xref: /openbmc/qemu/hw/ppc/spapr.c (revision a0c3747e)
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/qapi-events-machine.h"
32 #include "qapi/visitor.h"
33 #include "sysemu/sysemu.h"
34 #include "sysemu/hostmem.h"
35 #include "sysemu/numa.h"
36 #include "sysemu/qtest.h"
37 #include "sysemu/reset.h"
38 #include "sysemu/runstate.h"
39 #include "qemu/log.h"
40 #include "hw/fw-path-provider.h"
41 #include "elf.h"
42 #include "net/net.h"
43 #include "sysemu/device_tree.h"
44 #include "sysemu/cpus.h"
45 #include "sysemu/hw_accel.h"
46 #include "kvm_ppc.h"
47 #include "migration/misc.h"
48 #include "migration/qemu-file-types.h"
49 #include "migration/global_state.h"
50 #include "migration/register.h"
51 #include "migration/blocker.h"
52 #include "mmu-hash64.h"
53 #include "mmu-book3s-v3.h"
54 #include "cpu-models.h"
55 #include "hw/core/cpu.h"
56 
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/ram_addr.h"
73 #include "hw/usb.h"
74 #include "qemu/config-file.h"
75 #include "qemu/error-report.h"
76 #include "trace.h"
77 #include "hw/nmi.h"
78 #include "hw/intc/intc.h"
79 
80 #include "hw/ppc/spapr_cpu_core.h"
81 #include "hw/mem/memory-device.h"
82 #include "hw/ppc/spapr_tpm_proxy.h"
83 #include "hw/ppc/spapr_nvdimm.h"
84 #include "hw/ppc/spapr_numa.h"
85 #include "hw/ppc/pef.h"
86 
87 #include "monitor/monitor.h"
88 
89 #include <libfdt.h>
90 
91 /* SLOF memory layout:
92  *
93  * SLOF raw image loaded at 0, copies its romfs right below the flat
94  * device-tree, then position SLOF itself 31M below that
95  *
96  * So we set FW_OVERHEAD to 40MB which should account for all of that
97  * and more
98  *
99  * We load our kernel at 4M, leaving space for SLOF initial image
100  */
101 #define FDT_MAX_ADDR            0x80000000 /* FDT must stay below that */
102 #define FW_MAX_SIZE             0x400000
103 #define FW_FILE_NAME            "slof.bin"
104 #define FW_OVERHEAD             0x2800000
105 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
106 
107 #define MIN_RMA_SLOF            (128 * MiB)
108 
109 #define PHANDLE_INTC            0x00001111
110 
111 /* These two functions implement the VCPU id numbering: one to compute them
112  * all and one to identify thread 0 of a VCORE. Any change to the first one
113  * is likely to have an impact on the second one, so let's keep them close.
114  */
115 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
116 {
117     MachineState *ms = MACHINE(spapr);
118     unsigned int smp_threads = ms->smp.threads;
119 
120     assert(spapr->vsmt);
121     return
122         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
123 }
124 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
125                                       PowerPCCPU *cpu)
126 {
127     assert(spapr->vsmt);
128     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
129 }
130 
131 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
132 {
133     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
134      * and newer QEMUs don't even have them. In both cases, we don't want
135      * to send anything on the wire.
136      */
137     return false;
138 }
139 
140 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
141     .name = "icp/server",
142     .version_id = 1,
143     .minimum_version_id = 1,
144     .needed = pre_2_10_vmstate_dummy_icp_needed,
145     .fields = (VMStateField[]) {
146         VMSTATE_UNUSED(4), /* uint32_t xirr */
147         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
148         VMSTATE_UNUSED(1), /* uint8_t mfrr */
149         VMSTATE_END_OF_LIST()
150     },
151 };
152 
153 static void pre_2_10_vmstate_register_dummy_icp(int i)
154 {
155     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
156                      (void *)(uintptr_t) i);
157 }
158 
159 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
160 {
161     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
162                        (void *)(uintptr_t) i);
163 }
164 
165 int spapr_max_server_number(SpaprMachineState *spapr)
166 {
167     MachineState *ms = MACHINE(spapr);
168 
169     assert(spapr->vsmt);
170     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
171 }
172 
173 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
174                                   int smt_threads)
175 {
176     int i, ret = 0;
177     uint32_t servers_prop[smt_threads];
178     uint32_t gservers_prop[smt_threads * 2];
179     int index = spapr_get_vcpu_id(cpu);
180 
181     if (cpu->compat_pvr) {
182         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
183         if (ret < 0) {
184             return ret;
185         }
186     }
187 
188     /* Build interrupt servers and gservers properties */
189     for (i = 0; i < smt_threads; i++) {
190         servers_prop[i] = cpu_to_be32(index + i);
191         /* Hack, direct the group queues back to cpu 0 */
192         gservers_prop[i*2] = cpu_to_be32(index + i);
193         gservers_prop[i*2 + 1] = 0;
194     }
195     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
196                       servers_prop, sizeof(servers_prop));
197     if (ret < 0) {
198         return ret;
199     }
200     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
201                       gservers_prop, sizeof(gservers_prop));
202 
203     return ret;
204 }
205 
206 static void spapr_dt_pa_features(SpaprMachineState *spapr,
207                                  PowerPCCPU *cpu,
208                                  void *fdt, int offset)
209 {
210     uint8_t pa_features_206[] = { 6, 0,
211         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
212     uint8_t pa_features_207[] = { 24, 0,
213         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
214         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
215         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
216         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
217     uint8_t pa_features_300[] = { 66, 0,
218         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
219         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
220         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
221         /* 6: DS207 */
222         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
223         /* 16: Vector */
224         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
225         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
226         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
227         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
228         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
229         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
230         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
231         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
232         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
233         /* 42: PM, 44: PC RA, 46: SC vec'd */
234         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
235         /* 48: SIMD, 50: QP BFP, 52: String */
236         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
237         /* 54: DecFP, 56: DecI, 58: SHA */
238         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
239         /* 60: NM atomic, 62: RNG */
240         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
241     };
242     uint8_t *pa_features = NULL;
243     size_t pa_size;
244 
245     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
246         pa_features = pa_features_206;
247         pa_size = sizeof(pa_features_206);
248     }
249     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
250         pa_features = pa_features_207;
251         pa_size = sizeof(pa_features_207);
252     }
253     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
254         pa_features = pa_features_300;
255         pa_size = sizeof(pa_features_300);
256     }
257     if (!pa_features) {
258         return;
259     }
260 
261     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
262         /*
263          * Note: we keep CI large pages off by default because a 64K capable
264          * guest provisioned with large pages might otherwise try to map a qemu
265          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
266          * even if that qemu runs on a 4k host.
267          * We dd this bit back here if we are confident this is not an issue
268          */
269         pa_features[3] |= 0x20;
270     }
271     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
272         pa_features[24] |= 0x80;    /* Transactional memory support */
273     }
274     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
275         /* Workaround for broken kernels that attempt (guest) radix
276          * mode when they can't handle it, if they see the radix bit set
277          * in pa-features. So hide it from them. */
278         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
279     }
280 
281     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
282 }
283 
284 static hwaddr spapr_node0_size(MachineState *machine)
285 {
286     if (machine->numa_state->num_nodes) {
287         int i;
288         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
289             if (machine->numa_state->nodes[i].node_mem) {
290                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
291                            machine->ram_size);
292             }
293         }
294     }
295     return machine->ram_size;
296 }
297 
298 static void add_str(GString *s, const gchar *s1)
299 {
300     g_string_append_len(s, s1, strlen(s1) + 1);
301 }
302 
303 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid,
304                                 hwaddr start, hwaddr size)
305 {
306     char mem_name[32];
307     uint64_t mem_reg_property[2];
308     int off;
309 
310     mem_reg_property[0] = cpu_to_be64(start);
311     mem_reg_property[1] = cpu_to_be64(size);
312 
313     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
314     off = fdt_add_subnode(fdt, 0, mem_name);
315     _FDT(off);
316     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
317     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
318                       sizeof(mem_reg_property))));
319     spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid);
320     return off;
321 }
322 
323 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
324 {
325     MemoryDeviceInfoList *info;
326 
327     for (info = list; info; info = info->next) {
328         MemoryDeviceInfo *value = info->value;
329 
330         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
331             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
332 
333             if (addr >= pcdimm_info->addr &&
334                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
335                 return pcdimm_info->node;
336             }
337         }
338     }
339 
340     return -1;
341 }
342 
343 struct sPAPRDrconfCellV2 {
344      uint32_t seq_lmbs;
345      uint64_t base_addr;
346      uint32_t drc_index;
347      uint32_t aa_index;
348      uint32_t flags;
349 } QEMU_PACKED;
350 
351 typedef struct DrconfCellQueue {
352     struct sPAPRDrconfCellV2 cell;
353     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
354 } DrconfCellQueue;
355 
356 static DrconfCellQueue *
357 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
358                       uint32_t drc_index, uint32_t aa_index,
359                       uint32_t flags)
360 {
361     DrconfCellQueue *elem;
362 
363     elem = g_malloc0(sizeof(*elem));
364     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
365     elem->cell.base_addr = cpu_to_be64(base_addr);
366     elem->cell.drc_index = cpu_to_be32(drc_index);
367     elem->cell.aa_index = cpu_to_be32(aa_index);
368     elem->cell.flags = cpu_to_be32(flags);
369 
370     return elem;
371 }
372 
373 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
374                                       int offset, MemoryDeviceInfoList *dimms)
375 {
376     MachineState *machine = MACHINE(spapr);
377     uint8_t *int_buf, *cur_index;
378     int ret;
379     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
380     uint64_t addr, cur_addr, size;
381     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
382     uint64_t mem_end = machine->device_memory->base +
383                        memory_region_size(&machine->device_memory->mr);
384     uint32_t node, buf_len, nr_entries = 0;
385     SpaprDrc *drc;
386     DrconfCellQueue *elem, *next;
387     MemoryDeviceInfoList *info;
388     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
389         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
390 
391     /* Entry to cover RAM and the gap area */
392     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
393                                  SPAPR_LMB_FLAGS_RESERVED |
394                                  SPAPR_LMB_FLAGS_DRC_INVALID);
395     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
396     nr_entries++;
397 
398     cur_addr = machine->device_memory->base;
399     for (info = dimms; info; info = info->next) {
400         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
401 
402         addr = di->addr;
403         size = di->size;
404         node = di->node;
405 
406         /*
407          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
408          * area is marked hotpluggable in the next iteration for the bigger
409          * chunk including the NVDIMM occupied area.
410          */
411         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
412             continue;
413 
414         /* Entry for hot-pluggable area */
415         if (cur_addr < addr) {
416             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
417             g_assert(drc);
418             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
419                                          cur_addr, spapr_drc_index(drc), -1, 0);
420             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
421             nr_entries++;
422         }
423 
424         /* Entry for DIMM */
425         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
426         g_assert(drc);
427         elem = spapr_get_drconf_cell(size / lmb_size, addr,
428                                      spapr_drc_index(drc), node,
429                                      (SPAPR_LMB_FLAGS_ASSIGNED |
430                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
431         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
432         nr_entries++;
433         cur_addr = addr + size;
434     }
435 
436     /* Entry for remaining hotpluggable area */
437     if (cur_addr < mem_end) {
438         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
439         g_assert(drc);
440         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
441                                      cur_addr, spapr_drc_index(drc), -1, 0);
442         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
443         nr_entries++;
444     }
445 
446     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
447     int_buf = cur_index = g_malloc0(buf_len);
448     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
449     cur_index += sizeof(nr_entries);
450 
451     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
452         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
453         cur_index += sizeof(elem->cell);
454         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
455         g_free(elem);
456     }
457 
458     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
459     g_free(int_buf);
460     if (ret < 0) {
461         return -1;
462     }
463     return 0;
464 }
465 
466 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
467                                    int offset, MemoryDeviceInfoList *dimms)
468 {
469     MachineState *machine = MACHINE(spapr);
470     int i, ret;
471     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
472     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
473     uint32_t nr_lmbs = (machine->device_memory->base +
474                        memory_region_size(&machine->device_memory->mr)) /
475                        lmb_size;
476     uint32_t *int_buf, *cur_index, buf_len;
477 
478     /*
479      * Allocate enough buffer size to fit in ibm,dynamic-memory
480      */
481     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
482     cur_index = int_buf = g_malloc0(buf_len);
483     int_buf[0] = cpu_to_be32(nr_lmbs);
484     cur_index++;
485     for (i = 0; i < nr_lmbs; i++) {
486         uint64_t addr = i * lmb_size;
487         uint32_t *dynamic_memory = cur_index;
488 
489         if (i >= device_lmb_start) {
490             SpaprDrc *drc;
491 
492             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
493             g_assert(drc);
494 
495             dynamic_memory[0] = cpu_to_be32(addr >> 32);
496             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
497             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
498             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
499             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
500             if (memory_region_present(get_system_memory(), addr)) {
501                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
502             } else {
503                 dynamic_memory[5] = cpu_to_be32(0);
504             }
505         } else {
506             /*
507              * LMB information for RMA, boot time RAM and gap b/n RAM and
508              * device memory region -- all these are marked as reserved
509              * and as having no valid DRC.
510              */
511             dynamic_memory[0] = cpu_to_be32(addr >> 32);
512             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
513             dynamic_memory[2] = cpu_to_be32(0);
514             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
515             dynamic_memory[4] = cpu_to_be32(-1);
516             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
517                                             SPAPR_LMB_FLAGS_DRC_INVALID);
518         }
519 
520         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
521     }
522     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
523     g_free(int_buf);
524     if (ret < 0) {
525         return -1;
526     }
527     return 0;
528 }
529 
530 /*
531  * Adds ibm,dynamic-reconfiguration-memory node.
532  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
533  * of this device tree node.
534  */
535 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
536                                                    void *fdt)
537 {
538     MachineState *machine = MACHINE(spapr);
539     int ret, offset;
540     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
541     uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
542                                 cpu_to_be32(lmb_size & 0xffffffff)};
543     MemoryDeviceInfoList *dimms = NULL;
544 
545     /*
546      * Don't create the node if there is no device memory
547      */
548     if (machine->ram_size == machine->maxram_size) {
549         return 0;
550     }
551 
552     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
553 
554     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
555                     sizeof(prop_lmb_size));
556     if (ret < 0) {
557         return ret;
558     }
559 
560     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
561     if (ret < 0) {
562         return ret;
563     }
564 
565     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
566     if (ret < 0) {
567         return ret;
568     }
569 
570     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
571     dimms = qmp_memory_device_list();
572     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
573         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
574     } else {
575         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
576     }
577     qapi_free_MemoryDeviceInfoList(dimms);
578 
579     if (ret < 0) {
580         return ret;
581     }
582 
583     ret = spapr_numa_write_assoc_lookup_arrays(spapr, fdt, offset);
584 
585     return ret;
586 }
587 
588 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
589 {
590     MachineState *machine = MACHINE(spapr);
591     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
592     hwaddr mem_start, node_size;
593     int i, nb_nodes = machine->numa_state->num_nodes;
594     NodeInfo *nodes = machine->numa_state->nodes;
595 
596     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
597         if (!nodes[i].node_mem) {
598             continue;
599         }
600         if (mem_start >= machine->ram_size) {
601             node_size = 0;
602         } else {
603             node_size = nodes[i].node_mem;
604             if (node_size > machine->ram_size - mem_start) {
605                 node_size = machine->ram_size - mem_start;
606             }
607         }
608         if (!mem_start) {
609             /* spapr_machine_init() checks for rma_size <= node0_size
610              * already */
611             spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size);
612             mem_start += spapr->rma_size;
613             node_size -= spapr->rma_size;
614         }
615         for ( ; node_size; ) {
616             hwaddr sizetmp = pow2floor(node_size);
617 
618             /* mem_start != 0 here */
619             if (ctzl(mem_start) < ctzl(sizetmp)) {
620                 sizetmp = 1ULL << ctzl(mem_start);
621             }
622 
623             spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp);
624             node_size -= sizetmp;
625             mem_start += sizetmp;
626         }
627     }
628 
629     /* Generate ibm,dynamic-reconfiguration-memory node if required */
630     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
631         int ret;
632 
633         g_assert(smc->dr_lmb_enabled);
634         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
635         if (ret) {
636             return ret;
637         }
638     }
639 
640     return 0;
641 }
642 
643 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
644                          SpaprMachineState *spapr)
645 {
646     MachineState *ms = MACHINE(spapr);
647     PowerPCCPU *cpu = POWERPC_CPU(cs);
648     CPUPPCState *env = &cpu->env;
649     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
650     int index = spapr_get_vcpu_id(cpu);
651     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
652                        0xffffffff, 0xffffffff};
653     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
654         : SPAPR_TIMEBASE_FREQ;
655     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
656     uint32_t page_sizes_prop[64];
657     size_t page_sizes_prop_size;
658     unsigned int smp_threads = ms->smp.threads;
659     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
660     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
661     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
662     SpaprDrc *drc;
663     int drc_index;
664     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
665     int i;
666 
667     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
668     if (drc) {
669         drc_index = spapr_drc_index(drc);
670         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
671     }
672 
673     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
674     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
675 
676     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
677     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
678                            env->dcache_line_size)));
679     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
680                            env->dcache_line_size)));
681     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
682                            env->icache_line_size)));
683     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
684                            env->icache_line_size)));
685 
686     if (pcc->l1_dcache_size) {
687         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
688                                pcc->l1_dcache_size)));
689     } else {
690         warn_report("Unknown L1 dcache size for cpu");
691     }
692     if (pcc->l1_icache_size) {
693         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
694                                pcc->l1_icache_size)));
695     } else {
696         warn_report("Unknown L1 icache size for cpu");
697     }
698 
699     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
700     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
701     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
702     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
703     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
704     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
705 
706     if (ppc_has_spr(cpu, SPR_PURR)) {
707         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
708     }
709     if (ppc_has_spr(cpu, SPR_PURR)) {
710         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
711     }
712 
713     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
714         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
715                           segs, sizeof(segs))));
716     }
717 
718     /* Advertise VSX (vector extensions) if available
719      *   1               == VMX / Altivec available
720      *   2               == VSX available
721      *
722      * Only CPUs for which we create core types in spapr_cpu_core.c
723      * are possible, and all of those have VMX */
724     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
725         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
726     } else {
727         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
728     }
729 
730     /* Advertise DFP (Decimal Floating Point) if available
731      *   0 / no property == no DFP
732      *   1               == DFP available */
733     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
734         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
735     }
736 
737     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
738                                                       sizeof(page_sizes_prop));
739     if (page_sizes_prop_size) {
740         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
741                           page_sizes_prop, page_sizes_prop_size)));
742     }
743 
744     spapr_dt_pa_features(spapr, cpu, fdt, offset);
745 
746     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
747                            cs->cpu_index / vcpus_per_socket)));
748 
749     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
750                       pft_size_prop, sizeof(pft_size_prop))));
751 
752     if (ms->numa_state->num_nodes > 1) {
753         _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
754     }
755 
756     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
757 
758     if (pcc->radix_page_info) {
759         for (i = 0; i < pcc->radix_page_info->count; i++) {
760             radix_AP_encodings[i] =
761                 cpu_to_be32(pcc->radix_page_info->entries[i]);
762         }
763         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
764                           radix_AP_encodings,
765                           pcc->radix_page_info->count *
766                           sizeof(radix_AP_encodings[0]))));
767     }
768 
769     /*
770      * We set this property to let the guest know that it can use the large
771      * decrementer and its width in bits.
