xref: /openbmc/qemu/hw/ppc/spapr.c (revision eabfeb0c)
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 "qapi/error.h"
30 #include "qapi/visitor.h"
31 #include "sysemu/sysemu.h"
32 #include "sysemu/hostmem.h"
33 #include "sysemu/numa.h"
34 #include "sysemu/qtest.h"
35 #include "sysemu/reset.h"
36 #include "sysemu/runstate.h"
37 #include "qemu/log.h"
38 #include "hw/fw-path-provider.h"
39 #include "elf.h"
40 #include "net/net.h"
41 #include "sysemu/device_tree.h"
42 #include "sysemu/cpus.h"
43 #include "sysemu/hw_accel.h"
44 #include "kvm_ppc.h"
45 #include "migration/misc.h"
46 #include "migration/qemu-file-types.h"
47 #include "migration/global_state.h"
48 #include "migration/register.h"
49 #include "migration/blocker.h"
50 #include "mmu-hash64.h"
51 #include "mmu-book3s-v3.h"
52 #include "cpu-models.h"
53 #include "hw/core/cpu.h"
54 
55 #include "hw/boards.h"
56 #include "hw/ppc/ppc.h"
57 #include "hw/loader.h"
58 
59 #include "hw/ppc/fdt.h"
60 #include "hw/ppc/spapr.h"
61 #include "hw/ppc/spapr_vio.h"
62 #include "hw/qdev-properties.h"
63 #include "hw/pci-host/spapr.h"
64 #include "hw/pci/msi.h"
65 
66 #include "hw/pci/pci.h"
67 #include "hw/scsi/scsi.h"
68 #include "hw/virtio/virtio-scsi.h"
69 #include "hw/virtio/vhost-scsi-common.h"
70 
71 #include "exec/address-spaces.h"
72 #include "exec/ram_addr.h"
73 #include "hw/usb.h"
74 #include "qemu/config-file.h"
75 #include "qemu/error-report.h"
76 #include "trace.h"
77 #include "hw/nmi.h"
78 #include "hw/intc/intc.h"
79 
80 #include "hw/ppc/spapr_cpu_core.h"
81 #include "hw/mem/memory-device.h"
82 #include "hw/ppc/spapr_tpm_proxy.h"
83 #include "hw/ppc/spapr_nvdimm.h"
84 
85 #include "monitor/monitor.h"
86 
87 #include <libfdt.h>
88 
89 /* SLOF memory layout:
90  *
91  * SLOF raw image loaded at 0, copies its romfs right below the flat
92  * device-tree, then position SLOF itself 31M below that
93  *
94  * So we set FW_OVERHEAD to 40MB which should account for all of that
95  * and more
96  *
97  * We load our kernel at 4M, leaving space for SLOF initial image
98  */
99 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
100 #define FW_MAX_SIZE             0x400000
101 #define FW_FILE_NAME            "slof.bin"
102 #define FW_OVERHEAD             0x2800000
103 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
104 
105 #define MIN_RMA_SLOF            (128 * MiB)
106 
107 #define PHANDLE_INTC            0x00001111
108 
109 /* These two functions implement the VCPU id numbering: one to compute them
110  * all and one to identify thread 0 of a VCORE. Any change to the first one
111  * is likely to have an impact on the second one, so let's keep them close.
112  */
113 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
114 {
115     MachineState *ms = MACHINE(spapr);
116     unsigned int smp_threads = ms->smp.threads;
117 
118     assert(spapr->vsmt);
119     return
120         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
121 }
122 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
123                                       PowerPCCPU *cpu)
124 {
125     assert(spapr->vsmt);
126     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
127 }
128 
129 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
130 {
131     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
132      * and newer QEMUs don't even have them. In both cases, we don't want
133      * to send anything on the wire.
134      */
135     return false;
136 }
137 
138 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
139     .name = "icp/server",
140     .version_id = 1,
141     .minimum_version_id = 1,
142     .needed = pre_2_10_vmstate_dummy_icp_needed,
143     .fields = (VMStateField[]) {
144         VMSTATE_UNUSED(4), /* uint32_t xirr */
145         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
146         VMSTATE_UNUSED(1), /* uint8_t mfrr */
147         VMSTATE_END_OF_LIST()
148     },
149 };
150 
151 static void pre_2_10_vmstate_register_dummy_icp(int i)
152 {
153     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
154                      (void *)(uintptr_t) i);
155 }
156 
157 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
158 {
159     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
160                        (void *)(uintptr_t) i);
161 }
162 
163 int spapr_max_server_number(SpaprMachineState *spapr)
164 {
165     MachineState *ms = MACHINE(spapr);
166 
167     assert(spapr->vsmt);
168     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
169 }
170 
171 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
172                                   int smt_threads)
173 {
174     int i, ret = 0;
175     uint32_t servers_prop[smt_threads];
176     uint32_t gservers_prop[smt_threads * 2];
177     int index = spapr_get_vcpu_id(cpu);
178 
179     if (cpu->compat_pvr) {
180         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
181         if (ret < 0) {
182             return ret;
183         }
184     }
185 
186     /* Build interrupt servers and gservers properties */
187     for (i = 0; i < smt_threads; i++) {
188         servers_prop[i] = cpu_to_be32(index + i);
189         /* Hack, direct the group queues back to cpu 0 */
190         gservers_prop[i*2] = cpu_to_be32(index + i);
191         gservers_prop[i*2 + 1] = 0;
192     }
193     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
194                       servers_prop, sizeof(servers_prop));
195     if (ret < 0) {
196         return ret;
197     }
198     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
199                       gservers_prop, sizeof(gservers_prop));
200 
201     return ret;
202 }
203 
204 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
205 {
206     int index = spapr_get_vcpu_id(cpu);
207     uint32_t associativity[] = {cpu_to_be32(0x5),
208                                 cpu_to_be32(0x0),
209                                 cpu_to_be32(0x0),
210                                 cpu_to_be32(0x0),
211                                 cpu_to_be32(cpu->node_id),
212                                 cpu_to_be32(index)};
213 
214     /* Advertise NUMA via ibm,associativity */
215     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
216                           sizeof(associativity));
217 }
218 
219 static void spapr_dt_pa_features(SpaprMachineState *spapr,
220                                  PowerPCCPU *cpu,
221                                  void *fdt, int offset)
222 {
223     uint8_t pa_features_206[] = { 6, 0,
224         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
225     uint8_t pa_features_207[] = { 24, 0,
226         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
227         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
228         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
229         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
230     uint8_t pa_features_300[] = { 66, 0,
231         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
232         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
233         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
234         /* 6: DS207 */
235         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
236         /* 16: Vector */
237         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
238         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
239         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
240         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
241         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
242         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
243         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
244         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
245         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
246         /* 42: PM, 44: PC RA, 46: SC vec'd */
247         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
248         /* 48: SIMD, 50: QP BFP, 52: String */
249         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
250         /* 54: DecFP, 56: DecI, 58: SHA */
251         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
252         /* 60: NM atomic, 62: RNG */
253         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
254     };
255     uint8_t *pa_features = NULL;
256     size_t pa_size;
257 
258     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
259         pa_features = pa_features_206;
260         pa_size = sizeof(pa_features_206);
261     }
262     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
263         pa_features = pa_features_207;
264         pa_size = sizeof(pa_features_207);
265     }
266     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
267         pa_features = pa_features_300;
268         pa_size = sizeof(pa_features_300);
269     }
270     if (!pa_features) {
271         return;
272     }
273 
274     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
275         /*
276          * Note: we keep CI large pages off by default because a 64K capable
277          * guest provisioned with large pages might otherwise try to map a qemu
278          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
279          * even if that qemu runs on a 4k host.
280          * We dd this bit back here if we are confident this is not an issue
281          */
282         pa_features[3] |= 0x20;
283     }
284     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
285         pa_features[24] |= 0x80;    /* Transactional memory support */
286     }
287     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
288         /* Workaround for broken kernels that attempt (guest) radix
289          * mode when they can't handle it, if they see the radix bit set
290          * in pa-features. So hide it from them. */
291         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
292     }
293 
294     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
295 }
296 
297 static hwaddr spapr_node0_size(MachineState *machine)
298 {
299     if (machine->numa_state->num_nodes) {
300         int i;
301         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
302             if (machine->numa_state->nodes[i].node_mem) {
303                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
304                            machine->ram_size);
305             }
306         }
307     }
308     return machine->ram_size;
309 }
310 
311 static void add_str(GString *s, const gchar *s1)
312 {
313     g_string_append_len(s, s1, strlen(s1) + 1);
314 }
315 
316 static int spapr_dt_memory_node(void *fdt, int nodeid, hwaddr start,
317                                 hwaddr size)
318 {
319     uint32_t associativity[] = {
320         cpu_to_be32(0x4), /* length */
321         cpu_to_be32(0x0), cpu_to_be32(0x0),
322         cpu_to_be32(0x0), cpu_to_be32(nodeid)
323     };
324     char mem_name[32];
325     uint64_t mem_reg_property[2];
326     int off;
327 
328     mem_reg_property[0] = cpu_to_be64(start);
329     mem_reg_property[1] = cpu_to_be64(size);
330 
331     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
332     off = fdt_add_subnode(fdt, 0, mem_name);
333     _FDT(off);
334     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
335     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
336                       sizeof(mem_reg_property))));
337     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
338                       sizeof(associativity))));
339     return off;
340 }
341 
342 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
343 {
344     MemoryDeviceInfoList *info;
345 
346     for (info = list; info; info = info->next) {
347         MemoryDeviceInfo *value = info->value;
348 
349         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
350             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
351 
352             if (addr >= pcdimm_info->addr &&
353                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
354                 return pcdimm_info->node;
355             }
356         }
357     }
358 
359     return -1;
360 }
361 
362 struct sPAPRDrconfCellV2 {
363      uint32_t seq_lmbs;
364      uint64_t base_addr;
365      uint32_t drc_index;
366      uint32_t aa_index;
367      uint32_t flags;
368 } QEMU_PACKED;
369 
370 typedef struct DrconfCellQueue {
371     struct sPAPRDrconfCellV2 cell;
372     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
373 } DrconfCellQueue;
374 
375 static DrconfCellQueue *
376 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
377                       uint32_t drc_index, uint32_t aa_index,
378                       uint32_t flags)
379 {
380     DrconfCellQueue *elem;
381 
382     elem = g_malloc0(sizeof(*elem));
383     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
384     elem->cell.base_addr = cpu_to_be64(base_addr);
385     elem->cell.drc_index = cpu_to_be32(drc_index);
386     elem->cell.aa_index = cpu_to_be32(aa_index);
387     elem->cell.flags = cpu_to_be32(flags);
388 
389     return elem;
390 }
391 
392 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
393                                       int offset, MemoryDeviceInfoList *dimms)
394 {
395     MachineState *machine = MACHINE(spapr);
396     uint8_t *int_buf, *cur_index;
397     int ret;
398     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
399     uint64_t addr, cur_addr, size;
400     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
401     uint64_t mem_end = machine->device_memory->base +
402                        memory_region_size(&machine->device_memory->mr);
403     uint32_t node, buf_len, nr_entries = 0;
404     SpaprDrc *drc;
405     DrconfCellQueue *elem, *next;
406     MemoryDeviceInfoList *info;
407     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
408         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
409 
410     /* Entry to cover RAM and the gap area */
411     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
412                                  SPAPR_LMB_FLAGS_RESERVED |
413                                  SPAPR_LMB_FLAGS_DRC_INVALID);
414     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
415     nr_entries++;
416 
417     cur_addr = machine->device_memory->base;
418     for (info = dimms; info; info = info->next) {
419         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
420 
421         addr = di->addr;
422         size = di->size;
423         node = di->node;
424 
425         /*
426          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
427          * area is marked hotpluggable in the next iteration for the bigger
428          * chunk including the NVDIMM occupied area.
429          */
430         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
431             continue;
432 
433         /* Entry for hot-pluggable area */
434         if (cur_addr < addr) {
435             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
436             g_assert(drc);
437             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
438                                          cur_addr, spapr_drc_index(drc), -1, 0);
439             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
440             nr_entries++;
441         }
442 
443         /* Entry for DIMM */
444         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
445         g_assert(drc);
446         elem = spapr_get_drconf_cell(size / lmb_size, addr,
447                                      spapr_drc_index(drc), node,
448                                      (SPAPR_LMB_FLAGS_ASSIGNED |
449                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
450         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
451         nr_entries++;
452         cur_addr = addr + size;
453     }
454 
455     /* Entry for remaining hotpluggable area */
456     if (cur_addr < mem_end) {
457         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
458         g_assert(drc);
459         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
460                                      cur_addr, spapr_drc_index(drc), -1, 0);
461         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
462         nr_entries++;
463     }
464 
465     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
466     int_buf = cur_index = g_malloc0(buf_len);
467     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
468     cur_index += sizeof(nr_entries);
469 
470     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
471         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
472         cur_index += sizeof(elem->cell);
473         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
474         g_free(elem);
475     }
476 
477     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
478     g_free(int_buf);
479     if (ret < 0) {
480         return -1;
481     }
482     return 0;
483 }
484 
485 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
486                                    int offset, MemoryDeviceInfoList *dimms)
487 {
488     MachineState *machine = MACHINE(spapr);
489     int i, ret;
490     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
491     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
492     uint32_t nr_lmbs = (machine->device_memory->base +
493                        memory_region_size(&machine->device_memory->mr)) /
494                        lmb_size;
495     uint32_t *int_buf, *cur_index, buf_len;
496 
497     /*
498      * Allocate enough buffer size to fit in ibm,dynamic-memory
499      */
500     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
501     cur_index = int_buf = g_malloc0(buf_len);
502     int_buf[0] = cpu_to_be32(nr_lmbs);
503     cur_index++;
504     for (i = 0; i < nr_lmbs; i++) {
505         uint64_t addr = i * lmb_size;
506         uint32_t *dynamic_memory = cur_index;
507 
508         if (i >= device_lmb_start) {
509             SpaprDrc *drc;
510 
511             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
512             g_assert(drc);
513 
514             dynamic_memory[0] = cpu_to_be32(addr >> 32);
515             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
516             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
517             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
518             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
519             if (memory_region_present(get_system_memory(), addr)) {
520                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
521             } else {
522                 dynamic_memory[5] = cpu_to_be32(0);
523             }
524         } else {
525             /*
526              * LMB information for RMA, boot time RAM and gap b/n RAM and
527              * device memory region -- all these are marked as reserved
528              * and as having no valid DRC.
529              */
530             dynamic_memory[0] = cpu_to_be32(addr >> 32);
531             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
532             dynamic_memory[2] = cpu_to_be32(0);
533             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
534             dynamic_memory[4] = cpu_to_be32(-1);
535             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
536                                             SPAPR_LMB_FLAGS_DRC_INVALID);
537         }
538 
539         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
540     }
541     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
542     g_free(int_buf);
543     if (ret < 0) {
544         return -1;
545     }
546     return 0;
547 }
548 
549 /*
550  * Adds ibm,dynamic-reconfiguration-memory node.
551  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
552  * of this device tree node.
