xref: /openbmc/qemu/hw/ppc/spapr.c (revision 65a117da)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  */
26 
27 #include "qemu/osdep.h"
28 #include "qemu-common.h"
29 #include "qapi/error.h"
30 #include "qapi/visitor.h"
31 #include "sysemu/sysemu.h"
32 #include "sysemu/hostmem.h"
33 #include "sysemu/numa.h"
34 #include "sysemu/qtest.h"
35 #include "sysemu/reset.h"
36 #include "sysemu/runstate.h"
37 #include "qemu/log.h"
38 #include "hw/fw-path-provider.h"
39 #include "elf.h"
40 #include "net/net.h"
41 #include "sysemu/device_tree.h"
42 #include "sysemu/cpus.h"
43 #include "sysemu/hw_accel.h"
44 #include "kvm_ppc.h"
45 #include "migration/misc.h"
46 #include "migration/qemu-file-types.h"
47 #include "migration/global_state.h"
48 #include "migration/register.h"
49 #include "migration/blocker.h"
50 #include "mmu-hash64.h"
51 #include "mmu-book3s-v3.h"
52 #include "cpu-models.h"
53 #include "hw/core/cpu.h"
54 
55 #include "hw/boards.h"
56 #include "hw/ppc/ppc.h"
57 #include "hw/loader.h"
58 
59 #include "hw/ppc/fdt.h"
60 #include "hw/ppc/spapr.h"
61 #include "hw/ppc/spapr_vio.h"
62 #include "hw/qdev-properties.h"
63 #include "hw/pci-host/spapr.h"
64 #include "hw/pci/msi.h"
65 
66 #include "hw/pci/pci.h"
67 #include "hw/scsi/scsi.h"
68 #include "hw/virtio/virtio-scsi.h"
69 #include "hw/virtio/vhost-scsi-common.h"
70 
71 #include "exec/address-spaces.h"
72 #include "exec/ram_addr.h"
73 #include "hw/usb.h"
74 #include "qemu/config-file.h"
75 #include "qemu/error-report.h"
76 #include "trace.h"
77 #include "hw/nmi.h"
78 #include "hw/intc/intc.h"
79 
80 #include "hw/ppc/spapr_cpu_core.h"
81 #include "hw/mem/memory-device.h"
82 #include "hw/ppc/spapr_tpm_proxy.h"
83 #include "hw/ppc/spapr_nvdimm.h"
84 
85 #include "monitor/monitor.h"
86 
87 #include <libfdt.h>
88 
89 /* SLOF memory layout:
90  *
91  * SLOF raw image loaded at 0, copies its romfs right below the flat
92  * device-tree, then position SLOF itself 31M below that
93  *
94  * So we set FW_OVERHEAD to 40MB which should account for all of that
95  * and more
96  *
97  * We load our kernel at 4M, leaving space for SLOF initial image
98  */
99 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
100 #define FW_MAX_SIZE             0x400000
101 #define FW_FILE_NAME            "slof.bin"
102 #define FW_OVERHEAD             0x2800000
103 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
104 
105 #define MIN_RMA_SLOF            (128 * MiB)
106 
107 #define PHANDLE_INTC            0x00001111
108 
109 /* These two functions implement the VCPU id numbering: one to compute them
110  * all and one to identify thread 0 of a VCORE. Any change to the first one
111  * is likely to have an impact on the second one, so let's keep them close.
112  */
113 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
114 {
115     MachineState *ms = MACHINE(spapr);
116     unsigned int smp_threads = ms->smp.threads;
117 
118     assert(spapr->vsmt);
119     return
120         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
121 }
122 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
123                                       PowerPCCPU *cpu)
124 {
125     assert(spapr->vsmt);
126     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
127 }
128 
129 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
130 {
131     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
132      * and newer QEMUs don't even have them. In both cases, we don't want
133      * to send anything on the wire.
134      */
135     return false;
136 }
137 
138 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
139     .name = "icp/server",
140     .version_id = 1,
141     .minimum_version_id = 1,
142     .needed = pre_2_10_vmstate_dummy_icp_needed,
143     .fields = (VMStateField[]) {
144         VMSTATE_UNUSED(4), /* uint32_t xirr */
145         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
146         VMSTATE_UNUSED(1), /* uint8_t mfrr */
147         VMSTATE_END_OF_LIST()
148     },
149 };
150 
151 static void pre_2_10_vmstate_register_dummy_icp(int i)
152 {
153     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
154                      (void *)(uintptr_t) i);
155 }
156 
157 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
158 {
159     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
160                        (void *)(uintptr_t) i);
161 }
162 
163 int spapr_max_server_number(SpaprMachineState *spapr)
164 {
165     MachineState *ms = MACHINE(spapr);
166 
167     assert(spapr->vsmt);
168     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
169 }
170 
171 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
172                                   int smt_threads)
173 {
174     int i, ret = 0;
175     uint32_t servers_prop[smt_threads];
176     uint32_t gservers_prop[smt_threads * 2];
177     int index = spapr_get_vcpu_id(cpu);
178 
179     if (cpu->compat_pvr) {
180         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
181         if (ret < 0) {
182             return ret;
183         }
184     }
185 
186     /* Build interrupt servers and gservers properties */
187     for (i = 0; i < smt_threads; i++) {
188         servers_prop[i] = cpu_to_be32(index + i);
189         /* Hack, direct the group queues back to cpu 0 */
190         gservers_prop[i*2] = cpu_to_be32(index + i);
191         gservers_prop[i*2 + 1] = 0;
192     }
193     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
194                       servers_prop, sizeof(servers_prop));
195     if (ret < 0) {
196         return ret;
197     }
198     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
199                       gservers_prop, sizeof(gservers_prop));
200 
201     return ret;
202 }
203 
204 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
205 {
206     int index = spapr_get_vcpu_id(cpu);
207     uint32_t associativity[] = {cpu_to_be32(0x5),
208                                 cpu_to_be32(0x0),
209                                 cpu_to_be32(0x0),
210                                 cpu_to_be32(0x0),
211                                 cpu_to_be32(cpu->node_id),
212                                 cpu_to_be32(index)};
213 
214     /* Advertise NUMA via ibm,associativity */
215     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
216                           sizeof(associativity));
217 }
218 
219 static void spapr_dt_pa_features(SpaprMachineState *spapr,
220                                  PowerPCCPU *cpu,
221                                  void *fdt, int offset)
222 {
223     uint8_t pa_features_206[] = { 6, 0,
224         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
225     uint8_t pa_features_207[] = { 24, 0,
226         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
227         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
228         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
229         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
230     uint8_t pa_features_300[] = { 66, 0,
231         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
232         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
233         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
234         /* 6: DS207 */
235         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
236         /* 16: Vector */
237         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
238         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
239         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
240         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
241         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
242         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
243         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
244         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
245         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
246         /* 42: PM, 44: PC RA, 46: SC vec'd */
247         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
248         /* 48: SIMD, 50: QP BFP, 52: String */
249         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
250         /* 54: DecFP, 56: DecI, 58: SHA */
251         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
252         /* 60: NM atomic, 62: RNG */
253         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
254     };
255     uint8_t *pa_features = NULL;
256     size_t pa_size;
257 
258     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
259         pa_features = pa_features_206;
260         pa_size = sizeof(pa_features_206);
261     }
262     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
263         pa_features = pa_features_207;
264         pa_size = sizeof(pa_features_207);
265     }
266     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
267         pa_features = pa_features_300;
268         pa_size = sizeof(pa_features_300);
269     }
270     if (!pa_features) {
271         return;
272     }
273 
274     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
275         /*
276          * Note: we keep CI large pages off by default because a 64K capable
277          * guest provisioned with large pages might otherwise try to map a qemu
278          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
279          * even if that qemu runs on a 4k host.
280          * We dd this bit back here if we are confident this is not an issue
281          */
282         pa_features[3] |= 0x20;
283     }
284     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
285         pa_features[24] |= 0x80;    /* Transactional memory support */
286     }
287     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
288         /* Workaround for broken kernels that attempt (guest) radix
289          * mode when they can't handle it, if they see the radix bit set
290          * in pa-features. So hide it from them. */
291         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
292     }
293 
294     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
295 }
296 
297 static hwaddr spapr_node0_size(MachineState *machine)
298 {
299     if (machine->numa_state->num_nodes) {
300         int i;
301         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
302             if (machine->numa_state->nodes[i].node_mem) {
303                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
304                            machine->ram_size);
305             }
306         }
307     }
308     return machine->ram_size;
309 }
310 
311 static void add_str(GString *s, const gchar *s1)
312 {
313     g_string_append_len(s, s1, strlen(s1) + 1);
314 }
315 
316 static int spapr_dt_memory_node(void *fdt, int nodeid, hwaddr start,
317                                 hwaddr size)
318 {
319     uint32_t associativity[] = {
320         cpu_to_be32(0x4), /* length */
321         cpu_to_be32(0x0), cpu_to_be32(0x0),
322         cpu_to_be32(0x0), cpu_to_be32(nodeid)
323     };
324     char mem_name[32];
325     uint64_t mem_reg_property[2];
326     int off;
327 
328     mem_reg_property[0] = cpu_to_be64(start);
329     mem_reg_property[1] = cpu_to_be64(size);
330 
331     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
332     off = fdt_add_subnode(fdt, 0, mem_name);
333     _FDT(off);
334     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
335     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
336                       sizeof(mem_reg_property))));
337     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
338                       sizeof(associativity))));
339     return off;
340 }
341 
342 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
343 {
344     MemoryDeviceInfoList *info;
345 
346     for (info = list; info; info = info->next) {
347         MemoryDeviceInfo *value = info->value;
348 
349         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
350             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
351 
352             if (addr >= pcdimm_info->addr &&
353                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
354                 return pcdimm_info->node;
355             }
356         }
357     }
358 
359     return -1;
360 }
361 
362 struct sPAPRDrconfCellV2 {
363      uint32_t seq_lmbs;
364      uint64_t base_addr;
365      uint32_t drc_index;
366      uint32_t aa_index;
367      uint32_t flags;
368 } QEMU_PACKED;
369 
370 typedef struct DrconfCellQueue {
371     struct sPAPRDrconfCellV2 cell;
372     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
373 } DrconfCellQueue;
374 
375 static DrconfCellQueue *
376 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
377                       uint32_t drc_index, uint32_t aa_index,
378                       uint32_t flags)
379 {
380     DrconfCellQueue *elem;
381 
382     elem = g_malloc0(sizeof(*elem));
383     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
384     elem->cell.base_addr = cpu_to_be64(base_addr);
385     elem->cell.drc_index = cpu_to_be32(drc_index);
386     elem->cell.aa_index = cpu_to_be32(aa_index);
387     elem->cell.flags = cpu_to_be32(flags);
388 
389     return elem;
390 }
391 
392 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
393                                       int offset, MemoryDeviceInfoList *dimms)
394 {
395     MachineState *machine = MACHINE(spapr);
396     uint8_t *int_buf, *cur_index;
397     int ret;
398     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
399     uint64_t addr, cur_addr, size;
400     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
401     uint64_t mem_end = machine->device_memory->base +
402                        memory_region_size(&machine->device_memory->mr);
403     uint32_t node, buf_len, nr_entries = 0;
404     SpaprDrc *drc;
405     DrconfCellQueue *elem, *next;
406     MemoryDeviceInfoList *info;
407     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
408         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
409 
410     /* Entry to cover RAM and the gap area */
411     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
412                                  SPAPR_LMB_FLAGS_RESERVED |
413                                  SPAPR_LMB_FLAGS_DRC_INVALID);
414     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
415     nr_entries++;
416 
417     cur_addr = machine->device_memory->base;
418     for (info = dimms; info; info = info->next) {
419         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
420 
421         addr = di->addr;
422         size = di->size;
423         node = di->node;
424 
425         /*
426          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
427          * area is marked hotpluggable in the next iteration for the bigger
428          * chunk including the NVDIMM occupied area.
429          */
430         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
431             continue;
432 
433         /* Entry for hot-pluggable area */
434         if (cur_addr < addr) {
435             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
436             g_assert(drc);
437             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
438                                          cur_addr, spapr_drc_index(drc), -1, 0);
439             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
440             nr_entries++;
441         }
442 
443         /* Entry for DIMM */
444         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
445         g_assert(drc);
446         elem = spapr_get_drconf_cell(size / lmb_size, addr,
447                                      spapr_drc_index(drc), node,
448                                      (SPAPR_LMB_FLAGS_ASSIGNED |
449                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
450         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
451         nr_entries++;
452         cur_addr = addr + size;
453     }
454 
455     /* Entry for remaining hotpluggable area */
456     if (cur_addr < mem_end) {
457         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
458         g_assert(drc);
459         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
460                                      cur_addr, spapr_drc_index(drc), -1, 0);
461         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
462         nr_entries++;
463     }
464 
465     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
466     int_buf = cur_index = g_malloc0(buf_len);
467     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
468     cur_index += sizeof(nr_entries);
469 
470     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
471         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
472         cur_index += sizeof(elem->cell);
473         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
474         g_free(elem);
475     }
476 
477     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
478     g_free(int_buf);
479     if (ret < 0) {
480         return -1;
481     }
482     return 0;
483 }
484 
485 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
486                                    int offset, MemoryDeviceInfoList *dimms)
487 {
488     MachineState *machine = MACHINE(spapr);
489     int i, ret;
490     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
491     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
492     uint32_t nr_lmbs = (machine->device_memory->base +
493                        memory_region_size(&machine->device_memory->mr)) /
494                        lmb_size;
495     uint32_t *int_buf, *cur_index, buf_len;
496 
497     /*
498      * Allocate enough buffer size to fit in ibm,dynamic-memory
499      */
500     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
501     cur_index = int_buf = g_malloc0(buf_len);
502     int_buf[0] = cpu_to_be32(nr_lmbs);
503     cur_index++;
504     for (i = 0; i < nr_lmbs; i++) {
505         uint64_t addr = i * lmb_size;
506         uint32_t *dynamic_memory = cur_index;
507 
508         if (i >= device_lmb_start) {
509             SpaprDrc *drc;
510 
511             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
512             g_assert(drc);
513 
514             dynamic_memory[0] = cpu_to_be32(addr >> 32);
515             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
516             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
517             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
518             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
519             if (memory_region_present(get_system_memory(), addr)) {
520                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
521             } else {
522                 dynamic_memory[5] = cpu_to_be32(0);
523             }
524         } else {
525             /*
526              * LMB information for RMA, boot time RAM and gap b/n RAM and
527              * device memory region -- all these are marked as reserved
528              * and as having no valid DRC.
529              */
530             dynamic_memory[0] = cpu_to_be32(addr >> 32);
531             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
532             dynamic_memory[2] = cpu_to_be32(0);
533             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
534             dynamic_memory[4] = cpu_to_be32(-1);
535             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
536                                             SPAPR_LMB_FLAGS_DRC_INVALID);
537         }
538 
539         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
540     }
541     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
542     g_free(int_buf);
543     if (ret < 0) {
544         return -1;
545     }
546     return 0;
547 }
548 
549 /*
550  * Adds ibm,dynamic-reconfiguration-memory node.
551  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
552  * of this device tree node.
