xref: /openbmc/qemu/hw/ppc/spapr.c (revision 38e0b790)
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     char *nodename;
795     int i;
796 
797     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
798     _FDT(cpus_offset);
799     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
800     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
801 
802     /*
803      * We walk the CPUs in reverse order to ensure that CPU DT nodes
804      * created by fdt_add_subnode() end up in the right order in FDT
805      * for the guest kernel the enumerate the CPUs correctly.
806      *
807      * The CPU list cannot be traversed in reverse order, so we need
808      * to do extra work.
809      */
810     n_cpus = 0;
811     rev = NULL;
812     CPU_FOREACH(cs) {
813         rev = g_renew(CPUState *, rev, n_cpus + 1);
814         rev[n_cpus++] = cs;
815     }
816 
817     for (i = n_cpus - 1; i >= 0; i--) {
818         CPUState *cs = rev[i];
819         PowerPCCPU *cpu = POWERPC_CPU(cs);
820         int index = spapr_get_vcpu_id(cpu);
821         DeviceClass *dc = DEVICE_GET_CLASS(cs);
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         g_free(nodename);
831         _FDT(offset);
832         spapr_dt_cpu(cs, fdt, offset, spapr);
833     }
834 
835     g_free(rev);
836 }
837 
838 static int spapr_dt_rng(void *fdt)
839 {
840     int node;
841     int ret;
842 
843     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
844     if (node <= 0) {
845         return -1;
846     }
847     ret = fdt_setprop_string(fdt, node, "device_type",
848                              "ibm,platform-facilities");
849     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
850     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
851 
852     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
853     if (node <= 0) {
854         return -1;
855     }
856     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
857 
858     return ret ? -1 : 0;
859 }
860 
861 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
862 {
863     MachineState *ms = MACHINE(spapr);
864     int rtas;
865     GString *hypertas = g_string_sized_new(256);
866     GString *qemu_hypertas = g_string_sized_new(256);
867     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
868         memory_region_size(&MACHINE(spapr)->device_memory->mr);
869     uint32_t lrdr_capacity[] = {
870         cpu_to_be32(max_device_addr >> 32),
871         cpu_to_be32(max_device_addr & 0xffffffff),
872         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
873         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
874         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
875     };
876 
877     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
878 
879     /* hypertas */
880     add_str(hypertas, "hcall-pft");
881     add_str(hypertas, "hcall-term");
882     add_str(hypertas, "hcall-dabr");
883     add_str(hypertas, "hcall-interrupt");
884     add_str(hypertas, "hcall-tce");
885     add_str(hypertas, "hcall-vio");
886     add_str(hypertas, "hcall-splpar");
887     add_str(hypertas, "hcall-join");
888     add_str(hypertas, "hcall-bulk");
889     add_str(hypertas, "hcall-set-mode");
890     add_str(hypertas, "hcall-sprg0");
891     add_str(hypertas, "hcall-copy");
892     add_str(hypertas, "hcall-debug");
893     add_str(hypertas, "hcall-vphn");
894     add_str(qemu_hypertas, "hcall-memop1");
895 
896     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
897         add_str(hypertas, "hcall-multi-tce");
898     }
899 
900     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
901         add_str(hypertas, "hcall-hpt-resize");
902     }
903 
904     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
905                      hypertas->str, hypertas->len));
906     g_string_free(hypertas, TRUE);
907     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
908                      qemu_hypertas->str, qemu_hypertas->len));
909     g_string_free(qemu_hypertas, TRUE);
910 
911     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
912 
913     /*
914      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
915      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
916      *
917      * The system reset requirements are driven by existing Linux and PowerVM
918      * implementation which (contrary to PAPR) saves r3 in the error log
919      * structure like machine check, so Linux expects to find the saved r3
920      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
921      * does not look at the error value).
922      *
923      * System reset interrupts are not subject to interlock like machine
924      * check, so this memory area could be corrupted if the sreset is
925      * interrupted by a machine check (or vice versa) if it was shared. To
926      * prevent this, system reset uses per-CPU areas for the sreset save
927      * area. A system reset that interrupts a system reset handler could
928      * still overwrite this area, but Linux doesn't try to recover in that
929      * case anyway.
930      *
931      * The extra 8 bytes is required because Linux's FWNMI error log check
932      * is off-by-one.
933      */
934     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_ERROR_LOG_MAX +
935 			  ms->smp.max_cpus * sizeof(uint64_t)*2 + sizeof(uint64_t)));
936     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
937                           RTAS_ERROR_LOG_MAX));
938     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
939                           RTAS_EVENT_SCAN_RATE));
940 
941     g_assert(msi_nonbroken);
942     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
943 
944     /*
945      * According to PAPR, rtas ibm,os-term does not guarantee a return
946      * back to the guest cpu.
947      *
948      * While an additional ibm,extended-os-term property indicates
949      * that rtas call return will always occur. Set this property.
950      */
951     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
952 
953     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
954                      lrdr_capacity, sizeof(lrdr_capacity)));
955 
956     spapr_dt_rtas_tokens(fdt, rtas);
957 }
958 
959 /*
960  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
961  * and the XIVE features that the guest may request and thus the valid
962  * values for bytes 23..26 of option vector 5:
963  */
964 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
965                                           int chosen)
966 {
967     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
968 
969     char val[2 * 4] = {
970         23, 0x00, /* XICS / XIVE mode */
971         24, 0x00, /* Hash/Radix, filled in below. */
972         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
973         26, 0x40, /* Radix options: GTSE == yes. */
974     };
975 
976     if (spapr->irq->xics && spapr->irq->xive) {
977         val[1] = SPAPR_OV5_XIVE_BOTH;
978     } else if (spapr->irq->xive) {
979         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
980     } else {
981         assert(spapr->irq->xics);
982         val[1] = SPAPR_OV5_XIVE_LEGACY;
983     }
984 
985     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
986                           first_ppc_cpu->compat_pvr)) {
987         /*
988          * If we're in a pre POWER9 compat mode then the guest should
989          * do hash and use the legacy interrupt mode
990          */
991         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
992         val[3] = 0x00; /* Hash */
993     } else if (kvm_enabled()) {
994         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
995             val[3] = 0x80; /* OV5_MMU_BOTH */
996         } else if (kvmppc_has_cap_mmu_radix()) {
997             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
998         } else {
999             val[3] = 0x00; /* Hash */
1000         }
1001     } else {
1002         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1003         val[3] = 0xC0;
1004     }
1005     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1006                      val, sizeof(val)));
1007 }
1008 
1009 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1010 {
1011     MachineState *machine = MACHINE(spapr);
1012     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1013     int chosen;
1014 
1015     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1016 
1017     if (reset) {
1018         const char *boot_device = machine->boot_order;
1019         char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1020         size_t cb = 0;
1021         char *bootlist = get_boot_devices_list(&cb);
1022 
1023         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1024             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1025                                     machine->kernel_cmdline));
1026         }
1027 
1028         if (spapr->initrd_size) {
1029             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1030                                   spapr->initrd_base));
1031             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1032                                   spapr->initrd_base + spapr->initrd_size));
1033         }
1034 
1035         if (spapr->kernel_size) {
1036             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1037                                   cpu_to_be64(spapr->kernel_size) };
1038 
1039             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1040                          &kprop, sizeof(kprop)));
1041             if (spapr->kernel_le) {
1042                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1043             }
1044         }
1045         if (boot_menu) {
1046             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1047         }
1048         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1049         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1050         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1051 
1052         if (cb && bootlist) {
1053             int i;
1054 
1055             for (i = 0; i < cb; i++) {
1056                 if (bootlist[i] == '\n') {
1057                     bootlist[i] = ' ';
1058                 }
1059             }
1060             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1061         }
1062 
1063         if (boot_device && strlen(boot_device)) {
1064             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1065         }
1066 
1067         if (!spapr->has_graphics && stdout_path) {
1068             /*
1069              * "linux,stdout-path" and "stdout" properties are
1070              * deprecated by linux kernel. New platforms should only
1071              * use the "stdout-path" property. Set the new property
1072              * and continue using older property to remain compatible
1073              * with the existing firmware.
1074              */
1075             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1076             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1077         }
1078 
1079         /*
1080          * We can deal with BAR reallocation just fine, advertise it
1081          * to the guest
1082          */
1083         if (smc->linux_pci_probe) {
1084             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1085         }
1086 
1087         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1088 
1089         g_free(stdout_path);
1090         g_free(bootlist);
1091     }
1092 
1093     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1094 }
1095 
1096 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1097 {
1098     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1099      * KVM to work under pHyp with some guest co-operation */
1100     int hypervisor;
1101     uint8_t hypercall[16];
1102 
1103     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1104     /* indicate KVM hypercall interface */
1105     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1106     if (kvmppc_has_cap_fixup_hcalls()) {
1107         /*
1108          * Older KVM versions with older guest kernels were broken
1109          * with the magic page, don't allow the guest to map it.
1110          */
1111         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1112                                   sizeof(hypercall))) {
1113             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1114                              hypercall, sizeof(hypercall)));
1115         }
1116     }
1117 }
1118 
1119 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1120 {
1121     MachineState *machine = MACHINE(spapr);
1122     MachineClass *mc = MACHINE_GET_CLASS(machine);
1123     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1124     uint32_t root_drc_type_mask = 0;
1125     int ret;
1126     void *fdt;
1127     SpaprPhbState *phb;
1128     char *buf;
1129 
1130     fdt = g_malloc0(space);
1131     _FDT((fdt_create_empty_tree(fdt, space)));
1132 
1133     /* Root node */
1134     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1135     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1136     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1137 
1138     /* Guest UUID & Name*/
1139     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1140     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1141     if (qemu_uuid_set) {
1142         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1143     }
1144     g_free(buf);
1145 
1146     if (qemu_get_vm_name()) {
1147         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1148                                 qemu_get_vm_name()));
1149     }
1150 
1151     /* Host Model & Serial Number */
1152     if (spapr->host_model) {
1153         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1154     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1155         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1156         g_free(buf);
1157     }
1158 
1159     if (spapr->host_serial) {
1160         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1161     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1162         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1163         g_free(buf);
1164     }
1165 
1166     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1167     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1168 
1169     /* /interrupt controller */
1170     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1171 
1172     ret = spapr_dt_memory(spapr, fdt);
1173     if (ret < 0) {
1174         error_report("couldn't setup memory nodes in fdt");
1175         exit(1);
1176     }
1177 
1178     /* /vdevice */
1179     spapr_dt_vdevice(spapr->vio_bus, fdt);
1180 
1181     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1182         ret = spapr_dt_rng(fdt);
1183         if (ret < 0) {
1184             error_report("could not set up rng device in the fdt");
1185             exit(1);
1186         }
1187     }
1188 
1189     QLIST_FOREACH(phb, &spapr->phbs, list) {
1190         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1191         if (ret < 0) {
1192             error_report("couldn't setup PCI devices in fdt");
1193             exit(1);
1194         }
1195     }
1196 
1197     spapr_dt_cpus(fdt, spapr);
1198 
1199     /* ibm,drc-indexes and friends */
1200     if (smc->dr_lmb_enabled) {
1201         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_LMB;
1202     }
1203     if (smc->dr_phb_enabled) {
1204         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PHB;
1205     }
1206     if (mc->nvdimm_supported) {
1207         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PMEM;
1208     }
1209     if (root_drc_type_mask) {
1210         _FDT(spapr_dt_drc(fdt, 0, NULL, root_drc_type_mask));
1211     }
1212 
1213     if (mc->has_hotpluggable_cpus) {
1214         int offset = fdt_path_offset(fdt, "/cpus");
1215         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1216         if (ret < 0) {
1217             error_report("Couldn't set up CPU DR device tree properties");
1218             exit(1);
1219         }
1220     }
1221 
1222     /* /event-sources */
1223     spapr_dt_events(spapr, fdt);
1224 
1225     /* /rtas */
1226     spapr_dt_rtas(spapr, fdt);
1227 
1228     /* /chosen */
1229     spapr_dt_chosen(spapr, fdt, reset);
1230 
1231     /* /hypervisor */
1232     if (kvm_enabled()) {
1233         spapr_dt_hypervisor(spapr, fdt);
1234     }
1235 
1236     /* Build memory reserve map */
1237     if (reset) {
1238         if (spapr->kernel_size) {
1239             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1240                                   spapr->kernel_size)));
1241         }
1242         if (spapr->initrd_size) {
1243             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1244                                   spapr->initrd_size)));
1245         }
1246     }
1247 
1248     /* NVDIMM devices */
1249     if (mc->nvdimm_supported) {
1250         spapr_dt_persistent_memory(spapr, fdt);
1251     }
1252 
1253     return fdt;
1254 }
1255 
1256 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1257 {
1258     SpaprMachineState *spapr = opaque;
1259 
1260     return (addr & 0x0fffffff) + spapr->kernel_addr;
1261 }
1262 
1263 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1264                                     PowerPCCPU *cpu)
1265 {
1266     CPUPPCState *env = &cpu->env;
1267 
1268     /* The TCG path should also be holding the BQL at this point */
1269     g_assert(qemu_mutex_iothread_locked());
1270 
1271     if (msr_pr) {
1272         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1273         env->gpr[3] = H_PRIVILEGE;
1274     } else {
1275         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1276     }
1277 }
1278 
1279 struct LPCRSyncState {
1280     target_ulong value;
1281     target_ulong mask;
1282 };
1283 
1284 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1285 {
1286     struct LPCRSyncState *s = arg.host_ptr;
1287     PowerPCCPU *cpu = POWERPC_CPU(cs);
1288     CPUPPCState *env = &cpu->env;
1289     target_ulong lpcr;
1290 
1291     cpu_synchronize_state(cs);
1292     lpcr = env->spr[SPR_LPCR];
1293     lpcr &= ~s->mask;
1294     lpcr |= s->value;
1295     ppc_store_lpcr(cpu, lpcr);
1296 }
1297 
1298 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1299 {
1300     CPUState *cs;
1301     struct LPCRSyncState s = {
1302         .value = value,
1303         .mask = mask
1304     };
1305     CPU_FOREACH(cs) {
1306         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1307     }
1308 }
1309 
1310 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1311 {
1312     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1313 
1314     /* Copy PATE1:GR into PATE0:HR */
1315     entry->dw0 = spapr->patb_entry & PATE0_HR;
1316     entry->dw1 = spapr->patb_entry;
1317 }
1318 
1319 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1320 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1321 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1322 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1323 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1324 
1325 /*
1326  * Get the fd to access the kernel htab, re-opening it if necessary
1327  */
1328 static int get_htab_fd(SpaprMachineState *spapr)
1329 {
1330     Error *local_err = NULL;
1331 
1332     if (spapr->htab_fd >= 0) {
1333         return spapr->htab_fd;
1334     }
1335 
1336     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1337     if (spapr->htab_fd < 0) {
1338         error_report_err(local_err);
1339     }
1340 
1341     return spapr->htab_fd;
1342 }
1343 
1344 void close_htab_fd(SpaprMachineState *spapr)
1345 {
1346     if (spapr->htab_fd >= 0) {
1347         close(spapr->htab_fd);
1348     }
1349     spapr->htab_fd = -1;
1350 }
1351 
1352 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1353 {
1354     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1355 
1356     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1357 }
1358 
1359 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1360 {
1361     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1362 
1363     assert(kvm_enabled());
1364 
1365     if (!spapr->htab) {
1366         return 0;
1367     }
1368 
1369     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1370 }
1371 
1372 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1373                                                 hwaddr ptex, int n)
1374 {
1375     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1376     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1377 
1378     if (!spapr->htab) {
1379         /*
1380          * HTAB is controlled by KVM. Fetch into temporary buffer
1381          */
1382         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1383         kvmppc_read_hptes(hptes, ptex, n);
1384         return hptes;
1385     }
1386 
1387     /*
1388      * HTAB is controlled by QEMU. Just point to the internally
1389      * accessible PTEG.
