xref: /openbmc/qemu/hw/ppc/spapr.c (revision 19f4ed36)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  */
26 
27 #include "qemu/osdep.h"
28 #include "qemu-common.h"
29 #include "qemu/datadir.h"
30 #include "qapi/error.h"
31 #include "qapi/qapi-events-machine.h"
32 #include "qapi/visitor.h"
33 #include "sysemu/sysemu.h"
34 #include "sysemu/hostmem.h"
35 #include "sysemu/numa.h"
36 #include "sysemu/qtest.h"
37 #include "sysemu/reset.h"
38 #include "sysemu/runstate.h"
39 #include "qemu/log.h"
40 #include "hw/fw-path-provider.h"
41 #include "elf.h"
42 #include "net/net.h"
43 #include "sysemu/device_tree.h"
44 #include "sysemu/cpus.h"
45 #include "sysemu/hw_accel.h"
46 #include "kvm_ppc.h"
47 #include "migration/misc.h"
48 #include "migration/qemu-file-types.h"
49 #include "migration/global_state.h"
50 #include "migration/register.h"
51 #include "migration/blocker.h"
52 #include "mmu-hash64.h"
53 #include "mmu-book3s-v3.h"
54 #include "cpu-models.h"
55 #include "hw/core/cpu.h"
56 
57 #include "hw/boards.h"
58 #include "hw/ppc/ppc.h"
59 #include "hw/loader.h"
60 
61 #include "hw/ppc/fdt.h"
62 #include "hw/ppc/spapr.h"
63 #include "hw/ppc/spapr_vio.h"
64 #include "hw/qdev-properties.h"
65 #include "hw/pci-host/spapr.h"
66 #include "hw/pci/msi.h"
67 
68 #include "hw/pci/pci.h"
69 #include "hw/scsi/scsi.h"
70 #include "hw/virtio/virtio-scsi.h"
71 #include "hw/virtio/vhost-scsi-common.h"
72 
73 #include "exec/address-spaces.h"
74 #include "exec/ram_addr.h"
75 #include "hw/usb.h"
76 #include "qemu/config-file.h"
77 #include "qemu/error-report.h"
78 #include "trace.h"
79 #include "hw/nmi.h"
80 #include "hw/intc/intc.h"
81 
82 #include "hw/ppc/spapr_cpu_core.h"
83 #include "hw/mem/memory-device.h"
84 #include "hw/ppc/spapr_tpm_proxy.h"
85 #include "hw/ppc/spapr_nvdimm.h"
86 #include "hw/ppc/spapr_numa.h"
87 #include "hw/ppc/pef.h"
88 
89 #include "monitor/monitor.h"
90 
91 #include <libfdt.h>
92 
93 /* SLOF memory layout:
94  *
95  * SLOF raw image loaded at 0, copies its romfs right below the flat
96  * device-tree, then position SLOF itself 31M below that
97  *
98  * So we set FW_OVERHEAD to 40MB which should account for all of that
99  * and more
100  *
101  * We load our kernel at 4M, leaving space for SLOF initial image
102  */
103 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
104 #define FW_MAX_SIZE             0x400000
105 #define FW_FILE_NAME            "slof.bin"
106 #define FW_OVERHEAD             0x2800000
107 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
108 
109 #define MIN_RMA_SLOF            (128 * MiB)
110 
111 #define PHANDLE_INTC            0x00001111
112 
113 /* These two functions implement the VCPU id numbering: one to compute them
114  * all and one to identify thread 0 of a VCORE. Any change to the first one
115  * is likely to have an impact on the second one, so let's keep them close.
116  */
117 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
118 {
119     MachineState *ms = MACHINE(spapr);
120     unsigned int smp_threads = ms->smp.threads;
121 
122     assert(spapr->vsmt);
123     return
124         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
125 }
126 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
127                                       PowerPCCPU *cpu)
128 {
129     assert(spapr->vsmt);
130     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
131 }
132 
133 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
134 {
135     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
136      * and newer QEMUs don't even have them. In both cases, we don't want
137      * to send anything on the wire.
138      */
139     return false;
140 }
141 
142 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
143     .name = "icp/server",
144     .version_id = 1,
145     .minimum_version_id = 1,
146     .needed = pre_2_10_vmstate_dummy_icp_needed,
147     .fields = (VMStateField[]) {
148         VMSTATE_UNUSED(4), /* uint32_t xirr */
149         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
150         VMSTATE_UNUSED(1), /* uint8_t mfrr */
151         VMSTATE_END_OF_LIST()
152     },
153 };
154 
155 static void pre_2_10_vmstate_register_dummy_icp(int i)
156 {
157     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
158                      (void *)(uintptr_t) i);
159 }
160 
161 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
162 {
163     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
164                        (void *)(uintptr_t) i);
165 }
166 
167 int spapr_max_server_number(SpaprMachineState *spapr)
168 {
169     MachineState *ms = MACHINE(spapr);
170 
171     assert(spapr->vsmt);
172     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
173 }
174 
175 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
176                                   int smt_threads)
177 {
178     int i, ret = 0;
179     uint32_t servers_prop[smt_threads];
180     uint32_t gservers_prop[smt_threads * 2];
181     int index = spapr_get_vcpu_id(cpu);
182 
183     if (cpu->compat_pvr) {
184         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
185         if (ret < 0) {
186             return ret;
187         }
188     }
189 
190     /* Build interrupt servers and gservers properties */
191     for (i = 0; i < smt_threads; i++) {
192         servers_prop[i] = cpu_to_be32(index + i);
193         /* Hack, direct the group queues back to cpu 0 */
194         gservers_prop[i*2] = cpu_to_be32(index + i);
195         gservers_prop[i*2 + 1] = 0;
196     }
197     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
198                       servers_prop, sizeof(servers_prop));
199     if (ret < 0) {
200         return ret;
201     }
202     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
203                       gservers_prop, sizeof(gservers_prop));
204 
205     return ret;
206 }
207 
208 static void spapr_dt_pa_features(SpaprMachineState *spapr,
209                                  PowerPCCPU *cpu,
210                                  void *fdt, int offset)
211 {
212     uint8_t pa_features_206[] = { 6, 0,
213         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
214     uint8_t pa_features_207[] = { 24, 0,
215         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
216         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
217         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
218         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
219     uint8_t pa_features_300[] = { 66, 0,
220         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
221         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
222         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
223         /* 6: DS207 */
224         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
225         /* 16: Vector */
226         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
227         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
228         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
229         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
230         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
231         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
232         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
233         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
234         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
235         /* 42: PM, 44: PC RA, 46: SC vec'd */
236         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
237         /* 48: SIMD, 50: QP BFP, 52: String */
238         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
239         /* 54: DecFP, 56: DecI, 58: SHA */
240         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
241         /* 60: NM atomic, 62: RNG */
242         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
243     };
244     uint8_t *pa_features = NULL;
245     size_t pa_size;
246 
247     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
248         pa_features = pa_features_206;
249         pa_size = sizeof(pa_features_206);
250     }
251     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
252         pa_features = pa_features_207;
253         pa_size = sizeof(pa_features_207);
254     }
255     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
256         pa_features = pa_features_300;
257         pa_size = sizeof(pa_features_300);
258     }
259     if (!pa_features) {
260         return;
261     }
262 
263     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
264         /*
265          * Note: we keep CI large pages off by default because a 64K capable
266          * guest provisioned with large pages might otherwise try to map a qemu
267          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
268          * even if that qemu runs on a 4k host.
269          * We dd this bit back here if we are confident this is not an issue
270          */
271         pa_features[3] |= 0x20;
272     }
273     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
274         pa_features[24] |= 0x80;    /* Transactional memory support */
275     }
276     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
277         /* Workaround for broken kernels that attempt (guest) radix
278          * mode when they can't handle it, if they see the radix bit set
279          * in pa-features. So hide it from them. */
280         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
281     }
282 
283     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
284 }
285 
286 static hwaddr spapr_node0_size(MachineState *machine)
287 {
288     if (machine->numa_state->num_nodes) {
289         int i;
290         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
291             if (machine->numa_state->nodes[i].node_mem) {
292                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
293                            machine->ram_size);
294             }
295         }
296     }
297     return machine->ram_size;
298 }
299 
300 static void add_str(GString *s, const gchar *s1)
301 {
302     g_string_append_len(s, s1, strlen(s1) + 1);
303 }
304 
305 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid,
306                                 hwaddr start, hwaddr size)
307 {
308     char mem_name[32];
309     uint64_t mem_reg_property[2];
310     int off;
311 
312     mem_reg_property[0] = cpu_to_be64(start);
313     mem_reg_property[1] = cpu_to_be64(size);
314 
315     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
316     off = fdt_add_subnode(fdt, 0, mem_name);
317     _FDT(off);
318     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
319     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
320                       sizeof(mem_reg_property))));
321     spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid);
322     return off;
323 }
324 
325 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
326 {
327     MemoryDeviceInfoList *info;
328 
329     for (info = list; info; info = info->next) {
330         MemoryDeviceInfo *value = info->value;
331 
332         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
333             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
334 
335             if (addr >= pcdimm_info->addr &&
336                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
337                 return pcdimm_info->node;
338             }
339         }
340     }
341 
342     return -1;
343 }
344 
345 struct sPAPRDrconfCellV2 {
346      uint32_t seq_lmbs;
347      uint64_t base_addr;
348      uint32_t drc_index;
349      uint32_t aa_index;
350      uint32_t flags;
351 } QEMU_PACKED;
352 
353 typedef struct DrconfCellQueue {
354     struct sPAPRDrconfCellV2 cell;
355     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
356 } DrconfCellQueue;
357 
358 static DrconfCellQueue *
359 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
360                       uint32_t drc_index, uint32_t aa_index,
361                       uint32_t flags)
362 {
363     DrconfCellQueue *elem;
364 
365     elem = g_malloc0(sizeof(*elem));
366     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
367     elem->cell.base_addr = cpu_to_be64(base_addr);
368     elem->cell.drc_index = cpu_to_be32(drc_index);
369     elem->cell.aa_index = cpu_to_be32(aa_index);
370     elem->cell.flags = cpu_to_be32(flags);
371 
372     return elem;
373 }
374 
375 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
376                                       int offset, MemoryDeviceInfoList *dimms)
377 {
378     MachineState *machine = MACHINE(spapr);
379     uint8_t *int_buf, *cur_index;
380     int ret;
381     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
382     uint64_t addr, cur_addr, size;
383     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
384     uint64_t mem_end = machine->device_memory->base +
385                        memory_region_size(&machine->device_memory->mr);
386     uint32_t node, buf_len, nr_entries = 0;
387     SpaprDrc *drc;
388     DrconfCellQueue *elem, *next;
389     MemoryDeviceInfoList *info;
390     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
391         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
392 
393     /* Entry to cover RAM and the gap area */
394     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
395                                  SPAPR_LMB_FLAGS_RESERVED |
396                                  SPAPR_LMB_FLAGS_DRC_INVALID);
397     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
398     nr_entries++;
399 
400     cur_addr = machine->device_memory->base;
401     for (info = dimms; info; info = info->next) {
402         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
403 
404         addr = di->addr;
405         size = di->size;
406         node = di->node;
407 
408         /*
409          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
410          * area is marked hotpluggable in the next iteration for the bigger
411          * chunk including the NVDIMM occupied area.
412          */
413         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
414             continue;
415 
416         /* Entry for hot-pluggable area */
417         if (cur_addr < addr) {
418             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
419             g_assert(drc);
420             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
421                                          cur_addr, spapr_drc_index(drc), -1, 0);
422             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
423             nr_entries++;
424         }
425 
426         /* Entry for DIMM */
427         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
428         g_assert(drc);
429         elem = spapr_get_drconf_cell(size / lmb_size, addr,
430                                      spapr_drc_index(drc), node,
431                                      (SPAPR_LMB_FLAGS_ASSIGNED |
432                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
433         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
434         nr_entries++;
435         cur_addr = addr + size;
436     }
437 
438     /* Entry for remaining hotpluggable area */
439     if (cur_addr < mem_end) {
440         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
441         g_assert(drc);
442         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
443                                      cur_addr, spapr_drc_index(drc), -1, 0);
444         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
445         nr_entries++;
446     }
447 
448     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
449     int_buf = cur_index = g_malloc0(buf_len);
450     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
451     cur_index += sizeof(nr_entries);
452 
453     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
454         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
455         cur_index += sizeof(elem->cell);
456         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
457         g_free(elem);
458     }
459 
460     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
461     g_free(int_buf);
462     if (ret < 0) {
463         return -1;
464     }
465     return 0;
466 }
467 
468 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
469                                    int offset, MemoryDeviceInfoList *dimms)
470 {
471     MachineState *machine = MACHINE(spapr);
472     int i, ret;
473     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
474     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
475     uint32_t nr_lmbs = (machine->device_memory->base +
476                        memory_region_size(&machine->device_memory->mr)) /
477                        lmb_size;
478     uint32_t *int_buf, *cur_index, buf_len;
479 
480     /*
481      * Allocate enough buffer size to fit in ibm,dynamic-memory
482      */
483     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
484     cur_index = int_buf = g_malloc0(buf_len);
485     int_buf[0] = cpu_to_be32(nr_lmbs);
486     cur_index++;
487     for (i = 0; i < nr_lmbs; i++) {
488         uint64_t addr = i * lmb_size;
489         uint32_t *dynamic_memory = cur_index;
490 
491         if (i >= device_lmb_start) {
492             SpaprDrc *drc;
493 
494             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
495             g_assert(drc);
496 
497             dynamic_memory[0] = cpu_to_be32(addr >> 32);
498             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
499             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
500             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
501             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
502             if (memory_region_present(get_system_memory(), addr)) {
503                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
504             } else {
505                 dynamic_memory[5] = cpu_to_be32(0);
506             }
507         } else {
508             /*
509              * LMB information for RMA, boot time RAM and gap b/n RAM and
510              * device memory region -- all these are marked as reserved
511              * and as having no valid DRC.
512              */
513             dynamic_memory[0] = cpu_to_be32(addr >> 32);
514             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
515             dynamic_memory[2] = cpu_to_be32(0);
516             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
517             dynamic_memory[4] = cpu_to_be32(-1);
518             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
519                                             SPAPR_LMB_FLAGS_DRC_INVALID);
520         }
521 
522         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
523     }
524     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
525     g_free(int_buf);
526     if (ret < 0) {
527         return -1;
528     }
529     return 0;
530 }
531 
532 /*
533  * Adds ibm,dynamic-reconfiguration-memory node.
534  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
535  * of this device tree node.
536  */
537 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
538                                                    void *fdt)
539 {
540     MachineState *machine = MACHINE(spapr);
541     int ret, offset;
542     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
543     uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
544                                 cpu_to_be32(lmb_size & 0xffffffff)};
545     MemoryDeviceInfoList *dimms = NULL;
546 
547     /*
548      * Don't create the node if there is no device memory
549      */
550     if (machine->ram_size == machine->maxram_size) {
551         return 0;
552     }
553 
554     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
555 
556     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
557                     sizeof(prop_lmb_size));
558     if (ret < 0) {
559         return ret;
560     }
561 
562     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
563     if (ret < 0) {
564         return ret;
565     }
566 
567     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
568     if (ret < 0) {
569         return ret;
570     }
571 
572     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
573     dimms = qmp_memory_device_list();
574     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
575         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
576     } else {
577         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
578     }
579     qapi_free_MemoryDeviceInfoList(dimms);
580 
581     if (ret < 0) {
582         return ret;
583     }
584 
585     ret = spapr_numa_write_assoc_lookup_arrays(spapr, fdt, offset);
586 
587     return ret;
588 }
589 
590 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
591 {
592     MachineState *machine = MACHINE(spapr);
593     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
594     hwaddr mem_start, node_size;
595     int i, nb_nodes = machine->numa_state->num_nodes;
596     NodeInfo *nodes = machine->numa_state->nodes;
597 
598     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
599         if (!nodes[i].node_mem) {
600             continue;
601         }
602         if (mem_start >= machine->ram_size) {
603             node_size = 0;
604         } else {
605             node_size = nodes[i].node_mem;
606             if (node_size > machine->ram_size - mem_start) {
607                 node_size = machine->ram_size - mem_start;
608             }
609         }
610         if (!mem_start) {
611             /* spapr_machine_init() checks for rma_size <= node0_size
612              * already */
613             spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size);
614             mem_start += spapr->rma_size;
615             node_size -= spapr->rma_size;
616         }
617         for ( ; node_size; ) {
618             hwaddr sizetmp = pow2floor(node_size);
619 
620             /* mem_start != 0 here */
621             if (ctzl(mem_start) < ctzl(sizetmp)) {
622                 sizetmp = 1ULL << ctzl(mem_start);
623             }
624 
625             spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp);
626             node_size -= sizetmp;
627             mem_start += sizetmp;
628         }
629     }
630 
631     /* Generate ibm,dynamic-reconfiguration-memory node if required */
632     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
633         int ret;
634 
635         g_assert(smc->dr_lmb_enabled);
636         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
637         if (ret) {
638             return ret;
639         }
640     }
641 
642     return 0;
643 }
644 
645 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
646                          SpaprMachineState *spapr)
647 {
648     MachineState *ms = MACHINE(spapr);
649     PowerPCCPU *cpu = POWERPC_CPU(cs);
650     CPUPPCState *env = &cpu->env;
651     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
652     int index = spapr_get_vcpu_id(cpu);
653     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
654                        0xffffffff, 0xffffffff};
655     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
656         : SPAPR_TIMEBASE_FREQ;
657     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
658     uint32_t page_sizes_prop[64];
659     size_t page_sizes_prop_size;
660     unsigned int smp_threads = ms->smp.threads;
661     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
662     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
663     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
664     SpaprDrc *drc;
665     int drc_index;
666     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
667     int i;
668 
669     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
670     if (drc) {
671         drc_index = spapr_drc_index(drc);
672         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
673     }
674 
675     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
676     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
677 
678     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
679     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
680                            env->dcache_line_size)));
681     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
682                            env->dcache_line_size)));
683     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
684                            env->icache_line_size)));
685     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
686                            env->icache_line_size)));
687 
688     if (pcc->l1_dcache_size) {
689         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
690                                pcc->l1_dcache_size)));
691     } else {
692         warn_report("Unknown L1 dcache size for cpu");
693     }
694     if (pcc->l1_icache_size) {
695         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
696                                pcc->l1_icache_size)));
697     } else {
698         warn_report("Unknown L1 icache size for cpu");
699     }
700 
701     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
702     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
703     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
704     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
705     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
706     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
707 
708     if (env->spr_cb[SPR_PURR].oea_read) {
709         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
710     }
711     if (env->spr_cb[SPR_SPURR].oea_read) {
712         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
713     }
714 
715     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
716         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
717                           segs, sizeof(segs))));
718     }
719 
720     /* Advertise VSX (vector extensions) if available
721      *   1               == VMX / Altivec available
722      *   2               == VSX available
723      *
724      * Only CPUs for which we create core types in spapr_cpu_core.c
725      * are possible, and all of those have VMX */
726     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
727         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
728     } else {
729         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
730     }
731 
732     /* Advertise DFP (Decimal Floating Point) if available
733      *   0 / no property == no DFP
734      *   1               == DFP available */
735     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
736         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
737     }
738 
739     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
740                                                       sizeof(page_sizes_prop));
741     if (page_sizes_prop_size) {
742         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
743                           page_sizes_prop, page_sizes_prop_size)));
744     }
745 
746     spapr_dt_pa_features(spapr, cpu, fdt, offset);
747 
748     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
749                            cs->cpu_index / vcpus_per_socket)));
750 
751     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
752                       pft_size_prop, sizeof(pft_size_prop))));
753 
754     if (ms->numa_state->num_nodes > 1) {
755         _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
756     }
757 
758     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
759 
760     if (pcc->radix_page_info) {
761         for (i = 0; i < pcc->radix_page_info->count; i++) {
762             radix_AP_encodings[i] =
763                 cpu_to_be32(pcc->radix_page_info->entries[i]);
764         }
765         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
766                           radix_AP_encodings,
767                           pcc->radix_page_info->count *
768                           sizeof(radix_AP_encodings[0]))));
769     }
770 
771     /*
772      * We set this property to let the guest know that it can use the large
773      * decrementer and its width in bits.