772      */
773     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
774         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
775                               pcc->lrg_decr_bits)));
776 }
777 
778 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
779 {
780     CPUState **rev;
781     CPUState *cs;
782     int n_cpus;
783     int cpus_offset;
784     int i;
785 
786     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
787     _FDT(cpus_offset);
788     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
789     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
790 
791     /*
792      * We walk the CPUs in reverse order to ensure that CPU DT nodes
793      * created by fdt_add_subnode() end up in the right order in FDT
794      * for the guest kernel the enumerate the CPUs correctly.
795      *
796      * The CPU list cannot be traversed in reverse order, so we need
797      * to do extra work.
798      */
799     n_cpus = 0;
800     rev = NULL;
801     CPU_FOREACH(cs) {
802         rev = g_renew(CPUState *, rev, n_cpus + 1);
803         rev[n_cpus++] = cs;
804     }
805 
806     for (i = n_cpus - 1; i >= 0; i--) {
807         CPUState *cs = rev[i];
808         PowerPCCPU *cpu = POWERPC_CPU(cs);
809         int index = spapr_get_vcpu_id(cpu);
810         DeviceClass *dc = DEVICE_GET_CLASS(cs);
811         g_autofree char *nodename = NULL;
812         int offset;
813 
814         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
815             continue;
816         }
817 
818         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
819         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
820         _FDT(offset);
821         spapr_dt_cpu(cs, fdt, offset, spapr);
822     }
823 
824     g_free(rev);
825 }
826 
827 static int spapr_dt_rng(void *fdt)
828 {
829     int node;
830     int ret;
831 
832     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
833     if (node <= 0) {
834         return -1;
835     }
836     ret = fdt_setprop_string(fdt, node, "device_type",
837                              "ibm,platform-facilities");
838     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
839     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
840 
841     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
842     if (node <= 0) {
843         return -1;
844     }
845     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
846 
847     return ret ? -1 : 0;
848 }
849 
850 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
851 {
852     MachineState *ms = MACHINE(spapr);
853     int rtas;
854     GString *hypertas = g_string_sized_new(256);
855     GString *qemu_hypertas = g_string_sized_new(256);
856     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
857         memory_region_size(&MACHINE(spapr)->device_memory->mr);
858     uint32_t lrdr_capacity[] = {
859         cpu_to_be32(max_device_addr >> 32),
860         cpu_to_be32(max_device_addr & 0xffffffff),
861         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
862         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
863         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
864     };
865 
866     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
867 
868     /* hypertas */
869     add_str(hypertas, "hcall-pft");
870     add_str(hypertas, "hcall-term");
871     add_str(hypertas, "hcall-dabr");
872     add_str(hypertas, "hcall-interrupt");
873     add_str(hypertas, "hcall-tce");
874     add_str(hypertas, "hcall-vio");
875     add_str(hypertas, "hcall-splpar");
876     add_str(hypertas, "hcall-join");
877     add_str(hypertas, "hcall-bulk");
878     add_str(hypertas, "hcall-set-mode");
879     add_str(hypertas, "hcall-sprg0");
880     add_str(hypertas, "hcall-copy");
881     add_str(hypertas, "hcall-debug");
882     add_str(hypertas, "hcall-vphn");
883     add_str(qemu_hypertas, "hcall-memop1");
884 
885     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
886         add_str(hypertas, "hcall-multi-tce");
887     }
888 
889     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
890         add_str(hypertas, "hcall-hpt-resize");
891     }
892 
893     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
894                      hypertas->str, hypertas->len));
895     g_string_free(hypertas, TRUE);
896     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
897                      qemu_hypertas->str, qemu_hypertas->len));
898     g_string_free(qemu_hypertas, TRUE);
899 
900     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
901 
902     /*
903      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
904      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
905      *
906      * The system reset requirements are driven by existing Linux and PowerVM
907      * implementation which (contrary to PAPR) saves r3 in the error log
908      * structure like machine check, so Linux expects to find the saved r3
909      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
910      * does not look at the error value).
911      *
912      * System reset interrupts are not subject to interlock like machine
913      * check, so this memory area could be corrupted if the sreset is
914      * interrupted by a machine check (or vice versa) if it was shared. To
915      * prevent this, system reset uses per-CPU areas for the sreset save
916      * area. A system reset that interrupts a system reset handler could
917      * still overwrite this area, but Linux doesn't try to recover in that
918      * case anyway.
919      *
920      * The extra 8 bytes is required because Linux's FWNMI error log check
921      * is off-by-one.
922      *
923      * RTAS_MIN_SIZE is required for the RTAS blob itself.
924      */
925     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_MIN_SIZE +
926                           RTAS_ERROR_LOG_MAX +
927                           ms->smp.max_cpus * sizeof(uint64_t) * 2 +
928                           sizeof(uint64_t)));
929     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
930                           RTAS_ERROR_LOG_MAX));
931     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
932                           RTAS_EVENT_SCAN_RATE));
933 
934     g_assert(msi_nonbroken);
935     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
936 
937     /*
938      * According to PAPR, rtas ibm,os-term does not guarantee a return
939      * back to the guest cpu.
940      *
941      * While an additional ibm,extended-os-term property indicates
942      * that rtas call return will always occur. Set this property.
943      */
944     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
945 
946     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
947                      lrdr_capacity, sizeof(lrdr_capacity)));
948 
949     spapr_dt_rtas_tokens(fdt, rtas);
950 }
951 
952 /*
953  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
954  * and the XIVE features that the guest may request and thus the valid
955  * values for bytes 23..26 of option vector 5:
956  */
957 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
958                                           int chosen)
959 {
960     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
961 
962     char val[2 * 4] = {
963         23, 0x00, /* XICS / XIVE mode */
964         24, 0x00, /* Hash/Radix, filled in below. */
965         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
966         26, 0x40, /* Radix options: GTSE == yes. */
967     };
968 
969     if (spapr->irq->xics && spapr->irq->xive) {
970         val[1] = SPAPR_OV5_XIVE_BOTH;
971     } else if (spapr->irq->xive) {
972         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
973     } else {
974         assert(spapr->irq->xics);
975         val[1] = SPAPR_OV5_XIVE_LEGACY;
976     }
977 
978     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
979                           first_ppc_cpu->compat_pvr)) {
980         /*
981          * If we're in a pre POWER9 compat mode then the guest should
982          * do hash and use the legacy interrupt mode
983          */
984         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
985         val[3] = 0x00; /* Hash */
986         spapr_check_mmu_mode(false);
987     } else if (kvm_enabled()) {
988         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
989             val[3] = 0x80; /* OV5_MMU_BOTH */
990         } else if (kvmppc_has_cap_mmu_radix()) {
991             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
992         } else {
993             val[3] = 0x00; /* Hash */
994         }
995     } else {
996         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
997         val[3] = 0xC0;
998     }
999     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1000                      val, sizeof(val)));
1001 }
1002 
1003 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1004 {
1005     MachineState *machine = MACHINE(spapr);
1006     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1007     int chosen;
1008 
1009     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1010 
1011     if (reset) {
1012         const char *boot_device = spapr->boot_device;
1013         char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1014         size_t cb = 0;
1015         char *bootlist = get_boot_devices_list(&cb);
1016 
1017         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1018             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1019                                     machine->kernel_cmdline));
1020         }
1021 
1022         if (spapr->initrd_size) {
1023             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1024                                   spapr->initrd_base));
1025             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1026                                   spapr->initrd_base + spapr->initrd_size));
1027         }
1028 
1029         if (spapr->kernel_size) {
1030             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1031                                   cpu_to_be64(spapr->kernel_size) };
1032 
1033             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1034                          &kprop, sizeof(kprop)));
1035             if (spapr->kernel_le) {
1036                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1037             }
1038         }
1039         if (boot_menu) {
1040             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1041         }
1042         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1043         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1044         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1045 
1046         if (cb && bootlist) {
1047             int i;
1048 
1049             for (i = 0; i < cb; i++) {
1050                 if (bootlist[i] == '\n') {
1051                     bootlist[i] = ' ';
1052                 }
1053             }
1054             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1055         }
1056 
1057         if (boot_device && strlen(boot_device)) {
1058             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1059         }
1060 
1061         if (!spapr->has_graphics && stdout_path) {
1062             /*
1063              * "linux,stdout-path" and "stdout" properties are
1064              * deprecated by linux kernel. New platforms should only
1065              * use the "stdout-path" property. Set the new property
1066              * and continue using older property to remain compatible
1067              * with the existing firmware.
1068              */
1069             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1070             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1071         }
1072 
1073         /*
1074          * We can deal with BAR reallocation just fine, advertise it
1075          * to the guest
1076          */
1077         if (smc->linux_pci_probe) {
1078             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1079         }
1080 
1081         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1082 
1083         g_free(stdout_path);
1084         g_free(bootlist);
1085     }
1086 
1087     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1088 }
1089 
1090 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1091 {
1092     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1093      * KVM to work under pHyp with some guest co-operation */
1094     int hypervisor;
1095     uint8_t hypercall[16];
1096 
1097     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1098     /* indicate KVM hypercall interface */
1099     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1100     if (kvmppc_has_cap_fixup_hcalls()) {
1101         /*
1102          * Older KVM versions with older guest kernels were broken
1103          * with the magic page, don't allow the guest to map it.
1104          */
1105         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1106                                   sizeof(hypercall))) {
1107             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1108                              hypercall, sizeof(hypercall)));
1109         }
1110     }
1111 }
1112 
1113 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1114 {
1115     MachineState *machine = MACHINE(spapr);
1116     MachineClass *mc = MACHINE_GET_CLASS(machine);
1117     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1118     uint32_t root_drc_type_mask = 0;
1119     int ret;
1120     void *fdt;
1121     SpaprPhbState *phb;
1122     char *buf;
1123 
1124     fdt = g_malloc0(space);
1125     _FDT((fdt_create_empty_tree(fdt, space)));
1126 
1127     /* Root node */
1128     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1129     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1130     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1131 
1132     /* Guest UUID & Name*/
1133     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1134     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1135     if (qemu_uuid_set) {
1136         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1137     }
1138     g_free(buf);
1139 
1140     if (qemu_get_vm_name()) {
1141         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1142                                 qemu_get_vm_name()));
1143     }
1144 
1145     /* Host Model & Serial Number */
1146     if (spapr->host_model) {
1147         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1148     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1149         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1150         g_free(buf);
1151     }
1152 
1153     if (spapr->host_serial) {
1154         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1155     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1156         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1157         g_free(buf);
1158     }
1159 
1160     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1161     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1162 
1163     /* /interrupt controller */
1164     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1165 
1166     ret = spapr_dt_memory(spapr, fdt);
1167     if (ret < 0) {
1168         error_report("couldn't setup memory nodes in fdt");
1169         exit(1);
1170     }
1171 
1172     /* /vdevice */
1173     spapr_dt_vdevice(spapr->vio_bus, fdt);
1174 
1175     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1176         ret = spapr_dt_rng(fdt);
1177         if (ret < 0) {
1178             error_report("could not set up rng device in the fdt");
1179             exit(1);
1180         }
1181     }
1182 
1183     QLIST_FOREACH(phb, &spapr->phbs, list) {
1184         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1185         if (ret < 0) {
1186             error_report("couldn't setup PCI devices in fdt");
1187             exit(1);
1188         }
1189     }
1190 
1191     spapr_dt_cpus(fdt, spapr);
1192 
1193     /* ibm,drc-indexes and friends */
1194     if (smc->dr_lmb_enabled) {
1195         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_LMB;
1196     }
1197     if (smc->dr_phb_enabled) {
1198         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PHB;
1199     }
1200     if (mc->nvdimm_supported) {
1201         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PMEM;
1202     }
1203     if (root_drc_type_mask) {
1204         _FDT(spapr_dt_drc(fdt, 0, NULL, root_drc_type_mask));
1205     }
1206 
1207     if (mc->has_hotpluggable_cpus) {
1208         int offset = fdt_path_offset(fdt, "/cpus");
1209         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1210         if (ret < 0) {
1211             error_report("Couldn't set up CPU DR device tree properties");
1212             exit(1);
1213         }
1214     }
1215 
1216     /* /event-sources */
1217     spapr_dt_events(spapr, fdt);
1218 
1219     /* /rtas */
1220     spapr_dt_rtas(spapr, fdt);
1221 
1222     /* /chosen */
1223     spapr_dt_chosen(spapr, fdt, reset);
1224 
1225     /* /hypervisor */
1226     if (kvm_enabled()) {
1227         spapr_dt_hypervisor(spapr, fdt);
1228     }
1229 
1230     /* Build memory reserve map */
1231     if (reset) {
1232         if (spapr->kernel_size) {
1233             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1234                                   spapr->kernel_size)));
1235         }
1236         if (spapr->initrd_size) {
1237             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1238                                   spapr->initrd_size)));
1239         }
1240     }
1241 
1242     /* NVDIMM devices */
1243     if (mc->nvdimm_supported) {
1244         spapr_dt_persistent_memory(spapr, fdt);
1245     }
1246 
1247     return fdt;
1248 }
1249 
1250 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1251 {
1252     SpaprMachineState *spapr = opaque;
1253 
1254     return (addr & 0x0fffffff) + spapr->kernel_addr;
1255 }
1256 
1257 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1258                                     PowerPCCPU *cpu)
1259 {
1260     CPUPPCState *env = &cpu->env;
1261 
1262     /* The TCG path should also be holding the BQL at this point */
1263     g_assert(qemu_mutex_iothread_locked());
1264 
1265     if (msr_pr) {
1266         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1267         env->gpr[3] = H_PRIVILEGE;
1268     } else {
1269         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1270     }
1271 }
1272 
1273 struct LPCRSyncState {
1274     target_ulong value;
1275     target_ulong mask;
1276 };
1277 
1278 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1279 {
1280     struct LPCRSyncState *s = arg.host_ptr;
1281     PowerPCCPU *cpu = POWERPC_CPU(cs);
1282     CPUPPCState *env = &cpu->env;
1283     target_ulong lpcr;
1284 
1285     cpu_synchronize_state(cs);
1286     lpcr = env->spr[SPR_LPCR];
1287     lpcr &= ~s->mask;
1288     lpcr |= s->value;
1289     ppc_store_lpcr(cpu, lpcr);
1290 }
1291 
1292 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1293 {
1294     CPUState *cs;
1295     struct LPCRSyncState s = {
1296         .value = value,
1297         .mask = mask
1298     };
1299     CPU_FOREACH(cs) {
1300         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1301     }
1302 }
1303 
1304 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1305 {
1306     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1307 
1308     /* Copy PATE1:GR into PATE0:HR */
1309     entry->dw0 = spapr->patb_entry & PATE0_HR;
1310     entry->dw1 = spapr->patb_entry;
1311 }
1312 
1313 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1314 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1315 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1316 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1317 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1318 
1319 /*
1320  * Get the fd to access the kernel htab, re-opening it if necessary
1321  */
1322 static int get_htab_fd(SpaprMachineState *spapr)
1323 {
1324     Error *local_err = NULL;
1325 
1326     if (spapr->htab_fd >= 0) {
1327         return spapr->htab_fd;
1328     }
1329 
1330     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1331     if (spapr->htab_fd < 0) {
1332         error_report_err(local_err);
1333     }
1334 
1335     return spapr->htab_fd;
1336 }
1337 
1338 void close_htab_fd(SpaprMachineState *spapr)
1339 {
1340     if (spapr->htab_fd >= 0) {
1341         close(spapr->htab_fd);
1342     }
1343     spapr->htab_fd = -1;
1344 }
1345 
1346 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1347 {
1348     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1349 
1350     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1351 }
1352 
1353 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1354 {
1355     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1356 
1357     assert(kvm_enabled());
1358 
1359     if (!spapr->htab) {
1360         return 0;
1361     }
1362 
1363     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1364 }
1365 
1366 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1367                                                 hwaddr ptex, int n)
1368 {
1369     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1370     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1371 
1372     if (!spapr->htab) {
1373         /*
1374          * HTAB is controlled by KVM. Fetch into temporary buffer
1375          */
1376         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1377         kvmppc_read_hptes(hptes, ptex, n);
1378         return hptes;
1379     }
1380 
1381     /*
1382      * HTAB is controlled by QEMU. Just point to the internally
1383      * accessible PTEG.