553  */
554 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
555                                                    void *fdt)
556 {
557     MachineState *machine = MACHINE(spapr);
558     int nb_numa_nodes = machine->numa_state->num_nodes;
559     int ret, i, offset;
560     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
561     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
562     uint32_t *int_buf, *cur_index, buf_len;
563     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
564     MemoryDeviceInfoList *dimms = NULL;
565 
566     /*
567      * Don't create the node if there is no device memory
568      */
569     if (machine->ram_size == machine->maxram_size) {
570         return 0;
571     }
572 
573     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
574 
575     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
576                     sizeof(prop_lmb_size));
577     if (ret < 0) {
578         return ret;
579     }
580 
581     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
582     if (ret < 0) {
583         return ret;
584     }
585 
586     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
587     if (ret < 0) {
588         return ret;
589     }
590 
591     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
592     dimms = qmp_memory_device_list();
593     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
594         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
595     } else {
596         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
597     }
598     qapi_free_MemoryDeviceInfoList(dimms);
599 
600     if (ret < 0) {
601         return ret;
602     }
603 
604     /* ibm,associativity-lookup-arrays */
605     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
606     cur_index = int_buf = g_malloc0(buf_len);
607     int_buf[0] = cpu_to_be32(nr_nodes);
608     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
609     cur_index += 2;
610     for (i = 0; i < nr_nodes; i++) {
611         uint32_t associativity[] = {
612             cpu_to_be32(0x0),
613             cpu_to_be32(0x0),
614             cpu_to_be32(0x0),
615             cpu_to_be32(i)
616         };
617         memcpy(cur_index, associativity, sizeof(associativity));
618         cur_index += 4;
619     }
620     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
621             (cur_index - int_buf) * sizeof(uint32_t));
622     g_free(int_buf);
623 
624     return ret;
625 }
626 
627 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
628 {
629     MachineState *machine = MACHINE(spapr);
630     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
631     hwaddr mem_start, node_size;
632     int i, nb_nodes = machine->numa_state->num_nodes;
633     NodeInfo *nodes = machine->numa_state->nodes;
634 
635     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
636         if (!nodes[i].node_mem) {
637             continue;
638         }
639         if (mem_start >= machine->ram_size) {
640             node_size = 0;
641         } else {
642             node_size = nodes[i].node_mem;
643             if (node_size > machine->ram_size - mem_start) {
644                 node_size = machine->ram_size - mem_start;
645             }
646         }
647         if (!mem_start) {
648             /* spapr_machine_init() checks for rma_size <= node0_size
649              * already */
650             spapr_dt_memory_node(fdt, i, 0, spapr->rma_size);
651             mem_start += spapr->rma_size;
652             node_size -= spapr->rma_size;
653         }
654         for ( ; node_size; ) {
655             hwaddr sizetmp = pow2floor(node_size);
656 
657             /* mem_start != 0 here */
658             if (ctzl(mem_start) < ctzl(sizetmp)) {
659                 sizetmp = 1ULL << ctzl(mem_start);
660             }
661 
662             spapr_dt_memory_node(fdt, i, mem_start, sizetmp);
663             node_size -= sizetmp;
664             mem_start += sizetmp;
665         }
666     }
667 
668     /* Generate ibm,dynamic-reconfiguration-memory node if required */
669     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
670         int ret;
671 
672         g_assert(smc->dr_lmb_enabled);
673         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
674         if (ret) {
675             return ret;
676         }
677     }
678 
679     return 0;
680 }
681 
682 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
683                          SpaprMachineState *spapr)
684 {
685     MachineState *ms = MACHINE(spapr);
686     PowerPCCPU *cpu = POWERPC_CPU(cs);
687     CPUPPCState *env = &cpu->env;
688     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
689     int index = spapr_get_vcpu_id(cpu);
690     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
691                        0xffffffff, 0xffffffff};
692     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
693         : SPAPR_TIMEBASE_FREQ;
694     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
695     uint32_t page_sizes_prop[64];
696     size_t page_sizes_prop_size;
697     unsigned int smp_threads = ms->smp.threads;
698     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
699     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
700     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
701     SpaprDrc *drc;
702     int drc_index;
703     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
704     int i;
705 
706     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
707     if (drc) {
708         drc_index = spapr_drc_index(drc);
709         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
710     }
711 
712     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
713     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
714 
715     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
716     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
717                            env->dcache_line_size)));
718     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
719                            env->dcache_line_size)));
720     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
721                            env->icache_line_size)));
722     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
723                            env->icache_line_size)));
724 
725     if (pcc->l1_dcache_size) {
726         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
727                                pcc->l1_dcache_size)));
728     } else {
729         warn_report("Unknown L1 dcache size for cpu");
730     }
731     if (pcc->l1_icache_size) {
732         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
733                                pcc->l1_icache_size)));
734     } else {
735         warn_report("Unknown L1 icache size for cpu");
736     }
737 
738     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
739     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
740     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
741     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
742     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
743     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
744 
745     if (env->spr_cb[SPR_PURR].oea_read) {
746         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
747     }
748     if (env->spr_cb[SPR_SPURR].oea_read) {
749         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
750     }
751 
752     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
753         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
754                           segs, sizeof(segs))));
755     }
756 
757     /* Advertise VSX (vector extensions) if available
758      *   1               == VMX / Altivec available
759      *   2               == VSX available
760      *
761      * Only CPUs for which we create core types in spapr_cpu_core.c
762      * are possible, and all of those have VMX */
763     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
764         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
765     } else {
766         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
767     }
768 
769     /* Advertise DFP (Decimal Floating Point) if available
770      *   0 / no property == no DFP
771      *   1               == DFP available */
772     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
773         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
774     }
775 
776     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
777                                                       sizeof(page_sizes_prop));
778     if (page_sizes_prop_size) {
779         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
780                           page_sizes_prop, page_sizes_prop_size)));
781     }
782 
783     spapr_dt_pa_features(spapr, cpu, fdt, offset);
784 
785     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
786                            cs->cpu_index / vcpus_per_socket)));
787 
788     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
789                       pft_size_prop, sizeof(pft_size_prop))));
790 
791     if (ms->numa_state->num_nodes > 1) {
792         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
793     }
794 
795     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
796 
797     if (pcc->radix_page_info) {
798         for (i = 0; i < pcc->radix_page_info->count; i++) {
799             radix_AP_encodings[i] =
800                 cpu_to_be32(pcc->radix_page_info->entries[i]);
801         }
802         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
803                           radix_AP_encodings,
804                           pcc->radix_page_info->count *
805                           sizeof(radix_AP_encodings[0]))));
806     }
807 
808     /*
809      * We set this property to let the guest know that it can use the large
810      * decrementer and its width in bits.
811      */
812     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
813         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
814                               pcc->lrg_decr_bits)));
815 }
816 
817 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
818 {
819     CPUState **rev;
820     CPUState *cs;
821     int n_cpus;
822     int cpus_offset;
823     char *nodename;
824     int i;
825 
826     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
827     _FDT(cpus_offset);
828     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
829     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
830 
831     /*
832      * We walk the CPUs in reverse order to ensure that CPU DT nodes
833      * created by fdt_add_subnode() end up in the right order in FDT
834      * for the guest kernel the enumerate the CPUs correctly.
835      *
836      * The CPU list cannot be traversed in reverse order, so we need
837      * to do extra work.
838      */
839     n_cpus = 0;
840     rev = NULL;
841     CPU_FOREACH(cs) {
842         rev = g_renew(CPUState *, rev, n_cpus + 1);
843         rev[n_cpus++] = cs;
844     }
845 
846     for (i = n_cpus - 1; i >= 0; i--) {
847         CPUState *cs = rev[i];
848         PowerPCCPU *cpu = POWERPC_CPU(cs);
849         int index = spapr_get_vcpu_id(cpu);
850         DeviceClass *dc = DEVICE_GET_CLASS(cs);
851         int offset;
852 
853         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
854             continue;
855         }
856 
857         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
858         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
859         g_free(nodename);
860         _FDT(offset);
861         spapr_dt_cpu(cs, fdt, offset, spapr);
862     }
863 
864     g_free(rev);
865 }
866 
867 static int spapr_dt_rng(void *fdt)
868 {
869     int node;
870     int ret;
871 
872     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
873     if (node <= 0) {
874         return -1;
875     }
876     ret = fdt_setprop_string(fdt, node, "device_type",
877                              "ibm,platform-facilities");
878     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
879     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
880 
881     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
882     if (node <= 0) {
883         return -1;
884     }
885     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
886 
887     return ret ? -1 : 0;
888 }
889 
890 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
891 {
892     MachineState *ms = MACHINE(spapr);
893     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
894     int rtas;
895     GString *hypertas = g_string_sized_new(256);
896     GString *qemu_hypertas = g_string_sized_new(256);
897     uint32_t refpoints[] = {
898         cpu_to_be32(0x4),
899         cpu_to_be32(0x4),
900         cpu_to_be32(0x2),
901     };
902     uint32_t nr_refpoints = ARRAY_SIZE(refpoints);
903     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
904         memory_region_size(&MACHINE(spapr)->device_memory->mr);
905     uint32_t lrdr_capacity[] = {
906         cpu_to_be32(max_device_addr >> 32),
907         cpu_to_be32(max_device_addr & 0xffffffff),
908         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
909         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
910     };
911     uint32_t maxdomain = cpu_to_be32(spapr->gpu_numa_id > 1 ? 1 : 0);
912     uint32_t maxdomains[] = {
913         cpu_to_be32(4),
914         maxdomain,
915         maxdomain,
916         maxdomain,
917         cpu_to_be32(spapr->gpu_numa_id),
918     };
919 
920     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
921 
922     /* hypertas */
923     add_str(hypertas, "hcall-pft");
924     add_str(hypertas, "hcall-term");
925     add_str(hypertas, "hcall-dabr");
926     add_str(hypertas, "hcall-interrupt");
927     add_str(hypertas, "hcall-tce");
928     add_str(hypertas, "hcall-vio");
929     add_str(hypertas, "hcall-splpar");
930     add_str(hypertas, "hcall-join");
931     add_str(hypertas, "hcall-bulk");
932     add_str(hypertas, "hcall-set-mode");
933     add_str(hypertas, "hcall-sprg0");
934     add_str(hypertas, "hcall-copy");
935     add_str(hypertas, "hcall-debug");
936     add_str(hypertas, "hcall-vphn");
937     add_str(qemu_hypertas, "hcall-memop1");
938 
939     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
940         add_str(hypertas, "hcall-multi-tce");
941     }
942 
943     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
944         add_str(hypertas, "hcall-hpt-resize");
945     }
946 
947     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
948                      hypertas->str, hypertas->len));
949     g_string_free(hypertas, TRUE);
950     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
951                      qemu_hypertas->str, qemu_hypertas->len));
952     g_string_free(qemu_hypertas, TRUE);
953 
954     if (smc->pre_5_1_assoc_refpoints) {
955         nr_refpoints = 2;
956     }
957 
958     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
959                      refpoints, nr_refpoints * sizeof(refpoints[0])));
960 
961     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
962                      maxdomains, sizeof(maxdomains)));
963 
964     /*
965      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
966      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
967      *
968      * The system reset requirements are driven by existing Linux and PowerVM
969      * implementation which (contrary to PAPR) saves r3 in the error log
970      * structure like machine check, so Linux expects to find the saved r3
971      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
972      * does not look at the error value).
973      *
974      * System reset interrupts are not subject to interlock like machine
975      * check, so this memory area could be corrupted if the sreset is
976      * interrupted by a machine check (or vice versa) if it was shared. To
977      * prevent this, system reset uses per-CPU areas for the sreset save
978      * area. A system reset that interrupts a system reset handler could
979      * still overwrite this area, but Linux doesn't try to recover in that
980      * case anyway.
981      *
982      * The extra 8 bytes is required because Linux's FWNMI error log check
983      * is off-by-one.
984      */
985     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_ERROR_LOG_MAX +
986 			  ms->smp.max_cpus * sizeof(uint64_t)*2 + sizeof(uint64_t)));
987     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
988                           RTAS_ERROR_LOG_MAX));
989     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
990                           RTAS_EVENT_SCAN_RATE));
991 
992     g_assert(msi_nonbroken);
993     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
994 
995     /*
996      * According to PAPR, rtas ibm,os-term does not guarantee a return
997      * back to the guest cpu.
998      *
999      * While an additional ibm,extended-os-term property indicates
1000      * that rtas call return will always occur. Set this property.
1001      */
1002     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1003 
1004     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1005                      lrdr_capacity, sizeof(lrdr_capacity)));
1006 
1007     spapr_dt_rtas_tokens(fdt, rtas);
1008 }
1009 
1010 /*
1011  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
1012  * and the XIVE features that the guest may request and thus the valid
1013  * values for bytes 23..26 of option vector 5:
1014  */
1015 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
1016                                           int chosen)
1017 {
1018     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1019 
1020     char val[2 * 4] = {
1021         23, 0x00, /* XICS / XIVE mode */
1022         24, 0x00, /* Hash/Radix, filled in below. */
1023         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1024         26, 0x40, /* Radix options: GTSE == yes. */
1025     };
1026 
1027     if (spapr->irq->xics && spapr->irq->xive) {
1028         val[1] = SPAPR_OV5_XIVE_BOTH;
1029     } else if (spapr->irq->xive) {
1030         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
1031     } else {
1032         assert(spapr->irq->xics);
1033         val[1] = SPAPR_OV5_XIVE_LEGACY;
1034     }
1035 
1036     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1037                           first_ppc_cpu->compat_pvr)) {
1038         /*
1039          * If we're in a pre POWER9 compat mode then the guest should
1040          * do hash and use the legacy interrupt mode
1041          */
1042         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1043         val[3] = 0x00; /* Hash */
1044     } else if (kvm_enabled()) {
1045         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1046             val[3] = 0x80; /* OV5_MMU_BOTH */
1047         } else if (kvmppc_has_cap_mmu_radix()) {
1048             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1049         } else {
1050             val[3] = 0x00; /* Hash */
1051         }
1052     } else {
1053         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1054         val[3] = 0xC0;
1055     }
1056     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1057                      val, sizeof(val)));
1058 }
1059 
1060 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1061 {
1062     MachineState *machine = MACHINE(spapr);
1063     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1064     int chosen;
1065 
1066     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1067 
1068     if (reset) {
1069         const char *boot_device = machine->boot_order;
1070         char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1071         size_t cb = 0;
1072         char *bootlist = get_boot_devices_list(&cb);
1073 
1074         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1075             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1076                                     machine->kernel_cmdline));
1077         }
1078 
1079         if (spapr->initrd_size) {
1080             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1081                                   spapr->initrd_base));
1082             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1083                                   spapr->initrd_base + spapr->initrd_size));
1084         }
1085 
1086         if (spapr->kernel_size) {
1087             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1088                                   cpu_to_be64(spapr->kernel_size) };
1089 
1090             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1091                          &kprop, sizeof(kprop)));
1092             if (spapr->kernel_le) {
1093                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1094             }
1095         }
1096         if (boot_menu) {
1097             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1098         }
1099         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1100         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1101         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1102 
1103         if (cb && bootlist) {
1104             int i;
1105 
1106             for (i = 0; i < cb; i++) {
1107                 if (bootlist[i] == '\n') {
1108                     bootlist[i] = ' ';
1109                 }
1110             }
1111             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1112         }
1113 
1114         if (boot_device && strlen(boot_device)) {
1115             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1116         }
1117 
1118         if (!spapr->has_graphics && stdout_path) {
1119             /*
1120              * "linux,stdout-path" and "stdout" properties are
1121              * deprecated by linux kernel. New platforms should only
1122              * use the "stdout-path" property. Set the new property
1123              * and continue using older property to remain compatible
1124              * with the existing firmware.
1125              */
1126             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1127             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1128         }
1129 
1130         /*
1131          * We can deal with BAR reallocation just fine, advertise it
1132          * to the guest
1133          */
1134         if (smc->linux_pci_probe) {
1135             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1136         }
1137 
1138         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1139 
1140         g_free(stdout_path);
1141         g_free(bootlist);
1142     }
1143 
1144     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1145 }
1146 
1147 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1148 {
1149     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1150      * KVM to work under pHyp with some guest co-operation */
1151     int hypervisor;
1152     uint8_t hypercall[16];
1153 
1154     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1155     /* indicate KVM hypercall interface */
1156     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1157     if (kvmppc_has_cap_fixup_hcalls()) {
1158         /*
1159          * Older KVM versions with older guest kernels were broken
1160          * with the magic page, don't allow the guest to map it.
