xref: /openbmc/qemu/hw/ppc/spapr.c (revision 97841438)
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 #include "hw/ppc/spapr_numa.h"
85 
86 #include "monitor/monitor.h"
87 
88 #include <libfdt.h>
89 
90 /* SLOF memory layout:
91  *
92  * SLOF raw image loaded at 0, copies its romfs right below the flat
93  * device-tree, then position SLOF itself 31M below that
94  *
95  * So we set FW_OVERHEAD to 40MB which should account for all of that
96  * and more
97  *
98  * We load our kernel at 4M, leaving space for SLOF initial image
99  */
100 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
101 #define FW_MAX_SIZE             0x400000
102 #define FW_FILE_NAME            "slof.bin"
103 #define FW_OVERHEAD             0x2800000
104 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
105 
106 #define MIN_RMA_SLOF            (128 * MiB)
107 
108 #define PHANDLE_INTC            0x00001111
109 
110 /* These two functions implement the VCPU id numbering: one to compute them
111  * all and one to identify thread 0 of a VCORE. Any change to the first one
112  * is likely to have an impact on the second one, so let's keep them close.
113  */
114 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
115 {
116     MachineState *ms = MACHINE(spapr);
117     unsigned int smp_threads = ms->smp.threads;
118 
119     assert(spapr->vsmt);
120     return
121         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
122 }
123 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
124                                       PowerPCCPU *cpu)
125 {
126     assert(spapr->vsmt);
127     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
128 }
129 
130 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
131 {
132     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
133      * and newer QEMUs don't even have them. In both cases, we don't want
134      * to send anything on the wire.
135      */
136     return false;
137 }
138 
139 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
140     .name = "icp/server",
141     .version_id = 1,
142     .minimum_version_id = 1,
143     .needed = pre_2_10_vmstate_dummy_icp_needed,
144     .fields = (VMStateField[]) {
145         VMSTATE_UNUSED(4), /* uint32_t xirr */
146         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
147         VMSTATE_UNUSED(1), /* uint8_t mfrr */
148         VMSTATE_END_OF_LIST()
149     },
150 };
151 
152 static void pre_2_10_vmstate_register_dummy_icp(int i)
153 {
154     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
155                      (void *)(uintptr_t) i);
156 }
157 
158 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
159 {
160     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
161                        (void *)(uintptr_t) i);
162 }
163 
164 int spapr_max_server_number(SpaprMachineState *spapr)
165 {
166     MachineState *ms = MACHINE(spapr);
167 
168     assert(spapr->vsmt);
169     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
170 }
171 
172 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
173                                   int smt_threads)
174 {
175     int i, ret = 0;
176     uint32_t servers_prop[smt_threads];
177     uint32_t gservers_prop[smt_threads * 2];
178     int index = spapr_get_vcpu_id(cpu);
179 
180     if (cpu->compat_pvr) {
181         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
182         if (ret < 0) {
183             return ret;
184         }
185     }
186 
187     /* Build interrupt servers and gservers properties */
188     for (i = 0; i < smt_threads; i++) {
189         servers_prop[i] = cpu_to_be32(index + i);
190         /* Hack, direct the group queues back to cpu 0 */
191         gservers_prop[i*2] = cpu_to_be32(index + i);
192         gservers_prop[i*2 + 1] = 0;
193     }
194     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
195                       servers_prop, sizeof(servers_prop));
196     if (ret < 0) {
197         return ret;
198     }
199     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
200                       gservers_prop, sizeof(gservers_prop));
201 
202     return ret;
203 }
204 
205 static void spapr_dt_pa_features(SpaprMachineState *spapr,
206                                  PowerPCCPU *cpu,
207                                  void *fdt, int offset)
208 {
209     uint8_t pa_features_206[] = { 6, 0,
210         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
211     uint8_t pa_features_207[] = { 24, 0,
212         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
213         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
214         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
215         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
216     uint8_t pa_features_300[] = { 66, 0,
217         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
218         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
219         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
220         /* 6: DS207 */
221         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
222         /* 16: Vector */
223         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
224         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
225         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
226         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
227         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
228         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
229         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
230         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
231         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
232         /* 42: PM, 44: PC RA, 46: SC vec'd */
233         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
234         /* 48: SIMD, 50: QP BFP, 52: String */
235         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
236         /* 54: DecFP, 56: DecI, 58: SHA */
237         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
238         /* 60: NM atomic, 62: RNG */
239         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
240     };
241     uint8_t *pa_features = NULL;
242     size_t pa_size;
243 
244     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
245         pa_features = pa_features_206;
246         pa_size = sizeof(pa_features_206);
247     }
248     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
249         pa_features = pa_features_207;
250         pa_size = sizeof(pa_features_207);
251     }
252     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
253         pa_features = pa_features_300;
254         pa_size = sizeof(pa_features_300);
255     }
256     if (!pa_features) {
257         return;
258     }
259 
260     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
261         /*
262          * Note: we keep CI large pages off by default because a 64K capable
263          * guest provisioned with large pages might otherwise try to map a qemu
264          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
265          * even if that qemu runs on a 4k host.
266          * We dd this bit back here if we are confident this is not an issue
267          */
268         pa_features[3] |= 0x20;
269     }
270     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
271         pa_features[24] |= 0x80;    /* Transactional memory support */
272     }
273     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
274         /* Workaround for broken kernels that attempt (guest) radix
275          * mode when they can't handle it, if they see the radix bit set
276          * in pa-features. So hide it from them. */
277         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
278     }
279 
280     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
281 }
282 
283 static hwaddr spapr_node0_size(MachineState *machine)
284 {
285     if (machine->numa_state->num_nodes) {
286         int i;
287         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
288             if (machine->numa_state->nodes[i].node_mem) {
289                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
290                            machine->ram_size);
291             }
292         }
293     }
294     return machine->ram_size;
295 }
296 
297 bool spapr_machine_using_legacy_numa(SpaprMachineState *spapr)
298 {
299     MachineState *machine = MACHINE(spapr);
300     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
301 
302     return smc->pre_5_2_numa_associativity ||
303            machine->numa_state->num_nodes <= 1;
304 }
305 
306 static void add_str(GString *s, const gchar *s1)
307 {
308     g_string_append_len(s, s1, strlen(s1) + 1);
309 }
310 
311 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid,
312                                 hwaddr start, hwaddr size)
313 {
314     char mem_name[32];
315     uint64_t mem_reg_property[2];
316     int off;
317 
318     mem_reg_property[0] = cpu_to_be64(start);
319     mem_reg_property[1] = cpu_to_be64(size);
320 
321     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
322     off = fdt_add_subnode(fdt, 0, mem_name);
323     _FDT(off);
324     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
325     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
326                       sizeof(mem_reg_property))));
327     spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid);
328     return off;
329 }
330 
331 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
332 {
333     MemoryDeviceInfoList *info;
334 
335     for (info = list; info; info = info->next) {
336         MemoryDeviceInfo *value = info->value;
337 
338         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
339             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
340 
341             if (addr >= pcdimm_info->addr &&
342                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
343                 return pcdimm_info->node;
344             }
345         }
346     }
347 
348     return -1;
349 }
350 
351 struct sPAPRDrconfCellV2 {
352      uint32_t seq_lmbs;
353      uint64_t base_addr;
354      uint32_t drc_index;
355      uint32_t aa_index;
356      uint32_t flags;
357 } QEMU_PACKED;
358 
359 typedef struct DrconfCellQueue {
360     struct sPAPRDrconfCellV2 cell;
361     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
362 } DrconfCellQueue;
363 
364 static DrconfCellQueue *
365 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
366                       uint32_t drc_index, uint32_t aa_index,
367                       uint32_t flags)
368 {
369     DrconfCellQueue *elem;
370 
371     elem = g_malloc0(sizeof(*elem));
372     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
373     elem->cell.base_addr = cpu_to_be64(base_addr);
374     elem->cell.drc_index = cpu_to_be32(drc_index);
375     elem->cell.aa_index = cpu_to_be32(aa_index);
376     elem->cell.flags = cpu_to_be32(flags);
377 
378     return elem;
379 }
380 
381 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
382                                       int offset, MemoryDeviceInfoList *dimms)
383 {
384     MachineState *machine = MACHINE(spapr);
385     uint8_t *int_buf, *cur_index;
386     int ret;
387     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
388     uint64_t addr, cur_addr, size;
389     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
390     uint64_t mem_end = machine->device_memory->base +
391                        memory_region_size(&machine->device_memory->mr);
392     uint32_t node, buf_len, nr_entries = 0;
393     SpaprDrc *drc;
394     DrconfCellQueue *elem, *next;
395     MemoryDeviceInfoList *info;
396     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
397         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
398 
399     /* Entry to cover RAM and the gap area */
400     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
401                                  SPAPR_LMB_FLAGS_RESERVED |
402                                  SPAPR_LMB_FLAGS_DRC_INVALID);
403     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
404     nr_entries++;
405 
406     cur_addr = machine->device_memory->base;
407     for (info = dimms; info; info = info->next) {
408         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
409 
410         addr = di->addr;
411         size = di->size;
412         node = di->node;
413 
414         /*
415          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
416          * area is marked hotpluggable in the next iteration for the bigger
417          * chunk including the NVDIMM occupied area.
418          */
419         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
420             continue;
421 
422         /* Entry for hot-pluggable area */
423         if (cur_addr < addr) {
424             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
425             g_assert(drc);
426             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
427                                          cur_addr, spapr_drc_index(drc), -1, 0);
428             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
429             nr_entries++;
430         }
431 
432         /* Entry for DIMM */
433         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
434         g_assert(drc);
435         elem = spapr_get_drconf_cell(size / lmb_size, addr,
436                                      spapr_drc_index(drc), node,
437                                      (SPAPR_LMB_FLAGS_ASSIGNED |
438                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
439         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
440         nr_entries++;
441         cur_addr = addr + size;
442     }
443 
444     /* Entry for remaining hotpluggable area */
445     if (cur_addr < mem_end) {
446         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
447         g_assert(drc);
448         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
449                                      cur_addr, spapr_drc_index(drc), -1, 0);
450         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
451         nr_entries++;
452     }
453 
454     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
455     int_buf = cur_index = g_malloc0(buf_len);
456     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
457     cur_index += sizeof(nr_entries);
458 
459     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
460         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
461         cur_index += sizeof(elem->cell);
462         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
463         g_free(elem);
464     }
465 
466     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
467     g_free(int_buf);
468     if (ret < 0) {
469         return -1;
470     }
471     return 0;
472 }
473 
474 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
475                                    int offset, MemoryDeviceInfoList *dimms)
476 {
477     MachineState *machine = MACHINE(spapr);
478     int i, ret;
479     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
480     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
481     uint32_t nr_lmbs = (machine->device_memory->base +
482                        memory_region_size(&machine->device_memory->mr)) /
483                        lmb_size;
484     uint32_t *int_buf, *cur_index, buf_len;
485 
486     /*
487      * Allocate enough buffer size to fit in ibm,dynamic-memory
488      */
489     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
490     cur_index = int_buf = g_malloc0(buf_len);
491     int_buf[0] = cpu_to_be32(nr_lmbs);
492     cur_index++;
493     for (i = 0; i < nr_lmbs; i++) {
494         uint64_t addr = i * lmb_size;
495         uint32_t *dynamic_memory = cur_index;
496 
497         if (i >= device_lmb_start) {
498             SpaprDrc *drc;
499 
500             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
501             g_assert(drc);
502 
503             dynamic_memory[0] = cpu_to_be32(addr >> 32);
504             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
505             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
506             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
507             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
508             if (memory_region_present(get_system_memory(), addr)) {
509                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
510             } else {
511                 dynamic_memory[5] = cpu_to_be32(0);
512             }
513         } else {
514             /*
515              * LMB information for RMA, boot time RAM and gap b/n RAM and
516              * device memory region -- all these are marked as reserved
517              * and as having no valid DRC.
518              */
519             dynamic_memory[0] = cpu_to_be32(addr >> 32);
520             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
521             dynamic_memory[2] = cpu_to_be32(0);
522             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
523             dynamic_memory[4] = cpu_to_be32(-1);
524             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
525                                             SPAPR_LMB_FLAGS_DRC_INVALID);
526         }
527 
528         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
529     }
530     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
531     g_free(int_buf);
532     if (ret < 0) {
533         return -1;
534     }
535     return 0;
536 }
537 
538 /*
539  * Adds ibm,dynamic-reconfiguration-memory node.
540  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
541  * of this device tree node.
542  */
543 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
544                                                    void *fdt)
545 {
546     MachineState *machine = MACHINE(spapr);
547     int ret, offset;
548     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
549     uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
550                                 cpu_to_be32(lmb_size & 0xffffffff)};
551     MemoryDeviceInfoList *dimms = NULL;
552 
553     /*
554      * Don't create the node if there is no device memory
555      */
556     if (machine->ram_size == machine->maxram_size) {
557         return 0;
558     }
559 
560     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
561 
562     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
563                     sizeof(prop_lmb_size));
564     if (ret < 0) {
565         return ret;
566     }
567 
568     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
569     if (ret < 0) {
570         return ret;
571     }
572 
573     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
574     if (ret < 0) {
575         return ret;
576     }
577 
578     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
579     dimms = qmp_memory_device_list();
580     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
581         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
582     } else {
583         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
584     }
585     qapi_free_MemoryDeviceInfoList(dimms);
586 
587     if (ret < 0) {
588         return ret;
589     }
590 
591     ret = spapr_numa_write_assoc_lookup_arrays(spapr, fdt, offset);
592 
593     return ret;
594 }
595 
596 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
597 {
598     MachineState *machine = MACHINE(spapr);
599     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
600     hwaddr mem_start, node_size;
601     int i, nb_nodes = machine->numa_state->num_nodes;
602     NodeInfo *nodes = machine->numa_state->nodes;
603 
604     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
605         if (!nodes[i].node_mem) {
606             continue;
607         }
608         if (mem_start >= machine->ram_size) {
609             node_size = 0;
610         } else {
611             node_size = nodes[i].node_mem;
612             if (node_size > machine->ram_size - mem_start) {
613                 node_size = machine->ram_size - mem_start;
614             }
615         }
616         if (!mem_start) {
617             /* spapr_machine_init() checks for rma_size <= node0_size
618              * already */
619             spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size);
620             mem_start += spapr->rma_size;
621             node_size -= spapr->rma_size;
622         }
623         for ( ; node_size; ) {
624             hwaddr sizetmp = pow2floor(node_size);
625 
626             /* mem_start != 0 here */
627             if (ctzl(mem_start) < ctzl(sizetmp)) {
628                 sizetmp = 1ULL << ctzl(mem_start);
629             }
630 
631             spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp);
632             node_size -= sizetmp;
633             mem_start += sizetmp;
634         }
635     }
636 
637     /* Generate ibm,dynamic-reconfiguration-memory node if required */
638     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
639         int ret;
640 
641         g_assert(smc->dr_lmb_enabled);
642         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
643         if (ret) {
644             return ret;
645         }
646     }
647 
648     return 0;
649 }
650 
651 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
652                          SpaprMachineState *spapr)
653 {
654     MachineState *ms = MACHINE(spapr);
655     PowerPCCPU *cpu = POWERPC_CPU(cs);
656     CPUPPCState *env = &cpu->env;
657     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
658     int index = spapr_get_vcpu_id(cpu);
659     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
660                        0xffffffff, 0xffffffff};
661     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
662         : SPAPR_TIMEBASE_FREQ;
663     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
664     uint32_t page_sizes_prop[64];
665     size_t page_sizes_prop_size;
666     unsigned int smp_threads = ms->smp.threads;
667     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
668     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
669     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
670     SpaprDrc *drc;
671     int drc_index;
672     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
673     int i;
674 
675     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
676     if (drc) {
677         drc_index = spapr_drc_index(drc);
678         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
679     }
680 
681     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
682     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
683 
684     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
685     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
686                            env->dcache_line_size)));
687     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
688                            env->dcache_line_size)));
689     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
690                            env->icache_line_size)));
691     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
692                            env->icache_line_size)));
693 
694     if (pcc->l1_dcache_size) {
695         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
696                                pcc->l1_dcache_size)));
697     } else {
698         warn_report("Unknown L1 dcache size for cpu");
699     }
700     if (pcc->l1_icache_size) {
701         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
702                                pcc->l1_icache_size)));
703     } else {
704         warn_report("Unknown L1 icache size for cpu");
705     }
706 
707     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
708     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
709     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
710     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
711     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
712     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
713 
714     if (env->spr_cb[SPR_PURR].oea_read) {
715         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
716     }
717     if (env->spr_cb[SPR_SPURR].oea_read) {
718         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
719     }
720 
721     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
722         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
723                           segs, sizeof(segs))));
724     }
725 
726     /* Advertise VSX (vector extensions) if available
727      *   1               == VMX / Altivec available
728      *   2               == VSX available
729      *
730      * Only CPUs for which we create core types in spapr_cpu_core.c
731      * are possible, and all of those have VMX */
732     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
733         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
734     } else {
735         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
736     }
737 
738     /* Advertise DFP (Decimal Floating Point) if available
739      *   0 / no property == no DFP
740      *   1               == DFP available */
741     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
742         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
743     }
744 
745     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
746                                                       sizeof(page_sizes_prop));
747     if (page_sizes_prop_size) {
748         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
749                           page_sizes_prop, page_sizes_prop_size)));
750     }
751 
752     spapr_dt_pa_features(spapr, cpu, fdt, offset);
753 
754     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
755                            cs->cpu_index / vcpus_per_socket)));
756 
757     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
758                       pft_size_prop, sizeof(pft_size_prop))));
759 
760     if (ms->numa_state->num_nodes > 1) {
761         _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
762     }
763 
764     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
765 
766     if (pcc->radix_page_info) {
767         for (i = 0; i < pcc->radix_page_info->count; i++) {
768             radix_AP_encodings[i] =
769                 cpu_to_be32(pcc->radix_page_info->entries[i]);
770         }
771         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
772                           radix_AP_encodings,
773                           pcc->radix_page_info->count *
774                           sizeof(radix_AP_encodings[0]))));
775     }
776 
777     /*
778      * We set this property to let the guest know that it can use the large
779      * decrementer and its width in bits.