1384      */
1385     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1386 }
1387 
1388 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1389                               const ppc_hash_pte64_t *hptes,
1390                               hwaddr ptex, int n)
1391 {
1392     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1393 
1394     if (!spapr->htab) {
1395         g_free((void *)hptes);
1396     }
1397 
1398     /* Nothing to do for qemu managed HPT */
1399 }
1400 
1401 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1402                       uint64_t pte0, uint64_t pte1)
1403 {
1404     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1405     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1406 
1407     if (!spapr->htab) {
1408         kvmppc_write_hpte(ptex, pte0, pte1);
1409     } else {
1410         if (pte0 & HPTE64_V_VALID) {
1411             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1412             /*
1413              * When setting valid, we write PTE1 first. This ensures
1414              * proper synchronization with the reading code in
1415              * ppc_hash64_pteg_search()
1416              */
1417             smp_wmb();
1418             stq_p(spapr->htab + offset, pte0);
1419         } else {
1420             stq_p(spapr->htab + offset, pte0);
1421             /*
1422              * When clearing it we set PTE0 first. This ensures proper
1423              * synchronization with the reading code in
1424              * ppc_hash64_pteg_search()
1425              */
1426             smp_wmb();
1427             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1428         }
1429     }
1430 }
1431 
1432 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1433                              uint64_t pte1)
1434 {
1435     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1436     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1437 
1438     if (!spapr->htab) {
1439         /* There should always be a hash table when this is called */
1440         error_report("spapr_hpte_set_c called with no hash table !");
1441         return;
1442     }
1443 
1444     /* The HW performs a non-atomic byte update */
1445     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1446 }
1447 
1448 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1449                              uint64_t pte1)
1450 {
1451     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1452     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1453 
1454     if (!spapr->htab) {
1455         /* There should always be a hash table when this is called */
1456         error_report("spapr_hpte_set_r called with no hash table !");
1457         return;
1458     }
1459 
1460     /* The HW performs a non-atomic byte update */
1461     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1462 }
1463 
1464 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1465 {
1466     int shift;
1467 
1468     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1469      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1470      * that's much more than is needed for Linux guests */
1471     shift = ctz64(pow2ceil(ramsize)) - 7;
1472     shift = MAX(shift, 18); /* Minimum architected size */
1473     shift = MIN(shift, 46); /* Maximum architected size */
1474     return shift;
1475 }
1476 
1477 void spapr_free_hpt(SpaprMachineState *spapr)
1478 {
1479     g_free(spapr->htab);
1480     spapr->htab = NULL;
1481     spapr->htab_shift = 0;
1482     close_htab_fd(spapr);
1483 }
1484 
1485 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1486 {
1487     ERRP_GUARD();
1488     long rc;
1489 
1490     /* Clean up any HPT info from a previous boot */
1491     spapr_free_hpt(spapr);
1492 
1493     rc = kvmppc_reset_htab(shift);
1494 
1495     if (rc == -EOPNOTSUPP) {
1496         error_setg(errp, "HPT not supported in nested guests");
1497         return -EOPNOTSUPP;
1498     }
1499 
1500     if (rc < 0) {
1501         /* kernel-side HPT needed, but couldn't allocate one */
1502         error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1503                          shift);
1504         error_append_hint(errp, "Try smaller maxmem?\n");
1505         return -errno;
1506     } else if (rc > 0) {
1507         /* kernel-side HPT allocated */
1508         if (rc != shift) {
1509             error_setg(errp,
1510                        "Requested order %d HPT, but kernel allocated order %ld",
1511                        shift, rc);
1512             error_append_hint(errp, "Try smaller maxmem?\n");
1513             return -ENOSPC;
1514         }
1515 
1516         spapr->htab_shift = shift;
1517         spapr->htab = NULL;
1518     } else {
1519         /* kernel-side HPT not needed, allocate in userspace instead */
1520         size_t size = 1ULL << shift;
1521         int i;
1522 
1523         spapr->htab = qemu_memalign(size, size);
1524         memset(spapr->htab, 0, size);
1525         spapr->htab_shift = shift;
1526 
1527         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1528             DIRTY_HPTE(HPTE(spapr->htab, i));
1529         }
1530     }
1531     /* We're setting up a hash table, so that means we're not radix */
1532     spapr->patb_entry = 0;
1533     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1534     return 0;
1535 }
1536 
1537 void spapr_setup_hpt(SpaprMachineState *spapr)
1538 {
1539     int hpt_shift;
1540 
1541     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1542         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1543     } else {
1544         uint64_t current_ram_size;
1545 
1546         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1547         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1548     }
1549     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1550 
1551     if (kvm_enabled()) {
1552         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1553 
1554         /* Check our RMA fits in the possible VRMA */
1555         if (vrma_limit < spapr->rma_size) {
1556             error_report("Unable to create %" HWADDR_PRIu
1557                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1558                          spapr->rma_size / MiB, vrma_limit / MiB);
1559             exit(EXIT_FAILURE);
1560         }
1561     }
1562 }
1563 
1564 void spapr_check_mmu_mode(bool guest_radix)
1565 {
1566     if (guest_radix) {
1567         if (kvm_enabled() && !kvmppc_has_cap_mmu_radix()) {
1568             error_report("Guest requested unavailable MMU mode (radix).");
1569             exit(EXIT_FAILURE);
1570         }
1571     } else {
1572         if (kvm_enabled() && kvmppc_has_cap_mmu_radix()
1573             && !kvmppc_has_cap_mmu_hash_v3()) {
1574             error_report("Guest requested unavailable MMU mode (hash).");
1575             exit(EXIT_FAILURE);
1576         }
1577     }
1578 }
1579 
1580 static void spapr_machine_reset(MachineState *machine)
1581 {
1582     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1583     PowerPCCPU *first_ppc_cpu;
1584     hwaddr fdt_addr;
1585     void *fdt;
1586     int rc;
1587 
1588     pef_kvm_reset(machine->cgs, &error_fatal);
1589     spapr_caps_apply(spapr);
1590 
1591     first_ppc_cpu = POWERPC_CPU(first_cpu);
1592     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1593         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1594                               spapr->max_compat_pvr)) {
1595         /*
1596          * If using KVM with radix mode available, VCPUs can be started
1597          * without a HPT because KVM will start them in radix mode.
1598          * Set the GR bit in PATE so that we know there is no HPT.
1599          */
1600         spapr->patb_entry = PATE1_GR;
1601         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1602     } else {
1603         spapr_setup_hpt(spapr);
1604     }
1605 
1606     qemu_devices_reset();
1607 
1608     spapr_ovec_cleanup(spapr->ov5_cas);
1609     spapr->ov5_cas = spapr_ovec_new();
1610 
1611     ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1612 
1613     /*
1614      * This is fixing some of the default configuration of the XIVE
1615      * devices. To be called after the reset of the machine devices.
1616      */
1617     spapr_irq_reset(spapr, &error_fatal);
1618 
1619     /*
1620      * There is no CAS under qtest. Simulate one to please the code that
1621      * depends on spapr->ov5_cas. This is especially needed to test device
1622      * unplug, so we do that before resetting the DRCs.
1623      */
1624     if (qtest_enabled()) {
1625         spapr_ovec_cleanup(spapr->ov5_cas);
1626         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1627     }
1628 
1629     /* DRC reset may cause a device to be unplugged. This will cause troubles
1630      * if this device is used by another device (eg, a running vhost backend
1631      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1632      * situations, we reset DRCs after all devices have been reset.
1633      */
1634     spapr_drc_reset_all(spapr);
1635 
1636     spapr_clear_pending_events(spapr);
1637 
1638     /*
1639      * We place the device tree just below either the top of the RMA,
1640      * or just below 2GB, whichever is lower, so that it can be
1641      * processed with 32-bit real mode code if necessary
1642      */
1643     fdt_addr = MIN(spapr->rma_size, FDT_MAX_ADDR) - FDT_MAX_SIZE;
1644 
1645     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1646 
1647     rc = fdt_pack(fdt);
1648 
1649     /* Should only fail if we've built a corrupted tree */
1650     assert(rc == 0);
1651 
1652     /* Load the fdt */
1653     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1654     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1655     g_free(spapr->fdt_blob);
1656     spapr->fdt_size = fdt_totalsize(fdt);
1657     spapr->fdt_initial_size = spapr->fdt_size;
1658     spapr->fdt_blob = fdt;
1659 
1660     /* Set up the entry state */
1661     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, 0, fdt_addr, 0);
1662     first_ppc_cpu->env.gpr[5] = 0;
1663 
1664     spapr->fwnmi_system_reset_addr = -1;
1665     spapr->fwnmi_machine_check_addr = -1;
1666     spapr->fwnmi_machine_check_interlock = -1;
1667 
1668     /* Signal all vCPUs waiting on this condition */
1669     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1670 
1671     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1672 }
1673 
1674 static void spapr_create_nvram(SpaprMachineState *spapr)
1675 {
1676     DeviceState *dev = qdev_new("spapr-nvram");
1677     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1678 
1679     if (dinfo) {
1680         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1681                                 &error_fatal);
1682     }
1683 
1684     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1685 
1686     spapr->nvram = (struct SpaprNvram *)dev;
1687 }
1688 
1689 static void spapr_rtc_create(SpaprMachineState *spapr)
1690 {
1691     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1692                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1693                                        &error_fatal, NULL);
1694     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1695     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1696                               "date");
1697 }
1698 
1699 /* Returns whether we want to use VGA or not */
1700 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1701 {
1702     switch (vga_interface_type) {
1703     case VGA_NONE:
1704         return false;
1705     case VGA_DEVICE:
1706         return true;
1707     case VGA_STD:
1708     case VGA_VIRTIO:
1709     case VGA_CIRRUS:
1710         return pci_vga_init(pci_bus) != NULL;
1711     default:
1712         error_setg(errp,
1713                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1714         return false;
1715     }
1716 }
1717 
1718 static int spapr_pre_load(void *opaque)
1719 {
1720     int rc;
1721 
1722     rc = spapr_caps_pre_load(opaque);
1723     if (rc) {
1724         return rc;
1725     }
1726 
1727     return 0;
1728 }
1729 
1730 static int spapr_post_load(void *opaque, int version_id)
1731 {
1732     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1733     int err = 0;
1734 
1735     err = spapr_caps_post_migration(spapr);
1736     if (err) {
1737         return err;
1738     }
1739 
1740     /*
1741      * In earlier versions, there was no separate qdev for the PAPR
1742      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1743      * So when migrating from those versions, poke the incoming offset
1744      * value into the RTC device
1745      */
1746     if (version_id < 3) {
1747         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1748         if (err) {
1749             return err;
1750         }
1751     }
1752 
1753     if (kvm_enabled() && spapr->patb_entry) {
1754         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1755         bool radix = !!(spapr->patb_entry & PATE1_GR);
1756         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1757 
1758         /*
1759          * Update LPCR:HR and UPRT as they may not be set properly in
1760          * the stream
1761          */
1762         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1763                             LPCR_HR | LPCR_UPRT);
1764 
1765         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1766         if (err) {
1767             error_report("Process table config unsupported by the host");
1768             return -EINVAL;
1769         }
1770     }
1771 
1772     err = spapr_irq_post_load(spapr, version_id);
1773     if (err) {
1774         return err;
1775     }
1776 
1777     return err;
1778 }
1779 
1780 static int spapr_pre_save(void *opaque)
1781 {
1782     int rc;
1783 
1784     rc = spapr_caps_pre_save(opaque);
1785     if (rc) {
1786         return rc;
1787     }
1788 
1789     return 0;
1790 }
1791 
1792 static bool version_before_3(void *opaque, int version_id)
1793 {
1794     return version_id < 3;
1795 }
1796 
1797 static bool spapr_pending_events_needed(void *opaque)
1798 {
1799     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1800     return !QTAILQ_EMPTY(&spapr->pending_events);
1801 }
1802 
1803 static const VMStateDescription vmstate_spapr_event_entry = {
1804     .name = "spapr_event_log_entry",
1805     .version_id = 1,
1806     .minimum_version_id = 1,
1807     .fields = (VMStateField[]) {
1808         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1809         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1810         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1811                                      NULL, extended_length),
1812         VMSTATE_END_OF_LIST()
1813     },
1814 };
1815 
1816 static const VMStateDescription vmstate_spapr_pending_events = {
1817     .name = "spapr_pending_events",
1818     .version_id = 1,
1819     .minimum_version_id = 1,
1820     .needed = spapr_pending_events_needed,
1821     .fields = (VMStateField[]) {
1822         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1823                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1824         VMSTATE_END_OF_LIST()
1825     },
1826 };
1827 
1828 static bool spapr_ov5_cas_needed(void *opaque)
1829 {
1830     SpaprMachineState *spapr = opaque;
1831     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1832     bool cas_needed;
1833 
1834     /* Prior to the introduction of SpaprOptionVector, we had two option
1835      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1836      * Both of these options encode machine topology into the device-tree
1837      * in such a way that the now-booted OS should still be able to interact
1838      * appropriately with QEMU regardless of what options were actually
1839      * negotiatied on the source side.
1840      *
1841      * As such, we can avoid migrating the CAS-negotiated options if these
1842      * are the only options available on the current machine/platform.
1843      * Since these are the only options available for pseries-2.7 and
1844      * earlier, this allows us to maintain old->new/new->old migration
1845      * compatibility.
1846      *
1847      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1848      * via default pseries-2.8 machines and explicit command-line parameters.
1849      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1850      * of the actual CAS-negotiated values to continue working properly. For
1851      * example, availability of memory unplug depends on knowing whether
1852      * OV5_HP_EVT was negotiated via CAS.
1853      *
1854      * Thus, for any cases where the set of available CAS-negotiatable
1855      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1856      * include the CAS-negotiated options in the migration stream, unless
1857      * if they affect boot time behaviour only.
1858      */
1859     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1860     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1861     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1862 
1863     /* We need extra information if we have any bits outside the mask
1864      * defined above */
1865     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1866 
1867     spapr_ovec_cleanup(ov5_mask);
1868 
1869     return cas_needed;
1870 }
1871 
1872 static const VMStateDescription vmstate_spapr_ov5_cas = {
1873     .name = "spapr_option_vector_ov5_cas",
1874     .version_id = 1,
1875     .minimum_version_id = 1,
1876     .needed = spapr_ov5_cas_needed,
1877     .fields = (VMStateField[]) {
1878         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1879                                  vmstate_spapr_ovec, SpaprOptionVector),
1880         VMSTATE_END_OF_LIST()
1881     },
1882 };
1883 
1884 static bool spapr_patb_entry_needed(void *opaque)
1885 {
1886     SpaprMachineState *spapr = opaque;
1887 
1888     return !!spapr->patb_entry;
1889 }
1890 
1891 static const VMStateDescription vmstate_spapr_patb_entry = {
1892     .name = "spapr_patb_entry",
1893     .version_id = 1,
1894     .minimum_version_id = 1,
1895     .needed = spapr_patb_entry_needed,
1896     .fields = (VMStateField[]) {
1897         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1898         VMSTATE_END_OF_LIST()
1899     },
1900 };
1901 
1902 static bool spapr_irq_map_needed(void *opaque)
1903 {
1904     SpaprMachineState *spapr = opaque;
1905 
1906     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1907 }
1908 
1909 static const VMStateDescription vmstate_spapr_irq_map = {
1910     .name = "spapr_irq_map",
1911     .version_id = 1,
1912     .minimum_version_id = 1,
1913     .needed = spapr_irq_map_needed,
1914     .fields = (VMStateField[]) {
1915         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1916         VMSTATE_END_OF_LIST()
1917     },
1918 };
1919 
1920 static bool spapr_dtb_needed(void *opaque)
1921 {
1922     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1923 
1924     return smc->update_dt_enabled;
1925 }
1926 
1927 static int spapr_dtb_pre_load(void *opaque)
1928 {
1929     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1930 
1931     g_free(spapr->fdt_blob);
1932     spapr->fdt_blob = NULL;
1933     spapr->fdt_size = 0;
1934 
1935     return 0;
1936 }
1937 
1938 static const VMStateDescription vmstate_spapr_dtb = {
1939     .name = "spapr_dtb",
1940     .version_id = 1,
1941     .minimum_version_id = 1,
1942     .needed = spapr_dtb_needed,
1943     .pre_load = spapr_dtb_pre_load,
1944     .fields = (VMStateField[]) {
1945         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1946         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1947         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1948                                      fdt_size),
1949         VMSTATE_END_OF_LIST()
1950     },
1951 };
1952 
1953 static bool spapr_fwnmi_needed(void *opaque)
1954 {
1955     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1956 
1957     return spapr->fwnmi_machine_check_addr != -1;
1958 }
1959 
1960 static int spapr_fwnmi_pre_save(void *opaque)
1961 {
1962     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1963 
1964     /*
1965      * Check if machine check handling is in progress and print a
1966      * warning message.
