xref: /openbmc/qemu/hw/ppc/spapr.c (revision b8012ecf)
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     g_assert(!vhyp_cpu_in_nested(cpu));
1274 
1275     if (msr_pr) {
1276         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1277         env->gpr[3] = H_PRIVILEGE;
1278     } else {
1279         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1280     }
1281 }
1282 
1283 struct LPCRSyncState {
1284     target_ulong value;
1285     target_ulong mask;
1286 };
1287 
1288 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1289 {
1290     struct LPCRSyncState *s = arg.host_ptr;
1291     PowerPCCPU *cpu = POWERPC_CPU(cs);
1292     CPUPPCState *env = &cpu->env;
1293     target_ulong lpcr;
1294 
1295     cpu_synchronize_state(cs);
1296     lpcr = env->spr[SPR_LPCR];
1297     lpcr &= ~s->mask;
1298     lpcr |= s->value;
1299     ppc_store_lpcr(cpu, lpcr);
1300 }
1301 
1302 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1303 {
1304     CPUState *cs;
1305     struct LPCRSyncState s = {
1306         .value = value,
1307         .mask = mask
1308     };
1309     CPU_FOREACH(cs) {
1310         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1311     }
1312 }
1313 
1314 static bool spapr_get_pate(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu,
1315                            target_ulong lpid, ppc_v3_pate_t *entry)
1316 {
1317     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1318     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
1319 
1320     if (!spapr_cpu->in_nested) {
1321         assert(lpid == 0);
1322 
1323         /* Copy PATE1:GR into PATE0:HR */
1324         entry->dw0 = spapr->patb_entry & PATE0_HR;
1325         entry->dw1 = spapr->patb_entry;
1326 
1327     } else {
1328         uint64_t patb, pats;
1329 
1330         assert(lpid != 0);
1331 
1332         patb = spapr->nested_ptcr & PTCR_PATB;
1333         pats = spapr->nested_ptcr & PTCR_PATS;
1334 
1335         /* Calculate number of entries */
1336         pats = 1ull << (pats + 12 - 4);
1337         if (pats <= lpid) {
1338             return false;
1339         }
1340 
1341         /* Grab entry */
1342         patb += 16 * lpid;
1343         entry->dw0 = ldq_phys(CPU(cpu)->as, patb);
1344         entry->dw1 = ldq_phys(CPU(cpu)->as, patb + 8);
1345     }
1346 
1347     return true;
1348 }
1349 
1350 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1351 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1352 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1353 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1354 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1355 
1356 /*
1357  * Get the fd to access the kernel htab, re-opening it if necessary
1358  */
1359 static int get_htab_fd(SpaprMachineState *spapr)
1360 {
1361     Error *local_err = NULL;
1362 
1363     if (spapr->htab_fd >= 0) {
1364         return spapr->htab_fd;
1365     }
1366 
1367     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1368     if (spapr->htab_fd < 0) {
1369         error_report_err(local_err);
1370     }
1371 
1372     return spapr->htab_fd;
1373 }
1374 
1375 void close_htab_fd(SpaprMachineState *spapr)
1376 {
1377     if (spapr->htab_fd >= 0) {
1378         close(spapr->htab_fd);
1379     }
1380     spapr->htab_fd = -1;
1381 }
1382 
1383 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1384 {
1385     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1386 
1387     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1388 }
1389 
1390 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1391 {
1392     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1393 
1394     assert(kvm_enabled());
1395 
1396     if (!spapr->htab) {
1397         return 0;
1398     }
1399 
1400     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1401 }
1402 
1403 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1404                                                 hwaddr ptex, int n)
1405 {
1406     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1407     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1408 
1409     if (!spapr->htab) {
1410         /*
1411          * HTAB is controlled by KVM. Fetch into temporary buffer
1412          */
1413         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1414         kvmppc_read_hptes(hptes, ptex, n);
1415         return hptes;
1416     }
1417 
1418     /*
1419      * HTAB is controlled by QEMU. Just point to the internally
1420      * accessible PTEG.
1421      */
1422     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1423 }
1424 
1425 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1426                               const ppc_hash_pte64_t *hptes,
1427                               hwaddr ptex, int n)
1428 {
1429     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1430 
1431     if (!spapr->htab) {
1432         g_free((void *)hptes);
1433     }
1434 
1435     /* Nothing to do for qemu managed HPT */
1436 }
1437 
1438 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1439                       uint64_t pte0, uint64_t pte1)
1440 {
1441     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1442     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1443 
1444     if (!spapr->htab) {
1445         kvmppc_write_hpte(ptex, pte0, pte1);
1446     } else {
1447         if (pte0 & HPTE64_V_VALID) {
1448             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1449             /*
1450              * When setting valid, we write PTE1 first. This ensures
1451              * proper synchronization with the reading code in
1452              * ppc_hash64_pteg_search()
1453              */
1454             smp_wmb();
1455             stq_p(spapr->htab + offset, pte0);
1456         } else {
1457             stq_p(spapr->htab + offset, pte0);
1458             /*
1459              * When clearing it we set PTE0 first. This ensures proper
1460              * synchronization with the reading code in
1461              * ppc_hash64_pteg_search()
1462              */
1463             smp_wmb();
1464             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1465         }
1466     }
1467 }
1468 
1469 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1470                              uint64_t pte1)
1471 {
1472     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_C;
1473     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1474 
1475     if (!spapr->htab) {
1476         /* There should always be a hash table when this is called */
1477         error_report("spapr_hpte_set_c called with no hash table !");
1478         return;
1479     }
1480 
1481     /* The HW performs a non-atomic byte update */
1482     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1483 }
1484 
1485 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1486                              uint64_t pte1)
1487 {
1488     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_R;
1489     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1490 
1491     if (!spapr->htab) {
1492         /* There should always be a hash table when this is called */
1493         error_report("spapr_hpte_set_r called with no hash table !");
1494         return;
1495     }
1496 
1497     /* The HW performs a non-atomic byte update */
1498     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1499 }
1500 
1501 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1502 {
1503     int shift;
1504 
1505     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1506      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1507      * that's much more than is needed for Linux guests */
1508     shift = ctz64(pow2ceil(ramsize)) - 7;
1509     shift = MAX(shift, 18); /* Minimum architected size */
1510     shift = MIN(shift, 46); /* Maximum architected size */
1511     return shift;
1512 }
1513 
1514 void spapr_free_hpt(SpaprMachineState *spapr)
1515 {
1516     g_free(spapr->htab);
1517     spapr->htab = NULL;
1518     spapr->htab_shift = 0;
1519     close_htab_fd(spapr);
1520 }
1521 
1522 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1523 {
1524     ERRP_GUARD();
1525     long rc;
1526 
1527     /* Clean up any HPT info from a previous boot */
1528     spapr_free_hpt(spapr);
1529 
1530     rc = kvmppc_reset_htab(shift);
1531 
1532     if (rc == -EOPNOTSUPP) {
1533         error_setg(errp, "HPT not supported in nested guests");
1534         return -EOPNOTSUPP;
1535     }
1536 
1537     if (rc < 0) {
1538         /* kernel-side HPT needed, but couldn't allocate one */
1539         error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1540                          shift);
1541         error_append_hint(errp, "Try smaller maxmem?\n");
1542         return -errno;
1543     } else if (rc > 0) {
1544         /* kernel-side HPT allocated */
1545         if (rc != shift) {
1546             error_setg(errp,
1547                        "Requested order %d HPT, but kernel allocated order %ld",
1548                        shift, rc);
1549             error_append_hint(errp, "Try smaller maxmem?\n");
1550             return -ENOSPC;
1551         }
1552 
1553         spapr->htab_shift = shift;
1554         spapr->htab = NULL;
1555     } else {
1556         /* kernel-side HPT not needed, allocate in userspace instead */
1557         size_t size = 1ULL << shift;
1558         int i;
1559 
1560         spapr->htab = qemu_memalign(size, size);
1561         memset(spapr->htab, 0, size);
1562         spapr->htab_shift = shift;
1563 
1564         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1565             DIRTY_HPTE(HPTE(spapr->htab, i));
1566         }
1567     }
1568     /* We're setting up a hash table, so that means we're not radix */
1569     spapr->patb_entry = 0;
1570     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1571     return 0;
1572 }
1573 
1574 void spapr_setup_hpt(SpaprMachineState *spapr)
1575 {
1576     int hpt_shift;
1577 
1578     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1579         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1580     } else {
1581         uint64_t current_ram_size;
1582 
1583         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1584         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1585     }
1586     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1587 
1588     if (kvm_enabled()) {
1589         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1590 
1591         /* Check our RMA fits in the possible VRMA */
1592         if (vrma_limit < spapr->rma_size) {
1593             error_report("Unable to create %" HWADDR_PRIu
1594                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1595                          spapr->rma_size / MiB, vrma_limit / MiB);
1596             exit(EXIT_FAILURE);
1597         }
1598     }
1599 }
1600 
1601 void spapr_check_mmu_mode(bool guest_radix)
1602 {
1603     if (guest_radix) {
1604         if (kvm_enabled() && !kvmppc_has_cap_mmu_radix()) {
1605             error_report("Guest requested unavailable MMU mode (radix).");
1606             exit(EXIT_FAILURE);
1607         }
1608     } else {
1609         if (kvm_enabled() && kvmppc_has_cap_mmu_radix()
1610             && !kvmppc_has_cap_mmu_hash_v3()) {
1611             error_report("Guest requested unavailable MMU mode (hash).");
1612             exit(EXIT_FAILURE);
1613         }
1614     }
1615 }
1616 
1617 static void spapr_machine_reset(MachineState *machine)
1618 {
1619     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1620     PowerPCCPU *first_ppc_cpu;
1621     hwaddr fdt_addr;
1622     void *fdt;
1623     int rc;
1624 
1625     pef_kvm_reset(machine->cgs, &error_fatal);
1626     spapr_caps_apply(spapr);
1627 
1628     first_ppc_cpu = POWERPC_CPU(first_cpu);
1629     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1630         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1631                               spapr->max_compat_pvr)) {
1632         /*
1633          * If using KVM with radix mode available, VCPUs can be started
1634          * without a HPT because KVM will start them in radix mode.
1635          * Set the GR bit in PATE so that we know there is no HPT.
1636          */
1637         spapr->patb_entry = PATE1_GR;
1638         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1639     } else {
1640         spapr_setup_hpt(spapr);
1641     }
1642 
1643     qemu_devices_reset();
1644 
1645     spapr_ovec_cleanup(spapr->ov5_cas);
1646     spapr->ov5_cas = spapr_ovec_new();
1647 
1648     ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1649 
1650     /*
1651      * This is fixing some of the default configuration of the XIVE
1652      * devices. To be called after the reset of the machine devices.
1653      */
1654     spapr_irq_reset(spapr, &error_fatal);
1655 
1656     /*
1657      * There is no CAS under qtest. Simulate one to please the code that
1658      * depends on spapr->ov5_cas. This is especially needed to test device
1659      * unplug, so we do that before resetting the DRCs.
1660      */
1661     if (qtest_enabled()) {
1662         spapr_ovec_cleanup(spapr->ov5_cas);
1663         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1664     }
1665 
1666     spapr_nvdimm_finish_flushes();
1667 
1668     /* DRC reset may cause a device to be unplugged. This will cause troubles
1669      * if this device is used by another device (eg, a running vhost backend
1670      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1671      * situations, we reset DRCs after all devices have been reset.
1672      */
1673     spapr_drc_reset_all(spapr);
1674 
1675     spapr_clear_pending_events(spapr);
1676 
1677     /*
1678      * We place the device tree just below either the top of the RMA,
1679      * or just below 2GB, whichever is lower, so that it can be
1680      * processed with 32-bit real mode code if necessary
1681      */
1682     fdt_addr = MIN(spapr->rma_size, FDT_MAX_ADDR) - FDT_MAX_SIZE;
1683 
1684     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1685     if (spapr->vof) {
1686         spapr_vof_reset(spapr, fdt, &error_fatal);
1687         /*
1688          * Do not pack the FDT as the client may change properties.
1689          * VOF client does not expect the FDT so we do not load it to the VM.