774      */
775     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
776         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
777                               pcc->lrg_decr_bits)));
778 }
779 
780 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
781 {
782     CPUState **rev;
783     CPUState *cs;
784     int n_cpus;
785     int cpus_offset;
786     int i;
787 
788     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
789     _FDT(cpus_offset);
790     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
791     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
792 
793     /*
794      * We walk the CPUs in reverse order to ensure that CPU DT nodes
795      * created by fdt_add_subnode() end up in the right order in FDT
796      * for the guest kernel the enumerate the CPUs correctly.
797      *
798      * The CPU list cannot be traversed in reverse order, so we need
799      * to do extra work.
800      */
801     n_cpus = 0;
802     rev = NULL;
803     CPU_FOREACH(cs) {
804         rev = g_renew(CPUState *, rev, n_cpus + 1);
805         rev[n_cpus++] = cs;
806     }
807 
808     for (i = n_cpus - 1; i >= 0; i--) {
809         CPUState *cs = rev[i];
810         PowerPCCPU *cpu = POWERPC_CPU(cs);
811         int index = spapr_get_vcpu_id(cpu);
812         DeviceClass *dc = DEVICE_GET_CLASS(cs);
813         g_autofree char *nodename = NULL;
814         int offset;
815 
816         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
817             continue;
818         }
819 
820         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
821         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
822         _FDT(offset);
823         spapr_dt_cpu(cs, fdt, offset, spapr);
824     }
825 
826     g_free(rev);
827 }
828 
829 static int spapr_dt_rng(void *fdt)
830 {
831     int node;
832     int ret;
833 
834     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
835     if (node <= 0) {
836         return -1;
837     }
838     ret = fdt_setprop_string(fdt, node, "device_type",
839                              "ibm,platform-facilities");
840     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
841     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
842 
843     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
844     if (node <= 0) {
845         return -1;
846     }
847     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
848 
849     return ret ? -1 : 0;
850 }
851 
852 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
853 {
854     MachineState *ms = MACHINE(spapr);
855     int rtas;
856     GString *hypertas = g_string_sized_new(256);
857     GString *qemu_hypertas = g_string_sized_new(256);
858     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
859         memory_region_size(&MACHINE(spapr)->device_memory->mr);
860     uint32_t lrdr_capacity[] = {
861         cpu_to_be32(max_device_addr >> 32),
862         cpu_to_be32(max_device_addr & 0xffffffff),
863         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
864         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
865         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
866     };
867 
868     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
869 
870     /* hypertas */
871     add_str(hypertas, "hcall-pft");
872     add_str(hypertas, "hcall-term");
873     add_str(hypertas, "hcall-dabr");
874     add_str(hypertas, "hcall-interrupt");
875     add_str(hypertas, "hcall-tce");
876     add_str(hypertas, "hcall-vio");
877     add_str(hypertas, "hcall-splpar");
878     add_str(hypertas, "hcall-join");
879     add_str(hypertas, "hcall-bulk");
880     add_str(hypertas, "hcall-set-mode");
881     add_str(hypertas, "hcall-sprg0");
882     add_str(hypertas, "hcall-copy");
883     add_str(hypertas, "hcall-debug");
884     add_str(hypertas, "hcall-vphn");
885     add_str(qemu_hypertas, "hcall-memop1");
886 
887     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
888         add_str(hypertas, "hcall-multi-tce");
889     }
890 
891     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
892         add_str(hypertas, "hcall-hpt-resize");
893     }
894 
895     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
896                      hypertas->str, hypertas->len));
897     g_string_free(hypertas, TRUE);
898     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
899                      qemu_hypertas->str, qemu_hypertas->len));
900     g_string_free(qemu_hypertas, TRUE);
901 
902     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
903 
904     /*
905      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
906      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
907      *
908      * The system reset requirements are driven by existing Linux and PowerVM
909      * implementation which (contrary to PAPR) saves r3 in the error log
910      * structure like machine check, so Linux expects to find the saved r3
911      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
912      * does not look at the error value).
913      *
914      * System reset interrupts are not subject to interlock like machine
915      * check, so this memory area could be corrupted if the sreset is
916      * interrupted by a machine check (or vice versa) if it was shared. To
917      * prevent this, system reset uses per-CPU areas for the sreset save
918      * area. A system reset that interrupts a system reset handler could
919      * still overwrite this area, but Linux doesn't try to recover in that
920      * case anyway.
921      *
922      * The extra 8 bytes is required because Linux's FWNMI error log check
923      * is off-by-one.
924      */
925     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_ERROR_LOG_MAX +
926 			  ms->smp.max_cpus * sizeof(uint64_t)*2 + sizeof(uint64_t)));
927     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
928                           RTAS_ERROR_LOG_MAX));
929     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
930                           RTAS_EVENT_SCAN_RATE));
931 
932     g_assert(msi_nonbroken);
933     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
934 
935     /*
936      * According to PAPR, rtas ibm,os-term does not guarantee a return
937      * back to the guest cpu.
938      *
939      * While an additional ibm,extended-os-term property indicates
940      * that rtas call return will always occur. Set this property.
941      */
942     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
943 
944     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
945                      lrdr_capacity, sizeof(lrdr_capacity)));
946 
947     spapr_dt_rtas_tokens(fdt, rtas);
948 }
949 
950 /*
951  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
952  * and the XIVE features that the guest may request and thus the valid
953  * values for bytes 23..26 of option vector 5:
954  */
955 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
956                                           int chosen)
957 {
958     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
959 
960     char val[2 * 4] = {
961         23, 0x00, /* XICS / XIVE mode */
962         24, 0x00, /* Hash/Radix, filled in below. */
963         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
964         26, 0x40, /* Radix options: GTSE == yes. */
965     };
966 
967     if (spapr->irq->xics && spapr->irq->xive) {
968         val[1] = SPAPR_OV5_XIVE_BOTH;
969     } else if (spapr->irq->xive) {
970         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
971     } else {
972         assert(spapr->irq->xics);
973         val[1] = SPAPR_OV5_XIVE_LEGACY;
974     }
975 
976     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
977                           first_ppc_cpu->compat_pvr)) {
978         /*
979          * If we're in a pre POWER9 compat mode then the guest should
980          * do hash and use the legacy interrupt mode
981          */
982         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
983         val[3] = 0x00; /* Hash */
984     } else if (kvm_enabled()) {
985         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
986             val[3] = 0x80; /* OV5_MMU_BOTH */
987         } else if (kvmppc_has_cap_mmu_radix()) {
988             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
989         } else {
990             val[3] = 0x00; /* Hash */
991         }
992     } else {
993         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
994         val[3] = 0xC0;
995     }
996     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
997                      val, sizeof(val)));
998 }
999 
1000 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1001 {
1002     MachineState *machine = MACHINE(spapr);
1003     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1004     int chosen;
1005 
1006     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1007 
1008     if (reset) {
1009         const char *boot_device = machine->boot_order;
1010         char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1011         size_t cb = 0;
1012         char *bootlist = get_boot_devices_list(&cb);
1013 
1014         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1015             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1016                                     machine->kernel_cmdline));
1017         }
1018 
1019         if (spapr->initrd_size) {
1020             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1021                                   spapr->initrd_base));
1022             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1023                                   spapr->initrd_base + spapr->initrd_size));
1024         }
1025 
1026         if (spapr->kernel_size) {
1027             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1028                                   cpu_to_be64(spapr->kernel_size) };
1029 
1030             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1031                          &kprop, sizeof(kprop)));
1032             if (spapr->kernel_le) {
1033                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1034             }
1035         }
1036         if (boot_menu) {
1037             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1038         }
1039         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1040         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1041         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1042 
1043         if (cb && bootlist) {
1044             int i;
1045 
1046             for (i = 0; i < cb; i++) {
1047                 if (bootlist[i] == '\n') {
1048                     bootlist[i] = ' ';
1049                 }
1050             }
1051             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1052         }
1053 
1054         if (boot_device && strlen(boot_device)) {
1055             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1056         }
1057 
1058         if (!spapr->has_graphics && stdout_path) {
1059             /*
1060              * "linux,stdout-path" and "stdout" properties are
1061              * deprecated by linux kernel. New platforms should only
1062              * use the "stdout-path" property. Set the new property
1063              * and continue using older property to remain compatible
1064              * with the existing firmware.
1065              */
1066             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1067             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1068         }
1069 
1070         /*
1071          * We can deal with BAR reallocation just fine, advertise it
1072          * to the guest
1073          */
1074         if (smc->linux_pci_probe) {
1075             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1076         }
1077 
1078         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1079 
1080         g_free(stdout_path);
1081         g_free(bootlist);
1082     }
1083 
1084     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1085 }
1086 
1087 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1088 {
1089     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1090      * KVM to work under pHyp with some guest co-operation */
1091     int hypervisor;
1092     uint8_t hypercall[16];
1093 
1094     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1095     /* indicate KVM hypercall interface */
1096     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1097     if (kvmppc_has_cap_fixup_hcalls()) {
1098         /*
1099          * Older KVM versions with older guest kernels were broken
1100          * with the magic page, don't allow the guest to map it.
1101          */
1102         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1103                                   sizeof(hypercall))) {
1104             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1105                              hypercall, sizeof(hypercall)));
1106         }
1107     }
1108 }
1109 
1110 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1111 {
1112     MachineState *machine = MACHINE(spapr);
1113     MachineClass *mc = MACHINE_GET_CLASS(machine);
1114     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1115     uint32_t root_drc_type_mask = 0;
1116     int ret;
1117     void *fdt;
1118     SpaprPhbState *phb;
1119     char *buf;
1120 
1121     fdt = g_malloc0(space);
1122     _FDT((fdt_create_empty_tree(fdt, space)));
1123 
1124     /* Root node */
1125     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1126     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1127     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1128 
1129     /* Guest UUID & Name*/
1130     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1131     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1132     if (qemu_uuid_set) {
1133         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1134     }
1135     g_free(buf);
1136 
1137     if (qemu_get_vm_name()) {
1138         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1139                                 qemu_get_vm_name()));
1140     }
1141 
1142     /* Host Model & Serial Number */
1143     if (spapr->host_model) {
1144         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1145     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1146         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1147         g_free(buf);
1148     }
1149 
1150     if (spapr->host_serial) {
1151         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1152     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1153         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1154         g_free(buf);
1155     }
1156 
1157     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1158     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1159 
1160     /* /interrupt controller */
1161     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1162 
1163     ret = spapr_dt_memory(spapr, fdt);
1164     if (ret < 0) {
1165         error_report("couldn't setup memory nodes in fdt");
1166         exit(1);
1167     }
1168 
1169     /* /vdevice */
1170     spapr_dt_vdevice(spapr->vio_bus, fdt);
1171 
1172     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1173         ret = spapr_dt_rng(fdt);
1174         if (ret < 0) {
1175             error_report("could not set up rng device in the fdt");
1176             exit(1);
1177         }
1178     }
1179 
1180     QLIST_FOREACH(phb, &spapr->phbs, list) {
1181         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1182         if (ret < 0) {
1183             error_report("couldn't setup PCI devices in fdt");
1184             exit(1);
1185         }
1186     }
1187 
1188     spapr_dt_cpus(fdt, spapr);
1189 
1190     /* ibm,drc-indexes and friends */
1191     if (smc->dr_lmb_enabled) {
1192         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_LMB;
1193     }
1194     if (smc->dr_phb_enabled) {
1195         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PHB;
1196     }
1197     if (mc->nvdimm_supported) {
1198         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PMEM;
1199     }
1200     if (root_drc_type_mask) {
1201         _FDT(spapr_dt_drc(fdt, 0, NULL, root_drc_type_mask));
1202     }
1203 
1204     if (mc->has_hotpluggable_cpus) {
1205         int offset = fdt_path_offset(fdt, "/cpus");
1206         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1207         if (ret < 0) {
1208             error_report("Couldn't set up CPU DR device tree properties");
1209             exit(1);
1210         }
1211     }
1212 
1213     /* /event-sources */
1214     spapr_dt_events(spapr, fdt);
1215 
1216     /* /rtas */
1217     spapr_dt_rtas(spapr, fdt);
1218 
1219     /* /chosen */
1220     spapr_dt_chosen(spapr, fdt, reset);
1221 
1222     /* /hypervisor */
1223     if (kvm_enabled()) {
1224         spapr_dt_hypervisor(spapr, fdt);
1225     }
1226 
1227     /* Build memory reserve map */
1228     if (reset) {
1229         if (spapr->kernel_size) {
1230             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1231                                   spapr->kernel_size)));
1232         }
1233         if (spapr->initrd_size) {
1234             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1235                                   spapr->initrd_size)));
1236         }
1237     }
1238 
1239     /* NVDIMM devices */
1240     if (mc->nvdimm_supported) {
1241         spapr_dt_persistent_memory(spapr, fdt);
1242     }
1243 
1244     return fdt;
1245 }
1246 
1247 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1248 {
1249     SpaprMachineState *spapr = opaque;
1250 
1251     return (addr & 0x0fffffff) + spapr->kernel_addr;
1252 }
1253 
1254 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1255                                     PowerPCCPU *cpu)
1256 {
1257     CPUPPCState *env = &cpu->env;
1258 
1259     /* The TCG path should also be holding the BQL at this point */
1260     g_assert(qemu_mutex_iothread_locked());
1261 
1262     if (msr_pr) {
1263         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1264         env->gpr[3] = H_PRIVILEGE;
1265     } else {
1266         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1267     }
1268 }
1269 
1270 struct LPCRSyncState {
1271     target_ulong value;
1272     target_ulong mask;
1273 };
1274 
1275 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1276 {
1277     struct LPCRSyncState *s = arg.host_ptr;
1278     PowerPCCPU *cpu = POWERPC_CPU(cs);
1279     CPUPPCState *env = &cpu->env;
1280     target_ulong lpcr;
1281 
1282     cpu_synchronize_state(cs);
1283     lpcr = env->spr[SPR_LPCR];
1284     lpcr &= ~s->mask;
1285     lpcr |= s->value;
1286     ppc_store_lpcr(cpu, lpcr);
1287 }
1288 
1289 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1290 {
1291     CPUState *cs;
1292     struct LPCRSyncState s = {
1293         .value = value,
1294         .mask = mask
1295     };
1296     CPU_FOREACH(cs) {
1297         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1298     }
1299 }
1300 
1301 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1302 {
1303     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1304 
1305     /* Copy PATE1:GR into PATE0:HR */
1306     entry->dw0 = spapr->patb_entry & PATE0_HR;
1307     entry->dw1 = spapr->patb_entry;
1308 }
1309 
1310 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1311 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1312 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1313 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1314 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1315 
1316 /*
1317  * Get the fd to access the kernel htab, re-opening it if necessary
1318  */
1319 static int get_htab_fd(SpaprMachineState *spapr)
1320 {
1321     Error *local_err = NULL;
1322 
1323     if (spapr->htab_fd >= 0) {
1324         return spapr->htab_fd;
1325     }
1326 
1327     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1328     if (spapr->htab_fd < 0) {
1329         error_report_err(local_err);
1330     }
1331 
1332     return spapr->htab_fd;
1333 }
1334 
1335 void close_htab_fd(SpaprMachineState *spapr)
1336 {
1337     if (spapr->htab_fd >= 0) {
1338         close(spapr->htab_fd);
1339     }
1340     spapr->htab_fd = -1;
1341 }
1342 
1343 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1344 {
1345     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1346 
1347     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1348 }
1349 
1350 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1351 {
1352     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1353 
1354     assert(kvm_enabled());
1355 
1356     if (!spapr->htab) {
1357         return 0;
1358     }
1359 
1360     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1361 }
1362 
1363 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1364                                                 hwaddr ptex, int n)
1365 {
1366     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1367     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1368 
1369     if (!spapr->htab) {
1370         /*
1371          * HTAB is controlled by KVM. Fetch into temporary buffer
1372          */
1373         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1374         kvmppc_read_hptes(hptes, ptex, n);
1375         return hptes;
1376     }
1377 
1378     /*
1379      * HTAB is controlled by QEMU. Just point to the internally
1380      * accessible PTEG.