1390      */
1391     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1392 }
1393 
1394 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1395                               const ppc_hash_pte64_t *hptes,
1396                               hwaddr ptex, int n)
1397 {
1398     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1399 
1400     if (!spapr->htab) {
1401         g_free((void *)hptes);
1402     }
1403 
1404     /* Nothing to do for qemu managed HPT */
1405 }
1406 
1407 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1408                       uint64_t pte0, uint64_t pte1)
1409 {
1410     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1411     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1412 
1413     if (!spapr->htab) {
1414         kvmppc_write_hpte(ptex, pte0, pte1);
1415     } else {
1416         if (pte0 & HPTE64_V_VALID) {
1417             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1418             /*
1419              * When setting valid, we write PTE1 first. This ensures
1420              * proper synchronization with the reading code in
1421              * ppc_hash64_pteg_search()
1422              */
1423             smp_wmb();
1424             stq_p(spapr->htab + offset, pte0);
1425         } else {
1426             stq_p(spapr->htab + offset, pte0);
1427             /*
1428              * When clearing it we set PTE0 first. This ensures proper
1429              * synchronization with the reading code in
1430              * ppc_hash64_pteg_search()
1431              */
1432             smp_wmb();
1433             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1434         }
1435     }
1436 }
1437 
1438 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1439                              uint64_t pte1)
1440 {
1441     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1442     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1443 
1444     if (!spapr->htab) {
1445         /* There should always be a hash table when this is called */
1446         error_report("spapr_hpte_set_c called with no hash table !");
1447         return;
1448     }
1449 
1450     /* The HW performs a non-atomic byte update */
1451     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1452 }
1453 
1454 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1455                              uint64_t pte1)
1456 {
1457     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1458     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1459 
1460     if (!spapr->htab) {
1461         /* There should always be a hash table when this is called */
1462         error_report("spapr_hpte_set_r called with no hash table !");
1463         return;
1464     }
1465 
1466     /* The HW performs a non-atomic byte update */
1467     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1468 }
1469 
1470 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1471 {
1472     int shift;
1473 
1474     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1475      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1476      * that's much more than is needed for Linux guests */
1477     shift = ctz64(pow2ceil(ramsize)) - 7;
1478     shift = MAX(shift, 18); /* Minimum architected size */
1479     shift = MIN(shift, 46); /* Maximum architected size */
1480     return shift;
1481 }
1482 
1483 void spapr_free_hpt(SpaprMachineState *spapr)
1484 {
1485     g_free(spapr->htab);
1486     spapr->htab = NULL;
1487     spapr->htab_shift = 0;
1488     close_htab_fd(spapr);
1489 }
1490 
1491 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1492 {
1493     ERRP_GUARD();
1494     long rc;
1495 
1496     /* Clean up any HPT info from a previous boot */
1497     spapr_free_hpt(spapr);
1498 
1499     rc = kvmppc_reset_htab(shift);
1500 
1501     if (rc == -EOPNOTSUPP) {
1502         error_setg(errp, "HPT not supported in nested guests");
1503         return -EOPNOTSUPP;
1504     }
1505 
1506     if (rc < 0) {
1507         /* kernel-side HPT needed, but couldn't allocate one */
1508         error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1509                          shift);
1510         error_append_hint(errp, "Try smaller maxmem?\n");
1511         return -errno;
1512     } else if (rc > 0) {
1513         /* kernel-side HPT allocated */
1514         if (rc != shift) {
1515             error_setg(errp,
1516                        "Requested order %d HPT, but kernel allocated order %ld",
1517                        shift, rc);
1518             error_append_hint(errp, "Try smaller maxmem?\n");
1519             return -ENOSPC;
1520         }
1521 
1522         spapr->htab_shift = shift;
1523         spapr->htab = NULL;
1524     } else {
1525         /* kernel-side HPT not needed, allocate in userspace instead */
1526         size_t size = 1ULL << shift;
1527         int i;
1528 
1529         spapr->htab = qemu_memalign(size, size);
1530         memset(spapr->htab, 0, size);
1531         spapr->htab_shift = shift;
1532 
1533         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1534             DIRTY_HPTE(HPTE(spapr->htab, i));
1535         }
1536     }
1537     /* We're setting up a hash table, so that means we're not radix */
1538     spapr->patb_entry = 0;
1539     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1540     return 0;
1541 }
1542 
1543 void spapr_setup_hpt(SpaprMachineState *spapr)
1544 {
1545     int hpt_shift;
1546 
1547     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1548         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1549     } else {
1550         uint64_t current_ram_size;
1551 
1552         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1553         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1554     }
1555     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1556 
1557     if (kvm_enabled()) {
1558         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1559 
1560         /* Check our RMA fits in the possible VRMA */
1561         if (vrma_limit < spapr->rma_size) {
1562             error_report("Unable to create %" HWADDR_PRIu
1563                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1564                          spapr->rma_size / MiB, vrma_limit / MiB);
1565             exit(EXIT_FAILURE);
1566         }
1567     }
1568 }
1569 
1570 static void spapr_machine_reset(MachineState *machine)
1571 {
1572     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1573     PowerPCCPU *first_ppc_cpu;
1574     hwaddr fdt_addr;
1575     void *fdt;
1576     int rc;
1577 
1578     pef_kvm_reset(machine->cgs, &error_fatal);
1579     spapr_caps_apply(spapr);
1580 
1581     first_ppc_cpu = POWERPC_CPU(first_cpu);
1582     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1583         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1584                               spapr->max_compat_pvr)) {
1585         /*
1586          * If using KVM with radix mode available, VCPUs can be started
1587          * without a HPT because KVM will start them in radix mode.
1588          * Set the GR bit in PATE so that we know there is no HPT.
1589          */
1590         spapr->patb_entry = PATE1_GR;
1591         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1592     } else {
1593         spapr_setup_hpt(spapr);
1594     }
1595 
1596     qemu_devices_reset();
1597 
1598     spapr_ovec_cleanup(spapr->ov5_cas);
1599     spapr->ov5_cas = spapr_ovec_new();
1600 
1601     ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1602 
1603     /*
1604      * This is fixing some of the default configuration of the XIVE
1605      * devices. To be called after the reset of the machine devices.
1606      */
1607     spapr_irq_reset(spapr, &error_fatal);
1608 
1609     /*
1610      * There is no CAS under qtest. Simulate one to please the code that
1611      * depends on spapr->ov5_cas. This is especially needed to test device
1612      * unplug, so we do that before resetting the DRCs.
1613      */
1614     if (qtest_enabled()) {
1615         spapr_ovec_cleanup(spapr->ov5_cas);
1616         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1617     }
1618 
1619     /* DRC reset may cause a device to be unplugged. This will cause troubles
1620      * if this device is used by another device (eg, a running vhost backend
1621      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1622      * situations, we reset DRCs after all devices have been reset.
1623      */
1624     spapr_drc_reset_all(spapr);
1625 
1626     spapr_clear_pending_events(spapr);
1627 
1628     /*
1629      * We place the device tree and RTAS just below either the top of the RMA,
1630      * or just below 2GB, whichever is lower, so that it can be
1631      * processed with 32-bit real mode code if necessary
1632      */
1633     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1634 
1635     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1636 
1637     rc = fdt_pack(fdt);
1638 
1639     /* Should only fail if we've built a corrupted tree */
1640     assert(rc == 0);
1641 
1642     /* Load the fdt */
1643     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1644     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1645     g_free(spapr->fdt_blob);
1646     spapr->fdt_size = fdt_totalsize(fdt);
1647     spapr->fdt_initial_size = spapr->fdt_size;
1648     spapr->fdt_blob = fdt;
1649 
1650     /* Set up the entry state */
1651     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, 0, fdt_addr, 0);
1652     first_ppc_cpu->env.gpr[5] = 0;
1653 
1654     spapr->fwnmi_system_reset_addr = -1;
1655     spapr->fwnmi_machine_check_addr = -1;
1656     spapr->fwnmi_machine_check_interlock = -1;
1657 
1658     /* Signal all vCPUs waiting on this condition */
1659     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1660 
1661     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1662 }
1663 
1664 static void spapr_create_nvram(SpaprMachineState *spapr)
1665 {
1666     DeviceState *dev = qdev_new("spapr-nvram");
1667     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1668 
1669     if (dinfo) {
1670         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1671                                 &error_fatal);
1672     }
1673 
1674     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1675 
1676     spapr->nvram = (struct SpaprNvram *)dev;
1677 }
1678 
1679 static void spapr_rtc_create(SpaprMachineState *spapr)
1680 {
1681     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1682                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1683                                        &error_fatal, NULL);
1684     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1685     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1686                               "date");
1687 }
1688 
1689 /* Returns whether we want to use VGA or not */
1690 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1691 {
1692     switch (vga_interface_type) {
1693     case VGA_NONE:
1694         return false;
1695     case VGA_DEVICE:
1696         return true;
1697     case VGA_STD:
1698     case VGA_VIRTIO:
1699     case VGA_CIRRUS:
1700         return pci_vga_init(pci_bus) != NULL;
1701     default:
1702         error_setg(errp,
1703                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1704         return false;
1705     }
1706 }
1707 
1708 static int spapr_pre_load(void *opaque)
1709 {
1710     int rc;
1711 
1712     rc = spapr_caps_pre_load(opaque);
1713     if (rc) {
1714         return rc;
1715     }
1716 
1717     return 0;
1718 }
1719 
1720 static int spapr_post_load(void *opaque, int version_id)
1721 {
1722     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1723     int err = 0;
1724 
1725     err = spapr_caps_post_migration(spapr);
1726     if (err) {
1727         return err;
1728     }
1729 
1730     /*
1731      * In earlier versions, there was no separate qdev for the PAPR
1732      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1733      * So when migrating from those versions, poke the incoming offset
1734      * value into the RTC device
1735      */
1736     if (version_id < 3) {
1737         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1738         if (err) {
1739             return err;
1740         }
1741     }
1742 
1743     if (kvm_enabled() && spapr->patb_entry) {
1744         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1745         bool radix = !!(spapr->patb_entry & PATE1_GR);
1746         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1747 
1748         /*
1749          * Update LPCR:HR and UPRT as they may not be set properly in
1750          * the stream
1751          */
1752         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1753                             LPCR_HR | LPCR_UPRT);
1754 
1755         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1756         if (err) {
1757             error_report("Process table config unsupported by the host");
1758             return -EINVAL;
1759         }
1760     }
1761 
1762     err = spapr_irq_post_load(spapr, version_id);
1763     if (err) {
1764         return err;
1765     }
1766 
1767     return err;
1768 }
1769 
1770 static int spapr_pre_save(void *opaque)
1771 {
1772     int rc;
1773 
1774     rc = spapr_caps_pre_save(opaque);
1775     if (rc) {
1776         return rc;
1777     }
1778 
1779     return 0;
1780 }
1781 
1782 static bool version_before_3(void *opaque, int version_id)
1783 {
1784     return version_id < 3;
1785 }
1786 
1787 static bool spapr_pending_events_needed(void *opaque)
1788 {
1789     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1790     return !QTAILQ_EMPTY(&spapr->pending_events);
1791 }
1792 
1793 static const VMStateDescription vmstate_spapr_event_entry = {
1794     .name = "spapr_event_log_entry",
1795     .version_id = 1,
1796     .minimum_version_id = 1,
1797     .fields = (VMStateField[]) {
1798         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1799         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1800         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1801                                      NULL, extended_length),
1802         VMSTATE_END_OF_LIST()
1803     },
1804 };
1805 
1806 static const VMStateDescription vmstate_spapr_pending_events = {
1807     .name = "spapr_pending_events",
1808     .version_id = 1,
1809     .minimum_version_id = 1,
1810     .needed = spapr_pending_events_needed,
1811     .fields = (VMStateField[]) {
1812         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1813                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1814         VMSTATE_END_OF_LIST()
1815     },
1816 };
1817 
1818 static bool spapr_ov5_cas_needed(void *opaque)
1819 {
1820     SpaprMachineState *spapr = opaque;
1821     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1822     bool cas_needed;
1823 
1824     /* Prior to the introduction of SpaprOptionVector, we had two option
1825      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1826      * Both of these options encode machine topology into the device-tree
1827      * in such a way that the now-booted OS should still be able to interact
1828      * appropriately with QEMU regardless of what options were actually
1829      * negotiatied on the source side.
1830      *
1831      * As such, we can avoid migrating the CAS-negotiated options if these
1832      * are the only options available on the current machine/platform.
1833      * Since these are the only options available for pseries-2.7 and
1834      * earlier, this allows us to maintain old->new/new->old migration
1835      * compatibility.
1836      *
1837      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1838      * via default pseries-2.8 machines and explicit command-line parameters.
1839      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1840      * of the actual CAS-negotiated values to continue working properly. For
1841      * example, availability of memory unplug depends on knowing whether
1842      * OV5_HP_EVT was negotiated via CAS.
1843      *
1844      * Thus, for any cases where the set of available CAS-negotiatable
1845      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1846      * include the CAS-negotiated options in the migration stream, unless
1847      * if they affect boot time behaviour only.
