
1364     if (!spapr->htab) {
1365         return 0;
1366     }
1367 
1368     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1369 }
1370 
1371 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1372                                                 hwaddr ptex, int n)
1373 {
1374     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1375     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1376 
1377     if (!spapr->htab) {
1378         /*
1379          * HTAB is controlled by KVM. Fetch into temporary buffer
1380          */
1381         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1382         kvmppc_read_hptes(hptes, ptex, n);
1383         return hptes;
1384     }
1385 
1386     /*
1387      * HTAB is controlled by QEMU. Just point to the internally
1388      * accessible PTEG.
1389      */
1390     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1391 }
1392 
1393 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1394                               const ppc_hash_pte64_t *hptes,
1395                               hwaddr ptex, int n)
1396 {
1397     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1398 
1399     if (!spapr->htab) {
1400         g_free((void *)hptes);
1401     }
1402 
1403     /* Nothing to do for qemu managed HPT */
1404 }
1405 
1406 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1407                       uint64_t pte0, uint64_t pte1)
1408 {
1409     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1410     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1411 
1412     if (!spapr->htab) {
1413         kvmppc_write_hpte(ptex, pte0, pte1);
1414     } else {
1415         if (pte0 & HPTE64_V_VALID) {
1416             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1417             /*
1418              * When setting valid, we write PTE1 first. This ensures
1419              * proper synchronization with the reading code in
1420              * ppc_hash64_pteg_search()
1421              */
1422             smp_wmb();
1423             stq_p(spapr->htab + offset, pte0);
1424         } else {
1425             stq_p(spapr->htab + offset, pte0);
1426             /*
1427              * When clearing it we set PTE0 first. This ensures proper
1428              * synchronization with the reading code in
1429              * ppc_hash64_pteg_search()
1430              */
1431             smp_wmb();
1432             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1433         }
1434     }
1435 }
1436 
1437 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1438                              uint64_t pte1)
1439 {
1440     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1441     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1442 
1443     if (!spapr->htab) {
1444         /* There should always be a hash table when this is called */
1445         error_report("spapr_hpte_set_c called with no hash table !");
1446         return;
1447     }
1448 
1449     /* The HW performs a non-atomic byte update */
1450     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1451 }
1452 
1453 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1454                              uint64_t pte1)
1455 {
1456     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1457     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1458 
1459     if (!spapr->htab) {
1460         /* There should always be a hash table when this is called */
1461         error_report("spapr_hpte_set_r called with no hash table !");
1462         return;
1463     }
1464 
1465     /* The HW performs a non-atomic byte update */
1466     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1467 }
1468 
1469 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1470 {
1471     int shift;
1472 
1473     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1474      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1475      * that's much more than is needed for Linux guests */
1476     shift = ctz64(pow2ceil(ramsize)) - 7;
1477     shift = MAX(shift, 18); /* Minimum architected size */
1478     shift = MIN(shift, 46); /* Maximum architected size */
1479     return shift;
1480 }
1481 
1482 void spapr_free_hpt(SpaprMachineState *spapr)
1483 {
1484     g_free(spapr->htab);
1485     spapr->htab = NULL;
1486     spapr->htab_shift = 0;
1487     close_htab_fd(spapr);
1488 }
1489 
1490 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1491 {
1492     ERRP_GUARD();
1493     long rc;
1494 
1495     /* Clean up any HPT info from a previous boot */
1496     spapr_free_hpt(spapr);
1497 
1498     rc = kvmppc_reset_htab(shift);
1499 
1500     if (rc == -EOPNOTSUPP) {
1501         error_setg(errp, "HPT not supported in nested guests");
1502         return -EOPNOTSUPP;
1503     }
1504 
1505     if (rc < 0) {
1506         /* kernel-side HPT needed, but couldn't allocate one */
1507         error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1508                          shift);
1509         error_append_hint(errp, "Try smaller maxmem?\n");
1510         return -errno;
1511     } else if (rc > 0) {
1512         /* kernel-side HPT allocated */
1513         if (rc != shift) {
1514             error_setg(errp,
1515                        "Requested order %d HPT, but kernel allocated order %ld",
1516                        shift, rc);
1517             error_append_hint(errp, "Try smaller maxmem?\n");
1518             return -ENOSPC;
1519         }
1520 
1521         spapr->htab_shift = shift;
1522         spapr->htab = NULL;
1523     } else {
1524         /* kernel-side HPT not needed, allocate in userspace instead */
1525         size_t size = 1ULL << shift;
1526         int i;
1527 
1528         spapr->htab = qemu_memalign(size, size);
1529         memset(spapr->htab, 0, size);
1530         spapr->htab_shift = shift;
1531 
1532         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1533             DIRTY_HPTE(HPTE(spapr->htab, i));
1534         }
1535     }
1536     /* We're setting up a hash table, so that means we're not radix */
1537     spapr->patb_entry = 0;
1538     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1539     return 0;
1540 }
1541 
1542 void spapr_setup_hpt(SpaprMachineState *spapr)
1543 {
1544     int hpt_shift;
1545 
1546     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1547         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1548     } else {
1549         uint64_t current_ram_size;
1550 
1551         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1552         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1553     }
1554     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1555 
1556     if (kvm_enabled()) {
1557         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1558 
1559         /* Check our RMA fits in the possible VRMA */
1560         if (vrma_limit < spapr->rma_size) {
1561             error_report("Unable to create %" HWADDR_PRIu
1562                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1563                          spapr->rma_size / MiB, vrma_limit / MiB);
1564             exit(EXIT_FAILURE);
1565         }
1566     }
1567 }
1568 
1569 static void spapr_machine_reset(MachineState *machine)
1570 {
1571     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1572     PowerPCCPU *first_ppc_cpu;
1573     hwaddr fdt_addr;
1574     void *fdt;
1575     int rc;
1576 
1577     pef_kvm_reset(machine->cgs, &error_fatal);
1578     spapr_caps_apply(spapr);
1579 
1580     first_ppc_cpu = POWERPC_CPU(first_cpu);
1581     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1582         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1583                               spapr->max_compat_pvr)) {
1584         /*
1585          * If using KVM with radix mode available, VCPUs can be started
1586          * without a HPT because KVM will start them in radix mode.
1587          * Set the GR bit in PATE so that we know there is no HPT.
1588          */
1589         spapr->patb_entry = PATE1_GR;
1590         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1591     } else {
1592         spapr_setup_hpt(spapr);
1593     }
1594 
1595     qemu_devices_reset();
1596 
1597     spapr_ovec_cleanup(spapr->ov5_cas);
1598     spapr->ov5_cas = spapr_ovec_new();
1599 
1600     ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1601 
1602     /*
1603      * This is fixing some of the default configuration of the XIVE
1604      * devices. To be called after the reset of the machine devices.
1605      */
1606     spapr_irq_reset(spapr, &error_fatal);
1607 
1608     /*
1609      * There is no CAS under qtest. Simulate one to please the code that
1610      * depends on spapr->ov5_cas. This is especially needed to test device
1611      * unplug, so we do that before resetting the DRCs.
1612      */
1613     if (qtest_enabled()) {
1614         spapr_ovec_cleanup(spapr->ov5_cas);
1615         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1616     }
1617 
1618     /* DRC reset may cause a device to be unplugged. This will cause troubles
1619      * if this device is used by another device (eg, a running vhost backend
1620      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1621      * situations, we reset DRCs after all devices have been reset.
1622      */
1623     spapr_drc_reset_all(spapr);
1624 
1625     spapr_clear_pending_events(spapr);
1626 
1627     /*
1628      * We place the device tree and RTAS just below either the top of the RMA,
1629      * or just below 2GB, whichever is lower, so that it can be
1630      * processed with 32-bit real mode code if necessary
1631      */
1632     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1633 
1634     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1635 
1636     rc = fdt_pack(fdt);
1637 
1638     /* Should only fail if we've built a corrupted tree */
1639     assert(rc == 0);
1640 
1641     /* Load the fdt */
1642     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1643     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1644     g_free(spapr->fdt_blob);
1645     spapr->fdt_size = fdt_totalsize(fdt);
1646     spapr->fdt_initial_size = spapr->fdt_size;
1647     spapr->fdt_blob = fdt;
1648 
1649     /* Set up the entry state */
1650     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, 0, fdt_addr, 0);
1651     first_ppc_cpu->env.gpr[5] = 0;
1652 
1653     spapr->fwnmi_system_reset_addr = -1;
1654     spapr->fwnmi_machine_check_addr = -1;
1655     spapr->fwnmi_machine_check_interlock = -1;
1656 
1657     /* Signal all vCPUs waiting on this condition */
1658     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1659 
1660     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1661 }
1662 
1663 static void spapr_create_nvram(SpaprMachineState *spapr)
1664 {
1665     DeviceState *dev = qdev_new("spapr-nvram");
1666     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1667 
1668     if (dinfo) {
1669         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1670                                 &error_fatal);
1671     }
1672 
1673     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1674 
1675     spapr->nvram = (struct SpaprNvram *)dev;
1676 }
1677 
1678 static void spapr_rtc_create(SpaprMachineState *spapr)
1679 {
1680     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1681                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1682                                        &error_fatal, NULL);
1683     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1684     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1685                               "date");
1686 }
1687 
1688 /* Returns whether we want to use VGA or not */
1689 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1690 {
1691     switch (vga_interface_type) {
1692     case VGA_NONE:
1693         return false;
1694     case VGA_DEVICE:
1695         return true;
1696     case VGA_STD:
1697     case VGA_VIRTIO:
1698     case VGA_CIRRUS:
1699         return pci_vga_init(pci_bus) != NULL;
1700     default:
1701         error_setg(errp,
1702                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1703         return false;
1704     }
1705 }
1706 
1707 static int spapr_pre_load(void *opaque)
1708 {
1709     int rc;
1710 
1711     rc = spapr_caps_pre_load(opaque);
1712     if (rc) {
1713         return rc;
1714     }
1715 
1716     return 0;
1717 }
1718 
1719 static int spapr_post_load(void *opaque, int version_id)
1720 {
1721     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1722     int err = 0;
1723 
1724     err = spapr_caps_post_migration(spapr);
1725     if (err) {
1726         return err;
1727     }
1728 
1729     /*
1730      * In earlier versions, there was no separate qdev for the PAPR
1731      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1732      * So when migrating from those versions, poke the incoming offset
1733      * value into the RTC device
1734      */
1735     if (version_id < 3) {
1736         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1737         if (err) {
1738             return err;
1739         }
1740     }
1741 
1742     if (kvm_enabled() && spapr->patb_entry) {
1743         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1744         bool radix = !!(spapr->patb_entry & PATE1_GR);
1745         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1746 
1747         /*
1748          * Update LPCR:HR and UPRT as they may not be set properly in
1749          * the stream
1750          */
1751         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1752                             LPCR_HR | LPCR_UPRT);
1753 
1754         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1755         if (err) {
1756             error_report("Process table config unsupported by the host");
1757             return -EINVAL;
1758         }
1759     }
1760 
1761     err = spapr_irq_post_load(spapr, version_id);
1762     if (err) {
1763         return err;
1764     }
1765 
1766     return err;
1767 }
1768 
1769 static int spapr_pre_save(void *opaque)
1770 {
1771     int rc;
1772 
1773     rc = spapr_caps_pre_save(opaque);
1774     if (rc) {
1775         return rc;
1776     }
1777 
1778     return 0;
1779 }
1780 
1781 static bool version_before_3(void *opaque, int version_id)
1782 {
1783     return version_id < 3;
1784 }
1785 
1786 static bool spapr_pending_events_needed(void *opaque)
1787 {
1788     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1789     return !QTAILQ_EMPTY(&spapr->pending_events);
1790 }
1791 
1792 static const VMStateDescription vmstate_spapr_event_entry = {
1793     .name = "spapr_event_log_entry",
1794     .version_id = 1,
1795     .minimum_version_id = 1,
1796     .fields = (VMStateField[]) {
1797         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1798         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1799         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1800                                      NULL, extended_length),
1801         VMSTATE_END_OF_LIST()
1802     },
1803 };
1804 
1805 static const VMStateDescription vmstate_spapr_pending_events = {
1806     .name = "spapr_pending_events",
1807     .version_id = 1,
1808     .minimum_version_id = 1,
1809     .needed = spapr_pending_events_needed,
1810     .fields = (VMStateField[]) {
1811         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1812                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1813         VMSTATE_END_OF_LIST()
1814     },
1815 };
1816 
1817 static bool spapr_ov5_cas_needed(void *opaque)
1818 {
1819     SpaprMachineState *spapr = opaque;
1820     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1821     bool cas_needed;
1822 
1823     /* Prior to the introduction of SpaprOptionVector, we had two option
1824      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1825      * Both of these options encode machine topology into the device-tree
1826      * in such a way that the now-booted OS should still be able to interact
1827      * appropriately with QEMU regardless of what options were actually
1828      * negotiatied on the source side.