1381      */
1382     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1383 }
1384 
1385 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1386                               const ppc_hash_pte64_t *hptes,
1387                               hwaddr ptex, int n)
1388 {
1389     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1390 
1391     if (!spapr->htab) {
1392         g_free((void *)hptes);
1393     }
1394 
1395     /* Nothing to do for qemu managed HPT */
1396 }
1397 
1398 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1399                       uint64_t pte0, uint64_t pte1)
1400 {
1401     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1402     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1403 
1404     if (!spapr->htab) {
1405         kvmppc_write_hpte(ptex, pte0, pte1);
1406     } else {
1407         if (pte0 & HPTE64_V_VALID) {
1408             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1409             /*
1410              * When setting valid, we write PTE1 first. This ensures
1411              * proper synchronization with the reading code in
1412              * ppc_hash64_pteg_search()
1413              */
1414             smp_wmb();
1415             stq_p(spapr->htab + offset, pte0);
1416         } else {
1417             stq_p(spapr->htab + offset, pte0);
1418             /*
1419              * When clearing it we set PTE0 first. This ensures proper
1420              * synchronization with the reading code in
1421              * ppc_hash64_pteg_search()
1422              */
1423             smp_wmb();
1424             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1425         }
1426     }
1427 }
1428 
1429 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1430                              uint64_t pte1)
1431 {
1432     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1433     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1434 
1435     if (!spapr->htab) {
1436         /* There should always be a hash table when this is called */
1437         error_report("spapr_hpte_set_c called with no hash table !");
1438         return;
1439     }
1440 
1441     /* The HW performs a non-atomic byte update */
1442     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1443 }
1444 
1445 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1446                              uint64_t pte1)
1447 {
1448     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1449     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1450 
1451     if (!spapr->htab) {
1452         /* There should always be a hash table when this is called */
1453         error_report("spapr_hpte_set_r called with no hash table !");
1454         return;
1455     }
1456 
1457     /* The HW performs a non-atomic byte update */
1458     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1459 }
1460 
1461 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1462 {
1463     int shift;
1464 
1465     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1466      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1467      * that's much more than is needed for Linux guests */
1468     shift = ctz64(pow2ceil(ramsize)) - 7;
1469     shift = MAX(shift, 18); /* Minimum architected size */
1470     shift = MIN(shift, 46); /* Maximum architected size */
1471     return shift;
1472 }
1473 
1474 void spapr_free_hpt(SpaprMachineState *spapr)
1475 {
1476     g_free(spapr->htab);
1477     spapr->htab = NULL;
1478     spapr->htab_shift = 0;
1479     close_htab_fd(spapr);
1480 }
1481 
1482 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1483 {
1484     ERRP_GUARD();
1485     long rc;
1486 
1487     /* Clean up any HPT info from a previous boot */
1488     spapr_free_hpt(spapr);
1489 
1490     rc = kvmppc_reset_htab(shift);
1491 
1492     if (rc == -EOPNOTSUPP) {
1493         error_setg(errp, "HPT not supported in nested guests");
1494         return -EOPNOTSUPP;
1495     }
1496 
1497     if (rc < 0) {
1498         /* kernel-side HPT needed, but couldn't allocate one */
1499         error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1500                          shift);
1501         error_append_hint(errp, "Try smaller maxmem?\n");
1502         return -errno;
1503     } else if (rc > 0) {
1504         /* kernel-side HPT allocated */
1505         if (rc != shift) {
1506             error_setg(errp,
1507                        "Requested order %d HPT, but kernel allocated order %ld",
1508                        shift, rc);
1509             error_append_hint(errp, "Try smaller maxmem?\n");
1510             return -ENOSPC;
1511         }
1512 
1513         spapr->htab_shift = shift;
1514         spapr->htab = NULL;
1515     } else {
1516         /* kernel-side HPT not needed, allocate in userspace instead */
1517         size_t size = 1ULL << shift;
1518         int i;
1519 
1520         spapr->htab = qemu_memalign(size, size);
1521         memset(spapr->htab, 0, size);
1522         spapr->htab_shift = shift;
1523 
1524         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1525             DIRTY_HPTE(HPTE(spapr->htab, i));
1526         }
1527     }
1528     /* We're setting up a hash table, so that means we're not radix */
1529     spapr->patb_entry = 0;
1530     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1531     return 0;
1532 }
1533 
1534 void spapr_setup_hpt(SpaprMachineState *spapr)
1535 {
1536     int hpt_shift;
1537 
1538     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1539         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1540     } else {
1541         uint64_t current_ram_size;
1542 
1543         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1544         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1545     }
1546     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1547 
1548     if (kvm_enabled()) {
1549         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1550 
1551         /* Check our RMA fits in the possible VRMA */
1552         if (vrma_limit < spapr->rma_size) {
1553             error_report("Unable to create %" HWADDR_PRIu
1554                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1555                          spapr->rma_size / MiB, vrma_limit / MiB);
1556             exit(EXIT_FAILURE);
1557         }
1558     }
1559 }
1560 
1561 static void spapr_machine_reset(MachineState *machine)
1562 {
1563     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1564     PowerPCCPU *first_ppc_cpu;
1565     hwaddr fdt_addr;
1566     void *fdt;
1567     int rc;
1568 
1569     pef_kvm_reset(machine->cgs, &error_fatal);
1570     spapr_caps_apply(spapr);
1571 
1572     first_ppc_cpu = POWERPC_CPU(first_cpu);
1573     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1574         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1575                               spapr->max_compat_pvr)) {
1576         /*
1577          * If using KVM with radix mode available, VCPUs can be started
1578          * without a HPT because KVM will start them in radix mode.
1579          * Set the GR bit in PATE so that we know there is no HPT.
1580          */
1581         spapr->patb_entry = PATE1_GR;
1582         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1583     } else {
1584         spapr_setup_hpt(spapr);
1585     }
1586 
1587     qemu_devices_reset();
1588 
1589     spapr_ovec_cleanup(spapr->ov5_cas);
1590     spapr->ov5_cas = spapr_ovec_new();
1591 
1592     ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1593 
1594     /*
1595      * This is fixing some of the default configuration of the XIVE
1596      * devices. To be called after the reset of the machine devices.
1597      */
1598     spapr_irq_reset(spapr, &error_fatal);
1599 
1600     /*
1601      * There is no CAS under qtest. Simulate one to please the code that
1602      * depends on spapr->ov5_cas. This is especially needed to test device
1603      * unplug, so we do that before resetting the DRCs.
1604      */
1605     if (qtest_enabled()) {
1606         spapr_ovec_cleanup(spapr->ov5_cas);
1607         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1608     }
1609 
1610     /* DRC reset may cause a device to be unplugged. This will cause troubles
1611      * if this device is used by another device (eg, a running vhost backend
1612      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1613      * situations, we reset DRCs after all devices have been reset.
1614      */
1615     spapr_drc_reset_all(spapr);
1616 
1617     spapr_clear_pending_events(spapr);
1618 
1619     /*
1620      * We place the device tree and RTAS just below either the top of the RMA,
1621      * or just below 2GB, whichever is lower, so that it can be
1622      * processed with 32-bit real mode code if necessary
1623      */
1624     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1625 
1626     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1627 
1628     rc = fdt_pack(fdt);
1629 
1630     /* Should only fail if we've built a corrupted tree */
1631     assert(rc == 0);
1632 
1633     /* Load the fdt */
1634     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1635     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1636     g_free(spapr->fdt_blob);
1637     spapr->fdt_size = fdt_totalsize(fdt);
1638     spapr->fdt_initial_size = spapr->fdt_size;
1639     spapr->fdt_blob = fdt;
1640 
1641     /* Set up the entry state */
1642     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, 0, fdt_addr, 0);
1643     first_ppc_cpu->env.gpr[5] = 0;
1644 
1645     spapr->fwnmi_system_reset_addr = -1;
1646     spapr->fwnmi_machine_check_addr = -1;
1647     spapr->fwnmi_machine_check_interlock = -1;
1648 
1649     /* Signal all vCPUs waiting on this condition */
1650     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1651 
1652     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1653 }
1654 
1655 static void spapr_create_nvram(SpaprMachineState *spapr)
1656 {
1657     DeviceState *dev = qdev_new("spapr-nvram");
1658     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1659 
1660     if (dinfo) {
1661         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1662                                 &error_fatal);
1663     }
1664 
1665     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1666 
1667     spapr->nvram = (struct SpaprNvram *)dev;
1668 }
1669 
1670 static void spapr_rtc_create(SpaprMachineState *spapr)
1671 {
1672     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1673                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1674                                        &error_fatal, NULL);
1675     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1676     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1677                               "date");
1678 }
1679 
1680 /* Returns whether we want to use VGA or not */
1681 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1682 {
1683     switch (vga_interface_type) {
1684     case VGA_NONE:
1685         return false;
1686     case VGA_DEVICE:
1687         return true;
1688     case VGA_STD:
1689     case VGA_VIRTIO:
1690     case VGA_CIRRUS:
1691         return pci_vga_init(pci_bus) != NULL;
1692     default:
1693         error_setg(errp,
1694                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1695         return false;
1696     }
1697 }
1698 
1699 static int spapr_pre_load(void *opaque)
1700 {
1701     int rc;
1702 
1703     rc = spapr_caps_pre_load(opaque);
1704     if (rc) {
1705         return rc;
1706     }
1707 
1708     return 0;
1709 }
1710 
1711 static int spapr_post_load(void *opaque, int version_id)
1712 {
1713     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1714     int err = 0;
1715 
1716     err = spapr_caps_post_migration(spapr);
1717     if (err) {
1718         return err;
1719     }
1720 
1721     /*
1722      * In earlier versions, there was no separate qdev for the PAPR
1723      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1724      * So when migrating from those versions, poke the incoming offset
1725      * value into the RTC device
1726      */
1727     if (version_id < 3) {
1728         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1729         if (err) {
1730             return err;
1731         }
1732     }
1733 
1734     if (kvm_enabled() && spapr->patb_entry) {
1735         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1736         bool radix = !!(spapr->patb_entry & PATE1_GR);
1737         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1738 
1739         /*
1740          * Update LPCR:HR and UPRT as they may not be set properly in
1741          * the stream
1742          */
1743         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1744                             LPCR_HR | LPCR_UPRT);
1745 
1746         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1747         if (err) {
1748             error_report("Process table config unsupported by the host");
1749             return -EINVAL;
1750         }
1751     }
1752 
1753     err = spapr_irq_post_load(spapr, version_id);
1754     if (err) {
1755         return err;
1756     }
1757 
1758     return err;
1759 }
1760 
1761 static int spapr_pre_save(void *opaque)
1762 {
1763     int rc;
1764 
1765     rc = spapr_caps_pre_save(opaque);
1766     if (rc) {
1767         return rc;
1768     }
1769 
1770     return 0;
1771 }
1772 
1773 static bool version_before_3(void *opaque, int version_id)
1774 {
1775     return version_id < 3;
1776 }
1777 
1778 static bool spapr_pending_events_needed(void *opaque)
1779 {
1780     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1781     return !QTAILQ_EMPTY(&spapr->pending_events);
1782 }
1783 
1784 static const VMStateDescription vmstate_spapr_event_entry = {
1785     .name = "spapr_event_log_entry",
1786     .version_id = 1,
1787     .minimum_version_id = 1,
1788     .fields = (VMStateField[]) {
1789         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1790         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1791         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1792                                      NULL, extended_length),
1793         VMSTATE_END_OF_LIST()
1794     },
1795 };
1796 
1797 static const VMStateDescription vmstate_spapr_pending_events = {
1798     .name = "spapr_pending_events",
1799     .version_id = 1,
1800     .minimum_version_id = 1,
1801     .needed = spapr_pending_events_needed,
1802     .fields = (VMStateField[]) {
1803         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1804                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1805         VMSTATE_END_OF_LIST()
1806     },
1807 };
1808 
1809 static bool spapr_ov5_cas_needed(void *opaque)
1810 {
1811     SpaprMachineState *spapr = opaque;
1812     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1813     bool cas_needed;
1814 
1815     /* Prior to the introduction of SpaprOptionVector, we had two option
1816      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1817      * Both of these options encode machine topology into the device-tree
1818      * in such a way that the now-booted OS should still be able to interact
1819      * appropriately with QEMU regardless of what options were actually
1820      * negotiatied on the source side.
1821      *
1822      * As such, we can avoid migrating the CAS-negotiated options if these
1823      * are the only options available on the current machine/platform.
1824      * Since these are the only options available for pseries-2.7 and
1825      * earlier, this allows us to maintain old->new/new->old migration
1826      * compatibility.
1827      *
1828      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1829      * via default pseries-2.8 machines and explicit command-line parameters.
1830      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1831      * of the actual CAS-negotiated values to continue working properly. For
1832      * example, availability of memory unplug depends on knowing whether
1833      * OV5_HP_EVT was negotiated via CAS.
1834      *
1835      * Thus, for any cases where the set of available CAS-negotiatable
1836      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1837      * include the CAS-negotiated options in the migration stream, unless
1838      * if they affect boot time behaviour only.
1839      */
1840     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1841     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1842     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1843 
1844     /* We need extra information if we have any bits outside the mask
1845      * defined above */
1846     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1847 
1848     spapr_ovec_cleanup(ov5_mask);
1849 
1850     return cas_needed;
1851 }
1852 
1853 static const VMStateDescription vmstate_spapr_ov5_cas = {
1854     .name = "spapr_option_vector_ov5_cas",
1855     .version_id = 1,
1856     .minimum_version_id = 1,
1857     .needed = spapr_ov5_cas_needed,
1858     .fields = (VMStateField[]) {
1859         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1860                                  vmstate_spapr_ovec, SpaprOptionVector),
1861         VMSTATE_END_OF_LIST()
1862     },
1863 };
1864 
1865 static bool spapr_patb_entry_needed(void *opaque)
1866 {
1867     SpaprMachineState *spapr = opaque;
1868 
1869     return !!spapr->patb_entry;
1870 }
1871 
1872 static const VMStateDescription vmstate_spapr_patb_entry = {
1873     .name = "spapr_patb_entry",
1874     .version_id = 1,
1875     .minimum_version_id = 1,
1876     .needed = spapr_patb_entry_needed,
1877     .fields = (VMStateField[]) {
1878         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1879         VMSTATE_END_OF_LIST()
1880     },
1881 };
1882 
1883 static bool spapr_irq_map_needed(void *opaque)
1884 {
1885     SpaprMachineState *spapr = opaque;
1886 
1887     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1888 }
1889 
1890 static const VMStateDescription vmstate_spapr_irq_map = {
1891     .name = "spapr_irq_map",
1892     .version_id = 1,
1893     .minimum_version_id = 1,
1894     .needed = spapr_irq_map_needed,
1895     .fields = (VMStateField[]) {
1896         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1897         VMSTATE_END_OF_LIST()
1898     },
1899 };
1900 
1901 static bool spapr_dtb_needed(void *opaque)
1902 {
1903     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1904 
1905     return smc->update_dt_enabled;
1906 }
1907 
1908 static int spapr_dtb_pre_load(void *opaque)
1909 {
1910     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1911 
1912     g_free(spapr->fdt_blob);
1913     spapr->fdt_blob = NULL;
1914     spapr->fdt_size = 0;
1915 
1916     return 0;
1917 }
1918 
1919 static const VMStateDescription vmstate_spapr_dtb = {
1920     .name = "spapr_dtb",
1921     .version_id = 1,
1922     .minimum_version_id = 1,
1923     .needed = spapr_dtb_needed,
1924     .pre_load = spapr_dtb_pre_load,
1925     .fields = (VMStateField[]) {
1926         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1927         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1928         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1929                                      fdt_size),
1930         VMSTATE_END_OF_LIST()
1931     },
1932 };
1933 
1934 static bool spapr_fwnmi_needed(void *opaque)
1935 {
1936     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1937 
1938     return spapr->fwnmi_machine_check_addr != -1;
1939 }
1940 
1941 static int spapr_fwnmi_pre_save(void *opaque)
1942 {
1943     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1944 
1945     /*
1946      * Check if machine check handling is in progress and print a
1947      * warning message.