1848      */
1849     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1850     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1851     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1852 
1853     /* We need extra information if we have any bits outside the mask
1854      * defined above */
1855     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1856 
1857     spapr_ovec_cleanup(ov5_mask);
1858 
1859     return cas_needed;
1860 }
1861 
1862 static const VMStateDescription vmstate_spapr_ov5_cas = {
1863     .name = "spapr_option_vector_ov5_cas",
1864     .version_id = 1,
1865     .minimum_version_id = 1,
1866     .needed = spapr_ov5_cas_needed,
1867     .fields = (VMStateField[]) {
1868         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1869                                  vmstate_spapr_ovec, SpaprOptionVector),
1870         VMSTATE_END_OF_LIST()
1871     },
1872 };
1873 
1874 static bool spapr_patb_entry_needed(void *opaque)
1875 {
1876     SpaprMachineState *spapr = opaque;
1877 
1878     return !!spapr->patb_entry;
1879 }
1880 
1881 static const VMStateDescription vmstate_spapr_patb_entry = {
1882     .name = "spapr_patb_entry",
1883     .version_id = 1,
1884     .minimum_version_id = 1,
1885     .needed = spapr_patb_entry_needed,
1886     .fields = (VMStateField[]) {
1887         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1888         VMSTATE_END_OF_LIST()
1889     },
1890 };
1891 
1892 static bool spapr_irq_map_needed(void *opaque)
1893 {
1894     SpaprMachineState *spapr = opaque;
1895 
1896     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1897 }
1898 
1899 static const VMStateDescription vmstate_spapr_irq_map = {
1900     .name = "spapr_irq_map",
1901     .version_id = 1,
1902     .minimum_version_id = 1,
1903     .needed = spapr_irq_map_needed,
1904     .fields = (VMStateField[]) {
1905         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1906         VMSTATE_END_OF_LIST()
1907     },
1908 };
1909 
1910 static bool spapr_dtb_needed(void *opaque)
1911 {
1912     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1913 
1914     return smc->update_dt_enabled;
1915 }
1916 
1917 static int spapr_dtb_pre_load(void *opaque)
1918 {
1919     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1920 
1921     g_free(spapr->fdt_blob);
1922     spapr->fdt_blob = NULL;
1923     spapr->fdt_size = 0;
1924 
1925     return 0;
1926 }
1927 
1928 static const VMStateDescription vmstate_spapr_dtb = {
1929     .name = "spapr_dtb",
1930     .version_id = 1,
1931     .minimum_version_id = 1,
1932     .needed = spapr_dtb_needed,
1933     .pre_load = spapr_dtb_pre_load,
1934     .fields = (VMStateField[]) {
1935         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1936         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1937         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1938                                      fdt_size),
1939         VMSTATE_END_OF_LIST()
1940     },
1941 };
1942 
1943 static bool spapr_fwnmi_needed(void *opaque)
1944 {
1945     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1946 
1947     return spapr->fwnmi_machine_check_addr != -1;
1948 }
1949 
1950 static int spapr_fwnmi_pre_save(void *opaque)
1951 {
1952     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1953 
1954     /*
1955      * Check if machine check handling is in progress and print a
1956      * warning message.
1957      */
1958     if (spapr->fwnmi_machine_check_interlock != -1) {
1959         warn_report("A machine check is being handled during migration. The"
1960                 "handler may run and log hardware error on the destination");
1961     }
1962 
1963     return 0;
1964 }
1965 
1966 static const VMStateDescription vmstate_spapr_fwnmi = {
1967     .name = "spapr_fwnmi",
1968     .version_id = 1,
1969     .minimum_version_id = 1,
1970     .needed = spapr_fwnmi_needed,
1971     .pre_save = spapr_fwnmi_pre_save,
1972     .fields = (VMStateField[]) {
1973         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
1974         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
1975         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
1976         VMSTATE_END_OF_LIST()
1977     },
1978 };
1979 
1980 static const VMStateDescription vmstate_spapr = {
1981     .name = "spapr",
1982     .version_id = 3,
1983     .minimum_version_id = 1,
1984     .pre_load = spapr_pre_load,
1985     .post_load = spapr_post_load,
1986     .pre_save = spapr_pre_save,
1987     .fields = (VMStateField[]) {
1988         /* used to be @next_irq */
1989         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1990 
1991         /* RTC offset */
1992         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
1993 
1994         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
1995         VMSTATE_END_OF_LIST()
1996     },
1997     .subsections = (const VMStateDescription*[]) {
1998         &vmstate_spapr_ov5_cas,
1999         &vmstate_spapr_patb_entry,
2000         &vmstate_spapr_pending_events,
2001         &vmstate_spapr_cap_htm,
2002         &vmstate_spapr_cap_vsx,
2003         &vmstate_spapr_cap_dfp,
2004         &vmstate_spapr_cap_cfpc,
2005         &vmstate_spapr_cap_sbbc,
2006         &vmstate_spapr_cap_ibs,
2007         &vmstate_spapr_cap_hpt_maxpagesize,
2008         &vmstate_spapr_irq_map,
2009         &vmstate_spapr_cap_nested_kvm_hv,
2010         &vmstate_spapr_dtb,
2011         &vmstate_spapr_cap_large_decr,
2012         &vmstate_spapr_cap_ccf_assist,
2013         &vmstate_spapr_cap_fwnmi,
2014         &vmstate_spapr_fwnmi,
2015         NULL
2016     }
2017 };
2018 
2019 static int htab_save_setup(QEMUFile *f, void *opaque)
2020 {
2021     SpaprMachineState *spapr = opaque;
2022 
2023     /* "Iteration" header */
2024     if (!spapr->htab_shift) {
2025         qemu_put_be32(f, -1);
2026     } else {
2027         qemu_put_be32(f, spapr->htab_shift);
2028     }
2029 
2030     if (spapr->htab) {
2031         spapr->htab_save_index = 0;
2032         spapr->htab_first_pass = true;
2033     } else {
2034         if (spapr->htab_shift) {
2035             assert(kvm_enabled());
2036         }
2037     }
2038 
2039 
2040     return 0;
2041 }
2042 
2043 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2044                             int chunkstart, int n_valid, int n_invalid)
2045 {
2046     qemu_put_be32(f, chunkstart);
2047     qemu_put_be16(f, n_valid);
2048     qemu_put_be16(f, n_invalid);
2049     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2050                     HASH_PTE_SIZE_64 * n_valid);
2051 }
2052 
2053 static void htab_save_end_marker(QEMUFile *f)
2054 {
2055     qemu_put_be32(f, 0);
2056     qemu_put_be16(f, 0);
2057     qemu_put_be16(f, 0);
2058 }
2059 
2060 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2061                                  int64_t max_ns)
2062 {
2063     bool has_timeout = max_ns != -1;
2064     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2065     int index = spapr->htab_save_index;
2066     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2067 
2068     assert(spapr->htab_first_pass);
2069 
2070     do {
2071         int chunkstart;
2072 
2073         /* Consume invalid HPTEs */
2074         while ((index < htabslots)
2075                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2076             CLEAN_HPTE(HPTE(spapr->htab, index));
2077             index++;
2078         }
2079 
2080         /* Consume valid HPTEs */
2081         chunkstart = index;
2082         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2083                && HPTE_VALID(HPTE(spapr->htab, index))) {
2084             CLEAN_HPTE(HPTE(spapr->htab, index));
2085             index++;
2086         }
2087 
2088         if (index > chunkstart) {
2089             int n_valid = index - chunkstart;
2090 
2091             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2092 
2093             if (has_timeout &&
2094                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2095                 break;
2096             }
2097         }
2098     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2099 
2100     if (index >= htabslots) {
2101         assert(index == htabslots);
2102         index = 0;
2103         spapr->htab_first_pass = false;
2104     }
2105     spapr->htab_save_index = index;
2106 }
2107 
2108 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2109                                 int64_t max_ns)
2110 {
2111     bool final = max_ns < 0;
2112     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2113     int examined = 0, sent = 0;
2114     int index = spapr->htab_save_index;
2115     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2116 
2117     assert(!spapr->htab_first_pass);
2118 
2119     do {
2120         int chunkstart, invalidstart;
2121 
2122         /* Consume non-dirty HPTEs */
2123         while ((index < htabslots)
2124                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2125             index++;
2126             examined++;
2127         }
2128 
2129         chunkstart = index;
2130         /* Consume valid dirty HPTEs */
2131         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2132                && HPTE_DIRTY(HPTE(spapr->htab, index))
2133                && HPTE_VALID(HPTE(spapr->htab, index))) {
2134             CLEAN_HPTE(HPTE(spapr->htab, index));
2135             index++;
2136             examined++;
2137         }
2138 
2139         invalidstart = index;
2140         /* Consume invalid dirty HPTEs */
2141         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2142                && HPTE_DIRTY(HPTE(spapr->htab, index))
2143                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2144             CLEAN_HPTE(HPTE(spapr->htab, index));
2145             index++;
2146             examined++;
2147         }
2148 
2149         if (index > chunkstart) {
2150             int n_valid = invalidstart - chunkstart;
2151             int n_invalid = index - invalidstart;
2152 
2153             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2154             sent += index - chunkstart;
2155 
2156             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2157                 break;
2158             }
2159         }
2160 
2161         if (examined >= htabslots) {
2162             break;
2163         }
2164 
2165         if (index >= htabslots) {
2166             assert(index == htabslots);
2167             index = 0;
2168         }
2169     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2170 
2171     if (index >= htabslots) {
2172         assert(index == htabslots);
2173         index = 0;
2174     }
2175 
2176     spapr->htab_save_index = index;
2177 
2178     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2179 }
2180 
2181 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2182 #define MAX_KVM_BUF_SIZE    2048
2183 
2184 static int htab_save_iterate(QEMUFile *f, void *opaque)
2185 {
2186     SpaprMachineState *spapr = opaque;
2187     int fd;
2188     int rc = 0;
2189 
2190     /* Iteration header */
2191     if (!spapr->htab_shift) {
2192         qemu_put_be32(f, -1);
2193         return 1;
2194     } else {
2195         qemu_put_be32(f, 0);
2196     }
2197 
2198     if (!spapr->htab) {
2199         assert(kvm_enabled());
2200 
2201         fd = get_htab_fd(spapr);
2202         if (fd < 0) {
2203             return fd;
2204         }
2205 
2206         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2207         if (rc < 0) {
2208             return rc;
2209         }
2210     } else  if (spapr->htab_first_pass) {
2211         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2212     } else {
2213         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2214     }
2215 
2216     htab_save_end_marker(f);
2217 
2218     return rc;
2219 }
2220 
2221 static int htab_save_complete(QEMUFile *f, void *opaque)
2222 {
2223     SpaprMachineState *spapr = opaque;
2224     int fd;
2225 
2226     /* Iteration header */
2227     if (!spapr->htab_shift) {
2228         qemu_put_be32(f, -1);
2229         return 0;
2230     } else {
2231         qemu_put_be32(f, 0);
2232     }
2233 
2234     if (!spapr->htab) {
2235         int rc;
2236 
2237         assert(kvm_enabled());
2238 
2239         fd = get_htab_fd(spapr);
2240         if (fd < 0) {
2241             return fd;
2242         }
2243 
2244         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2245         if (rc < 0) {
2246             return rc;
2247         }
2248     } else {
2249         if (spapr->htab_first_pass) {
2250             htab_save_first_pass(f, spapr, -1);
2251         }
2252         htab_save_later_pass(f, spapr, -1);
2253     }
2254 
2255     /* End marker */
2256     htab_save_end_marker(f);
2257 
2258     return 0;
2259 }
2260 
2261 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2262 {
2263     SpaprMachineState *spapr = opaque;
2264     uint32_t section_hdr;
2265     int fd = -1;
2266     Error *local_err = NULL;
2267 
2268     if (version_id < 1 || version_id > 1) {
2269         error_report("htab_load() bad version");
2270         return -EINVAL;
2271     }
2272 
2273     section_hdr = qemu_get_be32(f);
2274 
2275     if (section_hdr == -1) {
2276         spapr_free_hpt(spapr);
2277         return 0;
2278     }
2279 
2280     if (section_hdr) {
2281         int ret;
2282 
2283         /* First section gives the htab size */
2284         ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2285         if (ret < 0) {
2286             error_report_err(local_err);
2287             return ret;
2288         }
2289         return 0;
2290     }
2291 
2292     if (!spapr->htab) {
2293         assert(kvm_enabled());
2294 
2295         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2296         if (fd < 0) {
2297             error_report_err(local_err);
2298             return fd;
2299         }
2300     }
2301 
2302     while (true) {
2303         uint32_t index;
2304         uint16_t n_valid, n_invalid;
2305 
2306         index = qemu_get_be32(f);
2307         n_valid = qemu_get_be16(f);
2308         n_invalid = qemu_get_be16(f);
2309 
2310         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2311             /* End of Stream */
2312             break;
2313         }
2314 
2315         if ((index + n_valid + n_invalid) >
2316             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2317             /* Bad index in stream */
2318             error_report(
2319                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2320                 index, n_valid, n_invalid, spapr->htab_shift);
2321             return -EINVAL;
2322         }
2323 
2324         if (spapr->htab) {
2325             if (n_valid) {
2326                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2327                                 HASH_PTE_SIZE_64 * n_valid);
2328             }
2329             if (n_invalid) {
2330                 memset(HPTE(spapr->htab, index + n_valid), 0,
2331                        HASH_PTE_SIZE_64 * n_invalid);
2332             }
2333         } else {
2334             int rc;
2335 
2336             assert(fd >= 0);
2337 
2338             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2339                                         &local_err);
2340             if (rc < 0) {
2341                 error_report_err(local_err);
2342                 return rc;
2343             }
2344         }
2345     }
2346 
2347     if (!spapr->htab) {
2348         assert(fd >= 0);
2349         close(fd);
2350     }
2351 
2352     return 0;
2353 }
2354 
2355 static void htab_save_cleanup(void *opaque)
2356 {
2357     SpaprMachineState *spapr = opaque;
2358 
2359     close_htab_fd(spapr);
2360 }
2361 
2362 static SaveVMHandlers savevm_htab_handlers = {
2363     .save_setup = htab_save_setup,
2364     .save_live_iterate = htab_save_iterate,
2365     .save_live_complete_precopy = htab_save_complete,
2366     .save_cleanup = htab_save_cleanup,
2367     .load_state = htab_load,
2368 };
2369 
2370 static void spapr_boot_set(void *opaque, const char *boot_device,
2371                            Error **errp)
2372 {
2373     MachineState *machine = MACHINE(opaque);
2374     machine->boot_order = g_strdup(boot_device);
2375 }
2376 
2377 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2378 {
2379     MachineState *machine = MACHINE(spapr);
2380     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2381     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2382     int i;
2383 
2384     for (i = 0; i < nr_lmbs; i++) {
2385         uint64_t addr;
2386 
2387         addr = i * lmb_size + machine->device_memory->base;
2388         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2389                                addr / lmb_size);
2390     }
2391 }
2392 
2393 /*
2394  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2395  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2396  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2397  */
2398 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2399 {
2400     int i;
2401 
2402     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2403         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2404                    " is not aligned to %" PRIu64 " MiB",
2405                    machine->ram_size,
2406                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2407         return;
2408     }
2409 
2410     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2411         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2412                    " is not aligned to %" PRIu64 " MiB",
2413                    machine->ram_size,
2414                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2415         return;
2416     }
2417 
2418     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2419         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2420             error_setg(errp,
2421                        "Node %d memory size 0x%" PRIx64
2422                        " is not aligned to %" PRIu64 " MiB",
2423                        i, machine->numa_state->nodes[i].node_mem,
2424                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2425             return;
2426         }
2427     }
2428 }
2429 
2430 /* find cpu slot in machine->possible_cpus by core_id */
2431 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2432 {
2433     int index = id / ms->smp.threads;
2434 
2435     if (index >= ms->possible_cpus->len) {
2436         return NULL;
2437     }
2438     if (idx) {
2439         *idx = index;
2440     }
2441     return &ms->possible_cpus->cpus[index];
2442 }
2443 
2444 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2445 {
2446     MachineState *ms = MACHINE(spapr);
2447     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2448     Error *local_err = NULL;
2449     bool vsmt_user = !!spapr->vsmt;
2450     int kvm_smt = kvmppc_smt_threads();
2451     int ret;
2452     unsigned int smp_threads = ms->smp.threads;
2453 
2454     if (!kvm_enabled() && (smp_threads > 1)) {
2455         error_setg(errp, "TCG cannot support more than 1 thread/core "
2456                    "on a pseries machine");
2457         return;
2458     }
2459     if (!is_power_of_2(smp_threads)) {
2460         error_setg(errp, "Cannot support %d threads/core on a pseries "
2461                    "machine because it must be a power of 2", smp_threads);
2462         return;
2463     }
2464 
2465     /* Detemine the VSMT mode to use: */
2466     if (vsmt_user) {
2467         if (spapr->vsmt < smp_threads) {
2468             error_setg(errp, "Cannot support VSMT mode %d"
2469                        " because it must be >= threads/core (%d)",
2470                        spapr->vsmt, smp_threads);
2471             return;
2472         }
2473         /* In this case, spapr->vsmt has been set by the command line */
2474     } else if (!smc->smp_threads_vsmt) {
2475         /*
2476          * Default VSMT value is tricky, because we need it to be as
2477          * consistent as possible (for migration), but this requires
2478          * changing it for at least some existing cases.  We pick 8 as
2479          * the value that we'd get with KVM on POWER8, the
2480          * overwhelmingly common case in production systems.