1690          */
1691     } else {
1692         rc = fdt_pack(fdt);
1693         /* Should only fail if we've built a corrupted tree */
1694         assert(rc == 0);
1695 
1696         spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT,
1697                                   0, fdt_addr, 0);
1698         cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1699     }
1700     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1701 
1702     g_free(spapr->fdt_blob);
1703     spapr->fdt_size = fdt_totalsize(fdt);
1704     spapr->fdt_initial_size = spapr->fdt_size;
1705     spapr->fdt_blob = fdt;
1706 
1707     /* Set up the entry state */
1708     first_ppc_cpu->env.gpr[5] = 0;
1709 
1710     spapr->fwnmi_system_reset_addr = -1;
1711     spapr->fwnmi_machine_check_addr = -1;
1712     spapr->fwnmi_machine_check_interlock = -1;
1713 
1714     /* Signal all vCPUs waiting on this condition */
1715     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1716 
1717     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1718 }
1719 
1720 static void spapr_create_nvram(SpaprMachineState *spapr)
1721 {
1722     DeviceState *dev = qdev_new("spapr-nvram");
1723     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1724 
1725     if (dinfo) {
1726         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1727                                 &error_fatal);
1728     }
1729 
1730     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1731 
1732     spapr->nvram = (struct SpaprNvram *)dev;
1733 }
1734 
1735 static void spapr_rtc_create(SpaprMachineState *spapr)
1736 {
1737     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1738                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1739                                        &error_fatal, NULL);
1740     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1741     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1742                               "date");
1743 }
1744 
1745 /* Returns whether we want to use VGA or not */
1746 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1747 {
1748     switch (vga_interface_type) {
1749     case VGA_NONE:
1750         return false;
1751     case VGA_DEVICE:
1752         return true;
1753     case VGA_STD:
1754     case VGA_VIRTIO:
1755     case VGA_CIRRUS:
1756         return pci_vga_init(pci_bus) != NULL;
1757     default:
1758         error_setg(errp,
1759                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1760         return false;
1761     }
1762 }
1763 
1764 static int spapr_pre_load(void *opaque)
1765 {
1766     int rc;
1767 
1768     rc = spapr_caps_pre_load(opaque);
1769     if (rc) {
1770         return rc;
1771     }
1772 
1773     return 0;
1774 }
1775 
1776 static int spapr_post_load(void *opaque, int version_id)
1777 {
1778     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1779     int err = 0;
1780 
1781     err = spapr_caps_post_migration(spapr);
1782     if (err) {
1783         return err;
1784     }
1785 
1786     /*
1787      * In earlier versions, there was no separate qdev for the PAPR
1788      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1789      * So when migrating from those versions, poke the incoming offset
1790      * value into the RTC device
1791      */
1792     if (version_id < 3) {
1793         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1794         if (err) {
1795             return err;
1796         }
1797     }
1798 
1799     if (kvm_enabled() && spapr->patb_entry) {
1800         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1801         bool radix = !!(spapr->patb_entry & PATE1_GR);
1802         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1803 
1804         /*
1805          * Update LPCR:HR and UPRT as they may not be set properly in
1806          * the stream
1807          */
1808         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1809                             LPCR_HR | LPCR_UPRT);
1810 
1811         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1812         if (err) {
1813             error_report("Process table config unsupported by the host");
1814             return -EINVAL;
1815         }
1816     }
1817 
1818     err = spapr_irq_post_load(spapr, version_id);
1819     if (err) {
1820         return err;
1821     }
1822 
1823     return err;
1824 }
1825 
1826 static int spapr_pre_save(void *opaque)
1827 {
1828     int rc;
1829 
1830     rc = spapr_caps_pre_save(opaque);
1831     if (rc) {
1832         return rc;
1833     }
1834 
1835     return 0;
1836 }
1837 
1838 static bool version_before_3(void *opaque, int version_id)
1839 {
1840     return version_id < 3;
1841 }
1842 
1843 static bool spapr_pending_events_needed(void *opaque)
1844 {
1845     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1846     return !QTAILQ_EMPTY(&spapr->pending_events);
1847 }
1848 
1849 static const VMStateDescription vmstate_spapr_event_entry = {
1850     .name = "spapr_event_log_entry",
1851     .version_id = 1,
1852     .minimum_version_id = 1,
1853     .fields = (VMStateField[]) {
1854         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1855         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1856         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1857                                      NULL, extended_length),
1858         VMSTATE_END_OF_LIST()
1859     },
1860 };
1861 
1862 static const VMStateDescription vmstate_spapr_pending_events = {
1863     .name = "spapr_pending_events",
1864     .version_id = 1,
1865     .minimum_version_id = 1,
1866     .needed = spapr_pending_events_needed,
1867     .fields = (VMStateField[]) {
1868         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1869                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1870         VMSTATE_END_OF_LIST()
1871     },
1872 };
1873 
1874 static bool spapr_ov5_cas_needed(void *opaque)
1875 {
1876     SpaprMachineState *spapr = opaque;
1877     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1878     bool cas_needed;
1879 
1880     /* Prior to the introduction of SpaprOptionVector, we had two option
1881      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1882      * Both of these options encode machine topology into the device-tree
1883      * in such a way that the now-booted OS should still be able to interact
1884      * appropriately with QEMU regardless of what options were actually
1885      * negotiatied on the source side.
1886      *
1887      * As such, we can avoid migrating the CAS-negotiated options if these
1888      * are the only options available on the current machine/platform.
1889      * Since these are the only options available for pseries-2.7 and
1890      * earlier, this allows us to maintain old->new/new->old migration
1891      * compatibility.
1892      *
1893      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1894      * via default pseries-2.8 machines and explicit command-line parameters.
1895      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1896      * of the actual CAS-negotiated values to continue working properly. For
1897      * example, availability of memory unplug depends on knowing whether
1898      * OV5_HP_EVT was negotiated via CAS.
1899      *
1900      * Thus, for any cases where the set of available CAS-negotiatable
1901      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1902      * include the CAS-negotiated options in the migration stream, unless
1903      * if they affect boot time behaviour only.
1904      */
1905     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1906     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1907     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1908 
1909     /* We need extra information if we have any bits outside the mask
1910      * defined above */
1911     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1912 
1913     spapr_ovec_cleanup(ov5_mask);
1914 
1915     return cas_needed;
1916 }
1917 
1918 static const VMStateDescription vmstate_spapr_ov5_cas = {
1919     .name = "spapr_option_vector_ov5_cas",
1920     .version_id = 1,
1921     .minimum_version_id = 1,
1922     .needed = spapr_ov5_cas_needed,
1923     .fields = (VMStateField[]) {
1924         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1925                                  vmstate_spapr_ovec, SpaprOptionVector),
1926         VMSTATE_END_OF_LIST()
1927     },
1928 };
1929 
1930 static bool spapr_patb_entry_needed(void *opaque)
1931 {
1932     SpaprMachineState *spapr = opaque;
1933 
1934     return !!spapr->patb_entry;
1935 }
1936 
1937 static const VMStateDescription vmstate_spapr_patb_entry = {
1938     .name = "spapr_patb_entry",
1939     .version_id = 1,
1940     .minimum_version_id = 1,
1941     .needed = spapr_patb_entry_needed,
1942     .fields = (VMStateField[]) {
1943         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1944         VMSTATE_END_OF_LIST()
1945     },
1946 };
1947 
1948 static bool spapr_irq_map_needed(void *opaque)
1949 {
1950     SpaprMachineState *spapr = opaque;
1951 
1952     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1953 }
1954 
1955 static const VMStateDescription vmstate_spapr_irq_map = {
1956     .name = "spapr_irq_map",
1957     .version_id = 1,
1958     .minimum_version_id = 1,
1959     .needed = spapr_irq_map_needed,
1960     .fields = (VMStateField[]) {
1961         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1962         VMSTATE_END_OF_LIST()
1963     },
1964 };
1965 
1966 static bool spapr_dtb_needed(void *opaque)
1967 {
1968     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1969 
1970     return smc->update_dt_enabled;
1971 }
1972 
1973 static int spapr_dtb_pre_load(void *opaque)
1974 {
1975     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1976 
1977     g_free(spapr->fdt_blob);
1978     spapr->fdt_blob = NULL;
1979     spapr->fdt_size = 0;
1980 
1981     return 0;
1982 }
1983 
1984 static const VMStateDescription vmstate_spapr_dtb = {
1985     .name = "spapr_dtb",
1986     .version_id = 1,
1987     .minimum_version_id = 1,
1988     .needed = spapr_dtb_needed,
1989     .pre_load = spapr_dtb_pre_load,
1990     .fields = (VMStateField[]) {
1991         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1992         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1993         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1994                                      fdt_size),
1995         VMSTATE_END_OF_LIST()
1996     },
1997 };
1998 
1999 static bool spapr_fwnmi_needed(void *opaque)
2000 {
2001     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2002 
2003     return spapr->fwnmi_machine_check_addr != -1;
2004 }
2005 
2006 static int spapr_fwnmi_pre_save(void *opaque)
2007 {
2008     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2009 
2010     /*
2011      * Check if machine check handling is in progress and print a
2012      * warning message.
2013      */
2014     if (spapr->fwnmi_machine_check_interlock != -1) {
2015         warn_report("A machine check is being handled during migration. The"
2016                 "handler may run and log hardware error on the destination");
2017     }
2018 
2019     return 0;
2020 }
2021 
2022 static const VMStateDescription vmstate_spapr_fwnmi = {
2023     .name = "spapr_fwnmi",
2024     .version_id = 1,
2025     .minimum_version_id = 1,
2026     .needed = spapr_fwnmi_needed,
2027     .pre_save = spapr_fwnmi_pre_save,
2028     .fields = (VMStateField[]) {
2029         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
2030         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
2031         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
2032         VMSTATE_END_OF_LIST()
2033     },
2034 };
2035 
2036 static const VMStateDescription vmstate_spapr = {
2037     .name = "spapr",
2038     .version_id = 3,
2039     .minimum_version_id = 1,
2040     .pre_load = spapr_pre_load,
2041     .post_load = spapr_post_load,
2042     .pre_save = spapr_pre_save,
2043     .fields = (VMStateField[]) {
2044         /* used to be @next_irq */
2045         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2046 
2047         /* RTC offset */
2048         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2049 
2050         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2051         VMSTATE_END_OF_LIST()
2052     },
2053     .subsections = (const VMStateDescription*[]) {
2054         &vmstate_spapr_ov5_cas,
2055         &vmstate_spapr_patb_entry,
2056         &vmstate_spapr_pending_events,
2057         &vmstate_spapr_cap_htm,
2058         &vmstate_spapr_cap_vsx,
2059         &vmstate_spapr_cap_dfp,
2060         &vmstate_spapr_cap_cfpc,
2061         &vmstate_spapr_cap_sbbc,
2062         &vmstate_spapr_cap_ibs,
2063         &vmstate_spapr_cap_hpt_maxpagesize,
2064         &vmstate_spapr_irq_map,
2065         &vmstate_spapr_cap_nested_kvm_hv,
2066         &vmstate_spapr_dtb,
2067         &vmstate_spapr_cap_large_decr,
2068         &vmstate_spapr_cap_ccf_assist,
2069         &vmstate_spapr_cap_fwnmi,
2070         &vmstate_spapr_fwnmi,
2071         &vmstate_spapr_cap_rpt_invalidate,
2072         NULL
2073     }
2074 };
2075 
2076 static int htab_save_setup(QEMUFile *f, void *opaque)
2077 {
2078     SpaprMachineState *spapr = opaque;
2079 
2080     /* "Iteration" header */
2081     if (!spapr->htab_shift) {
2082         qemu_put_be32(f, -1);
2083     } else {
2084         qemu_put_be32(f, spapr->htab_shift);
2085     }
2086 
2087     if (spapr->htab) {
2088         spapr->htab_save_index = 0;
2089         spapr->htab_first_pass = true;
2090     } else {
2091         if (spapr->htab_shift) {
2092             assert(kvm_enabled());
2093         }
2094     }
2095 
2096 
2097     return 0;
2098 }
2099 
2100 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2101                             int chunkstart, int n_valid, int n_invalid)
2102 {
2103     qemu_put_be32(f, chunkstart);
2104     qemu_put_be16(f, n_valid);
2105     qemu_put_be16(f, n_invalid);
2106     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2107                     HASH_PTE_SIZE_64 * n_valid);
2108 }
2109 
2110 static void htab_save_end_marker(QEMUFile *f)
2111 {
2112     qemu_put_be32(f, 0);
2113     qemu_put_be16(f, 0);
2114     qemu_put_be16(f, 0);
2115 }
2116 
2117 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2118                                  int64_t max_ns)
2119 {
2120     bool has_timeout = max_ns != -1;
2121     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2122     int index = spapr->htab_save_index;
2123     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2124 
2125     assert(spapr->htab_first_pass);
2126 
2127     do {
2128         int chunkstart;
2129 
2130         /* Consume invalid HPTEs */
2131         while ((index < htabslots)
2132                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2133             CLEAN_HPTE(HPTE(spapr->htab, index));
2134             index++;
2135         }
2136 
2137         /* Consume valid HPTEs */
2138         chunkstart = index;
2139         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2140                && HPTE_VALID(HPTE(spapr->htab, index))) {
2141             CLEAN_HPTE(HPTE(spapr->htab, index));
2142             index++;
2143         }
2144 
2145         if (index > chunkstart) {
2146             int n_valid = index - chunkstart;
2147 
2148             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2149 
2150             if (has_timeout &&
2151                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2152                 break;
2153             }
2154         }
2155     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2156 
2157     if (index >= htabslots) {
2158         assert(index == htabslots);
2159         index = 0;
2160         spapr->htab_first_pass = false;
2161     }
2162     spapr->htab_save_index = index;
2163 }
2164 
2165 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2166                                 int64_t max_ns)
2167 {
2168     bool final = max_ns < 0;
2169     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2170     int examined = 0, sent = 0;
2171     int index = spapr->htab_save_index;
2172     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2173 
2174     assert(!spapr->htab_first_pass);
2175 
2176     do {
2177         int chunkstart, invalidstart;
2178 
2179         /* Consume non-dirty HPTEs */
2180         while ((index < htabslots)
2181                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2182             index++;
2183             examined++;
2184         }
2185 
2186         chunkstart = index;
2187         /* Consume valid dirty HPTEs */
2188         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2189                && HPTE_DIRTY(HPTE(spapr->htab, index))
2190                && HPTE_VALID(HPTE(spapr->htab, index))) {
2191             CLEAN_HPTE(HPTE(spapr->htab, index));
2192             index++;
2193             examined++;
2194         }
2195 
2196         invalidstart = index;
2197         /* Consume invalid dirty HPTEs */
2198         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2199                && HPTE_DIRTY(HPTE(spapr->htab, index))
2200                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2201             CLEAN_HPTE(HPTE(spapr->htab, index));
2202             index++;
2203             examined++;
2204         }
2205 
2206         if (index > chunkstart) {
2207             int n_valid = invalidstart - chunkstart;
2208             int n_invalid = index - invalidstart;
2209 
2210             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2211             sent += index - chunkstart;
2212 
2213             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2214                 break;
2215             }
2216         }
2217 
2218         if (examined >= htabslots) {
2219             break;
2220         }
2221 
2222         if (index >= htabslots) {
2223             assert(index == htabslots);
2224             index = 0;
2225         }
2226     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2227 
2228     if (index >= htabslots) {
2229         assert(index == htabslots);
2230         index = 0;
2231     }
2232 
2233     spapr->htab_save_index = index;
2234 
2235     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2236 }
2237 