553  */
554 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
555                                                    void *fdt)
556 {
557     MachineState *machine = MACHINE(spapr);
558     int nb_numa_nodes = machine->numa_state->num_nodes;
559     int ret, i, offset;
560     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
561     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
562     uint32_t *int_buf, *cur_index, buf_len;
563     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
564     MemoryDeviceInfoList *dimms = NULL;
565 
566     /*
567      * Don't create the node if there is no device memory
568      */
569     if (machine->ram_size == machine->maxram_size) {
570         return 0;
571     }
572 
573     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
574 
575     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
576                     sizeof(prop_lmb_size));
577     if (ret < 0) {
578         return ret;
579     }
580 
581     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
582     if (ret < 0) {
583         return ret;
584     }
585 
586     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
587     if (ret < 0) {
588         return ret;
589     }
590 
591     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
592     dimms = qmp_memory_device_list();
593     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
594         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
595     } else {
596         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
597     }
598     qapi_free_MemoryDeviceInfoList(dimms);
599 
600     if (ret < 0) {
601         return ret;
602     }
603 
604     /* ibm,associativity-lookup-arrays */
605     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
606     cur_index = int_buf = g_malloc0(buf_len);
607     int_buf[0] = cpu_to_be32(nr_nodes);
608     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
609     cur_index += 2;
610     for (i = 0; i < nr_nodes; i++) {
611         uint32_t associativity[] = {
612             cpu_to_be32(0x0),
613             cpu_to_be32(0x0),
614             cpu_to_be32(0x0),
615             cpu_to_be32(i)
616         };
617         memcpy(cur_index, associativity, sizeof(associativity));
618         cur_index += 4;
619     }
620     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
621             (cur_index - int_buf) * sizeof(uint32_t));
622     g_free(int_buf);
623 
624     return ret;
625 }
626 
627 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
628 {
629     MachineState *machine = MACHINE(spapr);
630     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
631     hwaddr mem_start, node_size;
632     int i, nb_nodes = machine->numa_state->num_nodes;
633     NodeInfo *nodes = machine->numa_state->nodes;
634 
635     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
636         if (!nodes[i].node_mem) {
637             continue;
638         }
639         if (mem_start >= machine->ram_size) {
640             node_size = 0;
641         } else {
642             node_size = nodes[i].node_mem;
643             if (node_size > machine->ram_size - mem_start) {
644                 node_size = machine->ram_size - mem_start;
645             }
646         }
647         if (!mem_start) {
648             /* spapr_machine_init() checks for rma_size <= node0_size
649              * already */
650             spapr_dt_memory_node(fdt, i, 0, spapr->rma_size);
651             mem_start += spapr->rma_size;
652             node_size -= spapr->rma_size;
653         }
654         for ( ; node_size; ) {
655             hwaddr sizetmp = pow2floor(node_size);
656 
657             /* mem_start != 0 here */
658             if (ctzl(mem_start) < ctzl(sizetmp)) {
659                 sizetmp = 1ULL << ctzl(mem_start);
660             }
661 
662             spapr_dt_memory_node(fdt, i, mem_start, sizetmp);
663             node_size -= sizetmp;
664             mem_start += sizetmp;
665         }
666     }
667 
668     /* Generate ibm,dynamic-reconfiguration-memory node if required */
669     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
670         int ret;
671 
672         g_assert(smc->dr_lmb_enabled);
673         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
674         if (ret) {
675             return ret;
676         }
677     }
678 
679     return 0;
680 }
681 
682 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
683                          SpaprMachineState *spapr)
684 {
685     MachineState *ms = MACHINE(spapr);
686     PowerPCCPU *cpu = POWERPC_CPU(cs);
687     CPUPPCState *env = &cpu->env;
688     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
689     int index = spapr_get_vcpu_id(cpu);
690     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
691                        0xffffffff, 0xffffffff};
692     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
693         : SPAPR_TIMEBASE_FREQ;
694     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
695     uint32_t page_sizes_prop[64];
696     size_t page_sizes_prop_size;
697     unsigned int smp_threads = ms->smp.threads;
698     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
699     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
700     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
701     SpaprDrc *drc;
702     int drc_index;
703     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
704     int i;
705 
706     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
707     if (drc) {
708         drc_index = spapr_drc_index(drc);
709         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
710     }
711 
712     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
713     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
714 
715     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
716     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
717                            env->dcache_line_size)));
718     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
719                            env->dcache_line_size)));
720     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
721                            env->icache_line_size)));
722     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
723                            env->icache_line_size)));
724 
725     if (pcc->l1_dcache_size) {
726         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
727                                pcc->l1_dcache_size)));
728     } else {
729         warn_report("Unknown L1 dcache size for cpu");
730     }
731     if (pcc->l1_icache_size) {
732         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
733                                pcc->l1_icache_size)));
734     } else {
735         warn_report("Unknown L1 icache size for cpu");
736     }
737 
738     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
739     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
740     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
741     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
742     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
743     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
744 
745     if (env->spr_cb[SPR_PURR].oea_read) {
746         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
747     }
748     if (env->spr_cb[SPR_SPURR].oea_read) {
749         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
750     }
751 
752     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
753         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
754                           segs, sizeof(segs))));
755     }
756 
757     /* Advertise VSX (vector extensions) if available
758      *   1               == VMX / Altivec available
759      *   2               == VSX available
760      *
761      * Only CPUs for which we create core types in spapr_cpu_core.c
762      * are possible, and all of those have VMX */
763     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
764         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
765     } else {
766         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
767     }
768 
769     /* Advertise DFP (Decimal Floating Point) if available
770      *   0 / no property == no DFP
771      *   1               == DFP available */
772     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
773         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
774     }
775 
776     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
777                                                       sizeof(page_sizes_prop));
778     if (page_sizes_prop_size) {
779         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
780                           page_sizes_prop, page_sizes_prop_size)));
781     }
782 
783     spapr_dt_pa_features(spapr, cpu, fdt, offset);
784 
785     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
786                            cs->cpu_index / vcpus_per_socket)));
787 
788     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
789                       pft_size_prop, sizeof(pft_size_prop))));
790 
791     if (ms->numa_state->num_nodes > 1) {
792         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
793     }
794 
795     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
796 
797     if (pcc->radix_page_info) {
798         for (i = 0; i < pcc->radix_page_info->count; i++) {
799             radix_AP_encodings[i] =
800                 cpu_to_be32(pcc->radix_page_info->entries[i]);
801         }
802         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
803                           radix_AP_encodings,
804                           pcc->radix_page_info->count *
805                           sizeof(radix_AP_encodings[0]))));
806     }
807 
808     /*
809      * We set this property to let the guest know that it can use the large
810      * decrementer and its width in bits.
811      */
812     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
813         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
814                               pcc->lrg_decr_bits)));
815 }
816 
817 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
818 {
819     CPUState **rev;
820     CPUState *cs;
821     int n_cpus;
822     int cpus_offset;
823     char *nodename;
824     int i;
825 
826     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
827     _FDT(cpus_offset);
828     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
829     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
830 
831     /*
832      * We walk the CPUs in reverse order to ensure that CPU DT nodes
833      * created by fdt_add_subnode() end up in the right order in FDT
834      * for the guest kernel the enumerate the CPUs correctly.
835      *
836      * The CPU list cannot be traversed in reverse order, so we need
837      * to do extra work.
838      */
839     n_cpus = 0;
840     rev = NULL;
841     CPU_FOREACH(cs) {
842         rev = g_renew(CPUState *, rev, n_cpus + 1);
843         rev[n_cpus++] = cs;
844     }
845 
846     for (i = n_cpus - 1; i >= 0; i--) {
847         CPUState *cs = rev[i];
848         PowerPCCPU *cpu = POWERPC_CPU(cs);
849         int index = spapr_get_vcpu_id(cpu);
850         DeviceClass *dc = DEVICE_GET_CLASS(cs);
851         int offset;
852 
853         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
854             continue;
855         }
856 
857         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
858         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
859         g_free(nodename);
860         _FDT(offset);
861         spapr_dt_cpu(cs, fdt, offset, spapr);
862     }
863 
864     g_free(rev);
865 }
866 
867 static int spapr_dt_rng(void *fdt)
868 {
869     int node;
870     int ret;
871 
872     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
873     if (node <= 0) {
874         return -1;
875     }
876     ret = fdt_setprop_string(fdt, node, "device_type",
877                              "ibm,platform-facilities");
878     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
879     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
880 
881     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
882     if (node <= 0) {
883         return -1;
884     }
885     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
886 
887     return ret ? -1 : 0;
888 }
889 
890 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
891 {
892     MachineState *ms = MACHINE(spapr);
893     int rtas;
894     GString *hypertas = g_string_sized_new(256);
895     GString *qemu_hypertas = g_string_sized_new(256);
896     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
897     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
898         memory_region_size(&MACHINE(spapr)->device_memory->mr);
899     uint32_t lrdr_capacity[] = {
900         cpu_to_be32(max_device_addr >> 32),
901         cpu_to_be32(max_device_addr & 0xffffffff),
902         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
903         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
904     };
905     uint32_t maxdomain = cpu_to_be32(spapr->gpu_numa_id > 1 ? 1 : 0);
906     uint32_t maxdomains[] = {
907         cpu_to_be32(4),
908         maxdomain,
909         maxdomain,
910         maxdomain,
911         cpu_to_be32(spapr->gpu_numa_id),
912     };
913 
914     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
915 
916     /* hypertas */
917     add_str(hypertas, "hcall-pft");
918     add_str(hypertas, "hcall-term");
919     add_str(hypertas, "hcall-dabr");
920     add_str(hypertas, "hcall-interrupt");
921     add_str(hypertas, "hcall-tce");
922     add_str(hypertas, "hcall-vio");
923     add_str(hypertas, "hcall-splpar");
924     add_str(hypertas, "hcall-join");
925     add_str(hypertas, "hcall-bulk");
926     add_str(hypertas, "hcall-set-mode");
927     add_str(hypertas, "hcall-sprg0");
928     add_str(hypertas, "hcall-copy");
929     add_str(hypertas, "hcall-debug");
930     add_str(hypertas, "hcall-vphn");
931     add_str(qemu_hypertas, "hcall-memop1");
932 
933     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
934         add_str(hypertas, "hcall-multi-tce");
935     }
936 
937     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
938         add_str(hypertas, "hcall-hpt-resize");
939     }
940 
941     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
942                      hypertas->str, hypertas->len));
943     g_string_free(hypertas, TRUE);
944     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
945                      qemu_hypertas->str, qemu_hypertas->len));
946     g_string_free(qemu_hypertas, TRUE);
947 
948     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
949                      refpoints, sizeof(refpoints)));
950 
951     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
952                      maxdomains, sizeof(maxdomains)));
953 
954     /*
955      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
956      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
957      *
958      * The system reset requirements are driven by existing Linux and PowerVM
959      * implementation which (contrary to PAPR) saves r3 in the error log
960      * structure like machine check, so Linux expects to find the saved r3
961      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
962      * does not look at the error value).
963      *
964      * System reset interrupts are not subject to interlock like machine
965      * check, so this memory area could be corrupted if the sreset is
966      * interrupted by a machine check (or vice versa) if it was shared. To
967      * prevent this, system reset uses per-CPU areas for the sreset save
968      * area. A system reset that interrupts a system reset handler could
969      * still overwrite this area, but Linux doesn't try to recover in that
970      * case anyway.
971      *
972      * The extra 8 bytes is required because Linux's FWNMI error log check
973      * is off-by-one.
974      */
975     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_ERROR_LOG_MAX +
976 			  ms->smp.max_cpus * sizeof(uint64_t)*2 + sizeof(uint64_t)));
977     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
978                           RTAS_ERROR_LOG_MAX));
979     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
980                           RTAS_EVENT_SCAN_RATE));
981 
982     g_assert(msi_nonbroken);
983     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
984 
985     /*
986      * According to PAPR, rtas ibm,os-term does not guarantee a return
987      * back to the guest cpu.
988      *
989      * While an additional ibm,extended-os-term property indicates
990      * that rtas call return will always occur. Set this property.
991      */
992     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
993 
994     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
995                      lrdr_capacity, sizeof(lrdr_capacity)));
996 
997     spapr_dt_rtas_tokens(fdt, rtas);
998 }
999 
1000 /*
1001  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
1002  * and the XIVE features that the guest may request and thus the valid
1003  * values for bytes 23..26 of option vector 5:
1004  */
1005 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
1006                                           int chosen)
1007 {
1008     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1009 
1010     char val[2 * 4] = {
1011         23, 0x00, /* XICS / XIVE mode */
1012         24, 0x00, /* Hash/Radix, filled in below. */
1013         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1014         26, 0x40, /* Radix options: GTSE == yes. */
1015     };
1016 
1017     if (spapr->irq->xics && spapr->irq->xive) {
1018         val[1] = SPAPR_OV5_XIVE_BOTH;
1019     } else if (spapr->irq->xive) {
1020         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
1021     } else {
1022         assert(spapr->irq->xics);
1023         val[1] = SPAPR_OV5_XIVE_LEGACY;
1024     }
1025 
1026     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1027                           first_ppc_cpu->compat_pvr)) {
1028         /*
1029          * If we're in a pre POWER9 compat mode then the guest should
1030          * do hash and use the legacy interrupt mode
1031          */
1032         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1033         val[3] = 0x00; /* Hash */
1034     } else if (kvm_enabled()) {
1035         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1036             val[3] = 0x80; /* OV5_MMU_BOTH */
1037         } else if (kvmppc_has_cap_mmu_radix()) {
1038             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1039         } else {
1040             val[3] = 0x00; /* Hash */
1041         }
1042     } else {
1043         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1044         val[3] = 0xC0;
1045     }
1046     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1047                      val, sizeof(val)));
1048 }
1049 
1050 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1051 {
1052     MachineState *machine = MACHINE(spapr);
1053     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1054     int chosen;
1055 
1056     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1057 
1058     if (reset) {
1059         const char *boot_device = machine->boot_order;
1060         char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1061         size_t cb = 0;
1062         char *bootlist = get_boot_devices_list(&cb);
1063 
1064         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1065             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1066                                     machine->kernel_cmdline));
1067         }
1068 
1069         if (spapr->initrd_size) {
1070             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1071                                   spapr->initrd_base));
1072             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1073                                   spapr->initrd_base + spapr->initrd_size));
1074         }
1075 
1076         if (spapr->kernel_size) {
1077             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1078                                   cpu_to_be64(spapr->kernel_size) };
1079 
1080             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1081                          &kprop, sizeof(kprop)));
1082             if (spapr->kernel_le) {
1083                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1084             }
1085         }
1086         if (boot_menu) {
1087             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1088         }
1089         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1090         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1091         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1092 
1093         if (cb && bootlist) {
1094             int i;
1095 
1096             for (i = 0; i < cb; i++) {
1097                 if (bootlist[i] == '\n') {
1098                     bootlist[i] = ' ';
1099                 }
1100             }
1101             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1102         }
1103 
1104         if (boot_device && strlen(boot_device)) {
1105             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1106         }
1107 
1108         if (!spapr->has_graphics && stdout_path) {
1109             /*
1110              * "linux,stdout-path" and "stdout" properties are
1111              * deprecated by linux kernel. New platforms should only
1112              * use the "stdout-path" property. Set the new property
1113              * and continue using older property to remain compatible
1114              * with the existing firmware.
1115              */
1116             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1117             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1118         }
1119 
1120         /*
1121          * We can deal with BAR reallocation just fine, advertise it
1122          * to the guest
1123          */
1124         if (smc->linux_pci_probe) {
1125             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1126         }
1127 
1128         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1129 
1130         g_free(stdout_path);
1131         g_free(bootlist);
1132     }
1133 
1134     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1135 }
1136 
1137 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1138 {
1139     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1140      * KVM to work under pHyp with some guest co-operation */
1141     int hypervisor;
1142     uint8_t hypercall[16];
1143 
1144     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1145     /* indicate KVM hypercall interface */
1146     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1147     if (kvmppc_has_cap_fixup_hcalls()) {
1148         /*
1149          * Older KVM versions with older guest kernels were broken
1150          * with the magic page, don't allow the guest to map it.
