xref: /openbmc/qemu/hw/ppc/spapr.c (revision ef17dd6a)
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/qapi-events-qdev.h"
33 #include "qapi/visitor.h"
34 #include "sysemu/sysemu.h"
35 #include "sysemu/hostmem.h"
36 #include "sysemu/numa.h"
37 #include "sysemu/qtest.h"
38 #include "sysemu/reset.h"
39 #include "sysemu/runstate.h"
40 #include "qemu/log.h"
41 #include "hw/fw-path-provider.h"
42 #include "elf.h"
43 #include "net/net.h"
44 #include "sysemu/device_tree.h"
45 #include "sysemu/cpus.h"
46 #include "sysemu/hw_accel.h"
47 #include "kvm_ppc.h"
48 #include "migration/misc.h"
49 #include "migration/qemu-file-types.h"
50 #include "migration/global_state.h"
51 #include "migration/register.h"
52 #include "migration/blocker.h"
53 #include "mmu-hash64.h"
54 #include "mmu-book3s-v3.h"
55 #include "cpu-models.h"
56 #include "hw/core/cpu.h"
57 
58 #include "hw/ppc/ppc.h"
59 #include "hw/loader.h"
60 
61 #include "hw/ppc/fdt.h"
62 #include "hw/ppc/spapr.h"
63 #include "hw/ppc/spapr_vio.h"
64 #include "hw/qdev-properties.h"
65 #include "hw/pci-host/spapr.h"
66 #include "hw/pci/msi.h"
67 
68 #include "hw/pci/pci.h"
69 #include "hw/scsi/scsi.h"
70 #include "hw/virtio/virtio-scsi.h"
71 #include "hw/virtio/vhost-scsi-common.h"
72 
73 #include "exec/ram_addr.h"
74 #include "hw/usb.h"
75 #include "qemu/config-file.h"
76 #include "qemu/error-report.h"
77 #include "trace.h"
78 #include "hw/nmi.h"
79 #include "hw/intc/intc.h"
80 
81 #include "hw/ppc/spapr_cpu_core.h"
82 #include "hw/mem/memory-device.h"
83 #include "hw/ppc/spapr_tpm_proxy.h"
84 #include "hw/ppc/spapr_nvdimm.h"
85 #include "hw/ppc/spapr_numa.h"
86 #include "hw/ppc/pef.h"
87 
88 #include "monitor/monitor.h"
89 
90 #include <libfdt.h>
91 
92 /* SLOF memory layout:
93  *
94  * SLOF raw image loaded at 0, copies its romfs right below the flat
95  * device-tree, then position SLOF itself 31M below that
96  *
97  * So we set FW_OVERHEAD to 40MB which should account for all of that
98  * and more
99  *
100  * We load our kernel at 4M, leaving space for SLOF initial image
101  */
102 #define FDT_MAX_ADDR            0x80000000 /* FDT must stay below that */
103 #define FW_MAX_SIZE             0x400000
104 #define FW_FILE_NAME            "slof.bin"
105 #define FW_FILE_NAME_VOF        "vof.bin"
106 #define FW_OVERHEAD             0x2800000
107 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
108 
109 #define MIN_RMA_SLOF            (128 * MiB)
110 
111 #define PHANDLE_INTC            0x00001111
112 
113 /* These two functions implement the VCPU id numbering: one to compute them
114  * all and one to identify thread 0 of a VCORE. Any change to the first one
115  * is likely to have an impact on the second one, so let's keep them close.
116  */
117 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
118 {
119     MachineState *ms = MACHINE(spapr);
120     unsigned int smp_threads = ms->smp.threads;
121 
122     assert(spapr->vsmt);
123     return
124         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
125 }
126 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
127                                       PowerPCCPU *cpu)
128 {
129     assert(spapr->vsmt);
130     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
131 }
132 
133 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
134 {
135     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
136      * and newer QEMUs don't even have them. In both cases, we don't want
137      * to send anything on the wire.
138      */
139     return false;
140 }
141 
142 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
143     .name = "icp/server",
144     .version_id = 1,
145     .minimum_version_id = 1,
146     .needed = pre_2_10_vmstate_dummy_icp_needed,
147     .fields = (VMStateField[]) {
148         VMSTATE_UNUSED(4), /* uint32_t xirr */
149         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
150         VMSTATE_UNUSED(1), /* uint8_t mfrr */
151         VMSTATE_END_OF_LIST()
152     },
153 };
154 
155 static void pre_2_10_vmstate_register_dummy_icp(int i)
156 {
157     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
158                      (void *)(uintptr_t) i);
159 }
160 
161 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
162 {
163     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
164                        (void *)(uintptr_t) i);
165 }
166 
167 int spapr_max_server_number(SpaprMachineState *spapr)
168 {
169     MachineState *ms = MACHINE(spapr);
170 
171     assert(spapr->vsmt);
172     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
173 }
174 
175 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
176                                   int smt_threads)
177 {
178     int i, ret = 0;
179     uint32_t servers_prop[smt_threads];
180     uint32_t gservers_prop[smt_threads * 2];
181     int index = spapr_get_vcpu_id(cpu);
182 
183     if (cpu->compat_pvr) {
184         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
185         if (ret < 0) {
186             return ret;
187         }
188     }
189 
190     /* Build interrupt servers and gservers properties */
191     for (i = 0; i < smt_threads; i++) {
192         servers_prop[i] = cpu_to_be32(index + i);
193         /* Hack, direct the group queues back to cpu 0 */
194         gservers_prop[i*2] = cpu_to_be32(index + i);
195         gservers_prop[i*2 + 1] = 0;
196     }
197     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
198                       servers_prop, sizeof(servers_prop));
199     if (ret < 0) {
200         return ret;
201     }
202     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
203                       gservers_prop, sizeof(gservers_prop));
204 
205     return ret;
206 }
207 
208 static void spapr_dt_pa_features(SpaprMachineState *spapr,
209                                  PowerPCCPU *cpu,
210                                  void *fdt, int offset)
211 {
212     uint8_t pa_features_206[] = { 6, 0,
213         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
214     uint8_t pa_features_207[] = { 24, 0,
215         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
216         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
217         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
218         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
219     uint8_t pa_features_300[] = { 66, 0,
220         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
221         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
222         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
223         /* 6: DS207 */
224         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
225         /* 16: Vector */
226         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
227         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
228         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
229         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
230         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
231         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
232         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
233         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
234         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
235         /* 42: PM, 44: PC RA, 46: SC vec'd */
236         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
237         /* 48: SIMD, 50: QP BFP, 52: String */
238         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
239         /* 54: DecFP, 56: DecI, 58: SHA */
240         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
241         /* 60: NM atomic, 62: RNG */
242         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
243     };
244     uint8_t *pa_features = NULL;
245     size_t pa_size;
246 
247     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
248         pa_features = pa_features_206;
249         pa_size = sizeof(pa_features_206);
250     }
251     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
252         pa_features = pa_features_207;
253         pa_size = sizeof(pa_features_207);
254     }
255     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
256         pa_features = pa_features_300;
257         pa_size = sizeof(pa_features_300);
258     }
259     if (!pa_features) {
260         return;
261     }
262 
263     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
264         /*
265          * Note: we keep CI large pages off by default because a 64K capable
266          * guest provisioned with large pages might otherwise try to map a qemu
267          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
268          * even if that qemu runs on a 4k host.
269          * We dd this bit back here if we are confident this is not an issue
270          */
271         pa_features[3] |= 0x20;
272     }
273     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
274         pa_features[24] |= 0x80;    /* Transactional memory support */
275     }
276     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
277         /* Workaround for broken kernels that attempt (guest) radix
278          * mode when they can't handle it, if they see the radix bit set
279          * in pa-features. So hide it from them. */
280         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
281     }
282 
283     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
284 }
285 
286 static hwaddr spapr_node0_size(MachineState *machine)
287 {
288     if (machine->numa_state->num_nodes) {
289         int i;
290         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
291             if (machine->numa_state->nodes[i].node_mem) {
292                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
293                            machine->ram_size);
294             }
295         }
296     }
297     return machine->ram_size;
298 }
299 
300 static void add_str(GString *s, const gchar *s1)
301 {
302     g_string_append_len(s, s1, strlen(s1) + 1);
303 }
304 
305 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid,
306                                 hwaddr start, hwaddr size)
307 {
308     char mem_name[32];
309     uint64_t mem_reg_property[2];
310     int off;
311 
312     mem_reg_property[0] = cpu_to_be64(start);
313     mem_reg_property[1] = cpu_to_be64(size);
314 
315     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
316     off = fdt_add_subnode(fdt, 0, mem_name);
317     _FDT(off);
318     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
319     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
320                       sizeof(mem_reg_property))));
321     spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid);
322     return off;
323 }
324 
325 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
326 {
327     MemoryDeviceInfoList *info;
328 
329     for (info = list; info; info = info->next) {
330         MemoryDeviceInfo *value = info->value;
331 
332         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
333             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
334 
335             if (addr >= pcdimm_info->addr &&
336                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
337                 return pcdimm_info->node;
338             }
339         }
340     }
341 
342     return -1;
343 }
344 
345 struct sPAPRDrconfCellV2 {
346      uint32_t seq_lmbs;
347      uint64_t base_addr;
348      uint32_t drc_index;
349      uint32_t aa_index;
350      uint32_t flags;
351 } QEMU_PACKED;
352 
353 typedef struct DrconfCellQueue {
354     struct sPAPRDrconfCellV2 cell;
355     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
356 } DrconfCellQueue;
357 
358 static DrconfCellQueue *
359 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
360                       uint32_t drc_index, uint32_t aa_index,
361                       uint32_t flags)
362 {
363     DrconfCellQueue *elem;
364 
365     elem = g_malloc0(sizeof(*elem));
366     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
367     elem->cell.base_addr = cpu_to_be64(base_addr);
368     elem->cell.drc_index = cpu_to_be32(drc_index);
369     elem->cell.aa_index = cpu_to_be32(aa_index);
370     elem->cell.flags = cpu_to_be32(flags);
371 
372     return elem;
373 }
374 
375 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
376                                       int offset, MemoryDeviceInfoList *dimms)
377 {
378     MachineState *machine = MACHINE(spapr);
379     uint8_t *int_buf, *cur_index;
380     int ret;
381     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
382     uint64_t addr, cur_addr, size;
383     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
384     uint64_t mem_end = machine->device_memory->base +
385                        memory_region_size(&machine->device_memory->mr);
386     uint32_t node, buf_len, nr_entries = 0;
387     SpaprDrc *drc;
388     DrconfCellQueue *elem, *next;
389     MemoryDeviceInfoList *info;
390     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
391         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
392 
393     /* Entry to cover RAM and the gap area */
394     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
395                                  SPAPR_LMB_FLAGS_RESERVED |
396                                  SPAPR_LMB_FLAGS_DRC_INVALID);
397     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
398     nr_entries++;
399 
400     cur_addr = machine->device_memory->base;
401     for (info = dimms; info; info = info->next) {
402         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
403 
404         addr = di->addr;
405         size = di->size;
406         node = di->node;
407 
408         /*
409          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
410          * area is marked hotpluggable in the next iteration for the bigger
411          * chunk including the NVDIMM occupied area.
412          */
413         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
414             continue;
415 
416         /* Entry for hot-pluggable area */
417         if (cur_addr < addr) {
418             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
419             g_assert(drc);
420             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
421                                          cur_addr, spapr_drc_index(drc), -1, 0);
422             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
423             nr_entries++;
424         }
425 
426         /* Entry for DIMM */
427         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
428         g_assert(drc);
429         elem = spapr_get_drconf_cell(size / lmb_size, addr,
430                                      spapr_drc_index(drc), node,
431                                      (SPAPR_LMB_FLAGS_ASSIGNED |
432                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
433         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
434         nr_entries++;
435         cur_addr = addr + size;
436     }
437 
438     /* Entry for remaining hotpluggable area */
439     if (cur_addr < mem_end) {
440         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
441         g_assert(drc);
442         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
443                                      cur_addr, spapr_drc_index(drc), -1, 0);
444         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
445         nr_entries++;
446     }
447 
448     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
449     int_buf = cur_index = g_malloc0(buf_len);
450     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
451     cur_index += sizeof(nr_entries);
452 
453     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
454         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
455         cur_index += sizeof(elem->cell);
456         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
457         g_free(elem);
458     }
459 
460     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
461     g_free(int_buf);
462     if (ret < 0) {
463         return -1;
464     }
465     return 0;
466 }
467 
468 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
469                                    int offset, MemoryDeviceInfoList *dimms)
470 {
471     MachineState *machine = MACHINE(spapr);
472     int i, ret;
473     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
474     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
475     uint32_t nr_lmbs = (machine->device_memory->base +
476                        memory_region_size(&machine->device_memory->mr)) /
477                        lmb_size;
478     uint32_t *int_buf, *cur_index, buf_len;
479 
480     /*
481      * Allocate enough buffer size to fit in ibm,dynamic-memory
482      */
483     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
484     cur_index = int_buf = g_malloc0(buf_len);
485     int_buf[0] = cpu_to_be32(nr_lmbs);
486     cur_index++;
487     for (i = 0; i < nr_lmbs; i++) {
488         uint64_t addr = i * lmb_size;
489         uint32_t *dynamic_memory = cur_index;
490 
491         if (i >= device_lmb_start) {
492             SpaprDrc *drc;
493 
494             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
495             g_assert(drc);
496 
497             dynamic_memory[0] = cpu_to_be32(addr >> 32);
498             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
499             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
500             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
501             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
502             if (memory_region_present(get_system_memory(), addr)) {
503                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
504             } else {
505                 dynamic_memory[5] = cpu_to_be32(0);
506             }
507         } else {
508             /*
509              * LMB information for RMA, boot time RAM and gap b/n RAM and
510              * device memory region -- all these are marked as reserved
511              * and as having no valid DRC.
512              */
513             dynamic_memory[0] = cpu_to_be32(addr >> 32);
514             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
515             dynamic_memory[2] = cpu_to_be32(0);
516             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
517             dynamic_memory[4] = cpu_to_be32(-1);
518             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
519                                             SPAPR_LMB_FLAGS_DRC_INVALID);
520         }
521 
522         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
523     }
524     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
525     g_free(int_buf);
526     if (ret < 0) {
527         return -1;
528     }
529     return 0;
530 }
531 
532 /*
533  * Adds ibm,dynamic-reconfiguration-memory node.
534  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
535  * of this device tree node.
536  */
537 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
538                                                    void *fdt)
539 {
540     MachineState *machine = MACHINE(spapr);
541     int ret, offset;
542     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
543     uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
544                                 cpu_to_be32(lmb_size & 0xffffffff)};
545     MemoryDeviceInfoList *dimms = NULL;
546 
547     /*
548      * Don't create the node if there is no device memory
549      */
550     if (machine->ram_size == machine->maxram_size) {
551         return 0;
552     }
553 
554     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
555 
556     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
557                     sizeof(prop_lmb_size));
558     if (ret < 0) {
559         return ret;
560     }
561 
562     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
563     if (ret < 0) {
564         return ret;
565     }
566 
567     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
568     if (ret < 0) {
569         return ret;
570     }
571 
572     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
573     dimms = qmp_memory_device_list();
574     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
575         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
576     } else {
577         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
578     }
579     qapi_free_MemoryDeviceInfoList(dimms);
580 
581     if (ret < 0) {
582         return ret;
583     }
584 
585     ret = spapr_numa_write_assoc_lookup_arrays(spapr, fdt, offset);
586 
587     return ret;
588 }
589 
590 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
591 {
592     MachineState *machine = MACHINE(spapr);
593     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
594     hwaddr mem_start, node_size;
595     int i, nb_nodes = machine->numa_state->num_nodes;
596     NodeInfo *nodes = machine->numa_state->nodes;
597 
598     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
599         if (!nodes[i].node_mem) {
600             continue;
601         }
602         if (mem_start >= machine->ram_size) {
603             node_size = 0;
604         } else {
605             node_size = nodes[i].node_mem;
606             if (node_size > machine->ram_size - mem_start) {
607                 node_size = machine->ram_size - mem_start;
608             }
609         }
610         if (!mem_start) {
611             /* spapr_machine_init() checks for rma_size <= node0_size
612              * already */
613             spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size);
614             mem_start += spapr->rma_size;
615             node_size -= spapr->rma_size;
616         }
617         for ( ; node_size; ) {
618             hwaddr sizetmp = pow2floor(node_size);
619 
620             /* mem_start != 0 here */
621             if (ctzl(mem_start) < ctzl(sizetmp)) {
622                 sizetmp = 1ULL << ctzl(mem_start);
623             }
624 
625             spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp);
626             node_size -= sizetmp;
627             mem_start += sizetmp;
628         }
629     }
630 
631     /* Generate ibm,dynamic-reconfiguration-memory node if required */
632     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
633         int ret;
634 
635         g_assert(smc->dr_lmb_enabled);
636         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
637         if (ret) {
638             return ret;
639         }
640     }
641 
642     return 0;
643 }
644 
645 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
646                          SpaprMachineState *spapr)
647 {
648     MachineState *ms = MACHINE(spapr);
649     PowerPCCPU *cpu = POWERPC_CPU(cs);
650     CPUPPCState *env = &cpu->env;
651     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
652     int index = spapr_get_vcpu_id(cpu);
653     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
654                        0xffffffff, 0xffffffff};
655     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
656         : SPAPR_TIMEBASE_FREQ;
657     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
658     uint32_t page_sizes_prop[64];
659     size_t page_sizes_prop_size;
660     unsigned int smp_threads = ms->smp.threads;
661     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
662     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
663     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
664     SpaprDrc *drc;
665     int drc_index;
666     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
667     int i;
668 
669     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
670     if (drc) {
671         drc_index = spapr_drc_index(drc);
672         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
673     }
674 
675     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
676     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
677 
678     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
679     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
680                            env->dcache_line_size)));
681     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
682                            env->dcache_line_size)));
683     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
684                            env->icache_line_size)));
685     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
686                            env->icache_line_size)));
687 
688     if (pcc->l1_dcache_size) {
689         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
690                                pcc->l1_dcache_size)));
691     } else {
692         warn_report("Unknown L1 dcache size for cpu");
693     }
694     if (pcc->l1_icache_size) {
695         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
696                                pcc->l1_icache_size)));
697     } else {
698         warn_report("Unknown L1 icache size for cpu");
699     }
700 
701     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
702     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
703     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
704     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
705     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
706     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
707 
708     if (ppc_has_spr(cpu, SPR_PURR)) {
709         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
710     }
711     if (ppc_has_spr(cpu, SPR_PURR)) {
712         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
713     }
714 
715     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
716         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
717                           segs, sizeof(segs))));
718     }
719 
720     /* Advertise VSX (vector extensions) if available
721      *   1               == VMX / Altivec available
722      *   2               == VSX available
723      *
724      * Only CPUs for which we create core types in spapr_cpu_core.c
725      * are possible, and all of those have VMX */
726     if (env->insns_flags & PPC_ALTIVEC) {
727         if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
728             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
729         } else {
730             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
731         }
732     }
733 
734     /* Advertise DFP (Decimal Floating Point) if available
735      *   0 / no property == no DFP
736      *   1               == DFP available */
737     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
738         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
739     }
740 
741     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
742                                                       sizeof(page_sizes_prop));
743     if (page_sizes_prop_size) {
744         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
745                           page_sizes_prop, page_sizes_prop_size)));
746     }
747 
748     spapr_dt_pa_features(spapr, cpu, fdt, offset);
749 
750     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
751                            cs->cpu_index / vcpus_per_socket)));
752 
753     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
754                       pft_size_prop, sizeof(pft_size_prop))));
755 
756     if (ms->numa_state->num_nodes > 1) {
757         _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
758     }
759 
760     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
761 
762     if (pcc->radix_page_info) {
763         for (i = 0; i < pcc->radix_page_info->count; i++) {
764             radix_AP_encodings[i] =
765                 cpu_to_be32(pcc->radix_page_info->entries[i]);
766         }
767         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
768                           radix_AP_encodings,
769                           pcc->radix_page_info->count *
770                           sizeof(radix_AP_encodings[0]))));
771     }
772 
773     /*
774      * We set this property to let the guest know that it can use the large
775      * decrementer and its width in bits.