1829      *
1830      * As such, we can avoid migrating the CAS-negotiated options if these
1831      * are the only options available on the current machine/platform.
1832      * Since these are the only options available for pseries-2.7 and
1833      * earlier, this allows us to maintain old->new/new->old migration
1834      * compatibility.
1835      *
1836      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1837      * via default pseries-2.8 machines and explicit command-line parameters.
1838      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1839      * of the actual CAS-negotiated values to continue working properly. For
1840      * example, availability of memory unplug depends on knowing whether
1841      * OV5_HP_EVT was negotiated via CAS.
1842      *
1843      * Thus, for any cases where the set of available CAS-negotiatable
1844      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1845      * include the CAS-negotiated options in the migration stream, unless
1846      * if they affect boot time behaviour only.
1847      */
1848     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1849     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1850     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1851 
1852     /* We need extra information if we have any bits outside the mask
1853      * defined above */
1854     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1855 
1856     spapr_ovec_cleanup(ov5_mask);
1857 
1858     return cas_needed;
1859 }
1860 
1861 static const VMStateDescription vmstate_spapr_ov5_cas = {
1862     .name = "spapr_option_vector_ov5_cas",
1863     .version_id = 1,
1864     .minimum_version_id = 1,
1865     .needed = spapr_ov5_cas_needed,
1866     .fields = (VMStateField[]) {
1867         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1868                                  vmstate_spapr_ovec, SpaprOptionVector),
1869         VMSTATE_END_OF_LIST()
1870     },
1871 };
1872 
1873 static bool spapr_patb_entry_needed(void *opaque)
1874 {
1875     SpaprMachineState *spapr = opaque;
1876 
1877     return !!spapr->patb_entry;
1878 }
1879 
1880 static const VMStateDescription vmstate_spapr_patb_entry = {
1881     .name = "spapr_patb_entry",
1882     .version_id = 1,
1883     .minimum_version_id = 1,
1884     .needed = spapr_patb_entry_needed,
1885     .fields = (VMStateField[]) {
1886         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1887         VMSTATE_END_OF_LIST()
1888     },
1889 };
1890 
1891 static bool spapr_irq_map_needed(void *opaque)
1892 {
1893     SpaprMachineState *spapr = opaque;
1894 
1895     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1896 }
1897 
1898 static const VMStateDescription vmstate_spapr_irq_map = {
1899     .name = "spapr_irq_map",
1900     .version_id = 1,
1901     .minimum_version_id = 1,
1902     .needed = spapr_irq_map_needed,
1903     .fields = (VMStateField[]) {
1904         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1905         VMSTATE_END_OF_LIST()
1906     },
1907 };
1908 
1909 static bool spapr_dtb_needed(void *opaque)
1910 {
1911     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1912 
1913     return smc->update_dt_enabled;
1914 }
1915 
1916 static int spapr_dtb_pre_load(void *opaque)
1917 {
1918     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1919 
1920     g_free(spapr->fdt_blob);
1921     spapr->fdt_blob = NULL;
1922     spapr->fdt_size = 0;
1923 
1924     return 0;
1925 }
1926 
1927 static const VMStateDescription vmstate_spapr_dtb = {
1928     .name = "spapr_dtb",
1929     .version_id = 1,
1930     .minimum_version_id = 1,
1931     .needed = spapr_dtb_needed,
1932     .pre_load = spapr_dtb_pre_load,
1933     .fields = (VMStateField[]) {
1934         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1935         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1936         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1937                                      fdt_size),
1938         VMSTATE_END_OF_LIST()
1939     },
1940 };
1941 
1942 static bool spapr_fwnmi_needed(void *opaque)
1943 {
1944     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1945 
1946     return spapr->fwnmi_machine_check_addr != -1;
1947 }
1948 
1949 static int spapr_fwnmi_pre_save(void *opaque)
1950 {
1951     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1952 
1953     /*
1954      * Check if machine check handling is in progress and print a
1955      * warning message.
1956      */
1957     if (spapr->fwnmi_machine_check_interlock != -1) {
1958         warn_report("A machine check is being handled during migration. The"
1959                 "handler may run and log hardware error on the destination");
1960     }
1961 
1962     return 0;
1963 }
1964 
1965 static const VMStateDescription vmstate_spapr_fwnmi = {
1966     .name = "spapr_fwnmi",
1967     .version_id = 1,
1968     .minimum_version_id = 1,
1969     .needed = spapr_fwnmi_needed,
1970     .pre_save = spapr_fwnmi_pre_save,
1971     .fields = (VMStateField[]) {
1972         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
1973         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
1974         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
1975         VMSTATE_END_OF_LIST()
1976     },
1977 };
1978 
1979 static const VMStateDescription vmstate_spapr = {
1980     .name = "spapr",
1981     .version_id = 3,
1982     .minimum_version_id = 1,
1983     .pre_load = spapr_pre_load,
1984     .post_load = spapr_post_load,
1985     .pre_save = spapr_pre_save,
1986     .fields = (VMStateField[]) {
1987         /* used to be @next_irq */
1988         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1989 
1990         /* RTC offset */
1991         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
1992 
1993         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
1994         VMSTATE_END_OF_LIST()
1995     },
1996     .subsections = (const VMStateDescription*[]) {
1997         &vmstate_spapr_ov5_cas,
1998         &vmstate_spapr_patb_entry,
1999         &vmstate_spapr_pending_events,
2000         &vmstate_spapr_cap_htm,
2001         &vmstate_spapr_cap_vsx,
2002         &vmstate_spapr_cap_dfp,
2003         &vmstate_spapr_cap_cfpc,
2004         &vmstate_spapr_cap_sbbc,
2005         &vmstate_spapr_cap_ibs,
2006         &vmstate_spapr_cap_hpt_maxpagesize,
2007         &vmstate_spapr_irq_map,
2008         &vmstate_spapr_cap_nested_kvm_hv,
2009         &vmstate_spapr_dtb,
2010         &vmstate_spapr_cap_large_decr,
2011         &vmstate_spapr_cap_ccf_assist,
2012         &vmstate_spapr_cap_fwnmi,
2013         &vmstate_spapr_fwnmi,
2014         NULL
2015     }
2016 };
2017 
2018 static int htab_save_setup(QEMUFile *f, void *opaque)
2019 {
2020     SpaprMachineState *spapr = opaque;
2021 
2022     /* "Iteration" header */
2023     if (!spapr->htab_shift) {
2024         qemu_put_be32(f, -1);
2025     } else {
2026         qemu_put_be32(f, spapr->htab_shift);
2027     }
2028 
2029     if (spapr->htab) {
2030         spapr->htab_save_index = 0;
2031         spapr->htab_first_pass = true;
2032     } else {
2033         if (spapr->htab_shift) {
2034             assert(kvm_enabled());
2035         }
2036     }
2037 
2038 
2039     return 0;
2040 }
2041 
2042 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2043                             int chunkstart, int n_valid, int n_invalid)
2044 {
2045     qemu_put_be32(f, chunkstart);
2046     qemu_put_be16(f, n_valid);
2047     qemu_put_be16(f, n_invalid);
2048     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2049                     HASH_PTE_SIZE_64 * n_valid);
2050 }
2051 
2052 static void htab_save_end_marker(QEMUFile *f)
2053 {
2054     qemu_put_be32(f, 0);
2055     qemu_put_be16(f, 0);
2056     qemu_put_be16(f, 0);
2057 }
2058 
2059 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2060                                  int64_t max_ns)
2061 {
2062     bool has_timeout = max_ns != -1;
2063     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2064     int index = spapr->htab_save_index;
2065     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2066 
2067     assert(spapr->htab_first_pass);
2068 
2069     do {
2070         int chunkstart;
2071 
2072         /* Consume invalid HPTEs */
2073         while ((index < htabslots)
2074                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2075             CLEAN_HPTE(HPTE(spapr->htab, index));
2076             index++;
2077         }
2078 
2079         /* Consume valid HPTEs */
2080         chunkstart = index;
2081         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2082                && HPTE_VALID(HPTE(spapr->htab, index))) {
2083             CLEAN_HPTE(HPTE(spapr->htab, index));
2084             index++;
2085         }
2086 
2087         if (index > chunkstart) {
2088             int n_valid = index - chunkstart;
2089 
2090             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2091 
2092             if (has_timeout &&
2093                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2094                 break;
2095             }
2096         }
2097     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2098 
2099     if (index >= htabslots) {
2100         assert(index == htabslots);
2101         index = 0;
2102         spapr->htab_first_pass = false;
2103     }
2104     spapr->htab_save_index = index;
2105 }
2106 
2107 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2108                                 int64_t max_ns)
2109 {
2110     bool final = max_ns < 0;
2111     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2112     int examined = 0, sent = 0;
2113     int index = spapr->htab_save_index;
2114     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2115 
2116     assert(!spapr->htab_first_pass);
2117 
2118     do {
2119         int chunkstart, invalidstart;
2120 
2121         /* Consume non-dirty HPTEs */
2122         while ((index < htabslots)
2123                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2124             index++;
2125             examined++;
2126         }
2127 
2128         chunkstart = index;
2129         /* Consume valid dirty HPTEs */
2130         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2131                && HPTE_DIRTY(HPTE(spapr->htab, index))
2132                && HPTE_VALID(HPTE(spapr->htab, index))) {
2133             CLEAN_HPTE(HPTE(spapr->htab, index));
2134             index++;
2135             examined++;
2136         }
2137 
2138         invalidstart = index;
2139         /* Consume invalid dirty HPTEs */
2140         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2141                && HPTE_DIRTY(HPTE(spapr->htab, index))
2142                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2143             CLEAN_HPTE(HPTE(spapr->htab, index));
2144             index++;
2145             examined++;
2146         }
2147 
2148         if (index > chunkstart) {
2149             int n_valid = invalidstart - chunkstart;
2150             int n_invalid = index - invalidstart;
2151 
2152             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2153             sent += index - chunkstart;
2154 
2155             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2156                 break;
2157             }
2158         }
2159 
2160         if (examined >= htabslots) {
2161             break;
2162         }
2163 
2164         if (index >= htabslots) {
2165             assert(index == htabslots);
2166             index = 0;
2167         }
2168     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2169 
2170     if (index >= htabslots) {
2171         assert(index == htabslots);
2172         index = 0;
2173     }
2174 
2175     spapr->htab_save_index = index;
2176 
2177     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2178 }
2179 
2180 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2181 #define MAX_KVM_BUF_SIZE    2048
2182 
2183 static int htab_save_iterate(QEMUFile *f, void *opaque)
2184 {
2185     SpaprMachineState *spapr = opaque;
2186     int fd;
2187     int rc = 0;
2188 
2189     /* Iteration header */
2190     if (!spapr->htab_shift) {
2191         qemu_put_be32(f, -1);
2192         return 1;
2193     } else {
2194         qemu_put_be32(f, 0);
2195     }
2196 
2197     if (!spapr->htab) {
2198         assert(kvm_enabled());
2199 
2200         fd = get_htab_fd(spapr);
2201         if (fd < 0) {
2202             return fd;
2203         }
2204 
2205         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2206         if (rc < 0) {
2207             return rc;
2208         }
2209     } else  if (spapr->htab_first_pass) {
2210         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2211     } else {
2212         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2213     }
2214 
2215     htab_save_end_marker(f);
2216 
2217     return rc;
2218 }
2219 
2220 static int htab_save_complete(QEMUFile *f, void *opaque)
2221 {
2222     SpaprMachineState *spapr = opaque;
2223     int fd;
2224 
2225     /* Iteration header */
2226     if (!spapr->htab_shift) {
2227         qemu_put_be32(f, -1);
2228         return 0;
2229     } else {
2230         qemu_put_be32(f, 0);
2231     }
2232 
2233     if (!spapr->htab) {
2234         int rc;
2235 
2236         assert(kvm_enabled());
2237 
2238         fd = get_htab_fd(spapr);
2239         if (fd < 0) {
2240             return fd;
2241         }
2242 
2243         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2244         if (rc < 0) {
2245             return rc;
2246         }
2247     } else {
2248         if (spapr->htab_first_pass) {
2249             htab_save_first_pass(f, spapr, -1);
2250         }
2251         htab_save_later_pass(f, spapr, -1);
2252     }
2253 
2254     /* End marker */
2255     htab_save_end_marker(f);
2256 
2257     return 0;
2258 }
2259 
2260 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2261 {
2262     SpaprMachineState *spapr = opaque;
2263     uint32_t section_hdr;
2264     int fd = -1;
2265     Error *local_err = NULL;
2266 
2267     if (version_id < 1 || version_id > 1) {
2268         error_report("htab_load() bad version");
2269         return -EINVAL;
2270     }
2271 
2272     section_hdr = qemu_get_be32(f);
2273 
2274     if (section_hdr == -1) {
2275         spapr_free_hpt(spapr);
2276         return 0;
2277     }
2278 
2279     if (section_hdr) {
2280         int ret;
2281 
2282         /* First section gives the htab size */
2283         ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2284         if (ret < 0) {
2285             error_report_err(local_err);
2286             return ret;
2287         }
2288         return 0;
2289     }
2290 
2291     if (!spapr->htab) {
2292         assert(kvm_enabled());
2293 
2294         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2295         if (fd < 0) {
2296             error_report_err(local_err);
2297             return fd;
2298         }
2299     }
2300 
2301     while (true) {
2302         uint32_t index;
2303         uint16_t n_valid, n_invalid;