2481          */
2482         spapr->vsmt = MAX(8, smp_threads);
2483     } else {
2484         spapr->vsmt = smp_threads;
2485     }
2486 
2487     /* KVM: If necessary, set the SMT mode: */
2488     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2489         ret = kvmppc_set_smt_threads(spapr->vsmt);
2490         if (ret) {
2491             /* Looks like KVM isn't able to change VSMT mode */
2492             error_setg(&local_err,
2493                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2494                        spapr->vsmt, ret);
2495             /* We can live with that if the default one is big enough
2496              * for the number of threads, and a submultiple of the one
2497              * we want.  In this case we'll waste some vcpu ids, but
2498              * behaviour will be correct */
2499             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2500                 warn_report_err(local_err);
2501             } else {
2502                 if (!vsmt_user) {
2503                     error_append_hint(&local_err,
2504                                       "On PPC, a VM with %d threads/core"
2505                                       " on a host with %d threads/core"
2506                                       " requires the use of VSMT mode %d.\n",
2507                                       smp_threads, kvm_smt, spapr->vsmt);
2508                 }
2509                 kvmppc_error_append_smt_possible_hint(&local_err);
2510                 error_propagate(errp, local_err);
2511             }
2512         }
2513     }
2514     /* else TCG: nothing to do currently */
2515 }
2516 
2517 static void spapr_init_cpus(SpaprMachineState *spapr)
2518 {
2519     MachineState *machine = MACHINE(spapr);
2520     MachineClass *mc = MACHINE_GET_CLASS(machine);
2521     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2522     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2523     const CPUArchIdList *possible_cpus;
2524     unsigned int smp_cpus = machine->smp.cpus;
2525     unsigned int smp_threads = machine->smp.threads;
2526     unsigned int max_cpus = machine->smp.max_cpus;
2527     int boot_cores_nr = smp_cpus / smp_threads;
2528     int i;
2529 
2530     possible_cpus = mc->possible_cpu_arch_ids(machine);
2531     if (mc->has_hotpluggable_cpus) {
2532         if (smp_cpus % smp_threads) {
2533             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2534                          smp_cpus, smp_threads);
2535             exit(1);
2536         }
2537         if (max_cpus % smp_threads) {
2538             error_report("max_cpus (%u) must be multiple of threads (%u)",
2539                          max_cpus, smp_threads);
2540             exit(1);
2541         }
2542     } else {
2543         if (max_cpus != smp_cpus) {
2544             error_report("This machine version does not support CPU hotplug");
2545             exit(1);
2546         }
2547         boot_cores_nr = possible_cpus->len;
2548     }
2549 
2550     if (smc->pre_2_10_has_unused_icps) {
2551         int i;
2552 
2553         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2554             /* Dummy entries get deregistered when real ICPState objects
2555              * are registered during CPU core hotplug.
2556              */
2557             pre_2_10_vmstate_register_dummy_icp(i);
2558         }
2559     }
2560 
2561     for (i = 0; i < possible_cpus->len; i++) {
2562         int core_id = i * smp_threads;
2563 
2564         if (mc->has_hotpluggable_cpus) {
2565             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2566                                    spapr_vcpu_id(spapr, core_id));
2567         }
2568 
2569         if (i < boot_cores_nr) {
2570             Object *core  = object_new(type);
2571             int nr_threads = smp_threads;
2572 
2573             /* Handle the partially filled core for older machine types */
2574             if ((i + 1) * smp_threads >= smp_cpus) {
2575                 nr_threads = smp_cpus - i * smp_threads;
2576             }
2577 
2578             object_property_set_int(core, "nr-threads", nr_threads,
2579                                     &error_fatal);
2580             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2581                                     &error_fatal);
2582             qdev_realize(DEVICE(core), NULL, &error_fatal);
2583 
2584             object_unref(core);
2585         }
2586     }
2587 }
2588 
2589 static PCIHostState *spapr_create_default_phb(void)
2590 {
2591     DeviceState *dev;
2592 
2593     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2594     qdev_prop_set_uint32(dev, "index", 0);
2595     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2596 
2597     return PCI_HOST_BRIDGE(dev);
2598 }
2599 
2600 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2601 {
2602     MachineState *machine = MACHINE(spapr);
2603     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2604     hwaddr rma_size = machine->ram_size;
2605     hwaddr node0_size = spapr_node0_size(machine);
2606 
2607     /* RMA has to fit in the first NUMA node */
2608     rma_size = MIN(rma_size, node0_size);
2609 
2610     /*
2611      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2612      * never exceed that
2613      */
2614     rma_size = MIN(rma_size, 1 * TiB);
2615 
2616     /*
2617      * Clamp the RMA size based on machine type.  This is for
2618      * migration compatibility with older qemu versions, which limited
2619      * the RMA size for complicated and mostly bad reasons.
2620      */
2621     if (smc->rma_limit) {
2622         rma_size = MIN(rma_size, smc->rma_limit);
2623     }
2624 
2625     if (rma_size < MIN_RMA_SLOF) {
2626         error_setg(errp,
2627                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2628                    "ldMiB guest RMA (Real Mode Area memory)",
2629                    MIN_RMA_SLOF / MiB);
2630         return 0;
2631     }
2632 
2633     return rma_size;
2634 }
2635 
2636 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2637 {
2638     MachineState *machine = MACHINE(spapr);
2639     int i;
2640 
2641     for (i = 0; i < machine->ram_slots; i++) {
2642         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2643     }
2644 }
2645 
2646 /* pSeries LPAR / sPAPR hardware init */
2647 static void spapr_machine_init(MachineState *machine)
2648 {
2649     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2650     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2651     MachineClass *mc = MACHINE_GET_CLASS(machine);
2652     const char *bios_name = machine->firmware ?: FW_FILE_NAME;
2653     const char *kernel_filename = machine->kernel_filename;
2654     const char *initrd_filename = machine->initrd_filename;
2655     PCIHostState *phb;
2656     int i;
2657     MemoryRegion *sysmem = get_system_memory();
2658     long load_limit, fw_size;
2659     char *filename;
2660     Error *resize_hpt_err = NULL;
2661 
2662     /*
2663      * if Secure VM (PEF) support is configured, then initialize it
2664      */
2665     pef_kvm_init(machine->cgs, &error_fatal);
2666 
2667     msi_nonbroken = true;
2668 
2669     QLIST_INIT(&spapr->phbs);
2670     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2671 
2672     /* Determine capabilities to run with */
2673     spapr_caps_init(spapr);
2674 
2675     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2676     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2677         /*
2678          * If the user explicitly requested a mode we should either
2679          * supply it, or fail completely (which we do below).  But if
2680          * it's not set explicitly, we reset our mode to something
2681          * that works
2682          */
2683         if (resize_hpt_err) {
2684             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2685             error_free(resize_hpt_err);
2686             resize_hpt_err = NULL;
2687         } else {
2688             spapr->resize_hpt = smc->resize_hpt_default;
2689         }
2690     }
2691 
2692     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2693 
2694     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2695         /*
2696          * User requested HPT resize, but this host can't supply it.  Bail out
2697          */
2698         error_report_err(resize_hpt_err);
2699         exit(1);
2700     }
2701     error_free(resize_hpt_err);
2702 
2703     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2704 
2705     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2706     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2707 
2708     /*
2709      * VSMT must be set in order to be able to compute VCPU ids, ie to
2710      * call spapr_max_server_number() or spapr_vcpu_id().
2711      */
2712     spapr_set_vsmt_mode(spapr, &error_fatal);
2713 
2714     /* Set up Interrupt Controller before we create the VCPUs */
2715     spapr_irq_init(spapr, &error_fatal);
2716 
2717     /* Set up containers for ibm,client-architecture-support negotiated options
2718      */
2719     spapr->ov5 = spapr_ovec_new();
2720     spapr->ov5_cas = spapr_ovec_new();
2721 
2722     if (smc->dr_lmb_enabled) {
2723         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2724         spapr_validate_node_memory(machine, &error_fatal);
2725     }
2726 
2727     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2728 
2729     /* advertise support for dedicated HP event source to guests */
2730     if (spapr->use_hotplug_event_source) {
2731         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2732     }
2733 
2734     /* advertise support for HPT resizing */
2735     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2736         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2737     }
2738 
2739     /* advertise support for ibm,dyamic-memory-v2 */
2740     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2741 
2742     /* advertise XIVE on POWER9 machines */
2743     if (spapr->irq->xive) {
2744         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2745     }
2746 
2747     /* init CPUs */
2748     spapr_init_cpus(spapr);
2749 
2750     /*
2751      * check we don't have a memory-less/cpu-less NUMA node
2752      * Firmware relies on the existing memory/cpu topology to provide the
2753      * NUMA topology to the kernel.
2754      * And the linux kernel needs to know the NUMA topology at start
2755      * to be able to hotplug CPUs later.
2756      */
2757     if (machine->numa_state->num_nodes) {
2758         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2759             /* check for memory-less node */
2760             if (machine->numa_state->nodes[i].node_mem == 0) {
2761                 CPUState *cs;
2762                 int found = 0;
2763                 /* check for cpu-less node */
2764                 CPU_FOREACH(cs) {
2765                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2766                     if (cpu->node_id == i) {
2767                         found = 1;
2768                         break;
2769                     }
2770                 }
2771                 /* memory-less and cpu-less node */
2772                 if (!found) {
2773                     error_report(
2774                        "Memory-less/cpu-less nodes are not supported (node %d)",
2775                                  i);
2776                     exit(1);
2777                 }
2778             }
2779         }
2780 
2781     }
2782 
2783     /*
2784      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2785      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2786      * called from vPHB reset handler so we initialize the counter here.
2787      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2788      * must be equally distant from any other node.
2789      * The final value of spapr->gpu_numa_id is going to be written to
2790      * max-associativity-domains in spapr_build_fdt().
2791      */
2792     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2793 
2794     /* Init numa_assoc_array */
2795     spapr_numa_associativity_init(spapr, machine);
2796 
2797     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2798         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2799                               spapr->max_compat_pvr)) {
2800         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2801         /* KVM and TCG always allow GTSE with radix... */
2802         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2803     }
2804     /* ... but not with hash (currently). */
2805 
2806     if (kvm_enabled()) {
2807         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2808         kvmppc_enable_logical_ci_hcalls();
2809         kvmppc_enable_set_mode_hcall();
2810 
2811         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2812         kvmppc_enable_clear_ref_mod_hcalls();
2813 
2814         /* Enable H_PAGE_INIT */
2815         kvmppc_enable_h_page_init();
2816     }
2817 
2818     /* map RAM */
2819     memory_region_add_subregion(sysmem, 0, machine->ram);
2820 
2821     /* always allocate the device memory information */
2822     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2823 
2824     /* initialize hotplug memory address space */
2825     if (machine->ram_size < machine->maxram_size) {
2826         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2827         /*
2828          * Limit the number of hotpluggable memory slots to half the number
2829          * slots that KVM supports, leaving the other half for PCI and other
2830          * devices. However ensure that number of slots doesn't drop below 32.