1948      */
1949     if (spapr->fwnmi_machine_check_interlock != -1) {
1950         warn_report("A machine check is being handled during migration. The"
1951                 "handler may run and log hardware error on the destination");
1952     }
1953 
1954     return 0;
1955 }
1956 
1957 static const VMStateDescription vmstate_spapr_fwnmi = {
1958     .name = "spapr_fwnmi",
1959     .version_id = 1,
1960     .minimum_version_id = 1,
1961     .needed = spapr_fwnmi_needed,
1962     .pre_save = spapr_fwnmi_pre_save,
1963     .fields = (VMStateField[]) {
1964         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
1965         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
1966         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
1967         VMSTATE_END_OF_LIST()
1968     },
1969 };
1970 
1971 static const VMStateDescription vmstate_spapr = {
1972     .name = "spapr",
1973     .version_id = 3,
1974     .minimum_version_id = 1,
1975     .pre_load = spapr_pre_load,
1976     .post_load = spapr_post_load,
1977     .pre_save = spapr_pre_save,
1978     .fields = (VMStateField[]) {
1979         /* used to be @next_irq */
1980         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1981 
1982         /* RTC offset */
1983         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
1984 
1985         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
1986         VMSTATE_END_OF_LIST()
1987     },
1988     .subsections = (const VMStateDescription*[]) {
1989         &vmstate_spapr_ov5_cas,
1990         &vmstate_spapr_patb_entry,
1991         &vmstate_spapr_pending_events,
1992         &vmstate_spapr_cap_htm,
1993         &vmstate_spapr_cap_vsx,
1994         &vmstate_spapr_cap_dfp,
1995         &vmstate_spapr_cap_cfpc,
1996         &vmstate_spapr_cap_sbbc,
1997         &vmstate_spapr_cap_ibs,
1998         &vmstate_spapr_cap_hpt_maxpagesize,
1999         &vmstate_spapr_irq_map,
2000         &vmstate_spapr_cap_nested_kvm_hv,
2001         &vmstate_spapr_dtb,
2002         &vmstate_spapr_cap_large_decr,
2003         &vmstate_spapr_cap_ccf_assist,
2004         &vmstate_spapr_cap_fwnmi,
2005         &vmstate_spapr_fwnmi,
2006         NULL
2007     }
2008 };
2009 
2010 static int htab_save_setup(QEMUFile *f, void *opaque)
2011 {
2012     SpaprMachineState *spapr = opaque;
2013 
2014     /* "Iteration" header */
2015     if (!spapr->htab_shift) {
2016         qemu_put_be32(f, -1);
2017     } else {
2018         qemu_put_be32(f, spapr->htab_shift);
2019     }
2020 
2021     if (spapr->htab) {
2022         spapr->htab_save_index = 0;
2023         spapr->htab_first_pass = true;
2024     } else {
2025         if (spapr->htab_shift) {
2026             assert(kvm_enabled());
2027         }
2028     }
2029 
2030 
2031     return 0;
2032 }
2033 
2034 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2035                             int chunkstart, int n_valid, int n_invalid)
2036 {
2037     qemu_put_be32(f, chunkstart);
2038     qemu_put_be16(f, n_valid);
2039     qemu_put_be16(f, n_invalid);
2040     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2041                     HASH_PTE_SIZE_64 * n_valid);
2042 }
2043 
2044 static void htab_save_end_marker(QEMUFile *f)
2045 {
2046     qemu_put_be32(f, 0);
2047     qemu_put_be16(f, 0);
2048     qemu_put_be16(f, 0);
2049 }
2050 
2051 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2052                                  int64_t max_ns)
2053 {
2054     bool has_timeout = max_ns != -1;
2055     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2056     int index = spapr->htab_save_index;
2057     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2058 
2059     assert(spapr->htab_first_pass);
2060 
2061     do {
2062         int chunkstart;
2063 
2064         /* Consume invalid HPTEs */
2065         while ((index < htabslots)
2066                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2067             CLEAN_HPTE(HPTE(spapr->htab, index));
2068             index++;
2069         }
2070 
2071         /* Consume valid HPTEs */
2072         chunkstart = index;
2073         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2074                && HPTE_VALID(HPTE(spapr->htab, index))) {
2075             CLEAN_HPTE(HPTE(spapr->htab, index));
2076             index++;
2077         }
2078 
2079         if (index > chunkstart) {
2080             int n_valid = index - chunkstart;
2081 
2082             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2083 
2084             if (has_timeout &&
2085                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2086                 break;
2087             }
2088         }
2089     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2090 
2091     if (index >= htabslots) {
2092         assert(index == htabslots);
2093         index = 0;
2094         spapr->htab_first_pass = false;
2095     }
2096     spapr->htab_save_index = index;
2097 }
2098 
2099 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2100                                 int64_t max_ns)
2101 {
2102     bool final = max_ns < 0;
2103     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2104     int examined = 0, sent = 0;
2105     int index = spapr->htab_save_index;
2106     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2107 
2108     assert(!spapr->htab_first_pass);
2109 
2110     do {
2111         int chunkstart, invalidstart;
2112 
2113         /* Consume non-dirty HPTEs */
2114         while ((index < htabslots)
2115                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2116             index++;
2117             examined++;
2118         }
2119 
2120         chunkstart = index;
2121         /* Consume valid dirty HPTEs */
2122         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2123                && HPTE_DIRTY(HPTE(spapr->htab, index))
2124                && HPTE_VALID(HPTE(spapr->htab, index))) {
2125             CLEAN_HPTE(HPTE(spapr->htab, index));
2126             index++;
2127             examined++;
2128         }
2129 
2130         invalidstart = index;
2131         /* Consume invalid dirty HPTEs */
2132         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2133                && HPTE_DIRTY(HPTE(spapr->htab, index))
2134                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2135             CLEAN_HPTE(HPTE(spapr->htab, index));
2136             index++;
2137             examined++;
2138         }
2139 
2140         if (index > chunkstart) {
2141             int n_valid = invalidstart - chunkstart;
2142             int n_invalid = index - invalidstart;
2143 
2144             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2145             sent += index - chunkstart;
2146 
2147             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2148                 break;
2149             }
2150         }
2151 
2152         if (examined >= htabslots) {
2153             break;
2154         }
2155 
2156         if (index >= htabslots) {
2157             assert(index == htabslots);
2158             index = 0;
2159         }
2160     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2161 
2162     if (index >= htabslots) {
2163         assert(index == htabslots);
2164         index = 0;
2165     }
2166 
2167     spapr->htab_save_index = index;
2168 
2169     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2170 }
2171 
2172 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2173 #define MAX_KVM_BUF_SIZE    2048
2174 
2175 static int htab_save_iterate(QEMUFile *f, void *opaque)
2176 {
2177     SpaprMachineState *spapr = opaque;
2178     int fd;
2179     int rc = 0;
2180 
2181     /* Iteration header */
2182     if (!spapr->htab_shift) {
2183         qemu_put_be32(f, -1);
2184         return 1;
2185     } else {
2186         qemu_put_be32(f, 0);
2187     }
2188 
2189     if (!spapr->htab) {
2190         assert(kvm_enabled());
2191 
2192         fd = get_htab_fd(spapr);
2193         if (fd < 0) {
2194             return fd;
2195         }
2196 
2197         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2198         if (rc < 0) {
2199             return rc;
2200         }
2201     } else  if (spapr->htab_first_pass) {
2202         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2203     } else {
2204         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2205     }
2206 
2207     htab_save_end_marker(f);
2208 
2209     return rc;
2210 }
2211 
2212 static int htab_save_complete(QEMUFile *f, void *opaque)
2213 {
2214     SpaprMachineState *spapr = opaque;
2215     int fd;
2216 
2217     /* Iteration header */
2218     if (!spapr->htab_shift) {
2219         qemu_put_be32(f, -1);
2220         return 0;
2221     } else {
2222         qemu_put_be32(f, 0);
2223     }
2224 
2225     if (!spapr->htab) {
2226         int rc;
2227 
2228         assert(kvm_enabled());
2229 
2230         fd = get_htab_fd(spapr);
2231         if (fd < 0) {
2232             return fd;
2233         }
2234 
2235         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2236         if (rc < 0) {
2237             return rc;
2238         }
2239     } else {
2240         if (spapr->htab_first_pass) {
2241             htab_save_first_pass(f, spapr, -1);
2242         }
2243         htab_save_later_pass(f, spapr, -1);
2244     }
2245 
2246     /* End marker */
2247     htab_save_end_marker(f);
2248 
2249     return 0;
2250 }
2251 
2252 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2253 {
2254     SpaprMachineState *spapr = opaque;
2255     uint32_t section_hdr;
2256     int fd = -1;
2257     Error *local_err = NULL;
2258 
2259     if (version_id < 1 || version_id > 1) {
2260         error_report("htab_load() bad version");
2261         return -EINVAL;
2262     }
2263 
2264     section_hdr = qemu_get_be32(f);
2265 
2266     if (section_hdr == -1) {
2267         spapr_free_hpt(spapr);
2268         return 0;
2269     }
2270 
2271     if (section_hdr) {
2272         int ret;
2273 
2274         /* First section gives the htab size */
2275         ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2276         if (ret < 0) {
2277             error_report_err(local_err);
2278             return ret;
2279         }
2280         return 0;
2281     }
2282 
2283     if (!spapr->htab) {
2284         assert(kvm_enabled());
2285 
2286         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2287         if (fd < 0) {
2288             error_report_err(local_err);
2289             return fd;
2290         }
2291     }
2292 
2293     while (true) {
2294         uint32_t index;
2295         uint16_t n_valid, n_invalid;
2296 
2297         index = qemu_get_be32(f);
2298         n_valid = qemu_get_be16(f);
2299         n_invalid = qemu_get_be16(f);
2300 
2301         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2302             /* End of Stream */
2303             break;
2304         }
2305 
2306         if ((index + n_valid + n_invalid) >
2307             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2308             /* Bad index in stream */
2309             error_report(
2310                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2311                 index, n_valid, n_invalid, spapr->htab_shift);
2312             return -EINVAL;
2313         }
2314 
2315         if (spapr->htab) {
2316             if (n_valid) {
2317                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2318                                 HASH_PTE_SIZE_64 * n_valid);
2319             }
2320             if (n_invalid) {
2321                 memset(HPTE(spapr->htab, index + n_valid), 0,
2322                        HASH_PTE_SIZE_64 * n_invalid);
2323             }
2324         } else {
2325             int rc;
2326 
2327             assert(fd >= 0);
2328 
2329             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2330                                         &local_err);
2331             if (rc < 0) {
2332                 error_report_err(local_err);
2333                 return rc;
2334             }
2335         }
2336     }
2337 
2338     if (!spapr->htab) {
2339         assert(fd >= 0);
2340         close(fd);
2341     }
2342 
2343     return 0;
2344 }
2345 
2346 static void htab_save_cleanup(void *opaque)
2347 {
2348     SpaprMachineState *spapr = opaque;
2349 
2350     close_htab_fd(spapr);
2351 }
2352 
2353 static SaveVMHandlers savevm_htab_handlers = {
2354     .save_setup = htab_save_setup,
2355     .save_live_iterate = htab_save_iterate,
2356     .save_live_complete_precopy = htab_save_complete,
2357     .save_cleanup = htab_save_cleanup,
2358     .load_state = htab_load,
2359 };
2360 
2361 static void spapr_boot_set(void *opaque, const char *boot_device,
2362                            Error **errp)
2363 {
2364     MachineState *machine = MACHINE(opaque);
2365     machine->boot_order = g_strdup(boot_device);
2366 }
2367 
2368 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2369 {
2370     MachineState *machine = MACHINE(spapr);
2371     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2372     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2373     int i;
2374 
2375     for (i = 0; i < nr_lmbs; i++) {
2376         uint64_t addr;
2377 
2378         addr = i * lmb_size + machine->device_memory->base;
2379         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2380                                addr / lmb_size);
2381     }
2382 }
2383 
2384 /*
2385  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2386  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2387  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2388  */
2389 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2390 {
2391     int i;
2392 
2393     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2394         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2395                    " is not aligned to %" PRIu64 " MiB",
2396                    machine->ram_size,
2397                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2398         return;
2399     }
2400 
2401     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2402         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2403                    " is not aligned to %" PRIu64 " MiB",
2404                    machine->ram_size,
2405                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2406         return;
2407     }
2408 
2409     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2410         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2411             error_setg(errp,
2412                        "Node %d memory size 0x%" PRIx64
2413                        " is not aligned to %" PRIu64 " MiB",
2414                        i, machine->numa_state->nodes[i].node_mem,
2415                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2416             return;
2417         }
2418     }
2419 }
2420 
2421 /* find cpu slot in machine->possible_cpus by core_id */
2422 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2423 {
2424     int index = id / ms->smp.threads;
2425 
2426     if (index >= ms->possible_cpus->len) {
2427         return NULL;
2428     }
2429     if (idx) {
2430         *idx = index;
2431     }
2432     return &ms->possible_cpus->cpus[index];
2433 }
2434 
2435 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2436 {
2437     MachineState *ms = MACHINE(spapr);
2438     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2439     Error *local_err = NULL;
2440     bool vsmt_user = !!spapr->vsmt;
2441     int kvm_smt = kvmppc_smt_threads();
2442     int ret;
2443     unsigned int smp_threads = ms->smp.threads;
2444 
2445     if (!kvm_enabled() && (smp_threads > 1)) {
2446         error_setg(errp, "TCG cannot support more than 1 thread/core "
2447                    "on a pseries machine");
2448         return;
2449     }
2450     if (!is_power_of_2(smp_threads)) {
2451         error_setg(errp, "Cannot support %d threads/core on a pseries "
2452                    "machine because it must be a power of 2", smp_threads);
2453         return;
2454     }
2455 
2456     /* Detemine the VSMT mode to use: */
2457     if (vsmt_user) {
2458         if (spapr->vsmt < smp_threads) {
2459             error_setg(errp, "Cannot support VSMT mode %d"
2460                        " because it must be >= threads/core (%d)",
2461                        spapr->vsmt, smp_threads);
2462             return;
2463         }
2464         /* In this case, spapr->vsmt has been set by the command line */
2465     } else if (!smc->smp_threads_vsmt) {
2466         /*
2467          * Default VSMT value is tricky, because we need it to be as
2468          * consistent as possible (for migration), but this requires
2469          * changing it for at least some existing cases.  We pick 8 as
2470          * the value that we'd get with KVM on POWER8, the
2471          * overwhelmingly common case in production systems.
2472          */
2473         spapr->vsmt = MAX(8, smp_threads);
2474     } else {
2475         spapr->vsmt = smp_threads;
2476     }
2477 
2478     /* KVM: If necessary, set the SMT mode: */
2479     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2480         ret = kvmppc_set_smt_threads(spapr->vsmt);
2481         if (ret) {
2482             /* Looks like KVM isn't able to change VSMT mode */
2483             error_setg(&local_err,
2484                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2485                        spapr->vsmt, ret);
2486             /* We can live with that if the default one is big enough
2487              * for the number of threads, and a submultiple of the one
2488              * we want.  In this case we'll waste some vcpu ids, but
2489              * behaviour will be correct */
2490             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2491                 warn_report_err(local_err);
2492             } else {
2493                 if (!vsmt_user) {
2494                     error_append_hint(&local_err,
2495                                       "On PPC, a VM with %d threads/core"
2496                                       " on a host with %d threads/core"
2497                                       " requires the use of VSMT mode %d.\n",
2498                                       smp_threads, kvm_smt, spapr->vsmt);
2499                 }
2500                 kvmppc_error_append_smt_possible_hint(&local_err);
2501                 error_propagate(errp, local_err);
2502             }
2503         }
2504     }
2505     /* else TCG: nothing to do currently */
2506 }
2507 
2508 static void spapr_init_cpus(SpaprMachineState *spapr)
2509 {
2510     MachineState *machine = MACHINE(spapr);
2511     MachineClass *mc = MACHINE_GET_CLASS(machine);
2512     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2513     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2514     const CPUArchIdList *possible_cpus;
2515     unsigned int smp_cpus = machine->smp.cpus;
2516     unsigned int smp_threads = machine->smp.threads;
2517     unsigned int max_cpus = machine->smp.max_cpus;
2518     int boot_cores_nr = smp_cpus / smp_threads;
2519     int i;
2520 
2521     possible_cpus = mc->possible_cpu_arch_ids(machine);
2522     if (mc->has_hotpluggable_cpus) {
2523         if (smp_cpus % smp_threads) {
2524             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2525                          smp_cpus, smp_threads);
2526             exit(1);
2527         }
2528         if (max_cpus % smp_threads) {
2529             error_report("max_cpus (%u) must be multiple of threads (%u)",
2530                          max_cpus, smp_threads);
2531             exit(1);
2532         }
2533     } else {
2534         if (max_cpus != smp_cpus) {
2535             error_report("This machine version does not support CPU hotplug");
2536             exit(1);
2537         }
2538         boot_cores_nr = possible_cpus->len;
2539     }
2540 
2541     if (smc->pre_2_10_has_unused_icps) {
2542         int i;
2543 
2544         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2545             /* Dummy entries get deregistered when real ICPState objects
2546              * are registered during CPU core hotplug.