1151          */
1152         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1153                                   sizeof(hypercall))) {
1154             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1155                              hypercall, sizeof(hypercall)));
1156         }
1157     }
1158 }
1159 
1160 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1161 {
1162     MachineState *machine = MACHINE(spapr);
1163     MachineClass *mc = MACHINE_GET_CLASS(machine);
1164     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1165     int ret;
1166     void *fdt;
1167     SpaprPhbState *phb;
1168     char *buf;
1169 
1170     fdt = g_malloc0(space);
1171     _FDT((fdt_create_empty_tree(fdt, space)));
1172 
1173     /* Root node */
1174     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1175     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1176     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1177 
1178     /* Guest UUID & Name*/
1179     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1180     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1181     if (qemu_uuid_set) {
1182         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1183     }
1184     g_free(buf);
1185 
1186     if (qemu_get_vm_name()) {
1187         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1188                                 qemu_get_vm_name()));
1189     }
1190 
1191     /* Host Model & Serial Number */
1192     if (spapr->host_model) {
1193         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1194     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1195         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1196         g_free(buf);
1197     }
1198 
1199     if (spapr->host_serial) {
1200         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1201     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1202         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1203         g_free(buf);
1204     }
1205 
1206     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1207     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1208 
1209     /* /interrupt controller */
1210     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1211 
1212     ret = spapr_dt_memory(spapr, fdt);
1213     if (ret < 0) {
1214         error_report("couldn't setup memory nodes in fdt");
1215         exit(1);
1216     }
1217 
1218     /* /vdevice */
1219     spapr_dt_vdevice(spapr->vio_bus, fdt);
1220 
1221     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1222         ret = spapr_dt_rng(fdt);
1223         if (ret < 0) {
1224             error_report("could not set up rng device in the fdt");
1225             exit(1);
1226         }
1227     }
1228 
1229     QLIST_FOREACH(phb, &spapr->phbs, list) {
1230         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1231         if (ret < 0) {
1232             error_report("couldn't setup PCI devices in fdt");
1233             exit(1);
1234         }
1235     }
1236 
1237     spapr_dt_cpus(fdt, spapr);
1238 
1239     if (smc->dr_lmb_enabled) {
1240         _FDT(spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1241     }
1242 
1243     if (mc->has_hotpluggable_cpus) {
1244         int offset = fdt_path_offset(fdt, "/cpus");
1245         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1246         if (ret < 0) {
1247             error_report("Couldn't set up CPU DR device tree properties");
1248             exit(1);
1249         }
1250     }
1251 
1252     /* /event-sources */
1253     spapr_dt_events(spapr, fdt);
1254 
1255     /* /rtas */
1256     spapr_dt_rtas(spapr, fdt);
1257 
1258     /* /chosen */
1259     spapr_dt_chosen(spapr, fdt, reset);
1260 
1261     /* /hypervisor */
1262     if (kvm_enabled()) {
1263         spapr_dt_hypervisor(spapr, fdt);
1264     }
1265 
1266     /* Build memory reserve map */
1267     if (reset) {
1268         if (spapr->kernel_size) {
1269             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1270                                   spapr->kernel_size)));
1271         }
1272         if (spapr->initrd_size) {
1273             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1274                                   spapr->initrd_size)));
1275         }
1276     }
1277 
1278     if (smc->dr_phb_enabled) {
1279         ret = spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1280         if (ret < 0) {
1281             error_report("Couldn't set up PHB DR device tree properties");
1282             exit(1);
1283         }
1284     }
1285 
1286     /* NVDIMM devices */
1287     if (mc->nvdimm_supported) {
1288         spapr_dt_persistent_memory(fdt);
1289     }
1290 
1291     return fdt;
1292 }
1293 
1294 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1295 {
1296     SpaprMachineState *spapr = opaque;
1297 
1298     return (addr & 0x0fffffff) + spapr->kernel_addr;
1299 }
1300 
1301 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1302                                     PowerPCCPU *cpu)
1303 {
1304     CPUPPCState *env = &cpu->env;
1305 
1306     /* The TCG path should also be holding the BQL at this point */
1307     g_assert(qemu_mutex_iothread_locked());
1308 
1309     if (msr_pr) {
1310         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1311         env->gpr[3] = H_PRIVILEGE;
1312     } else {
1313         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1314     }
1315 }
1316 
1317 struct LPCRSyncState {
1318     target_ulong value;
1319     target_ulong mask;
1320 };
1321 
1322 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1323 {
1324     struct LPCRSyncState *s = arg.host_ptr;
1325     PowerPCCPU *cpu = POWERPC_CPU(cs);
1326     CPUPPCState *env = &cpu->env;
1327     target_ulong lpcr;
1328 
1329     cpu_synchronize_state(cs);
1330     lpcr = env->spr[SPR_LPCR];
1331     lpcr &= ~s->mask;
1332     lpcr |= s->value;
1333     ppc_store_lpcr(cpu, lpcr);
1334 }
1335 
1336 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1337 {
1338     CPUState *cs;
1339     struct LPCRSyncState s = {
1340         .value = value,
1341         .mask = mask
1342     };
1343     CPU_FOREACH(cs) {
1344         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1345     }
1346 }
1347 
1348 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1349 {
1350     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1351 
1352     /* Copy PATE1:GR into PATE0:HR */
1353     entry->dw0 = spapr->patb_entry & PATE0_HR;
1354     entry->dw1 = spapr->patb_entry;
1355 }
1356 
1357 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1358 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1359 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1360 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1361 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1362 
1363 /*
1364  * Get the fd to access the kernel htab, re-opening it if necessary
1365  */
1366 static int get_htab_fd(SpaprMachineState *spapr)
1367 {
1368     Error *local_err = NULL;
1369 
1370     if (spapr->htab_fd >= 0) {
1371         return spapr->htab_fd;
1372     }
1373 
1374     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1375     if (spapr->htab_fd < 0) {
1376         error_report_err(local_err);
1377     }
1378 
1379     return spapr->htab_fd;
1380 }
1381 
1382 void close_htab_fd(SpaprMachineState *spapr)
1383 {
1384     if (spapr->htab_fd >= 0) {
1385         close(spapr->htab_fd);
1386     }
1387     spapr->htab_fd = -1;
1388 }
1389 
1390 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1391 {
1392     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1393 
1394     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1395 }
1396 
1397 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1398 {
1399     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1400 
1401     assert(kvm_enabled());
1402 
1403     if (!spapr->htab) {
1404         return 0;
1405     }
1406 
1407     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1408 }
1409 
1410 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1411                                                 hwaddr ptex, int n)
1412 {
1413     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1414     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1415 
1416     if (!spapr->htab) {
1417         /*
1418          * HTAB is controlled by KVM. Fetch into temporary buffer
1419          */
1420         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1421         kvmppc_read_hptes(hptes, ptex, n);
1422         return hptes;
1423     }
1424 
1425     /*
1426      * HTAB is controlled by QEMU. Just point to the internally
1427      * accessible PTEG.
1428      */
1429     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1430 }
1431 
1432 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1433                               const ppc_hash_pte64_t *hptes,
1434                               hwaddr ptex, int n)
1435 {
1436     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1437 
1438     if (!spapr->htab) {
1439         g_free((void *)hptes);
1440     }
1441 
1442     /* Nothing to do for qemu managed HPT */
1443 }
1444 
1445 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1446                       uint64_t pte0, uint64_t pte1)
1447 {
1448     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1449     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1450 
1451     if (!spapr->htab) {
1452         kvmppc_write_hpte(ptex, pte0, pte1);
1453     } else {
1454         if (pte0 & HPTE64_V_VALID) {
1455             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1456             /*
1457              * When setting valid, we write PTE1 first. This ensures
1458              * proper synchronization with the reading code in
1459              * ppc_hash64_pteg_search()
1460              */
1461             smp_wmb();
1462             stq_p(spapr->htab + offset, pte0);
1463         } else {
1464             stq_p(spapr->htab + offset, pte0);
1465             /*
1466              * When clearing it we set PTE0 first. This ensures proper
1467              * synchronization with the reading code in
1468              * ppc_hash64_pteg_search()
1469              */
1470             smp_wmb();
1471             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1472         }
1473     }
1474 }
1475 
1476 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1477                              uint64_t pte1)
1478 {
1479     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1480     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1481 
1482     if (!spapr->htab) {
1483         /* There should always be a hash table when this is called */
1484         error_report("spapr_hpte_set_c called with no hash table !");
1485         return;
1486     }
1487 
1488     /* The HW performs a non-atomic byte update */
1489     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1490 }
1491 
1492 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1493                              uint64_t pte1)
1494 {
1495     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1496     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1497 
1498     if (!spapr->htab) {
1499         /* There should always be a hash table when this is called */
1500         error_report("spapr_hpte_set_r called with no hash table !");
1501         return;
1502     }
1503 
1504     /* The HW performs a non-atomic byte update */
1505     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1506 }
1507 
1508 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1509 {
1510     int shift;
1511 
1512     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1513      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1514      * that's much more than is needed for Linux guests */
1515     shift = ctz64(pow2ceil(ramsize)) - 7;
1516     shift = MAX(shift, 18); /* Minimum architected size */
1517     shift = MIN(shift, 46); /* Maximum architected size */
1518     return shift;
1519 }
1520 
1521 void spapr_free_hpt(SpaprMachineState *spapr)
1522 {
1523     g_free(spapr->htab);
1524     spapr->htab = NULL;
1525     spapr->htab_shift = 0;
1526     close_htab_fd(spapr);
1527 }
1528 
1529 void spapr_reallocate_hpt(SpaprMachineState *spapr, int shift,
1530                           Error **errp)
1531 {
1532     long rc;
1533 
1534     /* Clean up any HPT info from a previous boot */
1535     spapr_free_hpt(spapr);
1536 
1537     rc = kvmppc_reset_htab(shift);
1538     if (rc < 0) {
1539         /* kernel-side HPT needed, but couldn't allocate one */
1540         error_setg_errno(errp, errno,
1541                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1542                          shift);
1543         /* This is almost certainly fatal, but if the caller really
1544          * wants to carry on with shift == 0, it's welcome to try */
1545     } else if (rc > 0) {
1546         /* kernel-side HPT allocated */
1547         if (rc != shift) {
1548             error_setg(errp,
1549                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1550                        shift, rc);
1551         }
1552 
1553         spapr->htab_shift = shift;
1554         spapr->htab = NULL;
1555     } else {
1556         /* kernel-side HPT not needed, allocate in userspace instead */
1557         size_t size = 1ULL << shift;
1558         int i;
1559 
1560         spapr->htab = qemu_memalign(size, size);
1561         if (!spapr->htab) {
1562             error_setg_errno(errp, errno,
1563                              "Could not allocate HPT of order %d", shift);
1564             return;
1565         }
1566 
1567         memset(spapr->htab, 0, size);
1568         spapr->htab_shift = shift;
1569 
1570         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1571             DIRTY_HPTE(HPTE(spapr->htab, i));
1572         }
1573     }
1574     /* We're setting up a hash table, so that means we're not radix */
1575     spapr->patb_entry = 0;
1576     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1577 }
1578 
1579 void spapr_setup_hpt(SpaprMachineState *spapr)
1580 {
1581     int hpt_shift;
1582 
1583     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1584         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1585     } else {
1586         uint64_t current_ram_size;
1587 
1588         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1589         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1590     }
1591     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1592 
1593     if (kvm_enabled()) {
1594         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1595 
1596         /* Check our RMA fits in the possible VRMA */
1597         if (vrma_limit < spapr->rma_size) {
1598             error_report("Unable to create %" HWADDR_PRIu
1599                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1600                          spapr->rma_size / MiB, vrma_limit / MiB);
1601             exit(EXIT_FAILURE);
1602         }
1603     }
1604 }
1605 
1606 static int spapr_reset_drcs(Object *child, void *opaque)
1607 {
1608     SpaprDrc *drc =
1609         (SpaprDrc *) object_dynamic_cast(child,
1610                                                  TYPE_SPAPR_DR_CONNECTOR);
1611 
1612     if (drc) {
1613         spapr_drc_reset(drc);
1614     }
1615 
1616     return 0;
1617 }
1618 
1619 static void spapr_machine_reset(MachineState *machine)
1620 {
1621     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1622     PowerPCCPU *first_ppc_cpu;
1623     hwaddr fdt_addr;
1624     void *fdt;
1625     int rc;
1626 
1627     kvmppc_svm_off(&error_fatal);
1628     spapr_caps_apply(spapr);
1629 
1630     first_ppc_cpu = POWERPC_CPU(first_cpu);
1631     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1632         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1633                               spapr->max_compat_pvr)) {
1634         /*
1635          * If using KVM with radix mode available, VCPUs can be started
1636          * without a HPT because KVM will start them in radix mode.
1637          * Set the GR bit in PATE so that we know there is no HPT.
1638          */
1639         spapr->patb_entry = PATE1_GR;
1640         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1641     } else {
1642         spapr_setup_hpt(spapr);
1643     }
1644 
1645     qemu_devices_reset();
1646 
1647     spapr_ovec_cleanup(spapr->ov5_cas);
1648     spapr->ov5_cas = spapr_ovec_new();
1649 
1650     ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1651 
1652     /*
1653      * This is fixing some of the default configuration of the XIVE
1654      * devices. To be called after the reset of the machine devices.
1655      */
1656     spapr_irq_reset(spapr, &error_fatal);
1657 
1658     /*
1659      * There is no CAS under qtest. Simulate one to please the code that
1660      * depends on spapr->ov5_cas. This is especially needed to test device
1661      * unplug, so we do that before resetting the DRCs.
1662      */
1663     if (qtest_enabled()) {
1664         spapr_ovec_cleanup(spapr->ov5_cas);
1665         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1666     }
1667 
1668     /* DRC reset may cause a device to be unplugged. This will cause troubles
1669      * if this device is used by another device (eg, a running vhost backend
1670      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1671      * situations, we reset DRCs after all devices have been reset.
1672      */
1673     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1674 
1675     spapr_clear_pending_events(spapr);
1676 
1677     /*
1678      * We place the device tree and RTAS just below either the top of the RMA,
1679      * or just below 2GB, whichever is lower, so that it can be
1680      * processed with 32-bit real mode code if necessary
1681      */
1682     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1683 
1684     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1685 
1686     rc = fdt_pack(fdt);
1687 
1688     /* Should only fail if we've built a corrupted tree */
1689     assert(rc == 0);
1690 
1691     /* Load the fdt */
1692     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1693     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1694     g_free(spapr->fdt_blob);
1695     spapr->fdt_size = fdt_totalsize(fdt);
1696     spapr->fdt_initial_size = spapr->fdt_size;
1697     spapr->fdt_blob = fdt;
1698 
1699     /* Set up the entry state */
1700     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, 0, fdt_addr, 0);
1701     first_ppc_cpu->env.gpr[5] = 0;
1702 
1703     spapr->fwnmi_system_reset_addr = -1;
1704     spapr->fwnmi_machine_check_addr = -1;
1705     spapr->fwnmi_machine_check_interlock = -1;
1706 
1707     /* Signal all vCPUs waiting on this condition */
1708     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1709 
1710     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1711 }
1712 
1713 static void spapr_create_nvram(SpaprMachineState *spapr)
1714 {
1715     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1716     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1717 
1718     if (dinfo) {
1719         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1720                             &error_fatal);
1721     }
1722 
1723     qdev_init_nofail(dev);
1724 
1725     spapr->nvram = (struct SpaprNvram *)dev;
1726 }
1727 
1728 static void spapr_rtc_create(SpaprMachineState *spapr)
1729 {
1730     object_initialize_child(OBJECT(spapr), "rtc",
1731                             &spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1732                             &error_fatal, NULL);
1733     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1734                               &error_fatal);
1735     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1736                               "date");
1737 }
1738 
1739 /* Returns whether we want to use VGA or not */
1740 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1741 {
1742     switch (vga_interface_type) {
1743     case VGA_NONE:
1744         return false;
1745     case VGA_DEVICE:
1746         return true;
1747     case VGA_STD:
1748     case VGA_VIRTIO:
1749     case VGA_CIRRUS:
1750         return pci_vga_init(pci_bus) != NULL;
1751     default:
1752         error_setg(errp,
1753                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1754         return false;
1755     }
1756 }
1757 
1758 static int spapr_pre_load(void *opaque)
1759 {
1760     int rc;
1761 
1762     rc = spapr_caps_pre_load(opaque);
1763     if (rc) {
1764         return rc;
1765     }
1766 
1767     return 0;
1768 }
1769 
1770 static int spapr_post_load(void *opaque, int version_id)
1771 {
1772     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1773     int err = 0;
1774 
1775     err = spapr_caps_post_migration(spapr);
1776     if (err) {
1777         return err;
1778     }
1779 
1780     /*
1781      * In earlier versions, there was no separate qdev for the PAPR
1782      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1783      * So when migrating from those versions, poke the incoming offset
1784      * value into the RTC device
1785      */
1786     if (version_id < 3) {
1787         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1788         if (err) {
1789             return err;
1790         }
1791     }
1792 
1793     if (kvm_enabled() && spapr->patb_entry) {
1794         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1795         bool radix = !!(spapr->patb_entry & PATE1_GR);
1796         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1797 
1798         /*
1799          * Update LPCR:HR and UPRT as they may not be set properly in
1800          * the stream
1801          */
1802         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1803                             LPCR_HR | LPCR_UPRT);
1804 
1805         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1806         if (err) {
1807             error_report("Process table config unsupported by the host");
1808             return -EINVAL;
1809         }
1810     }
1811 
1812     err = spapr_irq_post_load(spapr, version_id);
1813     if (err) {
1814         return err;
1815     }
1816 
1817     return err;
1818 }
1819 
1820 static int spapr_pre_save(void *opaque)
1821 {
1822     int rc;
1823 
1824     rc = spapr_caps_pre_save(opaque);
1825     if (rc) {
1826         return rc;
1827     }
1828 
1829     return 0;
1830 }
1831 
1832 static bool version_before_3(void *opaque, int version_id)
1833 {
1834     return version_id < 3;
1835 }
1836 
1837 static bool spapr_pending_events_needed(void *opaque)
1838 {
1839     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1840     return !QTAILQ_EMPTY(&spapr->pending_events);
1841 }
1842 
1843 static const VMStateDescription vmstate_spapr_event_entry = {
1844     .name = "spapr_event_log_entry",
1845     .version_id = 1,
1846     .minimum_version_id = 1,
1847     .fields = (VMStateField[]) {
1848         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1849         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1850         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1851                                      NULL, extended_length),
1852         VMSTATE_END_OF_LIST()
1853     },
1854 };
1855 
1856 static const VMStateDescription vmstate_spapr_pending_events = {
1857     .name = "spapr_pending_events",
1858     .version_id = 1,
1859     .minimum_version_id = 1,
1860     .needed = spapr_pending_events_needed,
1861     .fields = (VMStateField[]) {
1862         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1863                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1864         VMSTATE_END_OF_LIST()
1865     },
1866 };
1867 
1868 static bool spapr_ov5_cas_needed(void *opaque)
1869 {
1870     SpaprMachineState *spapr = opaque;
1871     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1872     bool cas_needed;
1873 
1874     /* Prior to the introduction of SpaprOptionVector, we had two option
1875      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1876      * Both of these options encode machine topology into the device-tree
1877      * in such a way that the now-booted OS should still be able to interact
1878      * appropriately with QEMU regardless of what options were actually
1879      * negotiatied on the source side.
1880      *
1881      * As such, we can avoid migrating the CAS-negotiated options if these
1882      * are the only options available on the current machine/platform.
1883      * Since these are the only options available for pseries-2.7 and
1884      * earlier, this allows us to maintain old->new/new->old migration
1885      * compatibility.
1886      *
1887      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1888      * via default pseries-2.8 machines and explicit command-line parameters.
1889      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1890      * of the actual CAS-negotiated values to continue working properly. For
1891      * example, availability of memory unplug depends on knowing whether
1892      * OV5_HP_EVT was negotiated via CAS.
1893      *
1894      * Thus, for any cases where the set of available CAS-negotiatable
1895      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1896      * include the CAS-negotiated options in the migration stream, unless
1897      * if they affect boot time behaviour only.