780      */
781     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
782         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
783                               pcc->lrg_decr_bits)));
784 }
785 
786 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
787 {
788     CPUState **rev;
789     CPUState *cs;
790     int n_cpus;
791     int cpus_offset;
792     char *nodename;
793     int i;
794 
795     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
796     _FDT(cpus_offset);
797     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
798     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
799 
800     /*
801      * We walk the CPUs in reverse order to ensure that CPU DT nodes
802      * created by fdt_add_subnode() end up in the right order in FDT
803      * for the guest kernel the enumerate the CPUs correctly.
804      *
805      * The CPU list cannot be traversed in reverse order, so we need
806      * to do extra work.
807      */
808     n_cpus = 0;
809     rev = NULL;
810     CPU_FOREACH(cs) {
811         rev = g_renew(CPUState *, rev, n_cpus + 1);
812         rev[n_cpus++] = cs;
813     }
814 
815     for (i = n_cpus - 1; i >= 0; i--) {
816         CPUState *cs = rev[i];
817         PowerPCCPU *cpu = POWERPC_CPU(cs);
818         int index = spapr_get_vcpu_id(cpu);
819         DeviceClass *dc = DEVICE_GET_CLASS(cs);
820         int offset;
821 
822         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
823             continue;
824         }
825 
826         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
827         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
828         g_free(nodename);
829         _FDT(offset);
830         spapr_dt_cpu(cs, fdt, offset, spapr);
831     }
832 
833     g_free(rev);
834 }
835 
836 static int spapr_dt_rng(void *fdt)
837 {
838     int node;
839     int ret;
840 
841     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
842     if (node <= 0) {
843         return -1;
844     }
845     ret = fdt_setprop_string(fdt, node, "device_type",
846                              "ibm,platform-facilities");
847     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
848     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
849 
850     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
851     if (node <= 0) {
852         return -1;
853     }
854     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
855 
856     return ret ? -1 : 0;
857 }
858 
859 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
860 {
861     MachineState *ms = MACHINE(spapr);
862     int rtas;
863     GString *hypertas = g_string_sized_new(256);
864     GString *qemu_hypertas = g_string_sized_new(256);
865     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
866         memory_region_size(&MACHINE(spapr)->device_memory->mr);
867     uint32_t lrdr_capacity[] = {
868         cpu_to_be32(max_device_addr >> 32),
869         cpu_to_be32(max_device_addr & 0xffffffff),
870         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
871         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
872         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
873     };
874 
875     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
876 
877     /* hypertas */
878     add_str(hypertas, "hcall-pft");
879     add_str(hypertas, "hcall-term");
880     add_str(hypertas, "hcall-dabr");
881     add_str(hypertas, "hcall-interrupt");
882     add_str(hypertas, "hcall-tce");
883     add_str(hypertas, "hcall-vio");
884     add_str(hypertas, "hcall-splpar");
885     add_str(hypertas, "hcall-join");
886     add_str(hypertas, "hcall-bulk");
887     add_str(hypertas, "hcall-set-mode");
888     add_str(hypertas, "hcall-sprg0");
889     add_str(hypertas, "hcall-copy");
890     add_str(hypertas, "hcall-debug");
891     add_str(hypertas, "hcall-vphn");
892     add_str(qemu_hypertas, "hcall-memop1");
893 
894     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
895         add_str(hypertas, "hcall-multi-tce");
896     }
897 
898     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
899         add_str(hypertas, "hcall-hpt-resize");
900     }
901 
902     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
903                      hypertas->str, hypertas->len));
904     g_string_free(hypertas, TRUE);
905     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
906                      qemu_hypertas->str, qemu_hypertas->len));
907     g_string_free(qemu_hypertas, TRUE);
908 
909     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
910 
911     /*
912      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
913      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
914      *
915      * The system reset requirements are driven by existing Linux and PowerVM
916      * implementation which (contrary to PAPR) saves r3 in the error log
917      * structure like machine check, so Linux expects to find the saved r3
918      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
919      * does not look at the error value).
920      *
921      * System reset interrupts are not subject to interlock like machine
922      * check, so this memory area could be corrupted if the sreset is
923      * interrupted by a machine check (or vice versa) if it was shared. To
924      * prevent this, system reset uses per-CPU areas for the sreset save
925      * area. A system reset that interrupts a system reset handler could
926      * still overwrite this area, but Linux doesn't try to recover in that
927      * case anyway.
928      *
929      * The extra 8 bytes is required because Linux's FWNMI error log check
930      * is off-by-one.
931      */
932     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_ERROR_LOG_MAX +
933 			  ms->smp.max_cpus * sizeof(uint64_t)*2 + sizeof(uint64_t)));
934     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
935                           RTAS_ERROR_LOG_MAX));
936     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
937                           RTAS_EVENT_SCAN_RATE));
938 
939     g_assert(msi_nonbroken);
940     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
941 
942     /*
943      * According to PAPR, rtas ibm,os-term does not guarantee a return
944      * back to the guest cpu.
945      *
946      * While an additional ibm,extended-os-term property indicates
947      * that rtas call return will always occur. Set this property.
948      */
949     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
950 
951     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
952                      lrdr_capacity, sizeof(lrdr_capacity)));
953 
954     spapr_dt_rtas_tokens(fdt, rtas);
955 }
956 
957 /*
958  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
959  * and the XIVE features that the guest may request and thus the valid
960  * values for bytes 23..26 of option vector 5:
961  */
962 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
963                                           int chosen)
964 {
965     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
966 
967     char val[2 * 4] = {
968         23, 0x00, /* XICS / XIVE mode */
969         24, 0x00, /* Hash/Radix, filled in below. */
970         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
971         26, 0x40, /* Radix options: GTSE == yes. */
972     };
973 
974     if (spapr->irq->xics && spapr->irq->xive) {
975         val[1] = SPAPR_OV5_XIVE_BOTH;
976     } else if (spapr->irq->xive) {
977         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
978     } else {
979         assert(spapr->irq->xics);
980         val[1] = SPAPR_OV5_XIVE_LEGACY;
981     }
982 
983     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
984                           first_ppc_cpu->compat_pvr)) {
985         /*
986          * If we're in a pre POWER9 compat mode then the guest should
987          * do hash and use the legacy interrupt mode
988          */
989         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
990         val[3] = 0x00; /* Hash */
991     } else if (kvm_enabled()) {
992         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
993             val[3] = 0x80; /* OV5_MMU_BOTH */
994         } else if (kvmppc_has_cap_mmu_radix()) {
995             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
996         } else {
997             val[3] = 0x00; /* Hash */
998         }
999     } else {
1000         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1001         val[3] = 0xC0;
1002     }
1003     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1004                      val, sizeof(val)));
1005 }
1006 
1007 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1008 {
1009     MachineState *machine = MACHINE(spapr);
1010     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1011     int chosen;
1012 
1013     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1014 
1015     if (reset) {
1016         const char *boot_device = machine->boot_order;
1017         char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1018         size_t cb = 0;
1019         char *bootlist = get_boot_devices_list(&cb);
1020 
1021         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1022             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1023                                     machine->kernel_cmdline));
1024         }
1025 
1026         if (spapr->initrd_size) {
1027             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1028                                   spapr->initrd_base));
1029             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1030                                   spapr->initrd_base + spapr->initrd_size));
1031         }
1032 
1033         if (spapr->kernel_size) {
1034             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1035                                   cpu_to_be64(spapr->kernel_size) };
1036 
1037             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1038                          &kprop, sizeof(kprop)));
1039             if (spapr->kernel_le) {
1040                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1041             }
1042         }
1043         if (boot_menu) {
1044             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1045         }
1046         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1047         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1048         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1049 
1050         if (cb && bootlist) {
1051             int i;
1052 
1053             for (i = 0; i < cb; i++) {
1054                 if (bootlist[i] == '\n') {
1055                     bootlist[i] = ' ';
1056                 }
1057             }
1058             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1059         }
1060 
1061         if (boot_device && strlen(boot_device)) {
1062             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1063         }
1064 
1065         if (!spapr->has_graphics && stdout_path) {
1066             /*
1067              * "linux,stdout-path" and "stdout" properties are
1068              * deprecated by linux kernel. New platforms should only
1069              * use the "stdout-path" property. Set the new property
1070              * and continue using older property to remain compatible
1071              * with the existing firmware.
1072              */
1073             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1074             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1075         }
1076 
1077         /*
1078          * We can deal with BAR reallocation just fine, advertise it
1079          * to the guest
1080          */
1081         if (smc->linux_pci_probe) {
1082             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1083         }
1084 
1085         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1086 
1087         g_free(stdout_path);
1088         g_free(bootlist);
1089     }
1090 
1091     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1092 }
1093 
1094 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1095 {
1096     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1097      * KVM to work under pHyp with some guest co-operation */
1098     int hypervisor;
1099     uint8_t hypercall[16];
1100 
1101     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1102     /* indicate KVM hypercall interface */
1103     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1104     if (kvmppc_has_cap_fixup_hcalls()) {
1105         /*
1106          * Older KVM versions with older guest kernels were broken
1107          * with the magic page, don't allow the guest to map it.
1108          */
1109         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1110                                   sizeof(hypercall))) {
1111             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1112                              hypercall, sizeof(hypercall)));
1113         }
1114     }
1115 }
1116 
1117 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1118 {
1119     MachineState *machine = MACHINE(spapr);
1120     MachineClass *mc = MACHINE_GET_CLASS(machine);
1121     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1122     int ret;
1123     void *fdt;
1124     SpaprPhbState *phb;
1125     char *buf;
1126 
1127     fdt = g_malloc0(space);
1128     _FDT((fdt_create_empty_tree(fdt, space)));
1129 
1130     /* Root node */
1131     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1132     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1133     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1134 
1135     /* Guest UUID & Name*/
1136     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1137     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1138     if (qemu_uuid_set) {
1139         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1140     }
1141     g_free(buf);
1142 
1143     if (qemu_get_vm_name()) {
1144         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1145                                 qemu_get_vm_name()));
1146     }
1147 
1148     /* Host Model & Serial Number */
1149     if (spapr->host_model) {
1150         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1151     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1152         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1153         g_free(buf);
1154     }
1155 
1156     if (spapr->host_serial) {
1157         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1158     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1159         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1160         g_free(buf);
1161     }
1162 
1163     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1164     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1165 
1166     /* /interrupt controller */
1167     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1168 
1169     ret = spapr_dt_memory(spapr, fdt);
1170     if (ret < 0) {
1171         error_report("couldn't setup memory nodes in fdt");
1172         exit(1);
1173     }
1174 
1175     /* /vdevice */
1176     spapr_dt_vdevice(spapr->vio_bus, fdt);
1177 
1178     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1179         ret = spapr_dt_rng(fdt);
1180         if (ret < 0) {
1181             error_report("could not set up rng device in the fdt");
1182             exit(1);
1183         }
1184     }
1185 
1186     QLIST_FOREACH(phb, &spapr->phbs, list) {
1187         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1188         if (ret < 0) {
1189             error_report("couldn't setup PCI devices in fdt");
1190             exit(1);
1191         }
1192     }
1193 
1194     spapr_dt_cpus(fdt, spapr);
1195 
1196     if (smc->dr_lmb_enabled) {
1197         _FDT(spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1198     }
1199 
1200     if (mc->has_hotpluggable_cpus) {
1201         int offset = fdt_path_offset(fdt, "/cpus");
1202         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1203         if (ret < 0) {
1204             error_report("Couldn't set up CPU DR device tree properties");
1205             exit(1);
1206         }
1207     }
1208 
1209     /* /event-sources */
1210     spapr_dt_events(spapr, fdt);
1211 
1212     /* /rtas */
1213     spapr_dt_rtas(spapr, fdt);
1214 
1215     /* /chosen */
1216     spapr_dt_chosen(spapr, fdt, reset);
1217 
1218     /* /hypervisor */
1219     if (kvm_enabled()) {
1220         spapr_dt_hypervisor(spapr, fdt);
1221     }
1222 
1223     /* Build memory reserve map */
1224     if (reset) {
1225         if (spapr->kernel_size) {
1226             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1227                                   spapr->kernel_size)));
1228         }
1229         if (spapr->initrd_size) {
1230             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1231                                   spapr->initrd_size)));
1232         }
1233     }
1234 
1235     if (smc->dr_phb_enabled) {
1236         ret = spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1237         if (ret < 0) {
1238             error_report("Couldn't set up PHB DR device tree properties");
1239             exit(1);
1240         }
1241     }
1242 
1243     /* NVDIMM devices */
1244     if (mc->nvdimm_supported) {
1245         spapr_dt_persistent_memory(spapr, fdt);
1246     }
1247 
1248     return fdt;
1249 }
1250 
1251 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1252 {
1253     SpaprMachineState *spapr = opaque;
1254 
1255     return (addr & 0x0fffffff) + spapr->kernel_addr;
1256 }
1257 
1258 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1259                                     PowerPCCPU *cpu)
1260 {
1261     CPUPPCState *env = &cpu->env;
1262 
1263     /* The TCG path should also be holding the BQL at this point */
1264     g_assert(qemu_mutex_iothread_locked());
1265 
1266     if (msr_pr) {
1267         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1268         env->gpr[3] = H_PRIVILEGE;
1269     } else {
1270         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1271     }
1272 }
1273 
1274 struct LPCRSyncState {
1275     target_ulong value;
1276     target_ulong mask;
1277 };
1278 
1279 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1280 {
1281     struct LPCRSyncState *s = arg.host_ptr;
1282     PowerPCCPU *cpu = POWERPC_CPU(cs);
1283     CPUPPCState *env = &cpu->env;
1284     target_ulong lpcr;
1285 
1286     cpu_synchronize_state(cs);
1287     lpcr = env->spr[SPR_LPCR];
1288     lpcr &= ~s->mask;
1289     lpcr |= s->value;
1290     ppc_store_lpcr(cpu, lpcr);
1291 }
1292 
1293 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1294 {
1295     CPUState *cs;
1296     struct LPCRSyncState s = {
1297         .value = value,
1298         .mask = mask
1299     };
1300     CPU_FOREACH(cs) {
1301         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1302     }
1303 }
1304 
1305 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1306 {
1307     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1308 
1309     /* Copy PATE1:GR into PATE0:HR */
1310     entry->dw0 = spapr->patb_entry & PATE0_HR;
1311     entry->dw1 = spapr->patb_entry;
1312 }
1313 
1314 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1315 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1316 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1317 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1318 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1319 
1320 /*
1321  * Get the fd to access the kernel htab, re-opening it if necessary
1322  */
1323 static int get_htab_fd(SpaprMachineState *spapr)
1324 {
1325     Error *local_err = NULL;
1326 
1327     if (spapr->htab_fd >= 0) {
1328         return spapr->htab_fd;
1329     }
1330 
1331     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1332     if (spapr->htab_fd < 0) {
1333         error_report_err(local_err);
1334     }
1335 
1336     return spapr->htab_fd;
1337 }
1338 
1339 void close_htab_fd(SpaprMachineState *spapr)
1340 {
1341     if (spapr->htab_fd >= 0) {
1342         close(spapr->htab_fd);
1343     }
1344     spapr->htab_fd = -1;
1345 }
1346 
1347 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1348 {
1349     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1350 
1351     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1352 }
1353 
1354 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1355 {
1356     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1357 
1358     assert(kvm_enabled());
1359 
1360     if (!spapr->htab) {
1361         return 0;
1362     }
1363 
1364     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1365 }
1366 
1367 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1368                                                 hwaddr ptex, int n)
1369 {
1370     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1371     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1372 
1373     if (!spapr->htab) {
1374         /*
1375          * HTAB is controlled by KVM. Fetch into temporary buffer
1376          */
1377         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1378         kvmppc_read_hptes(hptes, ptex, n);
1379         return hptes;
1380     }
1381 
1382     /*
1383      * HTAB is controlled by QEMU. Just point to the internally
1384      * accessible PTEG.