2304 
2305         index = qemu_get_be32(f);
2306         n_valid = qemu_get_be16(f);
2307         n_invalid = qemu_get_be16(f);
2308 
2309         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2310             /* End of Stream */
2311             break;
2312         }
2313 
2314         if ((index + n_valid + n_invalid) >
2315             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2316             /* Bad index in stream */
2317             error_report(
2318                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2319                 index, n_valid, n_invalid, spapr->htab_shift);
2320             return -EINVAL;
2321         }
2322 
2323         if (spapr->htab) {
2324             if (n_valid) {
2325                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2326                                 HASH_PTE_SIZE_64 * n_valid);
2327             }
2328             if (n_invalid) {
2329                 memset(HPTE(spapr->htab, index + n_valid), 0,
2330                        HASH_PTE_SIZE_64 * n_invalid);
2331             }
2332         } else {
2333             int rc;
2334 
2335             assert(fd >= 0);
2336 
2337             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2338                                         &local_err);
2339             if (rc < 0) {
2340                 error_report_err(local_err);
2341                 return rc;
2342             }
2343         }
2344     }
2345 
2346     if (!spapr->htab) {
2347         assert(fd >= 0);
2348         close(fd);
2349     }
2350 
2351     return 0;
2352 }
2353 
2354 static void htab_save_cleanup(void *opaque)
2355 {
2356     SpaprMachineState *spapr = opaque;
2357 
2358     close_htab_fd(spapr);
2359 }
2360 
2361 static SaveVMHandlers savevm_htab_handlers = {
2362     .save_setup = htab_save_setup,
2363     .save_live_iterate = htab_save_iterate,
2364     .save_live_complete_precopy = htab_save_complete,
2365     .save_cleanup = htab_save_cleanup,
2366     .load_state = htab_load,
2367 };
2368 
2369 static void spapr_boot_set(void *opaque, const char *boot_device,
2370                            Error **errp)
2371 {
2372     MachineState *machine = MACHINE(opaque);
2373     machine->boot_order = g_strdup(boot_device);
2374 }
2375 
2376 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2377 {
2378     MachineState *machine = MACHINE(spapr);
2379     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2380     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2381     int i;
2382 
2383     for (i = 0; i < nr_lmbs; i++) {
2384         uint64_t addr;
2385 
2386         addr = i * lmb_size + machine->device_memory->base;
2387         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2388                                addr / lmb_size);
2389     }
2390 }
2391 
2392 /*
2393  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2394  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2395  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2396  */
2397 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2398 {
2399     int i;
2400 
2401     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2402         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2403                    " is not aligned to %" PRIu64 " MiB",
2404                    machine->ram_size,
2405                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2406         return;
2407     }
2408 
2409     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2410         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2411                    " is not aligned to %" PRIu64 " MiB",
2412                    machine->ram_size,
2413                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2414         return;
2415     }
2416 
2417     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2418         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2419             error_setg(errp,
2420                        "Node %d memory size 0x%" PRIx64
2421                        " is not aligned to %" PRIu64 " MiB",
2422                        i, machine->numa_state->nodes[i].node_mem,
2423                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2424             return;
2425         }
2426     }
2427 }
2428 
2429 /* find cpu slot in machine->possible_cpus by core_id */
2430 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2431 {
2432     int index = id / ms->smp.threads;
2433 
2434     if (index >= ms->possible_cpus->len) {
2435         return NULL;
2436     }
2437     if (idx) {
2438         *idx = index;
2439     }
2440     return &ms->possible_cpus->cpus[index];
2441 }
2442 
2443 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2444 {
2445     MachineState *ms = MACHINE(spapr);
2446     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2447     Error *local_err = NULL;
2448     bool vsmt_user = !!spapr->vsmt;
2449     int kvm_smt = kvmppc_smt_threads();
2450     int ret;
2451     unsigned int smp_threads = ms->smp.threads;
2452 
2453     if (!kvm_enabled() && (smp_threads > 1)) {
2454         error_setg(errp, "TCG cannot support more than 1 thread/core "
2455                    "on a pseries machine");
2456         return;
2457     }
2458     if (!is_power_of_2(smp_threads)) {
2459         error_setg(errp, "Cannot support %d threads/core on a pseries "
2460                    "machine because it must be a power of 2", smp_threads);
2461         return;
2462     }
2463 
2464     /* Detemine the VSMT mode to use: */
2465     if (vsmt_user) {
2466         if (spapr->vsmt < smp_threads) {
2467             error_setg(errp, "Cannot support VSMT mode %d"
2468                        " because it must be >= threads/core (%d)",
2469                        spapr->vsmt, smp_threads);
2470             return;
2471         }
2472         /* In this case, spapr->vsmt has been set by the command line */
2473     } else if (!smc->smp_threads_vsmt) {
2474         /*
2475          * Default VSMT value is tricky, because we need it to be as
2476          * consistent as possible (for migration), but this requires
2477          * changing it for at least some existing cases.  We pick 8 as
2478          * the value that we'd get with KVM on POWER8, the
2479          * overwhelmingly common case in production systems.
2480          */
2481         spapr->vsmt = MAX(8, smp_threads);
2482     } else {
2483         spapr->vsmt = smp_threads;
2484     }
2485 
2486     /* KVM: If necessary, set the SMT mode: */
2487     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2488         ret = kvmppc_set_smt_threads(spapr->vsmt);
2489         if (ret) {
2490             /* Looks like KVM isn't able to change VSMT mode */
2491             error_setg(&local_err,
2492                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2493                        spapr->vsmt, ret);
2494             /* We can live with that if the default one is big enough
2495              * for the number of threads, and a submultiple of the one
2496              * we want.  In this case we'll waste some vcpu ids, but
2497              * behaviour will be correct */
2498             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2499                 warn_report_err(local_err);
2500             } else {
2501                 if (!vsmt_user) {
2502                     error_append_hint(&local_err,
2503                                       "On PPC, a VM with %d threads/core"
2504                                       " on a host with %d threads/core"
2505                                       " requires the use of VSMT mode %d.\n",
2506                                       smp_threads, kvm_smt, spapr->vsmt);
2507                 }
2508                 kvmppc_error_append_smt_possible_hint(&local_err);
2509                 error_propagate(errp, local_err);
2510             }
2511         }
2512     }
2513     /* else TCG: nothing to do currently */
2514 }
2515 
2516 static void spapr_init_cpus(SpaprMachineState *spapr)
2517 {
2518     MachineState *machine = MACHINE(spapr);
2519     MachineClass *mc = MACHINE_GET_CLASS(machine);
2520     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2521     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2522     const CPUArchIdList *possible_cpus;
2523     unsigned int smp_cpus = machine->smp.cpus;
2524     unsigned int smp_threads = machine->smp.threads;
2525     unsigned int max_cpus = machine->smp.max_cpus;
2526     int boot_cores_nr = smp_cpus / smp_threads;
2527     int i;
2528 
2529     possible_cpus = mc->possible_cpu_arch_ids(machine);
2530     if (mc->has_hotpluggable_cpus) {
2531         if (smp_cpus % smp_threads) {
2532             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2533                          smp_cpus, smp_threads);
2534             exit(1);
2535         }
2536         if (max_cpus % smp_threads) {
2537             error_report("max_cpus (%u) must be multiple of threads (%u)",
2538                          max_cpus, smp_threads);
2539             exit(1);
2540         }
2541     } else {
2542         if (max_cpus != smp_cpus) {
2543             error_report("This machine version does not support CPU hotplug");
2544             exit(1);
2545         }
2546         boot_cores_nr = possible_cpus->len;
2547     }
2548 
2549     if (smc->pre_2_10_has_unused_icps) {
2550         int i;
2551 
2552         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2553             /* Dummy entries get deregistered when real ICPState objects
2554              * are registered during CPU core hotplug.
2555              */
2556             pre_2_10_vmstate_register_dummy_icp(i);
2557         }
2558     }
2559 
2560     for (i = 0; i < possible_cpus->len; i++) {
2561         int core_id = i * smp_threads;
2562 
2563         if (mc->has_hotpluggable_cpus) {
2564             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2565                                    spapr_vcpu_id(spapr, core_id));
2566         }
2567 
2568         if (i < boot_cores_nr) {
2569             Object *core  = object_new(type);
2570             int nr_threads = smp_threads;
2571 
2572             /* Handle the partially filled core for older machine types */
2573             if ((i + 1) * smp_threads >= smp_cpus) {
2574                 nr_threads = smp_cpus - i * smp_threads;
2575             }
2576 
2577             object_property_set_int(core, "nr-threads", nr_threads,
2578                                     &error_fatal);
2579             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2580                                     &error_fatal);
2581             qdev_realize(DEVICE(core), NULL, &error_fatal);
2582 
2583             object_unref(core);
2584         }
2585     }
2586 }
2587 
2588 static PCIHostState *spapr_create_default_phb(void)
2589 {
2590     DeviceState *dev;
2591 
2592     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2593     qdev_prop_set_uint32(dev, "index", 0);
2594     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2595 
2596     return PCI_HOST_BRIDGE(dev);
2597 }
2598 
2599 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2600 {
2601     MachineState *machine = MACHINE(spapr);
2602     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2603     hwaddr rma_size = machine->ram_size;
2604     hwaddr node0_size = spapr_node0_size(machine);
2605 
2606     /* RMA has to fit in the first NUMA node */
2607     rma_size = MIN(rma_size, node0_size);
2608 
2609     /*
2610      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2611      * never exceed that
2612      */
2613     rma_size = MIN(rma_size, 1 * TiB);
2614 
2615     /*
2616      * Clamp the RMA size based on machine type.  This is for
2617      * migration compatibility with older qemu versions, which limited
2618      * the RMA size for complicated and mostly bad reasons.
2619      */
2620     if (smc->rma_limit) {
2621         rma_size = MIN(rma_size, smc->rma_limit);
2622     }
2623 
2624     if (rma_size < MIN_RMA_SLOF) {
2625         error_setg(errp,
2626                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2627                    "ldMiB guest RMA (Real Mode Area memory)",
2628                    MIN_RMA_SLOF / MiB);
2629         return 0;
2630     }
2631 
2632     return rma_size;
2633 }
2634 
2635 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2636 {
2637     MachineState *machine = MACHINE(spapr);
2638     int i;
2639 
2640     for (i = 0; i < machine->ram_slots; i++) {
2641         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2642     }
2643 }
2644 
2645 /* pSeries LPAR / sPAPR hardware init */
2646 static void spapr_machine_init(MachineState *machine)
2647 {
2648     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2649     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2650     MachineClass *mc = MACHINE_GET_CLASS(machine);
2651     const char *bios_name = machine->firmware ?: FW_FILE_NAME;
2652     const char *kernel_filename = machine->kernel_filename;
2653     const char *initrd_filename = machine->initrd_filename;
2654     PCIHostState *phb;
2655     int i;
2656     MemoryRegion *sysmem = get_system_memory();
2657     long load_limit, fw_size;
2658     char *filename;
2659     Error *resize_hpt_err = NULL;
2660 
2661     /*
2662      * if Secure VM (PEF) support is configured, then initialize it
2663      */
2664     pef_kvm_init(machine->cgs, &error_fatal);
2665 
2666     msi_nonbroken = true;
2667 
2668     QLIST_INIT(&spapr->phbs);
2669     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2670 
2671     /* Determine capabilities to run with */
2672     spapr_caps_init(spapr);
2673 
2674     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2675     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2676         /*
2677          * If the user explicitly requested a mode we should either
2678          * supply it, or fail completely (which we do below).  But if
2679          * it's not set explicitly, we reset our mode to something
2680          * that works
2681          */
2682         if (resize_hpt_err) {
2683             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2684             error_free(resize_hpt_err);
2685             resize_hpt_err = NULL;
2686         } else {
2687             spapr->resize_hpt = smc->resize_hpt_default;
2688         }
2689     }
2690 
2691     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2692 
2693     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2694         /*
2695          * User requested HPT resize, but this host can't supply it.  Bail out
2696          */
2697         error_report_err(resize_hpt_err);
2698         exit(1);
2699     }
2700     error_free(resize_hpt_err);
2701 
2702     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2703 
2704     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2705     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2706 
2707     /*
2708      * VSMT must be set in order to be able to compute VCPU ids, ie to
2709      * call spapr_max_server_number() or spapr_vcpu_id().