1898      */
1899     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1900     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1901     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1902 
1903     /* We need extra information if we have any bits outside the mask
1904      * defined above */
1905     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1906 
1907     spapr_ovec_cleanup(ov5_mask);
1908 
1909     return cas_needed;
1910 }
1911 
1912 static const VMStateDescription vmstate_spapr_ov5_cas = {
1913     .name = "spapr_option_vector_ov5_cas",
1914     .version_id = 1,
1915     .minimum_version_id = 1,
1916     .needed = spapr_ov5_cas_needed,
1917     .fields = (VMStateField[]) {
1918         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1919                                  vmstate_spapr_ovec, SpaprOptionVector),
1920         VMSTATE_END_OF_LIST()
1921     },
1922 };
1923 
1924 static bool spapr_patb_entry_needed(void *opaque)
1925 {
1926     SpaprMachineState *spapr = opaque;
1927 
1928     return !!spapr->patb_entry;
1929 }
1930 
1931 static const VMStateDescription vmstate_spapr_patb_entry = {
1932     .name = "spapr_patb_entry",
1933     .version_id = 1,
1934     .minimum_version_id = 1,
1935     .needed = spapr_patb_entry_needed,
1936     .fields = (VMStateField[]) {
1937         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1938         VMSTATE_END_OF_LIST()
1939     },
1940 };
1941 
1942 static bool spapr_irq_map_needed(void *opaque)
1943 {
1944     SpaprMachineState *spapr = opaque;
1945 
1946     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1947 }
1948 
1949 static const VMStateDescription vmstate_spapr_irq_map = {
1950     .name = "spapr_irq_map",
1951     .version_id = 1,
1952     .minimum_version_id = 1,
1953     .needed = spapr_irq_map_needed,
1954     .fields = (VMStateField[]) {
1955         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1956         VMSTATE_END_OF_LIST()
1957     },
1958 };
1959 
1960 static bool spapr_dtb_needed(void *opaque)
1961 {
1962     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1963 
1964     return smc->update_dt_enabled;
1965 }
1966 
1967 static int spapr_dtb_pre_load(void *opaque)
1968 {
1969     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1970 
1971     g_free(spapr->fdt_blob);
1972     spapr->fdt_blob = NULL;
1973     spapr->fdt_size = 0;
1974 
1975     return 0;
1976 }
1977 
1978 static const VMStateDescription vmstate_spapr_dtb = {
1979     .name = "spapr_dtb",
1980     .version_id = 1,
1981     .minimum_version_id = 1,
1982     .needed = spapr_dtb_needed,
1983     .pre_load = spapr_dtb_pre_load,
1984     .fields = (VMStateField[]) {
1985         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1986         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1987         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1988                                      fdt_size),
1989         VMSTATE_END_OF_LIST()
1990     },
1991 };
1992 
1993 static bool spapr_fwnmi_needed(void *opaque)
1994 {
1995     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1996 
1997     return spapr->fwnmi_machine_check_addr != -1;
1998 }
1999 
2000 static int spapr_fwnmi_pre_save(void *opaque)
2001 {
2002     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2003 
2004     /*
2005      * Check if machine check handling is in progress and print a
2006      * warning message.
2007      */
2008     if (spapr->fwnmi_machine_check_interlock != -1) {
2009         warn_report("A machine check is being handled during migration. The"
2010                 "handler may run and log hardware error on the destination");
2011     }
2012 
2013     return 0;
2014 }
2015 
2016 static const VMStateDescription vmstate_spapr_fwnmi = {
2017     .name = "spapr_fwnmi",
2018     .version_id = 1,
2019     .minimum_version_id = 1,
2020     .needed = spapr_fwnmi_needed,
2021     .pre_save = spapr_fwnmi_pre_save,
2022     .fields = (VMStateField[]) {
2023         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
2024         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
2025         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
2026         VMSTATE_END_OF_LIST()
2027     },
2028 };
2029 
2030 static const VMStateDescription vmstate_spapr = {
2031     .name = "spapr",
2032     .version_id = 3,
2033     .minimum_version_id = 1,
2034     .pre_load = spapr_pre_load,
2035     .post_load = spapr_post_load,
2036     .pre_save = spapr_pre_save,
2037     .fields = (VMStateField[]) {
2038         /* used to be @next_irq */
2039         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2040 
2041         /* RTC offset */
2042         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2043 
2044         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2045         VMSTATE_END_OF_LIST()
2046     },
2047     .subsections = (const VMStateDescription*[]) {
2048         &vmstate_spapr_ov5_cas,
2049         &vmstate_spapr_patb_entry,
2050         &vmstate_spapr_pending_events,
2051         &vmstate_spapr_cap_htm,
2052         &vmstate_spapr_cap_vsx,
2053         &vmstate_spapr_cap_dfp,
2054         &vmstate_spapr_cap_cfpc,
2055         &vmstate_spapr_cap_sbbc,
2056         &vmstate_spapr_cap_ibs,
2057         &vmstate_spapr_cap_hpt_maxpagesize,
2058         &vmstate_spapr_irq_map,
2059         &vmstate_spapr_cap_nested_kvm_hv,
2060         &vmstate_spapr_dtb,
2061         &vmstate_spapr_cap_large_decr,
2062         &vmstate_spapr_cap_ccf_assist,
2063         &vmstate_spapr_cap_fwnmi,
2064         &vmstate_spapr_fwnmi,
2065         NULL
2066     }
2067 };
2068 
2069 static int htab_save_setup(QEMUFile *f, void *opaque)
2070 {
2071     SpaprMachineState *spapr = opaque;
2072 
2073     /* "Iteration" header */
2074     if (!spapr->htab_shift) {
2075         qemu_put_be32(f, -1);
2076     } else {
2077         qemu_put_be32(f, spapr->htab_shift);
2078     }
2079 
2080     if (spapr->htab) {
2081         spapr->htab_save_index = 0;
2082         spapr->htab_first_pass = true;
2083     } else {
2084         if (spapr->htab_shift) {
2085             assert(kvm_enabled());
2086         }
2087     }
2088 
2089 
2090     return 0;
2091 }
2092 
2093 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2094                             int chunkstart, int n_valid, int n_invalid)
2095 {
2096     qemu_put_be32(f, chunkstart);
2097     qemu_put_be16(f, n_valid);
2098     qemu_put_be16(f, n_invalid);
2099     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2100                     HASH_PTE_SIZE_64 * n_valid);
2101 }
2102 
2103 static void htab_save_end_marker(QEMUFile *f)
2104 {
2105     qemu_put_be32(f, 0);
2106     qemu_put_be16(f, 0);
2107     qemu_put_be16(f, 0);
2108 }
2109 
2110 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2111                                  int64_t max_ns)
2112 {
2113     bool has_timeout = max_ns != -1;
2114     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2115     int index = spapr->htab_save_index;
2116     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2117 
2118     assert(spapr->htab_first_pass);
2119 
2120     do {
2121         int chunkstart;
2122 
2123         /* Consume invalid HPTEs */
2124         while ((index < htabslots)
2125                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2126             CLEAN_HPTE(HPTE(spapr->htab, index));
2127             index++;
2128         }
2129 
2130         /* Consume valid HPTEs */
2131         chunkstart = index;
2132         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2133                && HPTE_VALID(HPTE(spapr->htab, index))) {
2134             CLEAN_HPTE(HPTE(spapr->htab, index));
2135             index++;
2136         }
2137 
2138         if (index > chunkstart) {
2139             int n_valid = index - chunkstart;
2140 
2141             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2142 
2143             if (has_timeout &&
2144                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2145                 break;
2146             }
2147         }
2148     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2149 
2150     if (index >= htabslots) {
2151         assert(index == htabslots);
2152         index = 0;
2153         spapr->htab_first_pass = false;
2154     }
2155     spapr->htab_save_index = index;
2156 }
2157 
2158 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2159                                 int64_t max_ns)
2160 {
2161     bool final = max_ns < 0;
2162     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2163     int examined = 0, sent = 0;
2164     int index = spapr->htab_save_index;
2165     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2166 
2167     assert(!spapr->htab_first_pass);
2168 
2169     do {
2170         int chunkstart, invalidstart;
2171 
2172         /* Consume non-dirty HPTEs */
2173         while ((index < htabslots)
2174                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2175             index++;
2176             examined++;
2177         }
2178 
2179         chunkstart = index;
2180         /* Consume valid dirty HPTEs */
2181         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2182                && HPTE_DIRTY(HPTE(spapr->htab, index))
2183                && HPTE_VALID(HPTE(spapr->htab, index))) {
2184             CLEAN_HPTE(HPTE(spapr->htab, index));
2185             index++;
2186             examined++;
2187         }
2188 
2189         invalidstart = index;
2190         /* Consume invalid dirty HPTEs */
2191         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2192                && HPTE_DIRTY(HPTE(spapr->htab, index))
2193                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2194             CLEAN_HPTE(HPTE(spapr->htab, index));
2195             index++;
2196             examined++;
2197         }
2198 
2199         if (index > chunkstart) {
2200             int n_valid = invalidstart - chunkstart;
2201             int n_invalid = index - invalidstart;
2202 
2203             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2204             sent += index - chunkstart;
2205 
2206             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2207                 break;
2208             }
2209         }
2210 
2211         if (examined >= htabslots) {
2212             break;
2213         }
2214 
2215         if (index >= htabslots) {
2216             assert(index == htabslots);
2217             index = 0;
2218         }
2219     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2220 
2221     if (index >= htabslots) {
2222         assert(index == htabslots);
2223         index = 0;
2224     }
2225 
2226     spapr->htab_save_index = index;
2227 
2228     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2229 }
2230 
2231 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2232 #define MAX_KVM_BUF_SIZE    2048
2233 
2234 static int htab_save_iterate(QEMUFile *f, void *opaque)
2235 {
2236     SpaprMachineState *spapr = opaque;
2237     int fd;
2238     int rc = 0;
2239 
2240     /* Iteration header */
2241     if (!spapr->htab_shift) {
2242         qemu_put_be32(f, -1);
2243         return 1;
2244     } else {
2245         qemu_put_be32(f, 0);
2246     }
2247 
2248     if (!spapr->htab) {
2249         assert(kvm_enabled());
2250 
2251         fd = get_htab_fd(spapr);
2252         if (fd < 0) {
2253             return fd;
2254         }
2255 
2256         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2257         if (rc < 0) {
2258             return rc;
2259         }
2260     } else  if (spapr->htab_first_pass) {
2261         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2262     } else {
2263         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2264     }
2265 
2266     htab_save_end_marker(f);
2267 
2268     return rc;
2269 }
2270 
2271 static int htab_save_complete(QEMUFile *f, void *opaque)
2272 {
2273     SpaprMachineState *spapr = opaque;
2274     int fd;
2275 
2276     /* Iteration header */
2277     if (!spapr->htab_shift) {
2278         qemu_put_be32(f, -1);
2279         return 0;
2280     } else {
2281         qemu_put_be32(f, 0);
2282     }
2283 
2284     if (!spapr->htab) {
2285         int rc;
2286 
2287         assert(kvm_enabled());
2288 
2289         fd = get_htab_fd(spapr);
2290         if (fd < 0) {
2291             return fd;
2292         }
2293 
2294         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2295         if (rc < 0) {
2296             return rc;
2297         }
2298     } else {
2299         if (spapr->htab_first_pass) {
2300             htab_save_first_pass(f, spapr, -1);
2301         }
2302         htab_save_later_pass(f, spapr, -1);
2303     }
2304 
2305     /* End marker */
2306     htab_save_end_marker(f);
2307 
2308     return 0;
2309 }
2310 
2311 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2312 {
2313     SpaprMachineState *spapr = opaque;
2314     uint32_t section_hdr;
2315     int fd = -1;
2316     Error *local_err = NULL;
2317 
2318     if (version_id < 1 || version_id > 1) {
2319         error_report("htab_load() bad version");
2320         return -EINVAL;
2321     }
2322 
2323     section_hdr = qemu_get_be32(f);
2324 
2325     if (section_hdr == -1) {
2326         spapr_free_hpt(spapr);
2327         return 0;
2328     }
2329 
2330     if (section_hdr) {
2331         /* First section gives the htab size */
2332         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2333         if (local_err) {
2334             error_report_err(local_err);
2335             return -EINVAL;
2336         }
2337         return 0;
2338     }
2339 
2340     if (!spapr->htab) {
2341         assert(kvm_enabled());
2342 
2343         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2344         if (fd < 0) {
2345             error_report_err(local_err);
2346             return fd;
2347         }
2348     }
2349 
2350     while (true) {
2351         uint32_t index;
2352         uint16_t n_valid, n_invalid;
2353 
2354         index = qemu_get_be32(f);
2355         n_valid = qemu_get_be16(f);
2356         n_invalid = qemu_get_be16(f);
2357 
2358         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2359             /* End of Stream */
2360             break;
2361         }
2362 
2363         if ((index + n_valid + n_invalid) >
2364             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2365             /* Bad index in stream */
2366             error_report(
2367                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2368                 index, n_valid, n_invalid, spapr->htab_shift);
2369             return -EINVAL;
2370         }
2371 
2372         if (spapr->htab) {
2373             if (n_valid) {
2374                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2375                                 HASH_PTE_SIZE_64 * n_valid);
2376             }
2377             if (n_invalid) {
2378                 memset(HPTE(spapr->htab, index + n_valid), 0,
2379                        HASH_PTE_SIZE_64 * n_invalid);
2380             }
2381         } else {
2382             int rc;
2383 
2384             assert(fd >= 0);
2385 
2386             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2387             if (rc < 0) {
2388                 return rc;
2389             }
2390         }
2391     }
2392 
2393     if (!spapr->htab) {
2394         assert(fd >= 0);
2395         close(fd);
2396     }
2397 
2398     return 0;
2399 }
2400 
2401 static void htab_save_cleanup(void *opaque)
2402 {
2403     SpaprMachineState *spapr = opaque;
2404 
2405     close_htab_fd(spapr);
2406 }
2407 
2408 static SaveVMHandlers savevm_htab_handlers = {
2409     .save_setup = htab_save_setup,
2410     .save_live_iterate = htab_save_iterate,
2411     .save_live_complete_precopy = htab_save_complete,
2412     .save_cleanup = htab_save_cleanup,
2413     .load_state = htab_load,
2414 };
2415 
2416 static void spapr_boot_set(void *opaque, const char *boot_device,
2417                            Error **errp)
2418 {
2419     MachineState *machine = MACHINE(opaque);
2420     machine->boot_order = g_strdup(boot_device);
2421 }
2422 
2423 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2424 {
2425     MachineState *machine = MACHINE(spapr);
2426     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2427     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2428     int i;
2429 
2430     for (i = 0; i < nr_lmbs; i++) {
2431         uint64_t addr;
2432 
2433         addr = i * lmb_size + machine->device_memory->base;
2434         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2435                                addr / lmb_size);
2436     }
2437 }
2438 
2439 /*
2440  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2441  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2442  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2443  */
2444 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2445 {
2446     int i;
2447 
2448     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2449         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2450                    " is not aligned to %" PRIu64 " MiB",
2451                    machine->ram_size,
2452                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2453         return;
2454     }
2455 
2456     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2457         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2458                    " is not aligned to %" PRIu64 " MiB",
2459                    machine->ram_size,
2460                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2461         return;
2462     }
2463 
2464     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2465         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2466             error_setg(errp,
2467                        "Node %d memory size 0x%" PRIx64
2468                        " is not aligned to %" PRIu64 " MiB",
2469                        i, machine->numa_state->nodes[i].node_mem,
2470                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2471             return;
2472         }
2473     }
2474 }
2475 
2476 /* find cpu slot in machine->possible_cpus by core_id */
2477 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2478 {
2479     int index = id / ms->smp.threads;
2480 
2481     if (index >= ms->possible_cpus->len) {
2482         return NULL;
2483     }
2484     if (idx) {
2485         *idx = index;
2486     }
2487     return &ms->possible_cpus->cpus[index];
2488 }
2489 
2490 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2491 {
2492     MachineState *ms = MACHINE(spapr);
2493     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2494     Error *local_err = NULL;
2495     bool vsmt_user = !!spapr->vsmt;
2496     int kvm_smt = kvmppc_smt_threads();
2497     int ret;
2498     unsigned int smp_threads = ms->smp.threads;
2499 
2500     if (!kvm_enabled() && (smp_threads > 1)) {
2501         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2502                      "on a pseries machine");
2503         goto out;
2504     }
2505     if (!is_power_of_2(smp_threads)) {
2506         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2507                      "machine because it must be a power of 2", smp_threads);
2508         goto out;
2509     }
2510 
2511     /* Detemine the VSMT mode to use: */
2512     if (vsmt_user) {
2513         if (spapr->vsmt < smp_threads) {
2514             error_setg(&local_err, "Cannot support VSMT mode %d"
2515                          " because it must be >= threads/core (%d)",
2516                          spapr->vsmt, smp_threads);
2517             goto out;
2518         }
2519         /* In this case, spapr->vsmt has been set by the command line */
2520     } else if (!smc->smp_threads_vsmt) {
2521         /*
2522          * Default VSMT value is tricky, because we need it to be as
2523          * consistent as possible (for migration), but this requires
2524          * changing it for at least some existing cases.  We pick 8 as
2525          * the value that we'd get with KVM on POWER8, the
2526          * overwhelmingly common case in production systems.