1967      */
1968     if (spapr->fwnmi_machine_check_interlock != -1) {
1969         warn_report("A machine check is being handled during migration. The"
1970                 "handler may run and log hardware error on the destination");
1971     }
1972 
1973     return 0;
1974 }
1975 
1976 static const VMStateDescription vmstate_spapr_fwnmi = {
1977     .name = "spapr_fwnmi",
1978     .version_id = 1,
1979     .minimum_version_id = 1,
1980     .needed = spapr_fwnmi_needed,
1981     .pre_save = spapr_fwnmi_pre_save,
1982     .fields = (VMStateField[]) {
1983         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
1984         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
1985         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
1986         VMSTATE_END_OF_LIST()
1987     },
1988 };
1989 
1990 static const VMStateDescription vmstate_spapr = {
1991     .name = "spapr",
1992     .version_id = 3,
1993     .minimum_version_id = 1,
1994     .pre_load = spapr_pre_load,
1995     .post_load = spapr_post_load,
1996     .pre_save = spapr_pre_save,
1997     .fields = (VMStateField[]) {
1998         /* used to be @next_irq */
1999         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2000 
2001         /* RTC offset */
2002         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2003 
2004         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2005         VMSTATE_END_OF_LIST()
2006     },
2007     .subsections = (const VMStateDescription*[]) {
2008         &vmstate_spapr_ov5_cas,
2009         &vmstate_spapr_patb_entry,
2010         &vmstate_spapr_pending_events,
2011         &vmstate_spapr_cap_htm,
2012         &vmstate_spapr_cap_vsx,
2013         &vmstate_spapr_cap_dfp,
2014         &vmstate_spapr_cap_cfpc,
2015         &vmstate_spapr_cap_sbbc,
2016         &vmstate_spapr_cap_ibs,
2017         &vmstate_spapr_cap_hpt_maxpagesize,
2018         &vmstate_spapr_irq_map,
2019         &vmstate_spapr_cap_nested_kvm_hv,
2020         &vmstate_spapr_dtb,
2021         &vmstate_spapr_cap_large_decr,
2022         &vmstate_spapr_cap_ccf_assist,
2023         &vmstate_spapr_cap_fwnmi,
2024         &vmstate_spapr_fwnmi,
2025         NULL
2026     }
2027 };
2028 
2029 static int htab_save_setup(QEMUFile *f, void *opaque)
2030 {
2031     SpaprMachineState *spapr = opaque;
2032 
2033     /* "Iteration" header */
2034     if (!spapr->htab_shift) {
2035         qemu_put_be32(f, -1);
2036     } else {
2037         qemu_put_be32(f, spapr->htab_shift);
2038     }
2039 
2040     if (spapr->htab) {
2041         spapr->htab_save_index = 0;
2042         spapr->htab_first_pass = true;
2043     } else {
2044         if (spapr->htab_shift) {
2045             assert(kvm_enabled());
2046         }
2047     }
2048 
2049 
2050     return 0;
2051 }
2052 
2053 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2054                             int chunkstart, int n_valid, int n_invalid)
2055 {
2056     qemu_put_be32(f, chunkstart);
2057     qemu_put_be16(f, n_valid);
2058     qemu_put_be16(f, n_invalid);
2059     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2060                     HASH_PTE_SIZE_64 * n_valid);
2061 }
2062 
2063 static void htab_save_end_marker(QEMUFile *f)
2064 {
2065     qemu_put_be32(f, 0);
2066     qemu_put_be16(f, 0);
2067     qemu_put_be16(f, 0);
2068 }
2069 
2070 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2071                                  int64_t max_ns)
2072 {
2073     bool has_timeout = max_ns != -1;
2074     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2075     int index = spapr->htab_save_index;
2076     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2077 
2078     assert(spapr->htab_first_pass);
2079 
2080     do {
2081         int chunkstart;
2082 
2083         /* Consume invalid HPTEs */
2084         while ((index < htabslots)
2085                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2086             CLEAN_HPTE(HPTE(spapr->htab, index));
2087             index++;
2088         }
2089 
2090         /* Consume valid HPTEs */
2091         chunkstart = index;
2092         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2093                && HPTE_VALID(HPTE(spapr->htab, index))) {
2094             CLEAN_HPTE(HPTE(spapr->htab, index));
2095             index++;
2096         }
2097 
2098         if (index > chunkstart) {
2099             int n_valid = index - chunkstart;
2100 
2101             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2102 
2103             if (has_timeout &&
2104                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2105                 break;
2106             }
2107         }
2108     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2109 
2110     if (index >= htabslots) {
2111         assert(index == htabslots);
2112         index = 0;
2113         spapr->htab_first_pass = false;
2114     }
2115     spapr->htab_save_index = index;
2116 }
2117 
2118 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2119                                 int64_t max_ns)
2120 {
2121     bool final = max_ns < 0;
2122     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2123     int examined = 0, sent = 0;
2124     int index = spapr->htab_save_index;
2125     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2126 
2127     assert(!spapr->htab_first_pass);
2128 
2129     do {
2130         int chunkstart, invalidstart;
2131 
2132         /* Consume non-dirty HPTEs */
2133         while ((index < htabslots)
2134                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2135             index++;
2136             examined++;
2137         }
2138 
2139         chunkstart = index;
2140         /* Consume valid dirty HPTEs */
2141         while ((index < htabslots) && (index - chunkstart < 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         invalidstart = index;
2150         /* Consume invalid dirty HPTEs */
2151         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2152                && HPTE_DIRTY(HPTE(spapr->htab, index))
2153                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2154             CLEAN_HPTE(HPTE(spapr->htab, index));
2155             index++;
2156             examined++;
2157         }
2158 
2159         if (index > chunkstart) {
2160             int n_valid = invalidstart - chunkstart;
2161             int n_invalid = index - invalidstart;
2162 
2163             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2164             sent += index - chunkstart;
2165 
2166             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2167                 break;
2168             }
2169         }
2170 
2171         if (examined >= htabslots) {
2172             break;
2173         }
2174 
2175         if (index >= htabslots) {
2176             assert(index == htabslots);
2177             index = 0;
2178         }
2179     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2180 
2181     if (index >= htabslots) {
2182         assert(index == htabslots);
2183         index = 0;
2184     }
2185 
2186     spapr->htab_save_index = index;
2187 
2188     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2189 }
2190 
2191 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2192 #define MAX_KVM_BUF_SIZE    2048
2193 
2194 static int htab_save_iterate(QEMUFile *f, void *opaque)
2195 {
2196     SpaprMachineState *spapr = opaque;
2197     int fd;
2198     int rc = 0;
2199 
2200     /* Iteration header */
2201     if (!spapr->htab_shift) {
2202         qemu_put_be32(f, -1);
2203         return 1;
2204     } else {
2205         qemu_put_be32(f, 0);
2206     }
2207 
2208     if (!spapr->htab) {
2209         assert(kvm_enabled());
2210 
2211         fd = get_htab_fd(spapr);
2212         if (fd < 0) {
2213             return fd;
2214         }
2215 
2216         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2217         if (rc < 0) {
2218             return rc;
2219         }
2220     } else  if (spapr->htab_first_pass) {
2221         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2222     } else {
2223         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2224     }
2225 
2226     htab_save_end_marker(f);
2227 
2228     return rc;
2229 }
2230 
2231 static int htab_save_complete(QEMUFile *f, void *opaque)
2232 {
2233     SpaprMachineState *spapr = opaque;
2234     int fd;
2235 
2236     /* Iteration header */
2237     if (!spapr->htab_shift) {
2238         qemu_put_be32(f, -1);
2239         return 0;
2240     } else {
2241         qemu_put_be32(f, 0);
2242     }
2243 
2244     if (!spapr->htab) {
2245         int rc;
2246 
2247         assert(kvm_enabled());
2248 
2249         fd = get_htab_fd(spapr);
2250         if (fd < 0) {
2251             return fd;
2252         }
2253 
2254         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2255         if (rc < 0) {
2256             return rc;
2257         }
2258     } else {
2259         if (spapr->htab_first_pass) {
2260             htab_save_first_pass(f, spapr, -1);
2261         }
2262         htab_save_later_pass(f, spapr, -1);
2263     }
2264 
2265     /* End marker */
2266     htab_save_end_marker(f);
2267 
2268     return 0;
2269 }
2270 
2271 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2272 {
2273     SpaprMachineState *spapr = opaque;
2274     uint32_t section_hdr;
2275     int fd = -1;
2276     Error *local_err = NULL;
2277 
2278     if (version_id < 1 || version_id > 1) {
2279         error_report("htab_load() bad version");
2280         return -EINVAL;
2281     }
2282 
2283     section_hdr = qemu_get_be32(f);
2284 
2285     if (section_hdr == -1) {
2286         spapr_free_hpt(spapr);
2287         return 0;
2288     }
2289 
2290     if (section_hdr) {
2291         int ret;
2292 
2293         /* First section gives the htab size */
2294         ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2295         if (ret < 0) {
2296             error_report_err(local_err);
2297             return ret;
2298         }
2299         return 0;
2300     }
2301 
2302     if (!spapr->htab) {
2303         assert(kvm_enabled());
2304 
2305         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2306         if (fd < 0) {
2307             error_report_err(local_err);
2308             return fd;
2309         }
2310     }
2311 
2312     while (true) {
2313         uint32_t index;
2314         uint16_t n_valid, n_invalid;
2315 
2316         index = qemu_get_be32(f);
2317         n_valid = qemu_get_be16(f);
2318         n_invalid = qemu_get_be16(f);
2319 
2320         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2321             /* End of Stream */
2322             break;
2323         }
2324 
2325         if ((index + n_valid + n_invalid) >
2326             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2327             /* Bad index in stream */
2328             error_report(
2329                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2330                 index, n_valid, n_invalid, spapr->htab_shift);
2331             return -EINVAL;
2332         }
2333 
2334         if (spapr->htab) {
2335             if (n_valid) {
2336                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2337                                 HASH_PTE_SIZE_64 * n_valid);
2338             }
2339             if (n_invalid) {
2340                 memset(HPTE(spapr->htab, index + n_valid), 0,
2341                        HASH_PTE_SIZE_64 * n_invalid);
2342             }
2343         } else {
2344             int rc;
2345 
2346             assert(fd >= 0);
2347 
2348             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2349                                         &local_err);
2350             if (rc < 0) {
2351                 error_report_err(local_err);
2352                 return rc;
2353             }
2354         }
2355     }
2356 
2357     if (!spapr->htab) {
2358         assert(fd >= 0);
2359         close(fd);
2360     }
2361 
2362     return 0;
2363 }
2364 
2365 static void htab_save_cleanup(void *opaque)
2366 {
2367     SpaprMachineState *spapr = opaque;
2368 
2369     close_htab_fd(spapr);
2370 }
2371 
2372 static SaveVMHandlers savevm_htab_handlers = {
2373     .save_setup = htab_save_setup,
2374     .save_live_iterate = htab_save_iterate,
2375     .save_live_complete_precopy = htab_save_complete,
2376     .save_cleanup = htab_save_cleanup,
2377     .load_state = htab_load,
2378 };
2379 
2380 static void spapr_boot_set(void *opaque, const char *boot_device,
2381                            Error **errp)
2382 {
2383     SpaprMachineState *spapr = SPAPR_MACHINE(opaque);
2384 
2385     g_free(spapr->boot_device);
2386     spapr->boot_device = g_strdup(boot_device);
2387 }
2388 
2389 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2390 {
2391     MachineState *machine = MACHINE(spapr);
2392     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2393     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2394     int i;
2395 
2396     for (i = 0; i < nr_lmbs; i++) {
2397         uint64_t addr;
2398 
2399         addr = i * lmb_size + machine->device_memory->base;
2400         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2401                                addr / lmb_size);
2402     }
2403 }
2404 
2405 /*
2406  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2407  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2408  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2409  */
2410 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2411 {
2412     int i;
2413 
2414     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2415         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2416                    " is not aligned to %" PRIu64 " MiB",
2417                    machine->ram_size,
2418                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2419         return;
2420     }
2421 
2422     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2423         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2424                    " is not aligned to %" PRIu64 " MiB",
2425                    machine->ram_size,
2426                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2427         return;
2428     }
2429 
2430     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2431         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2432             error_setg(errp,
2433                        "Node %d memory size 0x%" PRIx64
2434                        " is not aligned to %" PRIu64 " MiB",
2435                        i, machine->numa_state->nodes[i].node_mem,
2436                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2437             return;
2438         }
2439     }
2440 }
2441 
2442 /* find cpu slot in machine->possible_cpus by core_id */
2443 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2444 {
2445     int index = id / ms->smp.threads;
2446 
2447     if (index >= ms->possible_cpus->len) {
2448         return NULL;
2449     }
2450     if (idx) {
2451         *idx = index;
2452     }
2453     return &ms->possible_cpus->cpus[index];
2454 }
2455 
2456 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2457 {
2458     MachineState *ms = MACHINE(spapr);
2459     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2460     Error *local_err = NULL;
2461     bool vsmt_user = !!spapr->vsmt;
2462     int kvm_smt = kvmppc_smt_threads();
2463     int ret;
2464     unsigned int smp_threads = ms->smp.threads;
2465 
2466     if (!kvm_enabled() && (smp_threads > 1)) {
2467         error_setg(errp, "TCG cannot support more than 1 thread/core "
2468                    "on a pseries machine");
2469         return;
2470     }
2471     if (!is_power_of_2(smp_threads)) {
2472         error_setg(errp, "Cannot support %d threads/core on a pseries "
2473                    "machine because it must be a power of 2", smp_threads);
2474         return;
2475     }
2476 
2477     /* Detemine the VSMT mode to use: */
2478     if (vsmt_user) {
2479         if (spapr->vsmt < smp_threads) {
2480             error_setg(errp, "Cannot support VSMT mode %d"
2481                        " because it must be >= threads/core (%d)",
2482                        spapr->vsmt, smp_threads);
2483             return;
2484         }
2485         /* In this case, spapr->vsmt has been set by the command line */
2486     } else if (!smc->smp_threads_vsmt) {
2487         /*
2488          * Default VSMT value is tricky, because we need it to be as
2489          * consistent as possible (for migration), but this requires
2490          * changing it for at least some existing cases.  We pick 8 as
2491          * the value that we'd get with KVM on POWER8, the
2492          * overwhelmingly common case in production systems.
2493          */
2494         spapr->vsmt = MAX(8, smp_threads);
2495     } else {
2496         spapr->vsmt = smp_threads;
2497     }
2498 
2499     /* KVM: If necessary, set the SMT mode: */
2500     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2501         ret = kvmppc_set_smt_threads(spapr->vsmt);
2502         if (ret) {
2503             /* Looks like KVM isn't able to change VSMT mode */
2504             error_setg(&local_err,
2505                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2506                        spapr->vsmt, ret);
2507             /* We can live with that if the default one is big enough
2508              * for the number of threads, and a submultiple of the one
2509              * we want.  In this case we'll waste some vcpu ids, but
2510              * behaviour will be correct */
2511             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2512                 warn_report_err(local_err);
2513             } else {
2514                 if (!vsmt_user) {
2515                     error_append_hint(&local_err,
2516                                       "On PPC, a VM with %d threads/core"
2517                                       " on a host with %d threads/core"
2518                                       " requires the use of VSMT mode %d.\n",
2519                                       smp_threads, kvm_smt, spapr->vsmt);
2520                 }
2521                 kvmppc_error_append_smt_possible_hint(&local_err);
2522                 error_propagate(errp, local_err);
2523             }
2524         }
2525     }
2526     /* else TCG: nothing to do currently */
2527 }
2528 
2529 static void spapr_init_cpus(SpaprMachineState *spapr)
2530 {
2531     MachineState *machine = MACHINE(spapr);
2532     MachineClass *mc = MACHINE_GET_CLASS(machine);
2533     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2534     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2535     const CPUArchIdList *possible_cpus;
2536     unsigned int smp_cpus = machine->smp.cpus;
2537     unsigned int smp_threads = machine->smp.threads;
2538     unsigned int max_cpus = machine->smp.max_cpus;
2539     int boot_cores_nr = smp_cpus / smp_threads;
2540     int i;
2541 
2542     possible_cpus = mc->possible_cpu_arch_ids(machine);
2543     if (mc->has_hotpluggable_cpus) {
2544         if (smp_cpus % smp_threads) {
2545             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2546                          smp_cpus, smp_threads);
2547             exit(1);
2548         }
2549         if (max_cpus % smp_threads) {
2550             error_report("max_cpus (%u) must be multiple of threads (%u)",
2551                          max_cpus, smp_threads);
2552             exit(1);
2553         }
2554     } else {
2555         if (max_cpus != smp_cpus) {
2556             error_report("This machine version does not support CPU hotplug");
2557             exit(1);
2558         }
2559         boot_cores_nr = possible_cpus->len;
2560     }
2561 
2562     if (smc->pre_2_10_has_unused_icps) {
2563         int i;
2564 
2565         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2566             /* Dummy entries get deregistered when real ICPState objects
2567              * are registered during CPU core hotplug.