1385      */
1386     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1387 }
1388 
1389 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1390                               const ppc_hash_pte64_t *hptes,
1391                               hwaddr ptex, int n)
1392 {
1393     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1394 
1395     if (!spapr->htab) {
1396         g_free((void *)hptes);
1397     }
1398 
1399     /* Nothing to do for qemu managed HPT */
1400 }
1401 
1402 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1403                       uint64_t pte0, uint64_t pte1)
1404 {
1405     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1406     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1407 
1408     if (!spapr->htab) {
1409         kvmppc_write_hpte(ptex, pte0, pte1);
1410     } else {
1411         if (pte0 & HPTE64_V_VALID) {
1412             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1413             /*
1414              * When setting valid, we write PTE1 first. This ensures
1415              * proper synchronization with the reading code in
1416              * ppc_hash64_pteg_search()
1417              */
1418             smp_wmb();
1419             stq_p(spapr->htab + offset, pte0);
1420         } else {
1421             stq_p(spapr->htab + offset, pte0);
1422             /*
1423              * When clearing it we set PTE0 first. This ensures proper
1424              * synchronization with the reading code in
1425              * ppc_hash64_pteg_search()
1426              */
1427             smp_wmb();
1428             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1429         }
1430     }
1431 }
1432 
1433 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1434                              uint64_t pte1)
1435 {
1436     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1437     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1438 
1439     if (!spapr->htab) {
1440         /* There should always be a hash table when this is called */
1441         error_report("spapr_hpte_set_c called with no hash table !");
1442         return;
1443     }
1444 
1445     /* The HW performs a non-atomic byte update */
1446     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1447 }
1448 
1449 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1450                              uint64_t pte1)
1451 {
1452     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1453     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1454 
1455     if (!spapr->htab) {
1456         /* There should always be a hash table when this is called */
1457         error_report("spapr_hpte_set_r called with no hash table !");
1458         return;
1459     }
1460 
1461     /* The HW performs a non-atomic byte update */
1462     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1463 }
1464 
1465 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1466 {
1467     int shift;
1468 
1469     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1470      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1471      * that's much more than is needed for Linux guests */
1472     shift = ctz64(pow2ceil(ramsize)) - 7;
1473     shift = MAX(shift, 18); /* Minimum architected size */
1474     shift = MIN(shift, 46); /* Maximum architected size */
1475     return shift;
1476 }
1477 
1478 void spapr_free_hpt(SpaprMachineState *spapr)
1479 {
1480     g_free(spapr->htab);
1481     spapr->htab = NULL;
1482     spapr->htab_shift = 0;
1483     close_htab_fd(spapr);
1484 }
1485 
1486 void spapr_reallocate_hpt(SpaprMachineState *spapr, int shift,
1487                           Error **errp)
1488 {
1489     long rc;
1490 
1491     /* Clean up any HPT info from a previous boot */
1492     spapr_free_hpt(spapr);
1493 
1494     rc = kvmppc_reset_htab(shift);
1495 
1496     if (rc == -EOPNOTSUPP) {
1497         error_setg(errp, "HPT not supported in nested guests");
1498         return;
1499     }
1500 
1501     if (rc < 0) {
1502         /* kernel-side HPT needed, but couldn't allocate one */
1503         error_setg_errno(errp, errno,
1504                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1505                          shift);
1506         /* This is almost certainly fatal, but if the caller really
1507          * wants to carry on with shift == 0, it's welcome to try */
1508     } else if (rc > 0) {
1509         /* kernel-side HPT allocated */
1510         if (rc != shift) {
1511             error_setg(errp,
1512                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1513                        shift, rc);
1514         }
1515 
1516         spapr->htab_shift = shift;
1517         spapr->htab = NULL;
1518     } else {
1519         /* kernel-side HPT not needed, allocate in userspace instead */
1520         size_t size = 1ULL << shift;
1521         int i;
1522 
1523         spapr->htab = qemu_memalign(size, size);
1524         if (!spapr->htab) {
1525             error_setg_errno(errp, errno,
1526                              "Could not allocate HPT of order %d", shift);
1527             return;
1528         }
1529 
1530         memset(spapr->htab, 0, size);
1531         spapr->htab_shift = shift;
1532 
1533         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1534             DIRTY_HPTE(HPTE(spapr->htab, i));
1535         }
1536     }
1537     /* We're setting up a hash table, so that means we're not radix */
1538     spapr->patb_entry = 0;
1539     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1540 }
1541 
1542 void spapr_setup_hpt(SpaprMachineState *spapr)
1543 {
1544     int hpt_shift;
1545 
1546     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1547         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1548     } else {
1549         uint64_t current_ram_size;
1550 
1551         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1552         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1553     }
1554     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1555 
1556     if (kvm_enabled()) {
1557         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1558 
1559         /* Check our RMA fits in the possible VRMA */
1560         if (vrma_limit < spapr->rma_size) {
1561             error_report("Unable to create %" HWADDR_PRIu
1562                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1563                          spapr->rma_size / MiB, vrma_limit / MiB);
1564             exit(EXIT_FAILURE);
1565         }
1566     }
1567 }
1568 
1569 static int spapr_reset_drcs(Object *child, void *opaque)
1570 {
1571     SpaprDrc *drc =
1572         (SpaprDrc *) object_dynamic_cast(child,
1573                                                  TYPE_SPAPR_DR_CONNECTOR);
1574 
1575     if (drc) {
1576         spapr_drc_reset(drc);
1577     }
1578 
1579     return 0;
1580 }
1581 
1582 static void spapr_machine_reset(MachineState *machine)
1583 {
1584     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1585     PowerPCCPU *first_ppc_cpu;
1586     hwaddr fdt_addr;
1587     void *fdt;
1588     int rc;
1589 
1590     kvmppc_svm_off(&error_fatal);
1591     spapr_caps_apply(spapr);
1592 
1593     first_ppc_cpu = POWERPC_CPU(first_cpu);
1594     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1595         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1596                               spapr->max_compat_pvr)) {
1597         /*
1598          * If using KVM with radix mode available, VCPUs can be started
1599          * without a HPT because KVM will start them in radix mode.
1600          * Set the GR bit in PATE so that we know there is no HPT.
1601          */
1602         spapr->patb_entry = PATE1_GR;
1603         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1604     } else {
1605         spapr_setup_hpt(spapr);
1606     }
1607 
1608     qemu_devices_reset();
1609 
1610     spapr_ovec_cleanup(spapr->ov5_cas);
1611     spapr->ov5_cas = spapr_ovec_new();
1612 
1613     ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1614 
1615     /*
1616      * This is fixing some of the default configuration of the XIVE
1617      * devices. To be called after the reset of the machine devices.
1618      */
1619     spapr_irq_reset(spapr, &error_fatal);
1620 
1621     /*
1622      * There is no CAS under qtest. Simulate one to please the code that
1623      * depends on spapr->ov5_cas. This is especially needed to test device
1624      * unplug, so we do that before resetting the DRCs.
1625      */
1626     if (qtest_enabled()) {
1627         spapr_ovec_cleanup(spapr->ov5_cas);
1628         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1629     }
1630 
1631     /* DRC reset may cause a device to be unplugged. This will cause troubles
1632      * if this device is used by another device (eg, a running vhost backend
1633      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1634      * situations, we reset DRCs after all devices have been reset.
1635      */
1636     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1637 
1638     spapr_clear_pending_events(spapr);
1639 
1640     /*
1641      * We place the device tree and RTAS just below either the top of the RMA,
1642      * or just below 2GB, whichever is lower, so that it can be
1643      * processed with 32-bit real mode code if necessary
1644      */
1645     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1646 
1647     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1648 
1649     rc = fdt_pack(fdt);
1650 
1651     /* Should only fail if we've built a corrupted tree */
1652     assert(rc == 0);
1653 
1654     /* Load the fdt */
1655     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1656     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1657     g_free(spapr->fdt_blob);
1658     spapr->fdt_size = fdt_totalsize(fdt);
1659     spapr->fdt_initial_size = spapr->fdt_size;
1660     spapr->fdt_blob = fdt;
1661 
1662     /* Set up the entry state */
1663     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, 0, fdt_addr, 0);
1664     first_ppc_cpu->env.gpr[5] = 0;
1665 
1666     spapr->fwnmi_system_reset_addr = -1;
1667     spapr->fwnmi_machine_check_addr = -1;
1668     spapr->fwnmi_machine_check_interlock = -1;
1669 
1670     /* Signal all vCPUs waiting on this condition */
1671     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1672 
1673     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1674 }
1675 
1676 static void spapr_create_nvram(SpaprMachineState *spapr)
1677 {
1678     DeviceState *dev = qdev_new("spapr-nvram");
1679     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1680 
1681     if (dinfo) {
1682         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1683                                 &error_fatal);
1684     }
1685 
1686     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1687 
1688     spapr->nvram = (struct SpaprNvram *)dev;
1689 }
1690 
1691 static void spapr_rtc_create(SpaprMachineState *spapr)
1692 {
1693     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1694                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1695                                        &error_fatal, NULL);
1696     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1697     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1698                               "date");
1699 }
1700 
1701 /* Returns whether we want to use VGA or not */
1702 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1703 {
1704     switch (vga_interface_type) {
1705     case VGA_NONE:
1706         return false;
1707     case VGA_DEVICE:
1708         return true;
1709     case VGA_STD:
1710     case VGA_VIRTIO:
1711     case VGA_CIRRUS:
1712         return pci_vga_init(pci_bus) != NULL;
1713     default:
1714         error_setg(errp,
1715                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1716         return false;
1717     }
1718 }
1719 
1720 static int spapr_pre_load(void *opaque)
1721 {
1722     int rc;
1723 
1724     rc = spapr_caps_pre_load(opaque);
1725     if (rc) {
1726         return rc;
1727     }
1728 
1729     return 0;
1730 }
1731 
1732 static int spapr_post_load(void *opaque, int version_id)
1733 {
1734     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1735     int err = 0;
1736 
1737     err = spapr_caps_post_migration(spapr);
1738     if (err) {
1739         return err;
1740     }
1741 
1742     /*
1743      * In earlier versions, there was no separate qdev for the PAPR
1744      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1745      * So when migrating from those versions, poke the incoming offset
1746      * value into the RTC device
1747      */
1748     if (version_id < 3) {
1749         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1750         if (err) {
1751             return err;
1752         }
1753     }
1754 
1755     if (kvm_enabled() && spapr->patb_entry) {
1756         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1757         bool radix = !!(spapr->patb_entry & PATE1_GR);
1758         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1759 
1760         /*
1761          * Update LPCR:HR and UPRT as they may not be set properly in
1762          * the stream
1763          */
1764         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1765                             LPCR_HR | LPCR_UPRT);
1766 
1767         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1768         if (err) {
1769             error_report("Process table config unsupported by the host");
1770             return -EINVAL;
1771         }
1772     }
1773 
1774     err = spapr_irq_post_load(spapr, version_id);
1775     if (err) {
1776         return err;
1777     }
1778 
1779     return err;
1780 }
1781 
1782 static int spapr_pre_save(void *opaque)
1783 {
1784     int rc;
1785 
1786     rc = spapr_caps_pre_save(opaque);
1787     if (rc) {
1788         return rc;
1789     }
1790 
1791     return 0;
1792 }
1793 
1794 static bool version_before_3(void *opaque, int version_id)
1795 {
1796     return version_id < 3;
1797 }
1798 
1799 static bool spapr_pending_events_needed(void *opaque)
1800 {
1801     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1802     return !QTAILQ_EMPTY(&spapr->pending_events);
1803 }
1804 
1805 static const VMStateDescription vmstate_spapr_event_entry = {
1806     .name = "spapr_event_log_entry",
1807     .version_id = 1,
1808     .minimum_version_id = 1,
1809     .fields = (VMStateField[]) {
1810         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1811         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1812         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1813                                      NULL, extended_length),
1814         VMSTATE_END_OF_LIST()
1815     },
1816 };
1817 
1818 static const VMStateDescription vmstate_spapr_pending_events = {
1819     .name = "spapr_pending_events",
1820     .version_id = 1,
1821     .minimum_version_id = 1,
1822     .needed = spapr_pending_events_needed,
1823     .fields = (VMStateField[]) {
1824         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1825                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1826         VMSTATE_END_OF_LIST()
1827     },
1828 };
1829 
1830 static bool spapr_ov5_cas_needed(void *opaque)
1831 {
1832     SpaprMachineState *spapr = opaque;
1833     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1834     bool cas_needed;
1835 
1836     /* Prior to the introduction of SpaprOptionVector, we had two option
1837      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1838      * Both of these options encode machine topology into the device-tree
1839      * in such a way that the now-booted OS should still be able to interact
1840      * appropriately with QEMU regardless of what options were actually
1841      * negotiatied on the source side.
1842      *
1843      * As such, we can avoid migrating the CAS-negotiated options if these
1844      * are the only options available on the current machine/platform.
1845      * Since these are the only options available for pseries-2.7 and
1846      * earlier, this allows us to maintain old->new/new->old migration
1847      * compatibility.
1848      *
1849      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1850      * via default pseries-2.8 machines and explicit command-line parameters.
1851      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1852      * of the actual CAS-negotiated values to continue working properly. For
1853      * example, availability of memory unplug depends on knowing whether
1854      * OV5_HP_EVT was negotiated via CAS.
1855      *
1856      * Thus, for any cases where the set of available CAS-negotiatable
1857      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1858      * include the CAS-negotiated options in the migration stream, unless
1859      * if they affect boot time behaviour only.
1860      */
1861     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1862     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1863     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1864 
1865     /* We need extra information if we have any bits outside the mask
1866      * defined above */
1867     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1868 
1869     spapr_ovec_cleanup(ov5_mask);
1870 
1871     return cas_needed;
1872 }
1873 
1874 static const VMStateDescription vmstate_spapr_ov5_cas = {
1875     .name = "spapr_option_vector_ov5_cas",
1876     .version_id = 1,
1877     .minimum_version_id = 1,
1878     .needed = spapr_ov5_cas_needed,
1879     .fields = (VMStateField[]) {
1880         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1881                                  vmstate_spapr_ovec, SpaprOptionVector),
1882         VMSTATE_END_OF_LIST()
1883     },
1884 };
1885 
1886 static bool spapr_patb_entry_needed(void *opaque)
1887 {
1888     SpaprMachineState *spapr = opaque;
1889 
1890     return !!spapr->patb_entry;
1891 }
1892 
1893 static const VMStateDescription vmstate_spapr_patb_entry = {
1894     .name = "spapr_patb_entry",
1895     .version_id = 1,
1896     .minimum_version_id = 1,
1897     .needed = spapr_patb_entry_needed,
1898     .fields = (VMStateField[]) {
1899         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1900         VMSTATE_END_OF_LIST()
1901     },
1902 };
1903 
1904 static bool spapr_irq_map_needed(void *opaque)
1905 {
1906     SpaprMachineState *spapr = opaque;
1907 
1908     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1909 }
1910 
1911 static const VMStateDescription vmstate_spapr_irq_map = {
1912     .name = "spapr_irq_map",
1913     .version_id = 1,
1914     .minimum_version_id = 1,
1915     .needed = spapr_irq_map_needed,
1916     .fields = (VMStateField[]) {
1917         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1918         VMSTATE_END_OF_LIST()
1919     },
1920 };
1921 
1922 static bool spapr_dtb_needed(void *opaque)
1923 {
1924     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1925 
1926     return smc->update_dt_enabled;
1927 }
1928 
1929 static int spapr_dtb_pre_load(void *opaque)
1930 {
1931     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1932 
1933     g_free(spapr->fdt_blob);
1934     spapr->fdt_blob = NULL;
1935     spapr->fdt_size = 0;
1936 
1937     return 0;
1938 }
1939 
1940 static const VMStateDescription vmstate_spapr_dtb = {
1941     .name = "spapr_dtb",
1942     .version_id = 1,
1943     .minimum_version_id = 1,
1944     .needed = spapr_dtb_needed,
1945     .pre_load = spapr_dtb_pre_load,
1946     .fields = (VMStateField[]) {
1947         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1948         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1949         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1950                                      fdt_size),
1951         VMSTATE_END_OF_LIST()
1952     },
1953 };
1954 
1955 static bool spapr_fwnmi_needed(void *opaque)
1956 {
1957     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1958 
1959     return spapr->fwnmi_machine_check_addr != -1;
1960 }
1961 
1962 static int spapr_fwnmi_pre_save(void *opaque)
1963 {
1964     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1965 
1966     /*
1967      * Check if machine check handling is in progress and print a
1968      * warning message.