1161          */
1162         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1163                                   sizeof(hypercall))) {
1164             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1165                              hypercall, sizeof(hypercall)));
1166         }
1167     }
1168 }
1169 
1170 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1171 {
1172     MachineState *machine = MACHINE(spapr);
1173     MachineClass *mc = MACHINE_GET_CLASS(machine);
1174     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1175     int ret;
1176     void *fdt;
1177     SpaprPhbState *phb;
1178     char *buf;
1179 
1180     fdt = g_malloc0(space);
1181     _FDT((fdt_create_empty_tree(fdt, space)));
1182 
1183     /* Root node */
1184     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1185     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1186     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1187 
1188     /* Guest UUID & Name*/
1189     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1190     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1191     if (qemu_uuid_set) {
1192         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1193     }
1194     g_free(buf);
1195 
1196     if (qemu_get_vm_name()) {
1197         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1198                                 qemu_get_vm_name()));
1199     }
1200 
1201     /* Host Model & Serial Number */
1202     if (spapr->host_model) {
1203         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1204     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1205         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1206         g_free(buf);
1207     }
1208 
1209     if (spapr->host_serial) {
1210         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1211     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1212         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1213         g_free(buf);
1214     }
1215 
1216     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1217     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1218 
1219     /* /interrupt controller */
1220     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1221 
1222     ret = spapr_dt_memory(spapr, fdt);
1223     if (ret < 0) {
1224         error_report("couldn't setup memory nodes in fdt");
1225         exit(1);
1226     }
1227 
1228     /* /vdevice */
1229     spapr_dt_vdevice(spapr->vio_bus, fdt);
1230 
1231     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1232         ret = spapr_dt_rng(fdt);
1233         if (ret < 0) {
1234             error_report("could not set up rng device in the fdt");
1235             exit(1);
1236         }
1237     }
1238 
1239     QLIST_FOREACH(phb, &spapr->phbs, list) {
1240         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1241         if (ret < 0) {
1242             error_report("couldn't setup PCI devices in fdt");
1243             exit(1);
1244         }
1245     }
1246 
1247     spapr_dt_cpus(fdt, spapr);
1248 
1249     if (smc->dr_lmb_enabled) {
1250         _FDT(spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1251     }
1252 
1253     if (mc->has_hotpluggable_cpus) {
1254         int offset = fdt_path_offset(fdt, "/cpus");
1255         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1256         if (ret < 0) {
1257             error_report("Couldn't set up CPU DR device tree properties");
1258             exit(1);
1259         }
1260     }
1261 
1262     /* /event-sources */
1263     spapr_dt_events(spapr, fdt);
1264 
1265     /* /rtas */
1266     spapr_dt_rtas(spapr, fdt);
1267 
1268     /* /chosen */
1269     spapr_dt_chosen(spapr, fdt, reset);
1270 
1271     /* /hypervisor */
1272     if (kvm_enabled()) {
1273         spapr_dt_hypervisor(spapr, fdt);
1274     }
1275 
1276     /* Build memory reserve map */
1277     if (reset) {
1278         if (spapr->kernel_size) {
1279             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1280                                   spapr->kernel_size)));
1281         }
1282         if (spapr->initrd_size) {
1283             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1284                                   spapr->initrd_size)));
1285         }
1286     }
1287 
1288     if (smc->dr_phb_enabled) {
1289         ret = spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1290         if (ret < 0) {
1291             error_report("Couldn't set up PHB DR device tree properties");
1292             exit(1);
1293         }
1294     }
1295 
1296     /* NVDIMM devices */
1297     if (mc->nvdimm_supported) {
1298         spapr_dt_persistent_memory(fdt);
1299     }
1300 
1301     return fdt;
1302 }
1303 
1304 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1305 {
1306     SpaprMachineState *spapr = opaque;
1307 
1308     return (addr & 0x0fffffff) + spapr->kernel_addr;
1309 }
1310 
1311 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1312                                     PowerPCCPU *cpu)
1313 {
1314     CPUPPCState *env = &cpu->env;
1315 
1316     /* The TCG path should also be holding the BQL at this point */
1317     g_assert(qemu_mutex_iothread_locked());
1318 
1319     if (msr_pr) {
1320         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1321         env->gpr[3] = H_PRIVILEGE;
1322     } else {
1323         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1324     }
1325 }
1326 
1327 struct LPCRSyncState {
1328     target_ulong value;
1329     target_ulong mask;
1330 };
1331 
1332 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1333 {
1334     struct LPCRSyncState *s = arg.host_ptr;
1335     PowerPCCPU *cpu = POWERPC_CPU(cs);
1336     CPUPPCState *env = &cpu->env;
1337     target_ulong lpcr;
1338 
1339     cpu_synchronize_state(cs);
1340     lpcr = env->spr[SPR_LPCR];
1341     lpcr &= ~s->mask;
1342     lpcr |= s->value;
1343     ppc_store_lpcr(cpu, lpcr);
1344 }
1345 
1346 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1347 {
1348     CPUState *cs;
1349     struct LPCRSyncState s = {
1350         .value = value,
1351         .mask = mask
1352     };
1353     CPU_FOREACH(cs) {
1354         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1355     }
1356 }
1357 
1358 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1359 {
1360     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1361 
1362     /* Copy PATE1:GR into PATE0:HR */
1363     entry->dw0 = spapr->patb_entry & PATE0_HR;
1364     entry->dw1 = spapr->patb_entry;
1365 }
1366 
1367 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1368 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1369 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1370 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1371 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1372 
1373 /*
1374  * Get the fd to access the kernel htab, re-opening it if necessary
1375  */
1376 static int get_htab_fd(SpaprMachineState *spapr)
1377 {
1378     Error *local_err = NULL;
1379 
1380     if (spapr->htab_fd >= 0) {
1381         return spapr->htab_fd;
1382     }
1383 
1384     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1385     if (spapr->htab_fd < 0) {
1386         error_report_err(local_err);
1387     }
1388 
1389     return spapr->htab_fd;
1390 }
1391 
1392 void close_htab_fd(SpaprMachineState *spapr)
1393 {
1394     if (spapr->htab_fd >= 0) {
1395         close(spapr->htab_fd);
1396     }
1397     spapr->htab_fd = -1;
1398 }
1399 
1400 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1401 {
1402     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1403 
1404     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1405 }
1406 
1407 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1408 {
1409     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1410 
1411     assert(kvm_enabled());
1412 
1413     if (!spapr->htab) {
1414         return 0;
1415     }
1416 
1417     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1418 }
1419 
1420 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1421                                                 hwaddr ptex, int n)
1422 {
1423     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1424     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1425 
1426     if (!spapr->htab) {
1427         /*
1428          * HTAB is controlled by KVM. Fetch into temporary buffer
1429          */
1430         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1431         kvmppc_read_hptes(hptes, ptex, n);
1432         return hptes;
1433     }
1434 
1435     /*
1436      * HTAB is controlled by QEMU. Just point to the internally
1437      * accessible PTEG.
1438      */
1439     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1440 }
1441 
1442 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1443                               const ppc_hash_pte64_t *hptes,
1444                               hwaddr ptex, int n)
1445 {
1446     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1447 
1448     if (!spapr->htab) {
1449         g_free((void *)hptes);
1450     }
1451 
1452     /* Nothing to do for qemu managed HPT */
1453 }
1454 
1455 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1456                       uint64_t pte0, uint64_t pte1)
1457 {
1458     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1459     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1460 
1461     if (!spapr->htab) {
1462         kvmppc_write_hpte(ptex, pte0, pte1);
1463     } else {
1464         if (pte0 & HPTE64_V_VALID) {
1465             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1466             /*
1467              * When setting valid, we write PTE1 first. This ensures
1468              * proper synchronization with the reading code in
1469              * ppc_hash64_pteg_search()
1470              */
1471             smp_wmb();
1472             stq_p(spapr->htab + offset, pte0);
1473         } else {
1474             stq_p(spapr->htab + offset, pte0);
1475             /*
1476              * When clearing it we set PTE0 first. This ensures proper
1477              * synchronization with the reading code in
1478              * ppc_hash64_pteg_search()
1479              */
1480             smp_wmb();
1481             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1482         }
1483     }
1484 }
1485 
1486 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1487                              uint64_t pte1)
1488 {
1489     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1490     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1491 
1492     if (!spapr->htab) {
1493         /* There should always be a hash table when this is called */
1494         error_report("spapr_hpte_set_c called with no hash table !");
1495         return;
1496     }
1497 
1498     /* The HW performs a non-atomic byte update */
1499     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1500 }
1501 
1502 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1503                              uint64_t pte1)
1504 {
1505     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1506     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1507 
1508     if (!spapr->htab) {
1509         /* There should always be a hash table when this is called */
1510         error_report("spapr_hpte_set_r called with no hash table !");
1511         return;
1512     }
1513 
1514     /* The HW performs a non-atomic byte update */
1515     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1516 }
1517 
1518 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1519 {
1520     int shift;
1521 
1522     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1523      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1524      * that's much more than is needed for Linux guests */
1525     shift = ctz64(pow2ceil(ramsize)) - 7;
1526     shift = MAX(shift, 18); /* Minimum architected size */
1527     shift = MIN(shift, 46); /* Maximum architected size */
1528     return shift;
1529 }
1530 
1531 void spapr_free_hpt(SpaprMachineState *spapr)
1532 {
1533     g_free(spapr->htab);
1534     spapr->htab = NULL;
1535     spapr->htab_shift = 0;
1536     close_htab_fd(spapr);
1537 }
1538 
1539 void spapr_reallocate_hpt(SpaprMachineState *spapr, int shift,
1540                           Error **errp)
1541 {
1542     long rc;
1543 
1544     /* Clean up any HPT info from a previous boot */
1545     spapr_free_hpt(spapr);
1546 
1547     rc = kvmppc_reset_htab(shift);
1548     if (rc < 0) {
1549         /* kernel-side HPT needed, but couldn't allocate one */
1550         error_setg_errno(errp, errno,
1551                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1552                          shift);
1553         /* This is almost certainly fatal, but if the caller really
1554          * wants to carry on with shift == 0, it's welcome to try */
1555     } else if (rc > 0) {
1556         /* kernel-side HPT allocated */
1557         if (rc != shift) {
1558             error_setg(errp,
1559                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1560                        shift, rc);
1561         }
1562 
1563         spapr->htab_shift = shift;
1564         spapr->htab = NULL;
1565     } else {
1566         /* kernel-side HPT not needed, allocate in userspace instead */
1567         size_t size = 1ULL << shift;
1568         int i;
1569 
1570         spapr->htab = qemu_memalign(size, size);
1571         if (!spapr->htab) {
1572             error_setg_errno(errp, errno,
1573                              "Could not allocate HPT of order %d", shift);
1574             return;
1575         }
1576 
1577         memset(spapr->htab, 0, size);
1578         spapr->htab_shift = shift;
1579 
1580         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1581             DIRTY_HPTE(HPTE(spapr->htab, i));
1582         }
1583     }
1584     /* We're setting up a hash table, so that means we're not radix */
1585     spapr->patb_entry = 0;
1586     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1587 }
1588 
1589 void spapr_setup_hpt(SpaprMachineState *spapr)
1590 {
1591     int hpt_shift;
1592 
1593     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1594         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1595     } else {
1596         uint64_t current_ram_size;
1597 
1598         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1599         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1600     }
1601     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1602 
1603     if (kvm_enabled()) {
1604         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1605 
1606         /* Check our RMA fits in the possible VRMA */
1607         if (vrma_limit < spapr->rma_size) {
1608             error_report("Unable to create %" HWADDR_PRIu
1609                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1610                          spapr->rma_size / MiB, vrma_limit / MiB);
1611             exit(EXIT_FAILURE);
1612         }
1613     }
1614 }
1615 
1616 static int spapr_reset_drcs(Object *child, void *opaque)
1617 {
1618     SpaprDrc *drc =
1619         (SpaprDrc *) object_dynamic_cast(child,
1620                                                  TYPE_SPAPR_DR_CONNECTOR);
1621 
1622     if (drc) {
1623         spapr_drc_reset(drc);
1624     }
1625 
1626     return 0;
1627 }
1628 
1629 static void spapr_machine_reset(MachineState *machine)
1630 {
1631     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1632     PowerPCCPU *first_ppc_cpu;
1633     hwaddr fdt_addr;
1634     void *fdt;
1635     int rc;
1636 
1637     kvmppc_svm_off(&error_fatal);
1638     spapr_caps_apply(spapr);
1639 
1640     first_ppc_cpu = POWERPC_CPU(first_cpu);
1641     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1642         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1643                               spapr->max_compat_pvr)) {
1644         /*
1645          * If using KVM with radix mode available, VCPUs can be started
1646          * without a HPT because KVM will start them in radix mode.
1647          * Set the GR bit in PATE so that we know there is no HPT.
1648          */
1649         spapr->patb_entry = PATE1_GR;
1650         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1651     } else {
1652         spapr_setup_hpt(spapr);
1653     }
1654 
1655     qemu_devices_reset();
1656 
1657     spapr_ovec_cleanup(spapr->ov5_cas);
1658     spapr->ov5_cas = spapr_ovec_new();
1659 
1660     ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1661 
1662     /*
1663      * This is fixing some of the default configuration of the XIVE
1664      * devices. To be called after the reset of the machine devices.
1665      */
1666     spapr_irq_reset(spapr, &error_fatal);
1667 
1668     /*
1669      * There is no CAS under qtest. Simulate one to please the code that
1670      * depends on spapr->ov5_cas. This is especially needed to test device
1671      * unplug, so we do that before resetting the DRCs.
1672      */
1673     if (qtest_enabled()) {
1674         spapr_ovec_cleanup(spapr->ov5_cas);
1675         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1676     }
1677 
1678     /* DRC reset may cause a device to be unplugged. This will cause troubles
1679      * if this device is used by another device (eg, a running vhost backend
1680      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1681      * situations, we reset DRCs after all devices have been reset.
1682      */
1683     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1684 
1685     spapr_clear_pending_events(spapr);
1686 
1687     /*
1688      * We place the device tree and RTAS just below either the top of the RMA,
1689      * or just below 2GB, whichever is lower, so that it can be
1690      * processed with 32-bit real mode code if necessary
1691      */
1692     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1693 
1694     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1695 
1696     rc = fdt_pack(fdt);
1697 
1698     /* Should only fail if we've built a corrupted tree */
1699     assert(rc == 0);
1700 
1701     /* Load the fdt */
1702     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1703     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1704     g_free(spapr->fdt_blob);
1705     spapr->fdt_size = fdt_totalsize(fdt);
1706     spapr->fdt_initial_size = spapr->fdt_size;
1707     spapr->fdt_blob = fdt;
1708 
1709     /* Set up the entry state */
1710     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, 0, fdt_addr, 0);
1711     first_ppc_cpu->env.gpr[5] = 0;
1712 
1713     spapr->fwnmi_system_reset_addr = -1;
1714     spapr->fwnmi_machine_check_addr = -1;
1715     spapr->fwnmi_machine_check_interlock = -1;
1716 
1717     /* Signal all vCPUs waiting on this condition */
1718     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1719 
1720     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1721 }
1722 
1723 static void spapr_create_nvram(SpaprMachineState *spapr)
1724 {
1725     DeviceState *dev = qdev_new("spapr-nvram");
1726     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1727 
1728     if (dinfo) {
1729         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1730                                 &error_fatal);
1731     }
1732 
1733     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1734 
1735     spapr->nvram = (struct SpaprNvram *)dev;
1736 }
1737 
1738 static void spapr_rtc_create(SpaprMachineState *spapr)
1739 {
1740     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1741                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1742                                        &error_fatal, NULL);
1743     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1744     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1745                               "date");
1746 }
1747 
1748 /* Returns whether we want to use VGA or not */
1749 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1750 {
1751     switch (vga_interface_type) {
1752     case VGA_NONE:
1753         return false;
1754     case VGA_DEVICE:
1755         return true;
1756     case VGA_STD:
1757     case VGA_VIRTIO:
1758     case VGA_CIRRUS:
1759         return pci_vga_init(pci_bus) != NULL;
1760     default:
1761         error_setg(errp,
1762                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1763         return false;
1764     }
1765 }
1766 
1767 static int spapr_pre_load(void *opaque)
1768 {
1769     int rc;
1770 
1771     rc = spapr_caps_pre_load(opaque);
1772     if (rc) {
1773         return rc;
1774     }
1775 
1776     return 0;
1777 }
1778 
1779 static int spapr_post_load(void *opaque, int version_id)
1780 {
1781     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1782     int err = 0;
1783 
1784     err = spapr_caps_post_migration(spapr);
1785     if (err) {
1786         return err;
1787     }
1788 
1789     /*
1790      * In earlier versions, there was no separate qdev for the PAPR
1791      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1792      * So when migrating from those versions, poke the incoming offset
1793      * value into the RTC device
1794      */
1795     if (version_id < 3) {
1796         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1797         if (err) {
1798             return err;
1799         }
1800     }
1801 
1802     if (kvm_enabled() && spapr->patb_entry) {
1803         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1804         bool radix = !!(spapr->patb_entry & PATE1_GR);
1805         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1806 
1807         /*
1808          * Update LPCR:HR and UPRT as they may not be set properly in
1809          * the stream
1810          */
1811         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1812                             LPCR_HR | LPCR_UPRT);
1813 
1814         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1815         if (err) {
1816             error_report("Process table config unsupported by the host");
1817             return -EINVAL;
1818         }
1819     }
1820 
1821     err = spapr_irq_post_load(spapr, version_id);
1822     if (err) {
1823         return err;
1824     }
1825 
1826     return err;
1827 }
1828 
1829 static int spapr_pre_save(void *opaque)
1830 {
1831     int rc;
1832 
1833     rc = spapr_caps_pre_save(opaque);
1834     if (rc) {
1835         return rc;
1836     }
1837 
1838     return 0;
1839 }
1840 
1841 static bool version_before_3(void *opaque, int version_id)
1842 {
1843     return version_id < 3;
1844 }
1845 
1846 static bool spapr_pending_events_needed(void *opaque)
1847 {
1848     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1849     return !QTAILQ_EMPTY(&spapr->pending_events);
1850 }
1851 
1852 static const VMStateDescription vmstate_spapr_event_entry = {
1853     .name = "spapr_event_log_entry",
1854     .version_id = 1,
1855     .minimum_version_id = 1,
1856     .fields = (VMStateField[]) {
1857         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1858         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1859         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1860                                      NULL, extended_length),
1861         VMSTATE_END_OF_LIST()
1862     },
1863 };
1864 
1865 static const VMStateDescription vmstate_spapr_pending_events = {
1866     .name = "spapr_pending_events",
1867     .version_id = 1,
1868     .minimum_version_id = 1,
1869     .needed = spapr_pending_events_needed,
1870     .fields = (VMStateField[]) {
1871         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1872                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1873         VMSTATE_END_OF_LIST()
1874     },
1875 };
1876 
1877 static bool spapr_ov5_cas_needed(void *opaque)
1878 {
1879     SpaprMachineState *spapr = opaque;
1880     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1881     bool cas_needed;
1882 
1883     /* Prior to the introduction of SpaprOptionVector, we had two option
1884      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1885      * Both of these options encode machine topology into the device-tree
1886      * in such a way that the now-booted OS should still be able to interact
1887      * appropriately with QEMU regardless of what options were actually
1888      * negotiatied on the source side.