776      */
777     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
778         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
779                               pcc->lrg_decr_bits)));
780 }
781 
782 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
783 {
784     CPUState **rev;
785     CPUState *cs;
786     int n_cpus;
787     int cpus_offset;
788     int i;
789 
790     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
791     _FDT(cpus_offset);
792     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
793     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
794 
795     /*
796      * We walk the CPUs in reverse order to ensure that CPU DT nodes
797      * created by fdt_add_subnode() end up in the right order in FDT
798      * for the guest kernel the enumerate the CPUs correctly.
799      *
800      * The CPU list cannot be traversed in reverse order, so we need
801      * to do extra work.
802      */
803     n_cpus = 0;
804     rev = NULL;
805     CPU_FOREACH(cs) {
806         rev = g_renew(CPUState *, rev, n_cpus + 1);
807         rev[n_cpus++] = cs;
808     }
809 
810     for (i = n_cpus - 1; i >= 0; i--) {
811         CPUState *cs = rev[i];
812         PowerPCCPU *cpu = POWERPC_CPU(cs);
813         int index = spapr_get_vcpu_id(cpu);
814         DeviceClass *dc = DEVICE_GET_CLASS(cs);
815         g_autofree char *nodename = NULL;
816         int offset;
817 
818         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
819             continue;
820         }
821 
822         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
823         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
824         _FDT(offset);
825         spapr_dt_cpu(cs, fdt, offset, spapr);
826     }
827 
828     g_free(rev);
829 }
830 
831 static int spapr_dt_rng(void *fdt)
832 {
833     int node;
834     int ret;
835 
836     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
837     if (node <= 0) {
838         return -1;
839     }
840     ret = fdt_setprop_string(fdt, node, "device_type",
841                              "ibm,platform-facilities");
842     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
843     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
844 
845     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
846     if (node <= 0) {
847         return -1;
848     }
849     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
850 
851     return ret ? -1 : 0;
852 }
853 
854 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
855 {
856     MachineState *ms = MACHINE(spapr);
857     int rtas;
858     GString *hypertas = g_string_sized_new(256);
859     GString *qemu_hypertas = g_string_sized_new(256);
860     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
861         memory_region_size(&MACHINE(spapr)->device_memory->mr);
862     uint32_t lrdr_capacity[] = {
863         cpu_to_be32(max_device_addr >> 32),
864         cpu_to_be32(max_device_addr & 0xffffffff),
865         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
866         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
867         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
868     };
869 
870     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
871 
872     /* hypertas */
873     add_str(hypertas, "hcall-pft");
874     add_str(hypertas, "hcall-term");
875     add_str(hypertas, "hcall-dabr");
876     add_str(hypertas, "hcall-interrupt");
877     add_str(hypertas, "hcall-tce");
878     add_str(hypertas, "hcall-vio");
879     add_str(hypertas, "hcall-splpar");
880     add_str(hypertas, "hcall-join");
881     add_str(hypertas, "hcall-bulk");
882     add_str(hypertas, "hcall-set-mode");
883     add_str(hypertas, "hcall-sprg0");
884     add_str(hypertas, "hcall-copy");
885     add_str(hypertas, "hcall-debug");
886     add_str(hypertas, "hcall-vphn");
887     if (spapr_get_cap(spapr, SPAPR_CAP_RPT_INVALIDATE) == SPAPR_CAP_ON) {
888         add_str(hypertas, "hcall-rpt-invalidate");
889     }
890 
891     add_str(qemu_hypertas, "hcall-memop1");
892 
893     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
894         add_str(hypertas, "hcall-multi-tce");
895     }
896 
897     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
898         add_str(hypertas, "hcall-hpt-resize");
899     }
900 
901     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
902                      hypertas->str, hypertas->len));
903     g_string_free(hypertas, TRUE);
904     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
905                      qemu_hypertas->str, qemu_hypertas->len));
906     g_string_free(qemu_hypertas, TRUE);
907 
908     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
909 
910     /*
911      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
912      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
913      *
914      * The system reset requirements are driven by existing Linux and PowerVM
915      * implementation which (contrary to PAPR) saves r3 in the error log
916      * structure like machine check, so Linux expects to find the saved r3
917      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
918      * does not look at the error value).
919      *
920      * System reset interrupts are not subject to interlock like machine
921      * check, so this memory area could be corrupted if the sreset is
922      * interrupted by a machine check (or vice versa) if it was shared. To
923      * prevent this, system reset uses per-CPU areas for the sreset save
924      * area. A system reset that interrupts a system reset handler could
925      * still overwrite this area, but Linux doesn't try to recover in that
926      * case anyway.
927      *
928      * The extra 8 bytes is required because Linux's FWNMI error log check
929      * is off-by-one.
930      *
931      * RTAS_MIN_SIZE is required for the RTAS blob itself.
932      */
933     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_MIN_SIZE +
934                           RTAS_ERROR_LOG_MAX +
935                           ms->smp.max_cpus * sizeof(uint64_t) * 2 +
936                           sizeof(uint64_t)));
937     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
938                           RTAS_ERROR_LOG_MAX));
939     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
940                           RTAS_EVENT_SCAN_RATE));
941 
942     g_assert(msi_nonbroken);
943     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
944 
945     /*
946      * According to PAPR, rtas ibm,os-term does not guarantee a return
947      * back to the guest cpu.
948      *
949      * While an additional ibm,extended-os-term property indicates
950      * that rtas call return will always occur. Set this property.
951      */
952     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
953 
954     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
955                      lrdr_capacity, sizeof(lrdr_capacity)));
956 
957     spapr_dt_rtas_tokens(fdt, rtas);
958 }
959 
960 /*
961  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
962  * and the XIVE features that the guest may request and thus the valid
963  * values for bytes 23..26 of option vector 5:
964  */
965 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
966                                           int chosen)
967 {
968     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
969 
970     char val[2 * 4] = {
971         23, 0x00, /* XICS / XIVE mode */
972         24, 0x00, /* Hash/Radix, filled in below. */
973         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
974         26, 0x40, /* Radix options: GTSE == yes. */
975     };
976 
977     if (spapr->irq->xics && spapr->irq->xive) {
978         val[1] = SPAPR_OV5_XIVE_BOTH;
979     } else if (spapr->irq->xive) {
980         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
981     } else {
982         assert(spapr->irq->xics);
983         val[1] = SPAPR_OV5_XIVE_LEGACY;
984     }
985 
986     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
987                           first_ppc_cpu->compat_pvr)) {
988         /*
989          * If we're in a pre POWER9 compat mode then the guest should
990          * do hash and use the legacy interrupt mode
991          */
992         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
993         val[3] = 0x00; /* Hash */
994         spapr_check_mmu_mode(false);
995     } else if (kvm_enabled()) {
996         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
997             val[3] = 0x80; /* OV5_MMU_BOTH */
998         } else if (kvmppc_has_cap_mmu_radix()) {
999             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1000         } else {
1001             val[3] = 0x00; /* Hash */
1002         }
1003     } else {
1004         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1005         val[3] = 0xC0;
1006     }
1007     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1008                      val, sizeof(val)));
1009 }
1010 
1011 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1012 {
1013     MachineState *machine = MACHINE(spapr);
1014     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1015     int chosen;
1016 
1017     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1018 
1019     if (reset) {
1020         const char *boot_device = spapr->boot_device;
1021         char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1022         size_t cb = 0;
1023         char *bootlist = get_boot_devices_list(&cb);
1024 
1025         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1026             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1027                                     machine->kernel_cmdline));
1028         }
1029 
1030         if (spapr->initrd_size) {
1031             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1032                                   spapr->initrd_base));
1033             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1034                                   spapr->initrd_base + spapr->initrd_size));
1035         }
1036 
1037         if (spapr->kernel_size) {
1038             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1039                                   cpu_to_be64(spapr->kernel_size) };
1040 
1041             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1042                          &kprop, sizeof(kprop)));
1043             if (spapr->kernel_le) {
1044                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1045             }
1046         }
1047         if (boot_menu) {
1048             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1049         }
1050         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1051         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1052         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1053 
1054         if (cb && bootlist) {
1055             int i;
1056 
1057             for (i = 0; i < cb; i++) {
1058                 if (bootlist[i] == '\n') {
1059                     bootlist[i] = ' ';
1060                 }
1061             }
1062             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1063         }
1064 
1065         if (boot_device && strlen(boot_device)) {
1066             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1067         }
1068 
1069         if (!spapr->has_graphics && stdout_path) {
1070             /*
1071              * "linux,stdout-path" and "stdout" properties are
1072              * deprecated by linux kernel. New platforms should only
1073              * use the "stdout-path" property. Set the new property
1074              * and continue using older property to remain compatible
1075              * with the existing firmware.
1076              */
1077             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1078             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1079         }
1080 
1081         /*
1082          * We can deal with BAR reallocation just fine, advertise it
1083          * to the guest
1084          */
1085         if (smc->linux_pci_probe) {
1086             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1087         }
1088 
1089         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1090 
1091         g_free(stdout_path);
1092         g_free(bootlist);
1093     }
1094 
1095     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1096 }
1097 
1098 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1099 {
1100     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1101      * KVM to work under pHyp with some guest co-operation */
1102     int hypervisor;
1103     uint8_t hypercall[16];
1104 
1105     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1106     /* indicate KVM hypercall interface */
1107     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1108     if (kvmppc_has_cap_fixup_hcalls()) {
1109         /*
1110          * Older KVM versions with older guest kernels were broken
1111          * with the magic page, don't allow the guest to map it.
1112          */
1113         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1114                                   sizeof(hypercall))) {
1115             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1116                              hypercall, sizeof(hypercall)));
1117         }
1118     }
1119 }
1120 
1121 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1122 {
1123     MachineState *machine = MACHINE(spapr);
1124     MachineClass *mc = MACHINE_GET_CLASS(machine);
1125     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1126     uint32_t root_drc_type_mask = 0;
1127     int ret;
1128     void *fdt;
1129     SpaprPhbState *phb;
1130     char *buf;
1131 
1132     fdt = g_malloc0(space);
1133     _FDT((fdt_create_empty_tree(fdt, space)));
1134 
1135     /* Root node */
1136     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1137     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1138     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1139 
1140     /* Guest UUID & Name*/
1141     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1142     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1143     if (qemu_uuid_set) {
1144         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1145     }
1146     g_free(buf);
1147 
1148     if (qemu_get_vm_name()) {
1149         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1150                                 qemu_get_vm_name()));
1151     }
1152 
1153     /* Host Model & Serial Number */
1154     if (spapr->host_model) {
1155         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1156     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1157         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1158         g_free(buf);
1159     }
1160 
1161     if (spapr->host_serial) {
1162         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1163     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1164         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1165         g_free(buf);
1166     }
1167 
1168     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1169     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1170 
1171     /* /interrupt controller */
1172     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1173 
1174     ret = spapr_dt_memory(spapr, fdt);
1175     if (ret < 0) {
1176         error_report("couldn't setup memory nodes in fdt");
1177         exit(1);
1178     }
1179 
1180     /* /vdevice */
1181     spapr_dt_vdevice(spapr->vio_bus, fdt);
1182 
1183     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1184         ret = spapr_dt_rng(fdt);
1185         if (ret < 0) {
1186             error_report("could not set up rng device in the fdt");
1187             exit(1);
1188         }
1189     }
1190 
1191     QLIST_FOREACH(phb, &spapr->phbs, list) {
1192         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1193         if (ret < 0) {
1194             error_report("couldn't setup PCI devices in fdt");
1195             exit(1);
1196         }
1197     }
1198 
1199     spapr_dt_cpus(fdt, spapr);
1200 
1201     /* ibm,drc-indexes and friends */
1202     if (smc->dr_lmb_enabled) {
1203         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_LMB;
1204     }
1205     if (smc->dr_phb_enabled) {
1206         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PHB;
1207     }
1208     if (mc->nvdimm_supported) {
1209         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PMEM;
1210     }
1211     if (root_drc_type_mask) {
1212         _FDT(spapr_dt_drc(fdt, 0, NULL, root_drc_type_mask));
1213     }
1214 
1215     if (mc->has_hotpluggable_cpus) {
1216         int offset = fdt_path_offset(fdt, "/cpus");
1217         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1218         if (ret < 0) {
1219             error_report("Couldn't set up CPU DR device tree properties");
1220             exit(1);
1221         }
1222     }
1223 
1224     /* /event-sources */
1225     spapr_dt_events(spapr, fdt);
1226 
1227     /* /rtas */
1228     spapr_dt_rtas(spapr, fdt);
1229 
1230     /* /chosen */
1231     spapr_dt_chosen(spapr, fdt, reset);
1232 
1233     /* /hypervisor */
1234     if (kvm_enabled()) {
1235         spapr_dt_hypervisor(spapr, fdt);
1236     }
1237 
1238     /* Build memory reserve map */
1239     if (reset) {
1240         if (spapr->kernel_size) {
1241             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1242                                   spapr->kernel_size)));
1243         }
1244         if (spapr->initrd_size) {
1245             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1246                                   spapr->initrd_size)));
1247         }
1248     }
1249 
1250     /* NVDIMM devices */
1251     if (mc->nvdimm_supported) {
1252         spapr_dt_persistent_memory(spapr, fdt);
1253     }
1254 
1255     return fdt;
1256 }
1257 
1258 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1259 {
1260     SpaprMachineState *spapr = opaque;
1261 
1262     return (addr & 0x0fffffff) + spapr->kernel_addr;
1263 }
1264 
1265 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1266                                     PowerPCCPU *cpu)
1267 {
1268     CPUPPCState *env = &cpu->env;
1269 
1270     /* The TCG path should also be holding the BQL at this point */
1271     g_assert(qemu_mutex_iothread_locked());
1272 
1273     if (msr_pr) {
1274         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1275         env->gpr[3] = H_PRIVILEGE;
1276     } else {
1277         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1278     }
1279 }
1280 
1281 struct LPCRSyncState {
1282     target_ulong value;
1283     target_ulong mask;
1284 };
1285 
1286 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1287 {
1288     struct LPCRSyncState *s = arg.host_ptr;
1289     PowerPCCPU *cpu = POWERPC_CPU(cs);
1290     CPUPPCState *env = &cpu->env;
1291     target_ulong lpcr;
1292 
1293     cpu_synchronize_state(cs);
1294     lpcr = env->spr[SPR_LPCR];
1295     lpcr &= ~s->mask;
1296     lpcr |= s->value;
1297     ppc_store_lpcr(cpu, lpcr);
1298 }
1299 
1300 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1301 {
1302     CPUState *cs;
1303     struct LPCRSyncState s = {
1304         .value = value,
1305         .mask = mask
1306     };
1307     CPU_FOREACH(cs) {
1308         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1309     }
1310 }
1311 
1312 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1313 {
1314     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1315 
1316     /* Copy PATE1:GR into PATE0:HR */
1317     entry->dw0 = spapr->patb_entry & PATE0_HR;
1318     entry->dw1 = spapr->patb_entry;
1319 }
1320 
1321 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1322 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1323 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1324 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1325 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1326 
1327 /*
1328  * Get the fd to access the kernel htab, re-opening it if necessary
1329  */
1330 static int get_htab_fd(SpaprMachineState *spapr)
1331 {
1332     Error *local_err = NULL;
1333 
1334     if (spapr->htab_fd >= 0) {
1335         return spapr->htab_fd;
1336     }
1337 
1338     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1339     if (spapr->htab_fd < 0) {
1340         error_report_err(local_err);
1341     }
1342 
1343     return spapr->htab_fd;
1344 }
1345 
1346 void close_htab_fd(SpaprMachineState *spapr)
1347 {
1348     if (spapr->htab_fd >= 0) {
1349         close(spapr->htab_fd);
1350     }
1351     spapr->htab_fd = -1;
1352 }
1353 
1354 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1355 {
1356     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1357 
1358     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1359 }
1360 
1361 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1362 {
1363     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1364 
1365     assert(kvm_enabled());
1366 
1367     if (!spapr->htab) {
1368         return 0;
1369     }
1370 
1371     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1372 }
1373 
1374 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1375                                                 hwaddr ptex, int n)
1376 {
1377     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1378     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1379 
1380     if (!spapr->htab) {
1381         /*
1382          * HTAB is controlled by KVM. Fetch into temporary buffer
1383          */
1384         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1385         kvmppc_read_hptes(hptes, ptex, n);
1386         return hptes;
1387     }
1388 
1389     /*
1390      * HTAB is controlled by QEMU. Just point to the internally
1391      * accessible PTEG.