2238 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2239 #define MAX_KVM_BUF_SIZE    2048
2240 
2241 static int htab_save_iterate(QEMUFile *f, void *opaque)
2242 {
2243     SpaprMachineState *spapr = opaque;
2244     int fd;
2245     int rc = 0;
2246 
2247     /* Iteration header */
2248     if (!spapr->htab_shift) {
2249         qemu_put_be32(f, -1);
2250         return 1;
2251     } else {
2252         qemu_put_be32(f, 0);
2253     }
2254 
2255     if (!spapr->htab) {
2256         assert(kvm_enabled());
2257 
2258         fd = get_htab_fd(spapr);
2259         if (fd < 0) {
2260             return fd;
2261         }
2262 
2263         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2264         if (rc < 0) {
2265             return rc;
2266         }
2267     } else  if (spapr->htab_first_pass) {
2268         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2269     } else {
2270         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2271     }
2272 
2273     htab_save_end_marker(f);
2274 
2275     return rc;
2276 }
2277 
2278 static int htab_save_complete(QEMUFile *f, void *opaque)
2279 {
2280     SpaprMachineState *spapr = opaque;
2281     int fd;
2282 
2283     /* Iteration header */
2284     if (!spapr->htab_shift) {
2285         qemu_put_be32(f, -1);
2286         return 0;
2287     } else {
2288         qemu_put_be32(f, 0);
2289     }
2290 
2291     if (!spapr->htab) {
2292         int rc;
2293 
2294         assert(kvm_enabled());
2295 
2296         fd = get_htab_fd(spapr);
2297         if (fd < 0) {
2298             return fd;
2299         }
2300 
2301         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2302         if (rc < 0) {
2303             return rc;
2304         }
2305     } else {
2306         if (spapr->htab_first_pass) {
2307             htab_save_first_pass(f, spapr, -1);
2308         }
2309         htab_save_later_pass(f, spapr, -1);
2310     }
2311 
2312     /* End marker */
2313     htab_save_end_marker(f);
2314 
2315     return 0;
2316 }
2317 
2318 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2319 {
2320     SpaprMachineState *spapr = opaque;
2321     uint32_t section_hdr;
2322     int fd = -1;
2323     Error *local_err = NULL;
2324 
2325     if (version_id < 1 || version_id > 1) {
2326         error_report("htab_load() bad version");
2327         return -EINVAL;
2328     }
2329 
2330     section_hdr = qemu_get_be32(f);
2331 
2332     if (section_hdr == -1) {
2333         spapr_free_hpt(spapr);
2334         return 0;
2335     }
2336 
2337     if (section_hdr) {
2338         int ret;
2339 
2340         /* First section gives the htab size */
2341         ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2342         if (ret < 0) {
2343             error_report_err(local_err);
2344             return ret;
2345         }
2346         return 0;
2347     }
2348 
2349     if (!spapr->htab) {
2350         assert(kvm_enabled());
2351 
2352         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2353         if (fd < 0) {
2354             error_report_err(local_err);
2355             return fd;
2356         }
2357     }
2358 
2359     while (true) {
2360         uint32_t index;
2361         uint16_t n_valid, n_invalid;
2362 
2363         index = qemu_get_be32(f);
2364         n_valid = qemu_get_be16(f);
2365         n_invalid = qemu_get_be16(f);
2366 
2367         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2368             /* End of Stream */
2369             break;
2370         }
2371 
2372         if ((index + n_valid + n_invalid) >
2373             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2374             /* Bad index in stream */
2375             error_report(
2376                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2377                 index, n_valid, n_invalid, spapr->htab_shift);
2378             return -EINVAL;
2379         }
2380 
2381         if (spapr->htab) {
2382             if (n_valid) {
2383                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2384                                 HASH_PTE_SIZE_64 * n_valid);
2385             }
2386             if (n_invalid) {
2387                 memset(HPTE(spapr->htab, index + n_valid), 0,
2388                        HASH_PTE_SIZE_64 * n_invalid);
2389             }
2390         } else {
2391             int rc;
2392 
2393             assert(fd >= 0);
2394 
2395             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2396                                         &local_err);
2397             if (rc < 0) {
2398                 error_report_err(local_err);
2399                 return rc;
2400             }
2401         }
2402     }
2403 
2404     if (!spapr->htab) {
2405         assert(fd >= 0);
2406         close(fd);
2407     }
2408 
2409     return 0;
2410 }
2411 
2412 static void htab_save_cleanup(void *opaque)
2413 {
2414     SpaprMachineState *spapr = opaque;
2415 
2416     close_htab_fd(spapr);
2417 }
2418 
2419 static SaveVMHandlers savevm_htab_handlers = {
2420     .save_setup = htab_save_setup,
2421     .save_live_iterate = htab_save_iterate,
2422     .save_live_complete_precopy = htab_save_complete,
2423     .save_cleanup = htab_save_cleanup,
2424     .load_state = htab_load,
2425 };
2426 
2427 static void spapr_boot_set(void *opaque, const char *boot_device,
2428                            Error **errp)
2429 {
2430     SpaprMachineState *spapr = SPAPR_MACHINE(opaque);
2431 
2432     g_free(spapr->boot_device);
2433     spapr->boot_device = g_strdup(boot_device);
2434 }
2435 
2436 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2437 {
2438     MachineState *machine = MACHINE(spapr);
2439     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2440     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2441     int i;
2442 
2443     for (i = 0; i < nr_lmbs; i++) {
2444         uint64_t addr;
2445 
2446         addr = i * lmb_size + machine->device_memory->base;
2447         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2448                                addr / lmb_size);
2449     }
2450 }
2451 
2452 /*
2453  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2454  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2455  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2456  */
2457 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2458 {
2459     int i;
2460 
2461     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2462         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2463                    " is not aligned to %" PRIu64 " MiB",
2464                    machine->ram_size,
2465                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2466         return;
2467     }
2468 
2469     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2470         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2471                    " is not aligned to %" PRIu64 " MiB",
2472                    machine->ram_size,
2473                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2474         return;
2475     }
2476 
2477     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2478         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2479             error_setg(errp,
2480                        "Node %d memory size 0x%" PRIx64
2481                        " is not aligned to %" PRIu64 " MiB",
2482                        i, machine->numa_state->nodes[i].node_mem,
2483                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2484             return;
2485         }
2486     }
2487 }
2488 
2489 /* find cpu slot in machine->possible_cpus by core_id */
2490 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2491 {
2492     int index = id / ms->smp.threads;
2493 
2494     if (index >= ms->possible_cpus->len) {
2495         return NULL;
2496     }
2497     if (idx) {
2498         *idx = index;
2499     }
2500     return &ms->possible_cpus->cpus[index];
2501 }
2502 
2503 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2504 {
2505     MachineState *ms = MACHINE(spapr);
2506     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2507     Error *local_err = NULL;
2508     bool vsmt_user = !!spapr->vsmt;
2509     int kvm_smt = kvmppc_smt_threads();
2510     int ret;
2511     unsigned int smp_threads = ms->smp.threads;
2512 
2513     if (!kvm_enabled() && (smp_threads > 1)) {
2514         error_setg(errp, "TCG cannot support more than 1 thread/core "
2515                    "on a pseries machine");
2516         return;
2517     }
2518     if (!is_power_of_2(smp_threads)) {
2519         error_setg(errp, "Cannot support %d threads/core on a pseries "
2520                    "machine because it must be a power of 2", smp_threads);
2521         return;
2522     }
2523 
2524     /* Detemine the VSMT mode to use: */
2525     if (vsmt_user) {
2526         if (spapr->vsmt < smp_threads) {
2527             error_setg(errp, "Cannot support VSMT mode %d"
2528                        " because it must be >= threads/core (%d)",
2529                        spapr->vsmt, smp_threads);
2530             return;
2531         }
2532         /* In this case, spapr->vsmt has been set by the command line */
2533     } else if (!smc->smp_threads_vsmt) {
2534         /*
2535          * Default VSMT value is tricky, because we need it to be as
2536          * consistent as possible (for migration), but this requires
2537          * changing it for at least some existing cases.  We pick 8 as
2538          * the value that we'd get with KVM on POWER8, the
2539          * overwhelmingly common case in production systems.
2540          */
2541         spapr->vsmt = MAX(8, smp_threads);
2542     } else {
2543         spapr->vsmt = smp_threads;
2544     }
2545 
2546     /* KVM: If necessary, set the SMT mode: */
2547     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2548         ret = kvmppc_set_smt_threads(spapr->vsmt);
2549         if (ret) {
2550             /* Looks like KVM isn't able to change VSMT mode */
2551             error_setg(&local_err,
2552                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2553                        spapr->vsmt, ret);
2554             /* We can live with that if the default one is big enough
2555              * for the number of threads, and a submultiple of the one
2556              * we want.  In this case we'll waste some vcpu ids, but
2557              * behaviour will be correct */
2558             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2559                 warn_report_err(local_err);
2560             } else {
2561                 if (!vsmt_user) {
2562                     error_append_hint(&local_err,
2563                                       "On PPC, a VM with %d threads/core"
2564                                       " on a host with %d threads/core"
2565                                       " requires the use of VSMT mode %d.\n",
2566                                       smp_threads, kvm_smt, spapr->vsmt);
2567                 }
2568                 kvmppc_error_append_smt_possible_hint(&local_err);
2569                 error_propagate(errp, local_err);
2570             }
2571         }
2572     }
2573     /* else TCG: nothing to do currently */
2574 }
2575 
2576 static void spapr_init_cpus(SpaprMachineState *spapr)
2577 {
2578     MachineState *machine = MACHINE(spapr);
2579     MachineClass *mc = MACHINE_GET_CLASS(machine);
2580     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2581     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2582     const CPUArchIdList *possible_cpus;
2583     unsigned int smp_cpus = machine->smp.cpus;
2584     unsigned int smp_threads = machine->smp.threads;
2585     unsigned int max_cpus = machine->smp.max_cpus;
2586     int boot_cores_nr = smp_cpus / smp_threads;
2587     int i;
2588 
2589     possible_cpus = mc->possible_cpu_arch_ids(machine);
2590     if (mc->has_hotpluggable_cpus) {
2591         if (smp_cpus % smp_threads) {
2592             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2593                          smp_cpus, smp_threads);
2594             exit(1);
2595         }
2596         if (max_cpus % smp_threads) {
2597             error_report("max_cpus (%u) must be multiple of threads (%u)",
2598                          max_cpus, smp_threads);
2599             exit(1);
2600         }
2601     } else {
2602         if (max_cpus != smp_cpus) {
2603             error_report("This machine version does not support CPU hotplug");
2604             exit(1);
2605         }
2606         boot_cores_nr = possible_cpus->len;
2607     }
2608 
2609     if (smc->pre_2_10_has_unused_icps) {
2610         int i;
2611 
2612         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2613             /* Dummy entries get deregistered when real ICPState objects
2614              * are registered during CPU core hotplug.
2615              */
2616             pre_2_10_vmstate_register_dummy_icp(i);
2617         }
2618     }
2619 
2620     for (i = 0; i < possible_cpus->len; i++) {
2621         int core_id = i * smp_threads;
2622 
2623         if (mc->has_hotpluggable_cpus) {
2624             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2625                                    spapr_vcpu_id(spapr, core_id));
2626         }
2627 
2628         if (i < boot_cores_nr) {
2629             Object *core  = object_new(type);
2630             int nr_threads = smp_threads;
2631 
2632             /* Handle the partially filled core for older machine types */
2633             if ((i + 1) * smp_threads >= smp_cpus) {
2634                 nr_threads = smp_cpus - i * smp_threads;
2635             }
2636 
2637             object_property_set_int(core, "nr-threads", nr_threads,
2638                                     &error_fatal);
2639             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2640                                     &error_fatal);
2641             qdev_realize(DEVICE(core), NULL, &error_fatal);
2642 
2643             object_unref(core);
2644         }
2645     }
2646 }
2647 
2648 static PCIHostState *spapr_create_default_phb(void)
2649 {
2650     DeviceState *dev;
2651 
2652     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2653     qdev_prop_set_uint32(dev, "index", 0);
2654     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2655 
2656     return PCI_HOST_BRIDGE(dev);
2657 }
2658 
2659 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2660 {
2661     MachineState *machine = MACHINE(spapr);
2662     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2663     hwaddr rma_size = machine->ram_size;
2664     hwaddr node0_size = spapr_node0_size(machine);
2665 
2666     /* RMA has to fit in the first NUMA node */
2667     rma_size = MIN(rma_size, node0_size);
2668 
2669     /*
2670      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2671      * never exceed that
2672      */
2673     rma_size = MIN(rma_size, 1 * TiB);
2674 
2675     /*
2676      * Clamp the RMA size based on machine type.  This is for
2677      * migration compatibility with older qemu versions, which limited
2678      * the RMA size for complicated and mostly bad reasons.
2679      */
2680     if (smc->rma_limit) {
2681         rma_size = MIN(rma_size, smc->rma_limit);
2682     }
2683 
2684     if (rma_size < MIN_RMA_SLOF) {
2685         error_setg(errp,
2686                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2687                    "ldMiB guest RMA (Real Mode Area memory)",
2688                    MIN_RMA_SLOF / MiB);
2689         return 0;
2690     }
2691 
2692     return rma_size;
2693 }
2694 
2695 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2696 {
2697     MachineState *machine = MACHINE(spapr);
2698     int i;
2699 
2700     for (i = 0; i < machine->ram_slots; i++) {
2701         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2702     }
2703 }
2704 
2705 /* pSeries LPAR / sPAPR hardware init */
2706 static void spapr_machine_init(MachineState *machine)
2707 {
2708     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2709     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2710     MachineClass *mc = MACHINE_GET_CLASS(machine);
2711     const char *bios_default = spapr->vof ? FW_FILE_NAME_VOF : FW_FILE_NAME;
2712     const char *bios_name = machine->firmware ?: bios_default;
2713     const char *kernel_filename = machine->kernel_filename;
2714     const char *initrd_filename = machine->initrd_filename;
2715     PCIHostState *phb;
2716     int i;
2717     MemoryRegion *sysmem = get_system_memory();
2718     long load_limit, fw_size;
2719     char *filename;
2720     Error *resize_hpt_err = NULL;
2721 
2722     /*
2723      * if Secure VM (PEF) support is configured, then initialize it
2724      */
2725     pef_kvm_init(machine->cgs, &error_fatal);
2726 
2727     msi_nonbroken = true;
2728 
2729     QLIST_INIT(&spapr->phbs);
2730     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2731 
2732     /* Determine capabilities to run with */
2733     spapr_caps_init(spapr);
2734 
2735     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2736     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2737         /*
2738          * If the user explicitly requested a mode we should either
2739          * supply it, or fail completely (which we do below).  But if
2740          * it's not set explicitly, we reset our mode to something
2741          * that works
2742          */
2743         if (resize_hpt_err) {
2744             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2745             error_free(resize_hpt_err);
2746             resize_hpt_err = NULL;
2747         } else {
2748             spapr->resize_hpt = smc->resize_hpt_default;
2749         }
2750     }
2751 
2752     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2753 
2754     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2755         /*
2756          * User requested HPT resize, but this host can't supply it.  Bail out
2757          */
2758         error_report_err(resize_hpt_err);
2759         exit(1);
2760     }
2761     error_free(resize_hpt_err);
2762 
2763     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2764 
2765     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2766     load_limit = MIN(spapr->rma_size, FDT_MAX_ADDR) - FW_OVERHEAD;
2767 
2768     /*
2769      * VSMT must be set in order to be able to compute VCPU ids, ie to
2770      * call spapr_max_server_number() or spapr_vcpu_id().