2831          */
2832         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2833                            SPAPR_MAX_RAM_SLOTS;
2834 
2835         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2836             max_memslots = SPAPR_MAX_RAM_SLOTS;
2837         }
2838         if (machine->ram_slots > max_memslots) {
2839             error_report("Specified number of memory slots %"
2840                          PRIu64" exceeds max supported %d",
2841                          machine->ram_slots, max_memslots);
2842             exit(1);
2843         }
2844 
2845         machine->device_memory->base = ROUND_UP(machine->ram_size,
2846                                                 SPAPR_DEVICE_MEM_ALIGN);
2847         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2848                            "device-memory", device_mem_size);
2849         memory_region_add_subregion(sysmem, machine->device_memory->base,
2850                                     &machine->device_memory->mr);
2851     }
2852 
2853     if (smc->dr_lmb_enabled) {
2854         spapr_create_lmb_dr_connectors(spapr);
2855     }
2856 
2857     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2858         /* Create the error string for live migration blocker */
2859         error_setg(&spapr->fwnmi_migration_blocker,
2860             "A machine check is being handled during migration. The handler"
2861             "may run and log hardware error on the destination");
2862     }
2863 
2864     if (mc->nvdimm_supported) {
2865         spapr_create_nvdimm_dr_connectors(spapr);
2866     }
2867 
2868     /* Set up RTAS event infrastructure */
2869     spapr_events_init(spapr);
2870 
2871     /* Set up the RTC RTAS interfaces */
2872     spapr_rtc_create(spapr);
2873 
2874     /* Set up VIO bus */
2875     spapr->vio_bus = spapr_vio_bus_init();
2876 
2877     for (i = 0; serial_hd(i); i++) {
2878         spapr_vty_create(spapr->vio_bus, serial_hd(i));
2879     }
2880 
2881     /* We always have at least the nvram device on VIO */
2882     spapr_create_nvram(spapr);
2883 
2884     /*
2885      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2886      * connectors (described in root DT node's "ibm,drc-types" property)
2887      * are pre-initialized here. additional child connectors (such as
2888      * connectors for a PHBs PCI slots) are added as needed during their
2889      * parent's realization.
2890      */
2891     if (smc->dr_phb_enabled) {
2892         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2893             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2894         }
2895     }
2896 
2897     /* Set up PCI */
2898     spapr_pci_rtas_init();
2899 
2900     phb = spapr_create_default_phb();
2901 
2902     for (i = 0; i < nb_nics; i++) {
2903         NICInfo *nd = &nd_table[i];
2904 
2905         if (!nd->model) {
2906             nd->model = g_strdup("spapr-vlan");
2907         }
2908 
2909         if (g_str_equal(nd->model, "spapr-vlan") ||
2910             g_str_equal(nd->model, "ibmveth")) {
2911             spapr_vlan_create(spapr->vio_bus, nd);
2912         } else {
2913             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2914         }
2915     }
2916 
2917     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2918         spapr_vscsi_create(spapr->vio_bus);
2919     }
2920 
2921     /* Graphics */
2922     if (spapr_vga_init(phb->bus, &error_fatal)) {
2923         spapr->has_graphics = true;
2924         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2925     }
2926 
2927     if (machine->usb) {
2928         if (smc->use_ohci_by_default) {
2929             pci_create_simple(phb->bus, -1, "pci-ohci");
2930         } else {
2931             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2932         }
2933 
2934         if (spapr->has_graphics) {
2935             USBBus *usb_bus = usb_bus_find(-1);
2936 
2937             usb_create_simple(usb_bus, "usb-kbd");
2938             usb_create_simple(usb_bus, "usb-mouse");
2939         }
2940     }
2941 
2942     if (kernel_filename) {
2943         spapr->kernel_size = load_elf(kernel_filename, NULL,
2944                                       translate_kernel_address, spapr,
2945                                       NULL, NULL, NULL, NULL, 1,
2946                                       PPC_ELF_MACHINE, 0, 0);
2947         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2948             spapr->kernel_size = load_elf(kernel_filename, NULL,
2949                                           translate_kernel_address, spapr,
2950                                           NULL, NULL, NULL, NULL, 0,
2951                                           PPC_ELF_MACHINE, 0, 0);
2952             spapr->kernel_le = spapr->kernel_size > 0;
2953         }
2954         if (spapr->kernel_size < 0) {
2955             error_report("error loading %s: %s", kernel_filename,
2956                          load_elf_strerror(spapr->kernel_size));
2957             exit(1);
2958         }
2959 
2960         /* load initrd */
2961         if (initrd_filename) {
2962             /* Try to locate the initrd in the gap between the kernel
2963              * and the firmware. Add a bit of space just in case
2964              */
2965             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
2966                                   + 0x1ffff) & ~0xffff;
2967             spapr->initrd_size = load_image_targphys(initrd_filename,
2968                                                      spapr->initrd_base,
2969                                                      load_limit
2970                                                      - spapr->initrd_base);
2971             if (spapr->initrd_size < 0) {
2972                 error_report("could not load initial ram disk '%s'",
2973                              initrd_filename);
2974                 exit(1);
2975             }
2976         }
2977     }
2978 
2979     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2980     if (!filename) {
2981         error_report("Could not find LPAR firmware '%s'", bios_name);
2982         exit(1);
2983     }
2984     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2985     if (fw_size <= 0) {
2986         error_report("Could not load LPAR firmware '%s'", filename);
2987         exit(1);
2988     }
2989     g_free(filename);
2990 
2991     /* FIXME: Should register things through the MachineState's qdev
2992      * interface, this is a legacy from the sPAPREnvironment structure
2993      * which predated MachineState but had a similar function */
2994     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2995     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
2996                          &savevm_htab_handlers, spapr);
2997 
2998     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
2999 
3000     qemu_register_boot_set(spapr_boot_set, spapr);
3001 
3002     /*
3003      * Nothing needs to be done to resume a suspended guest because
3004      * suspending does not change the machine state, so no need for
3005      * a ->wakeup method.
3006      */
3007     qemu_register_wakeup_support();
3008 
3009     if (kvm_enabled()) {
3010         /* to stop and start vmclock */
3011         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3012                                          &spapr->tb);
3013 
3014         kvmppc_spapr_enable_inkernel_multitce();
3015     }
3016 
3017     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3018 }
3019 
3020 #define DEFAULT_KVM_TYPE "auto"
3021 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3022 {
3023     /*
3024      * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to
3025      * accomodate the 'HV' and 'PV' formats that exists in the
3026      * wild. The 'auto' mode is being introduced already as
3027      * lower-case, thus we don't need to bother checking for
3028      * "AUTO".
3029      */
3030     if (!vm_type || !strcmp(vm_type, DEFAULT_KVM_TYPE)) {
3031         return 0;
3032     }
3033 
3034     if (!g_ascii_strcasecmp(vm_type, "hv")) {
3035         return 1;
3036     }
3037 
3038     if (!g_ascii_strcasecmp(vm_type, "pr")) {
3039         return 2;
3040     }
3041 
3042     error_report("Unknown kvm-type specified '%s'", vm_type);
3043     exit(1);
3044 }
3045 
3046 /*
3047  * Implementation of an interface to adjust firmware path
3048  * for the bootindex property handling.
3049  */
3050 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3051                                    DeviceState *dev)
3052 {
3053 #define CAST(type, obj, name) \
3054     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3055     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3056     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3057     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3058 
3059     if (d) {
3060         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3061         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3062         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3063 
3064         if (spapr) {
3065             /*
3066              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3067              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3068              * 0x8000 | (target << 8) | (bus << 5) | lun
3069              * (see the "Logical unit addressing format" table in SAM5)
3070              */
3071             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3072             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3073                                    (uint64_t)id << 48);
3074         } else if (virtio) {
3075             /*
3076              * We use SRP luns of the form 01000000 | (target << 8) | lun
3077              * in the top 32 bits of the 64-bit LUN
3078              * Note: the quote above is from SLOF and it is wrong,
3079              * the actual binding is:
3080              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3081              */
3082             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3083             if (d->lun >= 256) {
3084                 /* Use the LUN "flat space addressing method" */
3085                 id |= 0x4000;
3086             }
3087             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3088                                    (uint64_t)id << 32);
3089         } else if (usb) {
3090             /*
3091              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3092              * in the top 32 bits of the 64-bit LUN
3093              */
3094             unsigned usb_port = atoi(usb->port->path);
3095             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3096             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3097                                    (uint64_t)id << 32);
3098         }
3099     }
3100 
3101     /*
3102      * SLOF probes the USB devices, and if it recognizes that the device is a
3103      * storage device, it changes its name to "storage" instead of "usb-host",
3104      * and additionally adds a child node for the SCSI LUN, so the correct
3105      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3106      */
3107     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3108         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3109         if (usb_host_dev_is_scsi_storage(usbdev)) {
3110             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3111         }
3112     }
3113 
3114     if (phb) {
3115         /* Replace "pci" with "pci@800000020000000" */
3116         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3117     }
3118 
3119     if (vsc) {
3120         /* Same logic as virtio above */
3121         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3122         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3123     }
3124 
3125     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3126         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3127         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3128         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3129     }
3130 
3131     return NULL;
3132 }
3133 
3134 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3135 {
3136     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3137 
3138     return g_strdup(spapr->kvm_type);
3139 }
3140 
3141 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3142 {
3143     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3144 
3145     g_free(spapr->kvm_type);
3146     spapr->kvm_type = g_strdup(value);
3147 }
3148 
3149 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3150 {
3151     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3152 
3153     return spapr->use_hotplug_event_source;
3154 }
3155 
3156 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3157                                             Error **errp)
3158 {
3159     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3160 
3161     spapr->use_hotplug_event_source = value;
3162 }
3163 
3164 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3165 {
3166     return true;
3167 }
3168 
3169 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3170 {
3171     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3172 
3173     switch (spapr->resize_hpt) {
3174     case SPAPR_RESIZE_HPT_DEFAULT:
3175         return g_strdup("default");
3176     case SPAPR_RESIZE_HPT_DISABLED:
3177         return g_strdup("disabled");
3178     case SPAPR_RESIZE_HPT_ENABLED:
3179         return g_strdup("enabled");
3180     case SPAPR_RESIZE_HPT_REQUIRED:
3181         return g_strdup("required");
3182     }
3183     g_assert_not_reached();
3184 }
3185 
3186 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3187 {
3188     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3189 
3190     if (strcmp(value, "default") == 0) {
3191         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3192     } else if (strcmp(value, "disabled") == 0) {
3193         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3194     } else if (strcmp(value, "enabled") == 0) {
3195         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3196     } else if (strcmp(value, "required") == 0) {
3197         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3198     } else {
3199         error_setg(errp, "Bad value for \"resize-hpt\" property");
3200     }
3201 }
3202 
3203 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3204 {
3205     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3206 
3207     if (spapr->irq == &spapr_irq_xics_legacy) {
3208         return g_strdup("legacy");
3209     } else if (spapr->irq == &spapr_irq_xics) {
3210         return g_strdup("xics");
3211     } else if (spapr->irq == &spapr_irq_xive) {
3212         return g_strdup("xive");
3213     } else if (spapr->irq == &spapr_irq_dual) {
3214         return g_strdup("dual");
3215     }
3216     g_assert_not_reached();
3217 }
3218 
3219 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3220 {
3221     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3222 
3223     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3224         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3225         return;
3226     }
3227 
3228     /* The legacy IRQ backend can not be set */
3229     if (strcmp(value, "xics") == 0) {
3230         spapr->irq = &spapr_irq_xics;
3231     } else if (strcmp(value, "xive") == 0) {
3232         spapr->irq = &spapr_irq_xive;
3233     } else if (strcmp(value, "dual") == 0) {
3234         spapr->irq = &spapr_irq_dual;
3235     } else {
3236         error_setg(errp, "Bad value for \"ic-mode\" property");
3237     }
3238 }
3239 
3240 static char *spapr_get_host_model(Object *obj, Error **errp)
3241 {
3242     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3243 
3244     return g_strdup(spapr->host_model);
3245 }
3246 
3247 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3248 {
3249     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3250 
3251     g_free(spapr->host_model);
3252     spapr->host_model = g_strdup(value);
3253 }
3254 
3255 static char *spapr_get_host_serial(Object *obj, Error **errp)
3256 {
3257     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3258 
3259     return g_strdup(spapr->host_serial);
3260 }
3261 
3262 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3263 {
3264     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3265 
3266     g_free(spapr->host_serial);
3267     spapr->host_serial = g_strdup(value);
3268 }
3269 
3270 static void spapr_instance_init(Object *obj)
3271 {
3272     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3273     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3274     MachineState *ms = MACHINE(spapr);
3275     MachineClass *mc = MACHINE_GET_CLASS(ms);
3276 
3277     /*
3278      * NVDIMM support went live in 5.1 without considering that, in
3279      * other archs, the user needs to enable NVDIMM support with the
3280      * 'nvdimm' machine option and the default behavior is NVDIMM
3281      * support disabled. It is too late to roll back to the standard
3282      * behavior without breaking 5.1 guests.