2527          */
2528         spapr->vsmt = MAX(8, smp_threads);
2529     } else {
2530         spapr->vsmt = smp_threads;
2531     }
2532 
2533     /* KVM: If necessary, set the SMT mode: */
2534     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2535         ret = kvmppc_set_smt_threads(spapr->vsmt);
2536         if (ret) {
2537             /* Looks like KVM isn't able to change VSMT mode */
2538             error_setg(&local_err,
2539                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2540                        spapr->vsmt, ret);
2541             /* We can live with that if the default one is big enough
2542              * for the number of threads, and a submultiple of the one
2543              * we want.  In this case we'll waste some vcpu ids, but
2544              * behaviour will be correct */
2545             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2546                 warn_report_err(local_err);
2547                 local_err = NULL;
2548                 goto out;
2549             } else {
2550                 if (!vsmt_user) {
2551                     error_append_hint(&local_err,
2552                                       "On PPC, a VM with %d threads/core"
2553                                       " on a host with %d threads/core"
2554                                       " requires the use of VSMT mode %d.\n",
2555                                       smp_threads, kvm_smt, spapr->vsmt);
2556                 }
2557                 kvmppc_error_append_smt_possible_hint(&local_err);
2558                 goto out;
2559             }
2560         }
2561     }
2562     /* else TCG: nothing to do currently */
2563 out:
2564     error_propagate(errp, local_err);
2565 }
2566 
2567 static void spapr_init_cpus(SpaprMachineState *spapr)
2568 {
2569     MachineState *machine = MACHINE(spapr);
2570     MachineClass *mc = MACHINE_GET_CLASS(machine);
2571     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2572     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2573     const CPUArchIdList *possible_cpus;
2574     unsigned int smp_cpus = machine->smp.cpus;
2575     unsigned int smp_threads = machine->smp.threads;
2576     unsigned int max_cpus = machine->smp.max_cpus;
2577     int boot_cores_nr = smp_cpus / smp_threads;
2578     int i;
2579 
2580     possible_cpus = mc->possible_cpu_arch_ids(machine);
2581     if (mc->has_hotpluggable_cpus) {
2582         if (smp_cpus % smp_threads) {
2583             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2584                          smp_cpus, smp_threads);
2585             exit(1);
2586         }
2587         if (max_cpus % smp_threads) {
2588             error_report("max_cpus (%u) must be multiple of threads (%u)",
2589                          max_cpus, smp_threads);
2590             exit(1);
2591         }
2592     } else {
2593         if (max_cpus != smp_cpus) {
2594             error_report("This machine version does not support CPU hotplug");
2595             exit(1);
2596         }
2597         boot_cores_nr = possible_cpus->len;
2598     }
2599 
2600     if (smc->pre_2_10_has_unused_icps) {
2601         int i;
2602 
2603         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2604             /* Dummy entries get deregistered when real ICPState objects
2605              * are registered during CPU core hotplug.
2606              */
2607             pre_2_10_vmstate_register_dummy_icp(i);
2608         }
2609     }
2610 
2611     for (i = 0; i < possible_cpus->len; i++) {
2612         int core_id = i * smp_threads;
2613 
2614         if (mc->has_hotpluggable_cpus) {
2615             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2616                                    spapr_vcpu_id(spapr, core_id));
2617         }
2618 
2619         if (i < boot_cores_nr) {
2620             Object *core  = object_new(type);
2621             int nr_threads = smp_threads;
2622 
2623             /* Handle the partially filled core for older machine types */
2624             if ((i + 1) * smp_threads >= smp_cpus) {
2625                 nr_threads = smp_cpus - i * smp_threads;
2626             }
2627 
2628             object_property_set_int(core, nr_threads, "nr-threads",
2629                                     &error_fatal);
2630             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2631                                     &error_fatal);
2632             object_property_set_bool(core, true, "realized", &error_fatal);
2633 
2634             object_unref(core);
2635         }
2636     }
2637 }
2638 
2639 static PCIHostState *spapr_create_default_phb(void)
2640 {
2641     DeviceState *dev;
2642 
2643     dev = qdev_create(NULL, TYPE_SPAPR_PCI_HOST_BRIDGE);
2644     qdev_prop_set_uint32(dev, "index", 0);
2645     qdev_init_nofail(dev);
2646 
2647     return PCI_HOST_BRIDGE(dev);
2648 }
2649 
2650 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2651 {
2652     MachineState *machine = MACHINE(spapr);
2653     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2654     hwaddr rma_size = machine->ram_size;
2655     hwaddr node0_size = spapr_node0_size(machine);
2656 
2657     /* RMA has to fit in the first NUMA node */
2658     rma_size = MIN(rma_size, node0_size);
2659 
2660     /*
2661      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2662      * never exceed that
2663      */
2664     rma_size = MIN(rma_size, 1 * TiB);
2665 
2666     /*
2667      * Clamp the RMA size based on machine type.  This is for
2668      * migration compatibility with older qemu versions, which limited
2669      * the RMA size for complicated and mostly bad reasons.
2670      */
2671     if (smc->rma_limit) {
2672         rma_size = MIN(rma_size, smc->rma_limit);
2673     }
2674 
2675     if (rma_size < MIN_RMA_SLOF) {
2676         error_setg(errp,
2677                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2678                    "ldMiB guest RMA (Real Mode Area memory)",
2679                    MIN_RMA_SLOF / MiB);
2680         return 0;
2681     }
2682 
2683     return rma_size;
2684 }
2685 
2686 /* pSeries LPAR / sPAPR hardware init */
2687 static void spapr_machine_init(MachineState *machine)
2688 {
2689     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2690     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2691     MachineClass *mc = MACHINE_GET_CLASS(machine);
2692     const char *kernel_filename = machine->kernel_filename;
2693     const char *initrd_filename = machine->initrd_filename;
2694     PCIHostState *phb;
2695     int i;
2696     MemoryRegion *sysmem = get_system_memory();
2697     long load_limit, fw_size;
2698     char *filename;
2699     Error *resize_hpt_err = NULL;
2700 
2701     msi_nonbroken = true;
2702 
2703     QLIST_INIT(&spapr->phbs);
2704     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2705 
2706     /* Determine capabilities to run with */
2707     spapr_caps_init(spapr);
2708 
2709     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2710     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2711         /*
2712          * If the user explicitly requested a mode we should either
2713          * supply it, or fail completely (which we do below).  But if
2714          * it's not set explicitly, we reset our mode to something
2715          * that works
2716          */
2717         if (resize_hpt_err) {
2718             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2719             error_free(resize_hpt_err);
2720             resize_hpt_err = NULL;
2721         } else {
2722             spapr->resize_hpt = smc->resize_hpt_default;
2723         }
2724     }
2725 
2726     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2727 
2728     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2729         /*
2730          * User requested HPT resize, but this host can't supply it.  Bail out
2731          */
2732         error_report_err(resize_hpt_err);
2733         exit(1);
2734     }
2735 
2736     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2737 
2738     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2739     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2740 
2741     /*
2742      * VSMT must be set in order to be able to compute VCPU ids, ie to
2743      * call spapr_max_server_number() or spapr_vcpu_id().
2744      */
2745     spapr_set_vsmt_mode(spapr, &error_fatal);
2746 
2747     /* Set up Interrupt Controller before we create the VCPUs */
2748     spapr_irq_init(spapr, &error_fatal);
2749 
2750     /* Set up containers for ibm,client-architecture-support negotiated options
2751      */
2752     spapr->ov5 = spapr_ovec_new();
2753     spapr->ov5_cas = spapr_ovec_new();
2754 
2755     if (smc->dr_lmb_enabled) {
2756         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2757         spapr_validate_node_memory(machine, &error_fatal);
2758     }
2759 
2760     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2761 
2762     /* advertise support for dedicated HP event source to guests */
2763     if (spapr->use_hotplug_event_source) {
2764         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2765     }
2766 
2767     /* advertise support for HPT resizing */
2768     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2769         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2770     }
2771 
2772     /* advertise support for ibm,dyamic-memory-v2 */
2773     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2774 
2775     /* advertise XIVE on POWER9 machines */
2776     if (spapr->irq->xive) {
2777         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2778     }
2779 
2780     /* init CPUs */
2781     spapr_init_cpus(spapr);
2782 
2783     /*
2784      * check we don't have a memory-less/cpu-less NUMA node
2785      * Firmware relies on the existing memory/cpu topology to provide the
2786      * NUMA topology to the kernel.
2787      * And the linux kernel needs to know the NUMA topology at start
2788      * to be able to hotplug CPUs later.
2789      */
2790     if (machine->numa_state->num_nodes) {
2791         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2792             /* check for memory-less node */
2793             if (machine->numa_state->nodes[i].node_mem == 0) {
2794                 CPUState *cs;
2795                 int found = 0;
2796                 /* check for cpu-less node */
2797                 CPU_FOREACH(cs) {
2798                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2799                     if (cpu->node_id == i) {
2800                         found = 1;
2801                         break;
2802                     }
2803                 }
2804                 /* memory-less and cpu-less node */
2805                 if (!found) {
2806                     error_report(
2807                        "Memory-less/cpu-less nodes are not supported (node %d)",
2808                                  i);
2809                     exit(1);
2810                 }
2811             }
2812         }
2813 
2814     }
2815 
2816     /*
2817      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2818      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2819      * called from vPHB reset handler so we initialize the counter here.
2820      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2821      * must be equally distant from any other node.
2822      * The final value of spapr->gpu_numa_id is going to be written to
2823      * max-associativity-domains in spapr_build_fdt().
2824      */
2825     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2826 
2827     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2828         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2829                               spapr->max_compat_pvr)) {
2830         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2831         /* KVM and TCG always allow GTSE with radix... */
2832         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2833     }
2834     /* ... but not with hash (currently). */
2835 
2836     if (kvm_enabled()) {
2837         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2838         kvmppc_enable_logical_ci_hcalls();
2839         kvmppc_enable_set_mode_hcall();
2840 
2841         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2842         kvmppc_enable_clear_ref_mod_hcalls();
2843 
2844         /* Enable H_PAGE_INIT */
2845         kvmppc_enable_h_page_init();
2846     }
2847 
2848     /* map RAM */
2849     memory_region_add_subregion(sysmem, 0, machine->ram);
2850 
2851     /* always allocate the device memory information */
2852     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2853 
2854     /* initialize hotplug memory address space */
2855     if (machine->ram_size < machine->maxram_size) {
2856         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2857         /*
2858          * Limit the number of hotpluggable memory slots to half the number
2859          * slots that KVM supports, leaving the other half for PCI and other
2860          * devices. However ensure that number of slots doesn't drop below 32.
2861          */
2862         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2863                            SPAPR_MAX_RAM_SLOTS;
2864 
2865         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2866             max_memslots = SPAPR_MAX_RAM_SLOTS;
2867         }
2868         if (machine->ram_slots > max_memslots) {
2869             error_report("Specified number of memory slots %"
2870                          PRIu64" exceeds max supported %d",
2871                          machine->ram_slots, max_memslots);
2872             exit(1);
2873         }
2874 
2875         machine->device_memory->base = ROUND_UP(machine->ram_size,
2876                                                 SPAPR_DEVICE_MEM_ALIGN);
2877         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2878                            "device-memory", device_mem_size);
2879         memory_region_add_subregion(sysmem, machine->device_memory->base,
2880                                     &machine->device_memory->mr);
2881     }
2882 
2883     if (smc->dr_lmb_enabled) {
2884         spapr_create_lmb_dr_connectors(spapr);
2885     }
2886 
2887     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2888         /* Create the error string for live migration blocker */
2889         error_setg(&spapr->fwnmi_migration_blocker,
2890             "A machine check is being handled during migration. The handler"
2891             "may run and log hardware error on the destination");
2892     }
2893 
2894     if (mc->nvdimm_supported) {
2895         spapr_create_nvdimm_dr_connectors(spapr);
2896     }
2897 
2898     /* Set up RTAS event infrastructure */
2899     spapr_events_init(spapr);
2900 
2901     /* Set up the RTC RTAS interfaces */
2902     spapr_rtc_create(spapr);
2903 
2904     /* Set up VIO bus */
2905     spapr->vio_bus = spapr_vio_bus_init();
2906 
2907     for (i = 0; i < serial_max_hds(); i++) {
2908         if (serial_hd(i)) {
2909             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2910         }
2911     }
2912 
2913     /* We always have at least the nvram device on VIO */
2914     spapr_create_nvram(spapr);
2915 
2916     /*
2917      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2918      * connectors (described in root DT node's "ibm,drc-types" property)
2919      * are pre-initialized here. additional child connectors (such as
2920      * connectors for a PHBs PCI slots) are added as needed during their
2921      * parent's realization.
2922      */
2923     if (smc->dr_phb_enabled) {
2924         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2925             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2926         }
2927     }
2928 
2929     /* Set up PCI */
2930     spapr_pci_rtas_init();
2931 
2932     phb = spapr_create_default_phb();
2933 
2934     for (i = 0; i < nb_nics; i++) {
2935         NICInfo *nd = &nd_table[i];
2936 
2937         if (!nd->model) {
2938             nd->model = g_strdup("spapr-vlan");
2939         }
2940 
2941         if (g_str_equal(nd->model, "spapr-vlan") ||
2942             g_str_equal(nd->model, "ibmveth")) {
2943             spapr_vlan_create(spapr->vio_bus, nd);
2944         } else {
2945             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2946         }
2947     }
2948 
2949     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2950         spapr_vscsi_create(spapr->vio_bus);
2951     }
2952 
2953     /* Graphics */
2954     if (spapr_vga_init(phb->bus, &error_fatal)) {
2955         spapr->has_graphics = true;
2956         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2957     }
2958 
2959     if (machine->usb) {
2960         if (smc->use_ohci_by_default) {
2961             pci_create_simple(phb->bus, -1, "pci-ohci");
2962         } else {
2963             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2964         }
2965 
2966         if (spapr->has_graphics) {
2967             USBBus *usb_bus = usb_bus_find(-1);
2968 
2969             usb_create_simple(usb_bus, "usb-kbd");
2970             usb_create_simple(usb_bus, "usb-mouse");
2971         }
2972     }
2973 
2974     if (kernel_filename) {
2975         uint64_t lowaddr = 0;
2976 
2977         spapr->kernel_size = load_elf(kernel_filename, NULL,
2978                                       translate_kernel_address, spapr,
2979                                       NULL, &lowaddr, NULL, NULL, 1,
2980                                       PPC_ELF_MACHINE, 0, 0);
2981         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2982             spapr->kernel_size = load_elf(kernel_filename, NULL,
2983                                           translate_kernel_address, spapr, NULL,
2984                                           &lowaddr, NULL, NULL, 0,
2985                                           PPC_ELF_MACHINE,
2986                                           0, 0);
2987             spapr->kernel_le = spapr->kernel_size > 0;
2988         }
2989         if (spapr->kernel_size < 0) {
2990             error_report("error loading %s: %s", kernel_filename,
2991                          load_elf_strerror(spapr->kernel_size));
2992             exit(1);
2993         }
2994 
2995         /* load initrd */
2996         if (initrd_filename) {
2997             /* Try to locate the initrd in the gap between the kernel
2998              * and the firmware. Add a bit of space just in case
2999              */
3000             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
3001                                   + 0x1ffff) & ~0xffff;
3002             spapr->initrd_size = load_image_targphys(initrd_filename,
3003                                                      spapr->initrd_base,
3004                                                      load_limit
3005                                                      - spapr->initrd_base);
3006             if (spapr->initrd_size < 0) {
3007                 error_report("could not load initial ram disk '%s'",
3008                              initrd_filename);
3009                 exit(1);
3010             }
3011         }
3012     }
3013 
3014     if (bios_name == NULL) {
3015         bios_name = FW_FILE_NAME;
3016     }
3017     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
3018     if (!filename) {
3019         error_report("Could not find LPAR firmware '%s'", bios_name);
3020         exit(1);
3021     }
3022     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
3023     if (fw_size <= 0) {
3024         error_report("Could not load LPAR firmware '%s'", filename);
3025         exit(1);
3026     }
3027     g_free(filename);
3028 
3029     /* FIXME: Should register things through the MachineState's qdev
3030      * interface, this is a legacy from the sPAPREnvironment structure
3031      * which predated MachineState but had a similar function */
3032     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3033     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3034                          &savevm_htab_handlers, spapr);
3035 
3036     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine),
3037                              &error_fatal);
3038 
3039     qemu_register_boot_set(spapr_boot_set, spapr);
3040 
3041     /*
3042      * Nothing needs to be done to resume a suspended guest because
3043      * suspending does not change the machine state, so no need for
3044      * a ->wakeup method.