1392      */
1393     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1394 }
1395 
1396 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1397                               const ppc_hash_pte64_t *hptes,
1398                               hwaddr ptex, int n)
1399 {
1400     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1401 
1402     if (!spapr->htab) {
1403         g_free((void *)hptes);
1404     }
1405 
1406     /* Nothing to do for qemu managed HPT */
1407 }
1408 
1409 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1410                       uint64_t pte0, uint64_t pte1)
1411 {
1412     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1413     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1414 
1415     if (!spapr->htab) {
1416         kvmppc_write_hpte(ptex, pte0, pte1);
1417     } else {
1418         if (pte0 & HPTE64_V_VALID) {
1419             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1420             /*
1421              * When setting valid, we write PTE1 first. This ensures
1422              * proper synchronization with the reading code in
1423              * ppc_hash64_pteg_search()
1424              */
1425             smp_wmb();
1426             stq_p(spapr->htab + offset, pte0);
1427         } else {
1428             stq_p(spapr->htab + offset, pte0);
1429             /*
1430              * When clearing it we set PTE0 first. This ensures proper
1431              * synchronization with the reading code in
1432              * ppc_hash64_pteg_search()
1433              */
1434             smp_wmb();
1435             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1436         }
1437     }
1438 }
1439 
1440 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1441                              uint64_t pte1)
1442 {
1443     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_C;
1444     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1445 
1446     if (!spapr->htab) {
1447         /* There should always be a hash table when this is called */
1448         error_report("spapr_hpte_set_c called with no hash table !");
1449         return;
1450     }
1451 
1452     /* The HW performs a non-atomic byte update */
1453     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1454 }
1455 
1456 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1457                              uint64_t pte1)
1458 {
1459     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_R;
1460     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1461 
1462     if (!spapr->htab) {
1463         /* There should always be a hash table when this is called */
1464         error_report("spapr_hpte_set_r called with no hash table !");
1465         return;
1466     }
1467 
1468     /* The HW performs a non-atomic byte update */
1469     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1470 }
1471 
1472 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1473 {
1474     int shift;
1475 
1476     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1477      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1478      * that's much more than is needed for Linux guests */
1479     shift = ctz64(pow2ceil(ramsize)) - 7;
1480     shift = MAX(shift, 18); /* Minimum architected size */
1481     shift = MIN(shift, 46); /* Maximum architected size */
1482     return shift;
1483 }
1484 
1485 void spapr_free_hpt(SpaprMachineState *spapr)
1486 {
1487     g_free(spapr->htab);
1488     spapr->htab = NULL;
1489     spapr->htab_shift = 0;
1490     close_htab_fd(spapr);
1491 }
1492 
1493 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1494 {
1495     ERRP_GUARD();
1496     long rc;
1497 
1498     /* Clean up any HPT info from a previous boot */
1499     spapr_free_hpt(spapr);
1500 
1501     rc = kvmppc_reset_htab(shift);
1502 
1503     if (rc == -EOPNOTSUPP) {
1504         error_setg(errp, "HPT not supported in nested guests");
1505         return -EOPNOTSUPP;
1506     }
1507 
1508     if (rc < 0) {
1509         /* kernel-side HPT needed, but couldn't allocate one */
1510         error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1511                          shift);
1512         error_append_hint(errp, "Try smaller maxmem?\n");
1513         return -errno;
1514     } else if (rc > 0) {
1515         /* kernel-side HPT allocated */
1516         if (rc != shift) {
1517             error_setg(errp,
1518                        "Requested order %d HPT, but kernel allocated order %ld",
1519                        shift, rc);
1520             error_append_hint(errp, "Try smaller maxmem?\n");
1521             return -ENOSPC;
1522         }
1523 
1524         spapr->htab_shift = shift;
1525         spapr->htab = NULL;
1526     } else {
1527         /* kernel-side HPT not needed, allocate in userspace instead */
1528         size_t size = 1ULL << shift;
1529         int i;
1530 
1531         spapr->htab = qemu_memalign(size, size);
1532         memset(spapr->htab, 0, size);
1533         spapr->htab_shift = shift;
1534 
1535         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1536             DIRTY_HPTE(HPTE(spapr->htab, i));
1537         }
1538     }
1539     /* We're setting up a hash table, so that means we're not radix */
1540     spapr->patb_entry = 0;
1541     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1542     return 0;
1543 }
1544 
1545 void spapr_setup_hpt(SpaprMachineState *spapr)
1546 {
1547     int hpt_shift;
1548 
1549     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1550         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1551     } else {
1552         uint64_t current_ram_size;
1553 
1554         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1555         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1556     }
1557     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1558 
1559     if (kvm_enabled()) {
1560         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1561 
1562         /* Check our RMA fits in the possible VRMA */
1563         if (vrma_limit < spapr->rma_size) {
1564             error_report("Unable to create %" HWADDR_PRIu
1565                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1566                          spapr->rma_size / MiB, vrma_limit / MiB);
1567             exit(EXIT_FAILURE);
1568         }
1569     }
1570 }
1571 
1572 void spapr_check_mmu_mode(bool guest_radix)
1573 {
1574     if (guest_radix) {
1575         if (kvm_enabled() && !kvmppc_has_cap_mmu_radix()) {
1576             error_report("Guest requested unavailable MMU mode (radix).");
1577             exit(EXIT_FAILURE);
1578         }
1579     } else {
1580         if (kvm_enabled() && kvmppc_has_cap_mmu_radix()
1581             && !kvmppc_has_cap_mmu_hash_v3()) {
1582             error_report("Guest requested unavailable MMU mode (hash).");
1583             exit(EXIT_FAILURE);
1584         }
1585     }
1586 }
1587 
1588 static void spapr_machine_reset(MachineState *machine)
1589 {
1590     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1591     PowerPCCPU *first_ppc_cpu;
1592     hwaddr fdt_addr;
1593     void *fdt;
1594     int rc;
1595 
1596     pef_kvm_reset(machine->cgs, &error_fatal);
1597     spapr_caps_apply(spapr);
1598 
1599     first_ppc_cpu = POWERPC_CPU(first_cpu);
1600     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1601         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1602                               spapr->max_compat_pvr)) {
1603         /*
1604          * If using KVM with radix mode available, VCPUs can be started
1605          * without a HPT because KVM will start them in radix mode.
1606          * Set the GR bit in PATE so that we know there is no HPT.
1607          */
1608         spapr->patb_entry = PATE1_GR;
1609         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1610     } else {
1611         spapr_setup_hpt(spapr);
1612     }
1613 
1614     qemu_devices_reset();
1615 
1616     spapr_ovec_cleanup(spapr->ov5_cas);
1617     spapr->ov5_cas = spapr_ovec_new();
1618 
1619     ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1620 
1621     /*
1622      * This is fixing some of the default configuration of the XIVE
1623      * devices. To be called after the reset of the machine devices.
1624      */
1625     spapr_irq_reset(spapr, &error_fatal);
1626 
1627     /*
1628      * There is no CAS under qtest. Simulate one to please the code that
1629      * depends on spapr->ov5_cas. This is especially needed to test device
1630      * unplug, so we do that before resetting the DRCs.
1631      */
1632     if (qtest_enabled()) {
1633         spapr_ovec_cleanup(spapr->ov5_cas);
1634         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1635     }
1636 
1637     /* DRC reset may cause a device to be unplugged. This will cause troubles
1638      * if this device is used by another device (eg, a running vhost backend
1639      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1640      * situations, we reset DRCs after all devices have been reset.
1641      */
1642     spapr_drc_reset_all(spapr);
1643 
1644     spapr_clear_pending_events(spapr);
1645 
1646     /*
1647      * We place the device tree just below either the top of the RMA,
1648      * or just below 2GB, whichever is lower, so that it can be
1649      * processed with 32-bit real mode code if necessary
1650      */
1651     fdt_addr = MIN(spapr->rma_size, FDT_MAX_ADDR) - FDT_MAX_SIZE;
1652 
1653     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1654     if (spapr->vof) {
1655         spapr_vof_reset(spapr, fdt, &error_fatal);
1656         /*
1657          * Do not pack the FDT as the client may change properties.
1658          * VOF client does not expect the FDT so we do not load it to the VM.
1659          */
1660     } else {
1661         rc = fdt_pack(fdt);
1662         /* Should only fail if we've built a corrupted tree */
1663         assert(rc == 0);
1664 
1665         spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT,
1666                                   0, fdt_addr, 0);
1667         cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1668     }
1669     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1670 
1671     g_free(spapr->fdt_blob);
1672     spapr->fdt_size = fdt_totalsize(fdt);
1673     spapr->fdt_initial_size = spapr->fdt_size;
1674     spapr->fdt_blob = fdt;
1675 
1676     /* Set up the entry state */
1677     first_ppc_cpu->env.gpr[5] = 0;
1678 
1679     spapr->fwnmi_system_reset_addr = -1;
1680     spapr->fwnmi_machine_check_addr = -1;
1681     spapr->fwnmi_machine_check_interlock = -1;
1682 
1683     /* Signal all vCPUs waiting on this condition */
1684     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1685 
1686     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1687 }
1688 
1689 static void spapr_create_nvram(SpaprMachineState *spapr)
1690 {
1691     DeviceState *dev = qdev_new("spapr-nvram");
1692     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1693 
1694     if (dinfo) {
1695         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1696                                 &error_fatal);
1697     }
1698 
1699     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1700 
1701     spapr->nvram = (struct SpaprNvram *)dev;
1702 }
1703 
1704 static void spapr_rtc_create(SpaprMachineState *spapr)
1705 {
1706     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1707                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1708                                        &error_fatal, NULL);
1709     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1710     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1711                               "date");
1712 }
1713 
1714 /* Returns whether we want to use VGA or not */
1715 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1716 {
1717     switch (vga_interface_type) {
1718     case VGA_NONE:
1719         return false;
1720     case VGA_DEVICE:
1721         return true;
1722     case VGA_STD:
1723     case VGA_VIRTIO:
1724     case VGA_CIRRUS:
1725         return pci_vga_init(pci_bus) != NULL;
1726     default:
1727         error_setg(errp,
1728                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1729         return false;
1730     }
1731 }
1732 
1733 static int spapr_pre_load(void *opaque)
1734 {
1735     int rc;
1736 
1737     rc = spapr_caps_pre_load(opaque);
1738     if (rc) {
1739         return rc;
1740     }
1741 
1742     return 0;
1743 }
1744 
1745 static int spapr_post_load(void *opaque, int version_id)
1746 {
1747     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1748     int err = 0;
1749 
1750     err = spapr_caps_post_migration(spapr);
1751     if (err) {
1752         return err;
1753     }
1754 
1755     /*
1756      * In earlier versions, there was no separate qdev for the PAPR
1757      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1758      * So when migrating from those versions, poke the incoming offset
1759      * value into the RTC device
1760      */
1761     if (version_id < 3) {
1762         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1763         if (err) {
1764             return err;
1765         }
1766     }
1767 
1768     if (kvm_enabled() && spapr->patb_entry) {
1769         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1770         bool radix = !!(spapr->patb_entry & PATE1_GR);
1771         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1772 
1773         /*
1774          * Update LPCR:HR and UPRT as they may not be set properly in
1775          * the stream
1776          */
1777         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1778                             LPCR_HR | LPCR_UPRT);
1779 
1780         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1781         if (err) {
1782             error_report("Process table config unsupported by the host");
1783             return -EINVAL;
1784         }
1785     }
1786 
1787     err = spapr_irq_post_load(spapr, version_id);
1788     if (err) {
1789         return err;
1790     }
1791 
1792     return err;
1793 }
1794 
1795 static int spapr_pre_save(void *opaque)
1796 {
1797     int rc;
1798 
1799     rc = spapr_caps_pre_save(opaque);
1800     if (rc) {
1801         return rc;
1802     }
1803 
1804     return 0;
1805 }
1806 
1807 static bool version_before_3(void *opaque, int version_id)
1808 {
1809     return version_id < 3;
1810 }
1811 
1812 static bool spapr_pending_events_needed(void *opaque)
1813 {
1814     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1815     return !QTAILQ_EMPTY(&spapr->pending_events);
1816 }
1817 
1818 static const VMStateDescription vmstate_spapr_event_entry = {
1819     .name = "spapr_event_log_entry",
1820     .version_id = 1,
1821     .minimum_version_id = 1,
1822     .fields = (VMStateField[]) {
1823         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1824         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1825         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1826                                      NULL, extended_length),
1827         VMSTATE_END_OF_LIST()
1828     },
1829 };
1830 
1831 static const VMStateDescription vmstate_spapr_pending_events = {
1832     .name = "spapr_pending_events",
1833     .version_id = 1,
1834     .minimum_version_id = 1,
1835     .needed = spapr_pending_events_needed,
1836     .fields = (VMStateField[]) {
1837         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1838                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1839         VMSTATE_END_OF_LIST()
1840     },
1841 };
1842 
1843 static bool spapr_ov5_cas_needed(void *opaque)
1844 {
1845     SpaprMachineState *spapr = opaque;
1846     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1847     bool cas_needed;
1848 
1849     /* Prior to the introduction of SpaprOptionVector, we had two option
1850      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1851      * Both of these options encode machine topology into the device-tree
1852      * in such a way that the now-booted OS should still be able to interact
1853      * appropriately with QEMU regardless of what options were actually
1854      * negotiatied on the source side.
1855      *
1856      * As such, we can avoid migrating the CAS-negotiated options if these
1857      * are the only options available on the current machine/platform.
1858      * Since these are the only options available for pseries-2.7 and
1859      * earlier, this allows us to maintain old->new/new->old migration
1860      * compatibility.
1861      *
1862      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1863      * via default pseries-2.8 machines and explicit command-line parameters.
1864      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1865      * of the actual CAS-negotiated values to continue working properly. For
1866      * example, availability of memory unplug depends on knowing whether
1867      * OV5_HP_EVT was negotiated via CAS.
1868      *
1869      * Thus, for any cases where the set of available CAS-negotiatable
1870      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1871      * include the CAS-negotiated options in the migration stream, unless
1872      * if they affect boot time behaviour only.
1873      */
1874     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1875     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1876     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1877 
1878     /* We need extra information if we have any bits outside the mask
1879      * defined above */
1880     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1881 
1882     spapr_ovec_cleanup(ov5_mask);
1883 
1884     return cas_needed;
1885 }
1886 
1887 static const VMStateDescription vmstate_spapr_ov5_cas = {
1888     .name = "spapr_option_vector_ov5_cas",
1889     .version_id = 1,
1890     .minimum_version_id = 1,
1891     .needed = spapr_ov5_cas_needed,
1892     .fields = (VMStateField[]) {
1893         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1894                                  vmstate_spapr_ovec, SpaprOptionVector),
1895         VMSTATE_END_OF_LIST()
1896     },
1897 };
1898 
1899 static bool spapr_patb_entry_needed(void *opaque)
1900 {
1901     SpaprMachineState *spapr = opaque;
1902 
1903     return !!spapr->patb_entry;
1904 }
1905 
1906 static const VMStateDescription vmstate_spapr_patb_entry = {
1907     .name = "spapr_patb_entry",
1908     .version_id = 1,
1909     .minimum_version_id = 1,
1910     .needed = spapr_patb_entry_needed,
1911     .fields = (VMStateField[]) {
1912         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1913         VMSTATE_END_OF_LIST()
1914     },
1915 };
1916 
1917 static bool spapr_irq_map_needed(void *opaque)
1918 {
1919     SpaprMachineState *spapr = opaque;
1920 
1921     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1922 }
1923 
1924 static const VMStateDescription vmstate_spapr_irq_map = {
1925     .name = "spapr_irq_map",
1926     .version_id = 1,
1927     .minimum_version_id = 1,
1928     .needed = spapr_irq_map_needed,
1929     .fields = (VMStateField[]) {
1930         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1931         VMSTATE_END_OF_LIST()
1932     },
1933 };
1934 
1935 static bool spapr_dtb_needed(void *opaque)
1936 {
1937     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1938 
1939     return smc->update_dt_enabled;
1940 }
1941 
1942 static int spapr_dtb_pre_load(void *opaque)
1943 {
1944     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1945 
1946     g_free(spapr->fdt_blob);
1947     spapr->fdt_blob = NULL;
1948     spapr->fdt_size = 0;
1949 
1950     return 0;
1951 }
1952 
1953 static const VMStateDescription vmstate_spapr_dtb = {
1954     .name = "spapr_dtb",
1955     .version_id = 1,
1956     .minimum_version_id = 1,
1957     .needed = spapr_dtb_needed,
1958     .pre_load = spapr_dtb_pre_load,
1959     .fields = (VMStateField[]) {
1960         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1961         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1962         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1963                                      fdt_size),
1964         VMSTATE_END_OF_LIST()
1965     },
1966 };
1967 
1968 static bool spapr_fwnmi_needed(void *opaque)
1969 {
1970     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1971 
1972     return spapr->fwnmi_machine_check_addr != -1;
1973 }
1974 
1975 static int spapr_fwnmi_pre_save(void *opaque)
1976 {
1977     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1978 
1979     /*
1980      * Check if machine check handling is in progress and print a
1981      * warning message.