2547              */
2548             pre_2_10_vmstate_register_dummy_icp(i);
2549         }
2550     }
2551 
2552     for (i = 0; i < possible_cpus->len; i++) {
2553         int core_id = i * smp_threads;
2554 
2555         if (mc->has_hotpluggable_cpus) {
2556             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2557                                    spapr_vcpu_id(spapr, core_id));
2558         }
2559 
2560         if (i < boot_cores_nr) {
2561             Object *core  = object_new(type);
2562             int nr_threads = smp_threads;
2563 
2564             /* Handle the partially filled core for older machine types */
2565             if ((i + 1) * smp_threads >= smp_cpus) {
2566                 nr_threads = smp_cpus - i * smp_threads;
2567             }
2568 
2569             object_property_set_int(core, "nr-threads", nr_threads,
2570                                     &error_fatal);
2571             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2572                                     &error_fatal);
2573             qdev_realize(DEVICE(core), NULL, &error_fatal);
2574 
2575             object_unref(core);
2576         }
2577     }
2578 }
2579 
2580 static PCIHostState *spapr_create_default_phb(void)
2581 {
2582     DeviceState *dev;
2583 
2584     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2585     qdev_prop_set_uint32(dev, "index", 0);
2586     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2587 
2588     return PCI_HOST_BRIDGE(dev);
2589 }
2590 
2591 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2592 {
2593     MachineState *machine = MACHINE(spapr);
2594     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2595     hwaddr rma_size = machine->ram_size;
2596     hwaddr node0_size = spapr_node0_size(machine);
2597 
2598     /* RMA has to fit in the first NUMA node */
2599     rma_size = MIN(rma_size, node0_size);
2600 
2601     /*
2602      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2603      * never exceed that
2604      */
2605     rma_size = MIN(rma_size, 1 * TiB);
2606 
2607     /*
2608      * Clamp the RMA size based on machine type.  This is for
2609      * migration compatibility with older qemu versions, which limited
2610      * the RMA size for complicated and mostly bad reasons.
2611      */
2612     if (smc->rma_limit) {
2613         rma_size = MIN(rma_size, smc->rma_limit);
2614     }
2615 
2616     if (rma_size < MIN_RMA_SLOF) {
2617         error_setg(errp,
2618                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2619                    "ldMiB guest RMA (Real Mode Area memory)",
2620                    MIN_RMA_SLOF / MiB);
2621         return 0;
2622     }
2623 
2624     return rma_size;
2625 }
2626 
2627 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2628 {
2629     MachineState *machine = MACHINE(spapr);
2630     int i;
2631 
2632     for (i = 0; i < machine->ram_slots; i++) {
2633         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2634     }
2635 }
2636 
2637 /* pSeries LPAR / sPAPR hardware init */
2638 static void spapr_machine_init(MachineState *machine)
2639 {
2640     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2641     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2642     MachineClass *mc = MACHINE_GET_CLASS(machine);
2643     const char *bios_name = machine->firmware ?: FW_FILE_NAME;
2644     const char *kernel_filename = machine->kernel_filename;
2645     const char *initrd_filename = machine->initrd_filename;
2646     PCIHostState *phb;
2647     int i;
2648     MemoryRegion *sysmem = get_system_memory();
2649     long load_limit, fw_size;
2650     char *filename;
2651     Error *resize_hpt_err = NULL;
2652 
2653     /*
2654      * if Secure VM (PEF) support is configured, then initialize it
2655      */
2656     pef_kvm_init(machine->cgs, &error_fatal);
2657 
2658     msi_nonbroken = true;
2659 
2660     QLIST_INIT(&spapr->phbs);
2661     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2662 
2663     /* Determine capabilities to run with */
2664     spapr_caps_init(spapr);
2665 
2666     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2667     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2668         /*
2669          * If the user explicitly requested a mode we should either
2670          * supply it, or fail completely (which we do below).  But if
2671          * it's not set explicitly, we reset our mode to something
2672          * that works
2673          */
2674         if (resize_hpt_err) {
2675             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2676             error_free(resize_hpt_err);
2677             resize_hpt_err = NULL;
2678         } else {
2679             spapr->resize_hpt = smc->resize_hpt_default;
2680         }
2681     }
2682 
2683     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2684 
2685     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2686         /*
2687          * User requested HPT resize, but this host can't supply it.  Bail out
2688          */
2689         error_report_err(resize_hpt_err);
2690         exit(1);
2691     }
2692     error_free(resize_hpt_err);
2693 
2694     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2695 
2696     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2697     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2698 
2699     /*
2700      * VSMT must be set in order to be able to compute VCPU ids, ie to
2701      * call spapr_max_server_number() or spapr_vcpu_id().
2702      */
2703     spapr_set_vsmt_mode(spapr, &error_fatal);
2704 
2705     /* Set up Interrupt Controller before we create the VCPUs */
2706     spapr_irq_init(spapr, &error_fatal);
2707 
2708     /* Set up containers for ibm,client-architecture-support negotiated options
2709      */
2710     spapr->ov5 = spapr_ovec_new();
2711     spapr->ov5_cas = spapr_ovec_new();
2712 
2713     if (smc->dr_lmb_enabled) {
2714         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2715         spapr_validate_node_memory(machine, &error_fatal);
2716     }
2717 
2718     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2719 
2720     /* advertise support for dedicated HP event source to guests */
2721     if (spapr->use_hotplug_event_source) {
2722         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2723     }
2724 
2725     /* advertise support for HPT resizing */
2726     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2727         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2728     }
2729 
2730     /* advertise support for ibm,dyamic-memory-v2 */
2731     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2732 
2733     /* advertise XIVE on POWER9 machines */
2734     if (spapr->irq->xive) {
2735         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2736     }
2737 
2738     /* init CPUs */
2739     spapr_init_cpus(spapr);
2740 
2741     /*
2742      * check we don't have a memory-less/cpu-less NUMA node
2743      * Firmware relies on the existing memory/cpu topology to provide the
2744      * NUMA topology to the kernel.
2745      * And the linux kernel needs to know the NUMA topology at start
2746      * to be able to hotplug CPUs later.
2747      */
2748     if (machine->numa_state->num_nodes) {
2749         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2750             /* check for memory-less node */
2751             if (machine->numa_state->nodes[i].node_mem == 0) {
2752                 CPUState *cs;
2753                 int found = 0;
2754                 /* check for cpu-less node */
2755                 CPU_FOREACH(cs) {
2756                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2757                     if (cpu->node_id == i) {
2758                         found = 1;
2759                         break;
2760                     }
2761                 }
2762                 /* memory-less and cpu-less node */
2763                 if (!found) {
2764                     error_report(
2765                        "Memory-less/cpu-less nodes are not supported (node %d)",
2766                                  i);
2767                     exit(1);
2768                 }
2769             }
2770         }
2771 
2772     }
2773 
2774     spapr->gpu_numa_id = spapr_numa_initial_nvgpu_numa_id(machine);
2775 
2776     /* Init numa_assoc_array */
2777     spapr_numa_associativity_init(spapr, machine);
2778 
2779     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2780         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2781                               spapr->max_compat_pvr)) {
2782         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2783         /* KVM and TCG always allow GTSE with radix... */
2784         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2785     }
2786     /* ... but not with hash (currently). */
2787 
2788     if (kvm_enabled()) {
2789         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2790         kvmppc_enable_logical_ci_hcalls();
2791         kvmppc_enable_set_mode_hcall();
2792 
2793         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2794         kvmppc_enable_clear_ref_mod_hcalls();
2795 
2796         /* Enable H_PAGE_INIT */
2797         kvmppc_enable_h_page_init();
2798     }
2799 
2800     /* map RAM */
2801     memory_region_add_subregion(sysmem, 0, machine->ram);
2802 
2803     /* always allocate the device memory information */
2804     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2805 
2806     /* initialize hotplug memory address space */
2807     if (machine->ram_size < machine->maxram_size) {
2808         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2809         /*
2810          * Limit the number of hotpluggable memory slots to half the number
2811          * slots that KVM supports, leaving the other half for PCI and other
2812          * devices. However ensure that number of slots doesn't drop below 32.
2813          */
2814         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2815                            SPAPR_MAX_RAM_SLOTS;
2816 
2817         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2818             max_memslots = SPAPR_MAX_RAM_SLOTS;
2819         }
2820         if (machine->ram_slots > max_memslots) {
2821             error_report("Specified number of memory slots %"
2822                          PRIu64" exceeds max supported %d",
2823                          machine->ram_slots, max_memslots);
2824             exit(1);
2825         }
2826 
2827         machine->device_memory->base = ROUND_UP(machine->ram_size,
2828                                                 SPAPR_DEVICE_MEM_ALIGN);
2829         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2830                            "device-memory", device_mem_size);
2831         memory_region_add_subregion(sysmem, machine->device_memory->base,
2832                                     &machine->device_memory->mr);
2833     }
2834 
2835     if (smc->dr_lmb_enabled) {
2836         spapr_create_lmb_dr_connectors(spapr);
2837     }
2838 
2839     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2840         /* Create the error string for live migration blocker */
2841         error_setg(&spapr->fwnmi_migration_blocker,
2842             "A machine check is being handled during migration. The handler"
2843             "may run and log hardware error on the destination");
2844     }
2845 
2846     if (mc->nvdimm_supported) {
2847         spapr_create_nvdimm_dr_connectors(spapr);
2848     }
2849 
2850     /* Set up RTAS event infrastructure */
2851     spapr_events_init(spapr);
2852 
2853     /* Set up the RTC RTAS interfaces */
2854     spapr_rtc_create(spapr);
2855 
2856     /* Set up VIO bus */
2857     spapr->vio_bus = spapr_vio_bus_init();
2858 
2859     for (i = 0; serial_hd(i); i++) {
2860         spapr_vty_create(spapr->vio_bus, serial_hd(i));
2861     }
2862 
2863     /* We always have at least the nvram device on VIO */
2864     spapr_create_nvram(spapr);
2865 
2866     /*
2867      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2868      * connectors (described in root DT node's "ibm,drc-types" property)
2869      * are pre-initialized here. additional child connectors (such as
2870      * connectors for a PHBs PCI slots) are added as needed during their
2871      * parent's realization.
2872      */
2873     if (smc->dr_phb_enabled) {
2874         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2875             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2876         }
2877     }
2878 
2879     /* Set up PCI */
2880     spapr_pci_rtas_init();
2881 
2882     phb = spapr_create_default_phb();
2883 
2884     for (i = 0; i < nb_nics; i++) {
2885         NICInfo *nd = &nd_table[i];
2886 
2887         if (!nd->model) {
2888             nd->model = g_strdup("spapr-vlan");
2889         }
2890 
2891         if (g_str_equal(nd->model, "spapr-vlan") ||
2892             g_str_equal(nd->model, "ibmveth")) {
2893             spapr_vlan_create(spapr->vio_bus, nd);
2894         } else {
2895             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2896         }
2897     }
2898 
2899     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2900         spapr_vscsi_create(spapr->vio_bus);
2901     }
2902 
2903     /* Graphics */
2904     if (spapr_vga_init(phb->bus, &error_fatal)) {
2905         spapr->has_graphics = true;
2906         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2907     }
2908 
2909     if (machine->usb) {
2910         if (smc->use_ohci_by_default) {
2911             pci_create_simple(phb->bus, -1, "pci-ohci");
2912         } else {
2913             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2914         }
2915 
2916         if (spapr->has_graphics) {
2917             USBBus *usb_bus = usb_bus_find(-1);
2918 
2919             usb_create_simple(usb_bus, "usb-kbd");
2920             usb_create_simple(usb_bus, "usb-mouse");
2921         }
2922     }
2923 
2924     if (kernel_filename) {
2925         spapr->kernel_size = load_elf(kernel_filename, NULL,
2926                                       translate_kernel_address, spapr,
2927                                       NULL, NULL, NULL, NULL, 1,
2928                                       PPC_ELF_MACHINE, 0, 0);
2929         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2930             spapr->kernel_size = load_elf(kernel_filename, NULL,
2931                                           translate_kernel_address, spapr,
2932                                           NULL, NULL, NULL, NULL, 0,
2933                                           PPC_ELF_MACHINE, 0, 0);
2934             spapr->kernel_le = spapr->kernel_size > 0;
2935         }
2936         if (spapr->kernel_size < 0) {
2937             error_report("error loading %s: %s", kernel_filename,
2938                          load_elf_strerror(spapr->kernel_size));
2939             exit(1);
2940         }
2941 
2942         /* load initrd */
2943         if (initrd_filename) {
2944             /* Try to locate the initrd in the gap between the kernel
2945              * and the firmware. Add a bit of space just in case
2946              */
2947             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
2948                                   + 0x1ffff) & ~0xffff;
2949             spapr->initrd_size = load_image_targphys(initrd_filename,
2950                                                      spapr->initrd_base,
2951                                                      load_limit
2952                                                      - spapr->initrd_base);
2953             if (spapr->initrd_size < 0) {
2954                 error_report("could not load initial ram disk '%s'",
2955                              initrd_filename);
2956                 exit(1);
2957             }
2958         }
2959     }
2960 
2961     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2962     if (!filename) {
2963         error_report("Could not find LPAR firmware '%s'", bios_name);
2964         exit(1);
2965     }
2966     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2967     if (fw_size <= 0) {
2968         error_report("Could not load LPAR firmware '%s'", filename);
2969         exit(1);
2970     }
2971     g_free(filename);
2972 
2973     /* FIXME: Should register things through the MachineState's qdev
2974      * interface, this is a legacy from the sPAPREnvironment structure
2975      * which predated MachineState but had a similar function */
2976     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2977     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
2978                          &savevm_htab_handlers, spapr);
2979 
2980     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
2981 
2982     qemu_register_boot_set(spapr_boot_set, spapr);
2983 
2984     /*
2985      * Nothing needs to be done to resume a suspended guest because
2986      * suspending does not change the machine state, so no need for
2987      * a ->wakeup method.
2988      */
2989     qemu_register_wakeup_support();
2990 
2991     if (kvm_enabled()) {
2992         /* to stop and start vmclock */
2993         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2994                                          &spapr->tb);
2995 
2996         kvmppc_spapr_enable_inkernel_multitce();
2997     }
2998 
2999     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3000 }
3001 
3002 #define DEFAULT_KVM_TYPE "auto"
3003 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3004 {
3005     /*
3006      * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to
3007      * accomodate the 'HV' and 'PV' formats that exists in the
3008      * wild. The 'auto' mode is being introduced already as
3009      * lower-case, thus we don't need to bother checking for
3010      * "AUTO".
3011      */
3012     if (!vm_type || !strcmp(vm_type, DEFAULT_KVM_TYPE)) {
3013         return 0;
3014     }
3015 
3016     if (!g_ascii_strcasecmp(vm_type, "hv")) {
3017         return 1;
3018     }
3019 
3020     if (!g_ascii_strcasecmp(vm_type, "pr")) {
3021         return 2;
3022     }
3023 
3024     error_report("Unknown kvm-type specified '%s'", vm_type);
3025     exit(1);
3026 }
3027 
3028 /*
3029  * Implementation of an interface to adjust firmware path
3030  * for the bootindex property handling.