2568              */
2569             pre_2_10_vmstate_register_dummy_icp(i);
2570         }
2571     }
2572 
2573     for (i = 0; i < possible_cpus->len; i++) {
2574         int core_id = i * smp_threads;
2575 
2576         if (mc->has_hotpluggable_cpus) {
2577             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2578                                    spapr_vcpu_id(spapr, core_id));
2579         }
2580 
2581         if (i < boot_cores_nr) {
2582             Object *core  = object_new(type);
2583             int nr_threads = smp_threads;
2584 
2585             /* Handle the partially filled core for older machine types */
2586             if ((i + 1) * smp_threads >= smp_cpus) {
2587                 nr_threads = smp_cpus - i * smp_threads;
2588             }
2589 
2590             object_property_set_int(core, "nr-threads", nr_threads,
2591                                     &error_fatal);
2592             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2593                                     &error_fatal);
2594             qdev_realize(DEVICE(core), NULL, &error_fatal);
2595 
2596             object_unref(core);
2597         }
2598     }
2599 }
2600 
2601 static PCIHostState *spapr_create_default_phb(void)
2602 {
2603     DeviceState *dev;
2604 
2605     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2606     qdev_prop_set_uint32(dev, "index", 0);
2607     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2608 
2609     return PCI_HOST_BRIDGE(dev);
2610 }
2611 
2612 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2613 {
2614     MachineState *machine = MACHINE(spapr);
2615     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2616     hwaddr rma_size = machine->ram_size;
2617     hwaddr node0_size = spapr_node0_size(machine);
2618 
2619     /* RMA has to fit in the first NUMA node */
2620     rma_size = MIN(rma_size, node0_size);
2621 
2622     /*
2623      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2624      * never exceed that
2625      */
2626     rma_size = MIN(rma_size, 1 * TiB);
2627 
2628     /*
2629      * Clamp the RMA size based on machine type.  This is for
2630      * migration compatibility with older qemu versions, which limited
2631      * the RMA size for complicated and mostly bad reasons.
2632      */
2633     if (smc->rma_limit) {
2634         rma_size = MIN(rma_size, smc->rma_limit);
2635     }
2636 
2637     if (rma_size < MIN_RMA_SLOF) {
2638         error_setg(errp,
2639                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2640                    "ldMiB guest RMA (Real Mode Area memory)",
2641                    MIN_RMA_SLOF / MiB);
2642         return 0;
2643     }
2644 
2645     return rma_size;
2646 }
2647 
2648 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2649 {
2650     MachineState *machine = MACHINE(spapr);
2651     int i;
2652 
2653     for (i = 0; i < machine->ram_slots; i++) {
2654         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2655     }
2656 }
2657 
2658 /* pSeries LPAR / sPAPR hardware init */
2659 static void spapr_machine_init(MachineState *machine)
2660 {
2661     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2662     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2663     MachineClass *mc = MACHINE_GET_CLASS(machine);
2664     const char *bios_name = machine->firmware ?: FW_FILE_NAME;
2665     const char *kernel_filename = machine->kernel_filename;
2666     const char *initrd_filename = machine->initrd_filename;
2667     PCIHostState *phb;
2668     int i;
2669     MemoryRegion *sysmem = get_system_memory();
2670     long load_limit, fw_size;
2671     char *filename;
2672     Error *resize_hpt_err = NULL;
2673 
2674     /*
2675      * if Secure VM (PEF) support is configured, then initialize it
2676      */
2677     pef_kvm_init(machine->cgs, &error_fatal);
2678 
2679     msi_nonbroken = true;
2680 
2681     QLIST_INIT(&spapr->phbs);
2682     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2683 
2684     /* Determine capabilities to run with */
2685     spapr_caps_init(spapr);
2686 
2687     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2688     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2689         /*
2690          * If the user explicitly requested a mode we should either
2691          * supply it, or fail completely (which we do below).  But if
2692          * it's not set explicitly, we reset our mode to something
2693          * that works
2694          */
2695         if (resize_hpt_err) {
2696             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2697             error_free(resize_hpt_err);
2698             resize_hpt_err = NULL;
2699         } else {
2700             spapr->resize_hpt = smc->resize_hpt_default;
2701         }
2702     }
2703 
2704     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2705 
2706     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2707         /*
2708          * User requested HPT resize, but this host can't supply it.  Bail out
2709          */
2710         error_report_err(resize_hpt_err);
2711         exit(1);
2712     }
2713     error_free(resize_hpt_err);
2714 
2715     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2716 
2717     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2718     load_limit = MIN(spapr->rma_size, FDT_MAX_ADDR) - FW_OVERHEAD;
2719 
2720     /*
2721      * VSMT must be set in order to be able to compute VCPU ids, ie to
2722      * call spapr_max_server_number() or spapr_vcpu_id().
2723      */
2724     spapr_set_vsmt_mode(spapr, &error_fatal);
2725 
2726     /* Set up Interrupt Controller before we create the VCPUs */
2727     spapr_irq_init(spapr, &error_fatal);
2728 
2729     /* Set up containers for ibm,client-architecture-support negotiated options
2730      */
2731     spapr->ov5 = spapr_ovec_new();
2732     spapr->ov5_cas = spapr_ovec_new();
2733 
2734     if (smc->dr_lmb_enabled) {
2735         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2736         spapr_validate_node_memory(machine, &error_fatal);
2737     }
2738 
2739     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2740 
2741     /* advertise support for dedicated HP event source to guests */
2742     if (spapr->use_hotplug_event_source) {
2743         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2744     }
2745 
2746     /* advertise support for HPT resizing */
2747     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2748         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2749     }
2750 
2751     /* advertise support for ibm,dyamic-memory-v2 */
2752     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2753 
2754     /* advertise XIVE on POWER9 machines */
2755     if (spapr->irq->xive) {
2756         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2757     }
2758 
2759     /* init CPUs */
2760     spapr_init_cpus(spapr);
2761 
2762     /*
2763      * check we don't have a memory-less/cpu-less NUMA node
2764      * Firmware relies on the existing memory/cpu topology to provide the
2765      * NUMA topology to the kernel.
2766      * And the linux kernel needs to know the NUMA topology at start
2767      * to be able to hotplug CPUs later.
2768      */
2769     if (machine->numa_state->num_nodes) {
2770         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2771             /* check for memory-less node */
2772             if (machine->numa_state->nodes[i].node_mem == 0) {
2773                 CPUState *cs;
2774                 int found = 0;
2775                 /* check for cpu-less node */
2776                 CPU_FOREACH(cs) {
2777                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2778                     if (cpu->node_id == i) {
2779                         found = 1;
2780                         break;
2781                     }
2782                 }
2783                 /* memory-less and cpu-less node */
2784                 if (!found) {
2785                     error_report(
2786                        "Memory-less/cpu-less nodes are not supported (node %d)",
2787                                  i);
2788                     exit(1);
2789                 }
2790             }
2791         }
2792 
2793     }
2794 
2795     spapr->gpu_numa_id = spapr_numa_initial_nvgpu_numa_id(machine);
2796 
2797     /* Init numa_assoc_array */
2798     spapr_numa_associativity_init(spapr, machine);
2799 
2800     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2801         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2802                               spapr->max_compat_pvr)) {
2803         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2804         /* KVM and TCG always allow GTSE with radix... */
2805         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2806     }
2807     /* ... but not with hash (currently). */
2808 
2809     if (kvm_enabled()) {
2810         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2811         kvmppc_enable_logical_ci_hcalls();
2812         kvmppc_enable_set_mode_hcall();
2813 
2814         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2815         kvmppc_enable_clear_ref_mod_hcalls();
2816 
2817         /* Enable H_PAGE_INIT */
2818         kvmppc_enable_h_page_init();
2819     }
2820 
2821     /* map RAM */
2822     memory_region_add_subregion(sysmem, 0, machine->ram);
2823 
2824     /* always allocate the device memory information */
2825     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2826 
2827     /* initialize hotplug memory address space */
2828     if (machine->ram_size < machine->maxram_size) {
2829         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2830         /*
2831          * Limit the number of hotpluggable memory slots to half the number
2832          * slots that KVM supports, leaving the other half for PCI and other
2833          * devices. However ensure that number of slots doesn't drop below 32.
2834          */
2835         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2836                            SPAPR_MAX_RAM_SLOTS;
2837 
2838         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2839             max_memslots = SPAPR_MAX_RAM_SLOTS;
2840         }
2841         if (machine->ram_slots > max_memslots) {
2842             error_report("Specified number of memory slots %"
2843                          PRIu64" exceeds max supported %d",
2844                          machine->ram_slots, max_memslots);
2845             exit(1);
2846         }
2847 
2848         machine->device_memory->base = ROUND_UP(machine->ram_size,
2849                                                 SPAPR_DEVICE_MEM_ALIGN);
2850         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2851                            "device-memory", device_mem_size);
2852         memory_region_add_subregion(sysmem, machine->device_memory->base,
2853                                     &machine->device_memory->mr);
2854     }
2855 
2856     if (smc->dr_lmb_enabled) {
2857         spapr_create_lmb_dr_connectors(spapr);
2858     }
2859 
2860     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2861         /* Create the error string for live migration blocker */
2862         error_setg(&spapr->fwnmi_migration_blocker,
2863             "A machine check is being handled during migration. The handler"
2864             "may run and log hardware error on the destination");
2865     }
2866 
2867     if (mc->nvdimm_supported) {
2868         spapr_create_nvdimm_dr_connectors(spapr);
2869     }
2870 
2871     /* Set up RTAS event infrastructure */
2872     spapr_events_init(spapr);
2873 
2874     /* Set up the RTC RTAS interfaces */
2875     spapr_rtc_create(spapr);
2876 
2877     /* Set up VIO bus */
2878     spapr->vio_bus = spapr_vio_bus_init();
2879 
2880     for (i = 0; serial_hd(i); i++) {
2881         spapr_vty_create(spapr->vio_bus, serial_hd(i));
2882     }
2883 
2884     /* We always have at least the nvram device on VIO */
2885     spapr_create_nvram(spapr);
2886 
2887     /*
2888      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2889      * connectors (described in root DT node's "ibm,drc-types" property)
2890      * are pre-initialized here. additional child connectors (such as
2891      * connectors for a PHBs PCI slots) are added as needed during their
2892      * parent's realization.
2893      */
2894     if (smc->dr_phb_enabled) {
2895         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2896             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2897         }
2898     }
2899 
2900     /* Set up PCI */
2901     spapr_pci_rtas_init();
2902 
2903     phb = spapr_create_default_phb();
2904 
2905     for (i = 0; i < nb_nics; i++) {
2906         NICInfo *nd = &nd_table[i];
2907 
2908         if (!nd->model) {
2909             nd->model = g_strdup("spapr-vlan");
2910         }
2911 
2912         if (g_str_equal(nd->model, "spapr-vlan") ||
2913             g_str_equal(nd->model, "ibmveth")) {
2914             spapr_vlan_create(spapr->vio_bus, nd);
2915         } else {
2916             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2917         }
2918     }
2919 
2920     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2921         spapr_vscsi_create(spapr->vio_bus);
2922     }
2923 
2924     /* Graphics */
2925     if (spapr_vga_init(phb->bus, &error_fatal)) {
2926         spapr->has_graphics = true;
2927         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2928     }
2929 
2930     if (machine->usb) {
2931         if (smc->use_ohci_by_default) {
2932             pci_create_simple(phb->bus, -1, "pci-ohci");
2933         } else {
2934             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2935         }
2936 
2937         if (spapr->has_graphics) {
2938             USBBus *usb_bus = usb_bus_find(-1);
2939 
2940             usb_create_simple(usb_bus, "usb-kbd");
2941             usb_create_simple(usb_bus, "usb-mouse");
2942         }
2943     }
2944 
2945     if (kernel_filename) {
2946         spapr->kernel_size = load_elf(kernel_filename, NULL,
2947                                       translate_kernel_address, spapr,
2948                                       NULL, NULL, NULL, NULL, 1,
2949                                       PPC_ELF_MACHINE, 0, 0);
2950         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2951             spapr->kernel_size = load_elf(kernel_filename, NULL,
2952                                           translate_kernel_address, spapr,
2953                                           NULL, NULL, NULL, NULL, 0,
2954                                           PPC_ELF_MACHINE, 0, 0);
2955             spapr->kernel_le = spapr->kernel_size > 0;
2956         }
2957         if (spapr->kernel_size < 0) {
2958             error_report("error loading %s: %s", kernel_filename,
2959                          load_elf_strerror(spapr->kernel_size));
2960             exit(1);
2961         }
2962 
2963         /* load initrd */
2964         if (initrd_filename) {
2965             /* Try to locate the initrd in the gap between the kernel
2966              * and the firmware. Add a bit of space just in case
2967              */
2968             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
2969                                   + 0x1ffff) & ~0xffff;
2970             spapr->initrd_size = load_image_targphys(initrd_filename,
2971                                                      spapr->initrd_base,
2972                                                      load_limit
2973                                                      - spapr->initrd_base);
2974             if (spapr->initrd_size < 0) {
2975                 error_report("could not load initial ram disk '%s'",
2976                              initrd_filename);
2977                 exit(1);
2978             }
2979         }
2980     }
2981 
2982     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2983     if (!filename) {
2984         error_report("Could not find LPAR firmware '%s'", bios_name);
2985         exit(1);
2986     }
2987     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2988     if (fw_size <= 0) {
2989         error_report("Could not load LPAR firmware '%s'", filename);
2990         exit(1);
2991     }
2992     g_free(filename);
2993 
2994     /* FIXME: Should register things through the MachineState's qdev
2995      * interface, this is a legacy from the sPAPREnvironment structure
2996      * which predated MachineState but had a similar function */
2997     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2998     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
2999                          &savevm_htab_handlers, spapr);
3000 
3001     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
3002 
3003     qemu_register_boot_set(spapr_boot_set, spapr);
3004 
3005     /*
3006      * Nothing needs to be done to resume a suspended guest because
3007      * suspending does not change the machine state, so no need for
3008      * a ->wakeup method.
3009      */
3010     qemu_register_wakeup_support();
3011 
3012     if (kvm_enabled()) {
3013         /* to stop and start vmclock */
3014         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3015                                          &spapr->tb);
3016 
3017         kvmppc_spapr_enable_inkernel_multitce();
3018     }
3019 
3020     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3021 }
3022 
3023 #define DEFAULT_KVM_TYPE "auto"
3024 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3025 {
3026     /*
3027      * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to
3028      * accomodate the 'HV' and 'PV' formats that exists in the
3029      * wild. The 'auto' mode is being introduced already as
3030      * lower-case, thus we don't need to bother checking for
3031      * "AUTO".
3032      */
3033     if (!vm_type || !strcmp(vm_type, DEFAULT_KVM_TYPE)) {
3034         return 0;
3035     }
3036 
3037     if (!g_ascii_strcasecmp(vm_type, "hv")) {
3038         return 1;
3039     }
3040 
3041     if (!g_ascii_strcasecmp(vm_type, "pr")) {
3042         return 2;
3043     }
3044 
3045     error_report("Unknown kvm-type specified '%s'", vm_type);
3046     exit(1);
3047 }
3048 
3049 /*
3050  * Implementation of an interface to adjust firmware path
3051  * for the bootindex property handling.