1982      */
1983     if (spapr->fwnmi_machine_check_interlock != -1) {
1984         warn_report("A machine check is being handled during migration. The"
1985                 "handler may run and log hardware error on the destination");
1986     }
1987 
1988     return 0;
1989 }
1990 
1991 static const VMStateDescription vmstate_spapr_fwnmi = {
1992     .name = "spapr_fwnmi",
1993     .version_id = 1,
1994     .minimum_version_id = 1,
1995     .needed = spapr_fwnmi_needed,
1996     .pre_save = spapr_fwnmi_pre_save,
1997     .fields = (VMStateField[]) {
1998         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
1999         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
2000         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
2001         VMSTATE_END_OF_LIST()
2002     },
2003 };
2004 
2005 static const VMStateDescription vmstate_spapr = {
2006     .name = "spapr",
2007     .version_id = 3,
2008     .minimum_version_id = 1,
2009     .pre_load = spapr_pre_load,
2010     .post_load = spapr_post_load,
2011     .pre_save = spapr_pre_save,
2012     .fields = (VMStateField[]) {
2013         /* used to be @next_irq */
2014         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2015 
2016         /* RTC offset */
2017         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2018 
2019         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2020         VMSTATE_END_OF_LIST()
2021     },
2022     .subsections = (const VMStateDescription*[]) {
2023         &vmstate_spapr_ov5_cas,
2024         &vmstate_spapr_patb_entry,
2025         &vmstate_spapr_pending_events,
2026         &vmstate_spapr_cap_htm,
2027         &vmstate_spapr_cap_vsx,
2028         &vmstate_spapr_cap_dfp,
2029         &vmstate_spapr_cap_cfpc,
2030         &vmstate_spapr_cap_sbbc,
2031         &vmstate_spapr_cap_ibs,
2032         &vmstate_spapr_cap_hpt_maxpagesize,
2033         &vmstate_spapr_irq_map,
2034         &vmstate_spapr_cap_nested_kvm_hv,
2035         &vmstate_spapr_dtb,
2036         &vmstate_spapr_cap_large_decr,
2037         &vmstate_spapr_cap_ccf_assist,
2038         &vmstate_spapr_cap_fwnmi,
2039         &vmstate_spapr_fwnmi,
2040         &vmstate_spapr_cap_rpt_invalidate,
2041         NULL
2042     }
2043 };
2044 
2045 static int htab_save_setup(QEMUFile *f, void *opaque)
2046 {
2047     SpaprMachineState *spapr = opaque;
2048 
2049     /* "Iteration" header */
2050     if (!spapr->htab_shift) {
2051         qemu_put_be32(f, -1);
2052     } else {
2053         qemu_put_be32(f, spapr->htab_shift);
2054     }
2055 
2056     if (spapr->htab) {
2057         spapr->htab_save_index = 0;
2058         spapr->htab_first_pass = true;
2059     } else {
2060         if (spapr->htab_shift) {
2061             assert(kvm_enabled());
2062         }
2063     }
2064 
2065 
2066     return 0;
2067 }
2068 
2069 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2070                             int chunkstart, int n_valid, int n_invalid)
2071 {
2072     qemu_put_be32(f, chunkstart);
2073     qemu_put_be16(f, n_valid);
2074     qemu_put_be16(f, n_invalid);
2075     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2076                     HASH_PTE_SIZE_64 * n_valid);
2077 }
2078 
2079 static void htab_save_end_marker(QEMUFile *f)
2080 {
2081     qemu_put_be32(f, 0);
2082     qemu_put_be16(f, 0);
2083     qemu_put_be16(f, 0);
2084 }
2085 
2086 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2087                                  int64_t max_ns)
2088 {
2089     bool has_timeout = max_ns != -1;
2090     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2091     int index = spapr->htab_save_index;
2092     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2093 
2094     assert(spapr->htab_first_pass);
2095 
2096     do {
2097         int chunkstart;
2098 
2099         /* Consume invalid HPTEs */
2100         while ((index < htabslots)
2101                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2102             CLEAN_HPTE(HPTE(spapr->htab, index));
2103             index++;
2104         }
2105 
2106         /* Consume valid HPTEs */
2107         chunkstart = index;
2108         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2109                && HPTE_VALID(HPTE(spapr->htab, index))) {
2110             CLEAN_HPTE(HPTE(spapr->htab, index));
2111             index++;
2112         }
2113 
2114         if (index > chunkstart) {
2115             int n_valid = index - chunkstart;
2116 
2117             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2118 
2119             if (has_timeout &&
2120                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2121                 break;
2122             }
2123         }
2124     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2125 
2126     if (index >= htabslots) {
2127         assert(index == htabslots);
2128         index = 0;
2129         spapr->htab_first_pass = false;
2130     }
2131     spapr->htab_save_index = index;
2132 }
2133 
2134 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2135                                 int64_t max_ns)
2136 {
2137     bool final = max_ns < 0;
2138     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2139     int examined = 0, sent = 0;
2140     int index = spapr->htab_save_index;
2141     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2142 
2143     assert(!spapr->htab_first_pass);
2144 
2145     do {
2146         int chunkstart, invalidstart;
2147 
2148         /* Consume non-dirty HPTEs */
2149         while ((index < htabslots)
2150                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2151             index++;
2152             examined++;
2153         }
2154 
2155         chunkstart = index;
2156         /* Consume valid dirty HPTEs */
2157         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2158                && HPTE_DIRTY(HPTE(spapr->htab, index))
2159                && HPTE_VALID(HPTE(spapr->htab, index))) {
2160             CLEAN_HPTE(HPTE(spapr->htab, index));
2161             index++;
2162             examined++;
2163         }
2164 
2165         invalidstart = index;
2166         /* Consume invalid dirty HPTEs */
2167         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2168                && HPTE_DIRTY(HPTE(spapr->htab, index))
2169                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2170             CLEAN_HPTE(HPTE(spapr->htab, index));
2171             index++;
2172             examined++;
2173         }
2174 
2175         if (index > chunkstart) {
2176             int n_valid = invalidstart - chunkstart;
2177             int n_invalid = index - invalidstart;
2178 
2179             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2180             sent += index - chunkstart;
2181 
2182             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2183                 break;
2184             }
2185         }
2186 
2187         if (examined >= htabslots) {
2188             break;
2189         }
2190 
2191         if (index >= htabslots) {
2192             assert(index == htabslots);
2193             index = 0;
2194         }
2195     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2196 
2197     if (index >= htabslots) {
2198         assert(index == htabslots);
2199         index = 0;
2200     }
2201 
2202     spapr->htab_save_index = index;
2203 
2204     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2205 }
2206 
2207 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2208 #define MAX_KVM_BUF_SIZE    2048
2209 
2210 static int htab_save_iterate(QEMUFile *f, void *opaque)
2211 {
2212     SpaprMachineState *spapr = opaque;
2213     int fd;
2214     int rc = 0;
2215 
2216     /* Iteration header */
2217     if (!spapr->htab_shift) {
2218         qemu_put_be32(f, -1);
2219         return 1;
2220     } else {
2221         qemu_put_be32(f, 0);
2222     }
2223 
2224     if (!spapr->htab) {
2225         assert(kvm_enabled());
2226 
2227         fd = get_htab_fd(spapr);
2228         if (fd < 0) {
2229             return fd;
2230         }
2231 
2232         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2233         if (rc < 0) {
2234             return rc;
2235         }
2236     } else  if (spapr->htab_first_pass) {
2237         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2238     } else {
2239         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2240     }
2241 
2242     htab_save_end_marker(f);
2243 
2244     return rc;
2245 }
2246 
2247 static int htab_save_complete(QEMUFile *f, void *opaque)
2248 {
2249     SpaprMachineState *spapr = opaque;
2250     int fd;
2251 
2252     /* Iteration header */
2253     if (!spapr->htab_shift) {
2254         qemu_put_be32(f, -1);
2255         return 0;
2256     } else {
2257         qemu_put_be32(f, 0);
2258     }
2259 
2260     if (!spapr->htab) {
2261         int rc;
2262 
2263         assert(kvm_enabled());
2264 
2265         fd = get_htab_fd(spapr);
2266         if (fd < 0) {
2267             return fd;
2268         }
2269 
2270         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2271         if (rc < 0) {
2272             return rc;
2273         }
2274     } else {
2275         if (spapr->htab_first_pass) {
2276             htab_save_first_pass(f, spapr, -1);
2277         }
2278         htab_save_later_pass(f, spapr, -1);
2279     }
2280 
2281     /* End marker */
2282     htab_save_end_marker(f);
2283 
2284     return 0;
2285 }
2286 
2287 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2288 {
2289     SpaprMachineState *spapr = opaque;
2290     uint32_t section_hdr;
2291     int fd = -1;
2292     Error *local_err = NULL;
2293 
2294     if (version_id < 1 || version_id > 1) {
2295         error_report("htab_load() bad version");
2296         return -EINVAL;
2297     }
2298 
2299     section_hdr = qemu_get_be32(f);
2300 
2301     if (section_hdr == -1) {
2302         spapr_free_hpt(spapr);
2303         return 0;
2304     }
2305 
2306     if (section_hdr) {
2307         int ret;
2308 
2309         /* First section gives the htab size */
2310         ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2311         if (ret < 0) {
2312             error_report_err(local_err);
2313             return ret;
2314         }
2315         return 0;
2316     }
2317 
2318     if (!spapr->htab) {
2319         assert(kvm_enabled());
2320 
2321         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2322         if (fd < 0) {
2323             error_report_err(local_err);
2324             return fd;
2325         }
2326     }
2327 
2328     while (true) {
2329         uint32_t index;
2330         uint16_t n_valid, n_invalid;
2331 
2332         index = qemu_get_be32(f);
2333         n_valid = qemu_get_be16(f);
2334         n_invalid = qemu_get_be16(f);
2335 
2336         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2337             /* End of Stream */
2338             break;
2339         }
2340 
2341         if ((index + n_valid + n_invalid) >
2342             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2343             /* Bad index in stream */
2344             error_report(
2345                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2346                 index, n_valid, n_invalid, spapr->htab_shift);
2347             return -EINVAL;
2348         }
2349 
2350         if (spapr->htab) {
2351             if (n_valid) {
2352                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2353                                 HASH_PTE_SIZE_64 * n_valid);
2354             }
2355             if (n_invalid) {
2356                 memset(HPTE(spapr->htab, index + n_valid), 0,
2357                        HASH_PTE_SIZE_64 * n_invalid);
2358             }
2359         } else {
2360             int rc;
2361 
2362             assert(fd >= 0);
2363 
2364             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2365                                         &local_err);
2366             if (rc < 0) {
2367                 error_report_err(local_err);
2368                 return rc;
2369             }
2370         }
2371     }
2372 
2373     if (!spapr->htab) {
2374         assert(fd >= 0);
2375         close(fd);
2376     }
2377 
2378     return 0;
2379 }
2380 
2381 static void htab_save_cleanup(void *opaque)
2382 {
2383     SpaprMachineState *spapr = opaque;
2384 
2385     close_htab_fd(spapr);
2386 }
2387 
2388 static SaveVMHandlers savevm_htab_handlers = {
2389     .save_setup = htab_save_setup,
2390     .save_live_iterate = htab_save_iterate,
2391     .save_live_complete_precopy = htab_save_complete,
2392     .save_cleanup = htab_save_cleanup,
2393     .load_state = htab_load,
2394 };
2395 
2396 static void spapr_boot_set(void *opaque, const char *boot_device,
2397                            Error **errp)
2398 {
2399     SpaprMachineState *spapr = SPAPR_MACHINE(opaque);
2400 
2401     g_free(spapr->boot_device);
2402     spapr->boot_device = g_strdup(boot_device);
2403 }
2404 
2405 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2406 {
2407     MachineState *machine = MACHINE(spapr);
2408     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2409     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2410     int i;
2411 
2412     for (i = 0; i < nr_lmbs; i++) {
2413         uint64_t addr;
2414 
2415         addr = i * lmb_size + machine->device_memory->base;
2416         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2417                                addr / lmb_size);
2418     }
2419 }
2420 
2421 /*
2422  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2423  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2424  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2425  */
2426 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2427 {
2428     int i;
2429 
2430     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2431         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2432                    " is not aligned to %" PRIu64 " MiB",
2433                    machine->ram_size,
2434                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2435         return;
2436     }
2437 
2438     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2439         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2440                    " is not aligned to %" PRIu64 " MiB",
2441                    machine->ram_size,
2442                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2443         return;
2444     }
2445 
2446     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2447         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2448             error_setg(errp,
2449                        "Node %d memory size 0x%" PRIx64
2450                        " is not aligned to %" PRIu64 " MiB",
2451                        i, machine->numa_state->nodes[i].node_mem,
2452                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2453             return;
2454         }
2455     }
2456 }
2457 
2458 /* find cpu slot in machine->possible_cpus by core_id */
2459 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2460 {
2461     int index = id / ms->smp.threads;
2462 
2463     if (index >= ms->possible_cpus->len) {
2464         return NULL;
2465     }
2466     if (idx) {
2467         *idx = index;
2468     }
2469     return &ms->possible_cpus->cpus[index];
2470 }
2471 
2472 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2473 {
2474     MachineState *ms = MACHINE(spapr);
2475     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2476     Error *local_err = NULL;
2477     bool vsmt_user = !!spapr->vsmt;
2478     int kvm_smt = kvmppc_smt_threads();
2479     int ret;
2480     unsigned int smp_threads = ms->smp.threads;
2481 
2482     if (!kvm_enabled() && (smp_threads > 1)) {
2483         error_setg(errp, "TCG cannot support more than 1 thread/core "
2484                    "on a pseries machine");
2485         return;
2486     }
2487     if (!is_power_of_2(smp_threads)) {
2488         error_setg(errp, "Cannot support %d threads/core on a pseries "
2489                    "machine because it must be a power of 2", smp_threads);
2490         return;
2491     }
2492 
2493     /* Detemine the VSMT mode to use: */
2494     if (vsmt_user) {
2495         if (spapr->vsmt < smp_threads) {
2496             error_setg(errp, "Cannot support VSMT mode %d"
2497                        " because it must be >= threads/core (%d)",
2498                        spapr->vsmt, smp_threads);
2499             return;
2500         }
2501         /* In this case, spapr->vsmt has been set by the command line */
2502     } else if (!smc->smp_threads_vsmt) {
2503         /*
2504          * Default VSMT value is tricky, because we need it to be as
2505          * consistent as possible (for migration), but this requires
2506          * changing it for at least some existing cases.  We pick 8 as
2507          * the value that we'd get with KVM on POWER8, the
2508          * overwhelmingly common case in production systems.
2509          */
2510         spapr->vsmt = MAX(8, smp_threads);
2511     } else {
2512         spapr->vsmt = smp_threads;
2513     }
2514 
2515     /* KVM: If necessary, set the SMT mode: */
2516     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2517         ret = kvmppc_set_smt_threads(spapr->vsmt);
2518         if (ret) {
2519             /* Looks like KVM isn't able to change VSMT mode */
2520             error_setg(&local_err,
2521                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2522                        spapr->vsmt, ret);
2523             /* We can live with that if the default one is big enough
2524              * for the number of threads, and a submultiple of the one
2525              * we want.  In this case we'll waste some vcpu ids, but
2526              * behaviour will be correct */
2527             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2528                 warn_report_err(local_err);
2529             } else {
2530                 if (!vsmt_user) {
2531                     error_append_hint(&local_err,
2532                                       "On PPC, a VM with %d threads/core"
2533                                       " on a host with %d threads/core"
2534                                       " requires the use of VSMT mode %d.\n",
2535                                       smp_threads, kvm_smt, spapr->vsmt);
2536                 }
2537                 kvmppc_error_append_smt_possible_hint(&local_err);
2538                 error_propagate(errp, local_err);
2539             }
2540         }
2541     }
2542     /* else TCG: nothing to do currently */
2543 }
2544 
2545 static void spapr_init_cpus(SpaprMachineState *spapr)
2546 {
2547     MachineState *machine = MACHINE(spapr);
2548     MachineClass *mc = MACHINE_GET_CLASS(machine);
2549     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2550     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2551     const CPUArchIdList *possible_cpus;
2552     unsigned int smp_cpus = machine->smp.cpus;
2553     unsigned int smp_threads = machine->smp.threads;
2554     unsigned int max_cpus = machine->smp.max_cpus;
2555     int boot_cores_nr = smp_cpus / smp_threads;
2556     int i;
2557 
2558     possible_cpus = mc->possible_cpu_arch_ids(machine);
2559     if (mc->has_hotpluggable_cpus) {
2560         if (smp_cpus % smp_threads) {
2561             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2562                          smp_cpus, smp_threads);
2563             exit(1);
2564         }
2565         if (max_cpus % smp_threads) {
2566             error_report("max_cpus (%u) must be multiple of threads (%u)",
2567                          max_cpus, smp_threads);
2568             exit(1);
2569         }
2570     } else {
2571         if (max_cpus != smp_cpus) {
2572             error_report("This machine version does not support CPU hotplug");
2573             exit(1);
2574         }
2575         boot_cores_nr = possible_cpus->len;
2576     }
2577 
2578     if (smc->pre_2_10_has_unused_icps) {
2579         int i;
2580 
2581         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2582             /* Dummy entries get deregistered when real ICPState objects
2583              * are registered during CPU core hotplug.