3031  */
3032 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3033                                    DeviceState *dev)
3034 {
3035 #define CAST(type, obj, name) \
3036     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3037     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3038     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3039     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3040     PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3041 
3042     if (d) {
3043         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3044         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3045         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3046 
3047         if (spapr) {
3048             /*
3049              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3050              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3051              * 0x8000 | (target << 8) | (bus << 5) | lun
3052              * (see the "Logical unit addressing format" table in SAM5)
3053              */
3054             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3055             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3056                                    (uint64_t)id << 48);
3057         } else if (virtio) {
3058             /*
3059              * We use SRP luns of the form 01000000 | (target << 8) | lun
3060              * in the top 32 bits of the 64-bit LUN
3061              * Note: the quote above is from SLOF and it is wrong,
3062              * the actual binding is:
3063              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3064              */
3065             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3066             if (d->lun >= 256) {
3067                 /* Use the LUN "flat space addressing method" */
3068                 id |= 0x4000;
3069             }
3070             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3071                                    (uint64_t)id << 32);
3072         } else if (usb) {
3073             /*
3074              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3075              * in the top 32 bits of the 64-bit LUN
3076              */
3077             unsigned usb_port = atoi(usb->port->path);
3078             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3079             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3080                                    (uint64_t)id << 32);
3081         }
3082     }
3083 
3084     /*
3085      * SLOF probes the USB devices, and if it recognizes that the device is a
3086      * storage device, it changes its name to "storage" instead of "usb-host",
3087      * and additionally adds a child node for the SCSI LUN, so the correct
3088      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3089      */
3090     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3091         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3092         if (usb_host_dev_is_scsi_storage(usbdev)) {
3093             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3094         }
3095     }
3096 
3097     if (phb) {
3098         /* Replace "pci" with "pci@800000020000000" */
3099         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3100     }
3101 
3102     if (vsc) {
3103         /* Same logic as virtio above */
3104         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3105         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3106     }
3107 
3108     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3109         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3110         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3111         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3112     }
3113 
3114     if (pcidev) {
3115         return spapr_pci_fw_dev_name(pcidev);
3116     }
3117 
3118     return NULL;
3119 }
3120 
3121 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3122 {
3123     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3124 
3125     return g_strdup(spapr->kvm_type);
3126 }
3127 
3128 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3129 {
3130     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3131 
3132     g_free(spapr->kvm_type);
3133     spapr->kvm_type = g_strdup(value);
3134 }
3135 
3136 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3137 {
3138     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3139 
3140     return spapr->use_hotplug_event_source;
3141 }
3142 
3143 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3144                                             Error **errp)
3145 {
3146     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3147 
3148     spapr->use_hotplug_event_source = value;
3149 }
3150 
3151 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3152 {
3153     return true;
3154 }
3155 
3156 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3157 {
3158     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3159 
3160     switch (spapr->resize_hpt) {
3161     case SPAPR_RESIZE_HPT_DEFAULT:
3162         return g_strdup("default");
3163     case SPAPR_RESIZE_HPT_DISABLED:
3164         return g_strdup("disabled");
3165     case SPAPR_RESIZE_HPT_ENABLED:
3166         return g_strdup("enabled");
3167     case SPAPR_RESIZE_HPT_REQUIRED:
3168         return g_strdup("required");
3169     }
3170     g_assert_not_reached();
3171 }
3172 
3173 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3174 {
3175     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3176 
3177     if (strcmp(value, "default") == 0) {
3178         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3179     } else if (strcmp(value, "disabled") == 0) {
3180         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3181     } else if (strcmp(value, "enabled") == 0) {
3182         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3183     } else if (strcmp(value, "required") == 0) {
3184         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3185     } else {
3186         error_setg(errp, "Bad value for \"resize-hpt\" property");
3187     }
3188 }
3189 
3190 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3191 {
3192     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3193 
3194     if (spapr->irq == &spapr_irq_xics_legacy) {
3195         return g_strdup("legacy");
3196     } else if (spapr->irq == &spapr_irq_xics) {
3197         return g_strdup("xics");
3198     } else if (spapr->irq == &spapr_irq_xive) {
3199         return g_strdup("xive");
3200     } else if (spapr->irq == &spapr_irq_dual) {
3201         return g_strdup("dual");
3202     }
3203     g_assert_not_reached();
3204 }
3205 
3206 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3207 {
3208     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3209 
3210     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3211         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3212         return;
3213     }
3214 
3215     /* The legacy IRQ backend can not be set */
3216     if (strcmp(value, "xics") == 0) {
3217         spapr->irq = &spapr_irq_xics;
3218     } else if (strcmp(value, "xive") == 0) {
3219         spapr->irq = &spapr_irq_xive;
3220     } else if (strcmp(value, "dual") == 0) {
3221         spapr->irq = &spapr_irq_dual;
3222     } else {
3223         error_setg(errp, "Bad value for \"ic-mode\" property");
3224     }
3225 }
3226 
3227 static char *spapr_get_host_model(Object *obj, Error **errp)
3228 {
3229     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3230 
3231     return g_strdup(spapr->host_model);
3232 }
3233 
3234 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3235 {
3236     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3237 
3238     g_free(spapr->host_model);
3239     spapr->host_model = g_strdup(value);
3240 }
3241 
3242 static char *spapr_get_host_serial(Object *obj, Error **errp)
3243 {
3244     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3245 
3246     return g_strdup(spapr->host_serial);
3247 }
3248 
3249 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3250 {
3251     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3252 
3253     g_free(spapr->host_serial);
3254     spapr->host_serial = g_strdup(value);
3255 }
3256 
3257 static void spapr_instance_init(Object *obj)
3258 {
3259     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3260     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3261     MachineState *ms = MACHINE(spapr);
3262     MachineClass *mc = MACHINE_GET_CLASS(ms);
3263 
3264     /*
3265      * NVDIMM support went live in 5.1 without considering that, in
3266      * other archs, the user needs to enable NVDIMM support with the
3267      * 'nvdimm' machine option and the default behavior is NVDIMM
3268      * support disabled. It is too late to roll back to the standard
3269      * behavior without breaking 5.1 guests.
3270      */
3271     if (mc->nvdimm_supported) {
3272         ms->nvdimms_state->is_enabled = true;
3273     }
3274 
3275     spapr->htab_fd = -1;
3276     spapr->use_hotplug_event_source = true;
3277     spapr->kvm_type = g_strdup(DEFAULT_KVM_TYPE);
3278     object_property_add_str(obj, "kvm-type",
3279                             spapr_get_kvm_type, spapr_set_kvm_type);
3280     object_property_set_description(obj, "kvm-type",
3281                                     "Specifies the KVM virtualization mode (auto,"
3282                                     " hv, pr). Defaults to 'auto'. This mode will use"
3283                                     " any available KVM module loaded in the host,"
3284                                     " where kvm_hv takes precedence if both kvm_hv and"
3285                                     " kvm_pr are loaded.");
3286     object_property_add_bool(obj, "modern-hotplug-events",
3287                             spapr_get_modern_hotplug_events,
3288                             spapr_set_modern_hotplug_events);
3289     object_property_set_description(obj, "modern-hotplug-events",
3290                                     "Use dedicated hotplug event mechanism in"
3291                                     " place of standard EPOW events when possible"
3292                                     " (required for memory hot-unplug support)");
3293     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3294                             "Maximum permitted CPU compatibility mode");
3295 
3296     object_property_add_str(obj, "resize-hpt",
3297                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3298     object_property_set_description(obj, "resize-hpt",
3299                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3300     object_property_add_uint32_ptr(obj, "vsmt",
3301                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3302     object_property_set_description(obj, "vsmt",
3303                                     "Virtual SMT: KVM behaves as if this were"
3304                                     " the host's SMT mode");
3305 
3306     object_property_add_bool(obj, "vfio-no-msix-emulation",
3307                              spapr_get_msix_emulation, NULL);
3308 
3309     object_property_add_uint64_ptr(obj, "kernel-addr",
3310                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3311     object_property_set_description(obj, "kernel-addr",
3312                                     stringify(KERNEL_LOAD_ADDR)
3313                                     " for -kernel is the default");
3314     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3315     /* The machine class defines the default interrupt controller mode */
3316     spapr->irq = smc->irq;
3317     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3318                             spapr_set_ic_mode);
3319     object_property_set_description(obj, "ic-mode",
3320                  "Specifies the interrupt controller mode (xics, xive, dual)");
3321 
3322     object_property_add_str(obj, "host-model",
3323         spapr_get_host_model, spapr_set_host_model);
3324     object_property_set_description(obj, "host-model",
3325         "Host model to advertise in guest device tree");
3326     object_property_add_str(obj, "host-serial",
3327         spapr_get_host_serial, spapr_set_host_serial);
3328     object_property_set_description(obj, "host-serial",
3329         "Host serial number to advertise in guest device tree");
3330 }
3331 
3332 static void spapr_machine_finalizefn(Object *obj)
3333 {
3334     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3335 
3336     g_free(spapr->kvm_type);
3337 }
3338 
3339 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3340 {
3341     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3342     PowerPCCPU *cpu = POWERPC_CPU(cs);
3343     CPUPPCState *env = &cpu->env;
3344 
3345     cpu_synchronize_state(cs);
3346     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3347     if (spapr->fwnmi_system_reset_addr != -1) {
3348         uint64_t rtas_addr, addr;
3349 
3350         /* get rtas addr from fdt */
3351         rtas_addr = spapr_get_rtas_addr();
3352         if (!rtas_addr) {
3353             qemu_system_guest_panicked(NULL);
3354             return;
3355         }
3356 
3357         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3358         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3359         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3360         env->gpr[3] = addr;
3361     }
3362     ppc_cpu_do_system_reset(cs);
3363     if (spapr->fwnmi_system_reset_addr != -1) {
3364         env->nip = spapr->fwnmi_system_reset_addr;
3365     }
3366 }
3367 
3368 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3369 {
3370     CPUState *cs;
3371 
3372     CPU_FOREACH(cs) {
3373         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3374     }
3375 }
3376 
3377 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3378                           void *fdt, int *fdt_start_offset, Error **errp)
3379 {
3380     uint64_t addr;
3381     uint32_t node;
3382 
3383     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3384     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3385                                     &error_abort);
3386     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3387                                              SPAPR_MEMORY_BLOCK_SIZE);
3388     return 0;
3389 }
3390 
3391 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3392                            bool dedicated_hp_event_source)
3393 {
3394     SpaprDrc *drc;
3395     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3396     int i;
3397     uint64_t addr = addr_start;
3398     bool hotplugged = spapr_drc_hotplugged(dev);
3399 
3400     for (i = 0; i < nr_lmbs; i++) {
3401         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3402                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3403         g_assert(drc);
3404 
3405         /*
3406          * memory_device_get_free_addr() provided a range of free addresses
3407          * that doesn't overlap with any existing mapping at pre-plug. The
3408          * corresponding LMB DRCs are thus assumed to be all attachable.
3409          */
3410         spapr_drc_attach(drc, dev);
3411         if (!hotplugged) {
3412             spapr_drc_reset(drc);
3413         }
3414         addr += SPAPR_MEMORY_BLOCK_SIZE;
3415     }
3416     /* send hotplug notification to the
3417      * guest only in case of hotplugged memory
3418      */
3419     if (hotplugged) {
3420         if (dedicated_hp_event_source) {
3421             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3422                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3423             g_assert(drc);
3424             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3425                                                    nr_lmbs,
3426                                                    spapr_drc_index(drc));
3427         } else {
3428             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3429                                            nr_lmbs);
3430         }
3431     }
3432 }
3433 
3434 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3435 {
3436     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3437     PCDIMMDevice *dimm = PC_DIMM(dev);
3438     uint64_t size, addr;
3439     int64_t slot;
3440     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3441 
3442     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3443 
3444     pc_dimm_plug(dimm, MACHINE(ms));
3445 
3446     if (!is_nvdimm) {
3447         addr = object_property_get_uint(OBJECT(dimm),
3448                                         PC_DIMM_ADDR_PROP, &error_abort);
3449         spapr_add_lmbs(dev, addr, size,
3450                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT));
3451     } else {
3452         slot = object_property_get_int(OBJECT(dimm),
3453                                        PC_DIMM_SLOT_PROP, &error_abort);
3454         /* We should have valid slot number at this point */
3455         g_assert(slot >= 0);
3456         spapr_add_nvdimm(dev, slot);
3457     }
3458 }
3459 
3460 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3461                                   Error **errp)
3462 {
3463     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3464     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3465     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3466     PCDIMMDevice *dimm = PC_DIMM(dev);
3467     Error *local_err = NULL;
3468     uint64_t size;
3469     Object *memdev;
3470     hwaddr pagesize;
3471 
3472     if (!smc->dr_lmb_enabled) {
3473         error_setg(errp, "Memory hotplug not supported for this machine");
3474         return;
3475     }
3476 
3477     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3478     if (local_err) {
3479         error_propagate(errp, local_err);
3480         return;
3481     }
3482 
3483     if (is_nvdimm) {
3484         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3485             return;
3486         }
3487     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3488         error_setg(errp, "Hotplugged memory size must be a multiple of "
3489                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3490         return;
3491     }
3492 
3493     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3494                                       &error_abort);
3495     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3496     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3497         return;
3498     }
3499 
3500     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3501 }
3502 
3503 struct SpaprDimmState {
3504     PCDIMMDevice *dimm;
3505     uint32_t nr_lmbs;
3506     QTAILQ_ENTRY(SpaprDimmState) next;
3507 };
3508 
3509 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3510                                                        PCDIMMDevice *dimm)
3511 {
3512     SpaprDimmState *dimm_state = NULL;
3513 
3514     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3515         if (dimm_state->dimm == dimm) {
3516             break;
3517         }
3518     }
3519     return dimm_state;
3520 }
3521 
3522 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3523                                                       uint32_t nr_lmbs,
3524                                                       PCDIMMDevice *dimm)
3525 {
3526     SpaprDimmState *ds = NULL;
3527 
3528     /*
3529      * If this request is for a DIMM whose removal had failed earlier
3530      * (due to guest's refusal to remove the LMBs), we would have this
3531      * dimm already in the pending_dimm_unplugs list. In that
3532      * case don't add again.
3533      */
3534     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3535     if (!ds) {
3536         ds = g_malloc0(sizeof(SpaprDimmState));
3537         ds->nr_lmbs = nr_lmbs;
3538         ds->dimm = dimm;
3539         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3540     }
3541     return ds;
3542 }
3543 
3544 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3545                                               SpaprDimmState *dimm_state)
3546 {
3547     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3548     g_free(dimm_state);
3549 }
3550 
3551 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3552                                                         PCDIMMDevice *dimm)
3553 {
3554     SpaprDrc *drc;
3555     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3556                                                   &error_abort);
3557     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3558     uint32_t avail_lmbs = 0;
3559     uint64_t addr_start, addr;
3560     int i;
3561 
3562     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3563                                           &error_abort);
3564 
3565     addr = addr_start;
3566     for (i = 0; i < nr_lmbs; i++) {
3567         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3568                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3569         g_assert(drc);
3570         if (drc->dev) {
3571             avail_lmbs++;
3572         }
3573         addr += SPAPR_MEMORY_BLOCK_SIZE;
3574     }
3575 
3576     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3577 }
3578 
3579 void spapr_memory_unplug_rollback(SpaprMachineState *spapr, DeviceState *dev)
3580 {
3581     SpaprDimmState *ds;
3582     PCDIMMDevice *dimm;
3583     SpaprDrc *drc;
3584     uint32_t nr_lmbs;
3585     uint64_t size, addr_start, addr;
3586     g_autofree char *qapi_error = NULL;
3587     int i;
3588 
3589     if (!dev) {
3590         return;
3591     }
3592 
3593     dimm = PC_DIMM(dev);
3594     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3595 
3596     /*
3597      * 'ds == NULL' would mean that the DIMM doesn't have a pending
3598      * unplug state, but one of its DRC is marked as unplug_requested.
3599      * This is bad and weird enough to g_assert() out.
3600      */
3601     g_assert(ds);
3602 
3603     spapr_pending_dimm_unplugs_remove(spapr, ds);
3604 
3605     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3606     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3607 
3608     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3609                                           &error_abort);
3610 
3611     addr = addr_start;
3612     for (i = 0; i < nr_lmbs; i++) {
3613         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3614                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3615         g_assert(drc);
3616 
3617         drc->unplug_requested = false;
3618         addr += SPAPR_MEMORY_BLOCK_SIZE;
3619     }
3620 
3621     /*
3622      * Tell QAPI that something happened and the memory
3623      * hotunplug wasn't successful.
3624      */
3625     qapi_error = g_strdup_printf("Memory hotunplug rejected by the guest "
3626                                  "for device %s", dev->id);
3627     qapi_event_send_mem_unplug_error(dev->id, qapi_error);
3628 }
3629 
3630 /* Callback to be called during DRC release. */
3631 void spapr_lmb_release(DeviceState *dev)
3632 {
3633     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3634     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3635     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3636 
3637     /* This information will get lost if a migration occurs
3638      * during the unplug process. In this case recover it. */
3639     if (ds == NULL) {
3640         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3641         g_assert(ds);
3642         /* The DRC being examined by the caller at least must be counted */
3643         g_assert(ds->nr_lmbs);
3644     }
3645 
3646     if (--ds->nr_lmbs) {
3647         return;
3648     }
3649 
3650     /*
3651      * Now that all the LMBs have been removed by the guest, call the
3652      * unplug handler chain. This can never fail.
3653      */
3654     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3655     object_unparent(OBJECT(dev));
3656 }
3657 
3658 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3659 {
3660     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3661     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3662 
3663     /* We really shouldn't get this far without anything to unplug */
3664     g_assert(ds);
3665 
3666     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3667     qdev_unrealize(dev);
3668     spapr_pending_dimm_unplugs_remove(spapr, ds);
3669 }
3670 
3671 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3672                                         DeviceState *dev, Error **errp)
3673 {
3674     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3675     PCDIMMDevice *dimm = PC_DIMM(dev);
3676     uint32_t nr_lmbs;
3677     uint64_t size, addr_start, addr;
3678     int i;
3679     SpaprDrc *drc;
3680 
3681     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3682         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3683         return;
3684     }
3685 
3686     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3687     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3688 
3689     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3690                                           &error_abort);
3691 
3692     /*
3693      * An existing pending dimm state for this DIMM means that there is an
3694      * unplug operation in progress, waiting for the spapr_lmb_release
3695      * callback to complete the job (BQL can't cover that far). In this case,
3696      * bail out to avoid detaching DRCs that were already released.
3697      */
3698     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3699         error_setg(errp, "Memory unplug already in progress for device %s",
3700                    dev->id);
3701         return;
3702     }
3703 
3704     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3705 
3706     addr = addr_start;
3707     for (i = 0; i < nr_lmbs; i++) {
3708         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3709                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3710         g_assert(drc);
3711 
3712         spapr_drc_unplug_request(drc);
3713         addr += SPAPR_MEMORY_BLOCK_SIZE;
3714     }
3715 
3716     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3717                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3718     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3719                                               nr_lmbs, spapr_drc_index(drc));
3720 }
3721 
3722 /* Callback to be called during DRC release. */
3723 void spapr_core_release(DeviceState *dev)
3724 {
3725     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3726 
3727     /* Call the unplug handler chain. This can never fail. */
3728     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3729     object_unparent(OBJECT(dev));
3730 }
3731 
3732 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3733 {
3734     MachineState *ms = MACHINE(hotplug_dev);
3735     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3736     CPUCore *cc = CPU_CORE(dev);
3737     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3738 
3739     if (smc->pre_2_10_has_unused_icps) {
3740         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3741         int i;
3742 
3743         for (i = 0; i < cc->nr_threads; i++) {
3744             CPUState *cs = CPU(sc->threads[i]);
3745 
3746             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3747         }
3748     }
3749 
3750     assert(core_slot);
3751     core_slot->cpu = NULL;
3752     qdev_unrealize(dev);
3753 }
3754 
3755 static
3756 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3757                                Error **errp)
3758 {
3759     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3760     int index;
3761     SpaprDrc *drc;
3762     CPUCore *cc = CPU_CORE(dev);
3763 
3764     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3765         error_setg(errp, "Unable to find CPU core with core-id: %d",
3766                    cc->core_id);
3767         return;
3768     }
3769     if (index == 0) {
3770         error_setg(errp, "Boot CPU core may not be unplugged");
3771         return;
3772     }
3773 
3774     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3775                           spapr_vcpu_id(spapr, cc->core_id));
3776     g_assert(drc);
3777 
3778     if (!spapr_drc_unplug_requested(drc)) {
3779         spapr_drc_unplug_request(drc);
3780     }
3781 
3782     /*
3783      * spapr_hotplug_req_remove_by_index is left unguarded, out of the
3784      * "!spapr_drc_unplug_requested" check, to allow for multiple IRQ
3785      * pulses removing the same CPU. Otherwise, in an failed hotunplug
3786      * attempt (e.g. the kernel will refuse to remove the last online
3787      * CPU), we will never attempt it again because unplug_requested
3788      * will still be 'true' in that case.