3052  */
3053 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3054                                    DeviceState *dev)
3055 {
3056 #define CAST(type, obj, name) \
3057     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3058     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3059     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3060     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3061     PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3062 
3063     if (d) {
3064         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3065         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3066         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3067 
3068         if (spapr) {
3069             /*
3070              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3071              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3072              * 0x8000 | (target << 8) | (bus << 5) | lun
3073              * (see the "Logical unit addressing format" table in SAM5)
3074              */
3075             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3076             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3077                                    (uint64_t)id << 48);
3078         } else if (virtio) {
3079             /*
3080              * We use SRP luns of the form 01000000 | (target << 8) | lun
3081              * in the top 32 bits of the 64-bit LUN
3082              * Note: the quote above is from SLOF and it is wrong,
3083              * the actual binding is:
3084              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3085              */
3086             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3087             if (d->lun >= 256) {
3088                 /* Use the LUN "flat space addressing method" */
3089                 id |= 0x4000;
3090             }
3091             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3092                                    (uint64_t)id << 32);
3093         } else if (usb) {
3094             /*
3095              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3096              * in the top 32 bits of the 64-bit LUN
3097              */
3098             unsigned usb_port = atoi(usb->port->path);
3099             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3100             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3101                                    (uint64_t)id << 32);
3102         }
3103     }
3104 
3105     /*
3106      * SLOF probes the USB devices, and if it recognizes that the device is a
3107      * storage device, it changes its name to "storage" instead of "usb-host",
3108      * and additionally adds a child node for the SCSI LUN, so the correct
3109      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3110      */
3111     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3112         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3113         if (usb_host_dev_is_scsi_storage(usbdev)) {
3114             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3115         }
3116     }
3117 
3118     if (phb) {
3119         /* Replace "pci" with "pci@800000020000000" */
3120         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3121     }
3122 
3123     if (vsc) {
3124         /* Same logic as virtio above */
3125         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3126         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3127     }
3128 
3129     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3130         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3131         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3132         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3133     }
3134 
3135     if (pcidev) {
3136         return spapr_pci_fw_dev_name(pcidev);
3137     }
3138 
3139     return NULL;
3140 }
3141 
3142 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3143 {
3144     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3145 
3146     return g_strdup(spapr->kvm_type);
3147 }
3148 
3149 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3150 {
3151     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3152 
3153     g_free(spapr->kvm_type);
3154     spapr->kvm_type = g_strdup(value);
3155 }
3156 
3157 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3158 {
3159     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3160 
3161     return spapr->use_hotplug_event_source;
3162 }
3163 
3164 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3165                                             Error **errp)
3166 {
3167     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3168 
3169     spapr->use_hotplug_event_source = value;
3170 }
3171 
3172 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3173 {
3174     return true;
3175 }
3176 
3177 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3178 {
3179     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3180 
3181     switch (spapr->resize_hpt) {
3182     case SPAPR_RESIZE_HPT_DEFAULT:
3183         return g_strdup("default");
3184     case SPAPR_RESIZE_HPT_DISABLED:
3185         return g_strdup("disabled");
3186     case SPAPR_RESIZE_HPT_ENABLED:
3187         return g_strdup("enabled");
3188     case SPAPR_RESIZE_HPT_REQUIRED:
3189         return g_strdup("required");
3190     }
3191     g_assert_not_reached();
3192 }
3193 
3194 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3195 {
3196     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3197 
3198     if (strcmp(value, "default") == 0) {
3199         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3200     } else if (strcmp(value, "disabled") == 0) {
3201         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3202     } else if (strcmp(value, "enabled") == 0) {
3203         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3204     } else if (strcmp(value, "required") == 0) {
3205         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3206     } else {
3207         error_setg(errp, "Bad value for \"resize-hpt\" property");
3208     }
3209 }
3210 
3211 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3212 {
3213     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3214 
3215     if (spapr->irq == &spapr_irq_xics_legacy) {
3216         return g_strdup("legacy");
3217     } else if (spapr->irq == &spapr_irq_xics) {
3218         return g_strdup("xics");
3219     } else if (spapr->irq == &spapr_irq_xive) {
3220         return g_strdup("xive");
3221     } else if (spapr->irq == &spapr_irq_dual) {
3222         return g_strdup("dual");
3223     }
3224     g_assert_not_reached();
3225 }
3226 
3227 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3228 {
3229     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3230 
3231     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3232         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3233         return;
3234     }
3235 
3236     /* The legacy IRQ backend can not be set */
3237     if (strcmp(value, "xics") == 0) {
3238         spapr->irq = &spapr_irq_xics;
3239     } else if (strcmp(value, "xive") == 0) {
3240         spapr->irq = &spapr_irq_xive;
3241     } else if (strcmp(value, "dual") == 0) {
3242         spapr->irq = &spapr_irq_dual;
3243     } else {
3244         error_setg(errp, "Bad value for \"ic-mode\" property");
3245     }
3246 }
3247 
3248 static char *spapr_get_host_model(Object *obj, Error **errp)
3249 {
3250     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3251 
3252     return g_strdup(spapr->host_model);
3253 }
3254 
3255 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3256 {
3257     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3258 
3259     g_free(spapr->host_model);
3260     spapr->host_model = g_strdup(value);
3261 }
3262 
3263 static char *spapr_get_host_serial(Object *obj, Error **errp)
3264 {
3265     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3266 
3267     return g_strdup(spapr->host_serial);
3268 }
3269 
3270 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3271 {
3272     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3273 
3274     g_free(spapr->host_serial);
3275     spapr->host_serial = g_strdup(value);
3276 }
3277 
3278 static void spapr_instance_init(Object *obj)
3279 {
3280     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3281     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3282     MachineState *ms = MACHINE(spapr);
3283     MachineClass *mc = MACHINE_GET_CLASS(ms);
3284 
3285     /*
3286      * NVDIMM support went live in 5.1 without considering that, in
3287      * other archs, the user needs to enable NVDIMM support with the
3288      * 'nvdimm' machine option and the default behavior is NVDIMM
3289      * support disabled. It is too late to roll back to the standard
3290      * behavior without breaking 5.1 guests.
3291      */
3292     if (mc->nvdimm_supported) {
3293         ms->nvdimms_state->is_enabled = true;
3294     }
3295 
3296     spapr->htab_fd = -1;
3297     spapr->use_hotplug_event_source = true;
3298     spapr->kvm_type = g_strdup(DEFAULT_KVM_TYPE);
3299     object_property_add_str(obj, "kvm-type",
3300                             spapr_get_kvm_type, spapr_set_kvm_type);
3301     object_property_set_description(obj, "kvm-type",
3302                                     "Specifies the KVM virtualization mode (auto,"
3303                                     " hv, pr). Defaults to 'auto'. This mode will use"
3304                                     " any available KVM module loaded in the host,"
3305                                     " where kvm_hv takes precedence if both kvm_hv and"
3306                                     " kvm_pr are loaded.");
3307     object_property_add_bool(obj, "modern-hotplug-events",
3308                             spapr_get_modern_hotplug_events,
3309                             spapr_set_modern_hotplug_events);
3310     object_property_set_description(obj, "modern-hotplug-events",
3311                                     "Use dedicated hotplug event mechanism in"
3312                                     " place of standard EPOW events when possible"
3313                                     " (required for memory hot-unplug support)");
3314     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3315                             "Maximum permitted CPU compatibility mode");
3316 
3317     object_property_add_str(obj, "resize-hpt",
3318                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3319     object_property_set_description(obj, "resize-hpt",
3320                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3321     object_property_add_uint32_ptr(obj, "vsmt",
3322                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3323     object_property_set_description(obj, "vsmt",
3324                                     "Virtual SMT: KVM behaves as if this were"
3325                                     " the host's SMT mode");
3326 
3327     object_property_add_bool(obj, "vfio-no-msix-emulation",
3328                              spapr_get_msix_emulation, NULL);
3329 
3330     object_property_add_uint64_ptr(obj, "kernel-addr",
3331                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3332     object_property_set_description(obj, "kernel-addr",
3333                                     stringify(KERNEL_LOAD_ADDR)
3334                                     " for -kernel is the default");
3335     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3336     /* The machine class defines the default interrupt controller mode */
3337     spapr->irq = smc->irq;
3338     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3339                             spapr_set_ic_mode);
3340     object_property_set_description(obj, "ic-mode",
3341                  "Specifies the interrupt controller mode (xics, xive, dual)");
3342 
3343     object_property_add_str(obj, "host-model",
3344         spapr_get_host_model, spapr_set_host_model);
3345     object_property_set_description(obj, "host-model",
3346         "Host model to advertise in guest device tree");
3347     object_property_add_str(obj, "host-serial",
3348         spapr_get_host_serial, spapr_set_host_serial);
3349     object_property_set_description(obj, "host-serial",
3350         "Host serial number to advertise in guest device tree");
3351 }
3352 
3353 static void spapr_machine_finalizefn(Object *obj)
3354 {
3355     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3356 
3357     g_free(spapr->kvm_type);
3358 }
3359 
3360 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3361 {
3362     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3363     PowerPCCPU *cpu = POWERPC_CPU(cs);
3364     CPUPPCState *env = &cpu->env;
3365 
3366     cpu_synchronize_state(cs);
3367     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3368     if (spapr->fwnmi_system_reset_addr != -1) {
3369         uint64_t rtas_addr, addr;
3370 
3371         /* get rtas addr from fdt */
3372         rtas_addr = spapr_get_rtas_addr();
3373         if (!rtas_addr) {
3374             qemu_system_guest_panicked(NULL);
3375             return;
3376         }
3377 
3378         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3379         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3380         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3381         env->gpr[3] = addr;
3382     }
3383     ppc_cpu_do_system_reset(cs);
3384     if (spapr->fwnmi_system_reset_addr != -1) {
3385         env->nip = spapr->fwnmi_system_reset_addr;
3386     }
3387 }
3388 
3389 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3390 {
3391     CPUState *cs;
3392 
3393     CPU_FOREACH(cs) {
3394         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3395     }
3396 }
3397 
3398 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3399                           void *fdt, int *fdt_start_offset, Error **errp)
3400 {
3401     uint64_t addr;
3402     uint32_t node;
3403 
3404     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3405     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3406                                     &error_abort);
3407     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3408                                              SPAPR_MEMORY_BLOCK_SIZE);
3409     return 0;
3410 }
3411 
3412 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3413                            bool dedicated_hp_event_source)
3414 {
3415     SpaprDrc *drc;
3416     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3417     int i;
3418     uint64_t addr = addr_start;
3419     bool hotplugged = spapr_drc_hotplugged(dev);
3420 
3421     for (i = 0; i < nr_lmbs; i++) {
3422         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3423                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3424         g_assert(drc);
3425 
3426         /*
3427          * memory_device_get_free_addr() provided a range of free addresses
3428          * that doesn't overlap with any existing mapping at pre-plug. The
3429          * corresponding LMB DRCs are thus assumed to be all attachable.
3430          */
3431         spapr_drc_attach(drc, dev);
3432         if (!hotplugged) {
3433             spapr_drc_reset(drc);
3434         }
3435         addr += SPAPR_MEMORY_BLOCK_SIZE;
3436     }
3437     /* send hotplug notification to the
3438      * guest only in case of hotplugged memory
3439      */
3440     if (hotplugged) {
3441         if (dedicated_hp_event_source) {
3442             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3443                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3444             g_assert(drc);
3445             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3446                                                    nr_lmbs,
3447                                                    spapr_drc_index(drc));
3448         } else {
3449             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3450                                            nr_lmbs);
3451         }
3452     }
3453 }
3454 
3455 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3456 {
3457     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3458     PCDIMMDevice *dimm = PC_DIMM(dev);
3459     uint64_t size, addr;
3460     int64_t slot;
3461     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3462 
3463     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3464 
3465     pc_dimm_plug(dimm, MACHINE(ms));
3466 
3467     if (!is_nvdimm) {
3468         addr = object_property_get_uint(OBJECT(dimm),
3469                                         PC_DIMM_ADDR_PROP, &error_abort);
3470         spapr_add_lmbs(dev, addr, size,
3471                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT));
3472     } else {
3473         slot = object_property_get_int(OBJECT(dimm),
3474                                        PC_DIMM_SLOT_PROP, &error_abort);
3475         /* We should have valid slot number at this point */
3476         g_assert(slot >= 0);
3477         spapr_add_nvdimm(dev, slot);
3478     }
3479 }
3480 
3481 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3482                                   Error **errp)
3483 {
3484     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3485     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3486     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3487     PCDIMMDevice *dimm = PC_DIMM(dev);
3488     Error *local_err = NULL;
3489     uint64_t size;
3490     Object *memdev;
3491     hwaddr pagesize;
3492 
3493     if (!smc->dr_lmb_enabled) {
3494         error_setg(errp, "Memory hotplug not supported for this machine");
3495         return;
3496     }
3497 
3498     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3499     if (local_err) {
3500         error_propagate(errp, local_err);
3501         return;
3502     }
3503 
3504     if (is_nvdimm) {
3505         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3506             return;
3507         }
3508     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3509         error_setg(errp, "Hotplugged memory size must be a multiple of "
3510                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3511         return;
3512     }
3513 
3514     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3515                                       &error_abort);
3516     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3517     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3518         return;
3519     }
3520 
3521     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3522 }
3523 
3524 struct SpaprDimmState {
3525     PCDIMMDevice *dimm;
3526     uint32_t nr_lmbs;
3527     QTAILQ_ENTRY(SpaprDimmState) next;
3528 };
3529 
3530 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3531                                                        PCDIMMDevice *dimm)
3532 {
3533     SpaprDimmState *dimm_state = NULL;
3534 
3535     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3536         if (dimm_state->dimm == dimm) {
3537             break;
3538         }
3539     }
3540     return dimm_state;
3541 }
3542 
3543 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3544                                                       uint32_t nr_lmbs,
3545                                                       PCDIMMDevice *dimm)
3546 {
3547     SpaprDimmState *ds = NULL;
3548 
3549     /*
3550      * If this request is for a DIMM whose removal had failed earlier
3551      * (due to guest's refusal to remove the LMBs), we would have this
3552      * dimm already in the pending_dimm_unplugs list. In that
3553      * case don't add again.
3554      */
3555     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3556     if (!ds) {
3557         ds = g_malloc0(sizeof(SpaprDimmState));
3558         ds->nr_lmbs = nr_lmbs;
3559         ds->dimm = dimm;
3560         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3561     }
3562     return ds;
3563 }
3564 
3565 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3566                                               SpaprDimmState *dimm_state)
3567 {
3568     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3569     g_free(dimm_state);
3570 }
3571 
3572 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3573                                                         PCDIMMDevice *dimm)
3574 {
3575     SpaprDrc *drc;
3576     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3577                                                   &error_abort);
3578     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3579     uint32_t avail_lmbs = 0;
3580     uint64_t addr_start, addr;
3581     int i;
3582 
3583     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3584                                           &error_abort);
3585 
3586     addr = addr_start;
3587     for (i = 0; i < nr_lmbs; i++) {
3588         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3589                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3590         g_assert(drc);
3591         if (drc->dev) {
3592             avail_lmbs++;
3593         }
3594         addr += SPAPR_MEMORY_BLOCK_SIZE;
3595     }
3596 
3597     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3598 }
3599 
3600 void spapr_memory_unplug_rollback(SpaprMachineState *spapr, DeviceState *dev)
3601 {
3602     SpaprDimmState *ds;
3603     PCDIMMDevice *dimm;
3604     SpaprDrc *drc;
3605     uint32_t nr_lmbs;
3606     uint64_t size, addr_start, addr;
3607     g_autofree char *qapi_error = NULL;
3608     int i;
3609 
3610     if (!dev) {
3611         return;
3612     }
3613 
3614     dimm = PC_DIMM(dev);
3615     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3616 
3617     /*
3618      * 'ds == NULL' would mean that the DIMM doesn't have a pending
3619      * unplug state, but one of its DRC is marked as unplug_requested.
3620      * This is bad and weird enough to g_assert() out.
3621      */
3622     g_assert(ds);
3623 
3624     spapr_pending_dimm_unplugs_remove(spapr, ds);
3625 
3626     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3627     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3628 
3629     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3630                                           &error_abort);
3631 
3632     addr = addr_start;
3633     for (i = 0; i < nr_lmbs; i++) {
3634         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3635                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3636         g_assert(drc);
3637 
3638         drc->unplug_requested = false;
3639         addr += SPAPR_MEMORY_BLOCK_SIZE;
3640     }
3641 
3642     /*
3643      * Tell QAPI that something happened and the memory
3644      * hotunplug wasn't successful.
3645      */
3646     qapi_error = g_strdup_printf("Memory hotunplug rejected by the guest "
3647                                  "for device %s", dev->id);
3648     qapi_event_send_mem_unplug_error(dev->id, qapi_error);
3649 }
3650 
3651 /* Callback to be called during DRC release. */
3652 void spapr_lmb_release(DeviceState *dev)
3653 {
3654     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3655     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3656     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3657 
3658     /* This information will get lost if a migration occurs
3659      * during the unplug process. In this case recover it. */
3660     if (ds == NULL) {
3661         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3662         g_assert(ds);
3663         /* The DRC being examined by the caller at least must be counted */
3664         g_assert(ds->nr_lmbs);
3665     }
3666 
3667     if (--ds->nr_lmbs) {
3668         return;
3669     }
3670 
3671     /*
3672      * Now that all the LMBs have been removed by the guest, call the
3673      * unplug handler chain. This can never fail.
3674      */
3675     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3676     object_unparent(OBJECT(dev));
3677 }
3678 
3679 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3680 {
3681     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3682     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3683 
3684     /* We really shouldn't get this far without anything to unplug */
3685     g_assert(ds);
3686 
3687     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3688     qdev_unrealize(dev);
3689     spapr_pending_dimm_unplugs_remove(spapr, ds);
3690 }
3691 
3692 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3693                                         DeviceState *dev, Error **errp)
3694 {
3695     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3696     PCDIMMDevice *dimm = PC_DIMM(dev);
3697     uint32_t nr_lmbs;
3698     uint64_t size, addr_start, addr;
3699     int i;
3700     SpaprDrc *drc;
3701 
3702     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3703         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3704         return;
3705     }
3706 
3707     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3708     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3709 
3710     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3711                                           &error_abort);
3712 
3713     /*
3714      * An existing pending dimm state for this DIMM means that there is an
3715      * unplug operation in progress, waiting for the spapr_lmb_release
3716      * callback to complete the job (BQL can't cover that far). In this case,
3717      * bail out to avoid detaching DRCs that were already released.
3718      */
3719     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3720         error_setg(errp, "Memory unplug already in progress for device %s",
3721                    dev->id);
3722         return;
3723     }
3724 
3725     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3726 
3727     addr = addr_start;
3728     for (i = 0; i < nr_lmbs; i++) {
3729         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3730                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3731         g_assert(drc);
3732 
3733         spapr_drc_unplug_request(drc);
3734         addr += SPAPR_MEMORY_BLOCK_SIZE;
3735     }
3736 
3737     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3738                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3739     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3740                                               nr_lmbs, spapr_drc_index(drc));
3741 }
3742 
3743 /* Callback to be called during DRC release. */
3744 void spapr_core_release(DeviceState *dev)
3745 {
3746     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3747 
3748     /* Call the unplug handler chain. This can never fail. */
3749     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3750     object_unparent(OBJECT(dev));
3751 }
3752 
3753 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3754 {
3755     MachineState *ms = MACHINE(hotplug_dev);
3756     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3757     CPUCore *cc = CPU_CORE(dev);
3758     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3759 
3760     if (smc->pre_2_10_has_unused_icps) {
3761         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3762         int i;
3763 
3764         for (i = 0; i < cc->nr_threads; i++) {
3765             CPUState *cs = CPU(sc->threads[i]);
3766 
3767             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3768         }
3769     }
3770 
3771     assert(core_slot);
3772     core_slot->cpu = NULL;
3773     qdev_unrealize(dev);
3774 }
3775 
3776 static
3777 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3778                                Error **errp)
3779 {
3780     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3781     int index;
3782     SpaprDrc *drc;
3783     CPUCore *cc = CPU_CORE(dev);
3784 
3785     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3786         error_setg(errp, "Unable to find CPU core with core-id: %d",
3787                    cc->core_id);
3788         return;
3789     }
3790     if (index == 0) {
3791         error_setg(errp, "Boot CPU core may not be unplugged");
3792         return;
3793     }
3794 
3795     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3796                           spapr_vcpu_id(spapr, cc->core_id));
3797     g_assert(drc);
3798 
3799     if (!spapr_drc_unplug_requested(drc)) {
3800         spapr_drc_unplug_request(drc);
3801     }
3802 
3803     /*
3804      * spapr_hotplug_req_remove_by_index is left unguarded, out of the
3805      * "!spapr_drc_unplug_requested" check, to allow for multiple IRQ
3806      * pulses removing the same CPU. Otherwise, in an failed hotunplug
3807      * attempt (e.g. the kernel will refuse to remove the last online
3808      * CPU), we will never attempt it again because unplug_requested
3809      * will still be 'true' in that case.