2710      */
2711     spapr_set_vsmt_mode(spapr, &error_fatal);
2712 
2713     /* Set up Interrupt Controller before we create the VCPUs */
2714     spapr_irq_init(spapr, &error_fatal);
2715 
2716     /* Set up containers for ibm,client-architecture-support negotiated options
2717      */
2718     spapr->ov5 = spapr_ovec_new();
2719     spapr->ov5_cas = spapr_ovec_new();
2720 
2721     if (smc->dr_lmb_enabled) {
2722         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2723         spapr_validate_node_memory(machine, &error_fatal);
2724     }
2725 
2726     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2727 
2728     /* advertise support for dedicated HP event source to guests */
2729     if (spapr->use_hotplug_event_source) {
2730         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2731     }
2732 
2733     /* advertise support for HPT resizing */
2734     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2735         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2736     }
2737 
2738     /* advertise support for ibm,dyamic-memory-v2 */
2739     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2740 
2741     /* advertise XIVE on POWER9 machines */
2742     if (spapr->irq->xive) {
2743         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2744     }
2745 
2746     /* init CPUs */
2747     spapr_init_cpus(spapr);
2748 
2749     /*
2750      * check we don't have a memory-less/cpu-less NUMA node
2751      * Firmware relies on the existing memory/cpu topology to provide the
2752      * NUMA topology to the kernel.
2753      * And the linux kernel needs to know the NUMA topology at start
2754      * to be able to hotplug CPUs later.
2755      */
2756     if (machine->numa_state->num_nodes) {
2757         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2758             /* check for memory-less node */
2759             if (machine->numa_state->nodes[i].node_mem == 0) {
2760                 CPUState *cs;
2761                 int found = 0;
2762                 /* check for cpu-less node */
2763                 CPU_FOREACH(cs) {
2764                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2765                     if (cpu->node_id == i) {
2766                         found = 1;
2767                         break;
2768                     }
2769                 }
2770                 /* memory-less and cpu-less node */
2771                 if (!found) {
2772                     error_report(
2773                        "Memory-less/cpu-less nodes are not supported (node %d)",
2774                                  i);
2775                     exit(1);
2776                 }
2777             }
2778         }
2779 
2780     }
2781 
2782     /*
2783      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2784      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2785      * called from vPHB reset handler so we initialize the counter here.
2786      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2787      * must be equally distant from any other node.
2788      * The final value of spapr->gpu_numa_id is going to be written to
2789      * max-associativity-domains in spapr_build_fdt().
2790      */
2791     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2792 
2793     /* Init numa_assoc_array */
2794     spapr_numa_associativity_init(spapr, machine);
2795 
2796     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2797         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2798                               spapr->max_compat_pvr)) {
2799         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2800         /* KVM and TCG always allow GTSE with radix... */
2801         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2802     }
2803     /* ... but not with hash (currently). */
2804 
2805     if (kvm_enabled()) {
2806         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2807         kvmppc_enable_logical_ci_hcalls();
2808         kvmppc_enable_set_mode_hcall();
2809 
2810         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2811         kvmppc_enable_clear_ref_mod_hcalls();
2812 
2813         /* Enable H_PAGE_INIT */
2814         kvmppc_enable_h_page_init();
2815     }
2816 
2817     /* map RAM */
2818     memory_region_add_subregion(sysmem, 0, machine->ram);
2819 
2820     /* always allocate the device memory information */
2821     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2822 
2823     /* initialize hotplug memory address space */
2824     if (machine->ram_size < machine->maxram_size) {
2825         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2826         /*
2827          * Limit the number of hotpluggable memory slots to half the number
2828          * slots that KVM supports, leaving the other half for PCI and other
2829          * devices. However ensure that number of slots doesn't drop below 32.
2830          */
2831         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2832                            SPAPR_MAX_RAM_SLOTS;
2833 
2834         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2835             max_memslots = SPAPR_MAX_RAM_SLOTS;
2836         }
2837         if (machine->ram_slots > max_memslots) {
2838             error_report("Specified number of memory slots %"
2839                          PRIu64" exceeds max supported %d",
2840                          machine->ram_slots, max_memslots);
2841             exit(1);
2842         }
2843 
2844         machine->device_memory->base = ROUND_UP(machine->ram_size,
2845                                                 SPAPR_DEVICE_MEM_ALIGN);
2846         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2847                            "device-memory", device_mem_size);
2848         memory_region_add_subregion(sysmem, machine->device_memory->base,
2849                                     &machine->device_memory->mr);
2850     }
2851 
2852     if (smc->dr_lmb_enabled) {
2853         spapr_create_lmb_dr_connectors(spapr);
2854     }
2855 
2856     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2857         /* Create the error string for live migration blocker */
2858         error_setg(&spapr->fwnmi_migration_blocker,
2859             "A machine check is being handled during migration. The handler"
2860             "may run and log hardware error on the destination");
2861     }
2862 
2863     if (mc->nvdimm_supported) {
2864         spapr_create_nvdimm_dr_connectors(spapr);
2865     }
2866 
2867     /* Set up RTAS event infrastructure */
2868     spapr_events_init(spapr);
2869 
2870     /* Set up the RTC RTAS interfaces */
2871     spapr_rtc_create(spapr);
2872 
2873     /* Set up VIO bus */
2874     spapr->vio_bus = spapr_vio_bus_init();
2875 
2876     for (i = 0; serial_hd(i); i++) {
2877         spapr_vty_create(spapr->vio_bus, serial_hd(i));
2878     }
2879 
2880     /* We always have at least the nvram device on VIO */
2881     spapr_create_nvram(spapr);
2882 
2883     /*
2884      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2885      * connectors (described in root DT node's "ibm,drc-types" property)
2886      * are pre-initialized here. additional child connectors (such as
2887      * connectors for a PHBs PCI slots) are added as needed during their
2888      * parent's realization.
2889      */
2890     if (smc->dr_phb_enabled) {
2891         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2892             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2893         }
2894     }
2895 
2896     /* Set up PCI */
2897     spapr_pci_rtas_init();
2898 
2899     phb = spapr_create_default_phb();
2900 
2901     for (i = 0; i < nb_nics; i++) {
2902         NICInfo *nd = &nd_table[i];
2903 
2904         if (!nd->model) {
2905             nd->model = g_strdup("spapr-vlan");
2906         }
2907 
2908         if (g_str_equal(nd->model, "spapr-vlan") ||
2909             g_str_equal(nd->model, "ibmveth")) {
2910             spapr_vlan_create(spapr->vio_bus, nd);
2911         } else {
2912             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2913         }
2914     }
2915 
2916     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2917         spapr_vscsi_create(spapr->vio_bus);
2918     }
2919 
2920     /* Graphics */
2921     if (spapr_vga_init(phb->bus, &error_fatal)) {
2922         spapr->has_graphics = true;
2923         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2924     }
2925 
2926     if (machine->usb) {
2927         if (smc->use_ohci_by_default) {
2928             pci_create_simple(phb->bus, -1, "pci-ohci");
2929         } else {
2930             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2931         }
2932 
2933         if (spapr->has_graphics) {
2934             USBBus *usb_bus = usb_bus_find(-1);
2935 
2936             usb_create_simple(usb_bus, "usb-kbd");
2937             usb_create_simple(usb_bus, "usb-mouse");
2938         }
2939     }
2940 
2941     if (kernel_filename) {
2942         spapr->kernel_size = load_elf(kernel_filename, NULL,
2943                                       translate_kernel_address, spapr,
2944                                       NULL, NULL, NULL, NULL, 1,
2945                                       PPC_ELF_MACHINE, 0, 0);
2946         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2947             spapr->kernel_size = load_elf(kernel_filename, NULL,
2948                                           translate_kernel_address, spapr,
2949                                           NULL, NULL, NULL, NULL, 0,
2950                                           PPC_ELF_MACHINE, 0, 0);
2951             spapr->kernel_le = spapr->kernel_size > 0;
2952         }
2953         if (spapr->kernel_size < 0) {
2954             error_report("error loading %s: %s", kernel_filename,
2955                          load_elf_strerror(spapr->kernel_size));
2956             exit(1);
2957         }
2958 
2959         /* load initrd */
2960         if (initrd_filename) {
2961             /* Try to locate the initrd in the gap between the kernel
2962              * and the firmware. Add a bit of space just in case
2963              */
2964             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
2965                                   + 0x1ffff) & ~0xffff;
2966             spapr->initrd_size = load_image_targphys(initrd_filename,
2967                                                      spapr->initrd_base,
2968                                                      load_limit
2969                                                      - spapr->initrd_base);
2970             if (spapr->initrd_size < 0) {
2971                 error_report("could not load initial ram disk '%s'",
2972                              initrd_filename);
2973                 exit(1);
2974             }
2975         }
2976     }
2977 
2978     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2979     if (!filename) {
2980         error_report("Could not find LPAR firmware '%s'", bios_name);
2981         exit(1);
2982     }
2983     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2984     if (fw_size <= 0) {
2985         error_report("Could not load LPAR firmware '%s'", filename);
2986         exit(1);
2987     }
2988     g_free(filename);
2989 
2990     /* FIXME: Should register things through the MachineState's qdev
2991      * interface, this is a legacy from the sPAPREnvironment structure
2992      * which predated MachineState but had a similar function */
2993     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2994     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
2995                          &savevm_htab_handlers, spapr);
2996 
2997     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
2998 
2999     qemu_register_boot_set(spapr_boot_set, spapr);
3000 
3001     /*
3002      * Nothing needs to be done to resume a suspended guest because
3003      * suspending does not change the machine state, so no need for
3004      * a ->wakeup method.
3005      */
3006     qemu_register_wakeup_support();
3007 
3008     if (kvm_enabled()) {
3009         /* to stop and start vmclock */
3010         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3011                                          &spapr->tb);
3012 
3013         kvmppc_spapr_enable_inkernel_multitce();
3014     }
3015 
3016     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3017 }
3018 
3019 #define DEFAULT_KVM_TYPE "auto"
3020 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3021 {
3022     /*
3023      * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to
3024      * accomodate the 'HV' and 'PV' formats that exists in the
3025      * wild. The 'auto' mode is being introduced already as
3026      * lower-case, thus we don't need to bother checking for
3027      * "AUTO".
3028      */
3029     if (!vm_type || !strcmp(vm_type, DEFAULT_KVM_TYPE)) {
3030         return 0;
3031     }
3032 
3033     if (!g_ascii_strcasecmp(vm_type, "hv")) {
3034         return 1;
3035     }
3036 
3037     if (!g_ascii_strcasecmp(vm_type, "pr")) {
3038         return 2;
3039     }
3040 
3041     error_report("Unknown kvm-type specified '%s'", vm_type);
3042     exit(1);
3043 }
3044 
3045 /*
3046  * Implementation of an interface to adjust firmware path
3047  * for the bootindex property handling.