2771      */
2772     spapr_set_vsmt_mode(spapr, &error_fatal);
2773 
2774     /* Set up Interrupt Controller before we create the VCPUs */
2775     spapr_irq_init(spapr, &error_fatal);
2776 
2777     /* Set up containers for ibm,client-architecture-support negotiated options
2778      */
2779     spapr->ov5 = spapr_ovec_new();
2780     spapr->ov5_cas = spapr_ovec_new();
2781 
2782     if (smc->dr_lmb_enabled) {
2783         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2784         spapr_validate_node_memory(machine, &error_fatal);
2785     }
2786 
2787     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2788 
2789     /* Do not advertise FORM2 NUMA support for pseries-6.1 and older */
2790     if (!smc->pre_6_2_numa_affinity) {
2791         spapr_ovec_set(spapr->ov5, OV5_FORM2_AFFINITY);
2792     }
2793 
2794     /* advertise support for dedicated HP event source to guests */
2795     if (spapr->use_hotplug_event_source) {
2796         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2797     }
2798 
2799     /* advertise support for HPT resizing */
2800     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2801         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2802     }
2803 
2804     /* advertise support for ibm,dyamic-memory-v2 */
2805     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2806 
2807     /* advertise XIVE on POWER9 machines */
2808     if (spapr->irq->xive) {
2809         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2810     }
2811 
2812     /* init CPUs */
2813     spapr_init_cpus(spapr);
2814 
2815     spapr->gpu_numa_id = spapr_numa_initial_nvgpu_numa_id(machine);
2816 
2817     /* Init numa_assoc_array */
2818     spapr_numa_associativity_init(spapr, machine);
2819 
2820     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2821         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2822                               spapr->max_compat_pvr)) {
2823         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2824         /* KVM and TCG always allow GTSE with radix... */
2825         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2826     }
2827     /* ... but not with hash (currently). */
2828 
2829     if (kvm_enabled()) {
2830         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2831         kvmppc_enable_logical_ci_hcalls();
2832         kvmppc_enable_set_mode_hcall();
2833 
2834         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2835         kvmppc_enable_clear_ref_mod_hcalls();
2836 
2837         /* Enable H_PAGE_INIT */
2838         kvmppc_enable_h_page_init();
2839     }
2840 
2841     /* map RAM */
2842     memory_region_add_subregion(sysmem, 0, machine->ram);
2843 
2844     /* always allocate the device memory information */
2845     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2846 
2847     /* initialize hotplug memory address space */
2848     if (machine->ram_size < machine->maxram_size) {
2849         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2850         /*
2851          * Limit the number of hotpluggable memory slots to half the number
2852          * slots that KVM supports, leaving the other half for PCI and other
2853          * devices. However ensure that number of slots doesn't drop below 32.
2854          */
2855         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2856                            SPAPR_MAX_RAM_SLOTS;
2857 
2858         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2859             max_memslots = SPAPR_MAX_RAM_SLOTS;
2860         }
2861         if (machine->ram_slots > max_memslots) {
2862             error_report("Specified number of memory slots %"
2863                          PRIu64" exceeds max supported %d",
2864                          machine->ram_slots, max_memslots);
2865             exit(1);
2866         }
2867 
2868         machine->device_memory->base = ROUND_UP(machine->ram_size,
2869                                                 SPAPR_DEVICE_MEM_ALIGN);
2870         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2871                            "device-memory", device_mem_size);
2872         memory_region_add_subregion(sysmem, machine->device_memory->base,
2873                                     &machine->device_memory->mr);
2874     }
2875 
2876     if (smc->dr_lmb_enabled) {
2877         spapr_create_lmb_dr_connectors(spapr);
2878     }
2879 
2880     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2881         /* Create the error string for live migration blocker */
2882         error_setg(&spapr->fwnmi_migration_blocker,
2883             "A machine check is being handled during migration. The handler"
2884             "may run and log hardware error on the destination");
2885     }
2886 
2887     if (mc->nvdimm_supported) {
2888         spapr_create_nvdimm_dr_connectors(spapr);
2889     }
2890 
2891     /* Set up RTAS event infrastructure */
2892     spapr_events_init(spapr);
2893 
2894     /* Set up the RTC RTAS interfaces */
2895     spapr_rtc_create(spapr);
2896 
2897     /* Set up VIO bus */
2898     spapr->vio_bus = spapr_vio_bus_init();
2899 
2900     for (i = 0; serial_hd(i); i++) {
2901         spapr_vty_create(spapr->vio_bus, serial_hd(i));
2902     }
2903 
2904     /* We always have at least the nvram device on VIO */
2905     spapr_create_nvram(spapr);
2906 
2907     /*
2908      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2909      * connectors (described in root DT node's "ibm,drc-types" property)
2910      * are pre-initialized here. additional child connectors (such as
2911      * connectors for a PHBs PCI slots) are added as needed during their
2912      * parent's realization.
2913      */
2914     if (smc->dr_phb_enabled) {
2915         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2916             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2917         }
2918     }
2919 
2920     /* Set up PCI */
2921     spapr_pci_rtas_init();
2922 
2923     phb = spapr_create_default_phb();
2924 
2925     for (i = 0; i < nb_nics; i++) {
2926         NICInfo *nd = &nd_table[i];
2927 
2928         if (!nd->model) {
2929             nd->model = g_strdup("spapr-vlan");
2930         }
2931 
2932         if (g_str_equal(nd->model, "spapr-vlan") ||
2933             g_str_equal(nd->model, "ibmveth")) {
2934             spapr_vlan_create(spapr->vio_bus, nd);
2935         } else {
2936             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2937         }
2938     }
2939 
2940     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2941         spapr_vscsi_create(spapr->vio_bus);
2942     }
2943 
2944     /* Graphics */
2945     if (spapr_vga_init(phb->bus, &error_fatal)) {
2946         spapr->has_graphics = true;
2947         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2948     }
2949 
2950     if (machine->usb) {
2951         if (smc->use_ohci_by_default) {
2952             pci_create_simple(phb->bus, -1, "pci-ohci");
2953         } else {
2954             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2955         }
2956 
2957         if (spapr->has_graphics) {
2958             USBBus *usb_bus = usb_bus_find(-1);
2959 
2960             usb_create_simple(usb_bus, "usb-kbd");
2961             usb_create_simple(usb_bus, "usb-mouse");
2962         }
2963     }
2964 
2965     if (kernel_filename) {
2966         spapr->kernel_size = load_elf(kernel_filename, NULL,
2967                                       translate_kernel_address, spapr,
2968                                       NULL, NULL, NULL, NULL, 1,
2969                                       PPC_ELF_MACHINE, 0, 0);
2970         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2971             spapr->kernel_size = load_elf(kernel_filename, NULL,
2972                                           translate_kernel_address, spapr,
2973                                           NULL, NULL, NULL, NULL, 0,
2974                                           PPC_ELF_MACHINE, 0, 0);
2975             spapr->kernel_le = spapr->kernel_size > 0;
2976         }
2977         if (spapr->kernel_size < 0) {
2978             error_report("error loading %s: %s", kernel_filename,
2979                          load_elf_strerror(spapr->kernel_size));
2980             exit(1);
2981         }
2982 
2983         /* load initrd */
2984         if (initrd_filename) {
2985             /* Try to locate the initrd in the gap between the kernel
2986              * and the firmware. Add a bit of space just in case
2987              */
2988             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
2989                                   + 0x1ffff) & ~0xffff;
2990             spapr->initrd_size = load_image_targphys(initrd_filename,
2991                                                      spapr->initrd_base,
2992                                                      load_limit
2993                                                      - spapr->initrd_base);
2994             if (spapr->initrd_size < 0) {
2995                 error_report("could not load initial ram disk '%s'",
2996                              initrd_filename);
2997                 exit(1);
2998             }
2999         }
3000     }
3001 
3002     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
3003     if (!filename) {
3004         error_report("Could not find LPAR firmware '%s'", bios_name);
3005         exit(1);
3006     }
3007     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
3008     if (fw_size <= 0) {
3009         error_report("Could not load LPAR firmware '%s'", filename);
3010         exit(1);
3011     }
3012     g_free(filename);
3013 
3014     /* FIXME: Should register things through the MachineState's qdev
3015      * interface, this is a legacy from the sPAPREnvironment structure
3016      * which predated MachineState but had a similar function */
3017     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3018     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3019                          &savevm_htab_handlers, spapr);
3020 
3021     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
3022 
3023     qemu_register_boot_set(spapr_boot_set, spapr);
3024 
3025     /*
3026      * Nothing needs to be done to resume a suspended guest because
3027      * suspending does not change the machine state, so no need for
3028      * a ->wakeup method.
3029      */
3030     qemu_register_wakeup_support();
3031 
3032     if (kvm_enabled()) {
3033         /* to stop and start vmclock */
3034         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3035                                          &spapr->tb);
3036 
3037         kvmppc_spapr_enable_inkernel_multitce();
3038     }
3039 
3040     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3041     if (spapr->vof) {
3042         spapr->vof->fw_size = fw_size; /* for claim() on itself */
3043         spapr_register_hypercall(KVMPPC_H_VOF_CLIENT, spapr_h_vof_client);
3044     }
3045 }
3046 
3047 #define DEFAULT_KVM_TYPE "auto"
3048 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3049 {
3050     /*
3051      * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to
3052      * accomodate the 'HV' and 'PV' formats that exists in the
3053      * wild. The 'auto' mode is being introduced already as
3054      * lower-case, thus we don't need to bother checking for
3055      * "AUTO".
3056      */
3057     if (!vm_type || !strcmp(vm_type, DEFAULT_KVM_TYPE)) {
3058         return 0;
3059     }
3060 
3061     if (!g_ascii_strcasecmp(vm_type, "hv")) {
3062         return 1;
3063     }
3064 
3065     if (!g_ascii_strcasecmp(vm_type, "pr")) {
3066         return 2;
3067     }
3068 
3069     error_report("Unknown kvm-type specified '%s'", vm_type);
3070     exit(1);
3071 }
3072 
3073 /*
3074  * Implementation of an interface to adjust firmware path
3075  * for the bootindex property handling.