1889      *
1890      * As such, we can avoid migrating the CAS-negotiated options if these
1891      * are the only options available on the current machine/platform.
1892      * Since these are the only options available for pseries-2.7 and
1893      * earlier, this allows us to maintain old->new/new->old migration
1894      * compatibility.
1895      *
1896      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1897      * via default pseries-2.8 machines and explicit command-line parameters.
1898      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1899      * of the actual CAS-negotiated values to continue working properly. For
1900      * example, availability of memory unplug depends on knowing whether
1901      * OV5_HP_EVT was negotiated via CAS.
1902      *
1903      * Thus, for any cases where the set of available CAS-negotiatable
1904      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1905      * include the CAS-negotiated options in the migration stream, unless
1906      * if they affect boot time behaviour only.
1907      */
1908     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1909     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1910     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1911 
1912     /* We need extra information if we have any bits outside the mask
1913      * defined above */
1914     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1915 
1916     spapr_ovec_cleanup(ov5_mask);
1917 
1918     return cas_needed;
1919 }
1920 
1921 static const VMStateDescription vmstate_spapr_ov5_cas = {
1922     .name = "spapr_option_vector_ov5_cas",
1923     .version_id = 1,
1924     .minimum_version_id = 1,
1925     .needed = spapr_ov5_cas_needed,
1926     .fields = (VMStateField[]) {
1927         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1928                                  vmstate_spapr_ovec, SpaprOptionVector),
1929         VMSTATE_END_OF_LIST()
1930     },
1931 };
1932 
1933 static bool spapr_patb_entry_needed(void *opaque)
1934 {
1935     SpaprMachineState *spapr = opaque;
1936 
1937     return !!spapr->patb_entry;
1938 }
1939 
1940 static const VMStateDescription vmstate_spapr_patb_entry = {
1941     .name = "spapr_patb_entry",
1942     .version_id = 1,
1943     .minimum_version_id = 1,
1944     .needed = spapr_patb_entry_needed,
1945     .fields = (VMStateField[]) {
1946         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1947         VMSTATE_END_OF_LIST()
1948     },
1949 };
1950 
1951 static bool spapr_irq_map_needed(void *opaque)
1952 {
1953     SpaprMachineState *spapr = opaque;
1954 
1955     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1956 }
1957 
1958 static const VMStateDescription vmstate_spapr_irq_map = {
1959     .name = "spapr_irq_map",
1960     .version_id = 1,
1961     .minimum_version_id = 1,
1962     .needed = spapr_irq_map_needed,
1963     .fields = (VMStateField[]) {
1964         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1965         VMSTATE_END_OF_LIST()
1966     },
1967 };
1968 
1969 static bool spapr_dtb_needed(void *opaque)
1970 {
1971     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1972 
1973     return smc->update_dt_enabled;
1974 }
1975 
1976 static int spapr_dtb_pre_load(void *opaque)
1977 {
1978     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1979 
1980     g_free(spapr->fdt_blob);
1981     spapr->fdt_blob = NULL;
1982     spapr->fdt_size = 0;
1983 
1984     return 0;
1985 }
1986 
1987 static const VMStateDescription vmstate_spapr_dtb = {
1988     .name = "spapr_dtb",
1989     .version_id = 1,
1990     .minimum_version_id = 1,
1991     .needed = spapr_dtb_needed,
1992     .pre_load = spapr_dtb_pre_load,
1993     .fields = (VMStateField[]) {
1994         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1995         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1996         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1997                                      fdt_size),
1998         VMSTATE_END_OF_LIST()
1999     },
2000 };
2001 
2002 static bool spapr_fwnmi_needed(void *opaque)
2003 {
2004     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2005 
2006     return spapr->fwnmi_machine_check_addr != -1;
2007 }
2008 
2009 static int spapr_fwnmi_pre_save(void *opaque)
2010 {
2011     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2012 
2013     /*
2014      * Check if machine check handling is in progress and print a
2015      * warning message.
2016      */
2017     if (spapr->fwnmi_machine_check_interlock != -1) {
2018         warn_report("A machine check is being handled during migration. The"
2019                 "handler may run and log hardware error on the destination");
2020     }
2021 
2022     return 0;
2023 }
2024 
2025 static const VMStateDescription vmstate_spapr_fwnmi = {
2026     .name = "spapr_fwnmi",
2027     .version_id = 1,
2028     .minimum_version_id = 1,
2029     .needed = spapr_fwnmi_needed,
2030     .pre_save = spapr_fwnmi_pre_save,
2031     .fields = (VMStateField[]) {
2032         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
2033         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
2034         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
2035         VMSTATE_END_OF_LIST()
2036     },
2037 };
2038 
2039 static const VMStateDescription vmstate_spapr = {
2040     .name = "spapr",
2041     .version_id = 3,
2042     .minimum_version_id = 1,
2043     .pre_load = spapr_pre_load,
2044     .post_load = spapr_post_load,
2045     .pre_save = spapr_pre_save,
2046     .fields = (VMStateField[]) {
2047         /* used to be @next_irq */
2048         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2049 
2050         /* RTC offset */
2051         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2052 
2053         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2054         VMSTATE_END_OF_LIST()
2055     },
2056     .subsections = (const VMStateDescription*[]) {
2057         &vmstate_spapr_ov5_cas,
2058         &vmstate_spapr_patb_entry,
2059         &vmstate_spapr_pending_events,
2060         &vmstate_spapr_cap_htm,
2061         &vmstate_spapr_cap_vsx,
2062         &vmstate_spapr_cap_dfp,
2063         &vmstate_spapr_cap_cfpc,
2064         &vmstate_spapr_cap_sbbc,
2065         &vmstate_spapr_cap_ibs,
2066         &vmstate_spapr_cap_hpt_maxpagesize,
2067         &vmstate_spapr_irq_map,
2068         &vmstate_spapr_cap_nested_kvm_hv,
2069         &vmstate_spapr_dtb,
2070         &vmstate_spapr_cap_large_decr,
2071         &vmstate_spapr_cap_ccf_assist,
2072         &vmstate_spapr_cap_fwnmi,
2073         &vmstate_spapr_fwnmi,
2074         NULL
2075     }
2076 };
2077 
2078 static int htab_save_setup(QEMUFile *f, void *opaque)
2079 {
2080     SpaprMachineState *spapr = opaque;
2081 
2082     /* "Iteration" header */
2083     if (!spapr->htab_shift) {
2084         qemu_put_be32(f, -1);
2085     } else {
2086         qemu_put_be32(f, spapr->htab_shift);
2087     }
2088 
2089     if (spapr->htab) {
2090         spapr->htab_save_index = 0;
2091         spapr->htab_first_pass = true;
2092     } else {
2093         if (spapr->htab_shift) {
2094             assert(kvm_enabled());
2095         }
2096     }
2097 
2098 
2099     return 0;
2100 }
2101 
2102 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2103                             int chunkstart, int n_valid, int n_invalid)
2104 {
2105     qemu_put_be32(f, chunkstart);
2106     qemu_put_be16(f, n_valid);
2107     qemu_put_be16(f, n_invalid);
2108     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2109                     HASH_PTE_SIZE_64 * n_valid);
2110 }
2111 
2112 static void htab_save_end_marker(QEMUFile *f)
2113 {
2114     qemu_put_be32(f, 0);
2115     qemu_put_be16(f, 0);
2116     qemu_put_be16(f, 0);
2117 }
2118 
2119 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2120                                  int64_t max_ns)
2121 {
2122     bool has_timeout = max_ns != -1;
2123     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2124     int index = spapr->htab_save_index;
2125     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2126 
2127     assert(spapr->htab_first_pass);
2128 
2129     do {
2130         int chunkstart;
2131 
2132         /* Consume invalid HPTEs */
2133         while ((index < htabslots)
2134                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2135             CLEAN_HPTE(HPTE(spapr->htab, index));
2136             index++;
2137         }
2138 
2139         /* Consume valid HPTEs */
2140         chunkstart = index;
2141         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2142                && HPTE_VALID(HPTE(spapr->htab, index))) {
2143             CLEAN_HPTE(HPTE(spapr->htab, index));
2144             index++;
2145         }
2146 
2147         if (index > chunkstart) {
2148             int n_valid = index - chunkstart;
2149 
2150             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2151 
2152             if (has_timeout &&
2153                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2154                 break;
2155             }
2156         }
2157     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2158 
2159     if (index >= htabslots) {
2160         assert(index == htabslots);
2161         index = 0;
2162         spapr->htab_first_pass = false;
2163     }
2164     spapr->htab_save_index = index;
2165 }
2166 
2167 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2168                                 int64_t max_ns)
2169 {
2170     bool final = max_ns < 0;
2171     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2172     int examined = 0, sent = 0;
2173     int index = spapr->htab_save_index;
2174     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2175 
2176     assert(!spapr->htab_first_pass);
2177 
2178     do {
2179         int chunkstart, invalidstart;
2180 
2181         /* Consume non-dirty HPTEs */
2182         while ((index < htabslots)
2183                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2184             index++;
2185             examined++;
2186         }
2187 
2188         chunkstart = index;
2189         /* Consume valid dirty HPTEs */
2190         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2191                && HPTE_DIRTY(HPTE(spapr->htab, index))
2192                && HPTE_VALID(HPTE(spapr->htab, index))) {
2193             CLEAN_HPTE(HPTE(spapr->htab, index));
2194             index++;
2195             examined++;
2196         }
2197 
2198         invalidstart = index;
2199         /* Consume invalid dirty HPTEs */
2200         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2201                && HPTE_DIRTY(HPTE(spapr->htab, index))
2202                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2203             CLEAN_HPTE(HPTE(spapr->htab, index));
2204             index++;
2205             examined++;
2206         }
2207 
2208         if (index > chunkstart) {
2209             int n_valid = invalidstart - chunkstart;
2210             int n_invalid = index - invalidstart;
2211 
2212             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2213             sent += index - chunkstart;
2214 
2215             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2216                 break;
2217             }
2218         }
2219 
2220         if (examined >= htabslots) {
2221             break;
2222         }
2223 
2224         if (index >= htabslots) {
2225             assert(index == htabslots);
2226             index = 0;
2227         }
2228     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2229 
2230     if (index >= htabslots) {
2231         assert(index == htabslots);
2232         index = 0;
2233     }
2234 
2235     spapr->htab_save_index = index;
2236 
2237     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2238 }
2239 
2240 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2241 #define MAX_KVM_BUF_SIZE    2048
2242 
2243 static int htab_save_iterate(QEMUFile *f, void *opaque)
2244 {
2245     SpaprMachineState *spapr = opaque;
2246     int fd;
2247     int rc = 0;
2248 
2249     /* Iteration header */
2250     if (!spapr->htab_shift) {
2251         qemu_put_be32(f, -1);
2252         return 1;
2253     } else {
2254         qemu_put_be32(f, 0);
2255     }
2256 
2257     if (!spapr->htab) {
2258         assert(kvm_enabled());
2259 
2260         fd = get_htab_fd(spapr);
2261         if (fd < 0) {
2262             return fd;
2263         }
2264 
2265         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2266         if (rc < 0) {
2267             return rc;
2268         }
2269     } else  if (spapr->htab_first_pass) {
2270         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2271     } else {
2272         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2273     }
2274 
2275     htab_save_end_marker(f);
2276 
2277     return rc;
2278 }
2279 
2280 static int htab_save_complete(QEMUFile *f, void *opaque)
2281 {
2282     SpaprMachineState *spapr = opaque;
2283     int fd;
2284 
2285     /* Iteration header */
2286     if (!spapr->htab_shift) {
2287         qemu_put_be32(f, -1);
2288         return 0;
2289     } else {
2290         qemu_put_be32(f, 0);
2291     }
2292 
2293     if (!spapr->htab) {
2294         int rc;
2295 
2296         assert(kvm_enabled());
2297 
2298         fd = get_htab_fd(spapr);
2299         if (fd < 0) {
2300             return fd;
2301         }
2302 
2303         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2304         if (rc < 0) {
2305             return rc;
2306         }
2307     } else {
2308         if (spapr->htab_first_pass) {
2309             htab_save_first_pass(f, spapr, -1);
2310         }
2311         htab_save_later_pass(f, spapr, -1);
2312     }
2313 
2314     /* End marker */
2315     htab_save_end_marker(f);
2316 
2317     return 0;
2318 }
2319 
2320 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2321 {
2322     SpaprMachineState *spapr = opaque;
2323     uint32_t section_hdr;
2324     int fd = -1;
2325     Error *local_err = NULL;
2326 
2327     if (version_id < 1 || version_id > 1) {
2328         error_report("htab_load() bad version");
2329         return -EINVAL;
2330     }
2331 
2332     section_hdr = qemu_get_be32(f);
2333 
2334     if (section_hdr == -1) {
2335         spapr_free_hpt(spapr);
2336         return 0;
2337     }
2338 
2339     if (section_hdr) {
2340         /* First section gives the htab size */
2341         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2342         if (local_err) {
2343             error_report_err(local_err);
2344             return -EINVAL;
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             if (rc < 0) {
2397                 return rc;
2398             }
2399         }
2400     }
2401 
2402     if (!spapr->htab) {
2403         assert(fd >= 0);
2404         close(fd);
2405     }
2406 
2407     return 0;
2408 }
2409 
2410 static void htab_save_cleanup(void *opaque)
2411 {
2412     SpaprMachineState *spapr = opaque;
2413 
2414     close_htab_fd(spapr);
2415 }
2416 
2417 static SaveVMHandlers savevm_htab_handlers = {
2418     .save_setup = htab_save_setup,
2419     .save_live_iterate = htab_save_iterate,
2420     .save_live_complete_precopy = htab_save_complete,
2421     .save_cleanup = htab_save_cleanup,
2422     .load_state = htab_load,
2423 };
2424 
2425 static void spapr_boot_set(void *opaque, const char *boot_device,
2426                            Error **errp)
2427 {
2428     MachineState *machine = MACHINE(opaque);
2429     machine->boot_order = g_strdup(boot_device);
2430 }
2431 
2432 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2433 {
2434     MachineState *machine = MACHINE(spapr);
2435     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2436     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2437     int i;
2438 
2439     for (i = 0; i < nr_lmbs; i++) {
2440         uint64_t addr;
2441 
2442         addr = i * lmb_size + machine->device_memory->base;
2443         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2444                                addr / lmb_size);
2445     }
2446 }
2447 
2448 /*
2449  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2450  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2451  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2452  */
2453 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2454 {
2455     int i;
2456 
2457     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2458         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2459                    " is not aligned to %" PRIu64 " MiB",
2460                    machine->ram_size,
2461                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2462         return;
2463     }
2464 
2465     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2466         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2467                    " is not aligned to %" PRIu64 " MiB",
2468                    machine->ram_size,
2469                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2470         return;
2471     }
2472 
2473     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2474         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2475             error_setg(errp,
2476                        "Node %d memory size 0x%" PRIx64
2477                        " is not aligned to %" PRIu64 " MiB",
2478                        i, machine->numa_state->nodes[i].node_mem,
2479                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2480             return;
2481         }
2482     }
2483 }
2484 
2485 /* find cpu slot in machine->possible_cpus by core_id */
2486 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2487 {
2488     int index = id / ms->smp.threads;
2489 
2490     if (index >= ms->possible_cpus->len) {
2491         return NULL;
2492     }
2493     if (idx) {
2494         *idx = index;
2495     }
2496     return &ms->possible_cpus->cpus[index];
2497 }
2498 
2499 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2500 {
2501     MachineState *ms = MACHINE(spapr);
2502     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2503     Error *local_err = NULL;
2504     bool vsmt_user = !!spapr->vsmt;
2505     int kvm_smt = kvmppc_smt_threads();
2506     int ret;
2507     unsigned int smp_threads = ms->smp.threads;
2508 
2509     if (!kvm_enabled() && (smp_threads > 1)) {
2510         error_setg(errp, "TCG cannot support more than 1 thread/core "
2511                    "on a pseries machine");
2512         return;
2513     }
2514     if (!is_power_of_2(smp_threads)) {
2515         error_setg(errp, "Cannot support %d threads/core on a pseries "
2516                    "machine because it must be a power of 2", smp_threads);
2517         return;
2518     }
2519 
2520     /* Detemine the VSMT mode to use: */
2521     if (vsmt_user) {
2522         if (spapr->vsmt < smp_threads) {
2523             error_setg(errp, "Cannot support VSMT mode %d"
2524                        " because it must be >= threads/core (%d)",
2525                        spapr->vsmt, smp_threads);
2526             return;
2527         }
2528         /* In this case, spapr->vsmt has been set by the command line */
2529     } else if (!smc->smp_threads_vsmt) {
2530         /*
2531          * Default VSMT value is tricky, because we need it to be as
2532          * consistent as possible (for migration), but this requires
2533          * changing it for at least some existing cases.  We pick 8 as
2534          * the value that we'd get with KVM on POWER8, the
2535          * overwhelmingly common case in production systems.