3283      */
3284     if (mc->nvdimm_supported) {
3285         ms->nvdimms_state->is_enabled = true;
3286     }
3287 
3288     spapr->htab_fd = -1;
3289     spapr->use_hotplug_event_source = true;
3290     spapr->kvm_type = g_strdup(DEFAULT_KVM_TYPE);
3291     object_property_add_str(obj, "kvm-type",
3292                             spapr_get_kvm_type, spapr_set_kvm_type);
3293     object_property_set_description(obj, "kvm-type",
3294                                     "Specifies the KVM virtualization mode (auto,"
3295                                     " hv, pr). Defaults to 'auto'. This mode will use"
3296                                     " any available KVM module loaded in the host,"
3297                                     " where kvm_hv takes precedence if both kvm_hv and"
3298                                     " kvm_pr are loaded.");
3299     object_property_add_bool(obj, "modern-hotplug-events",
3300                             spapr_get_modern_hotplug_events,
3301                             spapr_set_modern_hotplug_events);
3302     object_property_set_description(obj, "modern-hotplug-events",
3303                                     "Use dedicated hotplug event mechanism in"
3304                                     " place of standard EPOW events when possible"
3305                                     " (required for memory hot-unplug support)");
3306     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3307                             "Maximum permitted CPU compatibility mode");
3308 
3309     object_property_add_str(obj, "resize-hpt",
3310                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3311     object_property_set_description(obj, "resize-hpt",
3312                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3313     object_property_add_uint32_ptr(obj, "vsmt",
3314                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3315     object_property_set_description(obj, "vsmt",
3316                                     "Virtual SMT: KVM behaves as if this were"
3317                                     " the host's SMT mode");
3318 
3319     object_property_add_bool(obj, "vfio-no-msix-emulation",
3320                              spapr_get_msix_emulation, NULL);
3321 
3322     object_property_add_uint64_ptr(obj, "kernel-addr",
3323                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3324     object_property_set_description(obj, "kernel-addr",
3325                                     stringify(KERNEL_LOAD_ADDR)
3326                                     " for -kernel is the default");
3327     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3328     /* The machine class defines the default interrupt controller mode */
3329     spapr->irq = smc->irq;
3330     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3331                             spapr_set_ic_mode);
3332     object_property_set_description(obj, "ic-mode",
3333                  "Specifies the interrupt controller mode (xics, xive, dual)");
3334 
3335     object_property_add_str(obj, "host-model",
3336         spapr_get_host_model, spapr_set_host_model);
3337     object_property_set_description(obj, "host-model",
3338         "Host model to advertise in guest device tree");
3339     object_property_add_str(obj, "host-serial",
3340         spapr_get_host_serial, spapr_set_host_serial);
3341     object_property_set_description(obj, "host-serial",
3342         "Host serial number to advertise in guest device tree");
3343 }
3344 
3345 static void spapr_machine_finalizefn(Object *obj)
3346 {
3347     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3348 
3349     g_free(spapr->kvm_type);
3350 }
3351 
3352 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3353 {
3354     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3355     PowerPCCPU *cpu = POWERPC_CPU(cs);
3356     CPUPPCState *env = &cpu->env;
3357 
3358     cpu_synchronize_state(cs);
3359     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3360     if (spapr->fwnmi_system_reset_addr != -1) {
3361         uint64_t rtas_addr, addr;
3362 
3363         /* get rtas addr from fdt */
3364         rtas_addr = spapr_get_rtas_addr();
3365         if (!rtas_addr) {
3366             qemu_system_guest_panicked(NULL);
3367             return;
3368         }
3369 
3370         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3371         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3372         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3373         env->gpr[3] = addr;
3374     }
3375     ppc_cpu_do_system_reset(cs);
3376     if (spapr->fwnmi_system_reset_addr != -1) {
3377         env->nip = spapr->fwnmi_system_reset_addr;
3378     }
3379 }
3380 
3381 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3382 {
3383     CPUState *cs;
3384 
3385     CPU_FOREACH(cs) {
3386         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3387     }
3388 }
3389 
3390 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3391                           void *fdt, int *fdt_start_offset, Error **errp)
3392 {
3393     uint64_t addr;
3394     uint32_t node;
3395 
3396     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3397     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3398                                     &error_abort);
3399     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3400                                              SPAPR_MEMORY_BLOCK_SIZE);
3401     return 0;
3402 }
3403 
3404 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3405                            bool dedicated_hp_event_source)
3406 {
3407     SpaprDrc *drc;
3408     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3409     int i;
3410     uint64_t addr = addr_start;
3411     bool hotplugged = spapr_drc_hotplugged(dev);
3412 
3413     for (i = 0; i < nr_lmbs; i++) {
3414         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3415                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3416         g_assert(drc);
3417 
3418         /*
3419          * memory_device_get_free_addr() provided a range of free addresses
3420          * that doesn't overlap with any existing mapping at pre-plug. The
3421          * corresponding LMB DRCs are thus assumed to be all attachable.
3422          */
3423         spapr_drc_attach(drc, dev);
3424         if (!hotplugged) {
3425             spapr_drc_reset(drc);
3426         }
3427         addr += SPAPR_MEMORY_BLOCK_SIZE;
3428     }
3429     /* send hotplug notification to the
3430      * guest only in case of hotplugged memory
3431      */
3432     if (hotplugged) {
3433         if (dedicated_hp_event_source) {
3434             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3435                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3436             g_assert(drc);
3437             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3438                                                    nr_lmbs,
3439                                                    spapr_drc_index(drc));
3440         } else {
3441             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3442                                            nr_lmbs);
3443         }
3444     }
3445 }
3446 
3447 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3448 {
3449     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3450     PCDIMMDevice *dimm = PC_DIMM(dev);
3451     uint64_t size, addr;
3452     int64_t slot;
3453     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3454 
3455     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3456 
3457     pc_dimm_plug(dimm, MACHINE(ms));
3458 
3459     if (!is_nvdimm) {
3460         addr = object_property_get_uint(OBJECT(dimm),
3461                                         PC_DIMM_ADDR_PROP, &error_abort);
3462         spapr_add_lmbs(dev, addr, size,
3463                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT));
3464     } else {
3465         slot = object_property_get_int(OBJECT(dimm),
3466                                        PC_DIMM_SLOT_PROP, &error_abort);
3467         /* We should have valid slot number at this point */
3468         g_assert(slot >= 0);
3469         spapr_add_nvdimm(dev, slot);
3470     }
3471 }
3472 
3473 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3474                                   Error **errp)
3475 {
3476     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3477     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3478     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3479     PCDIMMDevice *dimm = PC_DIMM(dev);
3480     Error *local_err = NULL;
3481     uint64_t size;
3482     Object *memdev;
3483     hwaddr pagesize;
3484 
3485     if (!smc->dr_lmb_enabled) {
3486         error_setg(errp, "Memory hotplug not supported for this machine");
3487         return;
3488     }
3489 
3490     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3491     if (local_err) {
3492         error_propagate(errp, local_err);
3493         return;
3494     }
3495 
3496     if (is_nvdimm) {
3497         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3498             return;
3499         }
3500     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3501         error_setg(errp, "Hotplugged memory size must be a multiple of "
3502                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3503         return;
3504     }
3505 
3506     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3507                                       &error_abort);
3508     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3509     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3510         return;
3511     }
3512 
3513     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3514 }
3515 
3516 struct SpaprDimmState {
3517     PCDIMMDevice *dimm;
3518     uint32_t nr_lmbs;
3519     QTAILQ_ENTRY(SpaprDimmState) next;
3520 };
3521 
3522 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3523                                                        PCDIMMDevice *dimm)
3524 {
3525     SpaprDimmState *dimm_state = NULL;
3526 
3527     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3528         if (dimm_state->dimm == dimm) {
3529             break;
3530         }
3531     }
3532     return dimm_state;
3533 }
3534 
3535 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3536                                                       uint32_t nr_lmbs,
3537                                                       PCDIMMDevice *dimm)
3538 {
3539     SpaprDimmState *ds = NULL;
3540 
3541     /*
3542      * If this request is for a DIMM whose removal had failed earlier
3543      * (due to guest's refusal to remove the LMBs), we would have this
3544      * dimm already in the pending_dimm_unplugs list. In that
3545      * case don't add again.
3546      */
3547     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3548     if (!ds) {
3549         ds = g_malloc0(sizeof(SpaprDimmState));
3550         ds->nr_lmbs = nr_lmbs;
3551         ds->dimm = dimm;
3552         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3553     }
3554     return ds;
3555 }
3556 
3557 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3558                                               SpaprDimmState *dimm_state)
3559 {
3560     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3561     g_free(dimm_state);
3562 }
3563 
3564 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3565                                                         PCDIMMDevice *dimm)
3566 {
3567     SpaprDrc *drc;
3568     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3569                                                   &error_abort);
3570     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3571     uint32_t avail_lmbs = 0;
3572     uint64_t addr_start, addr;
3573     int i;
3574 
3575     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3576                                           &error_abort);
3577 
3578     addr = addr_start;
3579     for (i = 0; i < nr_lmbs; i++) {
3580         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3581                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3582         g_assert(drc);
3583         if (drc->dev) {
3584             avail_lmbs++;
3585         }
3586         addr += SPAPR_MEMORY_BLOCK_SIZE;
3587     }
3588 
3589     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3590 }
3591 
3592 /* Callback to be called during DRC release. */
3593 void spapr_lmb_release(DeviceState *dev)
3594 {
3595     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3596     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3597     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3598 
3599     /* This information will get lost if a migration occurs
3600      * during the unplug process. In this case recover it. */
3601     if (ds == NULL) {
3602         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3603         g_assert(ds);
3604         /* The DRC being examined by the caller at least must be counted */
3605         g_assert(ds->nr_lmbs);
3606     }
3607 
3608     if (--ds->nr_lmbs) {
3609         return;
3610     }
3611 
3612     /*
3613      * Now that all the LMBs have been removed by the guest, call the
3614      * unplug handler chain. This can never fail.
3615      */
3616     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3617     object_unparent(OBJECT(dev));
3618 }
3619 
3620 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3621 {
3622     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3623     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3624 
3625     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3626     qdev_unrealize(dev);
3627     spapr_pending_dimm_unplugs_remove(spapr, ds);
3628 }
3629 
3630 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3631                                         DeviceState *dev, Error **errp)
3632 {
3633     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3634     PCDIMMDevice *dimm = PC_DIMM(dev);
3635     uint32_t nr_lmbs;
3636     uint64_t size, addr_start, addr;
3637     int i;
3638     SpaprDrc *drc;
3639 
3640     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3641         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3642         return;
3643     }
3644 
3645     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3646     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3647 
3648     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3649                                           &error_abort);
3650 
3651     /*
3652      * An existing pending dimm state for this DIMM means that there is an
3653      * unplug operation in progress, waiting for the spapr_lmb_release
3654      * callback to complete the job (BQL can't cover that far). In this case,
3655      * bail out to avoid detaching DRCs that were already released.
3656      */
3657     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3658         error_setg(errp, "Memory unplug already in progress for device %s",
3659                    dev->id);
3660         return;
3661     }
3662 
3663     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3664 
3665     addr = addr_start;
3666     for (i = 0; i < nr_lmbs; i++) {
3667         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3668                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3669         g_assert(drc);
3670 
3671         spapr_drc_detach(drc);
3672         addr += SPAPR_MEMORY_BLOCK_SIZE;
3673     }
3674 
3675     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3676                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3677     g_assert(drc);
3678     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3679                                               nr_lmbs, spapr_drc_index(drc));
3680 }
3681 
3682 /* Callback to be called during DRC release. */
3683 void spapr_core_release(DeviceState *dev)
3684 {
3685     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3686 
3687     /* Call the unplug handler chain. This can never fail. */
3688     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3689     object_unparent(OBJECT(dev));
3690 }
3691 
3692 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3693 {
3694     MachineState *ms = MACHINE(hotplug_dev);
3695     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3696     CPUCore *cc = CPU_CORE(dev);
3697     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3698 
3699     if (smc->pre_2_10_has_unused_icps) {
3700         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3701         int i;
3702 
3703         for (i = 0; i < cc->nr_threads; i++) {
3704             CPUState *cs = CPU(sc->threads[i]);
3705 
3706             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3707         }
3708     }
3709 
3710     assert(core_slot);
3711     core_slot->cpu = NULL;
3712     qdev_unrealize(dev);
3713 }
3714 
3715 static
3716 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3717                                Error **errp)
3718 {
3719     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3720     int index;
3721     SpaprDrc *drc;
3722     CPUCore *cc = CPU_CORE(dev);
3723 
3724     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3725         error_setg(errp, "Unable to find CPU core with core-id: %d",
3726                    cc->core_id);
3727         return;
3728     }
3729     if (index == 0) {
3730         error_setg(errp, "Boot CPU core may not be unplugged");
3731         return;
3732     }
3733 
3734     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3735                           spapr_vcpu_id(spapr, cc->core_id));
3736     g_assert(drc);
3737 
3738     if (!spapr_drc_unplug_requested(drc)) {
3739         spapr_drc_detach(drc);
3740         spapr_hotplug_req_remove_by_index(drc);
3741     }
3742 }
3743 
3744 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3745                            void *fdt, int *fdt_start_offset, Error **errp)
3746 {
3747     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3748     CPUState *cs = CPU(core->threads[0]);
3749     PowerPCCPU *cpu = POWERPC_CPU(cs);
3750     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3751     int id = spapr_get_vcpu_id(cpu);
3752     char *nodename;
3753     int offset;
3754 
3755     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3756     offset = fdt_add_subnode(fdt, 0, nodename);
3757     g_free(nodename);
3758 
3759     spapr_dt_cpu(cs, fdt, offset, spapr);
3760 
3761     *fdt_start_offset = offset;
3762     return 0;
3763 }
3764 
3765 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3766 {
3767     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3768     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3769     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3770     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3771     CPUCore *cc = CPU_CORE(dev);
3772     CPUState *cs;
3773     SpaprDrc *drc;
3774     CPUArchId *core_slot;
3775     int index;
3776     bool hotplugged = spapr_drc_hotplugged(dev);
3777     int i;
3778 
3779     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3780     g_assert(core_slot); /* Already checked in spapr_core_pre_plug() */
3781 
3782     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3783                           spapr_vcpu_id(spapr, cc->core_id));
3784 
3785     g_assert(drc || !mc->has_hotpluggable_cpus);
3786 
3787     if (drc) {
3788         /*
3789          * spapr_core_pre_plug() already buys us this is a brand new
3790          * core being plugged into a free slot. Nothing should already
3791          * be attached to the corresponding DRC.
3792          */
3793         spapr_drc_attach(drc, dev);
3794 
3795         if (hotplugged) {
3796             /*
3797              * Send hotplug notification interrupt to the guest only
3798              * in case of hotplugged CPUs.
3799              */
3800             spapr_hotplug_req_add_by_index(drc);
3801         } else {
3802             spapr_drc_reset(drc);
3803         }
3804     }
3805 
3806     core_slot->cpu = OBJECT(dev);
3807 
3808     /*
3809      * Set compatibility mode to match the boot CPU, which was either set
3810      * by the machine reset code or by CAS. This really shouldn't fail at
3811      * this point.
3812      */
3813     if (hotplugged) {
3814         for (i = 0; i < cc->nr_threads; i++) {
3815             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3816                            &error_abort);
3817         }
3818     }
3819 
3820     if (smc->pre_2_10_has_unused_icps) {
3821         for (i = 0; i < cc->nr_threads; i++) {
3822             cs = CPU(core->threads[i]);
3823             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3824         }
3825     }
3826 }
3827 
3828 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3829                                 Error **errp)
3830 {
3831     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3832     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3833     CPUCore *cc = CPU_CORE(dev);
3834     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3835     const char *type = object_get_typename(OBJECT(dev));
3836     CPUArchId *core_slot;
3837     int index;
3838     unsigned int smp_threads = machine->smp.threads;
3839 
3840     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3841         error_setg(errp, "CPU hotplug not supported for this machine");
3842         return;
3843     }
3844 
3845     if (strcmp(base_core_type, type)) {
3846         error_setg(errp, "CPU core type should be %s", base_core_type);
3847         return;
3848     }
3849 
3850     if (cc->core_id % smp_threads) {
3851         error_setg(errp, "invalid core id %d", cc->core_id);
3852         return;
3853     }
3854 
3855     /*
3856      * In general we should have homogeneous threads-per-core, but old
3857      * (pre hotplug support) machine types allow the last core to have
3858      * reduced threads as a compatibility hack for when we allowed
3859      * total vcpus not a multiple of threads-per-core.