1969      */
1970     if (spapr->fwnmi_machine_check_interlock != -1) {
1971         warn_report("A machine check is being handled during migration. The"
1972                 "handler may run and log hardware error on the destination");
1973     }
1974 
1975     return 0;
1976 }
1977 
1978 static const VMStateDescription vmstate_spapr_fwnmi = {
1979     .name = "spapr_fwnmi",
1980     .version_id = 1,
1981     .minimum_version_id = 1,
1982     .needed = spapr_fwnmi_needed,
1983     .pre_save = spapr_fwnmi_pre_save,
1984     .fields = (VMStateField[]) {
1985         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
1986         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
1987         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
1988         VMSTATE_END_OF_LIST()
1989     },
1990 };
1991 
1992 static const VMStateDescription vmstate_spapr = {
1993     .name = "spapr",
1994     .version_id = 3,
1995     .minimum_version_id = 1,
1996     .pre_load = spapr_pre_load,
1997     .post_load = spapr_post_load,
1998     .pre_save = spapr_pre_save,
1999     .fields = (VMStateField[]) {
2000         /* used to be @next_irq */
2001         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2002 
2003         /* RTC offset */
2004         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2005 
2006         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2007         VMSTATE_END_OF_LIST()
2008     },
2009     .subsections = (const VMStateDescription*[]) {
2010         &vmstate_spapr_ov5_cas,
2011         &vmstate_spapr_patb_entry,
2012         &vmstate_spapr_pending_events,
2013         &vmstate_spapr_cap_htm,
2014         &vmstate_spapr_cap_vsx,
2015         &vmstate_spapr_cap_dfp,
2016         &vmstate_spapr_cap_cfpc,
2017         &vmstate_spapr_cap_sbbc,
2018         &vmstate_spapr_cap_ibs,
2019         &vmstate_spapr_cap_hpt_maxpagesize,
2020         &vmstate_spapr_irq_map,
2021         &vmstate_spapr_cap_nested_kvm_hv,
2022         &vmstate_spapr_dtb,
2023         &vmstate_spapr_cap_large_decr,
2024         &vmstate_spapr_cap_ccf_assist,
2025         &vmstate_spapr_cap_fwnmi,
2026         &vmstate_spapr_fwnmi,
2027         NULL
2028     }
2029 };
2030 
2031 static int htab_save_setup(QEMUFile *f, void *opaque)
2032 {
2033     SpaprMachineState *spapr = opaque;
2034 
2035     /* "Iteration" header */
2036     if (!spapr->htab_shift) {
2037         qemu_put_be32(f, -1);
2038     } else {
2039         qemu_put_be32(f, spapr->htab_shift);
2040     }
2041 
2042     if (spapr->htab) {
2043         spapr->htab_save_index = 0;
2044         spapr->htab_first_pass = true;
2045     } else {
2046         if (spapr->htab_shift) {
2047             assert(kvm_enabled());
2048         }
2049     }
2050 
2051 
2052     return 0;
2053 }
2054 
2055 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2056                             int chunkstart, int n_valid, int n_invalid)
2057 {
2058     qemu_put_be32(f, chunkstart);
2059     qemu_put_be16(f, n_valid);
2060     qemu_put_be16(f, n_invalid);
2061     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2062                     HASH_PTE_SIZE_64 * n_valid);
2063 }
2064 
2065 static void htab_save_end_marker(QEMUFile *f)
2066 {
2067     qemu_put_be32(f, 0);
2068     qemu_put_be16(f, 0);
2069     qemu_put_be16(f, 0);
2070 }
2071 
2072 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2073                                  int64_t max_ns)
2074 {
2075     bool has_timeout = max_ns != -1;
2076     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2077     int index = spapr->htab_save_index;
2078     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2079 
2080     assert(spapr->htab_first_pass);
2081 
2082     do {
2083         int chunkstart;
2084 
2085         /* Consume invalid HPTEs */
2086         while ((index < htabslots)
2087                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2088             CLEAN_HPTE(HPTE(spapr->htab, index));
2089             index++;
2090         }
2091 
2092         /* Consume valid HPTEs */
2093         chunkstart = index;
2094         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2095                && HPTE_VALID(HPTE(spapr->htab, index))) {
2096             CLEAN_HPTE(HPTE(spapr->htab, index));
2097             index++;
2098         }
2099 
2100         if (index > chunkstart) {
2101             int n_valid = index - chunkstart;
2102 
2103             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2104 
2105             if (has_timeout &&
2106                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2107                 break;
2108             }
2109         }
2110     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2111 
2112     if (index >= htabslots) {
2113         assert(index == htabslots);
2114         index = 0;
2115         spapr->htab_first_pass = false;
2116     }
2117     spapr->htab_save_index = index;
2118 }
2119 
2120 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2121                                 int64_t max_ns)
2122 {
2123     bool final = max_ns < 0;
2124     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2125     int examined = 0, sent = 0;
2126     int index = spapr->htab_save_index;
2127     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2128 
2129     assert(!spapr->htab_first_pass);
2130 
2131     do {
2132         int chunkstart, invalidstart;
2133 
2134         /* Consume non-dirty HPTEs */
2135         while ((index < htabslots)
2136                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2137             index++;
2138             examined++;
2139         }
2140 
2141         chunkstart = index;
2142         /* Consume valid dirty HPTEs */
2143         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2144                && HPTE_DIRTY(HPTE(spapr->htab, index))
2145                && HPTE_VALID(HPTE(spapr->htab, index))) {
2146             CLEAN_HPTE(HPTE(spapr->htab, index));
2147             index++;
2148             examined++;
2149         }
2150 
2151         invalidstart = index;
2152         /* Consume invalid dirty HPTEs */
2153         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2154                && HPTE_DIRTY(HPTE(spapr->htab, index))
2155                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2156             CLEAN_HPTE(HPTE(spapr->htab, index));
2157             index++;
2158             examined++;
2159         }
2160 
2161         if (index > chunkstart) {
2162             int n_valid = invalidstart - chunkstart;
2163             int n_invalid = index - invalidstart;
2164 
2165             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2166             sent += index - chunkstart;
2167 
2168             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2169                 break;
2170             }
2171         }
2172 
2173         if (examined >= htabslots) {
2174             break;
2175         }
2176 
2177         if (index >= htabslots) {
2178             assert(index == htabslots);
2179             index = 0;
2180         }
2181     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2182 
2183     if (index >= htabslots) {
2184         assert(index == htabslots);
2185         index = 0;
2186     }
2187 
2188     spapr->htab_save_index = index;
2189 
2190     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2191 }
2192 
2193 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2194 #define MAX_KVM_BUF_SIZE    2048
2195 
2196 static int htab_save_iterate(QEMUFile *f, void *opaque)
2197 {
2198     SpaprMachineState *spapr = opaque;
2199     int fd;
2200     int rc = 0;
2201 
2202     /* Iteration header */
2203     if (!spapr->htab_shift) {
2204         qemu_put_be32(f, -1);
2205         return 1;
2206     } else {
2207         qemu_put_be32(f, 0);
2208     }
2209 
2210     if (!spapr->htab) {
2211         assert(kvm_enabled());
2212 
2213         fd = get_htab_fd(spapr);
2214         if (fd < 0) {
2215             return fd;
2216         }
2217 
2218         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2219         if (rc < 0) {
2220             return rc;
2221         }
2222     } else  if (spapr->htab_first_pass) {
2223         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2224     } else {
2225         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2226     }
2227 
2228     htab_save_end_marker(f);
2229 
2230     return rc;
2231 }
2232 
2233 static int htab_save_complete(QEMUFile *f, void *opaque)
2234 {
2235     SpaprMachineState *spapr = opaque;
2236     int fd;
2237 
2238     /* Iteration header */
2239     if (!spapr->htab_shift) {
2240         qemu_put_be32(f, -1);
2241         return 0;
2242     } else {
2243         qemu_put_be32(f, 0);
2244     }
2245 
2246     if (!spapr->htab) {
2247         int rc;
2248 
2249         assert(kvm_enabled());
2250 
2251         fd = get_htab_fd(spapr);
2252         if (fd < 0) {
2253             return fd;
2254         }
2255 
2256         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2257         if (rc < 0) {
2258             return rc;
2259         }
2260     } else {
2261         if (spapr->htab_first_pass) {
2262             htab_save_first_pass(f, spapr, -1);
2263         }
2264         htab_save_later_pass(f, spapr, -1);
2265     }
2266 
2267     /* End marker */
2268     htab_save_end_marker(f);
2269 
2270     return 0;
2271 }
2272 
2273 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2274 {
2275     SpaprMachineState *spapr = opaque;
2276     uint32_t section_hdr;
2277     int fd = -1;
2278     Error *local_err = NULL;
2279 
2280     if (version_id < 1 || version_id > 1) {
2281         error_report("htab_load() bad version");
2282         return -EINVAL;
2283     }
2284 
2285     section_hdr = qemu_get_be32(f);
2286 
2287     if (section_hdr == -1) {
2288         spapr_free_hpt(spapr);
2289         return 0;
2290     }
2291 
2292     if (section_hdr) {
2293         /* First section gives the htab size */
2294         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2295         if (local_err) {
2296             error_report_err(local_err);
2297             return -EINVAL;
2298         }
2299         return 0;
2300     }
2301 
2302     if (!spapr->htab) {
2303         assert(kvm_enabled());
2304 
2305         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2306         if (fd < 0) {
2307             error_report_err(local_err);
2308             return fd;
2309         }
2310     }
2311 
2312     while (true) {
2313         uint32_t index;
2314         uint16_t n_valid, n_invalid;
2315 
2316         index = qemu_get_be32(f);
2317         n_valid = qemu_get_be16(f);
2318         n_invalid = qemu_get_be16(f);
2319 
2320         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2321             /* End of Stream */
2322             break;
2323         }
2324 
2325         if ((index + n_valid + n_invalid) >
2326             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2327             /* Bad index in stream */
2328             error_report(
2329                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2330                 index, n_valid, n_invalid, spapr->htab_shift);
2331             return -EINVAL;
2332         }
2333 
2334         if (spapr->htab) {
2335             if (n_valid) {
2336                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2337                                 HASH_PTE_SIZE_64 * n_valid);
2338             }
2339             if (n_invalid) {
2340                 memset(HPTE(spapr->htab, index + n_valid), 0,
2341                        HASH_PTE_SIZE_64 * n_invalid);
2342             }
2343         } else {
2344             int rc;
2345 
2346             assert(fd >= 0);
2347 
2348             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2349             if (rc < 0) {
2350                 return rc;
2351             }
2352         }
2353     }
2354 
2355     if (!spapr->htab) {
2356         assert(fd >= 0);
2357         close(fd);
2358     }
2359 
2360     return 0;
2361 }
2362 
2363 static void htab_save_cleanup(void *opaque)
2364 {
2365     SpaprMachineState *spapr = opaque;
2366 
2367     close_htab_fd(spapr);
2368 }
2369 
2370 static SaveVMHandlers savevm_htab_handlers = {
2371     .save_setup = htab_save_setup,
2372     .save_live_iterate = htab_save_iterate,
2373     .save_live_complete_precopy = htab_save_complete,
2374     .save_cleanup = htab_save_cleanup,
2375     .load_state = htab_load,
2376 };
2377 
2378 static void spapr_boot_set(void *opaque, const char *boot_device,
2379                            Error **errp)
2380 {
2381     MachineState *machine = MACHINE(opaque);
2382     machine->boot_order = g_strdup(boot_device);
2383 }
2384 
2385 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2386 {
2387     MachineState *machine = MACHINE(spapr);
2388     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2389     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2390     int i;
2391 
2392     for (i = 0; i < nr_lmbs; i++) {
2393         uint64_t addr;
2394 
2395         addr = i * lmb_size + machine->device_memory->base;
2396         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2397                                addr / lmb_size);
2398     }
2399 }
2400 
2401 /*
2402  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2403  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2404  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2405  */
2406 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2407 {
2408     int i;
2409 
2410     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2411         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2412                    " is not aligned to %" PRIu64 " MiB",
2413                    machine->ram_size,
2414                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2415         return;
2416     }
2417 
2418     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2419         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2420                    " is not aligned to %" PRIu64 " MiB",
2421                    machine->ram_size,
2422                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2423         return;
2424     }
2425 
2426     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2427         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2428             error_setg(errp,
2429                        "Node %d memory size 0x%" PRIx64
2430                        " is not aligned to %" PRIu64 " MiB",
2431                        i, machine->numa_state->nodes[i].node_mem,
2432                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2433             return;
2434         }
2435     }
2436 }
2437 
2438 /* find cpu slot in machine->possible_cpus by core_id */
2439 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2440 {
2441     int index = id / ms->smp.threads;
2442 
2443     if (index >= ms->possible_cpus->len) {
2444         return NULL;
2445     }
2446     if (idx) {
2447         *idx = index;
2448     }
2449     return &ms->possible_cpus->cpus[index];
2450 }
2451 
2452 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2453 {
2454     MachineState *ms = MACHINE(spapr);
2455     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2456     Error *local_err = NULL;
2457     bool vsmt_user = !!spapr->vsmt;
2458     int kvm_smt = kvmppc_smt_threads();
2459     int ret;
2460     unsigned int smp_threads = ms->smp.threads;
2461 
2462     if (!kvm_enabled() && (smp_threads > 1)) {
2463         error_setg(errp, "TCG cannot support more than 1 thread/core "
2464                    "on a pseries machine");
2465         return;
2466     }
2467     if (!is_power_of_2(smp_threads)) {
2468         error_setg(errp, "Cannot support %d threads/core on a pseries "
2469                    "machine because it must be a power of 2", smp_threads);
2470         return;
2471     }
2472 
2473     /* Detemine the VSMT mode to use: */
2474     if (vsmt_user) {
2475         if (spapr->vsmt < smp_threads) {
2476             error_setg(errp, "Cannot support VSMT mode %d"
2477                        " because it must be >= threads/core (%d)",
2478                        spapr->vsmt, smp_threads);
2479             return;
2480         }
2481         /* In this case, spapr->vsmt has been set by the command line */
2482     } else if (!smc->smp_threads_vsmt) {
2483         /*
2484          * Default VSMT value is tricky, because we need it to be as
2485          * consistent as possible (for migration), but this requires
2486          * changing it for at least some existing cases.  We pick 8 as
2487          * the value that we'd get with KVM on POWER8, the
2488          * overwhelmingly common case in production systems.