3045      */
3046     qemu_register_wakeup_support();
3047 
3048     if (kvm_enabled()) {
3049         /* to stop and start vmclock */
3050         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3051                                          &spapr->tb);
3052 
3053         kvmppc_spapr_enable_inkernel_multitce();
3054     }
3055 
3056     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3057 }
3058 
3059 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3060 {
3061     if (!vm_type) {
3062         return 0;
3063     }
3064 
3065     if (!strcmp(vm_type, "HV")) {
3066         return 1;
3067     }
3068 
3069     if (!strcmp(vm_type, "PR")) {
3070         return 2;
3071     }
3072 
3073     error_report("Unknown kvm-type specified '%s'", vm_type);
3074     exit(1);
3075 }
3076 
3077 /*
3078  * Implementation of an interface to adjust firmware path
3079  * for the bootindex property handling.
3080  */
3081 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3082                                    DeviceState *dev)
3083 {
3084 #define CAST(type, obj, name) \
3085     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3086     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3087     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3088     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3089 
3090     if (d) {
3091         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3092         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3093         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3094 
3095         if (spapr) {
3096             /*
3097              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3098              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3099              * 0x8000 | (target << 8) | (bus << 5) | lun
3100              * (see the "Logical unit addressing format" table in SAM5)
3101              */
3102             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3103             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3104                                    (uint64_t)id << 48);
3105         } else if (virtio) {
3106             /*
3107              * We use SRP luns of the form 01000000 | (target << 8) | lun
3108              * in the top 32 bits of the 64-bit LUN
3109              * Note: the quote above is from SLOF and it is wrong,
3110              * the actual binding is:
3111              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3112              */
3113             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3114             if (d->lun >= 256) {
3115                 /* Use the LUN "flat space addressing method" */
3116                 id |= 0x4000;
3117             }
3118             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3119                                    (uint64_t)id << 32);
3120         } else if (usb) {
3121             /*
3122              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3123              * in the top 32 bits of the 64-bit LUN
3124              */
3125             unsigned usb_port = atoi(usb->port->path);
3126             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3127             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3128                                    (uint64_t)id << 32);
3129         }
3130     }
3131 
3132     /*
3133      * SLOF probes the USB devices, and if it recognizes that the device is a
3134      * storage device, it changes its name to "storage" instead of "usb-host",
3135      * and additionally adds a child node for the SCSI LUN, so the correct
3136      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3137      */
3138     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3139         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3140         if (usb_host_dev_is_scsi_storage(usbdev)) {
3141             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3142         }
3143     }
3144 
3145     if (phb) {
3146         /* Replace "pci" with "pci@800000020000000" */
3147         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3148     }
3149 
3150     if (vsc) {
3151         /* Same logic as virtio above */
3152         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3153         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3154     }
3155 
3156     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3157         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3158         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3159         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3160     }
3161 
3162     return NULL;
3163 }
3164 
3165 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3166 {
3167     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3168 
3169     return g_strdup(spapr->kvm_type);
3170 }
3171 
3172 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3173 {
3174     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3175 
3176     g_free(spapr->kvm_type);
3177     spapr->kvm_type = g_strdup(value);
3178 }
3179 
3180 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3181 {
3182     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3183 
3184     return spapr->use_hotplug_event_source;
3185 }
3186 
3187 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3188                                             Error **errp)
3189 {
3190     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3191 
3192     spapr->use_hotplug_event_source = value;
3193 }
3194 
3195 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3196 {
3197     return true;
3198 }
3199 
3200 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3201 {
3202     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3203 
3204     switch (spapr->resize_hpt) {
3205     case SPAPR_RESIZE_HPT_DEFAULT:
3206         return g_strdup("default");
3207     case SPAPR_RESIZE_HPT_DISABLED:
3208         return g_strdup("disabled");
3209     case SPAPR_RESIZE_HPT_ENABLED:
3210         return g_strdup("enabled");
3211     case SPAPR_RESIZE_HPT_REQUIRED:
3212         return g_strdup("required");
3213     }
3214     g_assert_not_reached();
3215 }
3216 
3217 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3218 {
3219     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3220 
3221     if (strcmp(value, "default") == 0) {
3222         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3223     } else if (strcmp(value, "disabled") == 0) {
3224         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3225     } else if (strcmp(value, "enabled") == 0) {
3226         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3227     } else if (strcmp(value, "required") == 0) {
3228         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3229     } else {
3230         error_setg(errp, "Bad value for \"resize-hpt\" property");
3231     }
3232 }
3233 
3234 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3235 {
3236     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3237 
3238     if (spapr->irq == &spapr_irq_xics_legacy) {
3239         return g_strdup("legacy");
3240     } else if (spapr->irq == &spapr_irq_xics) {
3241         return g_strdup("xics");
3242     } else if (spapr->irq == &spapr_irq_xive) {
3243         return g_strdup("xive");
3244     } else if (spapr->irq == &spapr_irq_dual) {
3245         return g_strdup("dual");
3246     }
3247     g_assert_not_reached();
3248 }
3249 
3250 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3251 {
3252     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3253 
3254     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3255         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3256         return;
3257     }
3258 
3259     /* The legacy IRQ backend can not be set */
3260     if (strcmp(value, "xics") == 0) {
3261         spapr->irq = &spapr_irq_xics;
3262     } else if (strcmp(value, "xive") == 0) {
3263         spapr->irq = &spapr_irq_xive;
3264     } else if (strcmp(value, "dual") == 0) {
3265         spapr->irq = &spapr_irq_dual;
3266     } else {
3267         error_setg(errp, "Bad value for \"ic-mode\" property");
3268     }
3269 }
3270 
3271 static char *spapr_get_host_model(Object *obj, Error **errp)
3272 {
3273     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3274 
3275     return g_strdup(spapr->host_model);
3276 }
3277 
3278 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3279 {
3280     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3281 
3282     g_free(spapr->host_model);
3283     spapr->host_model = g_strdup(value);
3284 }
3285 
3286 static char *spapr_get_host_serial(Object *obj, Error **errp)
3287 {
3288     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3289 
3290     return g_strdup(spapr->host_serial);
3291 }
3292 
3293 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3294 {
3295     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3296 
3297     g_free(spapr->host_serial);
3298     spapr->host_serial = g_strdup(value);
3299 }
3300 
3301 static void spapr_instance_init(Object *obj)
3302 {
3303     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3304     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3305 
3306     spapr->htab_fd = -1;
3307     spapr->use_hotplug_event_source = true;
3308     object_property_add_str(obj, "kvm-type",
3309                             spapr_get_kvm_type, spapr_set_kvm_type);
3310     object_property_set_description(obj, "kvm-type",
3311                                     "Specifies the KVM virtualization mode (HV, PR)");
3312     object_property_add_bool(obj, "modern-hotplug-events",
3313                             spapr_get_modern_hotplug_events,
3314                             spapr_set_modern_hotplug_events);
3315     object_property_set_description(obj, "modern-hotplug-events",
3316                                     "Use dedicated hotplug event mechanism in"
3317                                     " place of standard EPOW events when possible"
3318                                     " (required for memory hot-unplug support)");
3319     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3320                             "Maximum permitted CPU compatibility mode");
3321 
3322     object_property_add_str(obj, "resize-hpt",
3323                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3324     object_property_set_description(obj, "resize-hpt",
3325                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3326     object_property_add_uint32_ptr(obj, "vsmt",
3327                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3328     object_property_set_description(obj, "vsmt",
3329                                     "Virtual SMT: KVM behaves as if this were"
3330                                     " the host's SMT mode");
3331 
3332     object_property_add_bool(obj, "vfio-no-msix-emulation",
3333                              spapr_get_msix_emulation, NULL);
3334 
3335     object_property_add_uint64_ptr(obj, "kernel-addr",
3336                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3337     object_property_set_description(obj, "kernel-addr",
3338                                     stringify(KERNEL_LOAD_ADDR)
3339                                     " for -kernel is the default");
3340     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3341     /* The machine class defines the default interrupt controller mode */
3342     spapr->irq = smc->irq;
3343     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3344                             spapr_set_ic_mode);
3345     object_property_set_description(obj, "ic-mode",
3346                  "Specifies the interrupt controller mode (xics, xive, dual)");
3347 
3348     object_property_add_str(obj, "host-model",
3349         spapr_get_host_model, spapr_set_host_model);
3350     object_property_set_description(obj, "host-model",
3351         "Host model to advertise in guest device tree");
3352     object_property_add_str(obj, "host-serial",
3353         spapr_get_host_serial, spapr_set_host_serial);
3354     object_property_set_description(obj, "host-serial",
3355         "Host serial number to advertise in guest device tree");
3356 }
3357 
3358 static void spapr_machine_finalizefn(Object *obj)
3359 {
3360     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3361 
3362     g_free(spapr->kvm_type);
3363 }
3364 
3365 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3366 {
3367     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3368     PowerPCCPU *cpu = POWERPC_CPU(cs);
3369     CPUPPCState *env = &cpu->env;
3370 
3371     cpu_synchronize_state(cs);
3372     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3373     if (spapr->fwnmi_system_reset_addr != -1) {
3374         uint64_t rtas_addr, addr;
3375 
3376         /* get rtas addr from fdt */
3377         rtas_addr = spapr_get_rtas_addr();
3378         if (!rtas_addr) {
3379             qemu_system_guest_panicked(NULL);
3380             return;
3381         }
3382 
3383         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3384         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3385         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3386         env->gpr[3] = addr;
3387     }
3388     ppc_cpu_do_system_reset(cs);
3389     if (spapr->fwnmi_system_reset_addr != -1) {
3390         env->nip = spapr->fwnmi_system_reset_addr;
3391     }
3392 }
3393 
3394 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3395 {
3396     CPUState *cs;
3397 
3398     CPU_FOREACH(cs) {
3399         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3400     }
3401 }
3402 
3403 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3404                           void *fdt, int *fdt_start_offset, Error **errp)
3405 {
3406     uint64_t addr;
3407     uint32_t node;
3408 
3409     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3410     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3411                                     &error_abort);
3412     *fdt_start_offset = spapr_dt_memory_node(fdt, node, addr,
3413                                              SPAPR_MEMORY_BLOCK_SIZE);
3414     return 0;
3415 }
3416 
3417 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3418                            bool dedicated_hp_event_source, Error **errp)
3419 {
3420     SpaprDrc *drc;
3421     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3422     int i;
3423     uint64_t addr = addr_start;
3424     bool hotplugged = spapr_drc_hotplugged(dev);
3425     Error *local_err = NULL;
3426 
3427     for (i = 0; i < nr_lmbs; i++) {
3428         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3429                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3430         g_assert(drc);
3431 
3432         spapr_drc_attach(drc, dev, &local_err);
3433         if (local_err) {
3434             while (addr > addr_start) {
3435                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3436                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3437                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3438                 spapr_drc_detach(drc);
3439             }
3440             error_propagate(errp, local_err);
3441             return;
3442         }
3443         if (!hotplugged) {
3444             spapr_drc_reset(drc);
3445         }
3446         addr += SPAPR_MEMORY_BLOCK_SIZE;
3447     }
3448     /* send hotplug notification to the
3449      * guest only in case of hotplugged memory
3450      */
3451     if (hotplugged) {
3452         if (dedicated_hp_event_source) {
3453             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3454                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3455             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3456                                                    nr_lmbs,
3457                                                    spapr_drc_index(drc));
3458         } else {
3459             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3460                                            nr_lmbs);
3461         }
3462     }
3463 }
3464 
3465 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3466                               Error **errp)
3467 {
3468     Error *local_err = NULL;
3469     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3470     PCDIMMDevice *dimm = PC_DIMM(dev);
3471     uint64_t size, addr, slot;
3472     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3473 
3474     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3475 
3476     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3477     if (local_err) {
3478         goto out;
3479     }
3480 
3481     if (!is_nvdimm) {
3482         addr = object_property_get_uint(OBJECT(dimm),
3483                                         PC_DIMM_ADDR_PROP, &local_err);
3484         if (local_err) {
3485             goto out_unplug;
3486         }
3487         spapr_add_lmbs(dev, addr, size,
3488                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3489                        &local_err);
3490     } else {
3491         slot = object_property_get_uint(OBJECT(dimm),
3492                                         PC_DIMM_SLOT_PROP, &local_err);
3493         if (local_err) {
3494             goto out_unplug;
3495         }
3496         spapr_add_nvdimm(dev, slot, &local_err);
3497     }
3498 
3499     if (local_err) {
3500         goto out_unplug;
3501     }
3502 
3503     return;
3504 
3505 out_unplug:
3506     pc_dimm_unplug(dimm, MACHINE(ms));
3507 out:
3508     error_propagate(errp, local_err);
3509 }
3510 
3511 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3512                                   Error **errp)
3513 {
3514     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3515     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3516     const MachineClass *mc = MACHINE_CLASS(smc);
3517     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3518     PCDIMMDevice *dimm = PC_DIMM(dev);
3519     Error *local_err = NULL;
3520     uint64_t size;
3521     Object *memdev;
3522     hwaddr pagesize;
3523 
3524     if (!smc->dr_lmb_enabled) {
3525         error_setg(errp, "Memory hotplug not supported for this machine");
3526         return;
3527     }
3528 
3529     if (is_nvdimm && !mc->nvdimm_supported) {
3530         error_setg(errp, "NVDIMM hotplug not supported for this machine");
3531         return;
3532     }
3533 
3534     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3535     if (local_err) {
3536         error_propagate(errp, local_err);
3537         return;
3538     }
3539 
3540     if (!is_nvdimm && size % SPAPR_MEMORY_BLOCK_SIZE) {
3541         error_setg(errp, "Hotplugged memory size must be a multiple of "
3542                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3543         return;
3544     } else if (is_nvdimm) {
3545         spapr_nvdimm_validate_opts(NVDIMM(dev), size, &local_err);
3546         if (local_err) {
3547             error_propagate(errp, local_err);
3548             return;
3549         }
3550     }
3551 
3552     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3553                                       &error_abort);
3554     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3555     spapr_check_pagesize(spapr, pagesize, &local_err);
3556     if (local_err) {
3557         error_propagate(errp, local_err);
3558         return;
3559     }
3560 
3561     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3562 }
3563 
3564 struct SpaprDimmState {
3565     PCDIMMDevice *dimm;
3566     uint32_t nr_lmbs;
3567     QTAILQ_ENTRY(SpaprDimmState) next;
3568 };
3569 
3570 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3571                                                        PCDIMMDevice *dimm)
3572 {
3573     SpaprDimmState *dimm_state = NULL;
3574 
3575     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3576         if (dimm_state->dimm == dimm) {
3577             break;
3578         }
3579     }
3580     return dimm_state;
3581 }
3582 
3583 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3584                                                       uint32_t nr_lmbs,
3585                                                       PCDIMMDevice *dimm)
3586 {
3587     SpaprDimmState *ds = NULL;
3588 
3589     /*
3590      * If this request is for a DIMM whose removal had failed earlier
3591      * (due to guest's refusal to remove the LMBs), we would have this
3592      * dimm already in the pending_dimm_unplugs list. In that
3593      * case don't add again.
3594      */
3595     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3596     if (!ds) {
3597         ds = g_malloc0(sizeof(SpaprDimmState));
3598         ds->nr_lmbs = nr_lmbs;
3599         ds->dimm = dimm;
3600         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3601     }
3602     return ds;
3603 }
3604 
3605 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3606                                               SpaprDimmState *dimm_state)
3607 {
3608     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3609     g_free(dimm_state);
3610 }
3611 
3612 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3613                                                         PCDIMMDevice *dimm)
3614 {
3615     SpaprDrc *drc;
3616     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3617                                                   &error_abort);
3618     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3619     uint32_t avail_lmbs = 0;
3620     uint64_t addr_start, addr;
3621     int i;
3622 
3623     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3624                                          &error_abort);
3625 
3626     addr = addr_start;
3627     for (i = 0; i < nr_lmbs; i++) {
3628         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3629                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3630         g_assert(drc);
3631         if (drc->dev) {
3632             avail_lmbs++;
3633         }
3634         addr += SPAPR_MEMORY_BLOCK_SIZE;
3635     }
3636 
3637     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3638 }
3639 
3640 /* Callback to be called during DRC release. */
3641 void spapr_lmb_release(DeviceState *dev)
3642 {
3643     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3644     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3645     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3646 
3647     /* This information will get lost if a migration occurs
3648      * during the unplug process. In this case recover it. */
3649     if (ds == NULL) {
3650         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3651         g_assert(ds);
3652         /* The DRC being examined by the caller at least must be counted */
3653         g_assert(ds->nr_lmbs);
3654     }
3655 
3656     if (--ds->nr_lmbs) {
3657         return;
3658     }
3659 
3660     /*
3661      * Now that all the LMBs have been removed by the guest, call the
3662      * unplug handler chain. This can never fail.