3048  */
3049 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3050                                    DeviceState *dev)
3051 {
3052 #define CAST(type, obj, name) \
3053     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3054     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3055     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3056     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3057 
3058     if (d) {
3059         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3060         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3061         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3062 
3063         if (spapr) {
3064             /*
3065              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3066              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3067              * 0x8000 | (target << 8) | (bus << 5) | lun
3068              * (see the "Logical unit addressing format" table in SAM5)
3069              */
3070             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3071             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3072                                    (uint64_t)id << 48);
3073         } else if (virtio) {
3074             /*
3075              * We use SRP luns of the form 01000000 | (target << 8) | lun
3076              * in the top 32 bits of the 64-bit LUN
3077              * Note: the quote above is from SLOF and it is wrong,
3078              * the actual binding is:
3079              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3080              */
3081             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3082             if (d->lun >= 256) {
3083                 /* Use the LUN "flat space addressing method" */
3084                 id |= 0x4000;
3085             }
3086             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3087                                    (uint64_t)id << 32);
3088         } else if (usb) {
3089             /*
3090              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3091              * in the top 32 bits of the 64-bit LUN
3092              */
3093             unsigned usb_port = atoi(usb->port->path);
3094             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3095             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3096                                    (uint64_t)id << 32);
3097         }
3098     }
3099 
3100     /*
3101      * SLOF probes the USB devices, and if it recognizes that the device is a
3102      * storage device, it changes its name to "storage" instead of "usb-host",
3103      * and additionally adds a child node for the SCSI LUN, so the correct
3104      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3105      */
3106     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3107         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3108         if (usb_host_dev_is_scsi_storage(usbdev)) {
3109             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3110         }
3111     }
3112 
3113     if (phb) {
3114         /* Replace "pci" with "pci@800000020000000" */
3115         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3116     }
3117 
3118     if (vsc) {
3119         /* Same logic as virtio above */
3120         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3121         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3122     }
3123 
3124     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3125         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3126         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3127         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3128     }
3129 
3130     return NULL;
3131 }
3132 
3133 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3134 {
3135     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3136 
3137     return g_strdup(spapr->kvm_type);
3138 }
3139 
3140 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3141 {
3142     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3143 
3144     g_free(spapr->kvm_type);
3145     spapr->kvm_type = g_strdup(value);
3146 }
3147 
3148 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3149 {
3150     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3151 
3152     return spapr->use_hotplug_event_source;
3153 }
3154 
3155 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3156                                             Error **errp)
3157 {
3158     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3159 
3160     spapr->use_hotplug_event_source = value;
3161 }
3162 
3163 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3164 {
3165     return true;
3166 }
3167 
3168 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3169 {
3170     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3171 
3172     switch (spapr->resize_hpt) {
3173     case SPAPR_RESIZE_HPT_DEFAULT:
3174         return g_strdup("default");
3175     case SPAPR_RESIZE_HPT_DISABLED:
3176         return g_strdup("disabled");
3177     case SPAPR_RESIZE_HPT_ENABLED:
3178         return g_strdup("enabled");
3179     case SPAPR_RESIZE_HPT_REQUIRED:
3180         return g_strdup("required");
3181     }
3182     g_assert_not_reached();
3183 }
3184 
3185 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3186 {
3187     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3188 
3189     if (strcmp(value, "default") == 0) {
3190         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3191     } else if (strcmp(value, "disabled") == 0) {
3192         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3193     } else if (strcmp(value, "enabled") == 0) {
3194         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3195     } else if (strcmp(value, "required") == 0) {
3196         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3197     } else {
3198         error_setg(errp, "Bad value for \"resize-hpt\" property");
3199     }
3200 }
3201 
3202 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3203 {
3204     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3205 
3206     if (spapr->irq == &spapr_irq_xics_legacy) {
3207         return g_strdup("legacy");
3208     } else if (spapr->irq == &spapr_irq_xics) {
3209         return g_strdup("xics");
3210     } else if (spapr->irq == &spapr_irq_xive) {
3211         return g_strdup("xive");
3212     } else if (spapr->irq == &spapr_irq_dual) {
3213         return g_strdup("dual");
3214     }
3215     g_assert_not_reached();
3216 }
3217 
3218 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3219 {
3220     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3221 
3222     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3223         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3224         return;
3225     }
3226 
3227     /* The legacy IRQ backend can not be set */
3228     if (strcmp(value, "xics") == 0) {
3229         spapr->irq = &spapr_irq_xics;
3230     } else if (strcmp(value, "xive") == 0) {
3231         spapr->irq = &spapr_irq_xive;
3232     } else if (strcmp(value, "dual") == 0) {
3233         spapr->irq = &spapr_irq_dual;
3234     } else {
3235         error_setg(errp, "Bad value for \"ic-mode\" property");
3236     }
3237 }
3238 
3239 static char *spapr_get_host_model(Object *obj, Error **errp)
3240 {
3241     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3242 
3243     return g_strdup(spapr->host_model);
3244 }
3245 
3246 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3247 {
3248     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3249 
3250     g_free(spapr->host_model);
3251     spapr->host_model = g_strdup(value);
3252 }
3253 
3254 static char *spapr_get_host_serial(Object *obj, Error **errp)
3255 {
3256     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3257 
3258     return g_strdup(spapr->host_serial);
3259 }
3260 
3261 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3262 {
3263     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3264 
3265     g_free(spapr->host_serial);
3266     spapr->host_serial = g_strdup(value);
3267 }
3268 
3269 static void spapr_instance_init(Object *obj)
3270 {
3271     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3272     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3273     MachineState *ms = MACHINE(spapr);
3274     MachineClass *mc = MACHINE_GET_CLASS(ms);
3275 
3276     /*
3277      * NVDIMM support went live in 5.1 without considering that, in
3278      * other archs, the user needs to enable NVDIMM support with the
3279      * 'nvdimm' machine option and the default behavior is NVDIMM
3280      * support disabled. It is too late to roll back to the standard
3281      * behavior without breaking 5.1 guests.
3282      */
3283     if (mc->nvdimm_supported) {
3284         ms->nvdimms_state->is_enabled = true;
3285     }
3286 
3287     spapr->htab_fd = -1;
3288     spapr->use_hotplug_event_source = true;
3289     spapr->kvm_type = g_strdup(DEFAULT_KVM_TYPE);
3290     object_property_add_str(obj, "kvm-type",
3291                             spapr_get_kvm_type, spapr_set_kvm_type);
3292     object_property_set_description(obj, "kvm-type",
3293                                     "Specifies the KVM virtualization mode (auto,"
3294                                     " hv, pr). Defaults to 'auto'. This mode will use"
3295                                     " any available KVM module loaded in the host,"
3296                                     " where kvm_hv takes precedence if both kvm_hv and"
3297                                     " kvm_pr are loaded.");
3298     object_property_add_bool(obj, "modern-hotplug-events",
3299                             spapr_get_modern_hotplug_events,
3300                             spapr_set_modern_hotplug_events);
3301     object_property_set_description(obj, "modern-hotplug-events",
3302                                     "Use dedicated hotplug event mechanism in"
3303                                     " place of standard EPOW events when possible"
3304                                     " (required for memory hot-unplug support)");
3305     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3306                             "Maximum permitted CPU compatibility mode");
3307 
3308     object_property_add_str(obj, "resize-hpt",
3309                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3310     object_property_set_description(obj, "resize-hpt",
3311                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3312     object_property_add_uint32_ptr(obj, "vsmt",
3313                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3314     object_property_set_description(obj, "vsmt",
3315                                     "Virtual SMT: KVM behaves as if this were"
3316                                     " the host's SMT mode");
3317 
3318     object_property_add_bool(obj, "vfio-no-msix-emulation",
3319                              spapr_get_msix_emulation, NULL);
3320 
3321     object_property_add_uint64_ptr(obj, "kernel-addr",
3322                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3323     object_property_set_description(obj, "kernel-addr",
3324                                     stringify(KERNEL_LOAD_ADDR)
3325                                     " for -kernel is the default");
3326     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3327     /* The machine class defines the default interrupt controller mode */
3328     spapr->irq = smc->irq;
3329     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3330                             spapr_set_ic_mode);
3331     object_property_set_description(obj, "ic-mode",
3332                  "Specifies the interrupt controller mode (xics, xive, dual)");
3333 
3334     object_property_add_str(obj, "host-model",
3335         spapr_get_host_model, spapr_set_host_model);
3336     object_property_set_description(obj, "host-model",
3337         "Host model to advertise in guest device tree");
3338     object_property_add_str(obj, "host-serial",
3339         spapr_get_host_serial, spapr_set_host_serial);
3340     object_property_set_description(obj, "host-serial",
3341         "Host serial number to advertise in guest device tree");
3342 }
3343 
3344 static void spapr_machine_finalizefn(Object *obj)
3345 {
3346     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3347 
3348     g_free(spapr->kvm_type);
3349 }
3350 
3351 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3352 {
3353     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3354     PowerPCCPU *cpu = POWERPC_CPU(cs);
3355     CPUPPCState *env = &cpu->env;
3356 
3357     cpu_synchronize_state(cs);
3358     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3359     if (spapr->fwnmi_system_reset_addr != -1) {
3360         uint64_t rtas_addr, addr;
3361 
3362         /* get rtas addr from fdt */
3363         rtas_addr = spapr_get_rtas_addr();
3364         if (!rtas_addr) {
3365             qemu_system_guest_panicked(NULL);
3366             return;
3367         }
3368 
3369         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3370         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3371         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3372         env->gpr[3] = addr;
3373     }
3374     ppc_cpu_do_system_reset(cs);
3375     if (spapr->fwnmi_system_reset_addr != -1) {
3376         env->nip = spapr->fwnmi_system_reset_addr;
3377     }
3378 }
3379 
3380 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3381 {
3382     CPUState *cs;
3383 
3384     CPU_FOREACH(cs) {
3385         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3386     }
3387 }
3388 
3389 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3390                           void *fdt, int *fdt_start_offset, Error **errp)
3391 {
3392     uint64_t addr;
3393     uint32_t node;
3394 
3395     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3396     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3397                                     &error_abort);
3398     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3399                                              SPAPR_MEMORY_BLOCK_SIZE);
3400     return 0;
3401 }
3402 
3403 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3404                            bool dedicated_hp_event_source)
3405 {
3406     SpaprDrc *drc;
3407     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3408     int i;
3409     uint64_t addr = addr_start;
3410     bool hotplugged = spapr_drc_hotplugged(dev);
3411 
3412     for (i = 0; i < nr_lmbs; i++) {
3413         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3414                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3415         g_assert(drc);
3416 
3417         /*
3418          * memory_device_get_free_addr() provided a range of free addresses
3419          * that doesn't overlap with any existing mapping at pre-plug. The
3420          * corresponding LMB DRCs are thus assumed to be all attachable.
3421          */
3422         spapr_drc_attach(drc, dev);
3423         if (!hotplugged) {
3424             spapr_drc_reset(drc);
3425         }
3426         addr += SPAPR_MEMORY_BLOCK_SIZE;
3427     }
3428     /* send hotplug notification to the
3429      * guest only in case of hotplugged memory
3430      */
3431     if (hotplugged) {
3432         if (dedicated_hp_event_source) {
3433             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3434                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3435             g_assert(drc);
3436             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3437                                                    nr_lmbs,
3438                                                    spapr_drc_index(drc));
3439         } else {
3440             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3441                                            nr_lmbs);
3442         }
3443     }
3444 }
3445 
3446 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3447 {
3448     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3449     PCDIMMDevice *dimm = PC_DIMM(dev);
3450     uint64_t size, addr;
3451     int64_t slot;
3452     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3453 
3454     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3455 
3456     pc_dimm_plug(dimm, MACHINE(ms));
3457 
3458     if (!is_nvdimm) {
3459         addr = object_property_get_uint(OBJECT(dimm),
3460                                         PC_DIMM_ADDR_PROP, &error_abort);
3461         spapr_add_lmbs(dev, addr, size,
3462                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT));
3463     } else {
3464         slot = object_property_get_int(OBJECT(dimm),
3465                                        PC_DIMM_SLOT_PROP, &error_abort);
3466         /* We should have valid slot number at this point */
3467         g_assert(slot >= 0);
3468         spapr_add_nvdimm(dev, slot);
3469     }
3470 }
3471 
3472 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3473                                   Error **errp)
3474 {
3475     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3476     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3477     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3478     PCDIMMDevice *dimm = PC_DIMM(dev);
3479     Error *local_err = NULL;
3480     uint64_t size;
3481     Object *memdev;
3482     hwaddr pagesize;
3483 
3484     if (!smc->dr_lmb_enabled) {
3485         error_setg(errp, "Memory hotplug not supported for this machine");
3486         return;
3487     }
3488 
3489     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3490     if (local_err) {
3491         error_propagate(errp, local_err);
3492         return;
3493     }
3494 
3495     if (is_nvdimm) {
3496         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3497             return;
3498         }
3499     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3500         error_setg(errp, "Hotplugged memory size must be a multiple of "
3501                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3502         return;
3503     }
3504 
3505     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3506                                       &error_abort);
3507     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3508     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3509         return;
3510     }
3511 
3512     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3513 }
3514 
3515 struct SpaprDimmState {
3516     PCDIMMDevice *dimm;
3517     uint32_t nr_lmbs;
3518     QTAILQ_ENTRY(SpaprDimmState) next;
3519 };
3520 
3521 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3522                                                        PCDIMMDevice *dimm)
3523 {
3524     SpaprDimmState *dimm_state = NULL;
3525 
3526     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3527         if (dimm_state->dimm == dimm) {
3528             break;
3529         }
3530     }
3531     return dimm_state;
3532 }
3533 
3534 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3535                                                       uint32_t nr_lmbs,
3536                                                       PCDIMMDevice *dimm)
3537 {
3538     SpaprDimmState *ds = NULL;
3539 
3540     /*
3541      * If this request is for a DIMM whose removal had failed earlier
3542      * (due to guest's refusal to remove the LMBs), we would have this
3543      * dimm already in the pending_dimm_unplugs list. In that
3544      * case don't add again.