2584              */
2585             pre_2_10_vmstate_register_dummy_icp(i);
2586         }
2587     }
2588 
2589     for (i = 0; i < possible_cpus->len; i++) {
2590         int core_id = i * smp_threads;
2591 
2592         if (mc->has_hotpluggable_cpus) {
2593             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2594                                    spapr_vcpu_id(spapr, core_id));
2595         }
2596 
2597         if (i < boot_cores_nr) {
2598             Object *core  = object_new(type);
2599             int nr_threads = smp_threads;
2600 
2601             /* Handle the partially filled core for older machine types */
2602             if ((i + 1) * smp_threads >= smp_cpus) {
2603                 nr_threads = smp_cpus - i * smp_threads;
2604             }
2605 
2606             object_property_set_int(core, "nr-threads", nr_threads,
2607                                     &error_fatal);
2608             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2609                                     &error_fatal);
2610             qdev_realize(DEVICE(core), NULL, &error_fatal);
2611 
2612             object_unref(core);
2613         }
2614     }
2615 }
2616 
2617 static PCIHostState *spapr_create_default_phb(void)
2618 {
2619     DeviceState *dev;
2620 
2621     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2622     qdev_prop_set_uint32(dev, "index", 0);
2623     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2624 
2625     return PCI_HOST_BRIDGE(dev);
2626 }
2627 
2628 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2629 {
2630     MachineState *machine = MACHINE(spapr);
2631     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2632     hwaddr rma_size = machine->ram_size;
2633     hwaddr node0_size = spapr_node0_size(machine);
2634 
2635     /* RMA has to fit in the first NUMA node */
2636     rma_size = MIN(rma_size, node0_size);
2637 
2638     /*
2639      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2640      * never exceed that
2641      */
2642     rma_size = MIN(rma_size, 1 * TiB);
2643 
2644     /*
2645      * Clamp the RMA size based on machine type.  This is for
2646      * migration compatibility with older qemu versions, which limited
2647      * the RMA size for complicated and mostly bad reasons.
2648      */
2649     if (smc->rma_limit) {
2650         rma_size = MIN(rma_size, smc->rma_limit);
2651     }
2652 
2653     if (rma_size < MIN_RMA_SLOF) {
2654         error_setg(errp,
2655                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2656                    "ldMiB guest RMA (Real Mode Area memory)",
2657                    MIN_RMA_SLOF / MiB);
2658         return 0;
2659     }
2660 
2661     return rma_size;
2662 }
2663 
2664 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2665 {
2666     MachineState *machine = MACHINE(spapr);
2667     int i;
2668 
2669     for (i = 0; i < machine->ram_slots; i++) {
2670         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2671     }
2672 }
2673 
2674 /* pSeries LPAR / sPAPR hardware init */
2675 static void spapr_machine_init(MachineState *machine)
2676 {
2677     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2678     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2679     MachineClass *mc = MACHINE_GET_CLASS(machine);
2680     const char *bios_default = spapr->vof ? FW_FILE_NAME_VOF : FW_FILE_NAME;
2681     const char *bios_name = machine->firmware ?: bios_default;
2682     const char *kernel_filename = machine->kernel_filename;
2683     const char *initrd_filename = machine->initrd_filename;
2684     PCIHostState *phb;
2685     int i;
2686     MemoryRegion *sysmem = get_system_memory();
2687     long load_limit, fw_size;
2688     char *filename;
2689     Error *resize_hpt_err = NULL;
2690 
2691     /*
2692      * if Secure VM (PEF) support is configured, then initialize it
2693      */
2694     pef_kvm_init(machine->cgs, &error_fatal);
2695 
2696     msi_nonbroken = true;
2697 
2698     QLIST_INIT(&spapr->phbs);
2699     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2700 
2701     /* Determine capabilities to run with */
2702     spapr_caps_init(spapr);
2703 
2704     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2705     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2706         /*
2707          * If the user explicitly requested a mode we should either
2708          * supply it, or fail completely (which we do below).  But if
2709          * it's not set explicitly, we reset our mode to something
2710          * that works
2711          */
2712         if (resize_hpt_err) {
2713             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2714             error_free(resize_hpt_err);
2715             resize_hpt_err = NULL;
2716         } else {
2717             spapr->resize_hpt = smc->resize_hpt_default;
2718         }
2719     }
2720 
2721     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2722 
2723     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2724         /*
2725          * User requested HPT resize, but this host can't supply it.  Bail out
2726          */
2727         error_report_err(resize_hpt_err);
2728         exit(1);
2729     }
2730     error_free(resize_hpt_err);
2731 
2732     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2733 
2734     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2735     load_limit = MIN(spapr->rma_size, FDT_MAX_ADDR) - FW_OVERHEAD;
2736 
2737     /*
2738      * VSMT must be set in order to be able to compute VCPU ids, ie to
2739      * call spapr_max_server_number() or spapr_vcpu_id().
2740      */
2741     spapr_set_vsmt_mode(spapr, &error_fatal);
2742 
2743     /* Set up Interrupt Controller before we create the VCPUs */
2744     spapr_irq_init(spapr, &error_fatal);
2745 
2746     /* Set up containers for ibm,client-architecture-support negotiated options
2747      */
2748     spapr->ov5 = spapr_ovec_new();
2749     spapr->ov5_cas = spapr_ovec_new();
2750 
2751     if (smc->dr_lmb_enabled) {
2752         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2753         spapr_validate_node_memory(machine, &error_fatal);
2754     }
2755 
2756     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2757 
2758     /* Do not advertise FORM2 NUMA support for pseries-6.1 and older */
2759     if (!smc->pre_6_2_numa_affinity) {
2760         spapr_ovec_set(spapr->ov5, OV5_FORM2_AFFINITY);
2761     }
2762 
2763     /* advertise support for dedicated HP event source to guests */
2764     if (spapr->use_hotplug_event_source) {
2765         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2766     }
2767 
2768     /* advertise support for HPT resizing */
2769     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2770         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2771     }
2772 
2773     /* advertise support for ibm,dyamic-memory-v2 */
2774     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2775 
2776     /* advertise XIVE on POWER9 machines */
2777     if (spapr->irq->xive) {
2778         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2779     }
2780 
2781     /* init CPUs */
2782     spapr_init_cpus(spapr);
2783 
2784     spapr->gpu_numa_id = spapr_numa_initial_nvgpu_numa_id(machine);
2785 
2786     /* Init numa_assoc_array */
2787     spapr_numa_associativity_init(spapr, machine);
2788 
2789     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2790         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2791                               spapr->max_compat_pvr)) {
2792         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2793         /* KVM and TCG always allow GTSE with radix... */
2794         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2795     }
2796     /* ... but not with hash (currently). */
2797 
2798     if (kvm_enabled()) {
2799         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2800         kvmppc_enable_logical_ci_hcalls();
2801         kvmppc_enable_set_mode_hcall();
2802 
2803         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2804         kvmppc_enable_clear_ref_mod_hcalls();
2805 
2806         /* Enable H_PAGE_INIT */
2807         kvmppc_enable_h_page_init();
2808     }
2809 
2810     /* map RAM */
2811     memory_region_add_subregion(sysmem, 0, machine->ram);
2812 
2813     /* always allocate the device memory information */
2814     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2815 
2816     /* initialize hotplug memory address space */
2817     if (machine->ram_size < machine->maxram_size) {
2818         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2819         /*
2820          * Limit the number of hotpluggable memory slots to half the number
2821          * slots that KVM supports, leaving the other half for PCI and other
2822          * devices. However ensure that number of slots doesn't drop below 32.
2823          */
2824         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2825                            SPAPR_MAX_RAM_SLOTS;
2826 
2827         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2828             max_memslots = SPAPR_MAX_RAM_SLOTS;
2829         }
2830         if (machine->ram_slots > max_memslots) {
2831             error_report("Specified number of memory slots %"
2832                          PRIu64" exceeds max supported %d",
2833                          machine->ram_slots, max_memslots);
2834             exit(1);
2835         }
2836 
2837         machine->device_memory->base = ROUND_UP(machine->ram_size,
2838                                                 SPAPR_DEVICE_MEM_ALIGN);
2839         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2840                            "device-memory", device_mem_size);
2841         memory_region_add_subregion(sysmem, machine->device_memory->base,
2842                                     &machine->device_memory->mr);
2843     }
2844 
2845     if (smc->dr_lmb_enabled) {
2846         spapr_create_lmb_dr_connectors(spapr);
2847     }
2848 
2849     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2850         /* Create the error string for live migration blocker */
2851         error_setg(&spapr->fwnmi_migration_blocker,
2852             "A machine check is being handled during migration. The handler"
2853             "may run and log hardware error on the destination");
2854     }
2855 
2856     if (mc->nvdimm_supported) {
2857         spapr_create_nvdimm_dr_connectors(spapr);
2858     }
2859 
2860     /* Set up RTAS event infrastructure */
2861     spapr_events_init(spapr);
2862 
2863     /* Set up the RTC RTAS interfaces */
2864     spapr_rtc_create(spapr);
2865 
2866     /* Set up VIO bus */
2867     spapr->vio_bus = spapr_vio_bus_init();
2868 
2869     for (i = 0; serial_hd(i); i++) {
2870         spapr_vty_create(spapr->vio_bus, serial_hd(i));
2871     }
2872 
2873     /* We always have at least the nvram device on VIO */
2874     spapr_create_nvram(spapr);
2875 
2876     /*
2877      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2878      * connectors (described in root DT node's "ibm,drc-types" property)
2879      * are pre-initialized here. additional child connectors (such as
2880      * connectors for a PHBs PCI slots) are added as needed during their
2881      * parent's realization.
2882      */
2883     if (smc->dr_phb_enabled) {
2884         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2885             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2886         }
2887     }
2888 
2889     /* Set up PCI */
2890     spapr_pci_rtas_init();
2891 
2892     phb = spapr_create_default_phb();
2893 
2894     for (i = 0; i < nb_nics; i++) {
2895         NICInfo *nd = &nd_table[i];
2896 
2897         if (!nd->model) {
2898             nd->model = g_strdup("spapr-vlan");
2899         }
2900 
2901         if (g_str_equal(nd->model, "spapr-vlan") ||
2902             g_str_equal(nd->model, "ibmveth")) {
2903             spapr_vlan_create(spapr->vio_bus, nd);
2904         } else {
2905             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2906         }
2907     }
2908 
2909     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2910         spapr_vscsi_create(spapr->vio_bus);
2911     }
2912 
2913     /* Graphics */
2914     if (spapr_vga_init(phb->bus, &error_fatal)) {
2915         spapr->has_graphics = true;
2916         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2917     }
2918 
2919     if (machine->usb) {
2920         if (smc->use_ohci_by_default) {
2921             pci_create_simple(phb->bus, -1, "pci-ohci");
2922         } else {
2923             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2924         }
2925 
2926         if (spapr->has_graphics) {
2927             USBBus *usb_bus = usb_bus_find(-1);
2928 
2929             usb_create_simple(usb_bus, "usb-kbd");
2930             usb_create_simple(usb_bus, "usb-mouse");
2931         }
2932     }
2933 
2934     if (kernel_filename) {
2935         spapr->kernel_size = load_elf(kernel_filename, NULL,
2936                                       translate_kernel_address, spapr,
2937                                       NULL, NULL, NULL, NULL, 1,
2938                                       PPC_ELF_MACHINE, 0, 0);
2939         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2940             spapr->kernel_size = load_elf(kernel_filename, NULL,
2941                                           translate_kernel_address, spapr,
2942                                           NULL, NULL, NULL, NULL, 0,
2943                                           PPC_ELF_MACHINE, 0, 0);
2944             spapr->kernel_le = spapr->kernel_size > 0;
2945         }
2946         if (spapr->kernel_size < 0) {
2947             error_report("error loading %s: %s", kernel_filename,
2948                          load_elf_strerror(spapr->kernel_size));
2949             exit(1);
2950         }
2951 
2952         /* load initrd */
2953         if (initrd_filename) {
2954             /* Try to locate the initrd in the gap between the kernel
2955              * and the firmware. Add a bit of space just in case
2956              */
2957             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
2958                                   + 0x1ffff) & ~0xffff;
2959             spapr->initrd_size = load_image_targphys(initrd_filename,
2960                                                      spapr->initrd_base,
2961                                                      load_limit
2962                                                      - spapr->initrd_base);
2963             if (spapr->initrd_size < 0) {
2964                 error_report("could not load initial ram disk '%s'",
2965                              initrd_filename);
2966                 exit(1);
2967             }
2968         }
2969     }
2970 
2971     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2972     if (!filename) {
2973         error_report("Could not find LPAR firmware '%s'", bios_name);
2974         exit(1);
2975     }
2976     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2977     if (fw_size <= 0) {
2978         error_report("Could not load LPAR firmware '%s'", filename);
2979         exit(1);
2980     }
2981     g_free(filename);
2982 
2983     /* FIXME: Should register things through the MachineState's qdev
2984      * interface, this is a legacy from the sPAPREnvironment structure
2985      * which predated MachineState but had a similar function */
2986     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2987     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
2988                          &savevm_htab_handlers, spapr);
2989 
2990     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
2991 
2992     qemu_register_boot_set(spapr_boot_set, spapr);
2993 
2994     /*
2995      * Nothing needs to be done to resume a suspended guest because
2996      * suspending does not change the machine state, so no need for
2997      * a ->wakeup method.
2998      */
2999     qemu_register_wakeup_support();
3000 
3001     if (kvm_enabled()) {
3002         /* to stop and start vmclock */
3003         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3004                                          &spapr->tb);
3005 
3006         kvmppc_spapr_enable_inkernel_multitce();
3007     }
3008 
3009     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3010     if (spapr->vof) {
3011         spapr->vof->fw_size = fw_size; /* for claim() on itself */
3012         spapr_register_hypercall(KVMPPC_H_VOF_CLIENT, spapr_h_vof_client);
3013     }
3014 }
3015 
3016 #define DEFAULT_KVM_TYPE "auto"
3017 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3018 {
3019     /*
3020      * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to
3021      * accomodate the 'HV' and 'PV' formats that exists in the
3022      * wild. The 'auto' mode is being introduced already as
3023      * lower-case, thus we don't need to bother checking for
3024      * "AUTO".
3025      */
3026     if (!vm_type || !strcmp(vm_type, DEFAULT_KVM_TYPE)) {
3027         return 0;
3028     }
3029 
3030     if (!g_ascii_strcasecmp(vm_type, "hv")) {
3031         return 1;
3032     }
3033 
3034     if (!g_ascii_strcasecmp(vm_type, "pr")) {
3035         return 2;
3036     }
3037 
3038     error_report("Unknown kvm-type specified '%s'", vm_type);
3039     exit(1);
3040 }
3041 
3042 /*
3043  * Implementation of an interface to adjust firmware path
3044  * for the bootindex property handling.
3045  */
3046 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3047                                    DeviceState *dev)
3048 {
3049 #define CAST(type, obj, name) \
3050     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3051     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3052     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3053     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3054     PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3055 
3056     if (d && bus) {
3057         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3058         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3059         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3060 
3061         if (spapr) {
3062             /*
3063              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3064              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3065              * 0x8000 | (target << 8) | (bus << 5) | lun
3066              * (see the "Logical unit addressing format" table in SAM5)
3067              */
3068             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3069             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3070                                    (uint64_t)id << 48);
3071         } else if (virtio) {
3072             /*
3073              * We use SRP luns of the form 01000000 | (target << 8) | lun
3074              * in the top 32 bits of the 64-bit LUN
3075              * Note: the quote above is from SLOF and it is wrong,
3076              * the actual binding is:
3077              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3078              */
3079             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3080             if (d->lun >= 256) {
3081                 /* Use the LUN "flat space addressing method" */
3082                 id |= 0x4000;
3083             }
3084             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3085                                    (uint64_t)id << 32);
3086         } else if (usb) {
3087             /*
3088              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3089              * in the top 32 bits of the 64-bit LUN
3090              */
3091             unsigned usb_port = atoi(usb->port->path);
3092             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3093             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3094                                    (uint64_t)id << 32);
3095         }
3096     }
3097 
3098     /*
3099      * SLOF probes the USB devices, and if it recognizes that the device is a
3100      * storage device, it changes its name to "storage" instead of "usb-host",
3101      * and additionally adds a child node for the SCSI LUN, so the correct
3102      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3103      */
3104     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3105         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3106         if (usb_device_is_scsi_storage(usbdev)) {
3107             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3108         }
3109     }
3110 
3111     if (phb) {
3112         /* Replace "pci" with "pci@800000020000000" */
3113         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3114     }
3115 
3116     if (vsc) {
3117         /* Same logic as virtio above */
3118         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3119         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3120     }
3121 
3122     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3123         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3124         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3125         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3126     }
3127 
3128     if (pcidev) {
3129         return spapr_pci_fw_dev_name(pcidev);
3130     }
3131 
3132     return NULL;
3133 }
3134 
3135 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3136 {
3137     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3138 
3139     return g_strdup(spapr->kvm_type);
3140 }
3141 
3142 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3143 {
3144     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3145 
3146     g_free(spapr->kvm_type);
3147     spapr->kvm_type = g_strdup(value);
3148 }
3149 
3150 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3151 {
3152     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3153 
3154     return spapr->use_hotplug_event_source;
3155 }
3156 
3157 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3158                                             Error **errp)
3159 {
3160     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3161 
3162     spapr->use_hotplug_event_source = value;
3163 }
3164 
3165 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3166 {
3167     return true;
3168 }
3169 
3170 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3171 {
3172     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3173 
3174     switch (spapr->resize_hpt) {
3175     case SPAPR_RESIZE_HPT_DEFAULT:
3176         return g_strdup("default");
3177     case SPAPR_RESIZE_HPT_DISABLED:
3178         return g_strdup("disabled");
3179     case SPAPR_RESIZE_HPT_ENABLED:
3180         return g_strdup("enabled");
3181     case SPAPR_RESIZE_HPT_REQUIRED:
3182         return g_strdup("required");
3183     }
3184     g_assert_not_reached();
3185 }
3186 
3187 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3188 {
3189     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3190 
3191     if (strcmp(value, "default") == 0) {
3192         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3193     } else if (strcmp(value, "disabled") == 0) {
3194         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3195     } else if (strcmp(value, "enabled") == 0) {
3196         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3197     } else if (strcmp(value, "required") == 0) {
3198         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3199     } else {
3200         error_setg(errp, "Bad value for \"resize-hpt\" property");
3201     }
3202 }
3203 
3204 static bool spapr_get_vof(Object *obj, Error **errp)
3205 {
3206     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3207 
3208     return spapr->vof != NULL;
3209 }
3210 
3211 static void spapr_set_vof(Object *obj, bool value, Error **errp)
3212 {
3213     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3214 
3215     if (spapr->vof) {
3216         vof_cleanup(spapr->vof);
3217         g_free(spapr->vof);
3218         spapr->vof = NULL;
3219     }
3220     if (!value) {
3221         return;
3222     }
3223     spapr->vof = g_malloc0(sizeof(*spapr->vof));
3224 }
3225 
3226 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3227 {
3228     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3229 
3230     if (spapr->irq == &spapr_irq_xics_legacy) {
3231         return g_strdup("legacy");
3232     } else if (spapr->irq == &spapr_irq_xics) {
3233         return g_strdup("xics");
3234     } else if (spapr->irq == &spapr_irq_xive) {
3235         return g_strdup("xive");
3236     } else if (spapr->irq == &spapr_irq_dual) {
3237         return g_strdup("dual");
3238     }
3239     g_assert_not_reached();
3240 }
3241 
3242 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3243 {
3244     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3245 
3246     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3247         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3248         return;
3249     }
3250 
3251     /* The legacy IRQ backend can not be set */
3252     if (strcmp(value, "xics") == 0) {
3253         spapr->irq = &spapr_irq_xics;
3254     } else if (strcmp(value, "xive") == 0) {
3255         spapr->irq = &spapr_irq_xive;
3256     } else if (strcmp(value, "dual") == 0) {
3257         spapr->irq = &spapr_irq_dual;
3258     } else {
3259         error_setg(errp, "Bad value for \"ic-mode\" property");
3260     }
3261 }
3262 
3263 static char *spapr_get_host_model(Object *obj, Error **errp)
3264 {
3265     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3266 
3267     return g_strdup(spapr->host_model);
3268 }
3269 
3270 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3271 {
3272     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3273 
3274     g_free(spapr->host_model);
3275     spapr->host_model = g_strdup(value);
3276 }
3277 
3278 static char *spapr_get_host_serial(Object *obj, Error **errp)
3279 {
3280     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3281 
3282     return g_strdup(spapr->host_serial);
3283 }
3284 
3285 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3286 {
3287     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3288 
3289     g_free(spapr->host_serial);
3290     spapr->host_serial = g_strdup(value);
3291 }
3292 
3293 static void spapr_instance_init(Object *obj)
3294 {
3295     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3296     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3297     MachineState *ms = MACHINE(spapr);
3298     MachineClass *mc = MACHINE_GET_CLASS(ms);
3299 
3300     /*
3301      * NVDIMM support went live in 5.1 without considering that, in
3302      * other archs, the user needs to enable NVDIMM support with the
3303      * 'nvdimm' machine option and the default behavior is NVDIMM
3304      * support disabled. It is too late to roll back to the standard
3305      * behavior without breaking 5.1 guests.