2489          */
2490         spapr->vsmt = MAX(8, smp_threads);
2491     } else {
2492         spapr->vsmt = smp_threads;
2493     }
2494 
2495     /* KVM: If necessary, set the SMT mode: */
2496     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2497         ret = kvmppc_set_smt_threads(spapr->vsmt);
2498         if (ret) {
2499             /* Looks like KVM isn't able to change VSMT mode */
2500             error_setg(&local_err,
2501                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2502                        spapr->vsmt, ret);
2503             /* We can live with that if the default one is big enough
2504              * for the number of threads, and a submultiple of the one
2505              * we want.  In this case we'll waste some vcpu ids, but
2506              * behaviour will be correct */
2507             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2508                 warn_report_err(local_err);
2509             } else {
2510                 if (!vsmt_user) {
2511                     error_append_hint(&local_err,
2512                                       "On PPC, a VM with %d threads/core"
2513                                       " on a host with %d threads/core"
2514                                       " requires the use of VSMT mode %d.\n",
2515                                       smp_threads, kvm_smt, spapr->vsmt);
2516                 }
2517                 kvmppc_error_append_smt_possible_hint(&local_err);
2518                 error_propagate(errp, local_err);
2519             }
2520         }
2521     }
2522     /* else TCG: nothing to do currently */
2523 }
2524 
2525 static void spapr_init_cpus(SpaprMachineState *spapr)
2526 {
2527     MachineState *machine = MACHINE(spapr);
2528     MachineClass *mc = MACHINE_GET_CLASS(machine);
2529     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2530     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2531     const CPUArchIdList *possible_cpus;
2532     unsigned int smp_cpus = machine->smp.cpus;
2533     unsigned int smp_threads = machine->smp.threads;
2534     unsigned int max_cpus = machine->smp.max_cpus;
2535     int boot_cores_nr = smp_cpus / smp_threads;
2536     int i;
2537 
2538     possible_cpus = mc->possible_cpu_arch_ids(machine);
2539     if (mc->has_hotpluggable_cpus) {
2540         if (smp_cpus % smp_threads) {
2541             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2542                          smp_cpus, smp_threads);
2543             exit(1);
2544         }
2545         if (max_cpus % smp_threads) {
2546             error_report("max_cpus (%u) must be multiple of threads (%u)",
2547                          max_cpus, smp_threads);
2548             exit(1);
2549         }
2550     } else {
2551         if (max_cpus != smp_cpus) {
2552             error_report("This machine version does not support CPU hotplug");
2553             exit(1);
2554         }
2555         boot_cores_nr = possible_cpus->len;
2556     }
2557 
2558     if (smc->pre_2_10_has_unused_icps) {
2559         int i;
2560 
2561         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2562             /* Dummy entries get deregistered when real ICPState objects
2563              * are registered during CPU core hotplug.
2564              */
2565             pre_2_10_vmstate_register_dummy_icp(i);
2566         }
2567     }
2568 
2569     for (i = 0; i < possible_cpus->len; i++) {
2570         int core_id = i * smp_threads;
2571 
2572         if (mc->has_hotpluggable_cpus) {
2573             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2574                                    spapr_vcpu_id(spapr, core_id));
2575         }
2576 
2577         if (i < boot_cores_nr) {
2578             Object *core  = object_new(type);
2579             int nr_threads = smp_threads;
2580 
2581             /* Handle the partially filled core for older machine types */
2582             if ((i + 1) * smp_threads >= smp_cpus) {
2583                 nr_threads = smp_cpus - i * smp_threads;
2584             }
2585 
2586             object_property_set_int(core, "nr-threads", nr_threads,
2587                                     &error_fatal);
2588             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2589                                     &error_fatal);
2590             qdev_realize(DEVICE(core), NULL, &error_fatal);
2591 
2592             object_unref(core);
2593         }
2594     }
2595 }
2596 
2597 static PCIHostState *spapr_create_default_phb(void)
2598 {
2599     DeviceState *dev;
2600 
2601     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2602     qdev_prop_set_uint32(dev, "index", 0);
2603     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2604 
2605     return PCI_HOST_BRIDGE(dev);
2606 }
2607 
2608 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2609 {
2610     MachineState *machine = MACHINE(spapr);
2611     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2612     hwaddr rma_size = machine->ram_size;
2613     hwaddr node0_size = spapr_node0_size(machine);
2614 
2615     /* RMA has to fit in the first NUMA node */
2616     rma_size = MIN(rma_size, node0_size);
2617 
2618     /*
2619      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2620      * never exceed that
2621      */
2622     rma_size = MIN(rma_size, 1 * TiB);
2623 
2624     /*
2625      * Clamp the RMA size based on machine type.  This is for
2626      * migration compatibility with older qemu versions, which limited
2627      * the RMA size for complicated and mostly bad reasons.
2628      */
2629     if (smc->rma_limit) {
2630         rma_size = MIN(rma_size, smc->rma_limit);
2631     }
2632 
2633     if (rma_size < MIN_RMA_SLOF) {
2634         error_setg(errp,
2635                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2636                    "ldMiB guest RMA (Real Mode Area memory)",
2637                    MIN_RMA_SLOF / MiB);
2638         return 0;
2639     }
2640 
2641     return rma_size;
2642 }
2643 
2644 /* pSeries LPAR / sPAPR hardware init */
2645 static void spapr_machine_init(MachineState *machine)
2646 {
2647     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2648     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2649     MachineClass *mc = MACHINE_GET_CLASS(machine);
2650     const char *kernel_filename = machine->kernel_filename;
2651     const char *initrd_filename = machine->initrd_filename;
2652     PCIHostState *phb;
2653     int i;
2654     MemoryRegion *sysmem = get_system_memory();
2655     long load_limit, fw_size;
2656     char *filename;
2657     Error *resize_hpt_err = NULL;
2658 
2659     msi_nonbroken = true;
2660 
2661     QLIST_INIT(&spapr->phbs);
2662     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2663 
2664     /* Determine capabilities to run with */
2665     spapr_caps_init(spapr);
2666 
2667     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2668     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2669         /*
2670          * If the user explicitly requested a mode we should either
2671          * supply it, or fail completely (which we do below).  But if
2672          * it's not set explicitly, we reset our mode to something
2673          * that works
2674          */
2675         if (resize_hpt_err) {
2676             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2677             error_free(resize_hpt_err);
2678             resize_hpt_err = NULL;
2679         } else {
2680             spapr->resize_hpt = smc->resize_hpt_default;
2681         }
2682     }
2683 
2684     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2685 
2686     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2687         /*
2688          * User requested HPT resize, but this host can't supply it.  Bail out
2689          */
2690         error_report_err(resize_hpt_err);
2691         exit(1);
2692     }
2693     error_free(resize_hpt_err);
2694 
2695     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2696 
2697     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2698     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2699 
2700     /*
2701      * VSMT must be set in order to be able to compute VCPU ids, ie to
2702      * call spapr_max_server_number() or spapr_vcpu_id().
2703      */
2704     spapr_set_vsmt_mode(spapr, &error_fatal);
2705 
2706     /* Set up Interrupt Controller before we create the VCPUs */
2707     spapr_irq_init(spapr, &error_fatal);
2708 
2709     /* Set up containers for ibm,client-architecture-support negotiated options
2710      */
2711     spapr->ov5 = spapr_ovec_new();
2712     spapr->ov5_cas = spapr_ovec_new();
2713 
2714     if (smc->dr_lmb_enabled) {
2715         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2716         spapr_validate_node_memory(machine, &error_fatal);
2717     }
2718 
2719     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2720 
2721     /* advertise support for dedicated HP event source to guests */
2722     if (spapr->use_hotplug_event_source) {
2723         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2724     }
2725 
2726     /* advertise support for HPT resizing */
2727     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2728         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2729     }
2730 
2731     /* advertise support for ibm,dyamic-memory-v2 */
2732     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2733 
2734     /* advertise XIVE on POWER9 machines */
2735     if (spapr->irq->xive) {
2736         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2737     }
2738 
2739     /* init CPUs */
2740     spapr_init_cpus(spapr);
2741 
2742     /*
2743      * check we don't have a memory-less/cpu-less NUMA node
2744      * Firmware relies on the existing memory/cpu topology to provide the
2745      * NUMA topology to the kernel.
2746      * And the linux kernel needs to know the NUMA topology at start
2747      * to be able to hotplug CPUs later.
2748      */
2749     if (machine->numa_state->num_nodes) {
2750         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2751             /* check for memory-less node */
2752             if (machine->numa_state->nodes[i].node_mem == 0) {
2753                 CPUState *cs;
2754                 int found = 0;
2755                 /* check for cpu-less node */
2756                 CPU_FOREACH(cs) {
2757                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2758                     if (cpu->node_id == i) {
2759                         found = 1;
2760                         break;
2761                     }
2762                 }
2763                 /* memory-less and cpu-less node */
2764                 if (!found) {
2765                     error_report(
2766                        "Memory-less/cpu-less nodes are not supported (node %d)",
2767                                  i);
2768                     exit(1);
2769                 }
2770             }
2771         }
2772 
2773     }
2774 
2775     /*
2776      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2777      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2778      * called from vPHB reset handler so we initialize the counter here.
2779      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2780      * must be equally distant from any other node.
2781      * The final value of spapr->gpu_numa_id is going to be written to
2782      * max-associativity-domains in spapr_build_fdt().
2783      */
2784     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2785 
2786     /* Init numa_assoc_array */
2787     spapr_numa_associativity_init(spapr, machine);
2788 
2789     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2790         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2791                               spapr->max_compat_pvr)) {
2792         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2793         /* KVM and TCG always allow GTSE with radix... */
2794         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2795     }
2796     /* ... but not with hash (currently). */
2797 
2798     if (kvm_enabled()) {
2799         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2800         kvmppc_enable_logical_ci_hcalls();
2801         kvmppc_enable_set_mode_hcall();
2802 
2803         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2804         kvmppc_enable_clear_ref_mod_hcalls();
2805 
2806         /* Enable H_PAGE_INIT */
2807         kvmppc_enable_h_page_init();
2808     }
2809 
2810     /* map RAM */
2811     memory_region_add_subregion(sysmem, 0, machine->ram);
2812 
2813     /* always allocate the device memory information */
2814     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2815 
2816     /* initialize hotplug memory address space */
2817     if (machine->ram_size < machine->maxram_size) {
2818         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2819         /*
2820          * Limit the number of hotpluggable memory slots to half the number
2821          * slots that KVM supports, leaving the other half for PCI and other
2822          * devices. However ensure that number of slots doesn't drop below 32.
2823          */
2824         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2825                            SPAPR_MAX_RAM_SLOTS;
2826 
2827         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2828             max_memslots = SPAPR_MAX_RAM_SLOTS;
2829         }
2830         if (machine->ram_slots > max_memslots) {
2831             error_report("Specified number of memory slots %"
2832                          PRIu64" exceeds max supported %d",
2833                          machine->ram_slots, max_memslots);
2834             exit(1);
2835         }
2836 
2837         machine->device_memory->base = ROUND_UP(machine->ram_size,
2838                                                 SPAPR_DEVICE_MEM_ALIGN);
2839         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2840                            "device-memory", device_mem_size);
2841         memory_region_add_subregion(sysmem, machine->device_memory->base,
2842                                     &machine->device_memory->mr);
2843     }
2844 
2845     if (smc->dr_lmb_enabled) {
2846         spapr_create_lmb_dr_connectors(spapr);
2847     }
2848 
2849     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2850         /* Create the error string for live migration blocker */
2851         error_setg(&spapr->fwnmi_migration_blocker,
2852             "A machine check is being handled during migration. The handler"
2853             "may run and log hardware error on the destination");
2854     }
2855 
2856     if (mc->nvdimm_supported) {
2857         spapr_create_nvdimm_dr_connectors(spapr);
2858     }
2859 
2860     /* Set up RTAS event infrastructure */
2861     spapr_events_init(spapr);
2862 
2863     /* Set up the RTC RTAS interfaces */
2864     spapr_rtc_create(spapr);
2865 
2866     /* Set up VIO bus */
2867     spapr->vio_bus = spapr_vio_bus_init();
2868 
2869     for (i = 0; i < serial_max_hds(); i++) {
2870         if (serial_hd(i)) {
2871             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2872         }
2873     }
2874 
2875     /* We always have at least the nvram device on VIO */
2876     spapr_create_nvram(spapr);
2877 
2878     /*
2879      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2880      * connectors (described in root DT node's "ibm,drc-types" property)
2881      * are pre-initialized here. additional child connectors (such as
2882      * connectors for a PHBs PCI slots) are added as needed during their
2883      * parent's realization.
2884      */
2885     if (smc->dr_phb_enabled) {
2886         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2887             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2888         }
2889     }
2890 
2891     /* Set up PCI */
2892     spapr_pci_rtas_init();
2893 
2894     phb = spapr_create_default_phb();
2895 
2896     for (i = 0; i < nb_nics; i++) {
2897         NICInfo *nd = &nd_table[i];
2898 
2899         if (!nd->model) {
2900             nd->model = g_strdup("spapr-vlan");
2901         }
2902 
2903         if (g_str_equal(nd->model, "spapr-vlan") ||
2904             g_str_equal(nd->model, "ibmveth")) {
2905             spapr_vlan_create(spapr->vio_bus, nd);
2906         } else {
2907             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2908         }
2909     }
2910 
2911     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2912         spapr_vscsi_create(spapr->vio_bus);
2913     }
2914 
2915     /* Graphics */
2916     if (spapr_vga_init(phb->bus, &error_fatal)) {
2917         spapr->has_graphics = true;
2918         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2919     }
2920 
2921     if (machine->usb) {
2922         if (smc->use_ohci_by_default) {
2923             pci_create_simple(phb->bus, -1, "pci-ohci");
2924         } else {
2925             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2926         }
2927 
2928         if (spapr->has_graphics) {
2929             USBBus *usb_bus = usb_bus_find(-1);
2930 
2931             usb_create_simple(usb_bus, "usb-kbd");
2932             usb_create_simple(usb_bus, "usb-mouse");
2933         }
2934     }
2935 
2936     if (kernel_filename) {
2937         spapr->kernel_size = load_elf(kernel_filename, NULL,
2938                                       translate_kernel_address, spapr,
2939                                       NULL, NULL, NULL, NULL, 1,
2940                                       PPC_ELF_MACHINE, 0, 0);
2941         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2942             spapr->kernel_size = load_elf(kernel_filename, NULL,
2943                                           translate_kernel_address, spapr,
2944                                           NULL, NULL, NULL, NULL, 0,
2945                                           PPC_ELF_MACHINE, 0, 0);
2946             spapr->kernel_le = spapr->kernel_size > 0;
2947         }
2948         if (spapr->kernel_size < 0) {
2949             error_report("error loading %s: %s", kernel_filename,
2950                          load_elf_strerror(spapr->kernel_size));
2951             exit(1);
2952         }
2953 
2954         /* load initrd */
2955         if (initrd_filename) {
2956             /* Try to locate the initrd in the gap between the kernel
2957              * and the firmware. Add a bit of space just in case
2958              */
2959             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
2960                                   + 0x1ffff) & ~0xffff;
2961             spapr->initrd_size = load_image_targphys(initrd_filename,
2962                                                      spapr->initrd_base,
2963                                                      load_limit
2964                                                      - spapr->initrd_base);
2965             if (spapr->initrd_size < 0) {
2966                 error_report("could not load initial ram disk '%s'",
2967                              initrd_filename);
2968                 exit(1);
2969             }
2970         }
2971     }
2972 
2973     if (bios_name == NULL) {
2974         bios_name = FW_FILE_NAME;
2975     }
2976     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2977     if (!filename) {
2978         error_report("Could not find LPAR firmware '%s'", bios_name);
2979         exit(1);
2980     }
2981     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2982     if (fw_size <= 0) {
2983         error_report("Could not load LPAR firmware '%s'", filename);
2984         exit(1);
2985     }
2986     g_free(filename);
2987 
2988     /* FIXME: Should register things through the MachineState's qdev
2989      * interface, this is a legacy from the sPAPREnvironment structure
2990      * which predated MachineState but had a similar function */
2991     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2992     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
2993                          &savevm_htab_handlers, spapr);
2994 
2995     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
2996 
2997     qemu_register_boot_set(spapr_boot_set, spapr);
2998 
2999     /*
3000      * Nothing needs to be done to resume a suspended guest because
3001      * suspending does not change the machine state, so no need for
3002      * a ->wakeup method.
3003      */
3004     qemu_register_wakeup_support();
3005 
3006     if (kvm_enabled()) {
3007         /* to stop and start vmclock */
3008         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3009                                          &spapr->tb);
3010 
3011         kvmppc_spapr_enable_inkernel_multitce();
3012     }
3013 
3014     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3015 }
3016 
3017 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3018 {
3019     if (!vm_type) {
3020         return 0;
3021     }
3022 
3023     if (!strcmp(vm_type, "HV")) {
3024         return 1;
3025     }
3026 
3027     if (!strcmp(vm_type, "PR")) {
3028         return 2;
3029     }
3030 
3031     error_report("Unknown kvm-type specified '%s'", vm_type);
3032     exit(1);
3033 }
3034 
3035 /*
3036  * Implementation of an interface to adjust firmware path
3037  * for the bootindex property handling.