3810      */
3811     spapr_hotplug_req_remove_by_index(drc);
3812 }
3813 
3814 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3815                            void *fdt, int *fdt_start_offset, Error **errp)
3816 {
3817     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3818     CPUState *cs = CPU(core->threads[0]);
3819     PowerPCCPU *cpu = POWERPC_CPU(cs);
3820     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3821     int id = spapr_get_vcpu_id(cpu);
3822     g_autofree char *nodename = NULL;
3823     int offset;
3824 
3825     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3826     offset = fdt_add_subnode(fdt, 0, nodename);
3827 
3828     spapr_dt_cpu(cs, fdt, offset, spapr);
3829 
3830     /*
3831      * spapr_dt_cpu() does not fill the 'name' property in the
3832      * CPU node. The function is called during boot process, before
3833      * and after CAS, and overwriting the 'name' property written
3834      * by SLOF is not allowed.
3835      *
3836      * Write it manually after spapr_dt_cpu(). This makes the hotplug
3837      * CPUs more compatible with the coldplugged ones, which have
3838      * the 'name' property. Linux Kernel also relies on this
3839      * property to identify CPU nodes.
3840      */
3841     _FDT((fdt_setprop_string(fdt, offset, "name", nodename)));
3842 
3843     *fdt_start_offset = offset;
3844     return 0;
3845 }
3846 
3847 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3848 {
3849     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3850     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3851     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3852     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3853     CPUCore *cc = CPU_CORE(dev);
3854     CPUState *cs;
3855     SpaprDrc *drc;
3856     CPUArchId *core_slot;
3857     int index;
3858     bool hotplugged = spapr_drc_hotplugged(dev);
3859     int i;
3860 
3861     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3862     g_assert(core_slot); /* Already checked in spapr_core_pre_plug() */
3863 
3864     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3865                           spapr_vcpu_id(spapr, cc->core_id));
3866 
3867     g_assert(drc || !mc->has_hotpluggable_cpus);
3868 
3869     if (drc) {
3870         /*
3871          * spapr_core_pre_plug() already buys us this is a brand new
3872          * core being plugged into a free slot. Nothing should already
3873          * be attached to the corresponding DRC.
3874          */
3875         spapr_drc_attach(drc, dev);
3876 
3877         if (hotplugged) {
3878             /*
3879              * Send hotplug notification interrupt to the guest only
3880              * in case of hotplugged CPUs.
3881              */
3882             spapr_hotplug_req_add_by_index(drc);
3883         } else {
3884             spapr_drc_reset(drc);
3885         }
3886     }
3887 
3888     core_slot->cpu = OBJECT(dev);
3889 
3890     /*
3891      * Set compatibility mode to match the boot CPU, which was either set
3892      * by the machine reset code or by CAS. This really shouldn't fail at
3893      * this point.
3894      */
3895     if (hotplugged) {
3896         for (i = 0; i < cc->nr_threads; i++) {
3897             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3898                            &error_abort);
3899         }
3900     }
3901 
3902     if (smc->pre_2_10_has_unused_icps) {
3903         for (i = 0; i < cc->nr_threads; i++) {
3904             cs = CPU(core->threads[i]);
3905             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3906         }
3907     }
3908 }
3909 
3910 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3911                                 Error **errp)
3912 {
3913     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3914     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3915     CPUCore *cc = CPU_CORE(dev);
3916     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3917     const char *type = object_get_typename(OBJECT(dev));
3918     CPUArchId *core_slot;
3919     int index;
3920     unsigned int smp_threads = machine->smp.threads;
3921 
3922     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3923         error_setg(errp, "CPU hotplug not supported for this machine");
3924         return;
3925     }
3926 
3927     if (strcmp(base_core_type, type)) {
3928         error_setg(errp, "CPU core type should be %s", base_core_type);
3929         return;
3930     }
3931 
3932     if (cc->core_id % smp_threads) {
3933         error_setg(errp, "invalid core id %d", cc->core_id);
3934         return;
3935     }
3936 
3937     /*
3938      * In general we should have homogeneous threads-per-core, but old
3939      * (pre hotplug support) machine types allow the last core to have
3940      * reduced threads as a compatibility hack for when we allowed
3941      * total vcpus not a multiple of threads-per-core.
3942      */
3943     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3944         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
3945                    smp_threads);
3946         return;
3947     }
3948 
3949     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3950     if (!core_slot) {
3951         error_setg(errp, "core id %d out of range", cc->core_id);
3952         return;
3953     }
3954 
3955     if (core_slot->cpu) {
3956         error_setg(errp, "core %d already populated", cc->core_id);
3957         return;
3958     }
3959 
3960     numa_cpu_pre_plug(core_slot, dev, errp);
3961 }
3962 
3963 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3964                           void *fdt, int *fdt_start_offset, Error **errp)
3965 {
3966     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3967     int intc_phandle;
3968 
3969     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3970     if (intc_phandle <= 0) {
3971         return -1;
3972     }
3973 
3974     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3975         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3976         return -1;
3977     }
3978 
3979     /* generally SLOF creates these, for hotplug it's up to QEMU */
3980     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3981 
3982     return 0;
3983 }
3984 
3985 static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3986                                Error **errp)
3987 {
3988     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3989     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3990     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3991     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3992     SpaprDrc *drc;
3993 
3994     if (dev->hotplugged && !smc->dr_phb_enabled) {
3995         error_setg(errp, "PHB hotplug not supported for this machine");
3996         return false;
3997     }
3998 
3999     if (sphb->index == (uint32_t)-1) {
4000         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
4001         return false;
4002     }
4003 
4004     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4005     if (drc && drc->dev) {
4006         error_setg(errp, "PHB %d already attached", sphb->index);
4007         return false;
4008     }
4009 
4010     /*
4011      * This will check that sphb->index doesn't exceed the maximum number of
4012      * PHBs for the current machine type.
4013      */
4014     return
4015         smc->phb_placement(spapr, sphb->index,
4016                            &sphb->buid, &sphb->io_win_addr,
4017                            &sphb->mem_win_addr, &sphb->mem64_win_addr,
4018                            windows_supported, sphb->dma_liobn,
4019                            &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
4020                            errp);
4021 }
4022 
4023 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4024 {
4025     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4026     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4027     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4028     SpaprDrc *drc;
4029     bool hotplugged = spapr_drc_hotplugged(dev);
4030 
4031     if (!smc->dr_phb_enabled) {
4032         return;
4033     }
4034 
4035     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4036     /* hotplug hooks should check it's enabled before getting this far */
4037     assert(drc);
4038 
4039     /* spapr_phb_pre_plug() already checked the DRC is attachable */
4040     spapr_drc_attach(drc, dev);
4041 
4042     if (hotplugged) {
4043         spapr_hotplug_req_add_by_index(drc);
4044     } else {
4045         spapr_drc_reset(drc);
4046     }
4047 }
4048 
4049 void spapr_phb_release(DeviceState *dev)
4050 {
4051     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4052 
4053     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4054     object_unparent(OBJECT(dev));
4055 }
4056 
4057 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4058 {
4059     qdev_unrealize(dev);
4060 }
4061 
4062 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4063                                      DeviceState *dev, Error **errp)
4064 {
4065     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4066     SpaprDrc *drc;
4067 
4068     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4069     assert(drc);
4070 
4071     if (!spapr_drc_unplug_requested(drc)) {
4072         spapr_drc_unplug_request(drc);
4073         spapr_hotplug_req_remove_by_index(drc);
4074     } else {
4075         error_setg(errp,
4076                    "PCI Host Bridge unplug already in progress for device %s",
4077                    dev->id);
4078     }
4079 }
4080 
4081 static
4082 bool spapr_tpm_proxy_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4083                               Error **errp)
4084 {
4085     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4086 
4087     if (spapr->tpm_proxy != NULL) {
4088         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4089         return false;
4090     }
4091 
4092     return true;
4093 }
4094 
4095 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4096 {
4097     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4098     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4099 
4100     /* Already checked in spapr_tpm_proxy_pre_plug() */
4101     g_assert(spapr->tpm_proxy == NULL);
4102 
4103     spapr->tpm_proxy = tpm_proxy;
4104 }
4105 
4106 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4107 {
4108     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4109 
4110     qdev_unrealize(dev);
4111     object_unparent(OBJECT(dev));
4112     spapr->tpm_proxy = NULL;
4113 }
4114 
4115 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4116                                       DeviceState *dev, Error **errp)
4117 {
4118     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4119         spapr_memory_plug(hotplug_dev, dev);
4120     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4121         spapr_core_plug(hotplug_dev, dev);
4122     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4123         spapr_phb_plug(hotplug_dev, dev);
4124     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4125         spapr_tpm_proxy_plug(hotplug_dev, dev);
4126     }
4127 }
4128 
4129 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4130                                         DeviceState *dev, Error **errp)
4131 {
4132     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4133         spapr_memory_unplug(hotplug_dev, dev);
4134     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4135         spapr_core_unplug(hotplug_dev, dev);
4136     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4137         spapr_phb_unplug(hotplug_dev, dev);
4138     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4139         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4140     }
4141 }
4142 
4143 bool spapr_memory_hot_unplug_supported(SpaprMachineState *spapr)
4144 {
4145     return spapr_ovec_test(spapr->ov5_cas, OV5_HP_EVT) ||
4146         /*
4147          * CAS will process all pending unplug requests.
4148          *
4149          * HACK: a guest could theoretically have cleared all bits in OV5,
4150          * but none of the guests we care for do.
4151          */
4152         spapr_ovec_empty(spapr->ov5_cas);
4153 }
4154 
4155 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4156                                                 DeviceState *dev, Error **errp)
4157 {
4158     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4159     MachineClass *mc = MACHINE_GET_CLASS(sms);
4160     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4161 
4162     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4163         if (spapr_memory_hot_unplug_supported(sms)) {
4164             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4165         } else {
4166             error_setg(errp, "Memory hot unplug not supported for this guest");
4167         }
4168     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4169         if (!mc->has_hotpluggable_cpus) {
4170             error_setg(errp, "CPU hot unplug not supported on this machine");
4171             return;
4172         }
4173         spapr_core_unplug_request(hotplug_dev, dev, errp);
4174     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4175         if (!smc->dr_phb_enabled) {
4176             error_setg(errp, "PHB hot unplug not supported on this machine");
4177             return;
4178         }
4179         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4180     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4181         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4182     }
4183 }
4184 
4185 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4186                                           DeviceState *dev, Error **errp)
4187 {
4188     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4189         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4190     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4191         spapr_core_pre_plug(hotplug_dev, dev, errp);
4192     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4193         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4194     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4195         spapr_tpm_proxy_pre_plug(hotplug_dev, dev, errp);
4196     }
4197 }
4198 
4199 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4200                                                  DeviceState *dev)
4201 {
4202     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4203         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4204         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4205         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4206         return HOTPLUG_HANDLER(machine);
4207     }
4208     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4209         PCIDevice *pcidev = PCI_DEVICE(dev);
4210         PCIBus *root = pci_device_root_bus(pcidev);
4211         SpaprPhbState *phb =
4212             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4213                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4214 
4215         if (phb) {
4216             return HOTPLUG_HANDLER(phb);
4217         }
4218     }
4219     return NULL;
4220 }
4221 
4222 static CpuInstanceProperties
4223 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4224 {
4225     CPUArchId *core_slot;
4226     MachineClass *mc = MACHINE_GET_CLASS(machine);
4227 
4228     /* make sure possible_cpu are intialized */
4229     mc->possible_cpu_arch_ids(machine);
4230     /* get CPU core slot containing thread that matches cpu_index */
4231     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4232     assert(core_slot);
4233     return core_slot->props;
4234 }
4235 
4236 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4237 {
4238     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4239 }
4240 
4241 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4242 {
4243     int i;
4244     unsigned int smp_threads = machine->smp.threads;
4245     unsigned int smp_cpus = machine->smp.cpus;
4246     const char *core_type;
4247     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4248     MachineClass *mc = MACHINE_GET_CLASS(machine);
4249 
4250     if (!mc->has_hotpluggable_cpus) {
4251         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4252     }
4253     if (machine->possible_cpus) {
4254         assert(machine->possible_cpus->len == spapr_max_cores);
4255         return machine->possible_cpus;
4256     }
4257 
4258     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4259     if (!core_type) {
4260         error_report("Unable to find sPAPR CPU Core definition");
4261         exit(1);
4262     }
4263 
4264     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4265                              sizeof(CPUArchId) * spapr_max_cores);
4266     machine->possible_cpus->len = spapr_max_cores;
4267     for (i = 0; i < machine->possible_cpus->len; i++) {
4268         int core_id = i * smp_threads;
4269 
4270         machine->possible_cpus->cpus[i].type = core_type;
4271         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4272         machine->possible_cpus->cpus[i].arch_id = core_id;
4273         machine->possible_cpus->cpus[i].props.has_core_id = true;
4274         machine->possible_cpus->cpus[i].props.core_id = core_id;
4275     }
4276     return machine->possible_cpus;
4277 }
4278 
4279 static bool spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4280                                 uint64_t *buid, hwaddr *pio,
4281                                 hwaddr *mmio32, hwaddr *mmio64,
4282                                 unsigned n_dma, uint32_t *liobns,
4283                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4284 {
4285     /*
4286      * New-style PHB window placement.
4287      *
4288      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4289      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4290      * windows.
4291      *
4292      * Some guest kernels can't work with MMIO windows above 1<<46
4293      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4294      *
4295      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4296      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4297      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4298      * 1TiB 64-bit MMIO windows for each PHB.
4299      */
4300     const uint64_t base_buid = 0x800000020000000ULL;
4301     int i;
4302 
4303     /* Sanity check natural alignments */
4304     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4305     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4306     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4307     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4308     /* Sanity check bounds */
4309     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4310                       SPAPR_PCI_MEM32_WIN_SIZE);
4311     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4312                       SPAPR_PCI_MEM64_WIN_SIZE);
4313 
4314     if (index >= SPAPR_MAX_PHBS) {
4315         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4316                    SPAPR_MAX_PHBS - 1);
4317         return false;
4318     }
4319 
4320     *buid = base_buid + index;
4321     for (i = 0; i < n_dma; ++i) {
4322         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4323     }
4324 
4325     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4326     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4327     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4328 
4329     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4330     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4331     return true;
4332 }
4333 
4334 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4335 {
4336     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4337 
4338     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4339 }
4340 
4341 static void spapr_ics_resend(XICSFabric *dev)
4342 {
4343     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4344 
4345     ics_resend(spapr->ics);
4346 }
4347 
4348 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4349 {
4350     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4351 
4352     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4353 }
4354 
4355 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4356                                  Monitor *mon)
4357 {
4358     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4359 
4360     spapr_irq_print_info(spapr, mon);
4361     monitor_printf(mon, "irqchip: %s\n",
4362                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4363 }
4364 
4365 /*
4366  * This is a XIVE only operation
4367  */
4368 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4369                            uint8_t nvt_blk, uint32_t nvt_idx,
4370                            bool cam_ignore, uint8_t priority,
4371                            uint32_t logic_serv, XiveTCTXMatch *match)
4372 {
4373     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4374     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4375     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4376     int count;
4377 
4378     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4379                            priority, logic_serv, match);
4380     if (count < 0) {
4381         return count;
4382     }
4383 
4384     /*
4385      * When we implement the save and restore of the thread interrupt
4386      * contexts in the enter/exit CPU handlers of the machine and the
4387      * escalations in QEMU, we should be able to handle non dispatched
4388      * vCPUs.
4389      *
4390      * Until this is done, the sPAPR machine should find at least one
4391      * matching context always.