3306      */
3307     if (mc->nvdimm_supported) {
3308         ms->nvdimms_state->is_enabled = true;
3309     }
3310 
3311     spapr->htab_fd = -1;
3312     spapr->use_hotplug_event_source = true;
3313     spapr->kvm_type = g_strdup(DEFAULT_KVM_TYPE);
3314     object_property_add_str(obj, "kvm-type",
3315                             spapr_get_kvm_type, spapr_set_kvm_type);
3316     object_property_set_description(obj, "kvm-type",
3317                                     "Specifies the KVM virtualization mode (auto,"
3318                                     " hv, pr). Defaults to 'auto'. This mode will use"
3319                                     " any available KVM module loaded in the host,"
3320                                     " where kvm_hv takes precedence if both kvm_hv and"
3321                                     " kvm_pr are loaded.");
3322     object_property_add_bool(obj, "modern-hotplug-events",
3323                             spapr_get_modern_hotplug_events,
3324                             spapr_set_modern_hotplug_events);
3325     object_property_set_description(obj, "modern-hotplug-events",
3326                                     "Use dedicated hotplug event mechanism in"
3327                                     " place of standard EPOW events when possible"
3328                                     " (required for memory hot-unplug support)");
3329     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3330                             "Maximum permitted CPU compatibility mode");
3331 
3332     object_property_add_str(obj, "resize-hpt",
3333                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3334     object_property_set_description(obj, "resize-hpt",
3335                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3336     object_property_add_uint32_ptr(obj, "vsmt",
3337                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3338     object_property_set_description(obj, "vsmt",
3339                                     "Virtual SMT: KVM behaves as if this were"
3340                                     " the host's SMT mode");
3341 
3342     object_property_add_bool(obj, "vfio-no-msix-emulation",
3343                              spapr_get_msix_emulation, NULL);
3344 
3345     object_property_add_uint64_ptr(obj, "kernel-addr",
3346                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3347     object_property_set_description(obj, "kernel-addr",
3348                                     stringify(KERNEL_LOAD_ADDR)
3349                                     " for -kernel is the default");
3350     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3351 
3352     object_property_add_bool(obj, "x-vof", spapr_get_vof, spapr_set_vof);
3353     object_property_set_description(obj, "x-vof",
3354                                     "Enable Virtual Open Firmware (experimental)");
3355 
3356     /* The machine class defines the default interrupt controller mode */
3357     spapr->irq = smc->irq;
3358     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3359                             spapr_set_ic_mode);
3360     object_property_set_description(obj, "ic-mode",
3361                  "Specifies the interrupt controller mode (xics, xive, dual)");
3362 
3363     object_property_add_str(obj, "host-model",
3364         spapr_get_host_model, spapr_set_host_model);
3365     object_property_set_description(obj, "host-model",
3366         "Host model to advertise in guest device tree");
3367     object_property_add_str(obj, "host-serial",
3368         spapr_get_host_serial, spapr_set_host_serial);
3369     object_property_set_description(obj, "host-serial",
3370         "Host serial number to advertise in guest device tree");
3371 }
3372 
3373 static void spapr_machine_finalizefn(Object *obj)
3374 {
3375     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3376 
3377     g_free(spapr->kvm_type);
3378 }
3379 
3380 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3381 {
3382     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3383     PowerPCCPU *cpu = POWERPC_CPU(cs);
3384     CPUPPCState *env = &cpu->env;
3385 
3386     cpu_synchronize_state(cs);
3387     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3388     if (spapr->fwnmi_system_reset_addr != -1) {
3389         uint64_t rtas_addr, addr;
3390 
3391         /* get rtas addr from fdt */
3392         rtas_addr = spapr_get_rtas_addr();
3393         if (!rtas_addr) {
3394             qemu_system_guest_panicked(NULL);
3395             return;
3396         }
3397 
3398         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3399         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3400         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3401         env->gpr[3] = addr;
3402     }
3403     ppc_cpu_do_system_reset(cs);
3404     if (spapr->fwnmi_system_reset_addr != -1) {
3405         env->nip = spapr->fwnmi_system_reset_addr;
3406     }
3407 }
3408 
3409 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3410 {
3411     CPUState *cs;
3412 
3413     CPU_FOREACH(cs) {
3414         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3415     }
3416 }
3417 
3418 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3419                           void *fdt, int *fdt_start_offset, Error **errp)
3420 {
3421     uint64_t addr;
3422     uint32_t node;
3423 
3424     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3425     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3426                                     &error_abort);
3427     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3428                                              SPAPR_MEMORY_BLOCK_SIZE);
3429     return 0;
3430 }
3431 
3432 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3433                            bool dedicated_hp_event_source)
3434 {
3435     SpaprDrc *drc;
3436     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3437     int i;
3438     uint64_t addr = addr_start;
3439     bool hotplugged = spapr_drc_hotplugged(dev);
3440 
3441     for (i = 0; i < nr_lmbs; i++) {
3442         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3443                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3444         g_assert(drc);
3445 
3446         /*
3447          * memory_device_get_free_addr() provided a range of free addresses
3448          * that doesn't overlap with any existing mapping at pre-plug. The
3449          * corresponding LMB DRCs are thus assumed to be all attachable.
3450          */
3451         spapr_drc_attach(drc, dev);
3452         if (!hotplugged) {
3453             spapr_drc_reset(drc);
3454         }
3455         addr += SPAPR_MEMORY_BLOCK_SIZE;
3456     }
3457     /* send hotplug notification to the
3458      * guest only in case of hotplugged memory
3459      */
3460     if (hotplugged) {
3461         if (dedicated_hp_event_source) {
3462             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3463                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3464             g_assert(drc);
3465             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3466                                                    nr_lmbs,
3467                                                    spapr_drc_index(drc));
3468         } else {
3469             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3470                                            nr_lmbs);
3471         }
3472     }
3473 }
3474 
3475 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3476 {
3477     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3478     PCDIMMDevice *dimm = PC_DIMM(dev);
3479     uint64_t size, addr;
3480     int64_t slot;
3481     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3482 
3483     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3484 
3485     pc_dimm_plug(dimm, MACHINE(ms));
3486 
3487     if (!is_nvdimm) {
3488         addr = object_property_get_uint(OBJECT(dimm),
3489                                         PC_DIMM_ADDR_PROP, &error_abort);
3490         spapr_add_lmbs(dev, addr, size,
3491                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT));
3492     } else {
3493         slot = object_property_get_int(OBJECT(dimm),
3494                                        PC_DIMM_SLOT_PROP, &error_abort);
3495         /* We should have valid slot number at this point */
3496         g_assert(slot >= 0);
3497         spapr_add_nvdimm(dev, slot);
3498     }
3499 }
3500 
3501 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3502                                   Error **errp)
3503 {
3504     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3505     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3506     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3507     PCDIMMDevice *dimm = PC_DIMM(dev);
3508     Error *local_err = NULL;
3509     uint64_t size;
3510     Object *memdev;
3511     hwaddr pagesize;
3512 
3513     if (!smc->dr_lmb_enabled) {
3514         error_setg(errp, "Memory hotplug not supported for this machine");
3515         return;
3516     }
3517 
3518     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3519     if (local_err) {
3520         error_propagate(errp, local_err);
3521         return;
3522     }
3523 
3524     if (is_nvdimm) {
3525         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3526             return;
3527         }
3528     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3529         error_setg(errp, "Hotplugged memory size must be a multiple of "
3530                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3531         return;
3532     }
3533 
3534     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3535                                       &error_abort);
3536     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3537     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3538         return;
3539     }
3540 
3541     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3542 }
3543 
3544 struct SpaprDimmState {
3545     PCDIMMDevice *dimm;
3546     uint32_t nr_lmbs;
3547     QTAILQ_ENTRY(SpaprDimmState) next;
3548 };
3549 
3550 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3551                                                        PCDIMMDevice *dimm)
3552 {
3553     SpaprDimmState *dimm_state = NULL;
3554 
3555     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3556         if (dimm_state->dimm == dimm) {
3557             break;
3558         }
3559     }
3560     return dimm_state;
3561 }
3562 
3563 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3564                                                       uint32_t nr_lmbs,
3565                                                       PCDIMMDevice *dimm)
3566 {
3567     SpaprDimmState *ds = NULL;
3568 
3569     /*
3570      * If this request is for a DIMM whose removal had failed earlier
3571      * (due to guest's refusal to remove the LMBs), we would have this
3572      * dimm already in the pending_dimm_unplugs list. In that
3573      * case don't add again.
3574      */
3575     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3576     if (!ds) {
3577         ds = g_malloc0(sizeof(SpaprDimmState));
3578         ds->nr_lmbs = nr_lmbs;
3579         ds->dimm = dimm;
3580         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3581     }
3582     return ds;
3583 }
3584 
3585 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3586                                               SpaprDimmState *dimm_state)
3587 {
3588     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3589     g_free(dimm_state);
3590 }
3591 
3592 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3593                                                         PCDIMMDevice *dimm)
3594 {
3595     SpaprDrc *drc;
3596     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3597                                                   &error_abort);
3598     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3599     uint32_t avail_lmbs = 0;
3600     uint64_t addr_start, addr;
3601     int i;
3602 
3603     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3604                                           &error_abort);
3605 
3606     addr = addr_start;
3607     for (i = 0; i < nr_lmbs; i++) {
3608         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3609                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3610         g_assert(drc);
3611         if (drc->dev) {
3612             avail_lmbs++;
3613         }
3614         addr += SPAPR_MEMORY_BLOCK_SIZE;
3615     }
3616 
3617     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3618 }
3619 
3620 void spapr_memory_unplug_rollback(SpaprMachineState *spapr, DeviceState *dev)
3621 {
3622     SpaprDimmState *ds;
3623     PCDIMMDevice *dimm;
3624     SpaprDrc *drc;
3625     uint32_t nr_lmbs;
3626     uint64_t size, addr_start, addr;
3627     g_autofree char *qapi_error = NULL;
3628     int i;
3629 
3630     if (!dev) {
3631         return;
3632     }
3633 
3634     dimm = PC_DIMM(dev);
3635     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3636 
3637     /*
3638      * 'ds == NULL' would mean that the DIMM doesn't have a pending
3639      * unplug state, but one of its DRC is marked as unplug_requested.
3640      * This is bad and weird enough to g_assert() out.
3641      */
3642     g_assert(ds);
3643 
3644     spapr_pending_dimm_unplugs_remove(spapr, ds);
3645 
3646     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3647     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3648 
3649     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3650                                           &error_abort);
3651 
3652     addr = addr_start;
3653     for (i = 0; i < nr_lmbs; i++) {
3654         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3655                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3656         g_assert(drc);
3657 
3658         drc->unplug_requested = false;
3659         addr += SPAPR_MEMORY_BLOCK_SIZE;
3660     }
3661 
3662     /*
3663      * Tell QAPI that something happened and the memory
3664      * hotunplug wasn't successful. Keep sending
3665      * MEM_UNPLUG_ERROR even while sending
3666      * DEVICE_UNPLUG_GUEST_ERROR until the deprecation of
3667      * MEM_UNPLUG_ERROR is due.
3668      */
3669     qapi_error = g_strdup_printf("Memory hotunplug rejected by the guest "
3670                                  "for device %s", dev->id);
3671 
3672     qapi_event_send_mem_unplug_error(dev->id ? : "", qapi_error);
3673 
3674     qapi_event_send_device_unplug_guest_error(!!dev->id, dev->id,
3675                                               dev->canonical_path);
3676 }
3677 
3678 /* Callback to be called during DRC release. */
3679 void spapr_lmb_release(DeviceState *dev)
3680 {
3681     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3682     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3683     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3684 
3685     /* This information will get lost if a migration occurs
3686      * during the unplug process. In this case recover it. */
3687     if (ds == NULL) {
3688         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3689         g_assert(ds);
3690         /* The DRC being examined by the caller at least must be counted */
3691         g_assert(ds->nr_lmbs);
3692     }
3693 
3694     if (--ds->nr_lmbs) {
3695         return;
3696     }
3697 
3698     /*
3699      * Now that all the LMBs have been removed by the guest, call the
3700      * unplug handler chain. This can never fail.
3701      */
3702     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3703     object_unparent(OBJECT(dev));
3704 }
3705 
3706 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3707 {
3708     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3709     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3710 
3711     /* We really shouldn't get this far without anything to unplug */
3712     g_assert(ds);
3713 
3714     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3715     qdev_unrealize(dev);
3716     spapr_pending_dimm_unplugs_remove(spapr, ds);
3717 }
3718 
3719 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3720                                         DeviceState *dev, Error **errp)
3721 {
3722     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3723     PCDIMMDevice *dimm = PC_DIMM(dev);
3724     uint32_t nr_lmbs;
3725     uint64_t size, addr_start, addr;
3726     int i;
3727     SpaprDrc *drc;
3728 
3729     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3730         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3731         return;
3732     }
3733 
3734     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3735     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3736 
3737     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3738                                           &error_abort);
3739 
3740     /*
3741      * An existing pending dimm state for this DIMM means that there is an
3742      * unplug operation in progress, waiting for the spapr_lmb_release
3743      * callback to complete the job (BQL can't cover that far). In this case,
3744      * bail out to avoid detaching DRCs that were already released.
3745      */
3746     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3747         error_setg(errp, "Memory unplug already in progress for device %s",
3748                    dev->id);
3749         return;
3750     }
3751 
3752     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3753 
3754     addr = addr_start;
3755     for (i = 0; i < nr_lmbs; i++) {
3756         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3757                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3758         g_assert(drc);
3759 
3760         spapr_drc_unplug_request(drc);
3761         addr += SPAPR_MEMORY_BLOCK_SIZE;
3762     }
3763 
3764     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3765                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3766     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3767                                               nr_lmbs, spapr_drc_index(drc));
3768 }
3769 
3770 /* Callback to be called during DRC release. */
3771 void spapr_core_release(DeviceState *dev)
3772 {
3773     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3774 
3775     /* Call the unplug handler chain. This can never fail. */
3776     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3777     object_unparent(OBJECT(dev));
3778 }
3779 
3780 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3781 {
3782     MachineState *ms = MACHINE(hotplug_dev);
3783     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3784     CPUCore *cc = CPU_CORE(dev);
3785     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3786 
3787     if (smc->pre_2_10_has_unused_icps) {
3788         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3789         int i;
3790 
3791         for (i = 0; i < cc->nr_threads; i++) {
3792             CPUState *cs = CPU(sc->threads[i]);
3793 
3794             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3795         }
3796     }
3797 
3798     assert(core_slot);
3799     core_slot->cpu = NULL;
3800     qdev_unrealize(dev);
3801 }
3802 
3803 static
3804 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3805                                Error **errp)
3806 {
3807     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3808     int index;
3809     SpaprDrc *drc;
3810     CPUCore *cc = CPU_CORE(dev);
3811 
3812     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3813         error_setg(errp, "Unable to find CPU core with core-id: %d",
3814                    cc->core_id);
3815         return;
3816     }
3817     if (index == 0) {
3818         error_setg(errp, "Boot CPU core may not be unplugged");
3819         return;
3820     }
3821 
3822     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3823                           spapr_vcpu_id(spapr, cc->core_id));
3824     g_assert(drc);
3825 
3826     if (!spapr_drc_unplug_requested(drc)) {
3827         spapr_drc_unplug_request(drc);
3828     }
3829 
3830     /*
3831      * spapr_hotplug_req_remove_by_index is left unguarded, out of the
3832      * "!spapr_drc_unplug_requested" check, to allow for multiple IRQ
3833      * pulses removing the same CPU. Otherwise, in an failed hotunplug
3834      * attempt (e.g. the kernel will refuse to remove the last online
3835      * CPU), we will never attempt it again because unplug_requested
3836      * will still be 'true' in that case.