3076  */
3077 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3078                                    DeviceState *dev)
3079 {
3080 #define CAST(type, obj, name) \
3081     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3082     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3083     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3084     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3085     PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3086 
3087     if (d && bus) {
3088         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3089         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3090         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3091 
3092         if (spapr) {
3093             /*
3094              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3095              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3096              * 0x8000 | (target << 8) | (bus << 5) | lun
3097              * (see the "Logical unit addressing format" table in SAM5)
3098              */
3099             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3100             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3101                                    (uint64_t)id << 48);
3102         } else if (virtio) {
3103             /*
3104              * We use SRP luns of the form 01000000 | (target << 8) | lun
3105              * in the top 32 bits of the 64-bit LUN
3106              * Note: the quote above is from SLOF and it is wrong,
3107              * the actual binding is:
3108              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3109              */
3110             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3111             if (d->lun >= 256) {
3112                 /* Use the LUN "flat space addressing method" */
3113                 id |= 0x4000;
3114             }
3115             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3116                                    (uint64_t)id << 32);
3117         } else if (usb) {
3118             /*
3119              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3120              * in the top 32 bits of the 64-bit LUN
3121              */
3122             unsigned usb_port = atoi(usb->port->path);
3123             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3124             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3125                                    (uint64_t)id << 32);
3126         }
3127     }
3128 
3129     /*
3130      * SLOF probes the USB devices, and if it recognizes that the device is a
3131      * storage device, it changes its name to "storage" instead of "usb-host",
3132      * and additionally adds a child node for the SCSI LUN, so the correct
3133      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3134      */
3135     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3136         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3137         if (usb_device_is_scsi_storage(usbdev)) {
3138             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3139         }
3140     }
3141 
3142     if (phb) {
3143         /* Replace "pci" with "pci@800000020000000" */
3144         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3145     }
3146 
3147     if (vsc) {
3148         /* Same logic as virtio above */
3149         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3150         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3151     }
3152 
3153     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3154         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3155         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3156         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3157     }
3158 
3159     if (pcidev) {
3160         return spapr_pci_fw_dev_name(pcidev);
3161     }
3162 
3163     return NULL;
3164 }
3165 
3166 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3167 {
3168     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3169 
3170     return g_strdup(spapr->kvm_type);
3171 }
3172 
3173 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3174 {
3175     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3176 
3177     g_free(spapr->kvm_type);
3178     spapr->kvm_type = g_strdup(value);
3179 }
3180 
3181 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3182 {
3183     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3184 
3185     return spapr->use_hotplug_event_source;
3186 }
3187 
3188 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3189                                             Error **errp)
3190 {
3191     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3192 
3193     spapr->use_hotplug_event_source = value;
3194 }
3195 
3196 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3197 {
3198     return true;
3199 }
3200 
3201 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3202 {
3203     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3204 
3205     switch (spapr->resize_hpt) {
3206     case SPAPR_RESIZE_HPT_DEFAULT:
3207         return g_strdup("default");
3208     case SPAPR_RESIZE_HPT_DISABLED:
3209         return g_strdup("disabled");
3210     case SPAPR_RESIZE_HPT_ENABLED:
3211         return g_strdup("enabled");
3212     case SPAPR_RESIZE_HPT_REQUIRED:
3213         return g_strdup("required");
3214     }
3215     g_assert_not_reached();
3216 }
3217 
3218 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3219 {
3220     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3221 
3222     if (strcmp(value, "default") == 0) {
3223         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3224     } else if (strcmp(value, "disabled") == 0) {
3225         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3226     } else if (strcmp(value, "enabled") == 0) {
3227         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3228     } else if (strcmp(value, "required") == 0) {
3229         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3230     } else {
3231         error_setg(errp, "Bad value for \"resize-hpt\" property");
3232     }
3233 }
3234 
3235 static bool spapr_get_vof(Object *obj, Error **errp)
3236 {
3237     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3238 
3239     return spapr->vof != NULL;
3240 }
3241 
3242 static void spapr_set_vof(Object *obj, bool value, Error **errp)
3243 {
3244     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3245 
3246     if (spapr->vof) {
3247         vof_cleanup(spapr->vof);
3248         g_free(spapr->vof);
3249         spapr->vof = NULL;
3250     }
3251     if (!value) {
3252         return;
3253     }
3254     spapr->vof = g_malloc0(sizeof(*spapr->vof));
3255 }
3256 
3257 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3258 {
3259     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3260 
3261     if (spapr->irq == &spapr_irq_xics_legacy) {
3262         return g_strdup("legacy");
3263     } else if (spapr->irq == &spapr_irq_xics) {
3264         return g_strdup("xics");
3265     } else if (spapr->irq == &spapr_irq_xive) {
3266         return g_strdup("xive");
3267     } else if (spapr->irq == &spapr_irq_dual) {
3268         return g_strdup("dual");
3269     }
3270     g_assert_not_reached();
3271 }
3272 
3273 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3274 {
3275     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3276 
3277     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3278         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3279         return;
3280     }
3281 
3282     /* The legacy IRQ backend can not be set */
3283     if (strcmp(value, "xics") == 0) {
3284         spapr->irq = &spapr_irq_xics;
3285     } else if (strcmp(value, "xive") == 0) {
3286         spapr->irq = &spapr_irq_xive;
3287     } else if (strcmp(value, "dual") == 0) {
3288         spapr->irq = &spapr_irq_dual;
3289     } else {
3290         error_setg(errp, "Bad value for \"ic-mode\" property");
3291     }
3292 }
3293 
3294 static char *spapr_get_host_model(Object *obj, Error **errp)
3295 {
3296     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3297 
3298     return g_strdup(spapr->host_model);
3299 }
3300 
3301 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3302 {
3303     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3304 
3305     g_free(spapr->host_model);
3306     spapr->host_model = g_strdup(value);
3307 }
3308 
3309 static char *spapr_get_host_serial(Object *obj, Error **errp)
3310 {
3311     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3312 
3313     return g_strdup(spapr->host_serial);
3314 }
3315 
3316 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3317 {
3318     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3319 
3320     g_free(spapr->host_serial);
3321     spapr->host_serial = g_strdup(value);
3322 }
3323 
3324 static void spapr_instance_init(Object *obj)
3325 {
3326     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3327     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3328     MachineState *ms = MACHINE(spapr);
3329     MachineClass *mc = MACHINE_GET_CLASS(ms);
3330 
3331     /*
3332      * NVDIMM support went live in 5.1 without considering that, in
3333      * other archs, the user needs to enable NVDIMM support with the
3334      * 'nvdimm' machine option and the default behavior is NVDIMM
3335      * support disabled. It is too late to roll back to the standard
3336      * behavior without breaking 5.1 guests.
3337      */
3338     if (mc->nvdimm_supported) {
3339         ms->nvdimms_state->is_enabled = true;
3340     }
3341 
3342     spapr->htab_fd = -1;
3343     spapr->use_hotplug_event_source = true;
3344     spapr->kvm_type = g_strdup(DEFAULT_KVM_TYPE);
3345     object_property_add_str(obj, "kvm-type",
3346                             spapr_get_kvm_type, spapr_set_kvm_type);
3347     object_property_set_description(obj, "kvm-type",
3348                                     "Specifies the KVM virtualization mode (auto,"
3349                                     " hv, pr). Defaults to 'auto'. This mode will use"
3350                                     " any available KVM module loaded in the host,"
3351                                     " where kvm_hv takes precedence if both kvm_hv and"
3352                                     " kvm_pr are loaded.");
3353     object_property_add_bool(obj, "modern-hotplug-events",
3354                             spapr_get_modern_hotplug_events,
3355                             spapr_set_modern_hotplug_events);
3356     object_property_set_description(obj, "modern-hotplug-events",
3357                                     "Use dedicated hotplug event mechanism in"
3358                                     " place of standard EPOW events when possible"
3359                                     " (required for memory hot-unplug support)");
3360     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3361                             "Maximum permitted CPU compatibility mode");
3362 
3363     object_property_add_str(obj, "resize-hpt",
3364                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3365     object_property_set_description(obj, "resize-hpt",
3366                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3367     object_property_add_uint32_ptr(obj, "vsmt",
3368                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3369     object_property_set_description(obj, "vsmt",
3370                                     "Virtual SMT: KVM behaves as if this were"
3371                                     " the host's SMT mode");
3372 
3373     object_property_add_bool(obj, "vfio-no-msix-emulation",
3374                              spapr_get_msix_emulation, NULL);
3375 
3376     object_property_add_uint64_ptr(obj, "kernel-addr",
3377                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3378     object_property_set_description(obj, "kernel-addr",
3379                                     stringify(KERNEL_LOAD_ADDR)
3380                                     " for -kernel is the default");
3381     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3382 
3383     object_property_add_bool(obj, "x-vof", spapr_get_vof, spapr_set_vof);
3384     object_property_set_description(obj, "x-vof",
3385                                     "Enable Virtual Open Firmware (experimental)");
3386 
3387     /* The machine class defines the default interrupt controller mode */
3388     spapr->irq = smc->irq;
3389     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3390                             spapr_set_ic_mode);
3391     object_property_set_description(obj, "ic-mode",
3392                  "Specifies the interrupt controller mode (xics, xive, dual)");
3393 
3394     object_property_add_str(obj, "host-model",
3395         spapr_get_host_model, spapr_set_host_model);
3396     object_property_set_description(obj, "host-model",
3397         "Host model to advertise in guest device tree");
3398     object_property_add_str(obj, "host-serial",
3399         spapr_get_host_serial, spapr_set_host_serial);
3400     object_property_set_description(obj, "host-serial",
3401         "Host serial number to advertise in guest device tree");
3402 }
3403 
3404 static void spapr_machine_finalizefn(Object *obj)
3405 {
3406     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3407 
3408     g_free(spapr->kvm_type);
3409 }
3410 
3411 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3412 {
3413     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3414     PowerPCCPU *cpu = POWERPC_CPU(cs);
3415     CPUPPCState *env = &cpu->env;
3416 
3417     cpu_synchronize_state(cs);
3418     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3419     if (spapr->fwnmi_system_reset_addr != -1) {
3420         uint64_t rtas_addr, addr;
3421 
3422         /* get rtas addr from fdt */
3423         rtas_addr = spapr_get_rtas_addr();
3424         if (!rtas_addr) {
3425             qemu_system_guest_panicked(NULL);
3426             return;
3427         }
3428 
3429         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3430         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3431         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3432         env->gpr[3] = addr;
3433     }
3434     ppc_cpu_do_system_reset(cs);
3435     if (spapr->fwnmi_system_reset_addr != -1) {
3436         env->nip = spapr->fwnmi_system_reset_addr;
3437     }
3438 }
3439 
3440 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3441 {
3442     CPUState *cs;
3443 
3444     CPU_FOREACH(cs) {
3445         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3446     }
3447 }
3448 
3449 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3450                           void *fdt, int *fdt_start_offset, Error **errp)
3451 {
3452     uint64_t addr;
3453     uint32_t node;
3454 
3455     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3456     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3457                                     &error_abort);
3458     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3459                                              SPAPR_MEMORY_BLOCK_SIZE);
3460     return 0;
3461 }
3462 
3463 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3464                            bool dedicated_hp_event_source)
3465 {
3466     SpaprDrc *drc;
3467     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3468     int i;
3469     uint64_t addr = addr_start;
3470     bool hotplugged = spapr_drc_hotplugged(dev);
3471 
3472     for (i = 0; i < nr_lmbs; i++) {
3473         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3474                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3475         g_assert(drc);
3476 
3477         /*
3478          * memory_device_get_free_addr() provided a range of free addresses
3479          * that doesn't overlap with any existing mapping at pre-plug. The
3480          * corresponding LMB DRCs are thus assumed to be all attachable.
3481          */
3482         spapr_drc_attach(drc, dev);
3483         if (!hotplugged) {
3484             spapr_drc_reset(drc);
3485         }
3486         addr += SPAPR_MEMORY_BLOCK_SIZE;
3487     }
3488     /* send hotplug notification to the
3489      * guest only in case of hotplugged memory
3490      */
3491     if (hotplugged) {
3492         if (dedicated_hp_event_source) {
3493             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3494                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3495             g_assert(drc);
3496             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3497                                                    nr_lmbs,
3498                                                    spapr_drc_index(drc));
3499         } else {
3500             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3501                                            nr_lmbs);
3502         }
3503     }
3504 }
3505 
3506 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3507 {
3508     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3509     PCDIMMDevice *dimm = PC_DIMM(dev);
3510     uint64_t size, addr;
3511     int64_t slot;
3512     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3513 
3514     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3515 
3516     pc_dimm_plug(dimm, MACHINE(ms));
3517 
3518     if (!is_nvdimm) {
3519         addr = object_property_get_uint(OBJECT(dimm),
3520                                         PC_DIMM_ADDR_PROP, &error_abort);
3521         spapr_add_lmbs(dev, addr, size,
3522                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT));
3523     } else {
3524         slot = object_property_get_int(OBJECT(dimm),
3525                                        PC_DIMM_SLOT_PROP, &error_abort);
3526         /* We should have valid slot number at this point */
3527         g_assert(slot >= 0);
3528         spapr_add_nvdimm(dev, slot);
3529     }
3530 }
3531 
3532 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3533                                   Error **errp)
3534 {
3535     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3536     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3537     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3538     PCDIMMDevice *dimm = PC_DIMM(dev);
3539     Error *local_err = NULL;
3540     uint64_t size;
3541     Object *memdev;
3542     hwaddr pagesize;
3543 
3544     if (!smc->dr_lmb_enabled) {
3545         error_setg(errp, "Memory hotplug not supported for this machine");
3546         return;
3547     }
3548 
3549     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3550     if (local_err) {
3551         error_propagate(errp, local_err);
3552         return;
3553     }
3554 
3555     if (is_nvdimm) {
3556         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3557             return;
3558         }
3559     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3560         error_setg(errp, "Hotplugged memory size must be a multiple of "
3561                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3562         return;
3563     }
3564 
3565     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3566                                       &error_abort);
3567     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3568     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3569         return;
3570     }
3571 
3572     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3573 }
3574 
3575 struct SpaprDimmState {
3576     PCDIMMDevice *dimm;
3577     uint32_t nr_lmbs;
3578     QTAILQ_ENTRY(SpaprDimmState) next;
3579 };
3580 
3581 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3582                                                        PCDIMMDevice *dimm)
3583 {
3584     SpaprDimmState *dimm_state = NULL;
3585 
3586     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3587         if (dimm_state->dimm == dimm) {
3588             break;
3589         }
3590     }
3591     return dimm_state;
3592 }
3593 
3594 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3595                                                       uint32_t nr_lmbs,
3596                                                       PCDIMMDevice *dimm)
3597 {
3598     SpaprDimmState *ds = NULL;
3599 
3600     /*
3601      * If this request is for a DIMM whose removal had failed earlier
3602      * (due to guest's refusal to remove the LMBs), we would have this
3603      * dimm already in the pending_dimm_unplugs list. In that
3604      * case don't add again.
3605      */
3606     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3607     if (!ds) {
3608         ds = g_malloc0(sizeof(SpaprDimmState));
3609         ds->nr_lmbs = nr_lmbs;
3610         ds->dimm = dimm;
3611         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3612     }
3613     return ds;
3614 }
3615 
3616 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3617                                               SpaprDimmState *dimm_state)
3618 {
3619     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3620     g_free(dimm_state);
3621 }
3622 
3623 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3624                                                         PCDIMMDevice *dimm)
3625 {
3626     SpaprDrc *drc;
3627     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3628                                                   &error_abort);
3629     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3630     uint32_t avail_lmbs = 0;
3631     uint64_t addr_start, addr;
3632     int i;
3633 
3634     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3635                                           &error_abort);
3636 
3637     addr = addr_start;
3638     for (i = 0; i < nr_lmbs; i++) {
3639         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3640                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3641         g_assert(drc);
3642         if (drc->dev) {
3643             avail_lmbs++;
3644         }
3645         addr += SPAPR_MEMORY_BLOCK_SIZE;
3646     }
3647 
3648     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3649 }
3650 
3651 void spapr_memory_unplug_rollback(SpaprMachineState *spapr, DeviceState *dev)
3652 {
3653     SpaprDimmState *ds;
3654     PCDIMMDevice *dimm;
3655     SpaprDrc *drc;
3656     uint32_t nr_lmbs;
3657     uint64_t size, addr_start, addr;
3658     g_autofree char *qapi_error = NULL;
3659     int i;
3660 
3661     if (!dev) {
3662         return;
3663     }
3664 
3665     dimm = PC_DIMM(dev);
3666     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3667 
3668     /*
3669      * 'ds == NULL' would mean that the DIMM doesn't have a pending
3670      * unplug state, but one of its DRC is marked as unplug_requested.