3789      */
3790     spapr_hotplug_req_remove_by_index(drc);
3791 }
3792 
3793 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3794                            void *fdt, int *fdt_start_offset, Error **errp)
3795 {
3796     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3797     CPUState *cs = CPU(core->threads[0]);
3798     PowerPCCPU *cpu = POWERPC_CPU(cs);
3799     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3800     int id = spapr_get_vcpu_id(cpu);
3801     g_autofree char *nodename = NULL;
3802     int offset;
3803 
3804     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3805     offset = fdt_add_subnode(fdt, 0, nodename);
3806 
3807     spapr_dt_cpu(cs, fdt, offset, spapr);
3808 
3809     /*
3810      * spapr_dt_cpu() does not fill the 'name' property in the
3811      * CPU node. The function is called during boot process, before
3812      * and after CAS, and overwriting the 'name' property written
3813      * by SLOF is not allowed.
3814      *
3815      * Write it manually after spapr_dt_cpu(). This makes the hotplug
3816      * CPUs more compatible with the coldplugged ones, which have
3817      * the 'name' property. Linux Kernel also relies on this
3818      * property to identify CPU nodes.
3819      */
3820     _FDT((fdt_setprop_string(fdt, offset, "name", nodename)));
3821 
3822     *fdt_start_offset = offset;
3823     return 0;
3824 }
3825 
3826 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3827 {
3828     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3829     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3830     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3831     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3832     CPUCore *cc = CPU_CORE(dev);
3833     CPUState *cs;
3834     SpaprDrc *drc;
3835     CPUArchId *core_slot;
3836     int index;
3837     bool hotplugged = spapr_drc_hotplugged(dev);
3838     int i;
3839 
3840     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3841     g_assert(core_slot); /* Already checked in spapr_core_pre_plug() */
3842 
3843     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3844                           spapr_vcpu_id(spapr, cc->core_id));
3845 
3846     g_assert(drc || !mc->has_hotpluggable_cpus);
3847 
3848     if (drc) {
3849         /*
3850          * spapr_core_pre_plug() already buys us this is a brand new
3851          * core being plugged into a free slot. Nothing should already
3852          * be attached to the corresponding DRC.
3853          */
3854         spapr_drc_attach(drc, dev);
3855 
3856         if (hotplugged) {
3857             /*
3858              * Send hotplug notification interrupt to the guest only
3859              * in case of hotplugged CPUs.
3860              */
3861             spapr_hotplug_req_add_by_index(drc);
3862         } else {
3863             spapr_drc_reset(drc);
3864         }
3865     }
3866 
3867     core_slot->cpu = OBJECT(dev);
3868 
3869     /*
3870      * Set compatibility mode to match the boot CPU, which was either set
3871      * by the machine reset code or by CAS. This really shouldn't fail at
3872      * this point.
3873      */
3874     if (hotplugged) {
3875         for (i = 0; i < cc->nr_threads; i++) {
3876             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3877                            &error_abort);
3878         }
3879     }
3880 
3881     if (smc->pre_2_10_has_unused_icps) {
3882         for (i = 0; i < cc->nr_threads; i++) {
3883             cs = CPU(core->threads[i]);
3884             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3885         }
3886     }
3887 }
3888 
3889 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3890                                 Error **errp)
3891 {
3892     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3893     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3894     CPUCore *cc = CPU_CORE(dev);
3895     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3896     const char *type = object_get_typename(OBJECT(dev));
3897     CPUArchId *core_slot;
3898     int index;
3899     unsigned int smp_threads = machine->smp.threads;
3900 
3901     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3902         error_setg(errp, "CPU hotplug not supported for this machine");
3903         return;
3904     }
3905 
3906     if (strcmp(base_core_type, type)) {
3907         error_setg(errp, "CPU core type should be %s", base_core_type);
3908         return;
3909     }
3910 
3911     if (cc->core_id % smp_threads) {
3912         error_setg(errp, "invalid core id %d", cc->core_id);
3913         return;
3914     }
3915 
3916     /*
3917      * In general we should have homogeneous threads-per-core, but old
3918      * (pre hotplug support) machine types allow the last core to have
3919      * reduced threads as a compatibility hack for when we allowed
3920      * total vcpus not a multiple of threads-per-core.
3921      */
3922     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3923         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
3924                    smp_threads);
3925         return;
3926     }
3927 
3928     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3929     if (!core_slot) {
3930         error_setg(errp, "core id %d out of range", cc->core_id);
3931         return;
3932     }
3933 
3934     if (core_slot->cpu) {
3935         error_setg(errp, "core %d already populated", cc->core_id);
3936         return;
3937     }
3938 
3939     numa_cpu_pre_plug(core_slot, dev, errp);
3940 }
3941 
3942 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3943                           void *fdt, int *fdt_start_offset, Error **errp)
3944 {
3945     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3946     int intc_phandle;
3947 
3948     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3949     if (intc_phandle <= 0) {
3950         return -1;
3951     }
3952 
3953     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3954         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3955         return -1;
3956     }
3957 
3958     /* generally SLOF creates these, for hotplug it's up to QEMU */
3959     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3960 
3961     return 0;
3962 }
3963 
3964 static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3965                                Error **errp)
3966 {
3967     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3968     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3969     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3970     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3971     SpaprDrc *drc;
3972 
3973     if (dev->hotplugged && !smc->dr_phb_enabled) {
3974         error_setg(errp, "PHB hotplug not supported for this machine");
3975         return false;
3976     }
3977 
3978     if (sphb->index == (uint32_t)-1) {
3979         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3980         return false;
3981     }
3982 
3983     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3984     if (drc && drc->dev) {
3985         error_setg(errp, "PHB %d already attached", sphb->index);
3986         return false;
3987     }
3988 
3989     /*
3990      * This will check that sphb->index doesn't exceed the maximum number of
3991      * PHBs for the current machine type.
3992      */
3993     return
3994         smc->phb_placement(spapr, sphb->index,
3995                            &sphb->buid, &sphb->io_win_addr,
3996                            &sphb->mem_win_addr, &sphb->mem64_win_addr,
3997                            windows_supported, sphb->dma_liobn,
3998                            &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3999                            errp);
4000 }
4001 
4002 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4003 {
4004     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4005     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4006     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4007     SpaprDrc *drc;
4008     bool hotplugged = spapr_drc_hotplugged(dev);
4009 
4010     if (!smc->dr_phb_enabled) {
4011         return;
4012     }
4013 
4014     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4015     /* hotplug hooks should check it's enabled before getting this far */
4016     assert(drc);
4017 
4018     /* spapr_phb_pre_plug() already checked the DRC is attachable */
4019     spapr_drc_attach(drc, dev);
4020 
4021     if (hotplugged) {
4022         spapr_hotplug_req_add_by_index(drc);
4023     } else {
4024         spapr_drc_reset(drc);
4025     }
4026 }
4027 
4028 void spapr_phb_release(DeviceState *dev)
4029 {
4030     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4031 
4032     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4033     object_unparent(OBJECT(dev));
4034 }
4035 
4036 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4037 {
4038     qdev_unrealize(dev);
4039 }
4040 
4041 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4042                                      DeviceState *dev, Error **errp)
4043 {
4044     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4045     SpaprDrc *drc;
4046 
4047     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4048     assert(drc);
4049 
4050     if (!spapr_drc_unplug_requested(drc)) {
4051         spapr_drc_unplug_request(drc);
4052         spapr_hotplug_req_remove_by_index(drc);
4053     } else {
4054         error_setg(errp,
4055                    "PCI Host Bridge unplug already in progress for device %s",
4056                    dev->id);
4057     }
4058 }
4059 
4060 static
4061 bool spapr_tpm_proxy_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4062                               Error **errp)
4063 {
4064     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4065 
4066     if (spapr->tpm_proxy != NULL) {
4067         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4068         return false;
4069     }
4070 
4071     return true;
4072 }
4073 
4074 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4075 {
4076     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4077     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4078 
4079     /* Already checked in spapr_tpm_proxy_pre_plug() */
4080     g_assert(spapr->tpm_proxy == NULL);
4081 
4082     spapr->tpm_proxy = tpm_proxy;
4083 }
4084 
4085 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4086 {
4087     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4088 
4089     qdev_unrealize(dev);
4090     object_unparent(OBJECT(dev));
4091     spapr->tpm_proxy = NULL;
4092 }
4093 
4094 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4095                                       DeviceState *dev, Error **errp)
4096 {
4097     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4098         spapr_memory_plug(hotplug_dev, dev);
4099     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4100         spapr_core_plug(hotplug_dev, dev);
4101     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4102         spapr_phb_plug(hotplug_dev, dev);
4103     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4104         spapr_tpm_proxy_plug(hotplug_dev, dev);
4105     }
4106 }
4107 
4108 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4109                                         DeviceState *dev, Error **errp)
4110 {
4111     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4112         spapr_memory_unplug(hotplug_dev, dev);
4113     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4114         spapr_core_unplug(hotplug_dev, dev);
4115     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4116         spapr_phb_unplug(hotplug_dev, dev);
4117     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4118         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4119     }
4120 }
4121 
4122 bool spapr_memory_hot_unplug_supported(SpaprMachineState *spapr)
4123 {
4124     return spapr_ovec_test(spapr->ov5_cas, OV5_HP_EVT) ||
4125         /*
4126          * CAS will process all pending unplug requests.
4127          *
4128          * HACK: a guest could theoretically have cleared all bits in OV5,
4129          * but none of the guests we care for do.
4130          */
4131         spapr_ovec_empty(spapr->ov5_cas);
4132 }
4133 
4134 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4135                                                 DeviceState *dev, Error **errp)
4136 {
4137     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4138     MachineClass *mc = MACHINE_GET_CLASS(sms);
4139     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4140 
4141     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4142         if (spapr_memory_hot_unplug_supported(sms)) {
4143             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4144         } else {
4145             error_setg(errp, "Memory hot unplug not supported for this guest");
4146         }
4147     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4148         if (!mc->has_hotpluggable_cpus) {
4149             error_setg(errp, "CPU hot unplug not supported on this machine");
4150             return;
4151         }
4152         spapr_core_unplug_request(hotplug_dev, dev, errp);
4153     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4154         if (!smc->dr_phb_enabled) {
4155             error_setg(errp, "PHB hot unplug not supported on this machine");
4156             return;
4157         }
4158         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4159     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4160         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4161     }
4162 }
4163 
4164 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4165                                           DeviceState *dev, Error **errp)
4166 {
4167     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4168         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4169     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4170         spapr_core_pre_plug(hotplug_dev, dev, errp);
4171     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4172         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4173     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4174         spapr_tpm_proxy_pre_plug(hotplug_dev, dev, errp);
4175     }
4176 }
4177 
4178 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4179                                                  DeviceState *dev)
4180 {
4181     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4182         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4183         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4184         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4185         return HOTPLUG_HANDLER(machine);
4186     }
4187     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4188         PCIDevice *pcidev = PCI_DEVICE(dev);
4189         PCIBus *root = pci_device_root_bus(pcidev);
4190         SpaprPhbState *phb =
4191             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4192                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4193 
4194         if (phb) {
4195             return HOTPLUG_HANDLER(phb);
4196         }
4197     }
4198     return NULL;
4199 }
4200 
4201 static CpuInstanceProperties
4202 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4203 {
4204     CPUArchId *core_slot;
4205     MachineClass *mc = MACHINE_GET_CLASS(machine);
4206 
4207     /* make sure possible_cpu are intialized */
4208     mc->possible_cpu_arch_ids(machine);
4209     /* get CPU core slot containing thread that matches cpu_index */
4210     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4211     assert(core_slot);
4212     return core_slot->props;
4213 }
4214 
4215 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4216 {
4217     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4218 }
4219 
4220 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4221 {
4222     int i;
4223     unsigned int smp_threads = machine->smp.threads;
4224     unsigned int smp_cpus = machine->smp.cpus;
4225     const char *core_type;
4226     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4227     MachineClass *mc = MACHINE_GET_CLASS(machine);
4228 
4229     if (!mc->has_hotpluggable_cpus) {
4230         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4231     }
4232     if (machine->possible_cpus) {
4233         assert(machine->possible_cpus->len == spapr_max_cores);
4234         return machine->possible_cpus;
4235     }
4236 
4237     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4238     if (!core_type) {
4239         error_report("Unable to find sPAPR CPU Core definition");
4240         exit(1);
4241     }
4242 
4243     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4244                              sizeof(CPUArchId) * spapr_max_cores);
4245     machine->possible_cpus->len = spapr_max_cores;
4246     for (i = 0; i < machine->possible_cpus->len; i++) {
4247         int core_id = i * smp_threads;
4248 
4249         machine->possible_cpus->cpus[i].type = core_type;
4250         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4251         machine->possible_cpus->cpus[i].arch_id = core_id;
4252         machine->possible_cpus->cpus[i].props.has_core_id = true;
4253         machine->possible_cpus->cpus[i].props.core_id = core_id;
4254     }
4255     return machine->possible_cpus;
4256 }
4257 
4258 static bool spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4259                                 uint64_t *buid, hwaddr *pio,
4260                                 hwaddr *mmio32, hwaddr *mmio64,
4261                                 unsigned n_dma, uint32_t *liobns,
4262                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4263 {
4264     /*
4265      * New-style PHB window placement.
4266      *
4267      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4268      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4269      * windows.
4270      *
4271      * Some guest kernels can't work with MMIO windows above 1<<46
4272      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4273      *
4274      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4275      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4276      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4277      * 1TiB 64-bit MMIO windows for each PHB.
4278      */
4279     const uint64_t base_buid = 0x800000020000000ULL;
4280     int i;
4281 
4282     /* Sanity check natural alignments */
4283     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4284     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4285     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4286     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4287     /* Sanity check bounds */
4288     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4289                       SPAPR_PCI_MEM32_WIN_SIZE);
4290     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4291                       SPAPR_PCI_MEM64_WIN_SIZE);
4292 
4293     if (index >= SPAPR_MAX_PHBS) {
4294         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4295                    SPAPR_MAX_PHBS - 1);
4296         return false;
4297     }
4298 
4299     *buid = base_buid + index;
4300     for (i = 0; i < n_dma; ++i) {
4301         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4302     }
4303 
4304     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4305     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4306     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4307 
4308     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4309     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4310     return true;
4311 }
4312 
4313 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4314 {
4315     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4316 
4317     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4318 }
4319 
4320 static void spapr_ics_resend(XICSFabric *dev)
4321 {
4322     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4323 
4324     ics_resend(spapr->ics);
4325 }
4326 
4327 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4328 {
4329     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4330 
4331     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4332 }
4333 
4334 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4335                                  Monitor *mon)
4336 {
4337     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4338 
4339     spapr_irq_print_info(spapr, mon);
4340     monitor_printf(mon, "irqchip: %s\n",
4341                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4342 }
4343 
4344 /*
4345  * This is a XIVE only operation
4346  */
4347 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4348                            uint8_t nvt_blk, uint32_t nvt_idx,
4349                            bool cam_ignore, uint8_t priority,
4350                            uint32_t logic_serv, XiveTCTXMatch *match)
4351 {
4352     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4353     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4354     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4355     int count;
4356 
4357     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4358                            priority, logic_serv, match);
4359     if (count < 0) {
4360         return count;
4361     }
4362 
4363     /*
4364      * When we implement the save and restore of the thread interrupt
4365      * contexts in the enter/exit CPU handlers of the machine and the
4366      * escalations in QEMU, we should be able to handle non dispatched
4367      * vCPUs.
4368      *
4369      * Until this is done, the sPAPR machine should find at least one
4370      * matching context always.