3663      */
3664     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3665     object_unparent(OBJECT(dev));
3666 }
3667 
3668 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3669 {
3670     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3671     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3672 
3673     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3674     object_property_set_bool(OBJECT(dev), false, "realized", &error_abort);
3675     spapr_pending_dimm_unplugs_remove(spapr, ds);
3676 }
3677 
3678 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3679                                         DeviceState *dev, Error **errp)
3680 {
3681     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3682     Error *local_err = NULL;
3683     PCDIMMDevice *dimm = PC_DIMM(dev);
3684     uint32_t nr_lmbs;
3685     uint64_t size, addr_start, addr;
3686     int i;
3687     SpaprDrc *drc;
3688 
3689     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3690         error_setg(&local_err,
3691                    "nvdimm device hot unplug is not supported yet.");
3692         goto out;
3693     }
3694 
3695     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3696     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3697 
3698     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3699                                          &local_err);
3700     if (local_err) {
3701         goto out;
3702     }
3703 
3704     /*
3705      * An existing pending dimm state for this DIMM means that there is an
3706      * unplug operation in progress, waiting for the spapr_lmb_release
3707      * callback to complete the job (BQL can't cover that far). In this case,
3708      * bail out to avoid detaching DRCs that were already released.
3709      */
3710     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3711         error_setg(&local_err,
3712                    "Memory unplug already in progress for device %s",
3713                    dev->id);
3714         goto out;
3715     }
3716 
3717     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3718 
3719     addr = addr_start;
3720     for (i = 0; i < nr_lmbs; i++) {
3721         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3722                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3723         g_assert(drc);
3724 
3725         spapr_drc_detach(drc);
3726         addr += SPAPR_MEMORY_BLOCK_SIZE;
3727     }
3728 
3729     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3730                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3731     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3732                                               nr_lmbs, spapr_drc_index(drc));
3733 out:
3734     error_propagate(errp, local_err);
3735 }
3736 
3737 /* Callback to be called during DRC release. */
3738 void spapr_core_release(DeviceState *dev)
3739 {
3740     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3741 
3742     /* Call the unplug handler chain. This can never fail. */
3743     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3744     object_unparent(OBJECT(dev));
3745 }
3746 
3747 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3748 {
3749     MachineState *ms = MACHINE(hotplug_dev);
3750     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3751     CPUCore *cc = CPU_CORE(dev);
3752     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3753 
3754     if (smc->pre_2_10_has_unused_icps) {
3755         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3756         int i;
3757 
3758         for (i = 0; i < cc->nr_threads; i++) {
3759             CPUState *cs = CPU(sc->threads[i]);
3760 
3761             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3762         }
3763     }
3764 
3765     assert(core_slot);
3766     core_slot->cpu = NULL;
3767     object_property_set_bool(OBJECT(dev), false, "realized", &error_abort);
3768 }
3769 
3770 static
3771 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3772                                Error **errp)
3773 {
3774     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3775     int index;
3776     SpaprDrc *drc;
3777     CPUCore *cc = CPU_CORE(dev);
3778 
3779     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3780         error_setg(errp, "Unable to find CPU core with core-id: %d",
3781                    cc->core_id);
3782         return;
3783     }
3784     if (index == 0) {
3785         error_setg(errp, "Boot CPU core may not be unplugged");
3786         return;
3787     }
3788 
3789     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3790                           spapr_vcpu_id(spapr, cc->core_id));
3791     g_assert(drc);
3792 
3793     if (!spapr_drc_unplug_requested(drc)) {
3794         spapr_drc_detach(drc);
3795         spapr_hotplug_req_remove_by_index(drc);
3796     }
3797 }
3798 
3799 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3800                            void *fdt, int *fdt_start_offset, Error **errp)
3801 {
3802     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3803     CPUState *cs = CPU(core->threads[0]);
3804     PowerPCCPU *cpu = POWERPC_CPU(cs);
3805     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3806     int id = spapr_get_vcpu_id(cpu);
3807     char *nodename;
3808     int offset;
3809 
3810     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3811     offset = fdt_add_subnode(fdt, 0, nodename);
3812     g_free(nodename);
3813 
3814     spapr_dt_cpu(cs, fdt, offset, spapr);
3815 
3816     *fdt_start_offset = offset;
3817     return 0;
3818 }
3819 
3820 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3821                             Error **errp)
3822 {
3823     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3824     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3825     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3826     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3827     CPUCore *cc = CPU_CORE(dev);
3828     CPUState *cs;
3829     SpaprDrc *drc;
3830     Error *local_err = NULL;
3831     CPUArchId *core_slot;
3832     int index;
3833     bool hotplugged = spapr_drc_hotplugged(dev);
3834     int i;
3835 
3836     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3837     if (!core_slot) {
3838         error_setg(errp, "Unable to find CPU core with core-id: %d",
3839                    cc->core_id);
3840         return;
3841     }
3842     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3843                           spapr_vcpu_id(spapr, cc->core_id));
3844 
3845     g_assert(drc || !mc->has_hotpluggable_cpus);
3846 
3847     if (drc) {
3848         spapr_drc_attach(drc, dev, &local_err);
3849         if (local_err) {
3850             error_propagate(errp, local_err);
3851             return;
3852         }
3853 
3854         if (hotplugged) {
3855             /*
3856              * Send hotplug notification interrupt to the guest only
3857              * in case of hotplugged CPUs.
3858              */
3859             spapr_hotplug_req_add_by_index(drc);
3860         } else {
3861             spapr_drc_reset(drc);
3862         }
3863     }
3864 
3865     core_slot->cpu = OBJECT(dev);
3866 
3867     if (smc->pre_2_10_has_unused_icps) {
3868         for (i = 0; i < cc->nr_threads; i++) {
3869             cs = CPU(core->threads[i]);
3870             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3871         }
3872     }
3873 
3874     /*
3875      * Set compatibility mode to match the boot CPU, which was either set
3876      * by the machine reset code or by CAS.
3877      */
3878     if (hotplugged) {
3879         for (i = 0; i < cc->nr_threads; i++) {
3880             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3881                            &local_err);
3882             if (local_err) {
3883                 error_propagate(errp, local_err);
3884                 return;
3885             }
3886         }
3887     }
3888 }
3889 
3890 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3891                                 Error **errp)
3892 {
3893     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3894     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3895     Error *local_err = NULL;
3896     CPUCore *cc = CPU_CORE(dev);
3897     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3898     const char *type = object_get_typename(OBJECT(dev));
3899     CPUArchId *core_slot;
3900     int index;
3901     unsigned int smp_threads = machine->smp.threads;
3902 
3903     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3904         error_setg(&local_err, "CPU hotplug not supported for this machine");
3905         goto out;
3906     }
3907 
3908     if (strcmp(base_core_type, type)) {
3909         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3910         goto out;
3911     }
3912 
3913     if (cc->core_id % smp_threads) {
3914         error_setg(&local_err, "invalid core id %d", cc->core_id);
3915         goto out;
3916     }
3917 
3918     /*
3919      * In general we should have homogeneous threads-per-core, but old
3920      * (pre hotplug support) machine types allow the last core to have
3921      * reduced threads as a compatibility hack for when we allowed
3922      * total vcpus not a multiple of threads-per-core.
3923      */
3924     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3925         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3926                    cc->nr_threads, smp_threads);
3927         goto out;
3928     }
3929 
3930     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3931     if (!core_slot) {
3932         error_setg(&local_err, "core id %d out of range", cc->core_id);
3933         goto out;
3934     }
3935 
3936     if (core_slot->cpu) {
3937         error_setg(&local_err, "core %d already populated", cc->core_id);
3938         goto out;
3939     }
3940 
3941     numa_cpu_pre_plug(core_slot, dev, &local_err);
3942 
3943 out:
3944     error_propagate(errp, local_err);
3945 }
3946 
3947 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3948                           void *fdt, int *fdt_start_offset, Error **errp)
3949 {
3950     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3951     int intc_phandle;
3952 
3953     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3954     if (intc_phandle <= 0) {
3955         return -1;
3956     }
3957 
3958     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3959         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3960         return -1;
3961     }
3962 
3963     /* generally SLOF creates these, for hotplug it's up to QEMU */
3964     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3965 
3966     return 0;
3967 }
3968 
3969 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3970                                Error **errp)
3971 {
3972     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3973     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3974     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3975     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3976 
3977     if (dev->hotplugged && !smc->dr_phb_enabled) {
3978         error_setg(errp, "PHB hotplug not supported for this machine");
3979         return;
3980     }
3981 
3982     if (sphb->index == (uint32_t)-1) {
3983         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3984         return;
3985     }
3986 
3987     /*
3988      * This will check that sphb->index doesn't exceed the maximum number of
3989      * PHBs for the current machine type.
3990      */
3991     smc->phb_placement(spapr, sphb->index,
3992                        &sphb->buid, &sphb->io_win_addr,
3993                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
3994                        windows_supported, sphb->dma_liobn,
3995                        &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3996                        errp);
3997 }
3998 
3999 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4000                            Error **errp)
4001 {
4002     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4003     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4004     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4005     SpaprDrc *drc;
4006     bool hotplugged = spapr_drc_hotplugged(dev);
4007     Error *local_err = NULL;
4008 
4009     if (!smc->dr_phb_enabled) {
4010         return;
4011     }
4012 
4013     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4014     /* hotplug hooks should check it's enabled before getting this far */
4015     assert(drc);
4016 
4017     spapr_drc_attach(drc, dev, &local_err);
4018     if (local_err) {
4019         error_propagate(errp, local_err);
4020         return;
4021     }
4022 
4023     if (hotplugged) {
4024         spapr_hotplug_req_add_by_index(drc);
4025     } else {
4026         spapr_drc_reset(drc);
4027     }
4028 }
4029 
4030 void spapr_phb_release(DeviceState *dev)
4031 {
4032     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4033 
4034     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4035     object_unparent(OBJECT(dev));
4036 }
4037 
4038 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4039 {
4040     object_property_set_bool(OBJECT(dev), false, "realized", &error_abort);
4041 }
4042 
4043 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4044                                      DeviceState *dev, Error **errp)
4045 {
4046     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4047     SpaprDrc *drc;
4048 
4049     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4050     assert(drc);
4051 
4052     if (!spapr_drc_unplug_requested(drc)) {
4053         spapr_drc_detach(drc);
4054         spapr_hotplug_req_remove_by_index(drc);
4055     }
4056 }
4057 
4058 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4059                                  Error **errp)
4060 {
4061     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4062     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4063 
4064     if (spapr->tpm_proxy != NULL) {
4065         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4066         return;
4067     }
4068 
4069     spapr->tpm_proxy = tpm_proxy;
4070 }
4071 
4072 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4073 {
4074     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4075 
4076     object_property_set_bool(OBJECT(dev), false, "realized", &error_abort);
4077     object_unparent(OBJECT(dev));
4078     spapr->tpm_proxy = NULL;
4079 }
4080 
4081 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4082                                       DeviceState *dev, Error **errp)
4083 {
4084     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4085         spapr_memory_plug(hotplug_dev, dev, errp);
4086     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4087         spapr_core_plug(hotplug_dev, dev, errp);
4088     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4089         spapr_phb_plug(hotplug_dev, dev, errp);
4090     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4091         spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
4092     }
4093 }
4094 
4095 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4096                                         DeviceState *dev, Error **errp)
4097 {
4098     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4099         spapr_memory_unplug(hotplug_dev, dev);
4100     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4101         spapr_core_unplug(hotplug_dev, dev);
4102     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4103         spapr_phb_unplug(hotplug_dev, dev);
4104     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4105         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4106     }
4107 }
4108 
4109 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4110                                                 DeviceState *dev, Error **errp)
4111 {
4112     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4113     MachineClass *mc = MACHINE_GET_CLASS(sms);
4114     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4115 
4116     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4117         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4118             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4119         } else {
4120             /* NOTE: this means there is a window after guest reset, prior to
4121              * CAS negotiation, where unplug requests will fail due to the
4122              * capability not being detected yet. This is a bit different than
4123              * the case with PCI unplug, where the events will be queued and
4124              * eventually handled by the guest after boot
4125              */
4126             error_setg(errp, "Memory hot unplug not supported for this guest");
4127         }
4128     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4129         if (!mc->has_hotpluggable_cpus) {
4130             error_setg(errp, "CPU hot unplug not supported on this machine");
4131             return;
4132         }
4133         spapr_core_unplug_request(hotplug_dev, dev, errp);
4134     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4135         if (!smc->dr_phb_enabled) {
4136             error_setg(errp, "PHB hot unplug not supported on this machine");
4137             return;
4138         }
4139         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4140     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4141         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4142     }
4143 }
4144 
4145 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4146                                           DeviceState *dev, Error **errp)
4147 {
4148     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4149         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4150     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4151         spapr_core_pre_plug(hotplug_dev, dev, errp);
4152     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4153         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4154     }
4155 }
4156 
4157 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4158                                                  DeviceState *dev)
4159 {
4160     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4161         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4162         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4163         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4164         return HOTPLUG_HANDLER(machine);
4165     }
4166     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4167         PCIDevice *pcidev = PCI_DEVICE(dev);
4168         PCIBus *root = pci_device_root_bus(pcidev);
4169         SpaprPhbState *phb =
4170             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4171                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4172 
4173         if (phb) {
4174             return HOTPLUG_HANDLER(phb);
4175         }
4176     }
4177     return NULL;
4178 }
4179 
4180 static CpuInstanceProperties
4181 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4182 {
4183     CPUArchId *core_slot;
4184     MachineClass *mc = MACHINE_GET_CLASS(machine);
4185 
4186     /* make sure possible_cpu are intialized */
4187     mc->possible_cpu_arch_ids(machine);
4188     /* get CPU core slot containing thread that matches cpu_index */
4189     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4190     assert(core_slot);
4191     return core_slot->props;
4192 }
4193 
4194 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4195 {
4196     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4197 }
4198 
4199 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4200 {
4201     int i;
4202     unsigned int smp_threads = machine->smp.threads;
4203     unsigned int smp_cpus = machine->smp.cpus;
4204     const char *core_type;
4205     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4206     MachineClass *mc = MACHINE_GET_CLASS(machine);
4207 
4208     if (!mc->has_hotpluggable_cpus) {
4209         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4210     }
4211     if (machine->possible_cpus) {
4212         assert(machine->possible_cpus->len == spapr_max_cores);
4213         return machine->possible_cpus;
4214     }
4215 
4216     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4217     if (!core_type) {
4218         error_report("Unable to find sPAPR CPU Core definition");
4219         exit(1);
4220     }
4221 
4222     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4223                              sizeof(CPUArchId) * spapr_max_cores);
4224     machine->possible_cpus->len = spapr_max_cores;
4225     for (i = 0; i < machine->possible_cpus->len; i++) {
4226         int core_id = i * smp_threads;
4227 
4228         machine->possible_cpus->cpus[i].type = core_type;
4229         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4230         machine->possible_cpus->cpus[i].arch_id = core_id;
4231         machine->possible_cpus->cpus[i].props.has_core_id = true;
4232         machine->possible_cpus->cpus[i].props.core_id = core_id;
4233     }
4234     return machine->possible_cpus;
4235 }
4236 
4237 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4238                                 uint64_t *buid, hwaddr *pio,
4239                                 hwaddr *mmio32, hwaddr *mmio64,
4240                                 unsigned n_dma, uint32_t *liobns,
4241                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4242 {
4243     /*
4244      * New-style PHB window placement.
4245      *
4246      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4247      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4248      * windows.
4249      *
4250      * Some guest kernels can't work with MMIO windows above 1<<46
4251      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4252      *
4253      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4254      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4255      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4256      * 1TiB 64-bit MMIO windows for each PHB.
4257      */
4258     const uint64_t base_buid = 0x800000020000000ULL;
4259     int i;
4260 
4261     /* Sanity check natural alignments */
4262     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4263     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4264     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4265     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4266     /* Sanity check bounds */
4267     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4268                       SPAPR_PCI_MEM32_WIN_SIZE);
4269     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4270                       SPAPR_PCI_MEM64_WIN_SIZE);
4271 
4272     if (index >= SPAPR_MAX_PHBS) {
4273         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4274                    SPAPR_MAX_PHBS - 1);
4275         return;
4276     }
4277 
4278     *buid = base_buid + index;
4279     for (i = 0; i < n_dma; ++i) {
4280         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4281     }
4282 
4283     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4284     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4285     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4286 
4287     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4288     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4289 }
4290 
4291 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4292 {
4293     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4294 
4295     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4296 }
4297 
4298 static void spapr_ics_resend(XICSFabric *dev)
4299 {
4300     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4301 
4302     ics_resend(spapr->ics);
4303 }
4304 
4305 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4306 {
4307     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4308 
4309     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4310 }
4311 
4312 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4313                                  Monitor *mon)
4314 {
4315     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4316 
4317     spapr_irq_print_info(spapr, mon);
4318     monitor_printf(mon, "irqchip: %s\n",
4319                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4320 }
4321 
4322 /*
4323  * This is a XIVE only operation
4324  */
4325 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4326                            uint8_t nvt_blk, uint32_t nvt_idx,
4327                            bool cam_ignore, uint8_t priority,
4328                            uint32_t logic_serv, XiveTCTXMatch *match)
4329 {
4330     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4331     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4332     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4333     int count;
4334 
4335     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4336                            priority, logic_serv, match);
4337     if (count < 0) {
4338         return count;
4339     }
4340 
4341     /*
4342      * When we implement the save and restore of the thread interrupt
4343      * contexts in the enter/exit CPU handlers of the machine and the
4344      * escalations in QEMU, we should be able to handle non dispatched
4345      * vCPUs.