3545      */
3546     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3547     if (!ds) {
3548         ds = g_malloc0(sizeof(SpaprDimmState));
3549         ds->nr_lmbs = nr_lmbs;
3550         ds->dimm = dimm;
3551         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3552     }
3553     return ds;
3554 }
3555 
3556 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3557                                               SpaprDimmState *dimm_state)
3558 {
3559     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3560     g_free(dimm_state);
3561 }
3562 
3563 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3564                                                         PCDIMMDevice *dimm)
3565 {
3566     SpaprDrc *drc;
3567     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3568                                                   &error_abort);
3569     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3570     uint32_t avail_lmbs = 0;
3571     uint64_t addr_start, addr;
3572     int i;
3573 
3574     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3575                                           &error_abort);
3576 
3577     addr = addr_start;
3578     for (i = 0; i < nr_lmbs; i++) {
3579         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3580                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3581         g_assert(drc);
3582         if (drc->dev) {
3583             avail_lmbs++;
3584         }
3585         addr += SPAPR_MEMORY_BLOCK_SIZE;
3586     }
3587 
3588     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3589 }
3590 
3591 /* Callback to be called during DRC release. */
3592 void spapr_lmb_release(DeviceState *dev)
3593 {
3594     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3595     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3596     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3597 
3598     /* This information will get lost if a migration occurs
3599      * during the unplug process. In this case recover it. */
3600     if (ds == NULL) {
3601         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3602         g_assert(ds);
3603         /* The DRC being examined by the caller at least must be counted */
3604         g_assert(ds->nr_lmbs);
3605     }
3606 
3607     if (--ds->nr_lmbs) {
3608         return;
3609     }
3610 
3611     /*
3612      * Now that all the LMBs have been removed by the guest, call the
3613      * unplug handler chain. This can never fail.
3614      */
3615     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3616     object_unparent(OBJECT(dev));
3617 }
3618 
3619 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3620 {
3621     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3622     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3623 
3624     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3625     qdev_unrealize(dev);
3626     spapr_pending_dimm_unplugs_remove(spapr, ds);
3627 }
3628 
3629 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3630                                         DeviceState *dev, Error **errp)
3631 {
3632     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3633     PCDIMMDevice *dimm = PC_DIMM(dev);
3634     uint32_t nr_lmbs;
3635     uint64_t size, addr_start, addr;
3636     int i;
3637     SpaprDrc *drc;
3638 
3639     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3640         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3641         return;
3642     }
3643 
3644     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3645     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3646 
3647     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3648                                           &error_abort);
3649 
3650     /*
3651      * An existing pending dimm state for this DIMM means that there is an
3652      * unplug operation in progress, waiting for the spapr_lmb_release
3653      * callback to complete the job (BQL can't cover that far). In this case,
3654      * bail out to avoid detaching DRCs that were already released.
3655      */
3656     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3657         error_setg(errp, "Memory unplug already in progress for device %s",
3658                    dev->id);
3659         return;
3660     }
3661 
3662     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3663 
3664     addr = addr_start;
3665     for (i = 0; i < nr_lmbs; i++) {
3666         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3667                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3668         g_assert(drc);
3669 
3670         spapr_drc_detach(drc);
3671         addr += SPAPR_MEMORY_BLOCK_SIZE;
3672     }
3673 
3674     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3675                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3676     g_assert(drc);
3677     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3678                                               nr_lmbs, spapr_drc_index(drc));
3679 }
3680 
3681 /* Callback to be called during DRC release. */
3682 void spapr_core_release(DeviceState *dev)
3683 {
3684     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3685 
3686     /* Call the unplug handler chain. This can never fail. */
3687     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3688     object_unparent(OBJECT(dev));
3689 }
3690 
3691 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3692 {
3693     MachineState *ms = MACHINE(hotplug_dev);
3694     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3695     CPUCore *cc = CPU_CORE(dev);
3696     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3697 
3698     if (smc->pre_2_10_has_unused_icps) {
3699         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3700         int i;
3701 
3702         for (i = 0; i < cc->nr_threads; i++) {
3703             CPUState *cs = CPU(sc->threads[i]);
3704 
3705             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3706         }
3707     }
3708 
3709     assert(core_slot);
3710     core_slot->cpu = NULL;
3711     qdev_unrealize(dev);
3712 }
3713 
3714 static
3715 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3716                                Error **errp)
3717 {
3718     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3719     int index;
3720     SpaprDrc *drc;
3721     CPUCore *cc = CPU_CORE(dev);
3722 
3723     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3724         error_setg(errp, "Unable to find CPU core with core-id: %d",
3725                    cc->core_id);
3726         return;
3727     }
3728     if (index == 0) {
3729         error_setg(errp, "Boot CPU core may not be unplugged");
3730         return;
3731     }
3732 
3733     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3734                           spapr_vcpu_id(spapr, cc->core_id));
3735     g_assert(drc);
3736 
3737     if (!spapr_drc_unplug_requested(drc)) {
3738         spapr_drc_detach(drc);
3739         spapr_hotplug_req_remove_by_index(drc);
3740     }
3741 }
3742 
3743 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3744                            void *fdt, int *fdt_start_offset, Error **errp)
3745 {
3746     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3747     CPUState *cs = CPU(core->threads[0]);
3748     PowerPCCPU *cpu = POWERPC_CPU(cs);
3749     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3750     int id = spapr_get_vcpu_id(cpu);
3751     g_autofree char *nodename = NULL;
3752     int offset;
3753 
3754     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3755     offset = fdt_add_subnode(fdt, 0, nodename);
3756 
3757     spapr_dt_cpu(cs, fdt, offset, spapr);
3758 
3759     *fdt_start_offset = offset;
3760     return 0;
3761 }
3762 
3763 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3764 {
3765     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3766     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3767     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3768     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3769     CPUCore *cc = CPU_CORE(dev);
3770     CPUState *cs;
3771     SpaprDrc *drc;
3772     CPUArchId *core_slot;
3773     int index;
3774     bool hotplugged = spapr_drc_hotplugged(dev);
3775     int i;
3776 
3777     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3778     g_assert(core_slot); /* Already checked in spapr_core_pre_plug() */
3779 
3780     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3781                           spapr_vcpu_id(spapr, cc->core_id));
3782 
3783     g_assert(drc || !mc->has_hotpluggable_cpus);
3784 
3785     if (drc) {
3786         /*
3787          * spapr_core_pre_plug() already buys us this is a brand new
3788          * core being plugged into a free slot. Nothing should already
3789          * be attached to the corresponding DRC.
3790          */
3791         spapr_drc_attach(drc, dev);
3792 
3793         if (hotplugged) {
3794             /*
3795              * Send hotplug notification interrupt to the guest only
3796              * in case of hotplugged CPUs.
3797              */
3798             spapr_hotplug_req_add_by_index(drc);
3799         } else {
3800             spapr_drc_reset(drc);
3801         }
3802     }
3803 
3804     core_slot->cpu = OBJECT(dev);
3805 
3806     /*
3807      * Set compatibility mode to match the boot CPU, which was either set
3808      * by the machine reset code or by CAS. This really shouldn't fail at
3809      * this point.
3810      */
3811     if (hotplugged) {
3812         for (i = 0; i < cc->nr_threads; i++) {
3813             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3814                            &error_abort);
3815         }
3816     }
3817 
3818     if (smc->pre_2_10_has_unused_icps) {
3819         for (i = 0; i < cc->nr_threads; i++) {
3820             cs = CPU(core->threads[i]);
3821             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3822         }
3823     }
3824 }
3825 
3826 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3827                                 Error **errp)
3828 {
3829     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3830     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3831     CPUCore *cc = CPU_CORE(dev);
3832     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3833     const char *type = object_get_typename(OBJECT(dev));
3834     CPUArchId *core_slot;
3835     int index;
3836     unsigned int smp_threads = machine->smp.threads;
3837 
3838     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3839         error_setg(errp, "CPU hotplug not supported for this machine");
3840         return;
3841     }
3842 
3843     if (strcmp(base_core_type, type)) {
3844         error_setg(errp, "CPU core type should be %s", base_core_type);
3845         return;
3846     }
3847 
3848     if (cc->core_id % smp_threads) {
3849         error_setg(errp, "invalid core id %d", cc->core_id);
3850         return;
3851     }
3852 
3853     /*
3854      * In general we should have homogeneous threads-per-core, but old
3855      * (pre hotplug support) machine types allow the last core to have
3856      * reduced threads as a compatibility hack for when we allowed
3857      * total vcpus not a multiple of threads-per-core.
3858      */
3859     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3860         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
3861                    smp_threads);
3862         return;
3863     }
3864 
3865     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3866     if (!core_slot) {
3867         error_setg(errp, "core id %d out of range", cc->core_id);
3868         return;
3869     }
3870 
3871     if (core_slot->cpu) {
3872         error_setg(errp, "core %d already populated", cc->core_id);
3873         return;
3874     }
3875 
3876     numa_cpu_pre_plug(core_slot, dev, errp);
3877 }
3878 
3879 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3880                           void *fdt, int *fdt_start_offset, Error **errp)
3881 {
3882     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3883     int intc_phandle;
3884 
3885     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3886     if (intc_phandle <= 0) {
3887         return -1;
3888     }
3889 
3890     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3891         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3892         return -1;
3893     }
3894 
3895     /* generally SLOF creates these, for hotplug it's up to QEMU */
3896     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3897 
3898     return 0;
3899 }
3900 
3901 static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3902                                Error **errp)
3903 {
3904     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3905     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3906     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3907     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3908     SpaprDrc *drc;
3909 
3910     if (dev->hotplugged && !smc->dr_phb_enabled) {
3911         error_setg(errp, "PHB hotplug not supported for this machine");
3912         return false;
3913     }
3914 
3915     if (sphb->index == (uint32_t)-1) {
3916         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3917         return false;
3918     }
3919 
3920     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3921     if (drc && drc->dev) {
3922         error_setg(errp, "PHB %d already attached", sphb->index);
3923         return false;
3924     }
3925 
3926     /*
3927      * This will check that sphb->index doesn't exceed the maximum number of
3928      * PHBs for the current machine type.
3929      */
3930     return
3931         smc->phb_placement(spapr, sphb->index,
3932                            &sphb->buid, &sphb->io_win_addr,
3933                            &sphb->mem_win_addr, &sphb->mem64_win_addr,
3934                            windows_supported, sphb->dma_liobn,
3935                            &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3936                            errp);
3937 }
3938 
3939 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3940 {
3941     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3942     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3943     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3944     SpaprDrc *drc;
3945     bool hotplugged = spapr_drc_hotplugged(dev);
3946 
3947     if (!smc->dr_phb_enabled) {
3948         return;
3949     }
3950 
3951     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3952     /* hotplug hooks should check it's enabled before getting this far */
3953     assert(drc);
3954 
3955     /* spapr_phb_pre_plug() already checked the DRC is attachable */
3956     spapr_drc_attach(drc, dev);
3957 
3958     if (hotplugged) {
3959         spapr_hotplug_req_add_by_index(drc);
3960     } else {
3961         spapr_drc_reset(drc);
3962     }
3963 }
3964 
3965 void spapr_phb_release(DeviceState *dev)
3966 {
3967     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3968 
3969     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3970     object_unparent(OBJECT(dev));
3971 }
3972 
3973 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3974 {
3975     qdev_unrealize(dev);
3976 }
3977 
3978 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
3979                                      DeviceState *dev, Error **errp)
3980 {
3981     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3982     SpaprDrc *drc;
3983 
3984     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3985     assert(drc);
3986 
3987     if (!spapr_drc_unplug_requested(drc)) {
3988         spapr_drc_detach(drc);
3989         spapr_hotplug_req_remove_by_index(drc);
3990     }
3991 }
3992 
3993 static
3994 bool spapr_tpm_proxy_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3995                               Error **errp)
3996 {
3997     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3998 
3999     if (spapr->tpm_proxy != NULL) {
4000         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4001         return false;
4002     }
4003 
4004     return true;
4005 }
4006 
4007 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4008 {
4009     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4010     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4011 
4012     /* Already checked in spapr_tpm_proxy_pre_plug() */
4013     g_assert(spapr->tpm_proxy == NULL);
4014 
4015     spapr->tpm_proxy = tpm_proxy;
4016 }
4017 
4018 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4019 {
4020     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4021 
4022     qdev_unrealize(dev);
4023     object_unparent(OBJECT(dev));
4024     spapr->tpm_proxy = NULL;
4025 }
4026 
4027 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4028                                       DeviceState *dev, Error **errp)
4029 {
4030     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4031         spapr_memory_plug(hotplug_dev, dev);
4032     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4033         spapr_core_plug(hotplug_dev, dev);
4034     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4035         spapr_phb_plug(hotplug_dev, dev);
4036     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4037         spapr_tpm_proxy_plug(hotplug_dev, dev);
4038     }
4039 }
4040 
4041 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4042                                         DeviceState *dev, Error **errp)
4043 {
4044     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4045         spapr_memory_unplug(hotplug_dev, dev);
4046     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4047         spapr_core_unplug(hotplug_dev, dev);
4048     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4049         spapr_phb_unplug(hotplug_dev, dev);
4050     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4051         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4052     }
4053 }
4054 
4055 bool spapr_memory_hot_unplug_supported(SpaprMachineState *spapr)
4056 {
4057     return spapr_ovec_test(spapr->ov5_cas, OV5_HP_EVT) ||
4058         /*
4059          * CAS will process all pending unplug requests.