3038  */
3039 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3040                                    DeviceState *dev)
3041 {
3042 #define CAST(type, obj, name) \
3043     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3044     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3045     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3046     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3047 
3048     if (d) {
3049         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3050         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3051         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3052 
3053         if (spapr) {
3054             /*
3055              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3056              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3057              * 0x8000 | (target << 8) | (bus << 5) | lun
3058              * (see the "Logical unit addressing format" table in SAM5)
3059              */
3060             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3061             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3062                                    (uint64_t)id << 48);
3063         } else if (virtio) {
3064             /*
3065              * We use SRP luns of the form 01000000 | (target << 8) | lun
3066              * in the top 32 bits of the 64-bit LUN
3067              * Note: the quote above is from SLOF and it is wrong,
3068              * the actual binding is:
3069              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3070              */
3071             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3072             if (d->lun >= 256) {
3073                 /* Use the LUN "flat space addressing method" */
3074                 id |= 0x4000;
3075             }
3076             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3077                                    (uint64_t)id << 32);
3078         } else if (usb) {
3079             /*
3080              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3081              * in the top 32 bits of the 64-bit LUN
3082              */
3083             unsigned usb_port = atoi(usb->port->path);
3084             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3085             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3086                                    (uint64_t)id << 32);
3087         }
3088     }
3089 
3090     /*
3091      * SLOF probes the USB devices, and if it recognizes that the device is a
3092      * storage device, it changes its name to "storage" instead of "usb-host",
3093      * and additionally adds a child node for the SCSI LUN, so the correct
3094      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3095      */
3096     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3097         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3098         if (usb_host_dev_is_scsi_storage(usbdev)) {
3099             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3100         }
3101     }
3102 
3103     if (phb) {
3104         /* Replace "pci" with "pci@800000020000000" */
3105         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3106     }
3107 
3108     if (vsc) {
3109         /* Same logic as virtio above */
3110         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3111         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3112     }
3113 
3114     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3115         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3116         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3117         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3118     }
3119 
3120     return NULL;
3121 }
3122 
3123 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3124 {
3125     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3126 
3127     return g_strdup(spapr->kvm_type);
3128 }
3129 
3130 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3131 {
3132     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3133 
3134     g_free(spapr->kvm_type);
3135     spapr->kvm_type = g_strdup(value);
3136 }
3137 
3138 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3139 {
3140     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3141 
3142     return spapr->use_hotplug_event_source;
3143 }
3144 
3145 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3146                                             Error **errp)
3147 {
3148     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3149 
3150     spapr->use_hotplug_event_source = value;
3151 }
3152 
3153 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3154 {
3155     return true;
3156 }
3157 
3158 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3159 {
3160     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3161 
3162     switch (spapr->resize_hpt) {
3163     case SPAPR_RESIZE_HPT_DEFAULT:
3164         return g_strdup("default");
3165     case SPAPR_RESIZE_HPT_DISABLED:
3166         return g_strdup("disabled");
3167     case SPAPR_RESIZE_HPT_ENABLED:
3168         return g_strdup("enabled");
3169     case SPAPR_RESIZE_HPT_REQUIRED:
3170         return g_strdup("required");
3171     }
3172     g_assert_not_reached();
3173 }
3174 
3175 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3176 {
3177     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3178 
3179     if (strcmp(value, "default") == 0) {
3180         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3181     } else if (strcmp(value, "disabled") == 0) {
3182         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3183     } else if (strcmp(value, "enabled") == 0) {
3184         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3185     } else if (strcmp(value, "required") == 0) {
3186         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3187     } else {
3188         error_setg(errp, "Bad value for \"resize-hpt\" property");
3189     }
3190 }
3191 
3192 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3193 {
3194     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3195 
3196     if (spapr->irq == &spapr_irq_xics_legacy) {
3197         return g_strdup("legacy");
3198     } else if (spapr->irq == &spapr_irq_xics) {
3199         return g_strdup("xics");
3200     } else if (spapr->irq == &spapr_irq_xive) {
3201         return g_strdup("xive");
3202     } else if (spapr->irq == &spapr_irq_dual) {
3203         return g_strdup("dual");
3204     }
3205     g_assert_not_reached();
3206 }
3207 
3208 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3209 {
3210     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3211 
3212     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3213         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3214         return;
3215     }
3216 
3217     /* The legacy IRQ backend can not be set */
3218     if (strcmp(value, "xics") == 0) {
3219         spapr->irq = &spapr_irq_xics;
3220     } else if (strcmp(value, "xive") == 0) {
3221         spapr->irq = &spapr_irq_xive;
3222     } else if (strcmp(value, "dual") == 0) {
3223         spapr->irq = &spapr_irq_dual;
3224     } else {
3225         error_setg(errp, "Bad value for \"ic-mode\" property");
3226     }
3227 }
3228 
3229 static char *spapr_get_host_model(Object *obj, Error **errp)
3230 {
3231     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3232 
3233     return g_strdup(spapr->host_model);
3234 }
3235 
3236 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3237 {
3238     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3239 
3240     g_free(spapr->host_model);
3241     spapr->host_model = g_strdup(value);
3242 }
3243 
3244 static char *spapr_get_host_serial(Object *obj, Error **errp)
3245 {
3246     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3247 
3248     return g_strdup(spapr->host_serial);
3249 }
3250 
3251 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3252 {
3253     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3254 
3255     g_free(spapr->host_serial);
3256     spapr->host_serial = g_strdup(value);
3257 }
3258 
3259 static void spapr_instance_init(Object *obj)
3260 {
3261     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3262     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3263 
3264     spapr->htab_fd = -1;
3265     spapr->use_hotplug_event_source = true;
3266     object_property_add_str(obj, "kvm-type",
3267                             spapr_get_kvm_type, spapr_set_kvm_type);
3268     object_property_set_description(obj, "kvm-type",
3269                                     "Specifies the KVM virtualization mode (HV, PR)");
3270     object_property_add_bool(obj, "modern-hotplug-events",
3271                             spapr_get_modern_hotplug_events,
3272                             spapr_set_modern_hotplug_events);
3273     object_property_set_description(obj, "modern-hotplug-events",
3274                                     "Use dedicated hotplug event mechanism in"
3275                                     " place of standard EPOW events when possible"
3276                                     " (required for memory hot-unplug support)");
3277     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3278                             "Maximum permitted CPU compatibility mode");
3279 
3280     object_property_add_str(obj, "resize-hpt",
3281                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3282     object_property_set_description(obj, "resize-hpt",
3283                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3284     object_property_add_uint32_ptr(obj, "vsmt",
3285                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3286     object_property_set_description(obj, "vsmt",
3287                                     "Virtual SMT: KVM behaves as if this were"
3288                                     " the host's SMT mode");
3289 
3290     object_property_add_bool(obj, "vfio-no-msix-emulation",
3291                              spapr_get_msix_emulation, NULL);
3292 
3293     object_property_add_uint64_ptr(obj, "kernel-addr",
3294                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3295     object_property_set_description(obj, "kernel-addr",
3296                                     stringify(KERNEL_LOAD_ADDR)
3297                                     " for -kernel is the default");
3298     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3299     /* The machine class defines the default interrupt controller mode */
3300     spapr->irq = smc->irq;
3301     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3302                             spapr_set_ic_mode);
3303     object_property_set_description(obj, "ic-mode",
3304                  "Specifies the interrupt controller mode (xics, xive, dual)");
3305 
3306     object_property_add_str(obj, "host-model",
3307         spapr_get_host_model, spapr_set_host_model);
3308     object_property_set_description(obj, "host-model",
3309         "Host model to advertise in guest device tree");
3310     object_property_add_str(obj, "host-serial",
3311         spapr_get_host_serial, spapr_set_host_serial);
3312     object_property_set_description(obj, "host-serial",
3313         "Host serial number to advertise in guest device tree");
3314 }
3315 
3316 static void spapr_machine_finalizefn(Object *obj)
3317 {
3318     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3319 
3320     g_free(spapr->kvm_type);
3321 }
3322 
3323 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3324 {
3325     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3326     PowerPCCPU *cpu = POWERPC_CPU(cs);
3327     CPUPPCState *env = &cpu->env;
3328 
3329     cpu_synchronize_state(cs);
3330     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3331     if (spapr->fwnmi_system_reset_addr != -1) {
3332         uint64_t rtas_addr, addr;
3333 
3334         /* get rtas addr from fdt */
3335         rtas_addr = spapr_get_rtas_addr();
3336         if (!rtas_addr) {
3337             qemu_system_guest_panicked(NULL);
3338             return;
3339         }
3340 
3341         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3342         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3343         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3344         env->gpr[3] = addr;
3345     }
3346     ppc_cpu_do_system_reset(cs);
3347     if (spapr->fwnmi_system_reset_addr != -1) {
3348         env->nip = spapr->fwnmi_system_reset_addr;
3349     }
3350 }
3351 
3352 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3353 {
3354     CPUState *cs;
3355 
3356     CPU_FOREACH(cs) {
3357         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3358     }
3359 }
3360 
3361 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3362                           void *fdt, int *fdt_start_offset, Error **errp)
3363 {
3364     uint64_t addr;
3365     uint32_t node;
3366 
3367     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3368     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3369                                     &error_abort);
3370     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3371                                              SPAPR_MEMORY_BLOCK_SIZE);
3372     return 0;
3373 }
3374 
3375 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3376                            bool dedicated_hp_event_source, Error **errp)
3377 {
3378     SpaprDrc *drc;
3379     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3380     int i;
3381     uint64_t addr = addr_start;
3382     bool hotplugged = spapr_drc_hotplugged(dev);
3383 
3384     for (i = 0; i < nr_lmbs; i++) {
3385         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3386                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3387         g_assert(drc);
3388 
3389         if (!spapr_drc_attach(drc, dev, errp)) {
3390             while (addr > addr_start) {
3391                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3392                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3393                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3394                 spapr_drc_detach(drc);
3395             }
3396             return;
3397         }
3398         if (!hotplugged) {
3399             spapr_drc_reset(drc);
3400         }
3401         addr += SPAPR_MEMORY_BLOCK_SIZE;
3402     }
3403     /* send hotplug notification to the
3404      * guest only in case of hotplugged memory
3405      */
3406     if (hotplugged) {
3407         if (dedicated_hp_event_source) {
3408             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3409                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3410             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3411                                                    nr_lmbs,
3412                                                    spapr_drc_index(drc));
3413         } else {
3414             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3415                                            nr_lmbs);
3416         }
3417     }
3418 }
3419 
3420 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3421                               Error **errp)
3422 {
3423     Error *local_err = NULL;
3424     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3425     PCDIMMDevice *dimm = PC_DIMM(dev);
3426     uint64_t size, addr, slot;
3427     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3428 
3429     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3430 
3431     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3432     if (local_err) {
3433         goto out;
3434     }
3435 
3436     if (!is_nvdimm) {
3437         addr = object_property_get_uint(OBJECT(dimm),
3438                                         PC_DIMM_ADDR_PROP, &local_err);
3439         if (local_err) {
3440             goto out_unplug;
3441         }
3442         spapr_add_lmbs(dev, addr, size,
3443                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3444                        &local_err);
3445     } else {
3446         slot = object_property_get_uint(OBJECT(dimm),
3447                                         PC_DIMM_SLOT_PROP, &local_err);
3448         if (local_err) {
3449             goto out_unplug;
3450         }
3451         spapr_add_nvdimm(dev, slot, &local_err);
3452     }
3453 
3454     if (local_err) {
3455         goto out_unplug;
3456     }
3457 
3458     return;
3459 
3460 out_unplug:
3461     pc_dimm_unplug(dimm, MACHINE(ms));
3462 out:
3463     error_propagate(errp, local_err);
3464 }
3465 
3466 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3467                                   Error **errp)
3468 {
3469     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3470     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3471     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3472     PCDIMMDevice *dimm = PC_DIMM(dev);
3473     Error *local_err = NULL;
3474     uint64_t size;
3475     Object *memdev;
3476     hwaddr pagesize;
3477 
3478     if (!smc->dr_lmb_enabled) {
3479         error_setg(errp, "Memory hotplug not supported for this machine");
3480         return;
3481     }
3482 
3483     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3484     if (local_err) {
3485         error_propagate(errp, local_err);
3486         return;
3487     }
3488 
3489     if (is_nvdimm) {
3490         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3491             return;
3492         }
3493     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3494         error_setg(errp, "Hotplugged memory size must be a multiple of "
3495                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3496         return;
3497     }
3498 
3499     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3500                                       &error_abort);
3501     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3502     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3503         return;
3504     }
3505 
3506     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3507 }
3508 
3509 struct SpaprDimmState {
3510     PCDIMMDevice *dimm;
3511     uint32_t nr_lmbs;
3512     QTAILQ_ENTRY(SpaprDimmState) next;
3513 };
3514 
3515 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3516                                                        PCDIMMDevice *dimm)
3517 {
3518     SpaprDimmState *dimm_state = NULL;
3519 
3520     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3521         if (dimm_state->dimm == dimm) {
3522             break;
3523         }
3524     }
3525     return dimm_state;
3526 }
3527 
3528 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3529                                                       uint32_t nr_lmbs,
3530                                                       PCDIMMDevice *dimm)
3531 {
3532     SpaprDimmState *ds = NULL;
3533 
3534     /*
3535      * If this request is for a DIMM whose removal had failed earlier
3536      * (due to guest's refusal to remove the LMBs), we would have this
3537      * dimm already in the pending_dimm_unplugs list. In that
3538      * case don't add again.
3539      */
3540     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3541     if (!ds) {
3542         ds = g_malloc0(sizeof(SpaprDimmState));
3543         ds->nr_lmbs = nr_lmbs;
3544         ds->dimm = dimm;
3545         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3546     }
3547     return ds;
3548 }
3549 
3550 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3551                                               SpaprDimmState *dimm_state)
3552 {
3553     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3554     g_free(dimm_state);
3555 }
3556 
3557 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3558                                                         PCDIMMDevice *dimm)
3559 {
3560     SpaprDrc *drc;
3561     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3562                                                   &error_abort);
3563     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3564     uint32_t avail_lmbs = 0;
3565     uint64_t addr_start, addr;
3566     int i;
3567 
3568     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3569                                          &error_abort);
3570 
3571     addr = addr_start;
3572     for (i = 0; i < nr_lmbs; i++) {
3573         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3574                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3575         g_assert(drc);
3576         if (drc->dev) {
3577             avail_lmbs++;
3578         }
3579         addr += SPAPR_MEMORY_BLOCK_SIZE;
3580     }
3581 
3582     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3583 }
3584 
3585 /* Callback to be called during DRC release. */
3586 void spapr_lmb_release(DeviceState *dev)
3587 {
3588     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3589     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3590     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3591 
3592     /* This information will get lost if a migration occurs
3593      * during the unplug process. In this case recover it. */
3594     if (ds == NULL) {
3595         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3596         g_assert(ds);
3597         /* The DRC being examined by the caller at least must be counted */
3598         g_assert(ds->nr_lmbs);
3599     }
3600 
3601     if (--ds->nr_lmbs) {
3602         return;
3603     }
3604 
3605     /*
3606      * Now that all the LMBs have been removed by the guest, call the
3607      * unplug handler chain. This can never fail.
3608      */
3609     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3610     object_unparent(OBJECT(dev));
3611 }
3612 
3613 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3614 {
3615     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3616     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3617 
3618     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3619     qdev_unrealize(dev);
3620     spapr_pending_dimm_unplugs_remove(spapr, ds);
3621 }
3622 
3623 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3624                                         DeviceState *dev, Error **errp)
3625 {
3626     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3627     Error *local_err = NULL;
3628     PCDIMMDevice *dimm = PC_DIMM(dev);
3629     uint32_t nr_lmbs;
3630     uint64_t size, addr_start, addr;
3631     int i;
3632     SpaprDrc *drc;
3633 
3634     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3635         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3636         return;
3637     }
3638 
3639     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3640     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3641 
3642     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3643                                          &local_err);
3644     if (local_err) {
3645         error_propagate(errp, local_err);
3646         return;
3647     }
3648 
3649     /*
3650      * An existing pending dimm state for this DIMM means that there is an
3651      * unplug operation in progress, waiting for the spapr_lmb_release
3652      * callback to complete the job (BQL can't cover that far). In this case,
3653      * bail out to avoid detaching DRCs that were already released.