2536          */
2537         spapr->vsmt = MAX(8, smp_threads);
2538     } else {
2539         spapr->vsmt = smp_threads;
2540     }
2541 
2542     /* KVM: If necessary, set the SMT mode: */
2543     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2544         ret = kvmppc_set_smt_threads(spapr->vsmt);
2545         if (ret) {
2546             /* Looks like KVM isn't able to change VSMT mode */
2547             error_setg(&local_err,
2548                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2549                        spapr->vsmt, ret);
2550             /* We can live with that if the default one is big enough
2551              * for the number of threads, and a submultiple of the one
2552              * we want.  In this case we'll waste some vcpu ids, but
2553              * behaviour will be correct */
2554             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2555                 warn_report_err(local_err);
2556             } else {
2557                 if (!vsmt_user) {
2558                     error_append_hint(&local_err,
2559                                       "On PPC, a VM with %d threads/core"
2560                                       " on a host with %d threads/core"
2561                                       " requires the use of VSMT mode %d.\n",
2562                                       smp_threads, kvm_smt, spapr->vsmt);
2563                 }
2564                 kvmppc_error_append_smt_possible_hint(&local_err);
2565                 error_propagate(errp, local_err);
2566             }
2567         }
2568     }
2569     /* else TCG: nothing to do currently */
2570 }
2571 
2572 static void spapr_init_cpus(SpaprMachineState *spapr)
2573 {
2574     MachineState *machine = MACHINE(spapr);
2575     MachineClass *mc = MACHINE_GET_CLASS(machine);
2576     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2577     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2578     const CPUArchIdList *possible_cpus;
2579     unsigned int smp_cpus = machine->smp.cpus;
2580     unsigned int smp_threads = machine->smp.threads;
2581     unsigned int max_cpus = machine->smp.max_cpus;
2582     int boot_cores_nr = smp_cpus / smp_threads;
2583     int i;
2584 
2585     possible_cpus = mc->possible_cpu_arch_ids(machine);
2586     if (mc->has_hotpluggable_cpus) {
2587         if (smp_cpus % smp_threads) {
2588             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2589                          smp_cpus, smp_threads);
2590             exit(1);
2591         }
2592         if (max_cpus % smp_threads) {
2593             error_report("max_cpus (%u) must be multiple of threads (%u)",
2594                          max_cpus, smp_threads);
2595             exit(1);
2596         }
2597     } else {
2598         if (max_cpus != smp_cpus) {
2599             error_report("This machine version does not support CPU hotplug");
2600             exit(1);
2601         }
2602         boot_cores_nr = possible_cpus->len;
2603     }
2604 
2605     if (smc->pre_2_10_has_unused_icps) {
2606         int i;
2607 
2608         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2609             /* Dummy entries get deregistered when real ICPState objects
2610              * are registered during CPU core hotplug.
2611              */
2612             pre_2_10_vmstate_register_dummy_icp(i);
2613         }
2614     }
2615 
2616     for (i = 0; i < possible_cpus->len; i++) {
2617         int core_id = i * smp_threads;
2618 
2619         if (mc->has_hotpluggable_cpus) {
2620             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2621                                    spapr_vcpu_id(spapr, core_id));
2622         }
2623 
2624         if (i < boot_cores_nr) {
2625             Object *core  = object_new(type);
2626             int nr_threads = smp_threads;
2627 
2628             /* Handle the partially filled core for older machine types */
2629             if ((i + 1) * smp_threads >= smp_cpus) {
2630                 nr_threads = smp_cpus - i * smp_threads;
2631             }
2632 
2633             object_property_set_int(core, "nr-threads", nr_threads,
2634                                     &error_fatal);
2635             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2636                                     &error_fatal);
2637             qdev_realize(DEVICE(core), NULL, &error_fatal);
2638 
2639             object_unref(core);
2640         }
2641     }
2642 }
2643 
2644 static PCIHostState *spapr_create_default_phb(void)
2645 {
2646     DeviceState *dev;
2647 
2648     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2649     qdev_prop_set_uint32(dev, "index", 0);
2650     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2651 
2652     return PCI_HOST_BRIDGE(dev);
2653 }
2654 
2655 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2656 {
2657     MachineState *machine = MACHINE(spapr);
2658     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2659     hwaddr rma_size = machine->ram_size;
2660     hwaddr node0_size = spapr_node0_size(machine);
2661 
2662     /* RMA has to fit in the first NUMA node */
2663     rma_size = MIN(rma_size, node0_size);
2664 
2665     /*
2666      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2667      * never exceed that
2668      */
2669     rma_size = MIN(rma_size, 1 * TiB);
2670 
2671     /*
2672      * Clamp the RMA size based on machine type.  This is for
2673      * migration compatibility with older qemu versions, which limited
2674      * the RMA size for complicated and mostly bad reasons.
2675      */
2676     if (smc->rma_limit) {
2677         rma_size = MIN(rma_size, smc->rma_limit);
2678     }
2679 
2680     if (rma_size < MIN_RMA_SLOF) {
2681         error_setg(errp,
2682                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2683                    "ldMiB guest RMA (Real Mode Area memory)",
2684                    MIN_RMA_SLOF / MiB);
2685         return 0;
2686     }
2687 
2688     return rma_size;
2689 }
2690 
2691 /* pSeries LPAR / sPAPR hardware init */
2692 static void spapr_machine_init(MachineState *machine)
2693 {
2694     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2695     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2696     MachineClass *mc = MACHINE_GET_CLASS(machine);
2697     const char *kernel_filename = machine->kernel_filename;
2698     const char *initrd_filename = machine->initrd_filename;
2699     PCIHostState *phb;
2700     int i;
2701     MemoryRegion *sysmem = get_system_memory();
2702     long load_limit, fw_size;
2703     char *filename;
2704     Error *resize_hpt_err = NULL;
2705 
2706     msi_nonbroken = true;
2707 
2708     QLIST_INIT(&spapr->phbs);
2709     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2710 
2711     /* Determine capabilities to run with */
2712     spapr_caps_init(spapr);
2713 
2714     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2715     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2716         /*
2717          * If the user explicitly requested a mode we should either
2718          * supply it, or fail completely (which we do below).  But if
2719          * it's not set explicitly, we reset our mode to something
2720          * that works
2721          */
2722         if (resize_hpt_err) {
2723             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2724             error_free(resize_hpt_err);
2725             resize_hpt_err = NULL;
2726         } else {
2727             spapr->resize_hpt = smc->resize_hpt_default;
2728         }
2729     }
2730 
2731     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2732 
2733     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2734         /*
2735          * User requested HPT resize, but this host can't supply it.  Bail out
2736          */
2737         error_report_err(resize_hpt_err);
2738         exit(1);
2739     }
2740     error_free(resize_hpt_err);
2741 
2742     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2743 
2744     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2745     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2746 
2747     /*
2748      * VSMT must be set in order to be able to compute VCPU ids, ie to
2749      * call spapr_max_server_number() or spapr_vcpu_id().
2750      */
2751     spapr_set_vsmt_mode(spapr, &error_fatal);
2752 
2753     /* Set up Interrupt Controller before we create the VCPUs */
2754     spapr_irq_init(spapr, &error_fatal);
2755 
2756     /* Set up containers for ibm,client-architecture-support negotiated options
2757      */
2758     spapr->ov5 = spapr_ovec_new();
2759     spapr->ov5_cas = spapr_ovec_new();
2760 
2761     if (smc->dr_lmb_enabled) {
2762         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2763         spapr_validate_node_memory(machine, &error_fatal);
2764     }
2765 
2766     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2767 
2768     /* advertise support for dedicated HP event source to guests */
2769     if (spapr->use_hotplug_event_source) {
2770         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2771     }
2772 
2773     /* advertise support for HPT resizing */
2774     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2775         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2776     }
2777 
2778     /* advertise support for ibm,dyamic-memory-v2 */
2779     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2780 
2781     /* advertise XIVE on POWER9 machines */
2782     if (spapr->irq->xive) {
2783         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2784     }
2785 
2786     /* init CPUs */
2787     spapr_init_cpus(spapr);
2788 
2789     /*
2790      * check we don't have a memory-less/cpu-less NUMA node
2791      * Firmware relies on the existing memory/cpu topology to provide the
2792      * NUMA topology to the kernel.
2793      * And the linux kernel needs to know the NUMA topology at start
2794      * to be able to hotplug CPUs later.
2795      */
2796     if (machine->numa_state->num_nodes) {
2797         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2798             /* check for memory-less node */
2799             if (machine->numa_state->nodes[i].node_mem == 0) {
2800                 CPUState *cs;
2801                 int found = 0;
2802                 /* check for cpu-less node */
2803                 CPU_FOREACH(cs) {
2804                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2805                     if (cpu->node_id == i) {
2806                         found = 1;
2807                         break;
2808                     }
2809                 }
2810                 /* memory-less and cpu-less node */
2811                 if (!found) {
2812                     error_report(
2813                        "Memory-less/cpu-less nodes are not supported (node %d)",
2814                                  i);
2815                     exit(1);
2816                 }
2817             }
2818         }
2819 
2820     }
2821 
2822     /*
2823      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2824      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2825      * called from vPHB reset handler so we initialize the counter here.
2826      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2827      * must be equally distant from any other node.
2828      * The final value of spapr->gpu_numa_id is going to be written to
2829      * max-associativity-domains in spapr_build_fdt().
2830      */
2831     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2832 
2833     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2834         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2835                               spapr->max_compat_pvr)) {
2836         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2837         /* KVM and TCG always allow GTSE with radix... */
2838         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2839     }
2840     /* ... but not with hash (currently). */
2841 
2842     if (kvm_enabled()) {
2843         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2844         kvmppc_enable_logical_ci_hcalls();
2845         kvmppc_enable_set_mode_hcall();
2846 
2847         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2848         kvmppc_enable_clear_ref_mod_hcalls();
2849 
2850         /* Enable H_PAGE_INIT */
2851         kvmppc_enable_h_page_init();
2852     }
2853 
2854     /* map RAM */
2855     memory_region_add_subregion(sysmem, 0, machine->ram);
2856 
2857     /* always allocate the device memory information */
2858     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2859 
2860     /* initialize hotplug memory address space */
2861     if (machine->ram_size < machine->maxram_size) {
2862         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2863         /*
2864          * Limit the number of hotpluggable memory slots to half the number
2865          * slots that KVM supports, leaving the other half for PCI and other
2866          * devices. However ensure that number of slots doesn't drop below 32.
2867          */
2868         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2869                            SPAPR_MAX_RAM_SLOTS;
2870 
2871         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2872             max_memslots = SPAPR_MAX_RAM_SLOTS;
2873         }
2874         if (machine->ram_slots > max_memslots) {
2875             error_report("Specified number of memory slots %"
2876                          PRIu64" exceeds max supported %d",
2877                          machine->ram_slots, max_memslots);
2878             exit(1);
2879         }
2880 
2881         machine->device_memory->base = ROUND_UP(machine->ram_size,
2882                                                 SPAPR_DEVICE_MEM_ALIGN);
2883         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2884                            "device-memory", device_mem_size);
2885         memory_region_add_subregion(sysmem, machine->device_memory->base,
2886                                     &machine->device_memory->mr);
2887     }
2888 
2889     if (smc->dr_lmb_enabled) {
2890         spapr_create_lmb_dr_connectors(spapr);
2891     }
2892 
2893     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2894         /* Create the error string for live migration blocker */
2895         error_setg(&spapr->fwnmi_migration_blocker,
2896             "A machine check is being handled during migration. The handler"
2897             "may run and log hardware error on the destination");
2898     }
2899 
2900     if (mc->nvdimm_supported) {
2901         spapr_create_nvdimm_dr_connectors(spapr);
2902     }
2903 
2904     /* Set up RTAS event infrastructure */
2905     spapr_events_init(spapr);
2906 
2907     /* Set up the RTC RTAS interfaces */
2908     spapr_rtc_create(spapr);
2909 
2910     /* Set up VIO bus */
2911     spapr->vio_bus = spapr_vio_bus_init();
2912 
2913     for (i = 0; i < serial_max_hds(); i++) {
2914         if (serial_hd(i)) {
2915             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2916         }
2917     }
2918 
2919     /* We always have at least the nvram device on VIO */
2920     spapr_create_nvram(spapr);
2921 
2922     /*
2923      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2924      * connectors (described in root DT node's "ibm,drc-types" property)
2925      * are pre-initialized here. additional child connectors (such as
2926      * connectors for a PHBs PCI slots) are added as needed during their
2927      * parent's realization.
2928      */
2929     if (smc->dr_phb_enabled) {
2930         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2931             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2932         }
2933     }
2934 
2935     /* Set up PCI */
2936     spapr_pci_rtas_init();
2937 
2938     phb = spapr_create_default_phb();
2939 
2940     for (i = 0; i < nb_nics; i++) {
2941         NICInfo *nd = &nd_table[i];
2942 
2943         if (!nd->model) {
2944             nd->model = g_strdup("spapr-vlan");
2945         }
2946 
2947         if (g_str_equal(nd->model, "spapr-vlan") ||
2948             g_str_equal(nd->model, "ibmveth")) {
2949             spapr_vlan_create(spapr->vio_bus, nd);
2950         } else {
2951             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2952         }
2953     }
2954 
2955     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2956         spapr_vscsi_create(spapr->vio_bus);
2957     }
2958 
2959     /* Graphics */
2960     if (spapr_vga_init(phb->bus, &error_fatal)) {
2961         spapr->has_graphics = true;
2962         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2963     }
2964 
2965     if (machine->usb) {
2966         if (smc->use_ohci_by_default) {
2967             pci_create_simple(phb->bus, -1, "pci-ohci");
2968         } else {
2969             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2970         }
2971 
2972         if (spapr->has_graphics) {
2973             USBBus *usb_bus = usb_bus_find(-1);
2974 
2975             usb_create_simple(usb_bus, "usb-kbd");
2976             usb_create_simple(usb_bus, "usb-mouse");
2977         }
2978     }
2979 
2980     if (kernel_filename) {
2981         uint64_t lowaddr = 0;
2982 
2983         spapr->kernel_size = load_elf(kernel_filename, NULL,
2984                                       translate_kernel_address, spapr,
2985                                       NULL, &lowaddr, NULL, NULL, 1,
2986                                       PPC_ELF_MACHINE, 0, 0);
2987         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2988             spapr->kernel_size = load_elf(kernel_filename, NULL,
2989                                           translate_kernel_address, spapr, NULL,
2990                                           &lowaddr, NULL, NULL, 0,
2991                                           PPC_ELF_MACHINE,
2992                                           0, 0);
2993             spapr->kernel_le = spapr->kernel_size > 0;
2994         }
2995         if (spapr->kernel_size < 0) {
2996             error_report("error loading %s: %s", kernel_filename,
2997                          load_elf_strerror(spapr->kernel_size));
2998             exit(1);
2999         }
3000 
3001         /* load initrd */
3002         if (initrd_filename) {
3003             /* Try to locate the initrd in the gap between the kernel
3004              * and the firmware. Add a bit of space just in case
3005              */
3006             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
3007                                   + 0x1ffff) & ~0xffff;
3008             spapr->initrd_size = load_image_targphys(initrd_filename,
3009                                                      spapr->initrd_base,
3010                                                      load_limit
3011                                                      - spapr->initrd_base);
3012             if (spapr->initrd_size < 0) {
3013                 error_report("could not load initial ram disk '%s'",
3014                              initrd_filename);
3015                 exit(1);
3016             }
3017         }
3018     }
3019 
3020     if (bios_name == NULL) {
3021         bios_name = FW_FILE_NAME;
3022     }
3023     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
3024     if (!filename) {
3025         error_report("Could not find LPAR firmware '%s'", bios_name);
3026         exit(1);
3027     }
3028     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
3029     if (fw_size <= 0) {
3030         error_report("Could not load LPAR firmware '%s'", filename);
3031         exit(1);
3032     }
3033     g_free(filename);
3034 
3035     /* FIXME: Should register things through the MachineState's qdev
3036      * interface, this is a legacy from the sPAPREnvironment structure
3037      * which predated MachineState but had a similar function */
3038     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3039     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3040                          &savevm_htab_handlers, spapr);
3041 
3042     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
3043 
3044     qemu_register_boot_set(spapr_boot_set, spapr);
3045 
3046     /*
3047      * Nothing needs to be done to resume a suspended guest because
3048      * suspending does not change the machine state, so no need for
3049      * a ->wakeup method.
3050      */
3051     qemu_register_wakeup_support();
3052 
3053     if (kvm_enabled()) {
3054         /* to stop and start vmclock */
3055         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3056                                          &spapr->tb);
3057 
3058         kvmppc_spapr_enable_inkernel_multitce();
3059     }
3060 
3061     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3062 }
3063 
3064 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3065 {
3066     if (!vm_type) {
3067         return 0;
3068     }
3069 
3070     if (!strcmp(vm_type, "HV")) {
3071         return 1;
3072     }
3073 
3074     if (!strcmp(vm_type, "PR")) {
3075         return 2;
3076     }
3077 
3078     error_report("Unknown kvm-type specified '%s'", vm_type);
3079     exit(1);
3080 }
3081 
3082 /*
3083  * Implementation of an interface to adjust firmware path
3084  * for the bootindex property handling.