3671      * This is bad and weird enough to g_assert() out.
3672      */
3673     g_assert(ds);
3674 
3675     spapr_pending_dimm_unplugs_remove(spapr, ds);
3676 
3677     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3678     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3679 
3680     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3681                                           &error_abort);
3682 
3683     addr = addr_start;
3684     for (i = 0; i < nr_lmbs; i++) {
3685         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3686                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3687         g_assert(drc);
3688 
3689         drc->unplug_requested = false;
3690         addr += SPAPR_MEMORY_BLOCK_SIZE;
3691     }
3692 
3693     /*
3694      * Tell QAPI that something happened and the memory
3695      * hotunplug wasn't successful. Keep sending
3696      * MEM_UNPLUG_ERROR even while sending
3697      * DEVICE_UNPLUG_GUEST_ERROR until the deprecation of
3698      * MEM_UNPLUG_ERROR is due.
3699      */
3700     qapi_error = g_strdup_printf("Memory hotunplug rejected by the guest "
3701                                  "for device %s", dev->id);
3702 
3703     qapi_event_send_mem_unplug_error(dev->id ? : "", qapi_error);
3704 
3705     qapi_event_send_device_unplug_guest_error(!!dev->id, dev->id,
3706                                               dev->canonical_path);
3707 }
3708 
3709 /* Callback to be called during DRC release. */
3710 void spapr_lmb_release(DeviceState *dev)
3711 {
3712     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3713     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3714     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3715 
3716     /* This information will get lost if a migration occurs
3717      * during the unplug process. In this case recover it. */
3718     if (ds == NULL) {
3719         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3720         g_assert(ds);
3721         /* The DRC being examined by the caller at least must be counted */
3722         g_assert(ds->nr_lmbs);
3723     }
3724 
3725     if (--ds->nr_lmbs) {
3726         return;
3727     }
3728 
3729     /*
3730      * Now that all the LMBs have been removed by the guest, call the
3731      * unplug handler chain. This can never fail.
3732      */
3733     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3734     object_unparent(OBJECT(dev));
3735 }
3736 
3737 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3738 {
3739     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3740     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3741 
3742     /* We really shouldn't get this far without anything to unplug */
3743     g_assert(ds);
3744 
3745     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3746     qdev_unrealize(dev);
3747     spapr_pending_dimm_unplugs_remove(spapr, ds);
3748 }
3749 
3750 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3751                                         DeviceState *dev, Error **errp)
3752 {
3753     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3754     PCDIMMDevice *dimm = PC_DIMM(dev);
3755     uint32_t nr_lmbs;
3756     uint64_t size, addr_start, addr;
3757     int i;
3758     SpaprDrc *drc;
3759 
3760     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3761         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3762         return;
3763     }
3764 
3765     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3766     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3767 
3768     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3769                                           &error_abort);
3770 
3771     /*
3772      * An existing pending dimm state for this DIMM means that there is an
3773      * unplug operation in progress, waiting for the spapr_lmb_release
3774      * callback to complete the job (BQL can't cover that far). In this case,
3775      * bail out to avoid detaching DRCs that were already released.
3776      */
3777     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3778         error_setg(errp, "Memory unplug already in progress for device %s",
3779                    dev->id);
3780         return;
3781     }
3782 
3783     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3784 
3785     addr = addr_start;
3786     for (i = 0; i < nr_lmbs; i++) {
3787         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3788                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3789         g_assert(drc);
3790 
3791         spapr_drc_unplug_request(drc);
3792         addr += SPAPR_MEMORY_BLOCK_SIZE;
3793     }
3794 
3795     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3796                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3797     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3798                                               nr_lmbs, spapr_drc_index(drc));
3799 }
3800 
3801 /* Callback to be called during DRC release. */
3802 void spapr_core_release(DeviceState *dev)
3803 {
3804     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3805 
3806     /* Call the unplug handler chain. This can never fail. */
3807     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3808     object_unparent(OBJECT(dev));
3809 }
3810 
3811 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3812 {
3813     MachineState *ms = MACHINE(hotplug_dev);
3814     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3815     CPUCore *cc = CPU_CORE(dev);
3816     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3817 
3818     if (smc->pre_2_10_has_unused_icps) {
3819         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3820         int i;
3821 
3822         for (i = 0; i < cc->nr_threads; i++) {
3823             CPUState *cs = CPU(sc->threads[i]);
3824 
3825             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3826         }
3827     }
3828 
3829     assert(core_slot);
3830     core_slot->cpu = NULL;
3831     qdev_unrealize(dev);
3832 }
3833 
3834 static
3835 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3836                                Error **errp)
3837 {
3838     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3839     int index;
3840     SpaprDrc *drc;
3841     CPUCore *cc = CPU_CORE(dev);
3842 
3843     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3844         error_setg(errp, "Unable to find CPU core with core-id: %d",
3845                    cc->core_id);
3846         return;
3847     }
3848     if (index == 0) {
3849         error_setg(errp, "Boot CPU core may not be unplugged");
3850         return;
3851     }
3852 
3853     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3854                           spapr_vcpu_id(spapr, cc->core_id));
3855     g_assert(drc);
3856 
3857     if (!spapr_drc_unplug_requested(drc)) {
3858         spapr_drc_unplug_request(drc);
3859     }
3860 
3861     /*
3862      * spapr_hotplug_req_remove_by_index is left unguarded, out of the
3863      * "!spapr_drc_unplug_requested" check, to allow for multiple IRQ
3864      * pulses removing the same CPU. Otherwise, in an failed hotunplug
3865      * attempt (e.g. the kernel will refuse to remove the last online
3866      * CPU), we will never attempt it again because unplug_requested
3867      * will still be 'true' in that case.
3868      */
3869     spapr_hotplug_req_remove_by_index(drc);
3870 }
3871 
3872 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3873                            void *fdt, int *fdt_start_offset, Error **errp)
3874 {
3875     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3876     CPUState *cs = CPU(core->threads[0]);
3877     PowerPCCPU *cpu = POWERPC_CPU(cs);
3878     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3879     int id = spapr_get_vcpu_id(cpu);
3880     g_autofree char *nodename = NULL;
3881     int offset;
3882 
3883     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3884     offset = fdt_add_subnode(fdt, 0, nodename);
3885 
3886     spapr_dt_cpu(cs, fdt, offset, spapr);
3887 
3888     /*
3889      * spapr_dt_cpu() does not fill the 'name' property in the
3890      * CPU node. The function is called during boot process, before
3891      * and after CAS, and overwriting the 'name' property written
3892      * by SLOF is not allowed.
3893      *
3894      * Write it manually after spapr_dt_cpu(). This makes the hotplug
3895      * CPUs more compatible with the coldplugged ones, which have
3896      * the 'name' property. Linux Kernel also relies on this
3897      * property to identify CPU nodes.
3898      */
3899     _FDT((fdt_setprop_string(fdt, offset, "name", nodename)));
3900 
3901     *fdt_start_offset = offset;
3902     return 0;
3903 }
3904 
3905 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3906 {
3907     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3908     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3909     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3910     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3911     CPUCore *cc = CPU_CORE(dev);
3912     CPUState *cs;
3913     SpaprDrc *drc;
3914     CPUArchId *core_slot;
3915     int index;
3916     bool hotplugged = spapr_drc_hotplugged(dev);
3917     int i;
3918 
3919     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3920     g_assert(core_slot); /* Already checked in spapr_core_pre_plug() */
3921 
3922     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3923                           spapr_vcpu_id(spapr, cc->core_id));
3924 
3925     g_assert(drc || !mc->has_hotpluggable_cpus);
3926 
3927     if (drc) {
3928         /*
3929          * spapr_core_pre_plug() already buys us this is a brand new
3930          * core being plugged into a free slot. Nothing should already
3931          * be attached to the corresponding DRC.
3932          */
3933         spapr_drc_attach(drc, dev);
3934 
3935         if (hotplugged) {
3936             /*
3937              * Send hotplug notification interrupt to the guest only
3938              * in case of hotplugged CPUs.
3939              */
3940             spapr_hotplug_req_add_by_index(drc);
3941         } else {
3942             spapr_drc_reset(drc);
3943         }
3944     }
3945 
3946     core_slot->cpu = OBJECT(dev);
3947 
3948     /*
3949      * Set compatibility mode to match the boot CPU, which was either set
3950      * by the machine reset code or by CAS. This really shouldn't fail at
3951      * this point.
3952      */
3953     if (hotplugged) {
3954         for (i = 0; i < cc->nr_threads; i++) {
3955             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3956                            &error_abort);
3957         }
3958     }
3959 
3960     if (smc->pre_2_10_has_unused_icps) {
3961         for (i = 0; i < cc->nr_threads; i++) {
3962             cs = CPU(core->threads[i]);
3963             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3964         }
3965     }
3966 }
3967 
3968 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3969                                 Error **errp)
3970 {
3971     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3972     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3973     CPUCore *cc = CPU_CORE(dev);
3974     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3975     const char *type = object_get_typename(OBJECT(dev));
3976     CPUArchId *core_slot;
3977     int index;
3978     unsigned int smp_threads = machine->smp.threads;
3979 
3980     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3981         error_setg(errp, "CPU hotplug not supported for this machine");
3982         return;
3983     }
3984 
3985     if (strcmp(base_core_type, type)) {
3986         error_setg(errp, "CPU core type should be %s", base_core_type);
3987         return;
3988     }
3989 
3990     if (cc->core_id % smp_threads) {
3991         error_setg(errp, "invalid core id %d", cc->core_id);
3992         return;
3993     }
3994 
3995     /*
3996      * In general we should have homogeneous threads-per-core, but old
3997      * (pre hotplug support) machine types allow the last core to have
3998      * reduced threads as a compatibility hack for when we allowed
3999      * total vcpus not a multiple of threads-per-core.
4000      */
4001     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
4002         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
4003                    smp_threads);
4004         return;
4005     }
4006 
4007     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
4008     if (!core_slot) {
4009         error_setg(errp, "core id %d out of range", cc->core_id);
4010         return;
4011     }
4012 
4013     if (core_slot->cpu) {
4014         error_setg(errp, "core %d already populated", cc->core_id);
4015         return;
4016     }
4017 
4018     numa_cpu_pre_plug(core_slot, dev, errp);
4019 }
4020 
4021 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
4022                           void *fdt, int *fdt_start_offset, Error **errp)
4023 {
4024     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
4025     int intc_phandle;
4026 
4027     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
4028     if (intc_phandle <= 0) {
4029         return -1;
4030     }
4031 
4032     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
4033         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
4034         return -1;
4035     }
4036 
4037     /* generally SLOF creates these, for hotplug it's up to QEMU */
4038     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
4039 
4040     return 0;
4041 }
4042 
4043 static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4044                                Error **errp)
4045 {
4046     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4047     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4048     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4049     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
4050     SpaprDrc *drc;
4051 
4052     if (dev->hotplugged && !smc->dr_phb_enabled) {
4053         error_setg(errp, "PHB hotplug not supported for this machine");
4054         return false;
4055     }
4056 
4057     if (sphb->index == (uint32_t)-1) {
4058         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
4059         return false;
4060     }
4061 
4062     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4063     if (drc && drc->dev) {
4064         error_setg(errp, "PHB %d already attached", sphb->index);
4065         return false;
4066     }
4067 
4068     /*
4069      * This will check that sphb->index doesn't exceed the maximum number of
4070      * PHBs for the current machine type.
4071      */
4072     return
4073         smc->phb_placement(spapr, sphb->index,
4074                            &sphb->buid, &sphb->io_win_addr,
4075                            &sphb->mem_win_addr, &sphb->mem64_win_addr,
4076                            windows_supported, sphb->dma_liobn,
4077                            &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
4078                            errp);
4079 }
4080 
4081 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4082 {
4083     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4084     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4085     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4086     SpaprDrc *drc;
4087     bool hotplugged = spapr_drc_hotplugged(dev);
4088 
4089     if (!smc->dr_phb_enabled) {
4090         return;
4091     }
4092 
4093     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4094     /* hotplug hooks should check it's enabled before getting this far */
4095     assert(drc);
4096 
4097     /* spapr_phb_pre_plug() already checked the DRC is attachable */
4098     spapr_drc_attach(drc, dev);
4099 
4100     if (hotplugged) {
4101         spapr_hotplug_req_add_by_index(drc);
4102     } else {
4103         spapr_drc_reset(drc);
4104     }
4105 }
4106 
4107 void spapr_phb_release(DeviceState *dev)
4108 {
4109     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4110 
4111     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4112     object_unparent(OBJECT(dev));
4113 }
4114 
4115 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4116 {
4117     qdev_unrealize(dev);
4118 }
4119 
4120 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4121                                      DeviceState *dev, Error **errp)
4122 {
4123     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4124     SpaprDrc *drc;
4125 
4126     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4127     assert(drc);
4128 
4129     if (!spapr_drc_unplug_requested(drc)) {
4130         spapr_drc_unplug_request(drc);
4131         spapr_hotplug_req_remove_by_index(drc);
4132     } else {
4133         error_setg(errp,
4134                    "PCI Host Bridge unplug already in progress for device %s",
4135                    dev->id);
4136     }
4137 }
4138 
4139 static
4140 bool spapr_tpm_proxy_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4141                               Error **errp)
4142 {
4143     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4144 
4145     if (spapr->tpm_proxy != NULL) {
4146         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4147         return false;
4148     }
4149 
4150     return true;
4151 }
4152 
4153 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4154 {
4155     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4156     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4157 
4158     /* Already checked in spapr_tpm_proxy_pre_plug() */
4159     g_assert(spapr->tpm_proxy == NULL);
4160 
4161     spapr->tpm_proxy = tpm_proxy;
4162 }
4163 
4164 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4165 {
4166     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4167 
4168     qdev_unrealize(dev);
4169     object_unparent(OBJECT(dev));
4170     spapr->tpm_proxy = NULL;
4171 }
4172 
4173 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4174                                       DeviceState *dev, Error **errp)
4175 {
4176     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4177         spapr_memory_plug(hotplug_dev, dev);
4178     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4179         spapr_core_plug(hotplug_dev, dev);
4180     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4181         spapr_phb_plug(hotplug_dev, dev);
4182     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4183         spapr_tpm_proxy_plug(hotplug_dev, dev);
4184     }
4185 }
4186 
4187 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4188                                         DeviceState *dev, Error **errp)
4189 {
4190     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4191         spapr_memory_unplug(hotplug_dev, dev);
4192     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4193         spapr_core_unplug(hotplug_dev, dev);
4194     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4195         spapr_phb_unplug(hotplug_dev, dev);
4196     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4197         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4198     }
4199 }
4200 
4201 bool spapr_memory_hot_unplug_supported(SpaprMachineState *spapr)
4202 {
4203     return spapr_ovec_test(spapr->ov5_cas, OV5_HP_EVT) ||
4204         /*
4205          * CAS will process all pending unplug requests.