3860      */
3861     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3862         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
3863                    smp_threads);
3864         return;
3865     }
3866 
3867     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3868     if (!core_slot) {
3869         error_setg(errp, "core id %d out of range", cc->core_id);
3870         return;
3871     }
3872 
3873     if (core_slot->cpu) {
3874         error_setg(errp, "core %d already populated", cc->core_id);
3875         return;
3876     }
3877 
3878     numa_cpu_pre_plug(core_slot, dev, errp);
3879 }
3880 
3881 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3882                           void *fdt, int *fdt_start_offset, Error **errp)
3883 {
3884     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3885     int intc_phandle;
3886 
3887     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3888     if (intc_phandle <= 0) {
3889         return -1;
3890     }
3891 
3892     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3893         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3894         return -1;
3895     }
3896 
3897     /* generally SLOF creates these, for hotplug it's up to QEMU */
3898     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3899 
3900     return 0;
3901 }
3902 
3903 static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3904                                Error **errp)
3905 {
3906     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3907     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3908     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3909     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3910     SpaprDrc *drc;
3911 
3912     if (dev->hotplugged && !smc->dr_phb_enabled) {
3913         error_setg(errp, "PHB hotplug not supported for this machine");
3914         return false;
3915     }
3916 
3917     if (sphb->index == (uint32_t)-1) {
3918         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3919         return false;
3920     }
3921 
3922     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3923     if (drc && drc->dev) {
3924         error_setg(errp, "PHB %d already attached", sphb->index);
3925         return false;
3926     }
3927 
3928     /*
3929      * This will check that sphb->index doesn't exceed the maximum number of
3930      * PHBs for the current machine type.
3931      */
3932     return
3933         smc->phb_placement(spapr, sphb->index,
3934                            &sphb->buid, &sphb->io_win_addr,
3935                            &sphb->mem_win_addr, &sphb->mem64_win_addr,
3936                            windows_supported, sphb->dma_liobn,
3937                            &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3938                            errp);
3939 }
3940 
3941 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3942 {
3943     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3944     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3945     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3946     SpaprDrc *drc;
3947     bool hotplugged = spapr_drc_hotplugged(dev);
3948 
3949     if (!smc->dr_phb_enabled) {
3950         return;
3951     }
3952 
3953     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3954     /* hotplug hooks should check it's enabled before getting this far */
3955     assert(drc);
3956 
3957     /* spapr_phb_pre_plug() already checked the DRC is attachable */
3958     spapr_drc_attach(drc, dev);
3959 
3960     if (hotplugged) {
3961         spapr_hotplug_req_add_by_index(drc);
3962     } else {
3963         spapr_drc_reset(drc);
3964     }
3965 }
3966 
3967 void spapr_phb_release(DeviceState *dev)
3968 {
3969     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3970 
3971     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3972     object_unparent(OBJECT(dev));
3973 }
3974 
3975 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3976 {
3977     qdev_unrealize(dev);
3978 }
3979 
3980 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
3981                                      DeviceState *dev, Error **errp)
3982 {
3983     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3984     SpaprDrc *drc;
3985 
3986     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3987     assert(drc);
3988 
3989     if (!spapr_drc_unplug_requested(drc)) {
3990         spapr_drc_detach(drc);
3991         spapr_hotplug_req_remove_by_index(drc);
3992     }
3993 }
3994 
3995 static
3996 bool spapr_tpm_proxy_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3997                               Error **errp)
3998 {
3999     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4000 
4001     if (spapr->tpm_proxy != NULL) {
4002         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4003         return false;
4004     }
4005 
4006     return true;
4007 }
4008 
4009 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4010 {
4011     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4012     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4013 
4014     /* Already checked in spapr_tpm_proxy_pre_plug() */
4015     g_assert(spapr->tpm_proxy == NULL);
4016 
4017     spapr->tpm_proxy = tpm_proxy;
4018 }
4019 
4020 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4021 {
4022     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4023 
4024     qdev_unrealize(dev);
4025     object_unparent(OBJECT(dev));
4026     spapr->tpm_proxy = NULL;
4027 }
4028 
4029 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4030                                       DeviceState *dev, Error **errp)
4031 {
4032     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4033         spapr_memory_plug(hotplug_dev, dev);
4034     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4035         spapr_core_plug(hotplug_dev, dev);
4036     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4037         spapr_phb_plug(hotplug_dev, dev);
4038     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4039         spapr_tpm_proxy_plug(hotplug_dev, dev);
4040     }
4041 }
4042 
4043 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4044                                         DeviceState *dev, Error **errp)
4045 {
4046     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4047         spapr_memory_unplug(hotplug_dev, dev);
4048     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4049         spapr_core_unplug(hotplug_dev, dev);
4050     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4051         spapr_phb_unplug(hotplug_dev, dev);
4052     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4053         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4054     }
4055 }
4056 
4057 bool spapr_memory_hot_unplug_supported(SpaprMachineState *spapr)
4058 {
4059     return spapr_ovec_test(spapr->ov5_cas, OV5_HP_EVT) ||
4060         /*
4061          * CAS will process all pending unplug requests.
4062          *
4063          * HACK: a guest could theoretically have cleared all bits in OV5,
4064          * but none of the guests we care for do.
4065          */
4066         spapr_ovec_empty(spapr->ov5_cas);
4067 }
4068 
4069 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4070                                                 DeviceState *dev, Error **errp)
4071 {
4072     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4073     MachineClass *mc = MACHINE_GET_CLASS(sms);
4074     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4075 
4076     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4077         if (spapr_memory_hot_unplug_supported(sms)) {
4078             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4079         } else {
4080             error_setg(errp, "Memory hot unplug not supported for this guest");
4081         }
4082     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4083         if (!mc->has_hotpluggable_cpus) {
4084             error_setg(errp, "CPU hot unplug not supported on this machine");
4085             return;
4086         }
4087         spapr_core_unplug_request(hotplug_dev, dev, errp);
4088     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4089         if (!smc->dr_phb_enabled) {
4090             error_setg(errp, "PHB hot unplug not supported on this machine");
4091             return;
4092         }
4093         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4094     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4095         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4096     }
4097 }
4098 
4099 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4100                                           DeviceState *dev, Error **errp)
4101 {
4102     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4103         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4104     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4105         spapr_core_pre_plug(hotplug_dev, dev, errp);
4106     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4107         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4108     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4109         spapr_tpm_proxy_pre_plug(hotplug_dev, dev, errp);
4110     }
4111 }
4112 
4113 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4114                                                  DeviceState *dev)
4115 {
4116     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4117         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4118         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4119         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4120         return HOTPLUG_HANDLER(machine);
4121     }
4122     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4123         PCIDevice *pcidev = PCI_DEVICE(dev);
4124         PCIBus *root = pci_device_root_bus(pcidev);
4125         SpaprPhbState *phb =
4126             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4127                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4128 
4129         if (phb) {
4130             return HOTPLUG_HANDLER(phb);
4131         }
4132     }
4133     return NULL;
4134 }
4135 
4136 static CpuInstanceProperties
4137 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4138 {
4139     CPUArchId *core_slot;
4140     MachineClass *mc = MACHINE_GET_CLASS(machine);
4141 
4142     /* make sure possible_cpu are intialized */
4143     mc->possible_cpu_arch_ids(machine);
4144     /* get CPU core slot containing thread that matches cpu_index */
4145     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4146     assert(core_slot);
4147     return core_slot->props;
4148 }
4149 
4150 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4151 {
4152     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4153 }
4154 
4155 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4156 {
4157     int i;
4158     unsigned int smp_threads = machine->smp.threads;
4159     unsigned int smp_cpus = machine->smp.cpus;
4160     const char *core_type;
4161     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4162     MachineClass *mc = MACHINE_GET_CLASS(machine);
4163 
4164     if (!mc->has_hotpluggable_cpus) {
4165         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4166     }
4167     if (machine->possible_cpus) {
4168         assert(machine->possible_cpus->len == spapr_max_cores);
4169         return machine->possible_cpus;
4170     }
4171 
4172     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4173     if (!core_type) {
4174         error_report("Unable to find sPAPR CPU Core definition");
4175         exit(1);
4176     }
4177 
4178     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4179                              sizeof(CPUArchId) * spapr_max_cores);
4180     machine->possible_cpus->len = spapr_max_cores;
4181     for (i = 0; i < machine->possible_cpus->len; i++) {
4182         int core_id = i * smp_threads;
4183 
4184         machine->possible_cpus->cpus[i].type = core_type;
4185         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4186         machine->possible_cpus->cpus[i].arch_id = core_id;
4187         machine->possible_cpus->cpus[i].props.has_core_id = true;
4188         machine->possible_cpus->cpus[i].props.core_id = core_id;
4189     }
4190     return machine->possible_cpus;
4191 }
4192 
4193 static bool spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4194                                 uint64_t *buid, hwaddr *pio,
4195                                 hwaddr *mmio32, hwaddr *mmio64,
4196                                 unsigned n_dma, uint32_t *liobns,
4197                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4198 {
4199     /*
4200      * New-style PHB window placement.
4201      *
4202      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4203      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4204      * windows.
4205      *
4206      * Some guest kernels can't work with MMIO windows above 1<<46
4207      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4208      *
4209      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4210      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4211      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4212      * 1TiB 64-bit MMIO windows for each PHB.
4213      */
4214     const uint64_t base_buid = 0x800000020000000ULL;
4215     int i;
4216 
4217     /* Sanity check natural alignments */
4218     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4219     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4220     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4221     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4222     /* Sanity check bounds */
4223     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4224                       SPAPR_PCI_MEM32_WIN_SIZE);
4225     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4226                       SPAPR_PCI_MEM64_WIN_SIZE);
4227 
4228     if (index >= SPAPR_MAX_PHBS) {
4229         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4230                    SPAPR_MAX_PHBS - 1);
4231         return false;
4232     }
4233 
4234     *buid = base_buid + index;
4235     for (i = 0; i < n_dma; ++i) {
4236         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4237     }
4238 
4239     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4240     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4241     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4242 
4243     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4244     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4245     return true;
4246 }
4247 
4248 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4249 {
4250     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4251 
4252     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4253 }
4254 
4255 static void spapr_ics_resend(XICSFabric *dev)
4256 {
4257     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4258 
4259     ics_resend(spapr->ics);
4260 }
4261 
4262 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4263 {
4264     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4265 
4266     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4267 }
4268 
4269 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4270                                  Monitor *mon)
4271 {
4272     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4273 
4274     spapr_irq_print_info(spapr, mon);
4275     monitor_printf(mon, "irqchip: %s\n",
4276                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4277 }
4278 
4279 /*
4280  * This is a XIVE only operation
4281  */
4282 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4283                            uint8_t nvt_blk, uint32_t nvt_idx,
4284                            bool cam_ignore, uint8_t priority,
4285                            uint32_t logic_serv, XiveTCTXMatch *match)
4286 {
4287     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4288     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4289     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4290     int count;
4291 
4292     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4293                            priority, logic_serv, match);
4294     if (count < 0) {
4295         return count;
4296     }
4297 
4298     /*
4299      * When we implement the save and restore of the thread interrupt
4300      * contexts in the enter/exit CPU handlers of the machine and the
4301      * escalations in QEMU, we should be able to handle non dispatched
4302      * vCPUs.
4303      *
4304      * Until this is done, the sPAPR machine should find at least one
4305      * matching context always.