3085  */
3086 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3087                                    DeviceState *dev)
3088 {
3089 #define CAST(type, obj, name) \
3090     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3091     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3092     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3093     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3094 
3095     if (d) {
3096         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3097         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3098         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3099 
3100         if (spapr) {
3101             /*
3102              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3103              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3104              * 0x8000 | (target << 8) | (bus << 5) | lun
3105              * (see the "Logical unit addressing format" table in SAM5)
3106              */
3107             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3108             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3109                                    (uint64_t)id << 48);
3110         } else if (virtio) {
3111             /*
3112              * We use SRP luns of the form 01000000 | (target << 8) | lun
3113              * in the top 32 bits of the 64-bit LUN
3114              * Note: the quote above is from SLOF and it is wrong,
3115              * the actual binding is:
3116              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3117              */
3118             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3119             if (d->lun >= 256) {
3120                 /* Use the LUN "flat space addressing method" */
3121                 id |= 0x4000;
3122             }
3123             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3124                                    (uint64_t)id << 32);
3125         } else if (usb) {
3126             /*
3127              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3128              * in the top 32 bits of the 64-bit LUN
3129              */
3130             unsigned usb_port = atoi(usb->port->path);
3131             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3132             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3133                                    (uint64_t)id << 32);
3134         }
3135     }
3136 
3137     /*
3138      * SLOF probes the USB devices, and if it recognizes that the device is a
3139      * storage device, it changes its name to "storage" instead of "usb-host",
3140      * and additionally adds a child node for the SCSI LUN, so the correct
3141      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3142      */
3143     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3144         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3145         if (usb_host_dev_is_scsi_storage(usbdev)) {
3146             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3147         }
3148     }
3149 
3150     if (phb) {
3151         /* Replace "pci" with "pci@800000020000000" */
3152         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3153     }
3154 
3155     if (vsc) {
3156         /* Same logic as virtio above */
3157         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3158         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3159     }
3160 
3161     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3162         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3163         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3164         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3165     }
3166 
3167     return NULL;
3168 }
3169 
3170 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3171 {
3172     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3173 
3174     return g_strdup(spapr->kvm_type);
3175 }
3176 
3177 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3178 {
3179     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3180 
3181     g_free(spapr->kvm_type);
3182     spapr->kvm_type = g_strdup(value);
3183 }
3184 
3185 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3186 {
3187     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3188 
3189     return spapr->use_hotplug_event_source;
3190 }
3191 
3192 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3193                                             Error **errp)
3194 {
3195     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3196 
3197     spapr->use_hotplug_event_source = value;
3198 }
3199 
3200 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3201 {
3202     return true;
3203 }
3204 
3205 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3206 {
3207     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3208 
3209     switch (spapr->resize_hpt) {
3210     case SPAPR_RESIZE_HPT_DEFAULT:
3211         return g_strdup("default");
3212     case SPAPR_RESIZE_HPT_DISABLED:
3213         return g_strdup("disabled");
3214     case SPAPR_RESIZE_HPT_ENABLED:
3215         return g_strdup("enabled");
3216     case SPAPR_RESIZE_HPT_REQUIRED:
3217         return g_strdup("required");
3218     }
3219     g_assert_not_reached();
3220 }
3221 
3222 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3223 {
3224     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3225 
3226     if (strcmp(value, "default") == 0) {
3227         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3228     } else if (strcmp(value, "disabled") == 0) {
3229         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3230     } else if (strcmp(value, "enabled") == 0) {
3231         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3232     } else if (strcmp(value, "required") == 0) {
3233         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3234     } else {
3235         error_setg(errp, "Bad value for \"resize-hpt\" property");
3236     }
3237 }
3238 
3239 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3240 {
3241     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3242 
3243     if (spapr->irq == &spapr_irq_xics_legacy) {
3244         return g_strdup("legacy");
3245     } else if (spapr->irq == &spapr_irq_xics) {
3246         return g_strdup("xics");
3247     } else if (spapr->irq == &spapr_irq_xive) {
3248         return g_strdup("xive");
3249     } else if (spapr->irq == &spapr_irq_dual) {
3250         return g_strdup("dual");
3251     }
3252     g_assert_not_reached();
3253 }
3254 
3255 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3256 {
3257     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3258 
3259     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3260         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3261         return;
3262     }
3263 
3264     /* The legacy IRQ backend can not be set */
3265     if (strcmp(value, "xics") == 0) {
3266         spapr->irq = &spapr_irq_xics;
3267     } else if (strcmp(value, "xive") == 0) {
3268         spapr->irq = &spapr_irq_xive;
3269     } else if (strcmp(value, "dual") == 0) {
3270         spapr->irq = &spapr_irq_dual;
3271     } else {
3272         error_setg(errp, "Bad value for \"ic-mode\" property");
3273     }
3274 }
3275 
3276 static char *spapr_get_host_model(Object *obj, Error **errp)
3277 {
3278     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3279 
3280     return g_strdup(spapr->host_model);
3281 }
3282 
3283 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3284 {
3285     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3286 
3287     g_free(spapr->host_model);
3288     spapr->host_model = g_strdup(value);
3289 }
3290 
3291 static char *spapr_get_host_serial(Object *obj, Error **errp)
3292 {
3293     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3294 
3295     return g_strdup(spapr->host_serial);
3296 }
3297 
3298 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3299 {
3300     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3301 
3302     g_free(spapr->host_serial);
3303     spapr->host_serial = g_strdup(value);
3304 }
3305 
3306 static void spapr_instance_init(Object *obj)
3307 {
3308     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3309     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3310 
3311     spapr->htab_fd = -1;
3312     spapr->use_hotplug_event_source = true;
3313     object_property_add_str(obj, "kvm-type",
3314                             spapr_get_kvm_type, spapr_set_kvm_type);
3315     object_property_set_description(obj, "kvm-type",
3316                                     "Specifies the KVM virtualization mode (HV, PR)");
3317     object_property_add_bool(obj, "modern-hotplug-events",
3318                             spapr_get_modern_hotplug_events,
3319                             spapr_set_modern_hotplug_events);
3320     object_property_set_description(obj, "modern-hotplug-events",
3321                                     "Use dedicated hotplug event mechanism in"
3322                                     " place of standard EPOW events when possible"
3323                                     " (required for memory hot-unplug support)");
3324     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3325                             "Maximum permitted CPU compatibility mode");
3326 
3327     object_property_add_str(obj, "resize-hpt",
3328                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3329     object_property_set_description(obj, "resize-hpt",
3330                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3331     object_property_add_uint32_ptr(obj, "vsmt",
3332                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3333     object_property_set_description(obj, "vsmt",
3334                                     "Virtual SMT: KVM behaves as if this were"
3335                                     " the host's SMT mode");
3336 
3337     object_property_add_bool(obj, "vfio-no-msix-emulation",
3338                              spapr_get_msix_emulation, NULL);
3339 
3340     object_property_add_uint64_ptr(obj, "kernel-addr",
3341                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3342     object_property_set_description(obj, "kernel-addr",
3343                                     stringify(KERNEL_LOAD_ADDR)
3344                                     " for -kernel is the default");
3345     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3346     /* The machine class defines the default interrupt controller mode */
3347     spapr->irq = smc->irq;
3348     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3349                             spapr_set_ic_mode);
3350     object_property_set_description(obj, "ic-mode",
3351                  "Specifies the interrupt controller mode (xics, xive, dual)");
3352 
3353     object_property_add_str(obj, "host-model",
3354         spapr_get_host_model, spapr_set_host_model);
3355     object_property_set_description(obj, "host-model",
3356         "Host model to advertise in guest device tree");
3357     object_property_add_str(obj, "host-serial",
3358         spapr_get_host_serial, spapr_set_host_serial);
3359     object_property_set_description(obj, "host-serial",
3360         "Host serial number to advertise in guest device tree");
3361 }
3362 
3363 static void spapr_machine_finalizefn(Object *obj)
3364 {
3365     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3366 
3367     g_free(spapr->kvm_type);
3368 }
3369 
3370 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3371 {
3372     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3373     PowerPCCPU *cpu = POWERPC_CPU(cs);
3374     CPUPPCState *env = &cpu->env;
3375 
3376     cpu_synchronize_state(cs);
3377     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3378     if (spapr->fwnmi_system_reset_addr != -1) {
3379         uint64_t rtas_addr, addr;
3380 
3381         /* get rtas addr from fdt */
3382         rtas_addr = spapr_get_rtas_addr();
3383         if (!rtas_addr) {
3384             qemu_system_guest_panicked(NULL);
3385             return;
3386         }
3387 
3388         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3389         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3390         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3391         env->gpr[3] = addr;
3392     }
3393     ppc_cpu_do_system_reset(cs);
3394     if (spapr->fwnmi_system_reset_addr != -1) {
3395         env->nip = spapr->fwnmi_system_reset_addr;
3396     }
3397 }
3398 
3399 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3400 {
3401     CPUState *cs;
3402 
3403     CPU_FOREACH(cs) {
3404         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3405     }
3406 }
3407 
3408 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3409                           void *fdt, int *fdt_start_offset, Error **errp)
3410 {
3411     uint64_t addr;
3412     uint32_t node;
3413 
3414     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3415     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3416                                     &error_abort);
3417     *fdt_start_offset = spapr_dt_memory_node(fdt, node, addr,
3418                                              SPAPR_MEMORY_BLOCK_SIZE);
3419     return 0;
3420 }
3421 
3422 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3423                            bool dedicated_hp_event_source, Error **errp)
3424 {
3425     SpaprDrc *drc;
3426     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3427     int i;
3428     uint64_t addr = addr_start;
3429     bool hotplugged = spapr_drc_hotplugged(dev);
3430     Error *local_err = NULL;
3431 
3432     for (i = 0; i < nr_lmbs; i++) {
3433         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3434                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3435         g_assert(drc);
3436 
3437         spapr_drc_attach(drc, dev, &local_err);
3438         if (local_err) {
3439             while (addr > addr_start) {
3440                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3441                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3442                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3443                 spapr_drc_detach(drc);
3444             }
3445             error_propagate(errp, local_err);
3446             return;
3447         }
3448         if (!hotplugged) {
3449             spapr_drc_reset(drc);
3450         }
3451         addr += SPAPR_MEMORY_BLOCK_SIZE;
3452     }
3453     /* send hotplug notification to the
3454      * guest only in case of hotplugged memory
3455      */
3456     if (hotplugged) {
3457         if (dedicated_hp_event_source) {
3458             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3459                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3460             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3461                                                    nr_lmbs,
3462                                                    spapr_drc_index(drc));
3463         } else {
3464             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3465                                            nr_lmbs);
3466         }
3467     }
3468 }
3469 
3470 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3471                               Error **errp)
3472 {
3473     Error *local_err = NULL;
3474     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3475     PCDIMMDevice *dimm = PC_DIMM(dev);
3476     uint64_t size, addr, slot;
3477     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3478 
3479     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3480 
3481     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3482     if (local_err) {
3483         goto out;
3484     }
3485 
3486     if (!is_nvdimm) {
3487         addr = object_property_get_uint(OBJECT(dimm),
3488                                         PC_DIMM_ADDR_PROP, &local_err);
3489         if (local_err) {
3490             goto out_unplug;
3491         }
3492         spapr_add_lmbs(dev, addr, size,
3493                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3494                        &local_err);
3495     } else {
3496         slot = object_property_get_uint(OBJECT(dimm),
3497                                         PC_DIMM_SLOT_PROP, &local_err);
3498         if (local_err) {
3499             goto out_unplug;
3500         }
3501         spapr_add_nvdimm(dev, slot, &local_err);
3502     }
3503 
3504     if (local_err) {
3505         goto out_unplug;
3506     }
3507 
3508     return;
3509 
3510 out_unplug:
3511     pc_dimm_unplug(dimm, MACHINE(ms));
3512 out:
3513     error_propagate(errp, local_err);
3514 }
3515 
3516 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3517                                   Error **errp)
3518 {
3519     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3520     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3521     const MachineClass *mc = MACHINE_CLASS(smc);
3522     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3523     PCDIMMDevice *dimm = PC_DIMM(dev);
3524     Error *local_err = NULL;
3525     uint64_t size;
3526     Object *memdev;
3527     hwaddr pagesize;
3528 
3529     if (!smc->dr_lmb_enabled) {
3530         error_setg(errp, "Memory hotplug not supported for this machine");
3531         return;
3532     }
3533 
3534     if (is_nvdimm && !mc->nvdimm_supported) {
3535         error_setg(errp, "NVDIMM hotplug not supported for this machine");
3536         return;
3537     }
3538 
3539     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3540     if (local_err) {
3541         error_propagate(errp, local_err);
3542         return;
3543     }
3544 
3545     if (!is_nvdimm && size % SPAPR_MEMORY_BLOCK_SIZE) {
3546         error_setg(errp, "Hotplugged memory size must be a multiple of "
3547                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3548         return;
3549     } else if (is_nvdimm) {
3550         spapr_nvdimm_validate_opts(NVDIMM(dev), size, &local_err);
3551         if (local_err) {
3552             error_propagate(errp, local_err);
3553             return;
3554         }
3555     }
3556 
3557     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3558                                       &error_abort);
3559     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3560     spapr_check_pagesize(spapr, pagesize, &local_err);
3561     if (local_err) {
3562         error_propagate(errp, local_err);
3563         return;
3564     }
3565 
3566     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3567 }
3568 
3569 struct SpaprDimmState {
3570     PCDIMMDevice *dimm;
3571     uint32_t nr_lmbs;
3572     QTAILQ_ENTRY(SpaprDimmState) next;
3573 };
3574 
3575 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3576                                                        PCDIMMDevice *dimm)
3577 {
3578     SpaprDimmState *dimm_state = NULL;
3579 
3580     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3581         if (dimm_state->dimm == dimm) {
3582             break;
3583         }
3584     }
3585     return dimm_state;
3586 }
3587 
3588 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3589                                                       uint32_t nr_lmbs,
3590                                                       PCDIMMDevice *dimm)
3591 {
3592     SpaprDimmState *ds = NULL;
3593 
3594     /*
3595      * If this request is for a DIMM whose removal had failed earlier
3596      * (due to guest's refusal to remove the LMBs), we would have this
3597      * dimm already in the pending_dimm_unplugs list. In that
3598      * case don't add again.
3599      */
3600     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3601     if (!ds) {
3602         ds = g_malloc0(sizeof(SpaprDimmState));
3603         ds->nr_lmbs = nr_lmbs;
3604         ds->dimm = dimm;
3605         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3606     }
3607     return ds;
3608 }
3609 
3610 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3611                                               SpaprDimmState *dimm_state)
3612 {
3613     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3614     g_free(dimm_state);
3615 }
3616 
3617 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3618                                                         PCDIMMDevice *dimm)
3619 {
3620     SpaprDrc *drc;
3621     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3622                                                   &error_abort);
3623     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3624     uint32_t avail_lmbs = 0;
3625     uint64_t addr_start, addr;
3626     int i;
3627 
3628     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3629                                          &error_abort);
3630 
3631     addr = addr_start;
3632     for (i = 0; i < nr_lmbs; i++) {
3633         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3634                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3635         g_assert(drc);
3636         if (drc->dev) {
3637             avail_lmbs++;
3638         }
3639         addr += SPAPR_MEMORY_BLOCK_SIZE;
3640     }
3641 
3642     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3643 }
3644 
3645 /* Callback to be called during DRC release. */
3646 void spapr_lmb_release(DeviceState *dev)
3647 {
3648     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3649     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3650     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3651 
3652     /* This information will get lost if a migration occurs
3653      * during the unplug process. In this case recover it. */
3654     if (ds == NULL) {
3655         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3656         g_assert(ds);
3657         /* The DRC being examined by the caller at least must be counted */
3658         g_assert(ds->nr_lmbs);
3659     }
3660 
3661     if (--ds->nr_lmbs) {
3662         return;
3663     }
3664 
3665     /*
3666      * Now that all the LMBs have been removed by the guest, call the
3667      * unplug handler chain. This can never fail.
3668      */
3669     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3670     object_unparent(OBJECT(dev));
3671 }
3672 
3673 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3674 {
3675     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3676     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3677 
3678     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3679     qdev_unrealize(dev);
3680     spapr_pending_dimm_unplugs_remove(spapr, ds);
3681 }
3682 
3683 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3684                                         DeviceState *dev, Error **errp)
3685 {
3686     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3687     Error *local_err = NULL;
3688     PCDIMMDevice *dimm = PC_DIMM(dev);
3689     uint32_t nr_lmbs;
3690     uint64_t size, addr_start, addr;
3691     int i;
3692     SpaprDrc *drc;
3693 
3694     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3695         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3696         return;
3697     }
3698 
3699     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3700     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3701 
3702     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3703                                          &local_err);
3704     if (local_err) {
3705         error_propagate(errp, local_err);
3706         return;
3707     }
3708 
3709     /*
3710      * An existing pending dimm state for this DIMM means that there is an
3711      * unplug operation in progress, waiting for the spapr_lmb_release
3712      * callback to complete the job (BQL can't cover that far). In this case,
3713      * bail out to avoid detaching DRCs that were already released.