4346      *
4347      * Until this is done, the sPAPR machine should find at least one
4348      * matching context always.
4349      */
4350     if (count == 0) {
4351         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4352                       nvt_blk, nvt_idx);
4353     }
4354 
4355     return count;
4356 }
4357 
4358 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4359 {
4360     return cpu->vcpu_id;
4361 }
4362 
4363 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4364 {
4365     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4366     MachineState *ms = MACHINE(spapr);
4367     int vcpu_id;
4368 
4369     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4370 
4371     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4372         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4373         error_append_hint(errp, "Adjust the number of cpus to %d "
4374                           "or try to raise the number of threads per core\n",
4375                           vcpu_id * ms->smp.threads / spapr->vsmt);
4376         return;
4377     }
4378 
4379     cpu->vcpu_id = vcpu_id;
4380 }
4381 
4382 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4383 {
4384     CPUState *cs;
4385 
4386     CPU_FOREACH(cs) {
4387         PowerPCCPU *cpu = POWERPC_CPU(cs);
4388 
4389         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4390             return cpu;
4391         }
4392     }
4393 
4394     return NULL;
4395 }
4396 
4397 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4398 {
4399     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4400 
4401     /* These are only called by TCG, KVM maintains dispatch state */
4402 
4403     spapr_cpu->prod = false;
4404     if (spapr_cpu->vpa_addr) {
4405         CPUState *cs = CPU(cpu);
4406         uint32_t dispatch;
4407 
4408         dispatch = ldl_be_phys(cs->as,
4409                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4410         dispatch++;
4411         if ((dispatch & 1) != 0) {
4412             qemu_log_mask(LOG_GUEST_ERROR,
4413                           "VPA: incorrect dispatch counter value for "
4414                           "dispatched partition %u, correcting.\n", dispatch);
4415             dispatch++;
4416         }
4417         stl_be_phys(cs->as,
4418                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4419     }
4420 }
4421 
4422 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4423 {
4424     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4425 
4426     if (spapr_cpu->vpa_addr) {
4427         CPUState *cs = CPU(cpu);
4428         uint32_t dispatch;
4429 
4430         dispatch = ldl_be_phys(cs->as,
4431                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4432         dispatch++;
4433         if ((dispatch & 1) != 1) {
4434             qemu_log_mask(LOG_GUEST_ERROR,
4435                           "VPA: incorrect dispatch counter value for "
4436                           "preempted partition %u, correcting.\n", dispatch);
4437             dispatch++;
4438         }
4439         stl_be_phys(cs->as,
4440                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4441     }
4442 }
4443 
4444 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4445 {
4446     MachineClass *mc = MACHINE_CLASS(oc);
4447     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4448     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4449     NMIClass *nc = NMI_CLASS(oc);
4450     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4451     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4452     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4453     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4454     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4455 
4456     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4457     mc->ignore_boot_device_suffixes = true;
4458 
4459     /*
4460      * We set up the default / latest behaviour here.  The class_init
4461      * functions for the specific versioned machine types can override
4462      * these details for backwards compatibility
4463      */
4464     mc->init = spapr_machine_init;
4465     mc->reset = spapr_machine_reset;
4466     mc->block_default_type = IF_SCSI;
4467     mc->max_cpus = 1024;
4468     mc->no_parallel = 1;
4469     mc->default_boot_order = "";
4470     mc->default_ram_size = 512 * MiB;
4471     mc->default_ram_id = "ppc_spapr.ram";
4472     mc->default_display = "std";
4473     mc->kvm_type = spapr_kvm_type;
4474     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4475     mc->pci_allow_0_address = true;
4476     assert(!mc->get_hotplug_handler);
4477     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4478     hc->pre_plug = spapr_machine_device_pre_plug;
4479     hc->plug = spapr_machine_device_plug;
4480     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4481     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4482     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4483     hc->unplug_request = spapr_machine_device_unplug_request;
4484     hc->unplug = spapr_machine_device_unplug;
4485 
4486     smc->dr_lmb_enabled = true;
4487     smc->update_dt_enabled = true;
4488     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4489     mc->has_hotpluggable_cpus = true;
4490     mc->nvdimm_supported = true;
4491     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4492     fwc->get_dev_path = spapr_get_fw_dev_path;
4493     nc->nmi_monitor_handler = spapr_nmi;
4494     smc->phb_placement = spapr_phb_placement;
4495     vhc->hypercall = emulate_spapr_hypercall;
4496     vhc->hpt_mask = spapr_hpt_mask;
4497     vhc->map_hptes = spapr_map_hptes;
4498     vhc->unmap_hptes = spapr_unmap_hptes;
4499     vhc->hpte_set_c = spapr_hpte_set_c;
4500     vhc->hpte_set_r = spapr_hpte_set_r;
4501     vhc->get_pate = spapr_get_pate;
4502     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4503     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4504     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4505     xic->ics_get = spapr_ics_get;
4506     xic->ics_resend = spapr_ics_resend;
4507     xic->icp_get = spapr_icp_get;
4508     ispc->print_info = spapr_pic_print_info;
4509     /* Force NUMA node memory size to be a multiple of
4510      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4511      * in which LMBs are represented and hot-added
4512      */
4513     mc->numa_mem_align_shift = 28;
4514     mc->numa_mem_supported = true;
4515     mc->auto_enable_numa = true;
4516 
4517     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4518     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4519     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4520     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4521     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4522     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4523     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4524     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4525     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4526     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4527     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4528     spapr_caps_add_properties(smc);
4529     smc->irq = &spapr_irq_dual;
4530     smc->dr_phb_enabled = true;
4531     smc->linux_pci_probe = true;
4532     smc->smp_threads_vsmt = true;
4533     smc->nr_xirqs = SPAPR_NR_XIRQS;
4534     xfc->match_nvt = spapr_match_nvt;
4535 }
4536 
4537 static const TypeInfo spapr_machine_info = {
4538     .name          = TYPE_SPAPR_MACHINE,
4539     .parent        = TYPE_MACHINE,
4540     .abstract      = true,
4541     .instance_size = sizeof(SpaprMachineState),
4542     .instance_init = spapr_instance_init,
4543     .instance_finalize = spapr_machine_finalizefn,
4544     .class_size    = sizeof(SpaprMachineClass),
4545     .class_init    = spapr_machine_class_init,
4546     .interfaces = (InterfaceInfo[]) {
4547         { TYPE_FW_PATH_PROVIDER },
4548         { TYPE_NMI },
4549         { TYPE_HOTPLUG_HANDLER },
4550         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4551         { TYPE_XICS_FABRIC },
4552         { TYPE_INTERRUPT_STATS_PROVIDER },
4553         { TYPE_XIVE_FABRIC },
4554         { }
4555     },
4556 };
4557 
4558 static void spapr_machine_latest_class_options(MachineClass *mc)
4559 {
4560     mc->alias = "pseries";
4561     mc->is_default = true;
4562 }
4563 
4564 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4565     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4566                                                     void *data)      \
4567     {                                                                \
4568         MachineClass *mc = MACHINE_CLASS(oc);                        \
4569         spapr_machine_##suffix##_class_options(mc);                  \
4570         if (latest) {                                                \
4571             spapr_machine_latest_class_options(mc);                  \
4572         }                                                            \
4573     }                                                                \
4574     static const TypeInfo spapr_machine_##suffix##_info = {          \
4575         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4576         .parent = TYPE_SPAPR_MACHINE,                                \
4577         .class_init = spapr_machine_##suffix##_class_init,           \
4578     };                                                               \
4579     static void spapr_machine_register_##suffix(void)                \
4580     {                                                                \
4581         type_register(&spapr_machine_##suffix##_info);               \
4582     }                                                                \
4583     type_init(spapr_machine_register_##suffix)
4584 
4585 /*
4586  * pseries-5.1
4587  */
4588 static void spapr_machine_5_1_class_options(MachineClass *mc)
4589 {
4590     /* Defaults for the latest behaviour inherited from the base class */
4591 }
4592 
4593 DEFINE_SPAPR_MACHINE(5_1, "5.1", true);
4594 
4595 /*
4596  * pseries-5.0
4597  */
4598 static void spapr_machine_5_0_class_options(MachineClass *mc)
4599 {
4600     spapr_machine_5_1_class_options(mc);
4601     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4602 }
4603 
4604 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4605 
4606 /*
4607  * pseries-4.2
4608  */
4609 static void spapr_machine_4_2_class_options(MachineClass *mc)
4610 {
4611     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4612 
4613     spapr_machine_5_0_class_options(mc);
4614     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4615     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4616     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4617     smc->rma_limit = 16 * GiB;
4618     mc->nvdimm_supported = false;
4619 }
4620 
4621 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4622 
4623 /*
4624  * pseries-4.1
4625  */
4626 static void spapr_machine_4_1_class_options(MachineClass *mc)
4627 {
4628     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4629     static GlobalProperty compat[] = {
4630         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4631         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4632     };
4633 
4634     spapr_machine_4_2_class_options(mc);
4635     smc->linux_pci_probe = false;
4636     smc->smp_threads_vsmt = false;
4637     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4638     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4639 }
4640 
4641 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4642 
4643 /*
4644  * pseries-4.0
4645  */
4646 static void phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4647                               uint64_t *buid, hwaddr *pio,
4648                               hwaddr *mmio32, hwaddr *mmio64,
4649                               unsigned n_dma, uint32_t *liobns,
4650                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4651 {
4652     spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, liobns,
4653                         nv2gpa, nv2atsd, errp);
4654     *nv2gpa = 0;
4655     *nv2atsd = 0;
4656 }
4657 
4658 static void spapr_machine_4_0_class_options(MachineClass *mc)
4659 {
4660     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4661 
4662     spapr_machine_4_1_class_options(mc);
4663     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4664     smc->phb_placement = phb_placement_4_0;
4665     smc->irq = &spapr_irq_xics;
4666     smc->pre_4_1_migration = true;
4667 }
4668 
4669 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4670 
4671 /*
4672  * pseries-3.1
4673  */
4674 static void spapr_machine_3_1_class_options(MachineClass *mc)
4675 {
4676     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4677 
4678     spapr_machine_4_0_class_options(mc);
4679     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4680 
4681     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4682     smc->update_dt_enabled = false;
4683     smc->dr_phb_enabled = false;
4684     smc->broken_host_serial_model = true;
4685     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4686     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4687     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4688     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4689 }
4690 
4691 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4692 
4693 /*
4694  * pseries-3.0
4695  */
4696 
4697 static void spapr_machine_3_0_class_options(MachineClass *mc)
4698 {
4699     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4700 
4701     spapr_machine_3_1_class_options(mc);
4702     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4703 
4704     smc->legacy_irq_allocation = true;
4705     smc->nr_xirqs = 0x400;
4706     smc->irq = &spapr_irq_xics_legacy;
4707 }
4708 
4709 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4710 
4711 /*
4712  * pseries-2.12
4713  */
4714 static void spapr_machine_2_12_class_options(MachineClass *mc)
4715 {
4716     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4717     static GlobalProperty compat[] = {
4718         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4719         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4720     };
4721 
4722     spapr_machine_3_0_class_options(mc);
4723     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4724     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4725 
4726     /* We depend on kvm_enabled() to choose a default value for the
4727      * hpt-max-page-size capability. Of course we can't do it here
4728      * because this is too early and the HW accelerator isn't initialzed
4729      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4730      */
4731     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4732 }
4733 
4734 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4735 
4736 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4737 {
4738     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4739 
4740     spapr_machine_2_12_class_options(mc);
4741     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4742     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4743     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4744 }
4745 
4746 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4747 
4748 /*
4749  * pseries-2.11
4750  */
4751 
4752 static void spapr_machine_2_11_class_options(MachineClass *mc)
4753 {
4754     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4755 
4756     spapr_machine_2_12_class_options(mc);
4757     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4758     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4759 }
4760 
4761 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4762 
4763 /*
4764  * pseries-2.10
4765  */
4766 
4767 static void spapr_machine_2_10_class_options(MachineClass *mc)
4768 {
4769     spapr_machine_2_11_class_options(mc);
4770     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4771 }
4772 
4773 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4774 
4775 /*
4776  * pseries-2.9
4777  */
4778 
4779 static void spapr_machine_2_9_class_options(MachineClass *mc)
4780 {
4781     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4782     static GlobalProperty compat[] = {
4783         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4784     };
4785 
4786     spapr_machine_2_10_class_options(mc);
4787     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4788     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4789     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4790     smc->pre_2_10_has_unused_icps = true;
4791     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4792 }
4793 
4794 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4795 
4796 /*
4797  * pseries-2.8
4798  */
4799 
4800 static void spapr_machine_2_8_class_options(MachineClass *mc)
4801 {
4802     static GlobalProperty compat[] = {
4803         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4804     };
4805 
4806     spapr_machine_2_9_class_options(mc);
4807     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4808     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4809     mc->numa_mem_align_shift = 23;
4810 }
4811 
4812 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4813 
4814 /*
4815  * pseries-2.7
4816  */
4817 
4818 static void phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4819                               uint64_t *buid, hwaddr *pio,
4820                               hwaddr *mmio32, hwaddr *mmio64,
4821                               unsigned n_dma, uint32_t *liobns,
4822                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4823 {
4824     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4825     const uint64_t base_buid = 0x800000020000000ULL;
4826     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4827     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4828     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4829     const uint32_t max_index = 255;
4830     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4831 
4832     uint64_t ram_top = MACHINE(spapr)->ram_size;
4833     hwaddr phb0_base, phb_base;
4834     int i;
4835 
4836     /* Do we have device memory? */
4837     if (MACHINE(spapr)->maxram_size > ram_top) {
4838         /* Can't just use maxram_size, because there may be an
4839          * alignment gap between normal and device memory regions
4840          */
4841         ram_top = MACHINE(spapr)->device_memory->base +
4842             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4843     }
4844 
4845     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4846 
4847     if (index > max_index) {
4848         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4849                    max_index);
4850         return;
4851     }
4852 
4853     *buid = base_buid + index;
4854     for (i = 0; i < n_dma; ++i) {
4855         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4856     }
4857 
4858     phb_base = phb0_base + index * phb_spacing;
4859     *pio = phb_base + pio_offset;
4860     *mmio32 = phb_base + mmio_offset;
4861     /*
4862      * We don't set the 64-bit MMIO window, relying on the PHB's
4863      * fallback behaviour of automatically splitting a large "32-bit"
4864      * window into contiguous 32-bit and 64-bit windows
4865      */
4866 
4867     *nv2gpa = 0;
4868     *nv2atsd = 0;
4869 }
4870 
4871 static void spapr_machine_2_7_class_options(MachineClass *mc)
4872 {
4873     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4874     static GlobalProperty compat[] = {
4875         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4876         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4877         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4878         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4879     };
4880 
4881     spapr_machine_2_8_class_options(mc);
4882     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4883     mc->default_machine_opts = "modern-hotplug-events=off";
4884     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4885     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4886     smc->phb_placement = phb_placement_2_7;
4887 }
4888 
4889 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4890 
4891 /*
4892  * pseries-2.6
4893  */
4894 
4895 static void spapr_machine_2_6_class_options(MachineClass *mc)
4896 {
4897     static GlobalProperty compat[] = {
4898         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4899     };
4900 
4901     spapr_machine_2_7_class_options(mc);
4902     mc->has_hotpluggable_cpus = false;
4903     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4904     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4905 }
4906 
4907 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4908 
4909 /*
4910  * pseries-2.5
4911  */
4912 
4913 static void spapr_machine_2_5_class_options(MachineClass *mc)
4914 {
4915     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4916     static GlobalProperty compat[] = {
4917         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4918     };
4919 
4920     spapr_machine_2_6_class_options(mc);
4921     smc->use_ohci_by_default = true;
4922     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4923     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4924 }
4925 
4926 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4927 
4928 /*
4929  * pseries-2.4
4930  */
4931 
4932 static void spapr_machine_2_4_class_options(MachineClass *mc)
4933 {
4934     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4935 
4936     spapr_machine_2_5_class_options(mc);
4937     smc->dr_lmb_enabled = false;
4938     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4939 }
4940 
4941 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4942 
4943 /*
4944  * pseries-2.3
4945  */
4946 
4947 static void spapr_machine_2_3_class_options(MachineClass *mc)
4948 {
4949     static GlobalProperty compat[] = {
4950         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4951     };
4952     spapr_machine_2_4_class_options(mc);
4953     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4954     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4955 }
4956 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4957 
4958 /*
4959  * pseries-2.2
4960  */
4961 
4962 static void spapr_machine_2_2_class_options(MachineClass *mc)
4963 {
4964     static GlobalProperty compat[] = {
4965         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4966     };
4967 
4968     spapr_machine_2_3_class_options(mc);
4969     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4970     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4971     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4972 }
4973 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4974 
4975 /*
4976  * pseries-2.1
4977  */
4978 
4979 static void spapr_machine_2_1_class_options(MachineClass *mc)
4980 {
4981     spapr_machine_2_2_class_options(mc);
4982     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4983 }
4984 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4985 
4986 static void spapr_machine_register_types(void)
4987 {
4988     type_register_static(&spapr_machine_info);
4989 }
4990 
4991 type_init(spapr_machine_register_types)
4992