4206          *
4207          * HACK: a guest could theoretically have cleared all bits in OV5,
4208          * but none of the guests we care for do.
4209          */
4210         spapr_ovec_empty(spapr->ov5_cas);
4211 }
4212 
4213 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4214                                                 DeviceState *dev, Error **errp)
4215 {
4216     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4217     MachineClass *mc = MACHINE_GET_CLASS(sms);
4218     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4219 
4220     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4221         if (spapr_memory_hot_unplug_supported(sms)) {
4222             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4223         } else {
4224             error_setg(errp, "Memory hot unplug not supported for this guest");
4225         }
4226     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4227         if (!mc->has_hotpluggable_cpus) {
4228             error_setg(errp, "CPU hot unplug not supported on this machine");
4229             return;
4230         }
4231         spapr_core_unplug_request(hotplug_dev, dev, errp);
4232     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4233         if (!smc->dr_phb_enabled) {
4234             error_setg(errp, "PHB hot unplug not supported on this machine");
4235             return;
4236         }
4237         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4238     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4239         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4240     }
4241 }
4242 
4243 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4244                                           DeviceState *dev, Error **errp)
4245 {
4246     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4247         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4248     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4249         spapr_core_pre_plug(hotplug_dev, dev, errp);
4250     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4251         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4252     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4253         spapr_tpm_proxy_pre_plug(hotplug_dev, dev, errp);
4254     }
4255 }
4256 
4257 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4258                                                  DeviceState *dev)
4259 {
4260     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4261         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4262         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4263         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4264         return HOTPLUG_HANDLER(machine);
4265     }
4266     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4267         PCIDevice *pcidev = PCI_DEVICE(dev);
4268         PCIBus *root = pci_device_root_bus(pcidev);
4269         SpaprPhbState *phb =
4270             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4271                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4272 
4273         if (phb) {
4274             return HOTPLUG_HANDLER(phb);
4275         }
4276     }
4277     return NULL;
4278 }
4279 
4280 static CpuInstanceProperties
4281 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4282 {
4283     CPUArchId *core_slot;
4284     MachineClass *mc = MACHINE_GET_CLASS(machine);
4285 
4286     /* make sure possible_cpu are intialized */
4287     mc->possible_cpu_arch_ids(machine);
4288     /* get CPU core slot containing thread that matches cpu_index */
4289     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4290     assert(core_slot);
4291     return core_slot->props;
4292 }
4293 
4294 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4295 {
4296     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4297 }
4298 
4299 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4300 {
4301     int i;
4302     unsigned int smp_threads = machine->smp.threads;
4303     unsigned int smp_cpus = machine->smp.cpus;
4304     const char *core_type;
4305     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4306     MachineClass *mc = MACHINE_GET_CLASS(machine);
4307 
4308     if (!mc->has_hotpluggable_cpus) {
4309         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4310     }
4311     if (machine->possible_cpus) {
4312         assert(machine->possible_cpus->len == spapr_max_cores);
4313         return machine->possible_cpus;
4314     }
4315 
4316     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4317     if (!core_type) {
4318         error_report("Unable to find sPAPR CPU Core definition");
4319         exit(1);
4320     }
4321 
4322     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4323                              sizeof(CPUArchId) * spapr_max_cores);
4324     machine->possible_cpus->len = spapr_max_cores;
4325     for (i = 0; i < machine->possible_cpus->len; i++) {
4326         int core_id = i * smp_threads;
4327 
4328         machine->possible_cpus->cpus[i].type = core_type;
4329         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4330         machine->possible_cpus->cpus[i].arch_id = core_id;
4331         machine->possible_cpus->cpus[i].props.has_core_id = true;
4332         machine->possible_cpus->cpus[i].props.core_id = core_id;
4333     }
4334     return machine->possible_cpus;
4335 }
4336 
4337 static bool spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4338                                 uint64_t *buid, hwaddr *pio,
4339                                 hwaddr *mmio32, hwaddr *mmio64,
4340                                 unsigned n_dma, uint32_t *liobns,
4341                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4342 {
4343     /*
4344      * New-style PHB window placement.
4345      *
4346      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4347      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4348      * windows.
4349      *
4350      * Some guest kernels can't work with MMIO windows above 1<<46
4351      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4352      *
4353      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4354      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4355      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4356      * 1TiB 64-bit MMIO windows for each PHB.
4357      */
4358     const uint64_t base_buid = 0x800000020000000ULL;
4359     int i;
4360 
4361     /* Sanity check natural alignments */
4362     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4363     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4364     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4365     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4366     /* Sanity check bounds */
4367     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4368                       SPAPR_PCI_MEM32_WIN_SIZE);
4369     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4370                       SPAPR_PCI_MEM64_WIN_SIZE);
4371 
4372     if (index >= SPAPR_MAX_PHBS) {
4373         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4374                    SPAPR_MAX_PHBS - 1);
4375         return false;
4376     }
4377 
4378     *buid = base_buid + index;
4379     for (i = 0; i < n_dma; ++i) {
4380         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4381     }
4382 
4383     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4384     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4385     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4386 
4387     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4388     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4389     return true;
4390 }
4391 
4392 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4393 {
4394     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4395 
4396     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4397 }
4398 
4399 static void spapr_ics_resend(XICSFabric *dev)
4400 {
4401     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4402 
4403     ics_resend(spapr->ics);
4404 }
4405 
4406 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4407 {
4408     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4409 
4410     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4411 }
4412 
4413 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4414                                  Monitor *mon)
4415 {
4416     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4417 
4418     spapr_irq_print_info(spapr, mon);
4419     monitor_printf(mon, "irqchip: %s\n",
4420                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4421 }
4422 
4423 /*
4424  * This is a XIVE only operation
4425  */
4426 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4427                            uint8_t nvt_blk, uint32_t nvt_idx,
4428                            bool cam_ignore, uint8_t priority,
4429                            uint32_t logic_serv, XiveTCTXMatch *match)
4430 {
4431     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4432     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4433     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4434     int count;
4435 
4436     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4437                            priority, logic_serv, match);
4438     if (count < 0) {
4439         return count;
4440     }
4441 
4442     /*
4443      * When we implement the save and restore of the thread interrupt
4444      * contexts in the enter/exit CPU handlers of the machine and the
4445      * escalations in QEMU, we should be able to handle non dispatched
4446      * vCPUs.
4447      *
4448      * Until this is done, the sPAPR machine should find at least one
4449      * matching context always.
4450      */
4451     if (count == 0) {
4452         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4453                       nvt_blk, nvt_idx);
4454     }
4455 
4456     return count;
4457 }
4458 
4459 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4460 {
4461     return cpu->vcpu_id;
4462 }
4463 
4464 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4465 {
4466     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4467     MachineState *ms = MACHINE(spapr);
4468     int vcpu_id;
4469 
4470     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4471 
4472     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4473         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4474         error_append_hint(errp, "Adjust the number of cpus to %d "
4475                           "or try to raise the number of threads per core\n",
4476                           vcpu_id * ms->smp.threads / spapr->vsmt);
4477         return false;
4478     }
4479 
4480     cpu->vcpu_id = vcpu_id;
4481     return true;
4482 }
4483 
4484 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4485 {
4486     CPUState *cs;
4487 
4488     CPU_FOREACH(cs) {
4489         PowerPCCPU *cpu = POWERPC_CPU(cs);
4490 
4491         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4492             return cpu;
4493         }
4494     }
4495 
4496     return NULL;
4497 }
4498 
4499 static bool spapr_cpu_in_nested(PowerPCCPU *cpu)
4500 {
4501     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4502 
4503     return spapr_cpu->in_nested;
4504 }
4505 
4506 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4507 {
4508     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4509 
4510     /* These are only called by TCG, KVM maintains dispatch state */
4511 
4512     spapr_cpu->prod = false;
4513     if (spapr_cpu->vpa_addr) {
4514         CPUState *cs = CPU(cpu);
4515         uint32_t dispatch;
4516 
4517         dispatch = ldl_be_phys(cs->as,
4518                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4519         dispatch++;
4520         if ((dispatch & 1) != 0) {
4521             qemu_log_mask(LOG_GUEST_ERROR,
4522                           "VPA: incorrect dispatch counter value for "
4523                           "dispatched partition %u, correcting.\n", dispatch);
4524             dispatch++;
4525         }
4526         stl_be_phys(cs->as,
4527                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4528     }
4529 }
4530 
4531 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4532 {
4533     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4534 
4535     if (spapr_cpu->vpa_addr) {
4536         CPUState *cs = CPU(cpu);
4537         uint32_t dispatch;
4538 
4539         dispatch = ldl_be_phys(cs->as,
4540                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4541         dispatch++;
4542         if ((dispatch & 1) != 1) {
4543             qemu_log_mask(LOG_GUEST_ERROR,
4544                           "VPA: incorrect dispatch counter value for "
4545                           "preempted partition %u, correcting.\n", dispatch);
4546             dispatch++;
4547         }
4548         stl_be_phys(cs->as,
4549                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4550     }
4551 }
4552 
4553 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4554 {
4555     MachineClass *mc = MACHINE_CLASS(oc);
4556     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4557     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4558     NMIClass *nc = NMI_CLASS(oc);
4559     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4560     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4561     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4562     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4563     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4564     VofMachineIfClass *vmc = VOF_MACHINE_CLASS(oc);
4565 
4566     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4567     mc->ignore_boot_device_suffixes = true;
4568 
4569     /*
4570      * We set up the default / latest behaviour here.  The class_init
4571      * functions for the specific versioned machine types can override
4572      * these details for backwards compatibility
4573      */
4574     mc->init = spapr_machine_init;
4575     mc->reset = spapr_machine_reset;
4576     mc->block_default_type = IF_SCSI;
4577 
4578     /*
4579      * Setting max_cpus to INT32_MAX. Both KVM and TCG max_cpus values
4580      * should be limited by the host capability instead of hardcoded.
4581      * max_cpus for KVM guests will be checked in kvm_init(), and TCG
4582      * guests are welcome to have as many CPUs as the host are capable
4583      * of emulate.