4060          *
4061          * HACK: a guest could theoretically have cleared all bits in OV5,
4062          * but none of the guests we care for do.
4063          */
4064         spapr_ovec_empty(spapr->ov5_cas);
4065 }
4066 
4067 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4068                                                 DeviceState *dev, Error **errp)
4069 {
4070     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4071     MachineClass *mc = MACHINE_GET_CLASS(sms);
4072     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4073 
4074     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4075         if (spapr_memory_hot_unplug_supported(sms)) {
4076             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4077         } else {
4078             error_setg(errp, "Memory hot unplug not supported for this guest");
4079         }
4080     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4081         if (!mc->has_hotpluggable_cpus) {
4082             error_setg(errp, "CPU hot unplug not supported on this machine");
4083             return;
4084         }
4085         spapr_core_unplug_request(hotplug_dev, dev, errp);
4086     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4087         if (!smc->dr_phb_enabled) {
4088             error_setg(errp, "PHB hot unplug not supported on this machine");
4089             return;
4090         }
4091         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4092     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4093         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4094     }
4095 }
4096 
4097 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4098                                           DeviceState *dev, Error **errp)
4099 {
4100     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4101         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4102     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4103         spapr_core_pre_plug(hotplug_dev, dev, errp);
4104     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4105         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4106     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4107         spapr_tpm_proxy_pre_plug(hotplug_dev, dev, errp);
4108     }
4109 }
4110 
4111 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4112                                                  DeviceState *dev)
4113 {
4114     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4115         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4116         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4117         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4118         return HOTPLUG_HANDLER(machine);
4119     }
4120     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4121         PCIDevice *pcidev = PCI_DEVICE(dev);
4122         PCIBus *root = pci_device_root_bus(pcidev);
4123         SpaprPhbState *phb =
4124             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4125                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4126 
4127         if (phb) {
4128             return HOTPLUG_HANDLER(phb);
4129         }
4130     }
4131     return NULL;
4132 }
4133 
4134 static CpuInstanceProperties
4135 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4136 {
4137     CPUArchId *core_slot;
4138     MachineClass *mc = MACHINE_GET_CLASS(machine);
4139 
4140     /* make sure possible_cpu are intialized */
4141     mc->possible_cpu_arch_ids(machine);
4142     /* get CPU core slot containing thread that matches cpu_index */
4143     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4144     assert(core_slot);
4145     return core_slot->props;
4146 }
4147 
4148 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4149 {
4150     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4151 }
4152 
4153 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4154 {
4155     int i;
4156     unsigned int smp_threads = machine->smp.threads;
4157     unsigned int smp_cpus = machine->smp.cpus;
4158     const char *core_type;
4159     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4160     MachineClass *mc = MACHINE_GET_CLASS(machine);
4161 
4162     if (!mc->has_hotpluggable_cpus) {
4163         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4164     }
4165     if (machine->possible_cpus) {
4166         assert(machine->possible_cpus->len == spapr_max_cores);
4167         return machine->possible_cpus;
4168     }
4169 
4170     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4171     if (!core_type) {
4172         error_report("Unable to find sPAPR CPU Core definition");
4173         exit(1);
4174     }
4175 
4176     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4177                              sizeof(CPUArchId) * spapr_max_cores);
4178     machine->possible_cpus->len = spapr_max_cores;
4179     for (i = 0; i < machine->possible_cpus->len; i++) {
4180         int core_id = i * smp_threads;
4181 
4182         machine->possible_cpus->cpus[i].type = core_type;
4183         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4184         machine->possible_cpus->cpus[i].arch_id = core_id;
4185         machine->possible_cpus->cpus[i].props.has_core_id = true;
4186         machine->possible_cpus->cpus[i].props.core_id = core_id;
4187     }
4188     return machine->possible_cpus;
4189 }
4190 
4191 static bool spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4192                                 uint64_t *buid, hwaddr *pio,
4193                                 hwaddr *mmio32, hwaddr *mmio64,
4194                                 unsigned n_dma, uint32_t *liobns,
4195                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4196 {
4197     /*
4198      * New-style PHB window placement.
4199      *
4200      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4201      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4202      * windows.
4203      *
4204      * Some guest kernels can't work with MMIO windows above 1<<46
4205      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4206      *
4207      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4208      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4209      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4210      * 1TiB 64-bit MMIO windows for each PHB.
4211      */
4212     const uint64_t base_buid = 0x800000020000000ULL;
4213     int i;
4214 
4215     /* Sanity check natural alignments */
4216     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4217     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4218     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4219     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4220     /* Sanity check bounds */
4221     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4222                       SPAPR_PCI_MEM32_WIN_SIZE);
4223     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4224                       SPAPR_PCI_MEM64_WIN_SIZE);
4225 
4226     if (index >= SPAPR_MAX_PHBS) {
4227         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4228                    SPAPR_MAX_PHBS - 1);
4229         return false;
4230     }
4231 
4232     *buid = base_buid + index;
4233     for (i = 0; i < n_dma; ++i) {
4234         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4235     }
4236 
4237     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4238     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4239     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4240 
4241     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4242     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4243     return true;
4244 }
4245 
4246 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4247 {
4248     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4249 
4250     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4251 }
4252 
4253 static void spapr_ics_resend(XICSFabric *dev)
4254 {
4255     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4256 
4257     ics_resend(spapr->ics);
4258 }
4259 
4260 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4261 {
4262     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4263 
4264     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4265 }
4266 
4267 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4268                                  Monitor *mon)
4269 {
4270     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4271 
4272     spapr_irq_print_info(spapr, mon);
4273     monitor_printf(mon, "irqchip: %s\n",
4274                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4275 }
4276 
4277 /*
4278  * This is a XIVE only operation
4279  */
4280 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4281                            uint8_t nvt_blk, uint32_t nvt_idx,
4282                            bool cam_ignore, uint8_t priority,
4283                            uint32_t logic_serv, XiveTCTXMatch *match)
4284 {
4285     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4286     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4287     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4288     int count;
4289 
4290     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4291                            priority, logic_serv, match);
4292     if (count < 0) {
4293         return count;
4294     }
4295 
4296     /*
4297      * When we implement the save and restore of the thread interrupt
4298      * contexts in the enter/exit CPU handlers of the machine and the
4299      * escalations in QEMU, we should be able to handle non dispatched
4300      * vCPUs.
4301      *
4302      * Until this is done, the sPAPR machine should find at least one
4303      * matching context always.
4304      */
4305     if (count == 0) {
4306         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4307                       nvt_blk, nvt_idx);
4308     }
4309 
4310     return count;
4311 }
4312 
4313 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4314 {
4315     return cpu->vcpu_id;
4316 }
4317 
4318 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4319 {
4320     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4321     MachineState *ms = MACHINE(spapr);
4322     int vcpu_id;
4323 
4324     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4325 
4326     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4327         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4328         error_append_hint(errp, "Adjust the number of cpus to %d "
4329                           "or try to raise the number of threads per core\n",
4330                           vcpu_id * ms->smp.threads / spapr->vsmt);
4331         return false;
4332     }
4333 
4334     cpu->vcpu_id = vcpu_id;
4335     return true;
4336 }
4337 
4338 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4339 {
4340     CPUState *cs;
4341 
4342     CPU_FOREACH(cs) {
4343         PowerPCCPU *cpu = POWERPC_CPU(cs);
4344 
4345         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4346             return cpu;
4347         }
4348     }
4349 
4350     return NULL;
4351 }
4352 
4353 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4354 {
4355     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4356 
4357     /* These are only called by TCG, KVM maintains dispatch state */
4358 
4359     spapr_cpu->prod = false;
4360     if (spapr_cpu->vpa_addr) {
4361         CPUState *cs = CPU(cpu);
4362         uint32_t dispatch;
4363 
4364         dispatch = ldl_be_phys(cs->as,
4365                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4366         dispatch++;
4367         if ((dispatch & 1) != 0) {
4368             qemu_log_mask(LOG_GUEST_ERROR,
4369                           "VPA: incorrect dispatch counter value for "
4370                           "dispatched partition %u, correcting.\n", dispatch);
4371             dispatch++;
4372         }
4373         stl_be_phys(cs->as,
4374                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4375     }
4376 }
4377 
4378 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4379 {
4380     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4381 
4382     if (spapr_cpu->vpa_addr) {
4383         CPUState *cs = CPU(cpu);
4384         uint32_t dispatch;
4385 
4386         dispatch = ldl_be_phys(cs->as,
4387                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4388         dispatch++;
4389         if ((dispatch & 1) != 1) {
4390             qemu_log_mask(LOG_GUEST_ERROR,
4391                           "VPA: incorrect dispatch counter value for "
4392                           "preempted partition %u, correcting.\n", dispatch);
4393             dispatch++;
4394         }
4395         stl_be_phys(cs->as,
4396                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4397     }
4398 }
4399 
4400 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4401 {
4402     MachineClass *mc = MACHINE_CLASS(oc);
4403     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4404     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4405     NMIClass *nc = NMI_CLASS(oc);
4406     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4407     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4408     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4409     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4410     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4411 
4412     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4413     mc->ignore_boot_device_suffixes = true;
4414 
4415     /*
4416      * We set up the default / latest behaviour here.  The class_init
4417      * functions for the specific versioned machine types can override
4418      * these details for backwards compatibility
4419      */
4420     mc->init = spapr_machine_init;
4421     mc->reset = spapr_machine_reset;
4422     mc->block_default_type = IF_SCSI;
4423     mc->max_cpus = 1024;
4424     mc->no_parallel = 1;
4425     mc->default_boot_order = "";
4426     mc->default_ram_size = 512 * MiB;
4427     mc->default_ram_id = "ppc_spapr.ram";
4428     mc->default_display = "std";
4429     mc->kvm_type = spapr_kvm_type;
4430     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4431     mc->pci_allow_0_address = true;
4432     assert(!mc->get_hotplug_handler);
4433     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4434     hc->pre_plug = spapr_machine_device_pre_plug;
4435     hc->plug = spapr_machine_device_plug;
4436     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4437     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4438     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4439     hc->unplug_request = spapr_machine_device_unplug_request;
4440     hc->unplug = spapr_machine_device_unplug;
4441 
4442     smc->dr_lmb_enabled = true;
4443     smc->update_dt_enabled = true;
4444     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4445     mc->has_hotpluggable_cpus = true;
4446     mc->nvdimm_supported = true;
4447     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4448     fwc->get_dev_path = spapr_get_fw_dev_path;
4449     nc->nmi_monitor_handler = spapr_nmi;
4450     smc->phb_placement = spapr_phb_placement;
4451     vhc->hypercall = emulate_spapr_hypercall;
4452     vhc->hpt_mask = spapr_hpt_mask;
4453     vhc->map_hptes = spapr_map_hptes;
4454     vhc->unmap_hptes = spapr_unmap_hptes;
4455     vhc->hpte_set_c = spapr_hpte_set_c;
4456     vhc->hpte_set_r = spapr_hpte_set_r;
4457     vhc->get_pate = spapr_get_pate;
4458     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4459     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4460     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4461     xic->ics_get = spapr_ics_get;
4462     xic->ics_resend = spapr_ics_resend;
4463     xic->icp_get = spapr_icp_get;
4464     ispc->print_info = spapr_pic_print_info;
4465     /* Force NUMA node memory size to be a multiple of
4466      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4467      * in which LMBs are represented and hot-added
4468      */
4469     mc->numa_mem_align_shift = 28;
4470     mc->auto_enable_numa = true;
4471 
4472     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4473     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4474     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4475     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4476     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4477     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4478     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4479     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4480     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4481     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4482     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4483     spapr_caps_add_properties(smc);
4484     smc->irq = &spapr_irq_dual;
4485     smc->dr_phb_enabled = true;
4486     smc->linux_pci_probe = true;
4487     smc->smp_threads_vsmt = true;
4488     smc->nr_xirqs = SPAPR_NR_XIRQS;