4306      */
4307     if (count == 0) {
4308         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4309                       nvt_blk, nvt_idx);
4310     }
4311 
4312     return count;
4313 }
4314 
4315 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4316 {
4317     return cpu->vcpu_id;
4318 }
4319 
4320 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4321 {
4322     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4323     MachineState *ms = MACHINE(spapr);
4324     int vcpu_id;
4325 
4326     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4327 
4328     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4329         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4330         error_append_hint(errp, "Adjust the number of cpus to %d "
4331                           "or try to raise the number of threads per core\n",
4332                           vcpu_id * ms->smp.threads / spapr->vsmt);
4333         return false;
4334     }
4335 
4336     cpu->vcpu_id = vcpu_id;
4337     return true;
4338 }
4339 
4340 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4341 {
4342     CPUState *cs;
4343 
4344     CPU_FOREACH(cs) {
4345         PowerPCCPU *cpu = POWERPC_CPU(cs);
4346 
4347         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4348             return cpu;
4349         }
4350     }
4351 
4352     return NULL;
4353 }
4354 
4355 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4356 {
4357     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4358 
4359     /* These are only called by TCG, KVM maintains dispatch state */
4360 
4361     spapr_cpu->prod = false;
4362     if (spapr_cpu->vpa_addr) {
4363         CPUState *cs = CPU(cpu);
4364         uint32_t dispatch;
4365 
4366         dispatch = ldl_be_phys(cs->as,
4367                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4368         dispatch++;
4369         if ((dispatch & 1) != 0) {
4370             qemu_log_mask(LOG_GUEST_ERROR,
4371                           "VPA: incorrect dispatch counter value for "
4372                           "dispatched partition %u, correcting.\n", dispatch);
4373             dispatch++;
4374         }
4375         stl_be_phys(cs->as,
4376                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4377     }
4378 }
4379 
4380 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4381 {
4382     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4383 
4384     if (spapr_cpu->vpa_addr) {
4385         CPUState *cs = CPU(cpu);
4386         uint32_t dispatch;
4387 
4388         dispatch = ldl_be_phys(cs->as,
4389                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4390         dispatch++;
4391         if ((dispatch & 1) != 1) {
4392             qemu_log_mask(LOG_GUEST_ERROR,
4393                           "VPA: incorrect dispatch counter value for "
4394                           "preempted partition %u, correcting.\n", dispatch);
4395             dispatch++;
4396         }
4397         stl_be_phys(cs->as,
4398                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4399     }
4400 }
4401 
4402 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4403 {
4404     MachineClass *mc = MACHINE_CLASS(oc);
4405     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4406     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4407     NMIClass *nc = NMI_CLASS(oc);
4408     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4409     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4410     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4411     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4412     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4413 
4414     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4415     mc->ignore_boot_device_suffixes = true;
4416 
4417     /*
4418      * We set up the default / latest behaviour here.  The class_init
4419      * functions for the specific versioned machine types can override
4420      * these details for backwards compatibility
4421      */
4422     mc->init = spapr_machine_init;
4423     mc->reset = spapr_machine_reset;
4424     mc->block_default_type = IF_SCSI;
4425     mc->max_cpus = 1024;
4426     mc->no_parallel = 1;
4427     mc->default_boot_order = "";
4428     mc->default_ram_size = 512 * MiB;
4429     mc->default_ram_id = "ppc_spapr.ram";
4430     mc->default_display = "std";
4431     mc->kvm_type = spapr_kvm_type;
4432     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4433     mc->pci_allow_0_address = true;
4434     assert(!mc->get_hotplug_handler);
4435     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4436     hc->pre_plug = spapr_machine_device_pre_plug;
4437     hc->plug = spapr_machine_device_plug;
4438     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4439     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4440     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4441     hc->unplug_request = spapr_machine_device_unplug_request;
4442     hc->unplug = spapr_machine_device_unplug;
4443 
4444     smc->dr_lmb_enabled = true;
4445     smc->update_dt_enabled = true;
4446     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4447     mc->has_hotpluggable_cpus = true;
4448     mc->nvdimm_supported = true;
4449     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4450     fwc->get_dev_path = spapr_get_fw_dev_path;
4451     nc->nmi_monitor_handler = spapr_nmi;
4452     smc->phb_placement = spapr_phb_placement;
4453     vhc->hypercall = emulate_spapr_hypercall;
4454     vhc->hpt_mask = spapr_hpt_mask;
4455     vhc->map_hptes = spapr_map_hptes;
4456     vhc->unmap_hptes = spapr_unmap_hptes;
4457     vhc->hpte_set_c = spapr_hpte_set_c;
4458     vhc->hpte_set_r = spapr_hpte_set_r;
4459     vhc->get_pate = spapr_get_pate;
4460     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4461     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4462     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4463     xic->ics_get = spapr_ics_get;
4464     xic->ics_resend = spapr_ics_resend;
4465     xic->icp_get = spapr_icp_get;
4466     ispc->print_info = spapr_pic_print_info;
4467     /* Force NUMA node memory size to be a multiple of
4468      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4469      * in which LMBs are represented and hot-added
4470      */
4471     mc->numa_mem_align_shift = 28;
4472     mc->auto_enable_numa = true;
4473 
4474     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4475     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4476     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4477     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4478     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4479     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4480     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4481     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4482     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4483     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4484     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4485     spapr_caps_add_properties(smc);
4486     smc->irq = &spapr_irq_dual;
4487     smc->dr_phb_enabled = true;
4488     smc->linux_pci_probe = true;
4489     smc->smp_threads_vsmt = true;
4490     smc->nr_xirqs = SPAPR_NR_XIRQS;
4491     xfc->match_nvt = spapr_match_nvt;
4492 }
4493 
4494 static const TypeInfo spapr_machine_info = {
4495     .name          = TYPE_SPAPR_MACHINE,
4496     .parent        = TYPE_MACHINE,
4497     .abstract      = true,
4498     .instance_size = sizeof(SpaprMachineState),
4499     .instance_init = spapr_instance_init,
4500     .instance_finalize = spapr_machine_finalizefn,
4501     .class_size    = sizeof(SpaprMachineClass),
4502     .class_init    = spapr_machine_class_init,
4503     .interfaces = (InterfaceInfo[]) {
4504         { TYPE_FW_PATH_PROVIDER },
4505         { TYPE_NMI },
4506         { TYPE_HOTPLUG_HANDLER },
4507         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4508         { TYPE_XICS_FABRIC },
4509         { TYPE_INTERRUPT_STATS_PROVIDER },
4510         { TYPE_XIVE_FABRIC },
4511         { }
4512     },
4513 };
4514 
4515 static void spapr_machine_latest_class_options(MachineClass *mc)
4516 {
4517     mc->alias = "pseries";
4518     mc->is_default = true;
4519 }
4520 
4521 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4522     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4523                                                     void *data)      \
4524     {                                                                \
4525         MachineClass *mc = MACHINE_CLASS(oc);                        \
4526         spapr_machine_##suffix##_class_options(mc);                  \
4527         if (latest) {                                                \
4528             spapr_machine_latest_class_options(mc);                  \
4529         }                                                            \
4530     }                                                                \
4531     static const TypeInfo spapr_machine_##suffix##_info = {          \
4532         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4533         .parent = TYPE_SPAPR_MACHINE,                                \
4534         .class_init = spapr_machine_##suffix##_class_init,           \
4535     };                                                               \
4536     static void spapr_machine_register_##suffix(void)                \
4537     {                                                                \
4538         type_register(&spapr_machine_##suffix##_info);               \
4539     }                                                                \
4540     type_init(spapr_machine_register_##suffix)
4541 
4542 /*
4543  * pseries-6.0
4544  */
4545 static void spapr_machine_6_0_class_options(MachineClass *mc)
4546 {
4547     /* Defaults for the latest behaviour inherited from the base class */
4548 }
4549 
4550 DEFINE_SPAPR_MACHINE(6_0, "6.0", true);
4551 
4552 /*
4553  * pseries-5.2
4554  */
4555 static void spapr_machine_5_2_class_options(MachineClass *mc)
4556 {
4557     spapr_machine_6_0_class_options(mc);
4558     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
4559 }
4560 
4561 DEFINE_SPAPR_MACHINE(5_2, "5.2", false);
4562 
4563 /*
4564  * pseries-5.1
4565  */
4566 static void spapr_machine_5_1_class_options(MachineClass *mc)
4567 {
4568     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4569 
4570     spapr_machine_5_2_class_options(mc);
4571     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4572     smc->pre_5_2_numa_associativity = true;
4573 }
4574 
4575 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4576 
4577 /*
4578  * pseries-5.0
4579  */
4580 static void spapr_machine_5_0_class_options(MachineClass *mc)
4581 {
4582     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4583     static GlobalProperty compat[] = {
4584         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4585     };
4586 
4587     spapr_machine_5_1_class_options(mc);
4588     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4589     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4590     mc->numa_mem_supported = true;
4591     smc->pre_5_1_assoc_refpoints = true;
4592 }
4593 
4594 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4595 
4596 /*
4597  * pseries-4.2
4598  */
4599 static void spapr_machine_4_2_class_options(MachineClass *mc)
4600 {
4601     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4602 
4603     spapr_machine_5_0_class_options(mc);
4604     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4605     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4606     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4607     smc->rma_limit = 16 * GiB;
4608     mc->nvdimm_supported = false;
4609 }
4610 
4611 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4612 
4613 /*
4614  * pseries-4.1
4615  */
4616 static void spapr_machine_4_1_class_options(MachineClass *mc)
4617 {
4618     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4619     static GlobalProperty compat[] = {
4620         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4621         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4622     };
4623 
4624     spapr_machine_4_2_class_options(mc);
4625     smc->linux_pci_probe = false;
4626     smc->smp_threads_vsmt = false;
4627     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4628     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4629 }
4630 
4631 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4632 
4633 /*
4634  * pseries-4.0
4635  */
4636 static bool phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4637                               uint64_t *buid, hwaddr *pio,
4638                               hwaddr *mmio32, hwaddr *mmio64,
4639                               unsigned n_dma, uint32_t *liobns,
4640                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4641 {
4642     if (!spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma,
4643                              liobns, nv2gpa, nv2atsd, errp)) {
4644         return false;
4645     }
4646 
4647     *nv2gpa = 0;
4648     *nv2atsd = 0;
4649     return true;
4650 }
4651 static void spapr_machine_4_0_class_options(MachineClass *mc)
4652 {
4653     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4654 
4655     spapr_machine_4_1_class_options(mc);
4656     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4657     smc->phb_placement = phb_placement_4_0;
4658     smc->irq = &spapr_irq_xics;
4659     smc->pre_4_1_migration = true;
4660 }
4661 
4662 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4663 
4664 /*
4665  * pseries-3.1
4666  */
4667 static void spapr_machine_3_1_class_options(MachineClass *mc)
4668 {
4669     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4670 
4671     spapr_machine_4_0_class_options(mc);
4672     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4673 
4674     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4675     smc->update_dt_enabled = false;
4676     smc->dr_phb_enabled = false;
4677     smc->broken_host_serial_model = true;
4678     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4679     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4680     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4681     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4682 }
4683 
4684 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4685 
4686 /*
4687  * pseries-3.0
4688  */
4689 
4690 static void spapr_machine_3_0_class_options(MachineClass *mc)
4691 {
4692     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4693 
4694     spapr_machine_3_1_class_options(mc);
4695     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4696 
4697     smc->legacy_irq_allocation = true;
4698     smc->nr_xirqs = 0x400;
4699     smc->irq = &spapr_irq_xics_legacy;
4700 }
4701 
4702 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4703 
4704 /*
4705  * pseries-2.12
4706  */
4707 static void spapr_machine_2_12_class_options(MachineClass *mc)
4708 {
4709     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4710     static GlobalProperty compat[] = {
4711         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4712         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4713     };
4714 
4715     spapr_machine_3_0_class_options(mc);
4716     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4717     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4718 
4719     /* We depend on kvm_enabled() to choose a default value for the
4720      * hpt-max-page-size capability. Of course we can't do it here
4721      * because this is too early and the HW accelerator isn't initialzed
4722      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4723      */
4724     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4725 }
4726 
4727 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4728 
4729 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4730 {
4731     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4732 
4733     spapr_machine_2_12_class_options(mc);
4734     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4735     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4736     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4737 }
4738 
4739 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4740 
4741 /*
4742  * pseries-2.11
4743  */
4744 
4745 static void spapr_machine_2_11_class_options(MachineClass *mc)
4746 {
4747     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4748 
4749     spapr_machine_2_12_class_options(mc);
4750     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4751     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4752 }
4753 
4754 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4755 
4756 /*
4757  * pseries-2.10
4758  */
4759 
4760 static void spapr_machine_2_10_class_options(MachineClass *mc)
4761 {
4762     spapr_machine_2_11_class_options(mc);
4763     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4764 }
4765 
4766 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4767 
4768 /*
4769  * pseries-2.9
4770  */
4771 
4772 static void spapr_machine_2_9_class_options(MachineClass *mc)
4773 {
4774     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4775     static GlobalProperty compat[] = {
4776         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4777     };
4778 
4779     spapr_machine_2_10_class_options(mc);
4780     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4781     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4782     smc->pre_2_10_has_unused_icps = true;
4783     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4784 }
4785 
4786 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4787 
4788 /*
4789  * pseries-2.8
4790  */
4791 
4792 static void spapr_machine_2_8_class_options(MachineClass *mc)
4793 {
4794     static GlobalProperty compat[] = {
4795         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4796     };
4797 
4798     spapr_machine_2_9_class_options(mc);
4799     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4800     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4801     mc->numa_mem_align_shift = 23;
4802 }
4803 
4804 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4805 
4806 /*
4807  * pseries-2.7
4808  */
4809 
4810 static bool phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4811                               uint64_t *buid, hwaddr *pio,
4812                               hwaddr *mmio32, hwaddr *mmio64,
4813                               unsigned n_dma, uint32_t *liobns,
4814                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4815 {
4816     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4817     const uint64_t base_buid = 0x800000020000000ULL;
4818     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4819     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4820     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4821     const uint32_t max_index = 255;
4822     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4823 
4824     uint64_t ram_top = MACHINE(spapr)->ram_size;
4825     hwaddr phb0_base, phb_base;
4826     int i;
4827 
4828     /* Do we have device memory? */
4829     if (MACHINE(spapr)->maxram_size > ram_top) {
4830         /* Can't just use maxram_size, because there may be an
4831          * alignment gap between normal and device memory regions
4832          */
4833         ram_top = MACHINE(spapr)->device_memory->base +
4834             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4835     }
4836 
4837     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4838 
4839     if (index > max_index) {
4840         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4841                    max_index);
4842         return false;
4843     }
4844 
4845     *buid = base_buid + index;
4846     for (i = 0; i < n_dma; ++i) {
4847         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4848     }
4849 
4850     phb_base = phb0_base + index * phb_spacing;
4851     *pio = phb_base + pio_offset;
4852     *mmio32 = phb_base + mmio_offset;
4853     /*
4854      * We don't set the 64-bit MMIO window, relying on the PHB's
4855      * fallback behaviour of automatically splitting a large "32-bit"
4856      * window into contiguous 32-bit and 64-bit windows
4857      */
4858 
4859     *nv2gpa = 0;
4860     *nv2atsd = 0;
4861     return true;
4862 }
4863 
4864 static void spapr_machine_2_7_class_options(MachineClass *mc)
4865 {
4866     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4867     static GlobalProperty compat[] = {
4868         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4869         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4870         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4871         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4872     };
4873 
4874     spapr_machine_2_8_class_options(mc);
4875     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4876     mc->default_machine_opts = "modern-hotplug-events=off";
4877     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4878     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4879     smc->phb_placement = phb_placement_2_7;
4880 }
4881 
4882 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4883 
4884 /*
4885  * pseries-2.6
4886  */
4887 
4888 static void spapr_machine_2_6_class_options(MachineClass *mc)
4889 {
4890     static GlobalProperty compat[] = {
4891         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4892     };
4893 
4894     spapr_machine_2_7_class_options(mc);
4895     mc->has_hotpluggable_cpus = false;
4896     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4897     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4898 }
4899 
4900 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4901 
4902 /*
4903  * pseries-2.5
4904  */
4905 
4906 static void spapr_machine_2_5_class_options(MachineClass *mc)
4907 {
4908     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4909     static GlobalProperty compat[] = {
4910         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4911     };
4912 
4913     spapr_machine_2_6_class_options(mc);
4914     smc->use_ohci_by_default = true;
4915     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4916     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4917 }
4918 
4919 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4920 
4921 /*
4922  * pseries-2.4
4923  */
4924 
4925 static void spapr_machine_2_4_class_options(MachineClass *mc)
4926 {
4927     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4928 
4929     spapr_machine_2_5_class_options(mc);
4930     smc->dr_lmb_enabled = false;
4931     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4932 }
4933 
4934 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4935 
4936 /*
4937  * pseries-2.3
4938  */
4939 
4940 static void spapr_machine_2_3_class_options(MachineClass *mc)
4941 {
4942     static GlobalProperty compat[] = {
4943         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4944     };
4945     spapr_machine_2_4_class_options(mc);
4946     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4947     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4948 }
4949 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4950 
4951 /*
4952  * pseries-2.2
4953  */
4954 
4955 static void spapr_machine_2_2_class_options(MachineClass *mc)
4956 {
4957     static GlobalProperty compat[] = {
4958         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4959     };
4960 
4961     spapr_machine_2_3_class_options(mc);
4962     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4963     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4964     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4965 }
4966 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4967 
4968 /*
4969  * pseries-2.1
4970  */
4971 
4972 static void spapr_machine_2_1_class_options(MachineClass *mc)
4973 {
4974     spapr_machine_2_2_class_options(mc);
4975     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4976 }
4977 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4978 
4979 static void spapr_machine_register_types(void)
4980 {
4981     type_register_static(&spapr_machine_info);
4982 }
4983 
4984 type_init(spapr_machine_register_types)
4985