3654      */
3655     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3656         error_setg(errp, "Memory unplug already in progress for device %s",
3657                    dev->id);
3658         return;
3659     }
3660 
3661     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3662 
3663     addr = addr_start;
3664     for (i = 0; i < nr_lmbs; i++) {
3665         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3666                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3667         g_assert(drc);
3668 
3669         spapr_drc_detach(drc);
3670         addr += SPAPR_MEMORY_BLOCK_SIZE;
3671     }
3672 
3673     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3674                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3675     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3676                                               nr_lmbs, spapr_drc_index(drc));
3677 }
3678 
3679 /* Callback to be called during DRC release. */
3680 void spapr_core_release(DeviceState *dev)
3681 {
3682     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3683 
3684     /* Call the unplug handler chain. This can never fail. */
3685     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3686     object_unparent(OBJECT(dev));
3687 }
3688 
3689 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3690 {
3691     MachineState *ms = MACHINE(hotplug_dev);
3692     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3693     CPUCore *cc = CPU_CORE(dev);
3694     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3695 
3696     if (smc->pre_2_10_has_unused_icps) {
3697         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3698         int i;
3699 
3700         for (i = 0; i < cc->nr_threads; i++) {
3701             CPUState *cs = CPU(sc->threads[i]);
3702 
3703             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3704         }
3705     }
3706 
3707     assert(core_slot);
3708     core_slot->cpu = NULL;
3709     qdev_unrealize(dev);
3710 }
3711 
3712 static
3713 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3714                                Error **errp)
3715 {
3716     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3717     int index;
3718     SpaprDrc *drc;
3719     CPUCore *cc = CPU_CORE(dev);
3720 
3721     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3722         error_setg(errp, "Unable to find CPU core with core-id: %d",
3723                    cc->core_id);
3724         return;
3725     }
3726     if (index == 0) {
3727         error_setg(errp, "Boot CPU core may not be unplugged");
3728         return;
3729     }
3730 
3731     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3732                           spapr_vcpu_id(spapr, cc->core_id));
3733     g_assert(drc);
3734 
3735     if (!spapr_drc_unplug_requested(drc)) {
3736         spapr_drc_detach(drc);
3737         spapr_hotplug_req_remove_by_index(drc);
3738     }
3739 }
3740 
3741 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3742                            void *fdt, int *fdt_start_offset, Error **errp)
3743 {
3744     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3745     CPUState *cs = CPU(core->threads[0]);
3746     PowerPCCPU *cpu = POWERPC_CPU(cs);
3747     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3748     int id = spapr_get_vcpu_id(cpu);
3749     char *nodename;
3750     int offset;
3751 
3752     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3753     offset = fdt_add_subnode(fdt, 0, nodename);
3754     g_free(nodename);
3755 
3756     spapr_dt_cpu(cs, fdt, offset, spapr);
3757 
3758     *fdt_start_offset = offset;
3759     return 0;
3760 }
3761 
3762 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3763                             Error **errp)
3764 {
3765     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3766     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3767     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3768     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3769     CPUCore *cc = CPU_CORE(dev);
3770     CPUState *cs;
3771     SpaprDrc *drc;
3772     CPUArchId *core_slot;
3773     int index;
3774     bool hotplugged = spapr_drc_hotplugged(dev);
3775     int i;
3776 
3777     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3778     if (!core_slot) {
3779         error_setg(errp, "Unable to find CPU core with core-id: %d",
3780                    cc->core_id);
3781         return;
3782     }
3783     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3784                           spapr_vcpu_id(spapr, cc->core_id));
3785 
3786     g_assert(drc || !mc->has_hotpluggable_cpus);
3787 
3788     if (drc) {
3789         if (!spapr_drc_attach(drc, dev, errp)) {
3790             return;
3791         }
3792 
3793         if (hotplugged) {
3794             /*
3795              * Send hotplug notification interrupt to the guest only
3796              * in case of hotplugged CPUs.
3797              */
3798             spapr_hotplug_req_add_by_index(drc);
3799         } else {
3800             spapr_drc_reset(drc);
3801         }
3802     }
3803 
3804     core_slot->cpu = OBJECT(dev);
3805 
3806     if (smc->pre_2_10_has_unused_icps) {
3807         for (i = 0; i < cc->nr_threads; i++) {
3808             cs = CPU(core->threads[i]);
3809             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3810         }
3811     }
3812 
3813     /*
3814      * Set compatibility mode to match the boot CPU, which was either set
3815      * by the machine reset code or by CAS.
3816      */
3817     if (hotplugged) {
3818         for (i = 0; i < cc->nr_threads; i++) {
3819             if (ppc_set_compat(core->threads[i],
3820                                POWERPC_CPU(first_cpu)->compat_pvr,
3821                                errp) < 0) {
3822                 return;
3823             }
3824         }
3825     }
3826 }
3827 
3828 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3829                                 Error **errp)
3830 {
3831     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3832     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3833     CPUCore *cc = CPU_CORE(dev);
3834     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3835     const char *type = object_get_typename(OBJECT(dev));
3836     CPUArchId *core_slot;
3837     int index;
3838     unsigned int smp_threads = machine->smp.threads;
3839 
3840     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3841         error_setg(errp, "CPU hotplug not supported for this machine");
3842         return;
3843     }
3844 
3845     if (strcmp(base_core_type, type)) {
3846         error_setg(errp, "CPU core type should be %s", base_core_type);
3847         return;
3848     }
3849 
3850     if (cc->core_id % smp_threads) {
3851         error_setg(errp, "invalid core id %d", cc->core_id);
3852         return;
3853     }
3854 
3855     /*
3856      * In general we should have homogeneous threads-per-core, but old
3857      * (pre hotplug support) machine types allow the last core to have
3858      * reduced threads as a compatibility hack for when we allowed
3859      * total vcpus not a multiple of threads-per-core.
3860      */
3861     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3862         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
3863                    smp_threads);
3864         return;
3865     }
3866 
3867     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3868     if (!core_slot) {
3869         error_setg(errp, "core id %d out of range", cc->core_id);
3870         return;
3871     }
3872 
3873     if (core_slot->cpu) {
3874         error_setg(errp, "core %d already populated", cc->core_id);
3875         return;
3876     }
3877 
3878     numa_cpu_pre_plug(core_slot, dev, errp);
3879 }
3880 
3881 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3882                           void *fdt, int *fdt_start_offset, Error **errp)
3883 {
3884     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3885     int intc_phandle;
3886 
3887     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3888     if (intc_phandle <= 0) {
3889         return -1;
3890     }
3891 
3892     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3893         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3894         return -1;
3895     }
3896 
3897     /* generally SLOF creates these, for hotplug it's up to QEMU */
3898     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3899 
3900     return 0;
3901 }
3902 
3903 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3904                                Error **errp)
3905 {
3906     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3907     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3908     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3909     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3910 
3911     if (dev->hotplugged && !smc->dr_phb_enabled) {
3912         error_setg(errp, "PHB hotplug not supported for this machine");
3913         return;
3914     }
3915 
3916     if (sphb->index == (uint32_t)-1) {
3917         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3918         return;
3919     }
3920 
3921     /*
3922      * This will check that sphb->index doesn't exceed the maximum number of
3923      * PHBs for the current machine type.
3924      */
3925     smc->phb_placement(spapr, sphb->index,
3926                        &sphb->buid, &sphb->io_win_addr,
3927                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
3928                        windows_supported, sphb->dma_liobn,
3929                        &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3930                        errp);
3931 }
3932 
3933 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3934                            Error **errp)
3935 {
3936     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3937     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3938     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3939     SpaprDrc *drc;
3940     bool hotplugged = spapr_drc_hotplugged(dev);
3941 
3942     if (!smc->dr_phb_enabled) {
3943         return;
3944     }
3945 
3946     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3947     /* hotplug hooks should check it's enabled before getting this far */
3948     assert(drc);
3949 
3950     if (!spapr_drc_attach(drc, dev, errp)) {
3951         return;
3952     }
3953 
3954     if (hotplugged) {
3955         spapr_hotplug_req_add_by_index(drc);
3956     } else {
3957         spapr_drc_reset(drc);
3958     }
3959 }
3960 
3961 void spapr_phb_release(DeviceState *dev)
3962 {
3963     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3964 
3965     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3966     object_unparent(OBJECT(dev));
3967 }
3968 
3969 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3970 {
3971     qdev_unrealize(dev);
3972 }
3973 
3974 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
3975                                      DeviceState *dev, Error **errp)
3976 {
3977     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3978     SpaprDrc *drc;
3979 
3980     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3981     assert(drc);
3982 
3983     if (!spapr_drc_unplug_requested(drc)) {
3984         spapr_drc_detach(drc);
3985         spapr_hotplug_req_remove_by_index(drc);
3986     }
3987 }
3988 
3989 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3990                                  Error **errp)
3991 {
3992     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3993     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
3994 
3995     if (spapr->tpm_proxy != NULL) {
3996         error_setg(errp, "Only one TPM proxy can be specified for this machine");
3997         return;
3998     }
3999 
4000     spapr->tpm_proxy = tpm_proxy;
4001 }
4002 
4003 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4004 {
4005     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4006 
4007     qdev_unrealize(dev);
4008     object_unparent(OBJECT(dev));
4009     spapr->tpm_proxy = NULL;
4010 }
4011 
4012 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4013                                       DeviceState *dev, Error **errp)
4014 {
4015     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4016         spapr_memory_plug(hotplug_dev, dev, errp);
4017     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4018         spapr_core_plug(hotplug_dev, dev, errp);
4019     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4020         spapr_phb_plug(hotplug_dev, dev, errp);
4021     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4022         spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
4023     }
4024 }
4025 
4026 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4027                                         DeviceState *dev, Error **errp)
4028 {
4029     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4030         spapr_memory_unplug(hotplug_dev, dev);
4031     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4032         spapr_core_unplug(hotplug_dev, dev);
4033     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4034         spapr_phb_unplug(hotplug_dev, dev);
4035     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4036         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4037     }
4038 }
4039 
4040 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4041                                                 DeviceState *dev, Error **errp)
4042 {
4043     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4044     MachineClass *mc = MACHINE_GET_CLASS(sms);
4045     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4046 
4047     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4048         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4049             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4050         } else {
4051             /* NOTE: this means there is a window after guest reset, prior to
4052              * CAS negotiation, where unplug requests will fail due to the
4053              * capability not being detected yet. This is a bit different than
4054              * the case with PCI unplug, where the events will be queued and
4055              * eventually handled by the guest after boot
4056              */
4057             error_setg(errp, "Memory hot unplug not supported for this guest");
4058         }
4059     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4060         if (!mc->has_hotpluggable_cpus) {
4061             error_setg(errp, "CPU hot unplug not supported on this machine");
4062             return;
4063         }
4064         spapr_core_unplug_request(hotplug_dev, dev, errp);
4065     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4066         if (!smc->dr_phb_enabled) {
4067             error_setg(errp, "PHB hot unplug not supported on this machine");
4068             return;
4069         }
4070         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4071     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4072         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4073     }
4074 }
4075 
4076 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4077                                           DeviceState *dev, Error **errp)
4078 {
4079     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4080         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4081     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4082         spapr_core_pre_plug(hotplug_dev, dev, errp);
4083     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4084         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4085     }
4086 }
4087 
4088 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4089                                                  DeviceState *dev)
4090 {
4091     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4092         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4093         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4094         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4095         return HOTPLUG_HANDLER(machine);
4096     }
4097     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4098         PCIDevice *pcidev = PCI_DEVICE(dev);
4099         PCIBus *root = pci_device_root_bus(pcidev);
4100         SpaprPhbState *phb =
4101             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4102                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4103 
4104         if (phb) {
4105             return HOTPLUG_HANDLER(phb);
4106         }
4107     }
4108     return NULL;
4109 }
4110 
4111 static CpuInstanceProperties
4112 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4113 {
4114     CPUArchId *core_slot;
4115     MachineClass *mc = MACHINE_GET_CLASS(machine);
4116 
4117     /* make sure possible_cpu are intialized */
4118     mc->possible_cpu_arch_ids(machine);
4119     /* get CPU core slot containing thread that matches cpu_index */
4120     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4121     assert(core_slot);
4122     return core_slot->props;
4123 }
4124 
4125 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4126 {
4127     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4128 }
4129 
4130 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4131 {
4132     int i;
4133     unsigned int smp_threads = machine->smp.threads;
4134     unsigned int smp_cpus = machine->smp.cpus;
4135     const char *core_type;
4136     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4137     MachineClass *mc = MACHINE_GET_CLASS(machine);
4138 
4139     if (!mc->has_hotpluggable_cpus) {
4140         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4141     }
4142     if (machine->possible_cpus) {
4143         assert(machine->possible_cpus->len == spapr_max_cores);
4144         return machine->possible_cpus;
4145     }
4146 
4147     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4148     if (!core_type) {
4149         error_report("Unable to find sPAPR CPU Core definition");
4150         exit(1);
4151     }
4152 
4153     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4154                              sizeof(CPUArchId) * spapr_max_cores);
4155     machine->possible_cpus->len = spapr_max_cores;
4156     for (i = 0; i < machine->possible_cpus->len; i++) {
4157         int core_id = i * smp_threads;
4158 
4159         machine->possible_cpus->cpus[i].type = core_type;
4160         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4161         machine->possible_cpus->cpus[i].arch_id = core_id;
4162         machine->possible_cpus->cpus[i].props.has_core_id = true;
4163         machine->possible_cpus->cpus[i].props.core_id = core_id;
4164     }
4165     return machine->possible_cpus;
4166 }
4167 
4168 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4169                                 uint64_t *buid, hwaddr *pio,
4170                                 hwaddr *mmio32, hwaddr *mmio64,
4171                                 unsigned n_dma, uint32_t *liobns,
4172                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4173 {
4174     /*
4175      * New-style PHB window placement.
4176      *
4177      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4178      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4179      * windows.
4180      *
4181      * Some guest kernels can't work with MMIO windows above 1<<46
4182      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4183      *
4184      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4185      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4186      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4187      * 1TiB 64-bit MMIO windows for each PHB.
4188      */
4189     const uint64_t base_buid = 0x800000020000000ULL;
4190     int i;
4191 
4192     /* Sanity check natural alignments */
4193     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4194     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4195     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4196     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4197     /* Sanity check bounds */
4198     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4199                       SPAPR_PCI_MEM32_WIN_SIZE);
4200     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4201                       SPAPR_PCI_MEM64_WIN_SIZE);
4202 
4203     if (index >= SPAPR_MAX_PHBS) {
4204         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4205                    SPAPR_MAX_PHBS - 1);
4206         return;
4207     }
4208 
4209     *buid = base_buid + index;
4210     for (i = 0; i < n_dma; ++i) {
4211         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4212     }
4213 
4214     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4215     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4216     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4217 
4218     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4219     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4220 }
4221 
4222 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4223 {
4224     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4225 
4226     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4227 }
4228 
4229 static void spapr_ics_resend(XICSFabric *dev)
4230 {
4231     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4232 
4233     ics_resend(spapr->ics);
4234 }
4235 
4236 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4237 {
4238     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4239 
4240     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4241 }
4242 
4243 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4244                                  Monitor *mon)
4245 {
4246     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4247 
4248     spapr_irq_print_info(spapr, mon);
4249     monitor_printf(mon, "irqchip: %s\n",
4250                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4251 }
4252 
4253 /*
4254  * This is a XIVE only operation
4255  */
4256 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4257                            uint8_t nvt_blk, uint32_t nvt_idx,
4258                            bool cam_ignore, uint8_t priority,
4259                            uint32_t logic_serv, XiveTCTXMatch *match)
4260 {
4261     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4262     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4263     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4264     int count;
4265 
4266     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4267                            priority, logic_serv, match);
4268     if (count < 0) {
4269         return count;
4270     }
4271 
4272     /*
4273      * When we implement the save and restore of the thread interrupt
4274      * contexts in the enter/exit CPU handlers of the machine and the
4275      * escalations in QEMU, we should be able to handle non dispatched
4276      * vCPUs.
4277      *
4278      * Until this is done, the sPAPR machine should find at least one
4279      * matching context always.