3837      */
3838     spapr_hotplug_req_remove_by_index(drc);
3839 }
3840 
3841 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3842                            void *fdt, int *fdt_start_offset, Error **errp)
3843 {
3844     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3845     CPUState *cs = CPU(core->threads[0]);
3846     PowerPCCPU *cpu = POWERPC_CPU(cs);
3847     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3848     int id = spapr_get_vcpu_id(cpu);
3849     g_autofree char *nodename = NULL;
3850     int offset;
3851 
3852     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3853     offset = fdt_add_subnode(fdt, 0, nodename);
3854 
3855     spapr_dt_cpu(cs, fdt, offset, spapr);
3856 
3857     /*
3858      * spapr_dt_cpu() does not fill the 'name' property in the
3859      * CPU node. The function is called during boot process, before
3860      * and after CAS, and overwriting the 'name' property written
3861      * by SLOF is not allowed.
3862      *
3863      * Write it manually after spapr_dt_cpu(). This makes the hotplug
3864      * CPUs more compatible with the coldplugged ones, which have
3865      * the 'name' property. Linux Kernel also relies on this
3866      * property to identify CPU nodes.
3867      */
3868     _FDT((fdt_setprop_string(fdt, offset, "name", nodename)));
3869 
3870     *fdt_start_offset = offset;
3871     return 0;
3872 }
3873 
3874 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3875 {
3876     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3877     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3878     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3879     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3880     CPUCore *cc = CPU_CORE(dev);
3881     CPUState *cs;
3882     SpaprDrc *drc;
3883     CPUArchId *core_slot;
3884     int index;
3885     bool hotplugged = spapr_drc_hotplugged(dev);
3886     int i;
3887 
3888     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3889     g_assert(core_slot); /* Already checked in spapr_core_pre_plug() */
3890 
3891     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3892                           spapr_vcpu_id(spapr, cc->core_id));
3893 
3894     g_assert(drc || !mc->has_hotpluggable_cpus);
3895 
3896     if (drc) {
3897         /*
3898          * spapr_core_pre_plug() already buys us this is a brand new
3899          * core being plugged into a free slot. Nothing should already
3900          * be attached to the corresponding DRC.
3901          */
3902         spapr_drc_attach(drc, dev);
3903 
3904         if (hotplugged) {
3905             /*
3906              * Send hotplug notification interrupt to the guest only
3907              * in case of hotplugged CPUs.
3908              */
3909             spapr_hotplug_req_add_by_index(drc);
3910         } else {
3911             spapr_drc_reset(drc);
3912         }
3913     }
3914 
3915     core_slot->cpu = OBJECT(dev);
3916 
3917     /*
3918      * Set compatibility mode to match the boot CPU, which was either set
3919      * by the machine reset code or by CAS. This really shouldn't fail at
3920      * this point.
3921      */
3922     if (hotplugged) {
3923         for (i = 0; i < cc->nr_threads; i++) {
3924             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3925                            &error_abort);
3926         }
3927     }
3928 
3929     if (smc->pre_2_10_has_unused_icps) {
3930         for (i = 0; i < cc->nr_threads; i++) {
3931             cs = CPU(core->threads[i]);
3932             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3933         }
3934     }
3935 }
3936 
3937 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3938                                 Error **errp)
3939 {
3940     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3941     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3942     CPUCore *cc = CPU_CORE(dev);
3943     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3944     const char *type = object_get_typename(OBJECT(dev));
3945     CPUArchId *core_slot;
3946     int index;
3947     unsigned int smp_threads = machine->smp.threads;
3948 
3949     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3950         error_setg(errp, "CPU hotplug not supported for this machine");
3951         return;
3952     }
3953 
3954     if (strcmp(base_core_type, type)) {
3955         error_setg(errp, "CPU core type should be %s", base_core_type);
3956         return;
3957     }
3958 
3959     if (cc->core_id % smp_threads) {
3960         error_setg(errp, "invalid core id %d", cc->core_id);
3961         return;
3962     }
3963 
3964     /*
3965      * In general we should have homogeneous threads-per-core, but old
3966      * (pre hotplug support) machine types allow the last core to have
3967      * reduced threads as a compatibility hack for when we allowed
3968      * total vcpus not a multiple of threads-per-core.
3969      */
3970     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3971         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
3972                    smp_threads);
3973         return;
3974     }
3975 
3976     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3977     if (!core_slot) {
3978         error_setg(errp, "core id %d out of range", cc->core_id);
3979         return;
3980     }
3981 
3982     if (core_slot->cpu) {
3983         error_setg(errp, "core %d already populated", cc->core_id);
3984         return;
3985     }
3986 
3987     numa_cpu_pre_plug(core_slot, dev, errp);
3988 }
3989 
3990 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3991                           void *fdt, int *fdt_start_offset, Error **errp)
3992 {
3993     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3994     int intc_phandle;
3995 
3996     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3997     if (intc_phandle <= 0) {
3998         return -1;
3999     }
4000 
4001     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
4002         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
4003         return -1;
4004     }
4005 
4006     /* generally SLOF creates these, for hotplug it's up to QEMU */
4007     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
4008 
4009     return 0;
4010 }
4011 
4012 static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4013                                Error **errp)
4014 {
4015     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4016     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4017     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4018     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
4019     SpaprDrc *drc;
4020 
4021     if (dev->hotplugged && !smc->dr_phb_enabled) {
4022         error_setg(errp, "PHB hotplug not supported for this machine");
4023         return false;
4024     }
4025 
4026     if (sphb->index == (uint32_t)-1) {
4027         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
4028         return false;
4029     }
4030 
4031     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4032     if (drc && drc->dev) {
4033         error_setg(errp, "PHB %d already attached", sphb->index);
4034         return false;
4035     }
4036 
4037     /*
4038      * This will check that sphb->index doesn't exceed the maximum number of
4039      * PHBs for the current machine type.
4040      */
4041     return
4042         smc->phb_placement(spapr, sphb->index,
4043                            &sphb->buid, &sphb->io_win_addr,
4044                            &sphb->mem_win_addr, &sphb->mem64_win_addr,
4045                            windows_supported, sphb->dma_liobn,
4046                            &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
4047                            errp);
4048 }
4049 
4050 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4051 {
4052     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4053     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4054     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4055     SpaprDrc *drc;
4056     bool hotplugged = spapr_drc_hotplugged(dev);
4057 
4058     if (!smc->dr_phb_enabled) {
4059         return;
4060     }
4061 
4062     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4063     /* hotplug hooks should check it's enabled before getting this far */
4064     assert(drc);
4065 
4066     /* spapr_phb_pre_plug() already checked the DRC is attachable */
4067     spapr_drc_attach(drc, dev);
4068 
4069     if (hotplugged) {
4070         spapr_hotplug_req_add_by_index(drc);
4071     } else {
4072         spapr_drc_reset(drc);
4073     }
4074 }
4075 
4076 void spapr_phb_release(DeviceState *dev)
4077 {
4078     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4079 
4080     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4081     object_unparent(OBJECT(dev));
4082 }
4083 
4084 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4085 {
4086     qdev_unrealize(dev);
4087 }
4088 
4089 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4090                                      DeviceState *dev, Error **errp)
4091 {
4092     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4093     SpaprDrc *drc;
4094 
4095     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4096     assert(drc);
4097 
4098     if (!spapr_drc_unplug_requested(drc)) {
4099         spapr_drc_unplug_request(drc);
4100         spapr_hotplug_req_remove_by_index(drc);
4101     } else {
4102         error_setg(errp,
4103                    "PCI Host Bridge unplug already in progress for device %s",
4104                    dev->id);
4105     }
4106 }
4107 
4108 static
4109 bool spapr_tpm_proxy_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4110                               Error **errp)
4111 {
4112     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4113 
4114     if (spapr->tpm_proxy != NULL) {
4115         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4116         return false;
4117     }
4118 
4119     return true;
4120 }
4121 
4122 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4123 {
4124     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4125     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4126 
4127     /* Already checked in spapr_tpm_proxy_pre_plug() */
4128     g_assert(spapr->tpm_proxy == NULL);
4129 
4130     spapr->tpm_proxy = tpm_proxy;
4131 }
4132 
4133 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4134 {
4135     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4136 
4137     qdev_unrealize(dev);
4138     object_unparent(OBJECT(dev));
4139     spapr->tpm_proxy = NULL;
4140 }
4141 
4142 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4143                                       DeviceState *dev, Error **errp)
4144 {
4145     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4146         spapr_memory_plug(hotplug_dev, dev);
4147     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4148         spapr_core_plug(hotplug_dev, dev);
4149     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4150         spapr_phb_plug(hotplug_dev, dev);
4151     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4152         spapr_tpm_proxy_plug(hotplug_dev, dev);
4153     }
4154 }
4155 
4156 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4157                                         DeviceState *dev, Error **errp)
4158 {
4159     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4160         spapr_memory_unplug(hotplug_dev, dev);
4161     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4162         spapr_core_unplug(hotplug_dev, dev);
4163     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4164         spapr_phb_unplug(hotplug_dev, dev);
4165     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4166         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4167     }
4168 }
4169 
4170 bool spapr_memory_hot_unplug_supported(SpaprMachineState *spapr)
4171 {
4172     return spapr_ovec_test(spapr->ov5_cas, OV5_HP_EVT) ||
4173         /*
4174          * CAS will process all pending unplug requests.
4175          *
4176          * HACK: a guest could theoretically have cleared all bits in OV5,
4177          * but none of the guests we care for do.
4178          */
4179         spapr_ovec_empty(spapr->ov5_cas);
4180 }
4181 
4182 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4183                                                 DeviceState *dev, Error **errp)
4184 {
4185     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4186     MachineClass *mc = MACHINE_GET_CLASS(sms);
4187     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4188 
4189     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4190         if (spapr_memory_hot_unplug_supported(sms)) {
4191             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4192         } else {
4193             error_setg(errp, "Memory hot unplug not supported for this guest");
4194         }
4195     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4196         if (!mc->has_hotpluggable_cpus) {
4197             error_setg(errp, "CPU hot unplug not supported on this machine");
4198             return;
4199         }
4200         spapr_core_unplug_request(hotplug_dev, dev, errp);
4201     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4202         if (!smc->dr_phb_enabled) {
4203             error_setg(errp, "PHB hot unplug not supported on this machine");
4204             return;
4205         }
4206         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4207     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4208         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4209     }
4210 }
4211 
4212 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4213                                           DeviceState *dev, Error **errp)
4214 {
4215     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4216         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4217     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4218         spapr_core_pre_plug(hotplug_dev, dev, errp);
4219     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4220         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4221     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4222         spapr_tpm_proxy_pre_plug(hotplug_dev, dev, errp);
4223     }
4224 }
4225 
4226 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4227                                                  DeviceState *dev)
4228 {
4229     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4230         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4231         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4232         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4233         return HOTPLUG_HANDLER(machine);
4234     }
4235     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4236         PCIDevice *pcidev = PCI_DEVICE(dev);
4237         PCIBus *root = pci_device_root_bus(pcidev);
4238         SpaprPhbState *phb =
4239             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4240                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4241 
4242         if (phb) {
4243             return HOTPLUG_HANDLER(phb);
4244         }
4245     }
4246     return NULL;
4247 }
4248 
4249 static CpuInstanceProperties
4250 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4251 {
4252     CPUArchId *core_slot;
4253     MachineClass *mc = MACHINE_GET_CLASS(machine);
4254 
4255     /* make sure possible_cpu are intialized */
4256     mc->possible_cpu_arch_ids(machine);
4257     /* get CPU core slot containing thread that matches cpu_index */
4258     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4259     assert(core_slot);
4260     return core_slot->props;
4261 }
4262 
4263 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4264 {
4265     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4266 }
4267 
4268 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4269 {
4270     int i;
4271     unsigned int smp_threads = machine->smp.threads;
4272     unsigned int smp_cpus = machine->smp.cpus;
4273     const char *core_type;
4274     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4275     MachineClass *mc = MACHINE_GET_CLASS(machine);
4276 
4277     if (!mc->has_hotpluggable_cpus) {
4278         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4279     }
4280     if (machine->possible_cpus) {
4281         assert(machine->possible_cpus->len == spapr_max_cores);
4282         return machine->possible_cpus;
4283     }
4284 
4285     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4286     if (!core_type) {
4287         error_report("Unable to find sPAPR CPU Core definition");
4288         exit(1);
4289     }
4290 
4291     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4292                              sizeof(CPUArchId) * spapr_max_cores);
4293     machine->possible_cpus->len = spapr_max_cores;
4294     for (i = 0; i < machine->possible_cpus->len; i++) {
4295         int core_id = i * smp_threads;
4296 
4297         machine->possible_cpus->cpus[i].type = core_type;
4298         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4299         machine->possible_cpus->cpus[i].arch_id = core_id;
4300         machine->possible_cpus->cpus[i].props.has_core_id = true;
4301         machine->possible_cpus->cpus[i].props.core_id = core_id;
4302     }
4303     return machine->possible_cpus;
4304 }
4305 
4306 static bool spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4307                                 uint64_t *buid, hwaddr *pio,
4308                                 hwaddr *mmio32, hwaddr *mmio64,
4309                                 unsigned n_dma, uint32_t *liobns,
4310                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4311 {
4312     /*
4313      * New-style PHB window placement.
4314      *
4315      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4316      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4317      * windows.
4318      *
4319      * Some guest kernels can't work with MMIO windows above 1<<46
4320      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4321      *
4322      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4323      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4324      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4325      * 1TiB 64-bit MMIO windows for each PHB.
4326      */
4327     const uint64_t base_buid = 0x800000020000000ULL;
4328     int i;
4329 
4330     /* Sanity check natural alignments */
4331     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4332     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4333     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4334     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4335     /* Sanity check bounds */
4336     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4337                       SPAPR_PCI_MEM32_WIN_SIZE);
4338     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4339                       SPAPR_PCI_MEM64_WIN_SIZE);
4340 
4341     if (index >= SPAPR_MAX_PHBS) {
4342         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4343                    SPAPR_MAX_PHBS - 1);
4344         return false;
4345     }
4346 
4347     *buid = base_buid + index;
4348     for (i = 0; i < n_dma; ++i) {
4349         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4350     }
4351 
4352     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4353     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4354     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4355 
4356     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4357     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4358     return true;
4359 }
4360 
4361 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4362 {
4363     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4364 
4365     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4366 }
4367 
4368 static void spapr_ics_resend(XICSFabric *dev)
4369 {
4370     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4371 
4372     ics_resend(spapr->ics);
4373 }
4374 
4375 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4376 {
4377     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4378 
4379     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4380 }
4381 
4382 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4383                                  Monitor *mon)
4384 {
4385     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4386 
4387     spapr_irq_print_info(spapr, mon);
4388     monitor_printf(mon, "irqchip: %s\n",
4389                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4390 }
4391 
4392 /*
4393  * This is a XIVE only operation
4394  */
4395 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4396                            uint8_t nvt_blk, uint32_t nvt_idx,
4397                            bool cam_ignore, uint8_t priority,
4398                            uint32_t logic_serv, XiveTCTXMatch *match)
4399 {
4400     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4401     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4402     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4403     int count;
4404 
4405     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4406                            priority, logic_serv, match);
4407     if (count < 0) {
4408         return count;
4409     }
4410 
4411     /*
4412      * When we implement the save and restore of the thread interrupt
4413      * contexts in the enter/exit CPU handlers of the machine and the
4414      * escalations in QEMU, we should be able to handle non dispatched
4415      * vCPUs.
4416      *
4417      * Until this is done, the sPAPR machine should find at least one
4418      * matching context always.
4419      */
4420     if (count == 0) {
4421         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4422                       nvt_blk, nvt_idx);
4423     }
4424 
4425     return count;
4426 }
4427 
4428 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4429 {
4430     return cpu->vcpu_id;
4431 }
4432 
4433 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4434 {
4435     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4436     MachineState *ms = MACHINE(spapr);
4437     int vcpu_id;
4438 
4439     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4440 
4441     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4442         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4443         error_append_hint(errp, "Adjust the number of cpus to %d "
4444                           "or try to raise the number of threads per core\n",
4445                           vcpu_id * ms->smp.threads / spapr->vsmt);
4446         return false;
4447     }
4448 
4449     cpu->vcpu_id = vcpu_id;
4450     return true;
4451 }
4452 
4453 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4454 {
4455     CPUState *cs;
4456 
4457     CPU_FOREACH(cs) {
4458         PowerPCCPU *cpu = POWERPC_CPU(cs);
4459 
4460         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4461             return cpu;
4462         }
4463     }
4464 
4465     return NULL;
4466 }
4467 
4468 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4469 {
4470     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4471 
4472     /* These are only called by TCG, KVM maintains dispatch state */
4473 
4474     spapr_cpu->prod = false;
4475     if (spapr_cpu->vpa_addr) {
4476         CPUState *cs = CPU(cpu);
4477         uint32_t dispatch;
4478 
4479         dispatch = ldl_be_phys(cs->as,
4480                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4481         dispatch++;
4482         if ((dispatch & 1) != 0) {
4483             qemu_log_mask(LOG_GUEST_ERROR,
4484                           "VPA: incorrect dispatch counter value for "
4485                           "dispatched partition %u, correcting.\n", dispatch);
4486             dispatch++;
4487         }
4488         stl_be_phys(cs->as,
4489                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4490     }
4491 }
4492 
4493 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4494 {
4495     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4496 
4497     if (spapr_cpu->vpa_addr) {
4498         CPUState *cs = CPU(cpu);
4499         uint32_t dispatch;
4500 
4501         dispatch = ldl_be_phys(cs->as,
4502                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4503         dispatch++;
4504         if ((dispatch & 1) != 1) {
4505             qemu_log_mask(LOG_GUEST_ERROR,
4506                           "VPA: incorrect dispatch counter value for "
4507                           "preempted partition %u, correcting.\n", dispatch);
4508             dispatch++;
4509         }
4510         stl_be_phys(cs->as,
4511                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4512     }
4513 }
4514 
4515 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4516 {
4517     MachineClass *mc = MACHINE_CLASS(oc);
4518     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4519     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4520     NMIClass *nc = NMI_CLASS(oc);
4521     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4522     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4523     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4524     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4525     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4526     VofMachineIfClass *vmc = VOF_MACHINE_CLASS(oc);
4527 
4528     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4529     mc->ignore_boot_device_suffixes = true;
4530 
4531     /*
4532      * We set up the default / latest behaviour here.  The class_init
4533      * functions for the specific versioned machine types can override
4534      * these details for backwards compatibility
4535      */
4536     mc->init = spapr_machine_init;
4537     mc->reset = spapr_machine_reset;
4538     mc->block_default_type = IF_SCSI;
4539 
4540     /*
4541      * Setting max_cpus to INT32_MAX. Both KVM and TCG max_cpus values
4542      * should be limited by the host capability instead of hardcoded.