3714      */
3715     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3716         error_setg(errp, "Memory unplug already in progress for device %s",
3717                    dev->id);
3718         return;
3719     }
3720 
3721     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3722 
3723     addr = addr_start;
3724     for (i = 0; i < nr_lmbs; i++) {
3725         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3726                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3727         g_assert(drc);
3728 
3729         spapr_drc_detach(drc);
3730         addr += SPAPR_MEMORY_BLOCK_SIZE;
3731     }
3732 
3733     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3734                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3735     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3736                                               nr_lmbs, spapr_drc_index(drc));
3737 }
3738 
3739 /* Callback to be called during DRC release. */
3740 void spapr_core_release(DeviceState *dev)
3741 {
3742     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3743 
3744     /* Call the unplug handler chain. This can never fail. */
3745     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3746     object_unparent(OBJECT(dev));
3747 }
3748 
3749 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3750 {
3751     MachineState *ms = MACHINE(hotplug_dev);
3752     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3753     CPUCore *cc = CPU_CORE(dev);
3754     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3755 
3756     if (smc->pre_2_10_has_unused_icps) {
3757         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3758         int i;
3759 
3760         for (i = 0; i < cc->nr_threads; i++) {
3761             CPUState *cs = CPU(sc->threads[i]);
3762 
3763             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3764         }
3765     }
3766 
3767     assert(core_slot);
3768     core_slot->cpu = NULL;
3769     qdev_unrealize(dev);
3770 }
3771 
3772 static
3773 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3774                                Error **errp)
3775 {
3776     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3777     int index;
3778     SpaprDrc *drc;
3779     CPUCore *cc = CPU_CORE(dev);
3780 
3781     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3782         error_setg(errp, "Unable to find CPU core with core-id: %d",
3783                    cc->core_id);
3784         return;
3785     }
3786     if (index == 0) {
3787         error_setg(errp, "Boot CPU core may not be unplugged");
3788         return;
3789     }
3790 
3791     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3792                           spapr_vcpu_id(spapr, cc->core_id));
3793     g_assert(drc);
3794 
3795     if (!spapr_drc_unplug_requested(drc)) {
3796         spapr_drc_detach(drc);
3797         spapr_hotplug_req_remove_by_index(drc);
3798     }
3799 }
3800 
3801 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3802                            void *fdt, int *fdt_start_offset, Error **errp)
3803 {
3804     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3805     CPUState *cs = CPU(core->threads[0]);
3806     PowerPCCPU *cpu = POWERPC_CPU(cs);
3807     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3808     int id = spapr_get_vcpu_id(cpu);
3809     char *nodename;
3810     int offset;
3811 
3812     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3813     offset = fdt_add_subnode(fdt, 0, nodename);
3814     g_free(nodename);
3815 
3816     spapr_dt_cpu(cs, fdt, offset, spapr);
3817 
3818     *fdt_start_offset = offset;
3819     return 0;
3820 }
3821 
3822 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3823                             Error **errp)
3824 {
3825     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3826     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3827     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3828     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3829     CPUCore *cc = CPU_CORE(dev);
3830     CPUState *cs;
3831     SpaprDrc *drc;
3832     Error *local_err = NULL;
3833     CPUArchId *core_slot;
3834     int index;
3835     bool hotplugged = spapr_drc_hotplugged(dev);
3836     int i;
3837 
3838     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3839     if (!core_slot) {
3840         error_setg(errp, "Unable to find CPU core with core-id: %d",
3841                    cc->core_id);
3842         return;
3843     }
3844     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3845                           spapr_vcpu_id(spapr, cc->core_id));
3846 
3847     g_assert(drc || !mc->has_hotpluggable_cpus);
3848 
3849     if (drc) {
3850         spapr_drc_attach(drc, dev, &local_err);
3851         if (local_err) {
3852             error_propagate(errp, local_err);
3853             return;
3854         }
3855 
3856         if (hotplugged) {
3857             /*
3858              * Send hotplug notification interrupt to the guest only
3859              * in case of hotplugged CPUs.
3860              */
3861             spapr_hotplug_req_add_by_index(drc);
3862         } else {
3863             spapr_drc_reset(drc);
3864         }
3865     }
3866 
3867     core_slot->cpu = OBJECT(dev);
3868 
3869     if (smc->pre_2_10_has_unused_icps) {
3870         for (i = 0; i < cc->nr_threads; i++) {
3871             cs = CPU(core->threads[i]);
3872             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3873         }
3874     }
3875 
3876     /*
3877      * Set compatibility mode to match the boot CPU, which was either set
3878      * by the machine reset code or by CAS.
3879      */
3880     if (hotplugged) {
3881         for (i = 0; i < cc->nr_threads; i++) {
3882             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3883                            &local_err);
3884             if (local_err) {
3885                 error_propagate(errp, local_err);
3886                 return;
3887             }
3888         }
3889     }
3890 }
3891 
3892 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3893                                 Error **errp)
3894 {
3895     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3896     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3897     CPUCore *cc = CPU_CORE(dev);
3898     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3899     const char *type = object_get_typename(OBJECT(dev));
3900     CPUArchId *core_slot;
3901     int index;
3902     unsigned int smp_threads = machine->smp.threads;
3903 
3904     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3905         error_setg(errp, "CPU hotplug not supported for this machine");
3906         return;
3907     }
3908 
3909     if (strcmp(base_core_type, type)) {
3910         error_setg(errp, "CPU core type should be %s", base_core_type);
3911         return;
3912     }
3913 
3914     if (cc->core_id % smp_threads) {
3915         error_setg(errp, "invalid core id %d", cc->core_id);
3916         return;
3917     }
3918 
3919     /*
3920      * In general we should have homogeneous threads-per-core, but old
3921      * (pre hotplug support) machine types allow the last core to have
3922      * reduced threads as a compatibility hack for when we allowed
3923      * total vcpus not a multiple of threads-per-core.
3924      */
3925     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3926         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
3927                    smp_threads);
3928         return;
3929     }
3930 
3931     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3932     if (!core_slot) {
3933         error_setg(errp, "core id %d out of range", cc->core_id);
3934         return;
3935     }
3936 
3937     if (core_slot->cpu) {
3938         error_setg(errp, "core %d already populated", cc->core_id);
3939         return;
3940     }
3941 
3942     numa_cpu_pre_plug(core_slot, dev, errp);
3943 }
3944 
3945 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3946                           void *fdt, int *fdt_start_offset, Error **errp)
3947 {
3948     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3949     int intc_phandle;
3950 
3951     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3952     if (intc_phandle <= 0) {
3953         return -1;
3954     }
3955 
3956     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3957         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3958         return -1;
3959     }
3960 
3961     /* generally SLOF creates these, for hotplug it's up to QEMU */
3962     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3963 
3964     return 0;
3965 }
3966 
3967 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3968                                Error **errp)
3969 {
3970     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3971     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3972     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3973     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3974 
3975     if (dev->hotplugged && !smc->dr_phb_enabled) {
3976         error_setg(errp, "PHB hotplug not supported for this machine");
3977         return;
3978     }
3979 
3980     if (sphb->index == (uint32_t)-1) {
3981         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3982         return;
3983     }
3984 
3985     /*
3986      * This will check that sphb->index doesn't exceed the maximum number of
3987      * PHBs for the current machine type.
3988      */
3989     smc->phb_placement(spapr, sphb->index,
3990                        &sphb->buid, &sphb->io_win_addr,
3991                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
3992                        windows_supported, sphb->dma_liobn,
3993                        &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3994                        errp);
3995 }
3996 
3997 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3998                            Error **errp)
3999 {
4000     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4001     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4002     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4003     SpaprDrc *drc;
4004     bool hotplugged = spapr_drc_hotplugged(dev);
4005     Error *local_err = NULL;
4006 
4007     if (!smc->dr_phb_enabled) {
4008         return;
4009     }
4010 
4011     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4012     /* hotplug hooks should check it's enabled before getting this far */
4013     assert(drc);
4014 
4015     spapr_drc_attach(drc, dev, &local_err);
4016     if (local_err) {
4017         error_propagate(errp, local_err);
4018         return;
4019     }
4020 
4021     if (hotplugged) {
4022         spapr_hotplug_req_add_by_index(drc);
4023     } else {
4024         spapr_drc_reset(drc);
4025     }
4026 }
4027 
4028 void spapr_phb_release(DeviceState *dev)
4029 {
4030     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4031 
4032     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4033     object_unparent(OBJECT(dev));
4034 }
4035 
4036 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4037 {
4038     qdev_unrealize(dev);
4039 }
4040 
4041 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4042                                      DeviceState *dev, Error **errp)
4043 {
4044     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4045     SpaprDrc *drc;
4046 
4047     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4048     assert(drc);
4049 
4050     if (!spapr_drc_unplug_requested(drc)) {
4051         spapr_drc_detach(drc);
4052         spapr_hotplug_req_remove_by_index(drc);
4053     }
4054 }
4055 
4056 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4057                                  Error **errp)
4058 {
4059     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4060     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4061 
4062     if (spapr->tpm_proxy != NULL) {
4063         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4064         return;
4065     }
4066 
4067     spapr->tpm_proxy = tpm_proxy;
4068 }
4069 
4070 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4071 {
4072     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4073 
4074     qdev_unrealize(dev);
4075     object_unparent(OBJECT(dev));
4076     spapr->tpm_proxy = NULL;
4077 }
4078 
4079 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4080                                       DeviceState *dev, Error **errp)
4081 {
4082     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4083         spapr_memory_plug(hotplug_dev, dev, errp);
4084     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4085         spapr_core_plug(hotplug_dev, dev, errp);
4086     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4087         spapr_phb_plug(hotplug_dev, dev, errp);
4088     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4089         spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
4090     }
4091 }
4092 
4093 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4094                                         DeviceState *dev, Error **errp)
4095 {
4096     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4097         spapr_memory_unplug(hotplug_dev, dev);
4098     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4099         spapr_core_unplug(hotplug_dev, dev);
4100     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4101         spapr_phb_unplug(hotplug_dev, dev);
4102     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4103         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4104     }
4105 }
4106 
4107 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4108                                                 DeviceState *dev, Error **errp)
4109 {
4110     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4111     MachineClass *mc = MACHINE_GET_CLASS(sms);
4112     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4113 
4114     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4115         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4116             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4117         } else {
4118             /* NOTE: this means there is a window after guest reset, prior to
4119              * CAS negotiation, where unplug requests will fail due to the
4120              * capability not being detected yet. This is a bit different than
4121              * the case with PCI unplug, where the events will be queued and
4122              * eventually handled by the guest after boot
4123              */
4124             error_setg(errp, "Memory hot unplug not supported for this guest");
4125         }
4126     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4127         if (!mc->has_hotpluggable_cpus) {
4128             error_setg(errp, "CPU hot unplug not supported on this machine");
4129             return;
4130         }
4131         spapr_core_unplug_request(hotplug_dev, dev, errp);
4132     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4133         if (!smc->dr_phb_enabled) {
4134             error_setg(errp, "PHB hot unplug not supported on this machine");
4135             return;
4136         }
4137         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4138     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4139         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4140     }
4141 }
4142 
4143 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4144                                           DeviceState *dev, Error **errp)
4145 {
4146     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4147         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4148     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4149         spapr_core_pre_plug(hotplug_dev, dev, errp);
4150     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4151         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4152     }
4153 }
4154 
4155 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4156                                                  DeviceState *dev)
4157 {
4158     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4159         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4160         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4161         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4162         return HOTPLUG_HANDLER(machine);
4163     }
4164     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4165         PCIDevice *pcidev = PCI_DEVICE(dev);
4166         PCIBus *root = pci_device_root_bus(pcidev);
4167         SpaprPhbState *phb =
4168             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4169                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4170 
4171         if (phb) {
4172             return HOTPLUG_HANDLER(phb);
4173         }
4174     }
4175     return NULL;
4176 }
4177 
4178 static CpuInstanceProperties
4179 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4180 {
4181     CPUArchId *core_slot;
4182     MachineClass *mc = MACHINE_GET_CLASS(machine);
4183 
4184     /* make sure possible_cpu are intialized */
4185     mc->possible_cpu_arch_ids(machine);
4186     /* get CPU core slot containing thread that matches cpu_index */
4187     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4188     assert(core_slot);
4189     return core_slot->props;
4190 }
4191 
4192 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4193 {
4194     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4195 }
4196 
4197 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4198 {
4199     int i;
4200     unsigned int smp_threads = machine->smp.threads;
4201     unsigned int smp_cpus = machine->smp.cpus;
4202     const char *core_type;
4203     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4204     MachineClass *mc = MACHINE_GET_CLASS(machine);
4205 
4206     if (!mc->has_hotpluggable_cpus) {
4207         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4208     }
4209     if (machine->possible_cpus) {
4210         assert(machine->possible_cpus->len == spapr_max_cores);
4211         return machine->possible_cpus;
4212     }
4213 
4214     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4215     if (!core_type) {
4216         error_report("Unable to find sPAPR CPU Core definition");
4217         exit(1);
4218     }
4219 
4220     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4221                              sizeof(CPUArchId) * spapr_max_cores);
4222     machine->possible_cpus->len = spapr_max_cores;
4223     for (i = 0; i < machine->possible_cpus->len; i++) {
4224         int core_id = i * smp_threads;
4225 
4226         machine->possible_cpus->cpus[i].type = core_type;
4227         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4228         machine->possible_cpus->cpus[i].arch_id = core_id;
4229         machine->possible_cpus->cpus[i].props.has_core_id = true;
4230         machine->possible_cpus->cpus[i].props.core_id = core_id;
4231     }
4232     return machine->possible_cpus;
4233 }
4234 
4235 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4236                                 uint64_t *buid, hwaddr *pio,
4237                                 hwaddr *mmio32, hwaddr *mmio64,
4238                                 unsigned n_dma, uint32_t *liobns,
4239                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4240 {
4241     /*
4242      * New-style PHB window placement.
4243      *
4244      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4245      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4246      * windows.
4247      *
4248      * Some guest kernels can't work with MMIO windows above 1<<46
4249      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4250      *
4251      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4252      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4253      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4254      * 1TiB 64-bit MMIO windows for each PHB.
4255      */
4256     const uint64_t base_buid = 0x800000020000000ULL;
4257     int i;
4258 
4259     /* Sanity check natural alignments */
4260     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4261     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4262     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4263     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4264     /* Sanity check bounds */
4265     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4266                       SPAPR_PCI_MEM32_WIN_SIZE);
4267     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4268                       SPAPR_PCI_MEM64_WIN_SIZE);
4269 
4270     if (index >= SPAPR_MAX_PHBS) {
4271         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4272                    SPAPR_MAX_PHBS - 1);
4273         return;
4274     }
4275 
4276     *buid = base_buid + index;
4277     for (i = 0; i < n_dma; ++i) {
4278         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4279     }
4280 
4281     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4282     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4283     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4284 
4285     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4286     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4287 }
4288 
4289 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4290 {
4291     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4292 
4293     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4294 }
4295 
4296 static void spapr_ics_resend(XICSFabric *dev)
4297 {
4298     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4299 
4300     ics_resend(spapr->ics);
4301 }
4302 
4303 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4304 {
4305     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4306 
4307     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4308 }
4309 
4310 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4311                                  Monitor *mon)
4312 {
4313     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4314 
4315     spapr_irq_print_info(spapr, mon);
4316     monitor_printf(mon, "irqchip: %s\n",
4317                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4318 }
4319 
4320 /*
4321  * This is a XIVE only operation
4322  */
4323 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4324                            uint8_t nvt_blk, uint32_t nvt_idx,
4325                            bool cam_ignore, uint8_t priority,
4326                            uint32_t logic_serv, XiveTCTXMatch *match)
4327 {
4328     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4329     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4330     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4331     int count;
4332 
4333     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4334                            priority, logic_serv, match);
4335     if (count < 0) {
4336         return count;
4337     }
4338 
4339     /*
4340      * When we implement the save and restore of the thread interrupt
4341      * contexts in the enter/exit CPU handlers of the machine and the
4342      * escalations in QEMU, we should be able to handle non dispatched
4343      * vCPUs.
4344      *
4345      * Until this is done, the sPAPR machine should find at least one
4346      * matching context always.