4280      */
4281     if (count == 0) {
4282         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4283                       nvt_blk, nvt_idx);
4284     }
4285 
4286     return count;
4287 }
4288 
4289 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4290 {
4291     return cpu->vcpu_id;
4292 }
4293 
4294 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4295 {
4296     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4297     MachineState *ms = MACHINE(spapr);
4298     int vcpu_id;
4299 
4300     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4301 
4302     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4303         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4304         error_append_hint(errp, "Adjust the number of cpus to %d "
4305                           "or try to raise the number of threads per core\n",
4306                           vcpu_id * ms->smp.threads / spapr->vsmt);
4307         return false;
4308     }
4309 
4310     cpu->vcpu_id = vcpu_id;
4311     return true;
4312 }
4313 
4314 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4315 {
4316     CPUState *cs;
4317 
4318     CPU_FOREACH(cs) {
4319         PowerPCCPU *cpu = POWERPC_CPU(cs);
4320 
4321         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4322             return cpu;
4323         }
4324     }
4325 
4326     return NULL;
4327 }
4328 
4329 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4330 {
4331     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4332 
4333     /* These are only called by TCG, KVM maintains dispatch state */
4334 
4335     spapr_cpu->prod = false;
4336     if (spapr_cpu->vpa_addr) {
4337         CPUState *cs = CPU(cpu);
4338         uint32_t dispatch;
4339 
4340         dispatch = ldl_be_phys(cs->as,
4341                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4342         dispatch++;
4343         if ((dispatch & 1) != 0) {
4344             qemu_log_mask(LOG_GUEST_ERROR,
4345                           "VPA: incorrect dispatch counter value for "
4346                           "dispatched partition %u, correcting.\n", dispatch);
4347             dispatch++;
4348         }
4349         stl_be_phys(cs->as,
4350                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4351     }
4352 }
4353 
4354 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4355 {
4356     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4357 
4358     if (spapr_cpu->vpa_addr) {
4359         CPUState *cs = CPU(cpu);
4360         uint32_t dispatch;
4361 
4362         dispatch = ldl_be_phys(cs->as,
4363                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4364         dispatch++;
4365         if ((dispatch & 1) != 1) {
4366             qemu_log_mask(LOG_GUEST_ERROR,
4367                           "VPA: incorrect dispatch counter value for "
4368                           "preempted partition %u, correcting.\n", dispatch);
4369             dispatch++;
4370         }
4371         stl_be_phys(cs->as,
4372                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4373     }
4374 }
4375 
4376 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4377 {
4378     MachineClass *mc = MACHINE_CLASS(oc);
4379     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4380     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4381     NMIClass *nc = NMI_CLASS(oc);
4382     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4383     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4384     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4385     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4386     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4387 
4388     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4389     mc->ignore_boot_device_suffixes = true;
4390 
4391     /*
4392      * We set up the default / latest behaviour here.  The class_init
4393      * functions for the specific versioned machine types can override
4394      * these details for backwards compatibility
4395      */
4396     mc->init = spapr_machine_init;
4397     mc->reset = spapr_machine_reset;
4398     mc->block_default_type = IF_SCSI;
4399     mc->max_cpus = 1024;
4400     mc->no_parallel = 1;
4401     mc->default_boot_order = "";
4402     mc->default_ram_size = 512 * MiB;
4403     mc->default_ram_id = "ppc_spapr.ram";
4404     mc->default_display = "std";
4405     mc->kvm_type = spapr_kvm_type;
4406     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4407     mc->pci_allow_0_address = true;
4408     assert(!mc->get_hotplug_handler);
4409     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4410     hc->pre_plug = spapr_machine_device_pre_plug;
4411     hc->plug = spapr_machine_device_plug;
4412     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4413     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4414     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4415     hc->unplug_request = spapr_machine_device_unplug_request;
4416     hc->unplug = spapr_machine_device_unplug;
4417 
4418     smc->dr_lmb_enabled = true;
4419     smc->update_dt_enabled = true;
4420     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4421     mc->has_hotpluggable_cpus = true;
4422     mc->nvdimm_supported = true;
4423     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4424     fwc->get_dev_path = spapr_get_fw_dev_path;
4425     nc->nmi_monitor_handler = spapr_nmi;
4426     smc->phb_placement = spapr_phb_placement;
4427     vhc->hypercall = emulate_spapr_hypercall;
4428     vhc->hpt_mask = spapr_hpt_mask;
4429     vhc->map_hptes = spapr_map_hptes;
4430     vhc->unmap_hptes = spapr_unmap_hptes;
4431     vhc->hpte_set_c = spapr_hpte_set_c;
4432     vhc->hpte_set_r = spapr_hpte_set_r;
4433     vhc->get_pate = spapr_get_pate;
4434     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4435     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4436     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4437     xic->ics_get = spapr_ics_get;
4438     xic->ics_resend = spapr_ics_resend;
4439     xic->icp_get = spapr_icp_get;
4440     ispc->print_info = spapr_pic_print_info;
4441     /* Force NUMA node memory size to be a multiple of
4442      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4443      * in which LMBs are represented and hot-added
4444      */
4445     mc->numa_mem_align_shift = 28;
4446     mc->auto_enable_numa = true;
4447 
4448     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4449     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4450     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4451     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4452     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4453     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4454     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4455     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4456     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4457     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4458     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4459     spapr_caps_add_properties(smc);
4460     smc->irq = &spapr_irq_dual;
4461     smc->dr_phb_enabled = true;
4462     smc->linux_pci_probe = true;
4463     smc->smp_threads_vsmt = true;
4464     smc->nr_xirqs = SPAPR_NR_XIRQS;
4465     xfc->match_nvt = spapr_match_nvt;
4466 }
4467 
4468 static const TypeInfo spapr_machine_info = {
4469     .name          = TYPE_SPAPR_MACHINE,
4470     .parent        = TYPE_MACHINE,
4471     .abstract      = true,
4472     .instance_size = sizeof(SpaprMachineState),
4473     .instance_init = spapr_instance_init,
4474     .instance_finalize = spapr_machine_finalizefn,
4475     .class_size    = sizeof(SpaprMachineClass),
4476     .class_init    = spapr_machine_class_init,
4477     .interfaces = (InterfaceInfo[]) {
4478         { TYPE_FW_PATH_PROVIDER },
4479         { TYPE_NMI },
4480         { TYPE_HOTPLUG_HANDLER },
4481         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4482         { TYPE_XICS_FABRIC },
4483         { TYPE_INTERRUPT_STATS_PROVIDER },
4484         { TYPE_XIVE_FABRIC },
4485         { }
4486     },
4487 };
4488 
4489 static void spapr_machine_latest_class_options(MachineClass *mc)
4490 {
4491     mc->alias = "pseries";
4492     mc->is_default = true;
4493 }
4494 
4495 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4496     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4497                                                     void *data)      \
4498     {                                                                \
4499         MachineClass *mc = MACHINE_CLASS(oc);                        \
4500         spapr_machine_##suffix##_class_options(mc);                  \
4501         if (latest) {                                                \
4502             spapr_machine_latest_class_options(mc);                  \
4503         }                                                            \
4504     }                                                                \
4505     static const TypeInfo spapr_machine_##suffix##_info = {          \
4506         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4507         .parent = TYPE_SPAPR_MACHINE,                                \
4508         .class_init = spapr_machine_##suffix##_class_init,           \
4509     };                                                               \
4510     static void spapr_machine_register_##suffix(void)                \
4511     {                                                                \
4512         type_register(&spapr_machine_##suffix##_info);               \
4513     }                                                                \
4514     type_init(spapr_machine_register_##suffix)
4515 
4516 /*
4517  * pseries-5.2
4518  */
4519 static void spapr_machine_5_2_class_options(MachineClass *mc)
4520 {
4521     /* Defaults for the latest behaviour inherited from the base class */
4522 }
4523 
4524 DEFINE_SPAPR_MACHINE(5_2, "5.2", true);
4525 
4526 /*
4527  * pseries-5.1
4528  */
4529 static void spapr_machine_5_1_class_options(MachineClass *mc)
4530 {
4531     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4532 
4533     spapr_machine_5_2_class_options(mc);
4534     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4535     smc->pre_5_2_numa_associativity = true;
4536 }
4537 
4538 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4539 
4540 /*
4541  * pseries-5.0
4542  */
4543 static void spapr_machine_5_0_class_options(MachineClass *mc)
4544 {
4545     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4546     static GlobalProperty compat[] = {
4547         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4548     };
4549 
4550     spapr_machine_5_1_class_options(mc);
4551     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4552     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4553     mc->numa_mem_supported = true;
4554     smc->pre_5_1_assoc_refpoints = true;
4555 }
4556 
4557 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4558 
4559 /*
4560  * pseries-4.2
4561  */
4562 static void spapr_machine_4_2_class_options(MachineClass *mc)
4563 {
4564     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4565 
4566     spapr_machine_5_0_class_options(mc);
4567     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4568     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4569     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4570     smc->rma_limit = 16 * GiB;
4571     mc->nvdimm_supported = false;
4572 }
4573 
4574 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4575 
4576 /*
4577  * pseries-4.1
4578  */
4579 static void spapr_machine_4_1_class_options(MachineClass *mc)
4580 {
4581     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4582     static GlobalProperty compat[] = {
4583         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4584         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4585     };
4586 
4587     spapr_machine_4_2_class_options(mc);
4588     smc->linux_pci_probe = false;
4589     smc->smp_threads_vsmt = false;
4590     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4591     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4592 }
4593 
4594 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4595 
4596 /*
4597  * pseries-4.0
4598  */
4599 static void phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4600                               uint64_t *buid, hwaddr *pio,
4601                               hwaddr *mmio32, hwaddr *mmio64,
4602                               unsigned n_dma, uint32_t *liobns,
4603                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4604 {
4605     spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, liobns,
4606                         nv2gpa, nv2atsd, errp);
4607     *nv2gpa = 0;
4608     *nv2atsd = 0;
4609 }
4610 
4611 static void spapr_machine_4_0_class_options(MachineClass *mc)
4612 {
4613     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4614 
4615     spapr_machine_4_1_class_options(mc);
4616     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4617     smc->phb_placement = phb_placement_4_0;
4618     smc->irq = &spapr_irq_xics;
4619     smc->pre_4_1_migration = true;
4620 }
4621 
4622 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4623 
4624 /*
4625  * pseries-3.1
4626  */
4627 static void spapr_machine_3_1_class_options(MachineClass *mc)
4628 {
4629     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4630 
4631     spapr_machine_4_0_class_options(mc);
4632     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4633 
4634     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4635     smc->update_dt_enabled = false;
4636     smc->dr_phb_enabled = false;
4637     smc->broken_host_serial_model = true;
4638     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4639     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4640     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4641     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4642 }
4643 
4644 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4645 
4646 /*
4647  * pseries-3.0
4648  */
4649 
4650 static void spapr_machine_3_0_class_options(MachineClass *mc)
4651 {
4652     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4653 
4654     spapr_machine_3_1_class_options(mc);
4655     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4656 
4657     smc->legacy_irq_allocation = true;
4658     smc->nr_xirqs = 0x400;
4659     smc->irq = &spapr_irq_xics_legacy;
4660 }
4661 
4662 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4663 
4664 /*
4665  * pseries-2.12
4666  */
4667 static void spapr_machine_2_12_class_options(MachineClass *mc)
4668 {
4669     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4670     static GlobalProperty compat[] = {
4671         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4672         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4673     };
4674 
4675     spapr_machine_3_0_class_options(mc);
4676     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4677     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4678 
4679     /* We depend on kvm_enabled() to choose a default value for the
4680      * hpt-max-page-size capability. Of course we can't do it here
4681      * because this is too early and the HW accelerator isn't initialzed
4682      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4683      */
4684     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4685 }
4686 
4687 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4688 
4689 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4690 {
4691     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4692 
4693     spapr_machine_2_12_class_options(mc);
4694     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4695     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4696     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4697 }
4698 
4699 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4700 
4701 /*
4702  * pseries-2.11
4703  */
4704 
4705 static void spapr_machine_2_11_class_options(MachineClass *mc)
4706 {
4707     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4708 
4709     spapr_machine_2_12_class_options(mc);
4710     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4711     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4712 }
4713 
4714 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4715 
4716 /*
4717  * pseries-2.10
4718  */
4719 
4720 static void spapr_machine_2_10_class_options(MachineClass *mc)
4721 {
4722     spapr_machine_2_11_class_options(mc);
4723     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4724 }
4725 
4726 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4727 
4728 /*
4729  * pseries-2.9
4730  */
4731 
4732 static void spapr_machine_2_9_class_options(MachineClass *mc)
4733 {
4734     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4735     static GlobalProperty compat[] = {
4736         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4737     };
4738 
4739     spapr_machine_2_10_class_options(mc);
4740     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4741     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4742     smc->pre_2_10_has_unused_icps = true;
4743     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4744 }
4745 
4746 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4747 
4748 /*
4749  * pseries-2.8
4750  */
4751 
4752 static void spapr_machine_2_8_class_options(MachineClass *mc)
4753 {
4754     static GlobalProperty compat[] = {
4755         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4756     };
4757 
4758     spapr_machine_2_9_class_options(mc);
4759     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4760     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4761     mc->numa_mem_align_shift = 23;
4762 }
4763 
4764 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4765 
4766 /*
4767  * pseries-2.7
4768  */
4769 
4770 static void phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4771                               uint64_t *buid, hwaddr *pio,
4772                               hwaddr *mmio32, hwaddr *mmio64,
4773                               unsigned n_dma, uint32_t *liobns,
4774                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4775 {
4776     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4777     const uint64_t base_buid = 0x800000020000000ULL;
4778     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4779     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4780     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4781     const uint32_t max_index = 255;
4782     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4783 
4784     uint64_t ram_top = MACHINE(spapr)->ram_size;
4785     hwaddr phb0_base, phb_base;
4786     int i;
4787 
4788     /* Do we have device memory? */
4789     if (MACHINE(spapr)->maxram_size > ram_top) {
4790         /* Can't just use maxram_size, because there may be an
4791          * alignment gap between normal and device memory regions
4792          */
4793         ram_top = MACHINE(spapr)->device_memory->base +
4794             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4795     }
4796 
4797     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4798 
4799     if (index > max_index) {
4800         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4801                    max_index);
4802         return;
4803     }
4804 
4805     *buid = base_buid + index;
4806     for (i = 0; i < n_dma; ++i) {
4807         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4808     }
4809 
4810     phb_base = phb0_base + index * phb_spacing;
4811     *pio = phb_base + pio_offset;
4812     *mmio32 = phb_base + mmio_offset;
4813     /*
4814      * We don't set the 64-bit MMIO window, relying on the PHB's
4815      * fallback behaviour of automatically splitting a large "32-bit"
4816      * window into contiguous 32-bit and 64-bit windows
4817      */
4818 
4819     *nv2gpa = 0;
4820     *nv2atsd = 0;
4821 }
4822 
4823 static void spapr_machine_2_7_class_options(MachineClass *mc)
4824 {
4825     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4826     static GlobalProperty compat[] = {
4827         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4828         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4829         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4830         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4831     };
4832 
4833     spapr_machine_2_8_class_options(mc);
4834     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4835     mc->default_machine_opts = "modern-hotplug-events=off";
4836     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4837     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4838     smc->phb_placement = phb_placement_2_7;
4839 }
4840 
4841 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4842 
4843 /*
4844  * pseries-2.6
4845  */
4846 
4847 static void spapr_machine_2_6_class_options(MachineClass *mc)
4848 {
4849     static GlobalProperty compat[] = {
4850         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4851     };
4852 
4853     spapr_machine_2_7_class_options(mc);
4854     mc->has_hotpluggable_cpus = false;
4855     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4856     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4857 }
4858 
4859 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4860 
4861 /*
4862  * pseries-2.5
4863  */
4864 
4865 static void spapr_machine_2_5_class_options(MachineClass *mc)
4866 {
4867     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4868     static GlobalProperty compat[] = {
4869         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4870     };
4871 
4872     spapr_machine_2_6_class_options(mc);
4873     smc->use_ohci_by_default = true;
4874     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4875     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4876 }
4877 
4878 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4879 
4880 /*
4881  * pseries-2.4
4882  */
4883 
4884 static void spapr_machine_2_4_class_options(MachineClass *mc)
4885 {
4886     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4887 
4888     spapr_machine_2_5_class_options(mc);
4889     smc->dr_lmb_enabled = false;
4890     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4891 }
4892 
4893 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4894 
4895 /*
4896  * pseries-2.3
4897  */
4898 
4899 static void spapr_machine_2_3_class_options(MachineClass *mc)
4900 {
4901     static GlobalProperty compat[] = {
4902         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4903     };
4904     spapr_machine_2_4_class_options(mc);
4905     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4906     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4907 }
4908 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4909 
4910 /*
4911  * pseries-2.2
4912  */
4913 
4914 static void spapr_machine_2_2_class_options(MachineClass *mc)
4915 {
4916     static GlobalProperty compat[] = {
4917         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4918     };
4919 
4920     spapr_machine_2_3_class_options(mc);
4921     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4922     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4923     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4924 }
4925 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4926 
4927 /*
4928  * pseries-2.1
4929  */
4930 
4931 static void spapr_machine_2_1_class_options(MachineClass *mc)
4932 {
4933     spapr_machine_2_2_class_options(mc);
4934     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4935 }
4936 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4937 
4938 static void spapr_machine_register_types(void)
4939 {
4940     type_register_static(&spapr_machine_info);
4941 }
4942 
4943 type_init(spapr_machine_register_types)
4944