4392      */
4393     if (count == 0) {
4394         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4395                       nvt_blk, nvt_idx);
4396     }
4397 
4398     return count;
4399 }
4400 
4401 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4402 {
4403     return cpu->vcpu_id;
4404 }
4405 
4406 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4407 {
4408     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4409     MachineState *ms = MACHINE(spapr);
4410     int vcpu_id;
4411 
4412     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4413 
4414     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4415         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4416         error_append_hint(errp, "Adjust the number of cpus to %d "
4417                           "or try to raise the number of threads per core\n",
4418                           vcpu_id * ms->smp.threads / spapr->vsmt);
4419         return false;
4420     }
4421 
4422     cpu->vcpu_id = vcpu_id;
4423     return true;
4424 }
4425 
4426 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4427 {
4428     CPUState *cs;
4429 
4430     CPU_FOREACH(cs) {
4431         PowerPCCPU *cpu = POWERPC_CPU(cs);
4432 
4433         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4434             return cpu;
4435         }
4436     }
4437 
4438     return NULL;
4439 }
4440 
4441 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4442 {
4443     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4444 
4445     /* These are only called by TCG, KVM maintains dispatch state */
4446 
4447     spapr_cpu->prod = false;
4448     if (spapr_cpu->vpa_addr) {
4449         CPUState *cs = CPU(cpu);
4450         uint32_t dispatch;
4451 
4452         dispatch = ldl_be_phys(cs->as,
4453                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4454         dispatch++;
4455         if ((dispatch & 1) != 0) {
4456             qemu_log_mask(LOG_GUEST_ERROR,
4457                           "VPA: incorrect dispatch counter value for "
4458                           "dispatched partition %u, correcting.\n", dispatch);
4459             dispatch++;
4460         }
4461         stl_be_phys(cs->as,
4462                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4463     }
4464 }
4465 
4466 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4467 {
4468     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4469 
4470     if (spapr_cpu->vpa_addr) {
4471         CPUState *cs = CPU(cpu);
4472         uint32_t dispatch;
4473 
4474         dispatch = ldl_be_phys(cs->as,
4475                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4476         dispatch++;
4477         if ((dispatch & 1) != 1) {
4478             qemu_log_mask(LOG_GUEST_ERROR,
4479                           "VPA: incorrect dispatch counter value for "
4480                           "preempted partition %u, correcting.\n", dispatch);
4481             dispatch++;
4482         }
4483         stl_be_phys(cs->as,
4484                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4485     }
4486 }
4487 
4488 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4489 {
4490     MachineClass *mc = MACHINE_CLASS(oc);
4491     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4492     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4493     NMIClass *nc = NMI_CLASS(oc);
4494     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4495     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4496     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4497     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4498     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4499 
4500     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4501     mc->ignore_boot_device_suffixes = true;
4502 
4503     /*
4504      * We set up the default / latest behaviour here.  The class_init
4505      * functions for the specific versioned machine types can override
4506      * these details for backwards compatibility
4507      */
4508     mc->init = spapr_machine_init;
4509     mc->reset = spapr_machine_reset;
4510     mc->block_default_type = IF_SCSI;
4511 
4512     /*
4513      * Setting max_cpus to INT32_MAX. Both KVM and TCG max_cpus values
4514      * should be limited by the host capability instead of hardcoded.
4515      * max_cpus for KVM guests will be checked in kvm_init(), and TCG
4516      * guests are welcome to have as many CPUs as the host are capable
4517      * of emulate.
4518      */
4519     mc->max_cpus = INT32_MAX;
4520 
4521     mc->no_parallel = 1;
4522     mc->default_boot_order = "";
4523     mc->default_ram_size = 512 * MiB;
4524     mc->default_ram_id = "ppc_spapr.ram";
4525     mc->default_display = "std";
4526     mc->kvm_type = spapr_kvm_type;
4527     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4528     mc->pci_allow_0_address = true;
4529     assert(!mc->get_hotplug_handler);
4530     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4531     hc->pre_plug = spapr_machine_device_pre_plug;
4532     hc->plug = spapr_machine_device_plug;
4533     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4534     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4535     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4536     hc->unplug_request = spapr_machine_device_unplug_request;
4537     hc->unplug = spapr_machine_device_unplug;
4538 
4539     smc->dr_lmb_enabled = true;
4540     smc->update_dt_enabled = true;
4541     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4542     mc->has_hotpluggable_cpus = true;
4543     mc->nvdimm_supported = true;
4544     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4545     fwc->get_dev_path = spapr_get_fw_dev_path;
4546     nc->nmi_monitor_handler = spapr_nmi;
4547     smc->phb_placement = spapr_phb_placement;
4548     vhc->hypercall = emulate_spapr_hypercall;
4549     vhc->hpt_mask = spapr_hpt_mask;
4550     vhc->map_hptes = spapr_map_hptes;
4551     vhc->unmap_hptes = spapr_unmap_hptes;
4552     vhc->hpte_set_c = spapr_hpte_set_c;
4553     vhc->hpte_set_r = spapr_hpte_set_r;
4554     vhc->get_pate = spapr_get_pate;
4555     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4556     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4557     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4558     xic->ics_get = spapr_ics_get;
4559     xic->ics_resend = spapr_ics_resend;
4560     xic->icp_get = spapr_icp_get;
4561     ispc->print_info = spapr_pic_print_info;
4562     /* Force NUMA node memory size to be a multiple of
4563      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4564      * in which LMBs are represented and hot-added
4565      */
4566     mc->numa_mem_align_shift = 28;
4567     mc->auto_enable_numa = true;
4568 
4569     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4570     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4571     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4572     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4573     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4574     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4575     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4576     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4577     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4578     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4579     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4580     spapr_caps_add_properties(smc);
4581     smc->irq = &spapr_irq_dual;
4582     smc->dr_phb_enabled = true;
4583     smc->linux_pci_probe = true;
4584     smc->smp_threads_vsmt = true;
4585     smc->nr_xirqs = SPAPR_NR_XIRQS;
4586     xfc->match_nvt = spapr_match_nvt;
4587 }
4588 
4589 static const TypeInfo spapr_machine_info = {
4590     .name          = TYPE_SPAPR_MACHINE,
4591     .parent        = TYPE_MACHINE,
4592     .abstract      = true,
4593     .instance_size = sizeof(SpaprMachineState),
4594     .instance_init = spapr_instance_init,
4595     .instance_finalize = spapr_machine_finalizefn,
4596     .class_size    = sizeof(SpaprMachineClass),
4597     .class_init    = spapr_machine_class_init,
4598     .interfaces = (InterfaceInfo[]) {
4599         { TYPE_FW_PATH_PROVIDER },
4600         { TYPE_NMI },
4601         { TYPE_HOTPLUG_HANDLER },
4602         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4603         { TYPE_XICS_FABRIC },
4604         { TYPE_INTERRUPT_STATS_PROVIDER },
4605         { TYPE_XIVE_FABRIC },
4606         { }
4607     },
4608 };
4609 
4610 static void spapr_machine_latest_class_options(MachineClass *mc)
4611 {
4612     mc->alias = "pseries";
4613     mc->is_default = true;
4614 }
4615 
4616 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4617     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4618                                                     void *data)      \
4619     {                                                                \
4620         MachineClass *mc = MACHINE_CLASS(oc);                        \
4621         spapr_machine_##suffix##_class_options(mc);                  \
4622         if (latest) {                                                \
4623             spapr_machine_latest_class_options(mc);                  \
4624         }                                                            \
4625     }                                                                \
4626     static const TypeInfo spapr_machine_##suffix##_info = {          \
4627         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4628         .parent = TYPE_SPAPR_MACHINE,                                \
4629         .class_init = spapr_machine_##suffix##_class_init,           \
4630     };                                                               \
4631     static void spapr_machine_register_##suffix(void)                \
4632     {                                                                \
4633         type_register(&spapr_machine_##suffix##_info);               \
4634     }                                                                \
4635     type_init(spapr_machine_register_##suffix)
4636 
4637 /*
4638  * pseries-6.1
4639  */
4640 static void spapr_machine_6_1_class_options(MachineClass *mc)
4641 {
4642     /* Defaults for the latest behaviour inherited from the base class */
4643 }
4644 
4645 DEFINE_SPAPR_MACHINE(6_1, "6.1", true);
4646 
4647 /*
4648  * pseries-6.0
4649  */
4650 static void spapr_machine_6_0_class_options(MachineClass *mc)
4651 {
4652     spapr_machine_6_1_class_options(mc);
4653     compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len);
4654 }
4655 
4656 DEFINE_SPAPR_MACHINE(6_0, "6.0", false);
4657 
4658 /*
4659  * pseries-5.2
4660  */
4661 static void spapr_machine_5_2_class_options(MachineClass *mc)
4662 {
4663     spapr_machine_6_0_class_options(mc);
4664     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
4665 }
4666 
4667 DEFINE_SPAPR_MACHINE(5_2, "5.2", false);
4668 
4669 /*
4670  * pseries-5.1
4671  */
4672 static void spapr_machine_5_1_class_options(MachineClass *mc)
4673 {
4674     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4675 
4676     spapr_machine_5_2_class_options(mc);
4677     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4678     smc->pre_5_2_numa_associativity = true;
4679 }
4680 
4681 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4682 
4683 /*
4684  * pseries-5.0
4685  */
4686 static void spapr_machine_5_0_class_options(MachineClass *mc)
4687 {
4688     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4689     static GlobalProperty compat[] = {
4690         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4691     };
4692 
4693     spapr_machine_5_1_class_options(mc);
4694     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4695     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4696     mc->numa_mem_supported = true;
4697     smc->pre_5_1_assoc_refpoints = true;
4698 }
4699 
4700 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4701 
4702 /*
4703  * pseries-4.2
4704  */
4705 static void spapr_machine_4_2_class_options(MachineClass *mc)
4706 {
4707     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4708 
4709     spapr_machine_5_0_class_options(mc);
4710     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4711     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4712     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4713     smc->rma_limit = 16 * GiB;
4714     mc->nvdimm_supported = false;
4715 }
4716 
4717 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4718 
4719 /*
4720  * pseries-4.1
4721  */
4722 static void spapr_machine_4_1_class_options(MachineClass *mc)
4723 {
4724     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4725     static GlobalProperty compat[] = {
4726         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4727         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4728     };
4729 
4730     spapr_machine_4_2_class_options(mc);
4731     smc->linux_pci_probe = false;
4732     smc->smp_threads_vsmt = false;
4733     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4734     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4735 }
4736 
4737 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4738 
4739 /*
4740  * pseries-4.0
4741  */
4742 static bool phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4743                               uint64_t *buid, hwaddr *pio,
4744                               hwaddr *mmio32, hwaddr *mmio64,
4745                               unsigned n_dma, uint32_t *liobns,
4746                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4747 {
4748     if (!spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma,
4749                              liobns, nv2gpa, nv2atsd, errp)) {
4750         return false;
4751     }
4752 
4753     *nv2gpa = 0;
4754     *nv2atsd = 0;
4755     return true;
4756 }
4757 static void spapr_machine_4_0_class_options(MachineClass *mc)
4758 {
4759     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4760 
4761     spapr_machine_4_1_class_options(mc);
4762     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4763     smc->phb_placement = phb_placement_4_0;
4764     smc->irq = &spapr_irq_xics;
4765     smc->pre_4_1_migration = true;
4766 }
4767 
4768 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4769 
4770 /*
4771  * pseries-3.1
4772  */
4773 static void spapr_machine_3_1_class_options(MachineClass *mc)
4774 {
4775     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4776 
4777     spapr_machine_4_0_class_options(mc);
4778     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4779 
4780     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4781     smc->update_dt_enabled = false;
4782     smc->dr_phb_enabled = false;
4783     smc->broken_host_serial_model = true;
4784     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4785     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4786     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4787     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4788 }
4789 
4790 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4791 
4792 /*
4793  * pseries-3.0
4794  */
4795 
4796 static void spapr_machine_3_0_class_options(MachineClass *mc)
4797 {
4798     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4799 
4800     spapr_machine_3_1_class_options(mc);
4801     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4802 
4803     smc->legacy_irq_allocation = true;
4804     smc->nr_xirqs = 0x400;
4805     smc->irq = &spapr_irq_xics_legacy;
4806 }
4807 
4808 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4809 
4810 /*
4811  * pseries-2.12
4812  */
4813 static void spapr_machine_2_12_class_options(MachineClass *mc)
4814 {
4815     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4816     static GlobalProperty compat[] = {
4817         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4818         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4819     };
4820 
4821     spapr_machine_3_0_class_options(mc);
4822     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4823     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4824 
4825     /* We depend on kvm_enabled() to choose a default value for the
4826      * hpt-max-page-size capability. Of course we can't do it here
4827      * because this is too early and the HW accelerator isn't initialzed
4828      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4829      */
4830     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4831 }
4832 
4833 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4834 
4835 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4836 {
4837     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4838 
4839     spapr_machine_2_12_class_options(mc);
4840     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4841     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4842     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4843 }
4844 
4845 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4846 
4847 /*
4848  * pseries-2.11
4849  */
4850 
4851 static void spapr_machine_2_11_class_options(MachineClass *mc)
4852 {
4853     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4854 
4855     spapr_machine_2_12_class_options(mc);
4856     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4857     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4858 }
4859 
4860 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4861 
4862 /*
4863  * pseries-2.10
4864  */
4865 
4866 static void spapr_machine_2_10_class_options(MachineClass *mc)
4867 {
4868     spapr_machine_2_11_class_options(mc);
4869     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4870 }
4871 
4872 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4873 
4874 /*
4875  * pseries-2.9
4876  */
4877 
4878 static void spapr_machine_2_9_class_options(MachineClass *mc)
4879 {
4880     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4881     static GlobalProperty compat[] = {
4882         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4883     };
4884 
4885     spapr_machine_2_10_class_options(mc);
4886     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4887     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4888     smc->pre_2_10_has_unused_icps = true;
4889     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4890 }
4891 
4892 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4893 
4894 /*
4895  * pseries-2.8
4896  */
4897 
4898 static void spapr_machine_2_8_class_options(MachineClass *mc)
4899 {
4900     static GlobalProperty compat[] = {
4901         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4902     };
4903 
4904     spapr_machine_2_9_class_options(mc);
4905     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4906     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4907     mc->numa_mem_align_shift = 23;
4908 }
4909 
4910 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4911 
4912 /*
4913  * pseries-2.7
4914  */
4915 
4916 static bool phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4917                               uint64_t *buid, hwaddr *pio,
4918                               hwaddr *mmio32, hwaddr *mmio64,
4919                               unsigned n_dma, uint32_t *liobns,
4920                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4921 {
4922     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4923     const uint64_t base_buid = 0x800000020000000ULL;
4924     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4925     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4926     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4927     const uint32_t max_index = 255;
4928     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4929 
4930     uint64_t ram_top = MACHINE(spapr)->ram_size;
4931     hwaddr phb0_base, phb_base;
4932     int i;
4933 
4934     /* Do we have device memory? */
4935     if (MACHINE(spapr)->maxram_size > ram_top) {
4936         /* Can't just use maxram_size, because there may be an
4937          * alignment gap between normal and device memory regions
4938          */
4939         ram_top = MACHINE(spapr)->device_memory->base +
4940             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4941     }
4942 
4943     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4944 
4945     if (index > max_index) {
4946         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4947                    max_index);
4948         return false;
4949     }
4950 
4951     *buid = base_buid + index;
4952     for (i = 0; i < n_dma; ++i) {
4953         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4954     }
4955 
4956     phb_base = phb0_base + index * phb_spacing;
4957     *pio = phb_base + pio_offset;
4958     *mmio32 = phb_base + mmio_offset;
4959     /*
4960      * We don't set the 64-bit MMIO window, relying on the PHB's
4961      * fallback behaviour of automatically splitting a large "32-bit"
4962      * window into contiguous 32-bit and 64-bit windows
4963      */
4964 
4965     *nv2gpa = 0;
4966     *nv2atsd = 0;
4967     return true;
4968 }
4969 
4970 static void spapr_machine_2_7_class_options(MachineClass *mc)
4971 {
4972     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4973     static GlobalProperty compat[] = {
4974         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4975         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4976         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4977         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4978     };
4979 
4980     spapr_machine_2_8_class_options(mc);
4981     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4982     mc->default_machine_opts = "modern-hotplug-events=off";
4983     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4984     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4985     smc->phb_placement = phb_placement_2_7;
4986 }
4987 
4988 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4989 
4990 /*
4991  * pseries-2.6
4992  */
4993 
4994 static void spapr_machine_2_6_class_options(MachineClass *mc)
4995 {
4996     static GlobalProperty compat[] = {
4997         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4998     };
4999 
5000     spapr_machine_2_7_class_options(mc);
5001     mc->has_hotpluggable_cpus = false;
5002     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
5003     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5004 }
5005 
5006 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
5007 
5008 /*
5009  * pseries-2.5
5010  */
5011 
5012 static void spapr_machine_2_5_class_options(MachineClass *mc)
5013 {
5014     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5015     static GlobalProperty compat[] = {
5016         { "spapr-vlan", "use-rx-buffer-pools", "off" },
5017     };
5018 
5019     spapr_machine_2_6_class_options(mc);
5020     smc->use_ohci_by_default = true;
5021     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
5022     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5023 }
5024 
5025 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
5026 
5027 /*
5028  * pseries-2.4
5029  */
5030 
5031 static void spapr_machine_2_4_class_options(MachineClass *mc)
5032 {
5033     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5034 
5035     spapr_machine_2_5_class_options(mc);
5036     smc->dr_lmb_enabled = false;
5037     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
5038 }
5039 
5040 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
5041 
5042 /*
5043  * pseries-2.3
5044  */
5045 
5046 static void spapr_machine_2_3_class_options(MachineClass *mc)
5047 {
5048     static GlobalProperty compat[] = {
5049         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
5050     };
5051     spapr_machine_2_4_class_options(mc);
5052     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
5053     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5054 }
5055 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
5056 
5057 /*
5058  * pseries-2.2
5059  */
5060 
5061 static void spapr_machine_2_2_class_options(MachineClass *mc)
5062 {
5063     static GlobalProperty compat[] = {
5064         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
5065     };
5066 
5067     spapr_machine_2_3_class_options(mc);
5068     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
5069     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5070     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
5071 }
5072 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
5073 
5074 /*
5075  * pseries-2.1
5076  */
5077 
5078 static void spapr_machine_2_1_class_options(MachineClass *mc)
5079 {
5080     spapr_machine_2_2_class_options(mc);
5081     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
5082 }
5083 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
5084 
5085 static void spapr_machine_register_types(void)
5086 {
5087     type_register_static(&spapr_machine_info);
5088 }
5089 
5090 type_init(spapr_machine_register_types)
5091