4543      * max_cpus for KVM guests will be checked in kvm_init(), and TCG
4544      * guests are welcome to have as many CPUs as the host are capable
4545      * of emulate.
4546      */
4547     mc->max_cpus = INT32_MAX;
4548 
4549     mc->no_parallel = 1;
4550     mc->default_boot_order = "";
4551     mc->default_ram_size = 512 * MiB;
4552     mc->default_ram_id = "ppc_spapr.ram";
4553     mc->default_display = "std";
4554     mc->kvm_type = spapr_kvm_type;
4555     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4556     mc->pci_allow_0_address = true;
4557     assert(!mc->get_hotplug_handler);
4558     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4559     hc->pre_plug = spapr_machine_device_pre_plug;
4560     hc->plug = spapr_machine_device_plug;
4561     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4562     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4563     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4564     hc->unplug_request = spapr_machine_device_unplug_request;
4565     hc->unplug = spapr_machine_device_unplug;
4566 
4567     smc->dr_lmb_enabled = true;
4568     smc->update_dt_enabled = true;
4569     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4570     mc->has_hotpluggable_cpus = true;
4571     mc->nvdimm_supported = true;
4572     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4573     fwc->get_dev_path = spapr_get_fw_dev_path;
4574     nc->nmi_monitor_handler = spapr_nmi;
4575     smc->phb_placement = spapr_phb_placement;
4576     vhc->hypercall = emulate_spapr_hypercall;
4577     vhc->hpt_mask = spapr_hpt_mask;
4578     vhc->map_hptes = spapr_map_hptes;
4579     vhc->unmap_hptes = spapr_unmap_hptes;
4580     vhc->hpte_set_c = spapr_hpte_set_c;
4581     vhc->hpte_set_r = spapr_hpte_set_r;
4582     vhc->get_pate = spapr_get_pate;
4583     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4584     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4585     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4586     xic->ics_get = spapr_ics_get;
4587     xic->ics_resend = spapr_ics_resend;
4588     xic->icp_get = spapr_icp_get;
4589     ispc->print_info = spapr_pic_print_info;
4590     /* Force NUMA node memory size to be a multiple of
4591      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4592      * in which LMBs are represented and hot-added
4593      */
4594     mc->numa_mem_align_shift = 28;
4595     mc->auto_enable_numa = true;
4596 
4597     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4598     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4599     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4600     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4601     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4602     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4603     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4604     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4605     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4606     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4607     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4608     smc->default_caps.caps[SPAPR_CAP_RPT_INVALIDATE] = SPAPR_CAP_OFF;
4609     spapr_caps_add_properties(smc);
4610     smc->irq = &spapr_irq_dual;
4611     smc->dr_phb_enabled = true;
4612     smc->linux_pci_probe = true;
4613     smc->smp_threads_vsmt = true;
4614     smc->nr_xirqs = SPAPR_NR_XIRQS;
4615     xfc->match_nvt = spapr_match_nvt;
4616     vmc->client_architecture_support = spapr_vof_client_architecture_support;
4617     vmc->quiesce = spapr_vof_quiesce;
4618     vmc->setprop = spapr_vof_setprop;
4619 }
4620 
4621 static const TypeInfo spapr_machine_info = {
4622     .name          = TYPE_SPAPR_MACHINE,
4623     .parent        = TYPE_MACHINE,
4624     .abstract      = true,
4625     .instance_size = sizeof(SpaprMachineState),
4626     .instance_init = spapr_instance_init,
4627     .instance_finalize = spapr_machine_finalizefn,
4628     .class_size    = sizeof(SpaprMachineClass),
4629     .class_init    = spapr_machine_class_init,
4630     .interfaces = (InterfaceInfo[]) {
4631         { TYPE_FW_PATH_PROVIDER },
4632         { TYPE_NMI },
4633         { TYPE_HOTPLUG_HANDLER },
4634         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4635         { TYPE_XICS_FABRIC },
4636         { TYPE_INTERRUPT_STATS_PROVIDER },
4637         { TYPE_XIVE_FABRIC },
4638         { TYPE_VOF_MACHINE_IF },
4639         { }
4640     },
4641 };
4642 
4643 static void spapr_machine_latest_class_options(MachineClass *mc)
4644 {
4645     mc->alias = "pseries";
4646     mc->is_default = true;
4647 }
4648 
4649 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4650     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4651                                                     void *data)      \
4652     {                                                                \
4653         MachineClass *mc = MACHINE_CLASS(oc);                        \
4654         spapr_machine_##suffix##_class_options(mc);                  \
4655         if (latest) {                                                \
4656             spapr_machine_latest_class_options(mc);                  \
4657         }                                                            \
4658     }                                                                \
4659     static const TypeInfo spapr_machine_##suffix##_info = {          \
4660         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4661         .parent = TYPE_SPAPR_MACHINE,                                \
4662         .class_init = spapr_machine_##suffix##_class_init,           \
4663     };                                                               \
4664     static void spapr_machine_register_##suffix(void)                \
4665     {                                                                \
4666         type_register(&spapr_machine_##suffix##_info);               \
4667     }                                                                \
4668     type_init(spapr_machine_register_##suffix)
4669 
4670 /*
4671  * pseries-7.0
4672  */
4673 static void spapr_machine_7_0_class_options(MachineClass *mc)
4674 {
4675     /* Defaults for the latest behaviour inherited from the base class */
4676 }
4677 
4678 DEFINE_SPAPR_MACHINE(7_0, "7.0", true);
4679 
4680 /*
4681  * pseries-6.2
4682  */
4683 static void spapr_machine_6_2_class_options(MachineClass *mc)
4684 {
4685     spapr_machine_7_0_class_options(mc);
4686     compat_props_add(mc->compat_props, hw_compat_6_2, hw_compat_6_2_len);
4687 }
4688 
4689 DEFINE_SPAPR_MACHINE(6_2, "6.2", false);
4690 
4691 /*
4692  * pseries-6.1
4693  */
4694 static void spapr_machine_6_1_class_options(MachineClass *mc)
4695 {
4696     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4697 
4698     spapr_machine_6_2_class_options(mc);
4699     compat_props_add(mc->compat_props, hw_compat_6_1, hw_compat_6_1_len);
4700     smc->pre_6_2_numa_affinity = true;
4701     mc->smp_props.prefer_sockets = true;
4702 }
4703 
4704 DEFINE_SPAPR_MACHINE(6_1, "6.1", false);
4705 
4706 /*
4707  * pseries-6.0
4708  */
4709 static void spapr_machine_6_0_class_options(MachineClass *mc)
4710 {
4711     spapr_machine_6_1_class_options(mc);
4712     compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len);
4713 }
4714 
4715 DEFINE_SPAPR_MACHINE(6_0, "6.0", false);
4716 
4717 /*
4718  * pseries-5.2
4719  */
4720 static void spapr_machine_5_2_class_options(MachineClass *mc)
4721 {
4722     spapr_machine_6_0_class_options(mc);
4723     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
4724 }
4725 
4726 DEFINE_SPAPR_MACHINE(5_2, "5.2", false);
4727 
4728 /*
4729  * pseries-5.1
4730  */
4731 static void spapr_machine_5_1_class_options(MachineClass *mc)
4732 {
4733     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4734 
4735     spapr_machine_5_2_class_options(mc);
4736     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4737     smc->pre_5_2_numa_associativity = true;
4738 }
4739 
4740 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4741 
4742 /*
4743  * pseries-5.0
4744  */
4745 static void spapr_machine_5_0_class_options(MachineClass *mc)
4746 {
4747     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4748     static GlobalProperty compat[] = {
4749         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4750     };
4751 
4752     spapr_machine_5_1_class_options(mc);
4753     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4754     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4755     mc->numa_mem_supported = true;
4756     smc->pre_5_1_assoc_refpoints = true;
4757 }
4758 
4759 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4760 
4761 /*
4762  * pseries-4.2
4763  */
4764 static void spapr_machine_4_2_class_options(MachineClass *mc)
4765 {
4766     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4767 
4768     spapr_machine_5_0_class_options(mc);
4769     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4770     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4771     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4772     smc->rma_limit = 16 * GiB;
4773     mc->nvdimm_supported = false;
4774 }
4775 
4776 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4777 
4778 /*
4779  * pseries-4.1
4780  */
4781 static void spapr_machine_4_1_class_options(MachineClass *mc)
4782 {
4783     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4784     static GlobalProperty compat[] = {
4785         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4786         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4787     };
4788 
4789     spapr_machine_4_2_class_options(mc);
4790     smc->linux_pci_probe = false;
4791     smc->smp_threads_vsmt = false;
4792     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4793     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4794 }
4795 
4796 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4797 
4798 /*
4799  * pseries-4.0
4800  */
4801 static bool phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4802                               uint64_t *buid, hwaddr *pio,
4803                               hwaddr *mmio32, hwaddr *mmio64,
4804                               unsigned n_dma, uint32_t *liobns,
4805                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4806 {
4807     if (!spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma,
4808                              liobns, nv2gpa, nv2atsd, errp)) {
4809         return false;
4810     }
4811 
4812     *nv2gpa = 0;
4813     *nv2atsd = 0;
4814     return true;
4815 }
4816 static void spapr_machine_4_0_class_options(MachineClass *mc)
4817 {
4818     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4819 
4820     spapr_machine_4_1_class_options(mc);
4821     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4822     smc->phb_placement = phb_placement_4_0;
4823     smc->irq = &spapr_irq_xics;
4824     smc->pre_4_1_migration = true;
4825 }
4826 
4827 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4828 
4829 /*
4830  * pseries-3.1
4831  */
4832 static void spapr_machine_3_1_class_options(MachineClass *mc)
4833 {
4834     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4835 
4836     spapr_machine_4_0_class_options(mc);
4837     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4838 
4839     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4840     smc->update_dt_enabled = false;
4841     smc->dr_phb_enabled = false;
4842     smc->broken_host_serial_model = true;
4843     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4844     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4845     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4846     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4847 }
4848 
4849 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4850 
4851 /*
4852  * pseries-3.0
4853  */
4854 
4855 static void spapr_machine_3_0_class_options(MachineClass *mc)
4856 {
4857     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4858 
4859     spapr_machine_3_1_class_options(mc);
4860     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4861 
4862     smc->legacy_irq_allocation = true;
4863     smc->nr_xirqs = 0x400;
4864     smc->irq = &spapr_irq_xics_legacy;
4865 }
4866 
4867 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4868 
4869 /*
4870  * pseries-2.12
4871  */
4872 static void spapr_machine_2_12_class_options(MachineClass *mc)
4873 {
4874     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4875     static GlobalProperty compat[] = {
4876         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4877         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4878     };
4879 
4880     spapr_machine_3_0_class_options(mc);
4881     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4882     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4883 
4884     /* We depend on kvm_enabled() to choose a default value for the
4885      * hpt-max-page-size capability. Of course we can't do it here
4886      * because this is too early and the HW accelerator isn't initialzed
4887      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4888      */
4889     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4890 }
4891 
4892 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4893 
4894 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4895 {
4896     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4897 
4898     spapr_machine_2_12_class_options(mc);
4899     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4900     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4901     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4902 }
4903 
4904 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4905 
4906 /*
4907  * pseries-2.11
4908  */
4909 
4910 static void spapr_machine_2_11_class_options(MachineClass *mc)
4911 {
4912     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4913 
4914     spapr_machine_2_12_class_options(mc);
4915     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4916     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4917 }
4918 
4919 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4920 
4921 /*
4922  * pseries-2.10
4923  */
4924 
4925 static void spapr_machine_2_10_class_options(MachineClass *mc)
4926 {
4927     spapr_machine_2_11_class_options(mc);
4928     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4929 }
4930 
4931 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4932 
4933 /*
4934  * pseries-2.9
4935  */
4936 
4937 static void spapr_machine_2_9_class_options(MachineClass *mc)
4938 {
4939     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4940     static GlobalProperty compat[] = {
4941         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4942     };
4943 
4944     spapr_machine_2_10_class_options(mc);
4945     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4946     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4947     smc->pre_2_10_has_unused_icps = true;
4948     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4949 }
4950 
4951 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4952 
4953 /*
4954  * pseries-2.8
4955  */
4956 
4957 static void spapr_machine_2_8_class_options(MachineClass *mc)
4958 {
4959     static GlobalProperty compat[] = {
4960         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4961     };
4962 
4963     spapr_machine_2_9_class_options(mc);
4964     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4965     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4966     mc->numa_mem_align_shift = 23;
4967 }
4968 
4969 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4970 
4971 /*
4972  * pseries-2.7
4973  */
4974 
4975 static bool phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4976                               uint64_t *buid, hwaddr *pio,
4977                               hwaddr *mmio32, hwaddr *mmio64,
4978                               unsigned n_dma, uint32_t *liobns,
4979                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4980 {
4981     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4982     const uint64_t base_buid = 0x800000020000000ULL;
4983     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4984     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4985     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4986     const uint32_t max_index = 255;
4987     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4988 
4989     uint64_t ram_top = MACHINE(spapr)->ram_size;
4990     hwaddr phb0_base, phb_base;
4991     int i;
4992 
4993     /* Do we have device memory? */
4994     if (MACHINE(spapr)->maxram_size > ram_top) {
4995         /* Can't just use maxram_size, because there may be an
4996          * alignment gap between normal and device memory regions
4997          */
4998         ram_top = MACHINE(spapr)->device_memory->base +
4999             memory_region_size(&MACHINE(spapr)->device_memory->mr);
5000     }
5001 
5002     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
5003 
5004     if (index > max_index) {
5005         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
5006                    max_index);
5007         return false;
5008     }
5009 
5010     *buid = base_buid + index;
5011     for (i = 0; i < n_dma; ++i) {
5012         liobns[i] = SPAPR_PCI_LIOBN(index, i);
5013     }
5014 
5015     phb_base = phb0_base + index * phb_spacing;
5016     *pio = phb_base + pio_offset;
5017     *mmio32 = phb_base + mmio_offset;
5018     /*
5019      * We don't set the 64-bit MMIO window, relying on the PHB's
5020      * fallback behaviour of automatically splitting a large "32-bit"
5021      * window into contiguous 32-bit and 64-bit windows
5022      */
5023 
5024     *nv2gpa = 0;
5025     *nv2atsd = 0;
5026     return true;
5027 }
5028 
5029 static void spapr_machine_2_7_class_options(MachineClass *mc)
5030 {
5031     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5032     static GlobalProperty compat[] = {
5033         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
5034         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
5035         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
5036         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
5037     };
5038 
5039     spapr_machine_2_8_class_options(mc);
5040     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
5041     mc->default_machine_opts = "modern-hotplug-events=off";
5042     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
5043     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5044     smc->phb_placement = phb_placement_2_7;
5045 }
5046 
5047 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
5048 
5049 /*
5050  * pseries-2.6
5051  */
5052 
5053 static void spapr_machine_2_6_class_options(MachineClass *mc)
5054 {
5055     static GlobalProperty compat[] = {
5056         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
5057     };
5058 
5059     spapr_machine_2_7_class_options(mc);
5060     mc->has_hotpluggable_cpus = false;
5061     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
5062     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5063 }
5064 
5065 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
5066 
5067 /*
5068  * pseries-2.5
5069  */
5070 
5071 static void spapr_machine_2_5_class_options(MachineClass *mc)
5072 {
5073     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5074     static GlobalProperty compat[] = {
5075         { "spapr-vlan", "use-rx-buffer-pools", "off" },
5076     };
5077 
5078     spapr_machine_2_6_class_options(mc);
5079     smc->use_ohci_by_default = true;
5080     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
5081     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5082 }
5083 
5084 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
5085 
5086 /*
5087  * pseries-2.4
5088  */
5089 
5090 static void spapr_machine_2_4_class_options(MachineClass *mc)
5091 {
5092     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5093 
5094     spapr_machine_2_5_class_options(mc);
5095     smc->dr_lmb_enabled = false;
5096     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
5097 }
5098 
5099 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
5100 
5101 /*
5102  * pseries-2.3
5103  */
5104 
5105 static void spapr_machine_2_3_class_options(MachineClass *mc)
5106 {
5107     static GlobalProperty compat[] = {
5108         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
5109     };
5110     spapr_machine_2_4_class_options(mc);
5111     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
5112     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5113 }
5114 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
5115 
5116 /*
5117  * pseries-2.2
5118  */
5119 
5120 static void spapr_machine_2_2_class_options(MachineClass *mc)
5121 {
5122     static GlobalProperty compat[] = {
5123         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
5124     };
5125 
5126     spapr_machine_2_3_class_options(mc);
5127     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
5128     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5129     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
5130 }
5131 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
5132 
5133 /*
5134  * pseries-2.1
5135  */
5136 
5137 static void spapr_machine_2_1_class_options(MachineClass *mc)
5138 {
5139     spapr_machine_2_2_class_options(mc);
5140     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
5141 }
5142 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
5143 
5144 static void spapr_machine_register_types(void)
5145 {
5146     type_register_static(&spapr_machine_info);
5147 }
5148 
5149 type_init(spapr_machine_register_types)
5150