4584      */
4585     mc->max_cpus = INT32_MAX;
4586 
4587     mc->no_parallel = 1;
4588     mc->default_boot_order = "";
4589     mc->default_ram_size = 512 * MiB;
4590     mc->default_ram_id = "ppc_spapr.ram";
4591     mc->default_display = "std";
4592     mc->kvm_type = spapr_kvm_type;
4593     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4594     mc->pci_allow_0_address = true;
4595     assert(!mc->get_hotplug_handler);
4596     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4597     hc->pre_plug = spapr_machine_device_pre_plug;
4598     hc->plug = spapr_machine_device_plug;
4599     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4600     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4601     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4602     hc->unplug_request = spapr_machine_device_unplug_request;
4603     hc->unplug = spapr_machine_device_unplug;
4604 
4605     smc->dr_lmb_enabled = true;
4606     smc->update_dt_enabled = true;
4607     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4608     mc->has_hotpluggable_cpus = true;
4609     mc->nvdimm_supported = true;
4610     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4611     fwc->get_dev_path = spapr_get_fw_dev_path;
4612     nc->nmi_monitor_handler = spapr_nmi;
4613     smc->phb_placement = spapr_phb_placement;
4614     vhc->cpu_in_nested = spapr_cpu_in_nested;
4615     vhc->deliver_hv_excp = spapr_exit_nested;
4616     vhc->hypercall = emulate_spapr_hypercall;
4617     vhc->hpt_mask = spapr_hpt_mask;
4618     vhc->map_hptes = spapr_map_hptes;
4619     vhc->unmap_hptes = spapr_unmap_hptes;
4620     vhc->hpte_set_c = spapr_hpte_set_c;
4621     vhc->hpte_set_r = spapr_hpte_set_r;
4622     vhc->get_pate = spapr_get_pate;
4623     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4624     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4625     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4626     xic->ics_get = spapr_ics_get;
4627     xic->ics_resend = spapr_ics_resend;
4628     xic->icp_get = spapr_icp_get;
4629     ispc->print_info = spapr_pic_print_info;
4630     /* Force NUMA node memory size to be a multiple of
4631      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4632      * in which LMBs are represented and hot-added
4633      */
4634     mc->numa_mem_align_shift = 28;
4635     mc->auto_enable_numa = true;
4636 
4637     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4638     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4639     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4640     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4641     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4642     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4643     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4644     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4645     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4646     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4647     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4648     smc->default_caps.caps[SPAPR_CAP_RPT_INVALIDATE] = SPAPR_CAP_OFF;
4649     spapr_caps_add_properties(smc);
4650     smc->irq = &spapr_irq_dual;
4651     smc->dr_phb_enabled = true;
4652     smc->linux_pci_probe = true;
4653     smc->smp_threads_vsmt = true;
4654     smc->nr_xirqs = SPAPR_NR_XIRQS;
4655     xfc->match_nvt = spapr_match_nvt;
4656     vmc->client_architecture_support = spapr_vof_client_architecture_support;
4657     vmc->quiesce = spapr_vof_quiesce;
4658     vmc->setprop = spapr_vof_setprop;
4659 }
4660 
4661 static const TypeInfo spapr_machine_info = {
4662     .name          = TYPE_SPAPR_MACHINE,
4663     .parent        = TYPE_MACHINE,
4664     .abstract      = true,
4665     .instance_size = sizeof(SpaprMachineState),
4666     .instance_init = spapr_instance_init,
4667     .instance_finalize = spapr_machine_finalizefn,
4668     .class_size    = sizeof(SpaprMachineClass),
4669     .class_init    = spapr_machine_class_init,
4670     .interfaces = (InterfaceInfo[]) {
4671         { TYPE_FW_PATH_PROVIDER },
4672         { TYPE_NMI },
4673         { TYPE_HOTPLUG_HANDLER },
4674         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4675         { TYPE_XICS_FABRIC },
4676         { TYPE_INTERRUPT_STATS_PROVIDER },
4677         { TYPE_XIVE_FABRIC },
4678         { TYPE_VOF_MACHINE_IF },
4679         { }
4680     },
4681 };
4682 
4683 static void spapr_machine_latest_class_options(MachineClass *mc)
4684 {
4685     mc->alias = "pseries";
4686     mc->is_default = true;
4687 }
4688 
4689 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4690     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4691                                                     void *data)      \
4692     {                                                                \
4693         MachineClass *mc = MACHINE_CLASS(oc);                        \
4694         spapr_machine_##suffix##_class_options(mc);                  \
4695         if (latest) {                                                \
4696             spapr_machine_latest_class_options(mc);                  \
4697         }                                                            \
4698     }                                                                \
4699     static const TypeInfo spapr_machine_##suffix##_info = {          \
4700         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4701         .parent = TYPE_SPAPR_MACHINE,                                \
4702         .class_init = spapr_machine_##suffix##_class_init,           \
4703     };                                                               \
4704     static void spapr_machine_register_##suffix(void)                \
4705     {                                                                \
4706         type_register(&spapr_machine_##suffix##_info);               \
4707     }                                                                \
4708     type_init(spapr_machine_register_##suffix)
4709 
4710 /*
4711  * pseries-7.0
4712  */
4713 static void spapr_machine_7_0_class_options(MachineClass *mc)
4714 {
4715     /* Defaults for the latest behaviour inherited from the base class */
4716 }
4717 
4718 DEFINE_SPAPR_MACHINE(7_0, "7.0", true);
4719 
4720 /*
4721  * pseries-6.2
4722  */
4723 static void spapr_machine_6_2_class_options(MachineClass *mc)
4724 {
4725     spapr_machine_7_0_class_options(mc);
4726     compat_props_add(mc->compat_props, hw_compat_6_2, hw_compat_6_2_len);
4727 }
4728 
4729 DEFINE_SPAPR_MACHINE(6_2, "6.2", false);
4730 
4731 /*
4732  * pseries-6.1
4733  */
4734 static void spapr_machine_6_1_class_options(MachineClass *mc)
4735 {
4736     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4737 
4738     spapr_machine_6_2_class_options(mc);
4739     compat_props_add(mc->compat_props, hw_compat_6_1, hw_compat_6_1_len);
4740     smc->pre_6_2_numa_affinity = true;
4741     mc->smp_props.prefer_sockets = true;
4742 }
4743 
4744 DEFINE_SPAPR_MACHINE(6_1, "6.1", false);
4745 
4746 /*
4747  * pseries-6.0
4748  */
4749 static void spapr_machine_6_0_class_options(MachineClass *mc)
4750 {
4751     spapr_machine_6_1_class_options(mc);
4752     compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len);
4753 }
4754 
4755 DEFINE_SPAPR_MACHINE(6_0, "6.0", false);
4756 
4757 /*
4758  * pseries-5.2
4759  */
4760 static void spapr_machine_5_2_class_options(MachineClass *mc)
4761 {
4762     spapr_machine_6_0_class_options(mc);
4763     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
4764 }
4765 
4766 DEFINE_SPAPR_MACHINE(5_2, "5.2", false);
4767 
4768 /*
4769  * pseries-5.1
4770  */
4771 static void spapr_machine_5_1_class_options(MachineClass *mc)
4772 {
4773     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4774 
4775     spapr_machine_5_2_class_options(mc);
4776     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4777     smc->pre_5_2_numa_associativity = true;
4778 }
4779 
4780 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4781 
4782 /*
4783  * pseries-5.0
4784  */
4785 static void spapr_machine_5_0_class_options(MachineClass *mc)
4786 {
4787     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4788     static GlobalProperty compat[] = {
4789         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4790     };
4791 
4792     spapr_machine_5_1_class_options(mc);
4793     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4794     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4795     mc->numa_mem_supported = true;
4796     smc->pre_5_1_assoc_refpoints = true;
4797 }
4798 
4799 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4800 
4801 /*
4802  * pseries-4.2
4803  */
4804 static void spapr_machine_4_2_class_options(MachineClass *mc)
4805 {
4806     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4807 
4808     spapr_machine_5_0_class_options(mc);
4809     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4810     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4811     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4812     smc->rma_limit = 16 * GiB;
4813     mc->nvdimm_supported = false;
4814 }
4815 
4816 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4817 
4818 /*
4819  * pseries-4.1
4820  */
4821 static void spapr_machine_4_1_class_options(MachineClass *mc)
4822 {
4823     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4824     static GlobalProperty compat[] = {
4825         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4826         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4827     };
4828 
4829     spapr_machine_4_2_class_options(mc);
4830     smc->linux_pci_probe = false;
4831     smc->smp_threads_vsmt = false;
4832     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4833     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4834 }
4835 
4836 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4837 
4838 /*
4839  * pseries-4.0
4840  */
4841 static bool phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4842                               uint64_t *buid, hwaddr *pio,
4843                               hwaddr *mmio32, hwaddr *mmio64,
4844                               unsigned n_dma, uint32_t *liobns,
4845                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4846 {
4847     if (!spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma,
4848                              liobns, nv2gpa, nv2atsd, errp)) {
4849         return false;
4850     }
4851 
4852     *nv2gpa = 0;
4853     *nv2atsd = 0;
4854     return true;
4855 }
4856 static void spapr_machine_4_0_class_options(MachineClass *mc)
4857 {
4858     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4859 
4860     spapr_machine_4_1_class_options(mc);
4861     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4862     smc->phb_placement = phb_placement_4_0;
4863     smc->irq = &spapr_irq_xics;
4864     smc->pre_4_1_migration = true;
4865 }
4866 
4867 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4868 
4869 /*
4870  * pseries-3.1
4871  */
4872 static void spapr_machine_3_1_class_options(MachineClass *mc)
4873 {
4874     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4875 
4876     spapr_machine_4_0_class_options(mc);
4877     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4878 
4879     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4880     smc->update_dt_enabled = false;
4881     smc->dr_phb_enabled = false;
4882     smc->broken_host_serial_model = true;
4883     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4884     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4885     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4886     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4887 }
4888 
4889 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4890 
4891 /*
4892  * pseries-3.0
4893  */
4894 
4895 static void spapr_machine_3_0_class_options(MachineClass *mc)
4896 {
4897     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4898 
4899     spapr_machine_3_1_class_options(mc);
4900     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4901 
4902     smc->legacy_irq_allocation = true;
4903     smc->nr_xirqs = 0x400;
4904     smc->irq = &spapr_irq_xics_legacy;
4905 }
4906 
4907 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4908 
4909 /*
4910  * pseries-2.12
4911  */
4912 static void spapr_machine_2_12_class_options(MachineClass *mc)
4913 {
4914     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4915     static GlobalProperty compat[] = {
4916         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4917         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4918     };
4919 
4920     spapr_machine_3_0_class_options(mc);
4921     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4922     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4923 
4924     /* We depend on kvm_enabled() to choose a default value for the
4925      * hpt-max-page-size capability. Of course we can't do it here
4926      * because this is too early and the HW accelerator isn't initialzed
4927      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4928      */
4929     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4930 }
4931 
4932 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4933 
4934 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4935 {
4936     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4937 
4938     spapr_machine_2_12_class_options(mc);
4939     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4940     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4941     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4942 }
4943 
4944 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4945 
4946 /*
4947  * pseries-2.11
4948  */
4949 
4950 static void spapr_machine_2_11_class_options(MachineClass *mc)
4951 {
4952     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4953 
4954     spapr_machine_2_12_class_options(mc);
4955     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4956     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4957 }
4958 
4959 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4960 
4961 /*
4962  * pseries-2.10
4963  */
4964 
4965 static void spapr_machine_2_10_class_options(MachineClass *mc)
4966 {
4967     spapr_machine_2_11_class_options(mc);
4968     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4969 }
4970 
4971 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4972 
4973 /*
4974  * pseries-2.9
4975  */
4976 
4977 static void spapr_machine_2_9_class_options(MachineClass *mc)
4978 {
4979     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4980     static GlobalProperty compat[] = {
4981         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4982     };
4983 
4984     spapr_machine_2_10_class_options(mc);
4985     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4986     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4987     smc->pre_2_10_has_unused_icps = true;
4988     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4989 }
4990 
4991 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4992 
4993 /*
4994  * pseries-2.8
4995  */
4996 
4997 static void spapr_machine_2_8_class_options(MachineClass *mc)
4998 {
4999     static GlobalProperty compat[] = {
5000         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
5001     };
5002 
5003     spapr_machine_2_9_class_options(mc);
5004     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
5005     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5006     mc->numa_mem_align_shift = 23;
5007 }
5008 
5009 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
5010 
5011 /*
5012  * pseries-2.7
5013  */
5014 
5015 static bool phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
5016                               uint64_t *buid, hwaddr *pio,
5017                               hwaddr *mmio32, hwaddr *mmio64,
5018                               unsigned n_dma, uint32_t *liobns,
5019                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
5020 {
5021     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
5022     const uint64_t base_buid = 0x800000020000000ULL;
5023     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
5024     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
5025     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
5026     const uint32_t max_index = 255;
5027     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
5028 
5029     uint64_t ram_top = MACHINE(spapr)->ram_size;
5030     hwaddr phb0_base, phb_base;
5031     int i;
5032 
5033     /* Do we have device memory? */
5034     if (MACHINE(spapr)->maxram_size > ram_top) {
5035         /* Can't just use maxram_size, because there may be an
5036          * alignment gap between normal and device memory regions
5037          */
5038         ram_top = MACHINE(spapr)->device_memory->base +
5039             memory_region_size(&MACHINE(spapr)->device_memory->mr);
5040     }
5041 
5042     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
5043 
5044     if (index > max_index) {
5045         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
5046                    max_index);
5047         return false;
5048     }
5049 
5050     *buid = base_buid + index;
5051     for (i = 0; i < n_dma; ++i) {
5052         liobns[i] = SPAPR_PCI_LIOBN(index, i);
5053     }
5054 
5055     phb_base = phb0_base + index * phb_spacing;
5056     *pio = phb_base + pio_offset;
5057     *mmio32 = phb_base + mmio_offset;
5058     /*
5059      * We don't set the 64-bit MMIO window, relying on the PHB's
5060      * fallback behaviour of automatically splitting a large "32-bit"
5061      * window into contiguous 32-bit and 64-bit windows
5062      */
5063 
5064     *nv2gpa = 0;
5065     *nv2atsd = 0;
5066     return true;
5067 }
5068 
5069 static void spapr_machine_2_7_class_options(MachineClass *mc)
5070 {
5071     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5072     static GlobalProperty compat[] = {
5073         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
5074         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
5075         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
5076         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
5077     };
5078 
5079     spapr_machine_2_8_class_options(mc);
5080     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
5081     mc->default_machine_opts = "modern-hotplug-events=off";
5082     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
5083     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5084     smc->phb_placement = phb_placement_2_7;
5085 }
5086 
5087 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
5088 
5089 /*
5090  * pseries-2.6
5091  */
5092 
5093 static void spapr_machine_2_6_class_options(MachineClass *mc)
5094 {
5095     static GlobalProperty compat[] = {
5096         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
5097     };
5098 
5099     spapr_machine_2_7_class_options(mc);
5100     mc->has_hotpluggable_cpus = false;
5101     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
5102     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5103 }
5104 
5105 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
5106 
5107 /*
5108  * pseries-2.5
5109  */
5110 
5111 static void spapr_machine_2_5_class_options(MachineClass *mc)
5112 {
5113     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5114     static GlobalProperty compat[] = {
5115         { "spapr-vlan", "use-rx-buffer-pools", "off" },
5116     };
5117 
5118     spapr_machine_2_6_class_options(mc);
5119     smc->use_ohci_by_default = true;
5120     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
5121     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5122 }
5123 
5124 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
5125 
5126 /*
5127  * pseries-2.4
5128  */
5129 
5130 static void spapr_machine_2_4_class_options(MachineClass *mc)
5131 {
5132     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5133 
5134     spapr_machine_2_5_class_options(mc);
5135     smc->dr_lmb_enabled = false;
5136     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
5137 }
5138 
5139 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
5140 
5141 /*
5142  * pseries-2.3
5143  */
5144 
5145 static void spapr_machine_2_3_class_options(MachineClass *mc)
5146 {
5147     static GlobalProperty compat[] = {
5148         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
5149     };
5150     spapr_machine_2_4_class_options(mc);
5151     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
5152     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5153 }
5154 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
5155 
5156 /*
5157  * pseries-2.2
5158  */
5159 
5160 static void spapr_machine_2_2_class_options(MachineClass *mc)
5161 {
5162     static GlobalProperty compat[] = {
5163         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
5164     };
5165 
5166     spapr_machine_2_3_class_options(mc);
5167     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
5168     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5169     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
5170 }
5171 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
5172 
5173 /*
5174  * pseries-2.1
5175  */
5176 
5177 static void spapr_machine_2_1_class_options(MachineClass *mc)
5178 {
5179     spapr_machine_2_2_class_options(mc);
5180     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
5181 }
5182 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
5183 
5184 static void spapr_machine_register_types(void)
5185 {
5186     type_register_static(&spapr_machine_info);
5187 }
5188 
5189 type_init(spapr_machine_register_types)
5190