4489     xfc->match_nvt = spapr_match_nvt;
4490 }
4491 
4492 static const TypeInfo spapr_machine_info = {
4493     .name          = TYPE_SPAPR_MACHINE,
4494     .parent        = TYPE_MACHINE,
4495     .abstract      = true,
4496     .instance_size = sizeof(SpaprMachineState),
4497     .instance_init = spapr_instance_init,
4498     .instance_finalize = spapr_machine_finalizefn,
4499     .class_size    = sizeof(SpaprMachineClass),
4500     .class_init    = spapr_machine_class_init,
4501     .interfaces = (InterfaceInfo[]) {
4502         { TYPE_FW_PATH_PROVIDER },
4503         { TYPE_NMI },
4504         { TYPE_HOTPLUG_HANDLER },
4505         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4506         { TYPE_XICS_FABRIC },
4507         { TYPE_INTERRUPT_STATS_PROVIDER },
4508         { TYPE_XIVE_FABRIC },
4509         { }
4510     },
4511 };
4512 
4513 static void spapr_machine_latest_class_options(MachineClass *mc)
4514 {
4515     mc->alias = "pseries";
4516     mc->is_default = true;
4517 }
4518 
4519 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4520     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4521                                                     void *data)      \
4522     {                                                                \
4523         MachineClass *mc = MACHINE_CLASS(oc);                        \
4524         spapr_machine_##suffix##_class_options(mc);                  \
4525         if (latest) {                                                \
4526             spapr_machine_latest_class_options(mc);                  \
4527         }                                                            \
4528     }                                                                \
4529     static const TypeInfo spapr_machine_##suffix##_info = {          \
4530         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4531         .parent = TYPE_SPAPR_MACHINE,                                \
4532         .class_init = spapr_machine_##suffix##_class_init,           \
4533     };                                                               \
4534     static void spapr_machine_register_##suffix(void)                \
4535     {                                                                \
4536         type_register(&spapr_machine_##suffix##_info);               \
4537     }                                                                \
4538     type_init(spapr_machine_register_##suffix)
4539 
4540 /*
4541  * pseries-6.0
4542  */
4543 static void spapr_machine_6_0_class_options(MachineClass *mc)
4544 {
4545     /* Defaults for the latest behaviour inherited from the base class */
4546 }
4547 
4548 DEFINE_SPAPR_MACHINE(6_0, "6.0", true);
4549 
4550 /*
4551  * pseries-5.2
4552  */
4553 static void spapr_machine_5_2_class_options(MachineClass *mc)
4554 {
4555     spapr_machine_6_0_class_options(mc);
4556     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
4557 }
4558 
4559 DEFINE_SPAPR_MACHINE(5_2, "5.2", false);
4560 
4561 /*
4562  * pseries-5.1
4563  */
4564 static void spapr_machine_5_1_class_options(MachineClass *mc)
4565 {
4566     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4567 
4568     spapr_machine_5_2_class_options(mc);
4569     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4570     smc->pre_5_2_numa_associativity = true;
4571 }
4572 
4573 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4574 
4575 /*
4576  * pseries-5.0
4577  */
4578 static void spapr_machine_5_0_class_options(MachineClass *mc)
4579 {
4580     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4581     static GlobalProperty compat[] = {
4582         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4583     };
4584 
4585     spapr_machine_5_1_class_options(mc);
4586     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4587     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4588     mc->numa_mem_supported = true;
4589     smc->pre_5_1_assoc_refpoints = true;
4590 }
4591 
4592 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4593 
4594 /*
4595  * pseries-4.2
4596  */
4597 static void spapr_machine_4_2_class_options(MachineClass *mc)
4598 {
4599     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4600 
4601     spapr_machine_5_0_class_options(mc);
4602     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4603     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4604     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4605     smc->rma_limit = 16 * GiB;
4606     mc->nvdimm_supported = false;
4607 }
4608 
4609 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4610 
4611 /*
4612  * pseries-4.1
4613  */
4614 static void spapr_machine_4_1_class_options(MachineClass *mc)
4615 {
4616     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4617     static GlobalProperty compat[] = {
4618         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4619         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4620     };
4621 
4622     spapr_machine_4_2_class_options(mc);
4623     smc->linux_pci_probe = false;
4624     smc->smp_threads_vsmt = false;
4625     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4626     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4627 }
4628 
4629 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4630 
4631 /*
4632  * pseries-4.0
4633  */
4634 static bool phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4635                               uint64_t *buid, hwaddr *pio,
4636                               hwaddr *mmio32, hwaddr *mmio64,
4637                               unsigned n_dma, uint32_t *liobns,
4638                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4639 {
4640     if (!spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma,
4641                              liobns, nv2gpa, nv2atsd, errp)) {
4642         return false;
4643     }
4644 
4645     *nv2gpa = 0;
4646     *nv2atsd = 0;
4647     return true;
4648 }
4649 static void spapr_machine_4_0_class_options(MachineClass *mc)
4650 {
4651     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4652 
4653     spapr_machine_4_1_class_options(mc);
4654     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4655     smc->phb_placement = phb_placement_4_0;
4656     smc->irq = &spapr_irq_xics;
4657     smc->pre_4_1_migration = true;
4658 }
4659 
4660 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4661 
4662 /*
4663  * pseries-3.1
4664  */
4665 static void spapr_machine_3_1_class_options(MachineClass *mc)
4666 {
4667     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4668 
4669     spapr_machine_4_0_class_options(mc);
4670     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4671 
4672     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4673     smc->update_dt_enabled = false;
4674     smc->dr_phb_enabled = false;
4675     smc->broken_host_serial_model = true;
4676     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4677     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4678     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4679     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4680 }
4681 
4682 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4683 
4684 /*
4685  * pseries-3.0
4686  */
4687 
4688 static void spapr_machine_3_0_class_options(MachineClass *mc)
4689 {
4690     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4691 
4692     spapr_machine_3_1_class_options(mc);
4693     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4694 
4695     smc->legacy_irq_allocation = true;
4696     smc->nr_xirqs = 0x400;
4697     smc->irq = &spapr_irq_xics_legacy;
4698 }
4699 
4700 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4701 
4702 /*
4703  * pseries-2.12
4704  */
4705 static void spapr_machine_2_12_class_options(MachineClass *mc)
4706 {
4707     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4708     static GlobalProperty compat[] = {
4709         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4710         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4711     };
4712 
4713     spapr_machine_3_0_class_options(mc);
4714     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4715     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4716 
4717     /* We depend on kvm_enabled() to choose a default value for the
4718      * hpt-max-page-size capability. Of course we can't do it here
4719      * because this is too early and the HW accelerator isn't initialzed
4720      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4721      */
4722     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4723 }
4724 
4725 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4726 
4727 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4728 {
4729     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4730 
4731     spapr_machine_2_12_class_options(mc);
4732     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4733     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4734     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4735 }
4736 
4737 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4738 
4739 /*
4740  * pseries-2.11
4741  */
4742 
4743 static void spapr_machine_2_11_class_options(MachineClass *mc)
4744 {
4745     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4746 
4747     spapr_machine_2_12_class_options(mc);
4748     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4749     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4750 }
4751 
4752 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4753 
4754 /*
4755  * pseries-2.10
4756  */
4757 
4758 static void spapr_machine_2_10_class_options(MachineClass *mc)
4759 {
4760     spapr_machine_2_11_class_options(mc);
4761     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4762 }
4763 
4764 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4765 
4766 /*
4767  * pseries-2.9
4768  */
4769 
4770 static void spapr_machine_2_9_class_options(MachineClass *mc)
4771 {
4772     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4773     static GlobalProperty compat[] = {
4774         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4775     };
4776 
4777     spapr_machine_2_10_class_options(mc);
4778     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4779     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4780     smc->pre_2_10_has_unused_icps = true;
4781     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4782 }
4783 
4784 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4785 
4786 /*
4787  * pseries-2.8
4788  */
4789 
4790 static void spapr_machine_2_8_class_options(MachineClass *mc)
4791 {
4792     static GlobalProperty compat[] = {
4793         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4794     };
4795 
4796     spapr_machine_2_9_class_options(mc);
4797     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4798     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4799     mc->numa_mem_align_shift = 23;
4800 }
4801 
4802 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4803 
4804 /*
4805  * pseries-2.7
4806  */
4807 
4808 static bool phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4809                               uint64_t *buid, hwaddr *pio,
4810                               hwaddr *mmio32, hwaddr *mmio64,
4811                               unsigned n_dma, uint32_t *liobns,
4812                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4813 {
4814     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4815     const uint64_t base_buid = 0x800000020000000ULL;
4816     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4817     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4818     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4819     const uint32_t max_index = 255;
4820     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4821 
4822     uint64_t ram_top = MACHINE(spapr)->ram_size;
4823     hwaddr phb0_base, phb_base;
4824     int i;
4825 
4826     /* Do we have device memory? */
4827     if (MACHINE(spapr)->maxram_size > ram_top) {
4828         /* Can't just use maxram_size, because there may be an
4829          * alignment gap between normal and device memory regions
4830          */
4831         ram_top = MACHINE(spapr)->device_memory->base +
4832             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4833     }
4834 
4835     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4836 
4837     if (index > max_index) {
4838         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4839                    max_index);
4840         return false;
4841     }
4842 
4843     *buid = base_buid + index;
4844     for (i = 0; i < n_dma; ++i) {
4845         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4846     }
4847 
4848     phb_base = phb0_base + index * phb_spacing;
4849     *pio = phb_base + pio_offset;
4850     *mmio32 = phb_base + mmio_offset;
4851     /*
4852      * We don't set the 64-bit MMIO window, relying on the PHB's
4853      * fallback behaviour of automatically splitting a large "32-bit"
4854      * window into contiguous 32-bit and 64-bit windows
4855      */
4856 
4857     *nv2gpa = 0;
4858     *nv2atsd = 0;
4859     return true;
4860 }
4861 
4862 static void spapr_machine_2_7_class_options(MachineClass *mc)
4863 {
4864     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4865     static GlobalProperty compat[] = {
4866         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4867         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4868         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4869         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4870     };
4871 
4872     spapr_machine_2_8_class_options(mc);
4873     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4874     mc->default_machine_opts = "modern-hotplug-events=off";
4875     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4876     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4877     smc->phb_placement = phb_placement_2_7;
4878 }
4879 
4880 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4881 
4882 /*
4883  * pseries-2.6
4884  */
4885 
4886 static void spapr_machine_2_6_class_options(MachineClass *mc)
4887 {
4888     static GlobalProperty compat[] = {
4889         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4890     };
4891 
4892     spapr_machine_2_7_class_options(mc);
4893     mc->has_hotpluggable_cpus = false;
4894     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4895     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4896 }
4897 
4898 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4899 
4900 /*
4901  * pseries-2.5
4902  */
4903 
4904 static void spapr_machine_2_5_class_options(MachineClass *mc)
4905 {
4906     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4907     static GlobalProperty compat[] = {
4908         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4909     };
4910 
4911     spapr_machine_2_6_class_options(mc);
4912     smc->use_ohci_by_default = true;
4913     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4914     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4915 }
4916 
4917 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4918 
4919 /*
4920  * pseries-2.4
4921  */
4922 
4923 static void spapr_machine_2_4_class_options(MachineClass *mc)
4924 {
4925     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4926 
4927     spapr_machine_2_5_class_options(mc);
4928     smc->dr_lmb_enabled = false;
4929     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4930 }
4931 
4932 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4933 
4934 /*
4935  * pseries-2.3
4936  */
4937 
4938 static void spapr_machine_2_3_class_options(MachineClass *mc)
4939 {
4940     static GlobalProperty compat[] = {
4941         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4942     };
4943     spapr_machine_2_4_class_options(mc);
4944     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4945     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4946 }
4947 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4948 
4949 /*
4950  * pseries-2.2
4951  */
4952 
4953 static void spapr_machine_2_2_class_options(MachineClass *mc)
4954 {
4955     static GlobalProperty compat[] = {
4956         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4957     };
4958 
4959     spapr_machine_2_3_class_options(mc);
4960     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4961     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4962     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4963 }
4964 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4965 
4966 /*
4967  * pseries-2.1
4968  */
4969 
4970 static void spapr_machine_2_1_class_options(MachineClass *mc)
4971 {
4972     spapr_machine_2_2_class_options(mc);
4973     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4974 }
4975 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4976 
4977 static void spapr_machine_register_types(void)
4978 {
4979     type_register_static(&spapr_machine_info);
4980 }
4981 
4982 type_init(spapr_machine_register_types)
4983