4371      */
4372     if (count == 0) {
4373         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4374                       nvt_blk, nvt_idx);
4375     }
4376 
4377     return count;
4378 }
4379 
4380 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4381 {
4382     return cpu->vcpu_id;
4383 }
4384 
4385 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4386 {
4387     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4388     MachineState *ms = MACHINE(spapr);
4389     int vcpu_id;
4390 
4391     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4392 
4393     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4394         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4395         error_append_hint(errp, "Adjust the number of cpus to %d "
4396                           "or try to raise the number of threads per core\n",
4397                           vcpu_id * ms->smp.threads / spapr->vsmt);
4398         return false;
4399     }
4400 
4401     cpu->vcpu_id = vcpu_id;
4402     return true;
4403 }
4404 
4405 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4406 {
4407     CPUState *cs;
4408 
4409     CPU_FOREACH(cs) {
4410         PowerPCCPU *cpu = POWERPC_CPU(cs);
4411 
4412         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4413             return cpu;
4414         }
4415     }
4416 
4417     return NULL;
4418 }
4419 
4420 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4421 {
4422     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4423 
4424     /* These are only called by TCG, KVM maintains dispatch state */
4425 
4426     spapr_cpu->prod = false;
4427     if (spapr_cpu->vpa_addr) {
4428         CPUState *cs = CPU(cpu);
4429         uint32_t dispatch;
4430 
4431         dispatch = ldl_be_phys(cs->as,
4432                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4433         dispatch++;
4434         if ((dispatch & 1) != 0) {
4435             qemu_log_mask(LOG_GUEST_ERROR,
4436                           "VPA: incorrect dispatch counter value for "
4437                           "dispatched partition %u, correcting.\n", dispatch);
4438             dispatch++;
4439         }
4440         stl_be_phys(cs->as,
4441                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4442     }
4443 }
4444 
4445 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4446 {
4447     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4448 
4449     if (spapr_cpu->vpa_addr) {
4450         CPUState *cs = CPU(cpu);
4451         uint32_t dispatch;
4452 
4453         dispatch = ldl_be_phys(cs->as,
4454                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4455         dispatch++;
4456         if ((dispatch & 1) != 1) {
4457             qemu_log_mask(LOG_GUEST_ERROR,
4458                           "VPA: incorrect dispatch counter value for "
4459                           "preempted partition %u, correcting.\n", dispatch);
4460             dispatch++;
4461         }
4462         stl_be_phys(cs->as,
4463                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4464     }
4465 }
4466 
4467 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4468 {
4469     MachineClass *mc = MACHINE_CLASS(oc);
4470     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4471     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4472     NMIClass *nc = NMI_CLASS(oc);
4473     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4474     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4475     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4476     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4477     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4478 
4479     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4480     mc->ignore_boot_device_suffixes = true;
4481 
4482     /*
4483      * We set up the default / latest behaviour here.  The class_init
4484      * functions for the specific versioned machine types can override
4485      * these details for backwards compatibility
4486      */
4487     mc->init = spapr_machine_init;
4488     mc->reset = spapr_machine_reset;
4489     mc->block_default_type = IF_SCSI;
4490     mc->max_cpus = 1024;
4491     mc->no_parallel = 1;
4492     mc->default_boot_order = "";
4493     mc->default_ram_size = 512 * MiB;
4494     mc->default_ram_id = "ppc_spapr.ram";
4495     mc->default_display = "std";
4496     mc->kvm_type = spapr_kvm_type;
4497     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4498     mc->pci_allow_0_address = true;
4499     assert(!mc->get_hotplug_handler);
4500     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4501     hc->pre_plug = spapr_machine_device_pre_plug;
4502     hc->plug = spapr_machine_device_plug;
4503     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4504     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4505     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4506     hc->unplug_request = spapr_machine_device_unplug_request;
4507     hc->unplug = spapr_machine_device_unplug;
4508 
4509     smc->dr_lmb_enabled = true;
4510     smc->update_dt_enabled = true;
4511     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4512     mc->has_hotpluggable_cpus = true;
4513     mc->nvdimm_supported = true;
4514     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4515     fwc->get_dev_path = spapr_get_fw_dev_path;
4516     nc->nmi_monitor_handler = spapr_nmi;
4517     smc->phb_placement = spapr_phb_placement;
4518     vhc->hypercall = emulate_spapr_hypercall;
4519     vhc->hpt_mask = spapr_hpt_mask;
4520     vhc->map_hptes = spapr_map_hptes;
4521     vhc->unmap_hptes = spapr_unmap_hptes;
4522     vhc->hpte_set_c = spapr_hpte_set_c;
4523     vhc->hpte_set_r = spapr_hpte_set_r;
4524     vhc->get_pate = spapr_get_pate;
4525     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4526     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4527     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4528     xic->ics_get = spapr_ics_get;
4529     xic->ics_resend = spapr_ics_resend;
4530     xic->icp_get = spapr_icp_get;
4531     ispc->print_info = spapr_pic_print_info;
4532     /* Force NUMA node memory size to be a multiple of
4533      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4534      * in which LMBs are represented and hot-added
4535      */
4536     mc->numa_mem_align_shift = 28;
4537     mc->auto_enable_numa = true;
4538 
4539     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4540     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4541     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4542     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4543     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4544     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4545     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4546     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4547     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4548     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4549     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4550     spapr_caps_add_properties(smc);
4551     smc->irq = &spapr_irq_dual;
4552     smc->dr_phb_enabled = true;
4553     smc->linux_pci_probe = true;
4554     smc->smp_threads_vsmt = true;
4555     smc->nr_xirqs = SPAPR_NR_XIRQS;
4556     xfc->match_nvt = spapr_match_nvt;
4557 }
4558 
4559 static const TypeInfo spapr_machine_info = {
4560     .name          = TYPE_SPAPR_MACHINE,
4561     .parent        = TYPE_MACHINE,
4562     .abstract      = true,
4563     .instance_size = sizeof(SpaprMachineState),
4564     .instance_init = spapr_instance_init,
4565     .instance_finalize = spapr_machine_finalizefn,
4566     .class_size    = sizeof(SpaprMachineClass),
4567     .class_init    = spapr_machine_class_init,
4568     .interfaces = (InterfaceInfo[]) {
4569         { TYPE_FW_PATH_PROVIDER },
4570         { TYPE_NMI },
4571         { TYPE_HOTPLUG_HANDLER },
4572         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4573         { TYPE_XICS_FABRIC },
4574         { TYPE_INTERRUPT_STATS_PROVIDER },
4575         { TYPE_XIVE_FABRIC },
4576         { }
4577     },
4578 };
4579 
4580 static void spapr_machine_latest_class_options(MachineClass *mc)
4581 {
4582     mc->alias = "pseries";
4583     mc->is_default = true;
4584 }
4585 
4586 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4587     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4588                                                     void *data)      \
4589     {                                                                \
4590         MachineClass *mc = MACHINE_CLASS(oc);                        \
4591         spapr_machine_##suffix##_class_options(mc);                  \
4592         if (latest) {                                                \
4593             spapr_machine_latest_class_options(mc);                  \
4594         }                                                            \
4595     }                                                                \
4596     static const TypeInfo spapr_machine_##suffix##_info = {          \
4597         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4598         .parent = TYPE_SPAPR_MACHINE,                                \
4599         .class_init = spapr_machine_##suffix##_class_init,           \
4600     };                                                               \
4601     static void spapr_machine_register_##suffix(void)                \
4602     {                                                                \
4603         type_register(&spapr_machine_##suffix##_info);               \
4604     }                                                                \
4605     type_init(spapr_machine_register_##suffix)
4606 
4607 /*
4608  * pseries-6.1
4609  */
4610 static void spapr_machine_6_1_class_options(MachineClass *mc)
4611 {
4612     /* Defaults for the latest behaviour inherited from the base class */
4613 }
4614 
4615 DEFINE_SPAPR_MACHINE(6_1, "6.1", true);
4616 
4617 /*
4618  * pseries-6.0
4619  */
4620 static void spapr_machine_6_0_class_options(MachineClass *mc)
4621 {
4622     spapr_machine_6_1_class_options(mc);
4623     compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len);
4624 }
4625 
4626 DEFINE_SPAPR_MACHINE(6_0, "6.0", false);
4627 
4628 /*
4629  * pseries-5.2
4630  */
4631 static void spapr_machine_5_2_class_options(MachineClass *mc)
4632 {
4633     spapr_machine_6_0_class_options(mc);
4634     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
4635 }
4636 
4637 DEFINE_SPAPR_MACHINE(5_2, "5.2", false);
4638 
4639 /*
4640  * pseries-5.1
4641  */
4642 static void spapr_machine_5_1_class_options(MachineClass *mc)
4643 {
4644     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4645 
4646     spapr_machine_5_2_class_options(mc);
4647     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4648     smc->pre_5_2_numa_associativity = true;
4649 }
4650 
4651 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4652 
4653 /*
4654  * pseries-5.0
4655  */
4656 static void spapr_machine_5_0_class_options(MachineClass *mc)
4657 {
4658     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4659     static GlobalProperty compat[] = {
4660         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4661     };
4662 
4663     spapr_machine_5_1_class_options(mc);
4664     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4665     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4666     mc->numa_mem_supported = true;
4667     smc->pre_5_1_assoc_refpoints = true;
4668 }
4669 
4670 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4671 
4672 /*
4673  * pseries-4.2
4674  */
4675 static void spapr_machine_4_2_class_options(MachineClass *mc)
4676 {
4677     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4678 
4679     spapr_machine_5_0_class_options(mc);
4680     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4681     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4682     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4683     smc->rma_limit = 16 * GiB;
4684     mc->nvdimm_supported = false;
4685 }
4686 
4687 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4688 
4689 /*
4690  * pseries-4.1
4691  */
4692 static void spapr_machine_4_1_class_options(MachineClass *mc)
4693 {
4694     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4695     static GlobalProperty compat[] = {
4696         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4697         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4698     };
4699 
4700     spapr_machine_4_2_class_options(mc);
4701     smc->linux_pci_probe = false;
4702     smc->smp_threads_vsmt = false;
4703     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4704     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4705 }
4706 
4707 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4708 
4709 /*
4710  * pseries-4.0
4711  */
4712 static bool phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4713                               uint64_t *buid, hwaddr *pio,
4714                               hwaddr *mmio32, hwaddr *mmio64,
4715                               unsigned n_dma, uint32_t *liobns,
4716                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4717 {
4718     if (!spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma,
4719                              liobns, nv2gpa, nv2atsd, errp)) {
4720         return false;
4721     }
4722 
4723     *nv2gpa = 0;
4724     *nv2atsd = 0;
4725     return true;
4726 }
4727 static void spapr_machine_4_0_class_options(MachineClass *mc)
4728 {
4729     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4730 
4731     spapr_machine_4_1_class_options(mc);
4732     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4733     smc->phb_placement = phb_placement_4_0;
4734     smc->irq = &spapr_irq_xics;
4735     smc->pre_4_1_migration = true;
4736 }
4737 
4738 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4739 
4740 /*
4741  * pseries-3.1
4742  */
4743 static void spapr_machine_3_1_class_options(MachineClass *mc)
4744 {
4745     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4746 
4747     spapr_machine_4_0_class_options(mc);
4748     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4749 
4750     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4751     smc->update_dt_enabled = false;
4752     smc->dr_phb_enabled = false;
4753     smc->broken_host_serial_model = true;
4754     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4755     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4756     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4757     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4758 }
4759 
4760 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4761 
4762 /*
4763  * pseries-3.0
4764  */
4765 
4766 static void spapr_machine_3_0_class_options(MachineClass *mc)
4767 {
4768     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4769 
4770     spapr_machine_3_1_class_options(mc);
4771     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4772 
4773     smc->legacy_irq_allocation = true;
4774     smc->nr_xirqs = 0x400;
4775     smc->irq = &spapr_irq_xics_legacy;
4776 }
4777 
4778 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4779 
4780 /*
4781  * pseries-2.12
4782  */
4783 static void spapr_machine_2_12_class_options(MachineClass *mc)
4784 {
4785     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4786     static GlobalProperty compat[] = {
4787         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4788         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4789     };
4790 
4791     spapr_machine_3_0_class_options(mc);
4792     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4793     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4794 
4795     /* We depend on kvm_enabled() to choose a default value for the
4796      * hpt-max-page-size capability. Of course we can't do it here
4797      * because this is too early and the HW accelerator isn't initialzed
4798      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4799      */
4800     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4801 }
4802 
4803 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4804 
4805 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4806 {
4807     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4808 
4809     spapr_machine_2_12_class_options(mc);
4810     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4811     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4812     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4813 }
4814 
4815 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4816 
4817 /*
4818  * pseries-2.11
4819  */
4820 
4821 static void spapr_machine_2_11_class_options(MachineClass *mc)
4822 {
4823     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4824 
4825     spapr_machine_2_12_class_options(mc);
4826     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4827     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4828 }
4829 
4830 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4831 
4832 /*
4833  * pseries-2.10
4834  */
4835 
4836 static void spapr_machine_2_10_class_options(MachineClass *mc)
4837 {
4838     spapr_machine_2_11_class_options(mc);
4839     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4840 }
4841 
4842 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4843 
4844 /*
4845  * pseries-2.9
4846  */
4847 
4848 static void spapr_machine_2_9_class_options(MachineClass *mc)
4849 {
4850     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4851     static GlobalProperty compat[] = {
4852         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4853     };
4854 
4855     spapr_machine_2_10_class_options(mc);
4856     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4857     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4858     smc->pre_2_10_has_unused_icps = true;
4859     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4860 }
4861 
4862 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4863 
4864 /*
4865  * pseries-2.8
4866  */
4867 
4868 static void spapr_machine_2_8_class_options(MachineClass *mc)
4869 {
4870     static GlobalProperty compat[] = {
4871         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4872     };
4873 
4874     spapr_machine_2_9_class_options(mc);
4875     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4876     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4877     mc->numa_mem_align_shift = 23;
4878 }
4879 
4880 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4881 
4882 /*
4883  * pseries-2.7
4884  */
4885 
4886 static bool phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4887                               uint64_t *buid, hwaddr *pio,
4888                               hwaddr *mmio32, hwaddr *mmio64,
4889                               unsigned n_dma, uint32_t *liobns,
4890                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4891 {
4892     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4893     const uint64_t base_buid = 0x800000020000000ULL;
4894     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4895     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4896     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4897     const uint32_t max_index = 255;
4898     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4899 
4900     uint64_t ram_top = MACHINE(spapr)->ram_size;
4901     hwaddr phb0_base, phb_base;
4902     int i;
4903 
4904     /* Do we have device memory? */
4905     if (MACHINE(spapr)->maxram_size > ram_top) {
4906         /* Can't just use maxram_size, because there may be an
4907          * alignment gap between normal and device memory regions
4908          */
4909         ram_top = MACHINE(spapr)->device_memory->base +
4910             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4911     }
4912 
4913     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4914 
4915     if (index > max_index) {
4916         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4917                    max_index);
4918         return false;
4919     }
4920 
4921     *buid = base_buid + index;
4922     for (i = 0; i < n_dma; ++i) {
4923         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4924     }
4925 
4926     phb_base = phb0_base + index * phb_spacing;
4927     *pio = phb_base + pio_offset;
4928     *mmio32 = phb_base + mmio_offset;
4929     /*
4930      * We don't set the 64-bit MMIO window, relying on the PHB's
4931      * fallback behaviour of automatically splitting a large "32-bit"
4932      * window into contiguous 32-bit and 64-bit windows
4933      */
4934 
4935     *nv2gpa = 0;
4936     *nv2atsd = 0;
4937     return true;
4938 }
4939 
4940 static void spapr_machine_2_7_class_options(MachineClass *mc)
4941 {
4942     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4943     static GlobalProperty compat[] = {
4944         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4945         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4946         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4947         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4948     };
4949 
4950     spapr_machine_2_8_class_options(mc);
4951     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4952     mc->default_machine_opts = "modern-hotplug-events=off";
4953     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4954     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4955     smc->phb_placement = phb_placement_2_7;
4956 }
4957 
4958 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4959 
4960 /*
4961  * pseries-2.6
4962  */
4963 
4964 static void spapr_machine_2_6_class_options(MachineClass *mc)
4965 {
4966     static GlobalProperty compat[] = {
4967         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4968     };
4969 
4970     spapr_machine_2_7_class_options(mc);
4971     mc->has_hotpluggable_cpus = false;
4972     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4973     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4974 }
4975 
4976 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4977 
4978 /*
4979  * pseries-2.5
4980  */
4981 
4982 static void spapr_machine_2_5_class_options(MachineClass *mc)
4983 {
4984     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4985     static GlobalProperty compat[] = {
4986         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4987     };
4988 
4989     spapr_machine_2_6_class_options(mc);
4990     smc->use_ohci_by_default = true;
4991     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4992     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4993 }
4994 
4995 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4996 
4997 /*
4998  * pseries-2.4
4999  */
5000 
5001 static void spapr_machine_2_4_class_options(MachineClass *mc)
5002 {
5003     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5004 
5005     spapr_machine_2_5_class_options(mc);
5006     smc->dr_lmb_enabled = false;
5007     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
5008 }
5009 
5010 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
5011 
5012 /*
5013  * pseries-2.3
5014  */
5015 
5016 static void spapr_machine_2_3_class_options(MachineClass *mc)
5017 {
5018     static GlobalProperty compat[] = {
5019         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
5020     };
5021     spapr_machine_2_4_class_options(mc);
5022     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
5023     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5024 }
5025 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
5026 
5027 /*
5028  * pseries-2.2
5029  */
5030 
5031 static void spapr_machine_2_2_class_options(MachineClass *mc)
5032 {
5033     static GlobalProperty compat[] = {
5034         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
5035     };
5036 
5037     spapr_machine_2_3_class_options(mc);
5038     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
5039     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5040     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
5041 }
5042 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
5043 
5044 /*
5045  * pseries-2.1
5046  */
5047 
5048 static void spapr_machine_2_1_class_options(MachineClass *mc)
5049 {
5050     spapr_machine_2_2_class_options(mc);
5051     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
5052 }
5053 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
5054 
5055 static void spapr_machine_register_types(void)
5056 {
5057     type_register_static(&spapr_machine_info);
5058 }
5059 
5060 type_init(spapr_machine_register_types)
5061