4347      */
4348     if (count == 0) {
4349         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4350                       nvt_blk, nvt_idx);
4351     }
4352 
4353     return count;
4354 }
4355 
4356 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4357 {
4358     return cpu->vcpu_id;
4359 }
4360 
4361 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4362 {
4363     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4364     MachineState *ms = MACHINE(spapr);
4365     int vcpu_id;
4366 
4367     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4368 
4369     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4370         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4371         error_append_hint(errp, "Adjust the number of cpus to %d "
4372                           "or try to raise the number of threads per core\n",
4373                           vcpu_id * ms->smp.threads / spapr->vsmt);
4374         return;
4375     }
4376 
4377     cpu->vcpu_id = vcpu_id;
4378 }
4379 
4380 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4381 {
4382     CPUState *cs;
4383 
4384     CPU_FOREACH(cs) {
4385         PowerPCCPU *cpu = POWERPC_CPU(cs);
4386 
4387         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4388             return cpu;
4389         }
4390     }
4391 
4392     return NULL;
4393 }
4394 
4395 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4396 {
4397     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4398 
4399     /* These are only called by TCG, KVM maintains dispatch state */
4400 
4401     spapr_cpu->prod = false;
4402     if (spapr_cpu->vpa_addr) {
4403         CPUState *cs = CPU(cpu);
4404         uint32_t dispatch;
4405 
4406         dispatch = ldl_be_phys(cs->as,
4407                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4408         dispatch++;
4409         if ((dispatch & 1) != 0) {
4410             qemu_log_mask(LOG_GUEST_ERROR,
4411                           "VPA: incorrect dispatch counter value for "
4412                           "dispatched partition %u, correcting.\n", dispatch);
4413             dispatch++;
4414         }
4415         stl_be_phys(cs->as,
4416                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4417     }
4418 }
4419 
4420 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4421 {
4422     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4423 
4424     if (spapr_cpu->vpa_addr) {
4425         CPUState *cs = CPU(cpu);
4426         uint32_t dispatch;
4427 
4428         dispatch = ldl_be_phys(cs->as,
4429                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4430         dispatch++;
4431         if ((dispatch & 1) != 1) {
4432             qemu_log_mask(LOG_GUEST_ERROR,
4433                           "VPA: incorrect dispatch counter value for "
4434                           "preempted partition %u, correcting.\n", dispatch);
4435             dispatch++;
4436         }
4437         stl_be_phys(cs->as,
4438                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4439     }
4440 }
4441 
4442 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4443 {
4444     MachineClass *mc = MACHINE_CLASS(oc);
4445     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4446     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4447     NMIClass *nc = NMI_CLASS(oc);
4448     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4449     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4450     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4451     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4452     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4453 
4454     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4455     mc->ignore_boot_device_suffixes = true;
4456 
4457     /*
4458      * We set up the default / latest behaviour here.  The class_init
4459      * functions for the specific versioned machine types can override
4460      * these details for backwards compatibility
4461      */
4462     mc->init = spapr_machine_init;
4463     mc->reset = spapr_machine_reset;
4464     mc->block_default_type = IF_SCSI;
4465     mc->max_cpus = 1024;
4466     mc->no_parallel = 1;
4467     mc->default_boot_order = "";
4468     mc->default_ram_size = 512 * MiB;
4469     mc->default_ram_id = "ppc_spapr.ram";
4470     mc->default_display = "std";
4471     mc->kvm_type = spapr_kvm_type;
4472     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4473     mc->pci_allow_0_address = true;
4474     assert(!mc->get_hotplug_handler);
4475     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4476     hc->pre_plug = spapr_machine_device_pre_plug;
4477     hc->plug = spapr_machine_device_plug;
4478     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4479     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4480     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4481     hc->unplug_request = spapr_machine_device_unplug_request;
4482     hc->unplug = spapr_machine_device_unplug;
4483 
4484     smc->dr_lmb_enabled = true;
4485     smc->update_dt_enabled = true;
4486     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4487     mc->has_hotpluggable_cpus = true;
4488     mc->nvdimm_supported = true;
4489     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4490     fwc->get_dev_path = spapr_get_fw_dev_path;
4491     nc->nmi_monitor_handler = spapr_nmi;
4492     smc->phb_placement = spapr_phb_placement;
4493     vhc->hypercall = emulate_spapr_hypercall;
4494     vhc->hpt_mask = spapr_hpt_mask;
4495     vhc->map_hptes = spapr_map_hptes;
4496     vhc->unmap_hptes = spapr_unmap_hptes;
4497     vhc->hpte_set_c = spapr_hpte_set_c;
4498     vhc->hpte_set_r = spapr_hpte_set_r;
4499     vhc->get_pate = spapr_get_pate;
4500     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4501     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4502     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4503     xic->ics_get = spapr_ics_get;
4504     xic->ics_resend = spapr_ics_resend;
4505     xic->icp_get = spapr_icp_get;
4506     ispc->print_info = spapr_pic_print_info;
4507     /* Force NUMA node memory size to be a multiple of
4508      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4509      * in which LMBs are represented and hot-added
4510      */
4511     mc->numa_mem_align_shift = 28;
4512     mc->auto_enable_numa = true;
4513 
4514     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4515     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4516     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4517     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4518     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4519     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4520     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4521     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4522     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4523     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4524     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4525     spapr_caps_add_properties(smc);
4526     smc->irq = &spapr_irq_dual;
4527     smc->dr_phb_enabled = true;
4528     smc->linux_pci_probe = true;
4529     smc->smp_threads_vsmt = true;
4530     smc->nr_xirqs = SPAPR_NR_XIRQS;
4531     xfc->match_nvt = spapr_match_nvt;
4532 }
4533 
4534 static const TypeInfo spapr_machine_info = {
4535     .name          = TYPE_SPAPR_MACHINE,
4536     .parent        = TYPE_MACHINE,
4537     .abstract      = true,
4538     .instance_size = sizeof(SpaprMachineState),
4539     .instance_init = spapr_instance_init,
4540     .instance_finalize = spapr_machine_finalizefn,
4541     .class_size    = sizeof(SpaprMachineClass),
4542     .class_init    = spapr_machine_class_init,
4543     .interfaces = (InterfaceInfo[]) {
4544         { TYPE_FW_PATH_PROVIDER },
4545         { TYPE_NMI },
4546         { TYPE_HOTPLUG_HANDLER },
4547         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4548         { TYPE_XICS_FABRIC },
4549         { TYPE_INTERRUPT_STATS_PROVIDER },
4550         { TYPE_XIVE_FABRIC },
4551         { }
4552     },
4553 };
4554 
4555 static void spapr_machine_latest_class_options(MachineClass *mc)
4556 {
4557     mc->alias = "pseries";
4558     mc->is_default = true;
4559 }
4560 
4561 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4562     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4563                                                     void *data)      \
4564     {                                                                \
4565         MachineClass *mc = MACHINE_CLASS(oc);                        \
4566         spapr_machine_##suffix##_class_options(mc);                  \
4567         if (latest) {                                                \
4568             spapr_machine_latest_class_options(mc);                  \
4569         }                                                            \
4570     }                                                                \
4571     static const TypeInfo spapr_machine_##suffix##_info = {          \
4572         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4573         .parent = TYPE_SPAPR_MACHINE,                                \
4574         .class_init = spapr_machine_##suffix##_class_init,           \
4575     };                                                               \
4576     static void spapr_machine_register_##suffix(void)                \
4577     {                                                                \
4578         type_register(&spapr_machine_##suffix##_info);               \
4579     }                                                                \
4580     type_init(spapr_machine_register_##suffix)
4581 
4582 /*
4583  * pseries-5.1
4584  */
4585 static void spapr_machine_5_1_class_options(MachineClass *mc)
4586 {
4587     /* Defaults for the latest behaviour inherited from the base class */
4588 }
4589 
4590 DEFINE_SPAPR_MACHINE(5_1, "5.1", true);
4591 
4592 /*
4593  * pseries-5.0
4594  */
4595 static void spapr_machine_5_0_class_options(MachineClass *mc)
4596 {
4597     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4598     static GlobalProperty compat[] = {
4599         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4600     };
4601 
4602     spapr_machine_5_1_class_options(mc);
4603     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4604     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4605     mc->numa_mem_supported = true;
4606     smc->pre_5_1_assoc_refpoints = true;
4607 }
4608 
4609 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4610 
4611 /*
4612  * pseries-4.2
4613  */
4614 static void spapr_machine_4_2_class_options(MachineClass *mc)
4615 {
4616     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4617 
4618     spapr_machine_5_0_class_options(mc);
4619     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4620     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4621     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4622     smc->rma_limit = 16 * GiB;
4623     mc->nvdimm_supported = false;
4624 }
4625 
4626 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4627 
4628 /*
4629  * pseries-4.1
4630  */
4631 static void spapr_machine_4_1_class_options(MachineClass *mc)
4632 {
4633     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4634     static GlobalProperty compat[] = {
4635         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4636         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4637     };
4638 
4639     spapr_machine_4_2_class_options(mc);
4640     smc->linux_pci_probe = false;
4641     smc->smp_threads_vsmt = false;
4642     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4643     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4644 }
4645 
4646 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4647 
4648 /*
4649  * pseries-4.0
4650  */
4651 static void phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4652                               uint64_t *buid, hwaddr *pio,
4653                               hwaddr *mmio32, hwaddr *mmio64,
4654                               unsigned n_dma, uint32_t *liobns,
4655                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4656 {
4657     spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, liobns,
4658                         nv2gpa, nv2atsd, errp);
4659     *nv2gpa = 0;
4660     *nv2atsd = 0;
4661 }
4662 
4663 static void spapr_machine_4_0_class_options(MachineClass *mc)
4664 {
4665     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4666 
4667     spapr_machine_4_1_class_options(mc);
4668     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4669     smc->phb_placement = phb_placement_4_0;
4670     smc->irq = &spapr_irq_xics;
4671     smc->pre_4_1_migration = true;
4672 }
4673 
4674 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4675 
4676 /*
4677  * pseries-3.1
4678  */
4679 static void spapr_machine_3_1_class_options(MachineClass *mc)
4680 {
4681     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4682 
4683     spapr_machine_4_0_class_options(mc);
4684     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4685 
4686     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4687     smc->update_dt_enabled = false;
4688     smc->dr_phb_enabled = false;
4689     smc->broken_host_serial_model = true;
4690     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4691     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4692     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4693     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4694 }
4695 
4696 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4697 
4698 /*
4699  * pseries-3.0
4700  */
4701 
4702 static void spapr_machine_3_0_class_options(MachineClass *mc)
4703 {
4704     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4705 
4706     spapr_machine_3_1_class_options(mc);
4707     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4708 
4709     smc->legacy_irq_allocation = true;
4710     smc->nr_xirqs = 0x400;
4711     smc->irq = &spapr_irq_xics_legacy;
4712 }
4713 
4714 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4715 
4716 /*
4717  * pseries-2.12
4718  */
4719 static void spapr_machine_2_12_class_options(MachineClass *mc)
4720 {
4721     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4722     static GlobalProperty compat[] = {
4723         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4724         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4725     };
4726 
4727     spapr_machine_3_0_class_options(mc);
4728     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4729     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4730 
4731     /* We depend on kvm_enabled() to choose a default value for the
4732      * hpt-max-page-size capability. Of course we can't do it here
4733      * because this is too early and the HW accelerator isn't initialzed
4734      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4735      */
4736     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4737 }
4738 
4739 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4740 
4741 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4742 {
4743     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4744 
4745     spapr_machine_2_12_class_options(mc);
4746     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4747     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4748     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4749 }
4750 
4751 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4752 
4753 /*
4754  * pseries-2.11
4755  */
4756 
4757 static void spapr_machine_2_11_class_options(MachineClass *mc)
4758 {
4759     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4760 
4761     spapr_machine_2_12_class_options(mc);
4762     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4763     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4764 }
4765 
4766 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4767 
4768 /*
4769  * pseries-2.10
4770  */
4771 
4772 static void spapr_machine_2_10_class_options(MachineClass *mc)
4773 {
4774     spapr_machine_2_11_class_options(mc);
4775     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4776 }
4777 
4778 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4779 
4780 /*
4781  * pseries-2.9
4782  */
4783 
4784 static void spapr_machine_2_9_class_options(MachineClass *mc)
4785 {
4786     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4787     static GlobalProperty compat[] = {
4788         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4789     };
4790 
4791     spapr_machine_2_10_class_options(mc);
4792     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4793     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4794     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4795     smc->pre_2_10_has_unused_icps = true;
4796     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4797 }
4798 
4799 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4800 
4801 /*
4802  * pseries-2.8
4803  */
4804 
4805 static void spapr_machine_2_8_class_options(MachineClass *mc)
4806 {
4807     static GlobalProperty compat[] = {
4808         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4809     };
4810 
4811     spapr_machine_2_9_class_options(mc);
4812     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4813     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4814     mc->numa_mem_align_shift = 23;
4815 }
4816 
4817 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4818 
4819 /*
4820  * pseries-2.7
4821  */
4822 
4823 static void phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4824                               uint64_t *buid, hwaddr *pio,
4825                               hwaddr *mmio32, hwaddr *mmio64,
4826                               unsigned n_dma, uint32_t *liobns,
4827                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4828 {
4829     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4830     const uint64_t base_buid = 0x800000020000000ULL;
4831     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4832     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4833     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4834     const uint32_t max_index = 255;
4835     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4836 
4837     uint64_t ram_top = MACHINE(spapr)->ram_size;
4838     hwaddr phb0_base, phb_base;
4839     int i;
4840 
4841     /* Do we have device memory? */
4842     if (MACHINE(spapr)->maxram_size > ram_top) {
4843         /* Can't just use maxram_size, because there may be an
4844          * alignment gap between normal and device memory regions
4845          */
4846         ram_top = MACHINE(spapr)->device_memory->base +
4847             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4848     }
4849 
4850     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4851 
4852     if (index > max_index) {
4853         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4854                    max_index);
4855         return;
4856     }
4857 
4858     *buid = base_buid + index;
4859     for (i = 0; i < n_dma; ++i) {
4860         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4861     }
4862 
4863     phb_base = phb0_base + index * phb_spacing;
4864     *pio = phb_base + pio_offset;
4865     *mmio32 = phb_base + mmio_offset;
4866     /*
4867      * We don't set the 64-bit MMIO window, relying on the PHB's
4868      * fallback behaviour of automatically splitting a large "32-bit"
4869      * window into contiguous 32-bit and 64-bit windows
4870      */
4871 
4872     *nv2gpa = 0;
4873     *nv2atsd = 0;
4874 }
4875 
4876 static void spapr_machine_2_7_class_options(MachineClass *mc)
4877 {
4878     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4879     static GlobalProperty compat[] = {
4880         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4881         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4882         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4883         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4884     };
4885 
4886     spapr_machine_2_8_class_options(mc);
4887     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4888     mc->default_machine_opts = "modern-hotplug-events=off";
4889     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4890     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4891     smc->phb_placement = phb_placement_2_7;
4892 }
4893 
4894 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4895 
4896 /*
4897  * pseries-2.6
4898  */
4899 
4900 static void spapr_machine_2_6_class_options(MachineClass *mc)
4901 {
4902     static GlobalProperty compat[] = {
4903         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4904     };
4905 
4906     spapr_machine_2_7_class_options(mc);
4907     mc->has_hotpluggable_cpus = false;
4908     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4909     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4910 }
4911 
4912 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4913 
4914 /*
4915  * pseries-2.5
4916  */
4917 
4918 static void spapr_machine_2_5_class_options(MachineClass *mc)
4919 {
4920     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4921     static GlobalProperty compat[] = {
4922         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4923     };
4924 
4925     spapr_machine_2_6_class_options(mc);
4926     smc->use_ohci_by_default = true;
4927     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4928     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4929 }
4930 
4931 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4932 
4933 /*
4934  * pseries-2.4
4935  */
4936 
4937 static void spapr_machine_2_4_class_options(MachineClass *mc)
4938 {
4939     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4940 
4941     spapr_machine_2_5_class_options(mc);
4942     smc->dr_lmb_enabled = false;
4943     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4944 }
4945 
4946 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4947 
4948 /*
4949  * pseries-2.3
4950  */
4951 
4952 static void spapr_machine_2_3_class_options(MachineClass *mc)
4953 {
4954     static GlobalProperty compat[] = {
4955         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4956     };
4957     spapr_machine_2_4_class_options(mc);
4958     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4959     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4960 }
4961 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4962 
4963 /*
4964  * pseries-2.2
4965  */
4966 
4967 static void spapr_machine_2_2_class_options(MachineClass *mc)
4968 {
4969     static GlobalProperty compat[] = {
4970         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4971     };
4972 
4973     spapr_machine_2_3_class_options(mc);
4974     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4975     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4976     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4977 }
4978 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4979 
4980 /*
4981  * pseries-2.1
4982  */
4983 
4984 static void spapr_machine_2_1_class_options(MachineClass *mc)
4985 {
4986     spapr_machine_2_2_class_options(mc);
4987     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4988 }
4989 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4990 
4991 static void spapr_machine_register_types(void)
4992 {
4993     type_register_static(&spapr_machine_info);
4994 }
4995 
4996 type_init(spapr_machine_register_types)
4997